ps gis es Sl a Se re i ae. | ' PR sell) PROBABILITY CONCEPTS MCD st SIN TC UTES AUPA TUL USN CRETE IM LCM B A eT ALFREDO H-S. ANG e@ WILSON H. TANG Ne BCONVERSION FACTORS: U.S. CUSTOMARY AND SI METRIC UNITS To Convert from U.S. U.S. Units SI Metric Units Customary to SI, Multiply by inch (in.) meter (m) inch (in.) ‘centimeter (em) inch ¢in.) millimeter (mm) feet (i meter (a) yard (yd) meter (m) mile (mi kilometer (km) angle, degree (°) radian (rad) temp., degree (F) Celsius °C) acre (acre) hectare (ha) allon (gal) liter ) pound (Ib) ton (ton, 2000 Ib) pound force (Ib) ‘newton (N) poundéin? (psi) newton/m? (N/m?) poundi/ft® (psf) newtonim? (Nit?) Foot pound (ft-Ib) joule (J) horsepower (hp) watt (W) British thermal unit (BTU joule (3) British thermal unit (BTU) _kilowatt-hour (kwh) Definitions (000254 254 254 0.2305 os 1.609 0.0174 CP-3 0.40% 3.79 0.4536 907.2 4.448 6895 47.88 1 newton—force that will give a I-kg mass an acceleration of | m/sec 1 joule—work done by a force af 1 N over a displacement of 1 m 1 pascal = 1 N/m 1 kilogram force (kgf) = 9.807 N | gravity acceleration (g) = 9.807 misec? I acre (a) = 4052 m: Inectare (ha) = 10,000 m: kip (kip) = 1000 IbProbability Concepts in Engineering Emphasis on Applications in Civil & Environmental EngineeringQuo comon Probability Concepts in Engineering* Emphasis on Applications in Civil & Environmental Engineering WILSON H. TANG Chair Professor. Kong Unis Science d Techno ema 620.00151 aAS4P wiley JOHN WILEY & SONS, INC.ASSOCIATE PUBLISHER Daniel Saye Copyright © 2007 John Wiley & 5 al ‘No pat of his puiiation may be reproduce the appropri f onyrght Clerance m9 78-6008, website p/w wiley omgofpermsion 00 POR PROSE Pans 328964 “7Dedicated to Myrile Mae and BernadettePreface OBJECTIVES AND APPROACH The frst edition ofthis book ( ‘and Design, Vol. 1, Basic Princip probability and statistical cone the basics ally tiled Pr in Engin ublished in 19 3s in terms that are more easily understood by tengincering students, The basie principles are presented and illustrated through problems of relevance to engineering and the physical sciences (particulaely to civil and environmental designed to further enhance the engineering), and the exercise problems in each chapter a understanding of the basic concepts and reinforce a working knowledge of the concepts and methods. We firmly believe that the easiest and most effective way for engineers to new set of abstract principles, such as those of probability and statistics, to the extent as to be able to apply them in modeling and formulating engineering problems, is through varied illustrations of applications of such principles. Moreover, the first exposure of engine to probabilistic concepts and methods should be in physically meaningful terms; this is ificant roles of the necessary to properly emphasize and motivate the recognition of the s relevant mathematical concepts in engineering. NEW TO THIS SECOND EDITION This second edition follows the same general approach expounded in the first, faved and has been thorough ition updated, oF completely replaced from that of the first edition. In particular, almost all th illustrative examples, as well as the problems, in each chapter of the fist edition have been ;eplaced with new ones. Also, where actual data are illustrated, they have been updated: more recent or new data are presented or added in this edition, Several new topics and + Two Types of Uncertainty In this edition, 1 Gintroduced in Chapter 1) namely, the aleatory and the epistemic uncertainties, and the n hhave empha importance of distinguishing the two broad ty to evaluate their respective significances separately in engineering applications. Engineers and en neering students need 10 bs aware of this approach, especially in practical engineering decision making. Nonetheless, the tools for such evaluations require the + Extreme Values | Chapter 4 now includes the distributions of extreme values, which Jealing with natural and extreme hazards ‘+ Hypothesis Testing The topic of Hypothesis Testing is added as part of Chapter 6 Da ance when the tals of distributions are + Anderson-Darling Method The method for goodness-of-fit test is now included in Chapter 7 (of *+ Confidence Intervals in Regression Analysis Linear regression in Chapter 8 has been expanded to include the determination of confidence intervals, + Regression and Correlation Analyses Chapter 9 on Bayesian probability now includes Bayesian regression and corre ses.vili > Preface merical and Simulation Methods The new chapte sr Based Numerical and Sim + Computer-Based (Chapter 5) on Cor should make this seco The numerical and simulation methods presented in this chapter, particularly with nce to Monte Carlo simulations, extend the practical applicability of probabil ity concepts and methods for formulating and solving engineering problems, beyond powerful with the present-day availablity of personal computers and related com: mercial software, and should serve to augment the analytical methods, thus extending the general usefulness and utility of probability and statistics in engineering. + Quality Assurance | Thechapteron Ele ‘eand Accept Sampling (Chapter 9 of the fist edition) is now nts of Qu signated as Chapter 10 but isa il able only on the Web, The material in this chapter is beyond the scope of this book which is devoted to the basic fundamentals of probability and statistics; however, it is useful in the specialized area of quality assurance. For these reasons, the ma Fial in this chapter is made a college/ang. ilable online at the Wiley Web site, www.wiley.con INTENDED AUDIENCE The material in the book is intended fora first course on applied probability and statistics for engineering students at the sophomore or junior level, or for self-study, stressing probabilis tic modeling and the fundamentals of statistical inferences. The primary aim is to provide ofthe fundamentals for the proper application in engineering of elementary calculus is required, and thus the material can be taught to an in-depth understan: Only knoviled undergraduate engineering students at any level. tm: in the engineering departments or offered for engineers by the departments of mathematics and statisties. “The books self-contained and thus isalso suitable for self-study by practicing engineers who desire a reading and working knowledge of the basic concepts and tools of probability SUGGESTED SYLLABUS One-semester course A suge sted outline for a one-semester (or one-quarter) course may be as follows: Chapter I (assigned as required reading with guidance from instructor fhrough Chapter 5 stressing the modeling of probabilistic problems, plus Chapters 6 through 8stressing the fundamentals of inferential statistics, can be covered in aone-semester course One-quarter course Fora one-quarter course, the same chapters may be covered with les ‘emphasis on selected sections (¢ and limit the number of illustrations in each chapter. Senior-level course For a course at the senior level, all the chapters, including the first iscuss fewer types of useful probability distribution part of Chapter 9, may be covered! in one semeste The extensive variety of problems at the end of ments and also opportunities INSTRUCTOR RESOURCES These instructor resources are available on the Instructor section of the Web site at www.wiley.com/college/ang, They are available only to instructors who adopt the tex + Solutions Manual: Solutions to ll the exercise problems in the textPreface + ix + Image Gallery: All igures and tables from the text, appropriate for use in Powe! Point presentations, These resources are password protected. Visit the Instructor section of the book Web site ter for a password to access these materials, MATHEMATICAL RIGOR MOTIVATION We have not emphasized mathematical rigor throughout the book; such aay be stip mented with treatises on the mathematical theory of probability and statistics. We are concerned mainly with the practical applications and relevance of probability concepts 10 roblems and through illustrations of probabilistic modeling of physical situations id, only the essential principles of mathematical theory are iscussed, and these principles are explained in non-abstract terms in order to stress their ‘ecognition of the practical significance of probability concept Unce neering systems. Properly ‘ainties are unavoidable in the design and planni therefore, the tools of engineering analysis should include methods and concepts foi ficance of uncertainty on system performance and design. In this the prineiples of probability (and its allied fields of statistics and decision theory) offer the mathematical basis for modeling uncertainty and the analysis of its effects on er Probability and statistical decision theory have especially significant roles inal aspects, of engineering planning and design, including: (1) the modeling of engineering problems and evaluation of systems performance under conditions of uncertainty jevelopment of design criteria, explicitly taking into account the significance of uncertaint and (3) the logical framework for quantitative risk assessment and risk-benefit tradeott 3s relative to decision making, Our principal aim is to emphasize the probability and statistical decision theory in engineering, with special attention on problems. related to construction and industrial management; geotechnical, structural, and mechanical .cean engineering; transportation planni immetric and geodetic and problems of phot The principal motivation for developing this revised edition of the book is our fiem belief thatthe principles of probability and statistics are of fundamental importance to all branches of engineering, although the examples and exercise problems inclu in this tex are mostly from civil and environmental engineering. These principles are essential for fun central in the modern approach to decision making under un the quantitative analysis and model stainties in the assessment of risk, which is The concepts and methods expounded inthis book constitute only the basics necessary {ment of uncertainties. Thes ized applications. See Volume Il of Ang and for the proper yy need to be supple nented with more adva Tang (1984) for some o ed tools for specia professional colleagues regarding the way we elucidated the concepts and methods in the jtst edition, particularly for those wishing to learn and apply the principles of probability andtion ofthis book has contributed statistics. In this regard, we are encouraged thatthe first tions of engineering students, and of professional colleagues, through self-studies. The work for this second edition of the book is also inspired by the he education of fi hope that this work will continue t0 contribute t ature generations of engineering students in the practic engincering, enhanced further nowadays by the general availability of personal computers VOLUME I! The first edition of this text was published in two volumes. For the second edition, on! Volume 1 (this text) is being revised. If you would Tike to Volume Il, you may contact Professor Ang directly at ahang, ACKNOWLEDGEMENTS Finally, itis our pleasure to acknowleds the many constructive comments and su offered by the prepublication reviewers of our original manuseript, including: C,H, Aikens, University of Tennessee B. Bhattacharya, University of Delawar \. Cariapa, Marquette University A. Der Kiureghian, University of California, Berkeley S. Ekwaro-Osite, Texas Tech University B. Ellingwood, Georgia Institute of Technology TS. Hale, Ohio University P. A, Johnson, Pennsylvania State University J. Lee, University of Louisiana M. Maes, University of Calgary S. Mattingly, University of Texas, Arlington P. O'Shaughnessy, The University of Iowa C. Polito, Valparaiso University J.R, Rowland, University of Kansas Y.K, Wen, University of Illinois, Urbana-Champaign as well as @ number of other anonymous reviewers. Many of their suggestions have served the many compliments from nuscript, We also greatly appreci several ofthe reviewers, including the phrase “the authors seem to understand what Socrates knew a Tong time ago... "Analytical tools that are understood have a higher probability 0 scing used’ "; relating our work to Socrates certainly represents the height of compliments. Last but not least, our thanks to T. Hu, H, Lam, J. Zhang and L. Zhang for their assistance in the solutions 10 some of the examples, and in the preparation of the Solutions Manual for the problems in the book.Contents Prefix Roles of Probability and Statistics in Engineering 11 Introduction 12. Uncertainty in Engineering 1 Uncertainty Associated with Randomness ncertanty Associated with Imperfect Knowledge—The Epistemic Uncertain ind Decision Making under Uncertain arming and Design of Transportation Infrastructures Jesign of Structures and Machines 13.3 Planning and Design of Hydrosystems 134 Design of Geotechnical Systems 5. Construction Pan J Management 36 Photogrammetic, Geodetic, and Surveyin “Measurement Applications in Quality Control and >CHAPTER 2 Fundamentals of Probability Models Events and Probabil 2.1 Characteristic Probabilities 2 Estimating Probabilities ents of Set Theory —Tools for Definin 1 Important Definitions Mathematical Operations of Se ematies of Probabi The Addition Ru Conditional Probability The Multiplication Rul he Theorem of Total Probability 235. The Bayes’ Theorem 24 Concluding Summat Problem PCHAPTER 2 Analytical Models of Random Phenomena Random Variables and Probability Distibuto Random Events and Random Variables Probability Distribution of a Random Variable 3.1.3. Main Descriptors of « Random Variable ful Probability Distributions The Gaussian (or Normal) Distt normal Distribution The Bernoulli Sequence and the Binomial Distribution The Geometric Distribtio The Negative Binomial Distribu The Poisson Process and the Poss Distribution The Exponential Distibution The Gamma Distributior The Hypergeometre Distribution 2.10 The Beta Distribtior 2.11. Other Useful Distributions Multiple Random Variables ind Conditional Probability Distributions and Correlation Functions of Random Variables 4.1 Introduction 42. Derived Probability Distribui 42.1. Function ofa Single Random Variabl 4.2.2 Function of Multiple Random Variables 1.2.3. Extreme Value Distributions “Moments of Functions of Random Vatiab 43.1 Mathematical Expectations of a Function 43.2 Mean and V Concluding Summary Problems References Computer-Based Numerical and in Probability ulation Methods Intrxtion Numerical and Simlations Methods 5.2.1 Essentials of Monte Carlo Simulation 5.22 Numerical Examplespistemic Uncertainties 223.83. Comelation Analy 5.2.4 MCS Involving Corelted Random 8.3.1 Estimation ofthe Correlation Coefficient Concluding Summa 24284 Lincar Regression with Nonconstant Variance robles 242° §.5 Multiple Lincar Regression 8.7 Applications of Regression Analysis PcHAPTER 6 Engin 333 Statistical Inferonces from Observational Data 88 Concluding Summary 339 Problems 339 6.1 Role of Statistical Inference in Engineering i 62. Statistical Estimation of Parameters aie =i 62.2 Sampling Distibution PCHAPTER 9 Testing of Hypothes The Bayesian Approach M6 6.3.1 Inroduetion 1 Induction M6 6.32. Hypothesis Test Procedure 91.1 Estimation of Parameters M6 onfidence Interval 2. Basic Concepts —The Discrete Cse M7 {64.1 Confidence Interval of the Mer 64 limecomlnos Go oo 642 Confidence Interval ofthe Proportion 93.1 General Formulation 382 _ 643. Confidence Interval of the Variance D2, A Soedal Applica tf de Bayes 5 Measurement Theory Seiee eee “a7 6 Concluding Summa 94 Bayesian Concept in Sampling Theo 360) Probl 94.1 General Formulation 360, Referens 9.4.2. Sampling fore Noeml Populations 360 94.3 Erorin Estimation 362 PCHAPTER 7 9.4.4 The Lily of Conjugate Distibutions 36 Determination of Probability Distribution Models 9.5. Estimation of Two Parameters 368 1 Introduct 2.6 Bayesian Regression and Coreation Analyses 372 2 Probability Papers 26.1 Linear Regression 372 721 Unity ad Ploting Posts 9.6.2 Updating the Regression Parameters 374 The Normal Probability Paper 2.63 Conlation Analys a5 ne Lognormal Pobsbility Pap 3819.7 Concluding Summ ° 7.2.4 Construction of General Probaili Problems 377 Papers 284 Reference x1 3 Testing Goodness-ofFit of Disibution Models 289 7 Square Test for Goodness-of-Fit 289 PCHAPTER 10 The Kolmogorov-Smimov (K-S)T Elements of Quality Assuran Goodness-of Fit 293 33 The Anderson-Darling Test for Goodness-of Fit 296 APPENDICES: A OdGrAAgs in il Atpepennle Forma of Eres Appendix A: Probability Tables 388 Distribution 300 Table A.1_ Standard Normal Probabilities aa $ Concluding Summ 301 Table A2 CDF of the Binomial Distribution a7 Problem 302 Table A.} Critical Values of r-Distibution a otere 305 Contin 2 Table A Ceitical Values ofthe x >CHAPTER & probability Level « 393 Regression and Correlation Anelyses 206 “Table AS Cilial Vales of Dr af Sipificmce Level 8.1 Introduction 306 the K-S Te ws 8.2 Fundamentals of Linear Regression Analysis 306 8.2.1 Regression with Constant Variance 306 8.2.2. Variance in Regression Analysi 308 “Arial onlin at he Wiley Web site wow wiley omveollepeanContents ili Table A.6 Critical Values ofthe Anderson-Das 4: The Multinomiat Cooficient 399 I sooriness 19s BS:St mula 399 Appendix B: Combinatorial Formulas 397 Appendix C: Derivation of the Poisson Distribution 400 BLL: The Basie Relation 07 ! 3: The Binomial Coefficien gos Index 403 a 41 Roles of Probability and Statistics in Engineering 1.1 INTRODUCTION In dealing w portant th2. Chapter I, Roles of Probability and Statistics in Engineern 1.2 UNCERTAINTY IN.ENGINEERING The presence of uncertainty in engineering, therefore, is clearly unavoidable; the available often incomplete or insufficient and invariably contain variability. Moreover, en models with unknown degrees of imperfections relative to reality, and thus involve addi i) uncertainty associated withthe randomness ofthe underlying phenomenon that is exhib: served information, and (ii) uncertainty associated with imperfect models of the real world because of insufficient or imperfect knowledge of reality. As we id earlier, these two {ypes of uncertainty may be called, respectively, the aleaiory un See Ang (1970, 2004) for a basic framework fo be combined and analyzed as a total uncertainty, or treated separately. In either case principles of probability and statistics apply equally We might point out that there a «10 view the signifi the wo formation, which is part of the real world (within our ability to observe and describe). Much of the aleatory uncertainty that civil engineers must deal with are inherent in na knowledge-based) uncertainty is associated with imperfect knowledge of the real world, and may be reduced through application of better prediction models andor improved exper iments. Second, the respective consequences of these two types of uncertainty may also be different—the ef of the aleatory randomness leads to a calculated probability oF Fisk, Whereas the effect ofthe epistemic type expresses an uncertainty in the estimated probability cor risk. In many application areas of engineering and the physical sciences, the uncertain (or error bounds) of a calculated risk or probability i as important asthe risk itself eg. the National Research Couneil (1994) has emphasized the importance of quantifying the un- certainty in the calculated risk, and a number of U.S. government agencies, such as the U.S. Department of Energy (1996), the Environmental Protection Agency (1997), NASA (2002: NIH (1994), as well as in the UK (2000), have applied this approach in the quantitati ment of risk. In some practical applications, however, the two types of uncertainty a combined and their aggregate effects estimated 2c Again irrespective of w the two types of uncertainties are combined or tre ately, the concepts and methods covered in the ensuing chapters are equally appli Finally, there should be no problem in delineating between the two types of uncertainty —the aleatory type is essentially databased, whereas the epistemic type is knowledge based. For practical purposes, the epistemic uncertainty may be Timited to the the prescribed form of probability distributions and in all the parameters. 1.21 Uncertainty Associated with Randomness—the Aleatory Uncertainty Many phenomena or processe must contend with contain randomness; that is, the expected outcomes are unpredictable (10 some degree Such phenomena are characterized by field or experimental data that contain P P apparently identical conditions. In other words, there1.2 Uncertainty in Engineering 3 ~ TABLE 1.1 Rainfall Intensity Data Recorded over a Period of 29 Years Year Rainfall Year Rainfall Year Rainfall No. Intensity, in. No. Intensity, in. No. Intensity, in. 1 43.30 11 54.49 21 58.71 2 53.02 12 47.38 22 42.96 3 63.52 13 40.78 23 SSF 4 45.93 14 45.05 24 41.31 5 48.26 15 50.37 25 58.83 6 50.51 16 54.91 26 48.21 7 49.57 17 51.28 27 44.67 8 43.93 18 39.91 28 67.72 9 46.77 19 53.29 29 43.11 10 59.12 20 67.59 this range certain values may occur more frequently than others. The variability inherent in such data or information is statistical in nature, and the realization of a specific value (or range of values) involves probability. The inherent variability in the observed or measured data can be portrayed graphically in the form of a histogram or frequency diagram, such as those shown in Figs. 1.1 through 1.23, all of which demonstrate information on physical phenomena of relevance particularly to civil and environmental engineering. Furthermore, if two variables are involved, the joint variability may similarly be portrayed ina scattergram. A histogram simply shows the relative frequencies of the different observed values of a single variable. For example, for a specific set of experimental data, the corresponding histogram may be constructed as follows. From the range of the observed data set, we may select a range on one axis (for a two-dimensional graph) that is sufficient to cover the largest and smallest values among the set of data, and divide this range in convenient intervals. The other axis can then represent the number of observations within each interval among the total number of observations, or the fraction of the total number. For example, consider the 29 years of annual cumulative rainfall intensity in a watershed area recorded over a period of 29 years as presented in Table 1.1. An examination of these data will reveal that the observed rainfall intensities range from 39.91 to 67.72 in. Therefore, choosing a uniform interval of 4 in. between 38 and 70 in. the number of observations within each interval and the corresponding fraction of the total observations are calculated as summarized in Table 1.2. The uniform intervals indicated in Table 1.2 may then be scaled on the abscissa, and the corresponding number of observations (column 2 in Table 1.2) can be shown as a bar TABLE 1.2 Number and Fraction of Total Observations in Each Interval eee No. of Fraction of Interval Observations Total Observations 38-42 3 0.1034 42-46 A 0.2415 46-50 5 0.1724 50-54 3 0.1724 54-58 3 0.1034 58-62 3 0.1034 62-66 1 0.0345 66-70 z 0.06904p Chapter 1. Roles of Probability and Statistics in Engineering = ze é ae 5 Total number of 2 30 g observations = 29 2 e @ oe Ber @ 20;- © gL 3 3 Z 3 4 2 10 S c £ 2b 2 5 3 Zo - £0 é 34 38 42 46 50 54 58 62 66 70 '* 34 38 42 46 50 54 58 62 66 70 Annual rainfall intensity, in. Annual rainfall intensity, in. (a) In number of observations (b) In fraction of total observations Figure 1.1 Histograms of Annual Rainfall Intensity. on the vertical axis, as illustrated in the histogram of Fig. |.1a for the rainfall intensity of the watershed area. Alternatively, the vertical bar may be in terms of the fraction of the total observations (column 3 in Table 1.2) and would appear as shown in Fig. 1.1b. Oftentimes, there may be reasons to compare an empirical frequency diagram, such as a histogram, with a theoretical frequency distribution (such as a probability density function, PDF, discussed later in Chapter 3). For this purpose, the area under the empirical frequency diagram must be equal to unity; we obtain this by dividing each of the ordinates in a histogram by its total area; e.g., we obtain the empirical frequency function of Fig. 1.1a by dividing each of the ordinates by 29 x 4=116; whereas the corresponding empirical frequency function may also be obtained from Fig. 1.1b by dividing each of the ordinates by 4 x | = 4. In either case, we would obtain the empirical frequency function of Fig. |.1c¢ for the rainfall intensity in 4 % per in. bo F a Oo I 1E 0 34 38 42 46 50 54 58 62 66 70 ~ Annual rainfall intensity, in. (c) Figure 1.1¢ Empirical Frequency Function. the watershed area. We may then observe that the total area under the empirical frequency function is equal to 1.0, and thus the area over a given range may be used to estimate the probability of rainfall intensity within the given range. A large number of physical phenomena are represented in Figs. 1.1 through 1.23; these are purposely collected here to demonstrate and emphasize the fact that the state of most engineering information contains significant variability. For examples, the properties of most materials of construction vary widely; in Figs. 1.2 and 1.3 we present the histograms demonstrating the variabilities in the bulk density of soils and the water-cement (w/c) ratio of concrete specimens, respectively, whereas in Figs. 1.4 and 1.5 are shown the yield strength of reinforcing bars and the ultimate shear strength of steel fillet welds.Number of results TBAT 17-18 18-419 19-20 20-21 21-22 29-23 23-24 Bulk density (KN/m?) Figure 1.2 Bulk density of residual soils (after Winn et al., 2001). 120 Average = 0.42 Standard deviation = 0.026 Minimum = 0.36 Maximum = 0.54 Count = 522 Number of sections p QD oo ° o oS o So I T T I ry iS) T ° xe Ey S ele A BD NID plo PoPPoP PM Mo MOON oF Ma PoP PoP oF oF wic determined by fluorescent microscopy Figure 1.3 Water—cement (w/c) ratio in concrete (after Thoft—Christensen, 2003). 0.40;— S iw 6 T 171 tests on 3/8 in. to 1-1/4 in. diameter bars Fraction of tests ° i Ss 32 40 48 56 64 72 Yield strength, ksi Figure 1.4 Yield strength of reinforcing bars (data from Julian, 1957).6 » Chapter 1. Roles of Probability and Statistics in Engineering 30;— 20;- Frequency, percent OL. 10 50 60 70 Ultimate shear strength, ksi Figure 1.5 Ultimate shear strength of fillet welds (after Kulak, 1972). Similarly, in the case of construction timber, we see in Fig. 1.6 examples of the his- tograms of the modulus of elasticity of Southern pine and Douglas fir timber, whereas in Fig. 1.7 we show the histogram of the modulus of elasticity of grouted masonry. As ex- pected, there are wide variabilities in the moduli of elasticity of these two materials; timber is a natural organic material, whereas masonry is a highly heterogeneous mixture of cement and natural sand. 4 Southern pine lam stock 30;— ID grade xe 20}- oe 2 10;— 5 8 oF ae _ = Douglas fir lam stock 2 30 LI grade = > oO ww 20}- 10}— ol. 0 4.0 14 18 22 26 30 34 Modulus of elasticity, psi x 10° Figure 1.6 Modulus of elasticity of construction lumber (after Galligan and Snodgrass, 1970). Of great interest in reinforced concrete construction is the possible corrosion of the reinforcing steel; in Fig. 1.8, we show the histogram of the corrosion activity (mea- sured by the electrical current density) of steel in concrete structures. In rock mechanics, homogeneous rock mass contains fissures. We see in Fig. 1.9 that the trace lengths of discontinuities in a rock mass can vary widely.60 a so £ ° Number of cases 8 s o 0 500 1000 1500 2000 Modulus of elasticity/maximum compressive strength Figure 1.7 Modulus of elasticity of grouted masonry (after Brandow et al., 1997). 50 40;- ray ° T ey 3 I Frequency [%] T 10}/— -10-8 6-4-2024 6 8 10 12 14 16 18 20 Macrocell current density [mA/m?] Figure 1.8 Corrosion activity of steel in concrete (after Pruckner & Gjorv, 2002). 60 — 20/— 6 Figure 1.9 Trace lengths of discontinuities in rock mass (after Wu and Wang, 2002). oO 0. 3.0 Trace length, m8 » Chapter 1. Roles of Probability and Statistics in Engineering Number of occurrences 10 12 14 16 18 20 22 24 26 28 30 32 Residual friction angle (degrees) Figure 1.10 Residual friction angle of mudstone (after Becker et al., 1998) In geotechnical engineering, we can expect significant variability in the information pertaining to soil properties. For example, Fig. 1.10 shows the histogram of the residual angle of friction of soft mudstone, whereas Fig. 1.11 shows the variability of the compressive strength of sandstone underlying the foundation of the Confederation Bridge in Canada. onto 0.08 — 0.06 0.04 Relative frequency 0.02 0.00 Unconfined compressive strength (MPa) Figure 1.11 Compressive strength of sandstone (after Becker et al., 1998). Furthermore, the efficiencies of pile groups are also highly variable as shown in Figs. 1.12a and 1.12b, respectively, for pile groups in clay and sand. Significant variabilities are also present in the loads on structures; these are illustrated in Fig. 1.13 showing the wind-induced pressure fluctuations on tall buildings observed during two typhoons (hurricanes), and in Fig. 1.14 is shown the variability of earthquake- induced shear stresses in soils. In both of these figures, the dispersions are scaled in terms of respective standard deviations.1.2 Uncertainty in Engineering « 9 9 3.5 BL Free standing, 3D spacing, 2 ‘jl 3.0 8 oL 28 cases in clay 2 6 Measured frequency 42.5 = 3 5L — Log-normal distribution _|o 9 6 47 415 5 3b 2 41.0 Bot Z4b os oLLt fff dt L266) 03 05 0.7 09 1.1 1.3 15 1.7 1.9 21 Pile group efficiency (a) clay 9 3.5 BE Cap-ground contact, 3D 8 aL spacing, 17 cases in sand ~]3-0 2 Measured frequency 42.5 = = 5- i 2 8 5- Log-normal distribution_jo9 © 8 UD = Zz o 4— 15 2 2 3-- o 2 aL W108 2 a Z4b 70.5 oUt ! |_| Joo 07 09 11°13 15 1.7 19 21 23 25° Pile group efficiency (b) sand Figure 1.12 Pile group efficiencies in clay and sand (after Zhang et al., 2001). In environmental engineering, the water quality depends on certain parameters such as the stream temperature and dissolved oxygen (DO) deficit. Figure 1.15 illustrates the variability of dissolved oxygen (DO) deficit in the Ohio River, whereas in Fig. 1.16 is shown the variability of the weekly maximum temperature of four streams in four different states of the United States of America—Maine, Washington, Minnesota, and Oklahoma. In the case of transportation engineering, we show in Fig. 1.17 the empirical frequency distribution of O-D (origin—destination) trip lengths in Sioux Falls, SD, and in Fig. 1.18 the histogram of the estimated impact speed of passenger car accidents. —T- TOT TTT 1 TTT TT TI 4-32-1012 3 4 432-101 2 3 4 Number of standard deviation Number of standard deviation (a) Typhoon Ruby 16/7/1970 7SW-3 (b) Typhoon Georgia 13/9/1970 7NE-4 Figure 1.13 Pressure fluctuations on tall buildings during typhoons (after Lam Put, 1971).10 » Chapter 1. Roles of Probability and Statistics in Engineering 2.0 270 Distribution of largest Mea 1.6 instantaneous peak values in 4216 3 0.6 ry | 1 1 2 value 9 2 Distribution of 2 § 1.2-all values (x3) | Psy 162 3 2 7}, 8 = 2 a 0.4};- | 8 0.8- / 108 © S a I - oO 3 f 8 = a i € @ 0.4 543 * 0.2)- 7 0 0 a 4 0 0 | | l | Stresses in standard deviations from the mean 1.5 2.0 2.5 3.0 (Taft earthquake) DO deficit, mg/liter Figure 1.14 Earthquake-induced shear stresses in soils (after Figure 1.15 Dissolved oxygen (DO) deficit in Donovan, 1972). river (after Kothandaraman & Ewing, 1969). In Fig. 1.19 we see the histograms of the measured roughness profiles of a rough road and a smooth road, in terms of the respective rms (root-mean-square) values. The cost of injuries from highway work zone accidents in the United States can be expected to vary greatly; this is evidenced in Fig. 1.20. It is of interest to observe that engineered structures can sometimes fail and cause economic losses as well as loss of human lives. Figure 1.21 shows the statistics of dam failures in the United States as a function of life (in years) after completion. Clearly, we observe that most failures occur during the first year after completion of construction of adam. 35, 35, 30+ 01054140, ME 30 14238000, WA Ros & 25 > > S 20;- S 20}- § 15) 3 15 ® 10;- 2 10;— * 5E 5- 0 : o. : oe e e © © ee a e we we ae a? Oo eg 2 ee Maximum stream temperatures (°C) Maximum stream temperatures (°C) = 05124480, MN 36 07303400, OK 30{- x — 30 , Za5- S25 > > 3 20- S20 $15 $15 s s 2 10;- & 10 Si 5 0 0 » e Maximum stream temperatures (°C) Maximum stream temperatures (°C) Figure 1.16 Histograms of weekly maximum stream temperature (after Mohseni et al., 2002).a a 14|—- Work trips-25,193 = Vehicle hours—4,017 12|- Avg. trip length-9.57 — Total work trips, % 0 LE did Pit Vid Ett tit it tit ty 0 2 4 6 8 10 12 14 16 18 20 22 Travel time, min. Figure 1.17 Origin—destination trip length frequency in Sioux Falls, SD (U.S. Dept of Comm. 1965). oo oo 56 rT. Frequency, % np o Co So TT 3 T 0 Oo 10 20 30 40 50 60 70 80 Vehicle speed, mph Figure 1.18 Impact speeds of passenger car accidents (after Viner, 1972). Typical smooth road 25}- Oa rms= 20 Harms = 40 20 Ma aus = 30 Ra aus = 0:100 Occurrence (%) a T 3° I ( ~ 100 RMS (mm/m?) 50 Typical rough road 250 Figure 1.19 Measured road roughness profiles (after Rouillard et al., 2000). 35% 32% 30% 25% 25% 7 20% “15% 10% 5% - 0% $0-$1 $1-$1,000 $1,000-$5,000 $5,000-$7,500 Cost of work zone accidents Figure 1.20 Cost of work zone accidents (after Mohan & Gautam, 2002). 1112 ® Chapter 1. Roles of Probability and Statistics in Engineering 50 45 40 |. 35 | 30 | 25;- 20;- Number of cases 15;- 10;- 1 2 3 4 5 6 ih 8 9 10 Years after completion Figure 1.21 Statistics of dam failures in the United States (after van Gelder, 2000). Finally, the variabilities of information in construction engineering and construction management are illustrated in Figs. 1.22 and 1.23, respectively. Specifically, we show in the figures the completion time of building houses in England and the variability of bid prices in highway construction. All sites 500 1000 1500 2000 Speculative development TTT) Average 1080 man hours rt I 1 cH Oo t 500 1000 1500 2000 Local authority traditional dg To Average 1200 man hours I | T 500 1000 1500 2000 Local authority system Average 1070 man hours A 110 O 500 1000 1500 2000 Figure 1.22 Histograms of completion times in house building (after Forbes, 1969) In some of the figures, e.g., Figs. 1.11, 1.12, 1.14, 1.19, 1.21, and 1.23, the theoretical probability density functions (PDFs) are also shown; the significance of these theoretical functions and their relations to the corresponding experimental frequency diagrams will be discussed in greater depth in Chapters 3 and 7.1.2 Uncertainty in Engineering «@ 13 - Total no. of observations = 56 12h 10 % 2+ se 8 = oO D g °o 3 Ss = Ns 0 hid tid bs Pf) 0 0.975 1.075 14175 1.275 1.375 1.475 Bid ratio Figure 1.23 Distribution of bid prices in highway construction (after Cox, 1969). When two (or more) variables are involved, each variable may have its own variability, whereas there may also be joint variability of the two variables. Observed data of pairs of values of the two variables can be portrayed in a two-dimensional graph in the form of a scattergram of the observed data points. For example, in Fig. 1.24 is shown the scattergram of the modulus of elasticity versus the corresponding strength of Douglas fir timber, whereas, in Fig. 1.25 we see the scattergram of the tensile strength of concrete versus temperature. In Fig. 1.26 we observe the scatter of the data points of the mean annual discharge of a stream versus the corresponding drainage area near Honolulu. In Fig. 1.27 is shown the scattergram of the plasticity index versus the liquid limit of soils, which is of fundamental interest to geotechnical engineers. N a ° a o |_ MOE = 13,171.50 S.R. + 839,000 65 2° a Modulus of elasticity, 10° psi 0 20 40 60 80 100 Strength ratio, % Figure 1.24 Modulus of elasticity ys. strength of timber (after Littleford, 1967). In Fig. 1.28, we show the scattergram between the concentration of chlorophyll and phosphorous concentration; this information is important to environmental engineers con- cerned with the productivity of lakes. In Fig. 1.29 we see a typical scattergram of real estate land value plotted against population density.14 ® Chapter 1. Roles of Probability and Statistics in Engineering 2.8 26 24e Mean value = 2.4 MPa e ° R=0.791 ° ee 22 at 2.0 18 ig 1.6 fe o- ° 14Gee$ if 1.2 + 1.0 200 250 300 350 400 450 500 550 Temperature, °C Decrease in tensile strength below the mean value: MPa Figure 1.25 Tensile strength of concrete vs. temperature (after dos Santos et al., 2002). 14 12 So T © T Mean discharge, cfs 0 ! l | | | 0 a 2 3 4 5 6 Drainage area, sq. mile Figure 1.26 Mean annual discharge vs. drainage area of streams (after Todd & Meyer, 1971). 70 60+ ; gS 50 a x 40/;- 3 £ 2 30- * Woodlands 2 1 Serangoon rd & 4 Senoko rd a 20/- x Grange rd f gee = U Line 10 ;- ts —ALine fe 0 Ze 0 20 40 60 80 100 120 Liquid limit, LL (%) Figure 1.27 Plasticity index vs. liquid limit of solids (after Winn et al., 2001).1.2 Uncertainty in Engineering @ 15 1000 Log chlorophyll = ~1.09 + 1.48 log total P r=0.95 1000 100;— 100 Chlorophyll a (mg/m3) 6 T Land value, $1000 per acre (1970 dollars) 3 0.1 0.1 ! i 1 10 100 1000 0.4 1 10 100 Measured total phosphorus (mg/m) Population per acre Figure 1.28 Chlorophyll vs. phosphorous concentrations Figure 1.29 Land value vs. population density (after (after Jones & Bachmann, 1976). Wynn, 1969). In estimating the maximum wind speeds, the calculated speed may not be perfect; this is illustrated in Fig. 1.30, which shows the scattergram between the calculated and corresponding observed maximum wind speeds. Also, in traffic engineering, the relationship between the daily conflicts and the total entry volume at traffic intersections is illustrated in Fig. 1.31, which shows the scatter of the data points. Finally, in assessing the hazards from glacier lake outbursts, which can potentially cause dangerous outburst floods and debris flow in mountainous regions, the area and mean depth of a glacier lake can be used to estimate the volume of a lake. For this purpose, the Scattergram of Fig. 1.32 shows sucha relationship between the mean depth and the area. o S [ Ug>20m/s 2- S 20 E 2000 @ £ 2 4 = oO EE = [ ~@ 15; = 1500 8 oF ca > & Ss % v | 10 8 1000 ad © 5 3 500 }— y = 0.0713x — 76.188 2 R? = 0.9047 0 0 | | 0 5 10 15 20 25 30 0 10000 20000 30000 Ucac(m/s) Total entry volume (per day) Figure 1.30 Calculated vs. observed wind speeds Figure 1.31 Traffic conflicts vs. volume (after (after Matsui et al., 2002), Katamine, 2000).16 ® Chapter 1. Roles of Probability and Statistics in Engineering > EXAMPLE 1.1 100 Ts eo a = Q o + & = e = § q 8 = a tLe 10° 104 10° 10° 107 Area (m*) Figure 1.32 Mean depth vs. area of glacier lakes (after Huggel et al., 2002). A large number of histograms and scattergrams are shown above; the purpose of il- lustrating such a large number is to demonstrate clearly that variability of engineering information is invariably present and unavoidable in many areas of engineering applica- tions. We might emphasize that the variability exhibited in any histogram is due to randomness in nature, and thus is an aleatory type of uncertainty. The following example may serve to introduce how such information may be handled in practice. Quite often, when we have a finite set of observational data (called a sample), itis of interest to estimate the average of the sample (called the sample mean) and a measure of its variability or dispersion (called the sample variance); the latter is the aleatory uncertainty corresponding to the data set. Consider, for example, the set of 29 observed annual rainfall intensities in the watershed area tabulated earlier in Table 1. 1a. Clearly, we can estimate the sample average of the observations simply as follows: a 9 (43.30 + 53.02 + - + - + 67.72 + 43.11) = 50.70 in. Xe and the corresponding sample variance, which is the average of the squared deviation from the mean (we shall define the sample variance more thoroughly in Chapter 6), is = | 43.30 — 50.70)? + (53.02 — 50.70)? +--+ + (67.72 — 50.70)? + (43.11 — 50.70)] ele, = 57.34 Finally we obtain the corresponding sample standard deviation, 8x = 57.34 = 7.57 in. This sample standard deviation is, therefore, a measure of the dispersion of the annual rainfall intensity and represents the corresponding randomness or aleatory uncertainty of the rainfall intensity in the watershed area. As the above average annual rainfall intensity is estimated from the set of 29 observations, there is also epistemic uncertainty underlying the estimated average, called the sampling error. In this case, this is the uncertainty underlying the estimated average annual rainfall intensity of 50.70 in. This sampling error (defined in Chapter 6 as a function of the sample size) is equal to 7.57/./29 = 1.41 in.1.2 Uncertainty in Engineering “@ 17 In this example, we see that the variability’ (or randomness) in the observed data contains the aleatory uncertainty, whereas the sampling error in the estimated average contributes to the epistemic uncertainty, There are other sources of epistemic uncertainty as illustrated below in Examples 1.2 and 1.3. < 1.2.2 Uncertainty Associated with Imperfect Knowledge—the Epistemic Uncertainty > EXAMPLE 1.2 In engineering, we have to rely on idealized models of the real world in our analysis and predictions for the purpose of making decisions or for planning and developing criteria for the design of an engineering system. These idealized models, which may be mathemat- ical or simulation models (e.g., mathematical formulas, equations, numerical algorithms, computer programs) or even laboratory models, are imperfect representations of the real world. Consequently, the results of analysis, estimations, or predictions obtained on the basis of such models are inaccurate (with some unknown degree of error) and thus also contain uncertainty. Such uncertainties are, therefore, knowledge based and are of the epis- temic type. Quite often, this epistemic uncertainty may be more significant than the aleatory uncertainty. In performing a prediction or estimation with an idealized model, the objective is invariably to obtain a specific quantity of interest; this may be the mean-value or median value of a variable. Therefore, in considering the epistemic uncertainty it is reasonable (in practice) to limit our consideration to the inaccuracy in calculating or estimating the central value, such as the mean-value or median value. We might emphasize that we can expect errors also in the estimation of the other parameters as well as in the specification of the distribution, and thus there are also epistemic uncertainties in these latter estimations; however, such uncertainties are of second-order importance relative to that in the central value. For examples, consider the problems in the following illustrations. Consider the calculation of the deflection of a prismatic cantilever beam under the concentrated load Pas shown in Fig. E1.2. For engineering purposes, the deflection at the end of the beam B is usually calculated on the basis of the simple beam theory, which gives As= ae (1) as ; in which E = the modulus of elasticity of the material, and 7 = the moment of inertia of the beam cross section. ; se Y B A L Figure E1.2 A cantilever beam subjected to concentrated load P.18 » Chapter 1. Roles of Probability and Statistics in Engineering b> EXAMPLE 1.3 Equation 1.1 is based on several idealized assumptions, which are as follows: 1. The material is linearly elastic. 2. Under the load P, plane sections of the beam remain plane. 3. The support of the beam at A is perfectly rigid. In reality, each of the above assumptions may not be totally valid. For example, depending on the material of the beam and the magnitude of the load, the behavior may not be linearly elastic. Also, for high loads, plane sections of the beam will not remain plane; finally, it is seldom that the support at A can be perfectly rigid. Therefore, the calculation of the deflection, Ag, with Eq. 1.1 will involve some undetermined error and thus some uncertainty. Unless there is reason to believe otherwise, it is reasonable to assume that the calculated deflection is the mean deflection of the cantilever beam, implying that the error of Eq. 1.1 is symmetric about the average deflection. However, if necessary, the result may be corrected for any bias in the calculation. One way to evaluate or assess the error of Eq. 1.1 is to test a number of similar cantilever beams of the same material under the same conditions, and with precisely measured values of P, L, E, and /. The results of the tests should provide us with a basis for evaluating the error underlying the equation. For illustration, suppose (hypothetically) that 10 wood beams were tested and the results (in terms of the ratios of the measured to the calculated deflections) were as follows: measured As _ 1 95..0.95; 1.10; 0.983 1.15;0.97; 1.20; 1.00; 1.08; 1.12 Ag by Eq. 1.1 These test results would yield the following sample mean and sample standard deviation (see Chapter 6) for the ratio of the measured to the calculated deflection, Ap: 1.06 Standard deviation of ratio = 0.25 I Mean (or average) ratio According to the above test results, the deflection of the beam calculated with Eq. 1.1 tends to underestimate the correct deflection and, therefore, needs to be corrected by the bias factor of 1.06, whereas its coefficient of variation (defined as the standard deviation divided by the mean) representing the epistemic uncertainty is 0.25/1.06 = 0.24. Depending on the material, there may also be significant variability in the EJ of the beam. For example, the variability of the modulus of elasticity, E, of construction lumber is illustrated in Fig. 1.6. This variability in E7 will also lead to aleatory uncertainty in the calculated end deflection Ag. Finally, if the E/ of the beam is estimated from sampled observations, there will also be sampling error in the estimated mean £7, and this will add further epistemic uncertainty to the calculated deflection of the beam. < In Table E1.3, we show the data of the observed settlements of pile groups and the correspond- ing calculated settlements of the same pile groups reported by Viggiani (2001). The calculated settlements were obtained with a nonlinear model for predicting settlements of the respective pile groups. Using the data in Table E1.3, we develop later, in Example 8.3 of Chapter 8, the so-called linear regression equation showing the relationship between the observed and calculated settlements as follows: E(Y|x) = 0.038 + 1.064.x mm (1.2) in which E(Y|x) stands for the expected observed settlement Y of a pile group if the calculated settlement is x. In Example 8.3, we also show that the conditional standard deviation about the regression equation (i.e., the average dispersion along the regression line) is Syjy = 7.784 mm.1.3 Design and Decision Making Under Uncertainty @ 19 TABLE E1.3 Data of Observed and Calculated Settlements of Pile Groups (in mm) Obs. Calc. Obs. Cale. Obs. Cale. S'mt., yi S’mt., x; S’mt., yi S'mt., x S’mt., yi S’mt., x; 0.7 48 28.1 26.7 127 14.2 64 25.1 0.6 0.6 46 42.1 5 3.8 29.2. 31.5 3.8 1.58 25 26.4 32 31 49 5 225 27.8 5.4 Bed. 11.7 11.8 3.8 3.5 3.6 4 1.83 3.39 5.9 6.5 35,9 25.9, 943 3.24 38.1 2.6 11.6 11.3 6.6 43 185 174 Source: After Viggiani, 2001. Equation 1.2 has the following significance: The equation can be used to determine the mean-value (average) of the actual settlement, E(Y|x), if the calculated settlement is x. However, the conditional standard deviation of sy. =7.784 mm represents the error of the nonlinear model for predicting the actual settlements, and thus is the epistemic uncertainty of the proposed calculational method. The above Eq. 1.2 may be used to determine the expected settlements of similar pile groups at a site based on the settlements calculated by the same nonlinear model described in Viggiani (2001); however, according to the regression equation, the calculated settlements will tend to be on the low side and must be increased by a bias factor of 1.064. Moreover, there will be an average standard deviation (dispersion) of the true settlement of 7.784 mm for a given value of x. Depending on the value of the calculated settlement x, the corresponding coefficient of variation (c.o.v.) may also be estimated; i.e., the c.o.v. would be 7.784/E(Y|x). For example, if the calculated settlement for a particular pile group is x=45 mm, we can expect the actual settlement to be 45 x 1.064 = 47.88 mm with a c.0.v. of 7.784/47.88 = 0.16, which is the epistemic uncertainty underlying the model equation. < Figure 1.30 provides a similar example for assessing the epistemic uncertainty of a predictive model. All uncertainties, whether they are aleatoric or epistemic, can be assessed in statistical terms, and the evaluation of their significance on engineering planning and design can be performed systematically and logically using the concepts and methods that are embodied in the theory of probability. » 13 DESIGN AND DECISION MAKING UNDER UNCERTAINTY As we indicated earlier, engineering information is generally of the type illustrated in Figs. 1.1 through 1.32, in which no single observation or measurement is representative; and any evaluation or prediction must be performed on the basis of imperfect models of the real world. That is, uncertainty (aleatory and/or epistemic) is unavoidable in engineering. Under the preceding situation, how should engineering designs be formulated or deci- sions affecting a design and planning be determined? Presumably, we may assume consis- tently worst conditions (e.g., specify the highest possible flood, smallest observed fatigue life of materials) and develop conservative designs on this basis. From the standpoint of system performance and safety, this approach may be suitable, and indeed has been the basis for much of engineering designs and planning in the past and can be expected to continue in the future. However, this approach eschews any information on risk and lacks20 > Chapter 1. Roles of Probability and Statistics in Engineering a systematic basis for evaluating the degree of conservativeness; a resulting desi overly conservative may be excessively costly, whereas one with insufficient cor tay be inexpensive but will sacrifice performance or safety. The optimal decision ought to be based on a trade-off between cost and benefit, in order to achieve a balance between cos and system performance. AS the available information and evaluative models are invari ably imperfect or insufficient, and thas contain uncertainties, the required trade-off analysis, ought to be performed within the context of probability and risk Te situations described above are common to many problems in engineering; in the following we describe several examples illustrating some of these problems. The examples are idealized to simplify the presentations; nevertheless, they serve to illustrate the essence of the decision-making aspects of engineering under conditions of uncertainty. 1.3.1. Planning and Design of Transportation Infrastructures In planning the transportation system of a city, there are example, in the a bridge across a river, more than one type of bridge system may be feasible for the crossing. The cost for the design and construction of each type of system will contain uncertainty; moreover, depending on the political situation, the ability of the city to raise the level of available funding may also be uncertain. Consequently, the selection of the type of bridge may be based on the probability ofthe cost of a given system relative to the probabilities of realistic levels of funding In the design of pavements, such as a roadway pavement or airport pavement, one of the principal decision variubles (among several others) is the thickness of a pavement system consisting of several layers of subgrade base material and the finished pavement. In general, the usable life of a pavement will depend on the thickness of the system; ie the thicker the pavement system, the longer its useful life. Of course, forthe same material and workmanship quality, the cost will also increase with the thickness. A thin system will cost less initially, but the subsequent maintenance and replacement costs will be higher Therefore, the optimal thickness ofthe pavement system may be determined on the basis of a trade-off between high initial cost with low maintenance, versus low initial cost but high maintenance and replacement costs over the useful life of the pavement. For the purpose Of such a trade-off analysis, the relation between the life of a pavement system and its thickness is required. The pavement life, however, is a function also of other variables, including drainage condition, temperature ranges, density, and degree of compaction of the subgrade. Since all these factors are random and contain variability, as depicted in Fig. 1.33 for the compacted subgrade density the life of the pavement fora given thickne cannot be predicted with complete certainty, Hence, the total cost (including initial and ‘maintenance costs) associated with a given pavement thickness also cannot be estimated ‘with complete confidence; any meaningful trade-olT analysis, therefore, should proper include probability and statistical concepts, 132. Design of Structures and Machines The strengths of structural material and components are random and thus contain variabil tics as illustrated in Figs. 14 and 1.5 for concrete and steel; therefore, the calculation ofthe capacity of a structure (invariably based on an idealized model) will contain both aleatory and epistemic uncertainties. On the other hand, the applied loads on a structure are also invariably random, as illustrated in Fig, 1.13 for wind pressures on tall buildings. Even for this idealized situation, the design of a structure, ic. involving the determination of the load carrying capacity of the structure, must consider the question of “how safe is safe113 Design and Decision Making Under Uncertainty ¢ 21 21 sample a 85 80 «9§ «100 «105 «110 «+115 Maximum laboratory density, % Figure 133. Density of compacted vocanie tf subgrade (after Pettit ‘enough?"—a question that realistically requires the consideration of risk and the probability (of nonperformance or failure As a specific example, consider the design of an offshore drilling platform, which is subject to occasional hurricane forces. In such a case, we recognize that aside from the fact that the maximum wind and wave effects during a hurricane are random, as may be inferred om Fig. 1.13, the occurrence frequency of hurricanes in a given region of the ocean is also unpredictable. Hence, in determining the safety level for the design of the platform the probability of occurrence of strong hurricanes within the specified useful life of the structure must be considered, in addition to the survival probability of the structure under the highest wind and wave forces expected during the life of the structure. Consequently the level of hurricane force that should be specified in the design, and the required leve of protection that would be adequate during a hurricane, are decisions that may require a trade-off between cost and level of protection in terms of risk or failure probability within the lifetime of the platform, Similarly, in considering the design of structural or machine components that are sub- jected to repeated or cyclic loads, we recognize thatthe fatigue life (in number of load cycles ‘until fatigue failure) of a component is also highly variable, even under constant amplitude stress eycles, as illustrated in Fig. 1.34. For this reason, the failure-free life of a component s difficult to predict and may be described only in terms of probability. Therefore, such a component may be designed for a required operational life within a specified reliability (probability of no fatigue failure). As expected, the fatigue life is a function of the applied stress amplitudes; in general, the life will increase inversely with the applied stress ampli tudes, as we can see in Fig. 1,34, Consequently, if a desired failure-free operational life is specified, a component may be designed to be massive so that the stress amplitudes will be low and thus ensure a high reliability of achieving the desired life; the resulting design will, of course, require more material and higher expense. In contrast, ifthe component is, under-designed, high stresses will be induced resulting in shorter life and more frequer maintenance or replacement in order to maintain the required reliability within the opera ‘onal life, In this ease, the optimal operational life may be determined by minimizing the ‘otal expected life-cycle cost of a component, which would include the initial cost of the ‘component, the expected costs of maintenance and replacements (which are functions o22 > Chapter 1 Roles of Probabiity and Statistics in Engineering 0870-3, Figure 1.4 Fatigue life of welded b tr Fisher etal, the specified reliability), and the expected cost associated with the loss of revenue incurred during a repair (which is also a function of reliability). Having decided on the operational ni esi ‘and a stated reliability, the component may then be proportioned or designed 133. Planning and Design of Hydrosystems agricultural farm from flooding requires the construc -am, A decision on the arly, this will depend on the high st Suppose that the protection ofa large tion of a major culvert at the junction of a roadway and a st (ie., flow capacity) of the culvert is required; cle flow, which is a function of the rainfall intensity in the watershed area and the associate runoff. As we have observed in Fig. 1.1, the annual rainfall intensity is highly variable and the estimation of the stream runoff can be expected to be imperfect, Consequently. the pre: diction of the maximum stream flow over a period of concern (e-g., 10 years) will contain uncertainty, both aleatory and epistemic. Assuming thatthe low capacity of a given culvert size can be determined accurately, the size of the culvert would depend on the acceptable probability of flooding within Clearly, ifthe culvert i large enough to handle the would be no danger to flood prohibitively high, and even during the heaviest rainfall the culvert may be used only to a fraction of its capacity. That is, over-design would be rand, if the culvert is too small, its cost may be low, but the farm is likely to be subject to dama veriod of concern, ream flow, ther st possible however, the cost of constructing the culvert may be wasteful and costly, On the other serious flooding every time there is a heavy rainfall, causin c to crops and erosion the upstream soil: i., under-design can incur hi Therefore, the optimal size of the culvert may be determined so that the total expecte life-cycle cost of the project, consisting of the cost of constructing the culvert plus the and erosion, is minimized, As the expected loss is mination of the total expected life-cycle ted life-cyele loss from floodin ‘function of the probability of flooding, theDecision Making Under Uncertainty < 23 134 Design of Geotechnical Systems Properties of soil materials are inherently heterogeneous and highly variable, and natura carth deposits are characterized by irregular layers of various material (e.., clay, silt, sand gravel, or combinations thercof) with wide ranges of density, moisture content, and other Soil properties that affect the strength and compressibility of the deposit. Similarly, rock mations are often characterized by irregular systems of geologic faults and fissures th significantly affect the load-bearing capacity of the rock (see F esignin foundations on soils oF rocks to support structures or facilities, the capacity of the in situ subsoil and/or rock deposit must be determined. Invariably, this determination has t0 bx developed on the basis of available geologic information and data from site exploration with limited soil sampling results or rack core samples. Because of the natural heterogeneity and ieregularty of soil and rock deposits, the ity of the subsoil could vary widely (see Figs. 1.10 and 1.11) over a foundation site moreover, because the required load-carrying capacity ofa foundation or pile group capacity 1 in Fig. 1.12) must invariably be estimated on the basis of very limited in-situ such estimates are subject to considerable epistemic uncertainty. AS a consequence, an imate may involve some tisk of overestimating the actual capacity of the deposit at asi in this light, the safety of a foundation designed on the basis of such estimates may not be censured with much confidence unless a sufficient margin of safety is provided in the design. (On the other hand, an excessively large safety margin may yield an unnecessarily costly apport system. Therefore, the required optimal safety margin for design may be viewed as roblem involving the trade-off between cost and tolerable probability of failure Trade-off considerations could be extended to the planning of site exploration and soil testing programs. Obviously, more extensive site exploration and use of sophisticated testing programs will reduce the uncertainties ofthe site condition and in the estimation of the soil parameters forthe sit. At a certain point, however, the incremental benefit obtained from further exploration and testing will not yield sufficient increase inthe reliability ofthe 135 Construction Planning and Management Many factors in the planning and management of construction projects are subject to high variability and uncertainty, and may not be easily controlled. For example, the required durations of various activites in a construction project will depend on the availabilty of resources, including labor and equipment and their respective productivities, on weather conditions, and on availability of construction materials. As none of these factors is com etely predictable, the durations ofthe individual activities in an activity network as well as he project duration cannot be estimated with much precision or certainty (see, for exampl Fig. 1.22) and thus may have to be described as random variables Therefore, in preparing a bid for a project, if conservative or pessimistic completion me estimates are assumed, the bid price may be too high, thus reducing the chance of winning the bid, On the other hand, if the bid is prepared on the basis ptimistic estimate of the project duration, the contractor may lose money or sacrifice profit in a successful bid, What of conservativeness should the contractor exercise in order to maximize his orher potential profit? Realistically, because of variabilites and uncertainties in a number Fs, the answer may be based on a consideration of probability: ie. the bid price may be by sroject duration corresponding to a specified probability of24 > Chapter 1, Roles of Probability and in Engineering 1.36 Photogrammetric, Geodetic, and Surveying Measurements All practical engineering measurements are subject to errors, which can be classified as indom and systematic errors. Systematic errors may be eliminated or minimized by eval ating them and applying appropriate corrections. However, the magnitude and propagation ff random errors, inherent in making measurements, may be determined and analyzed properly on the basis of statistical methods. Such a probability-based statistical approach is the only reliable means for evaluating th instrament capabilities, The statistical basis for these purposes is presented in Chapter 6. 1.3.7 Applications in Quality Control and Assurance Inordertoensure some minimum level of quality, or performance, of engineering products or 5 {00 stringent, it may unnecessarily inc ‘or problems in compliance, and product ca enforcement may be difficult to achieve; on the other hand, if the standard is to0 Tax, the quality of the finished product may be overly compromised, Also, if the control variables or design variables are random, what constitutes a stringent or nonstringent standard is not immediately clear; in these cases, the standard of acceptance may be developed on the basis, of probability considerations. For example, in constructing anearth embankment, practical standards for acceptability Of the compaction need to recognize the variability in the density of compacted soil, as ‘on probability considerations and taking into account the inherent variability of compacted earth material To control the quality of stream water of the parameters commonly used as a measure of pollution is the concentration of dissolved oxygen (DO) in the stream water which is also highly variable, as illustrated in Fig. 1.15. Environmental engineers have nized the need for probabilistic standards for controlling pollution of streams; fo example, Loucks and Lynn (1966) proposed the following probabilistic stream standard To ensure the quality of concrete material in reinforced concrete construction, the building code of the American Concrete Institute (ACI 318.93) requires the followin, test result falls below the required f.” by more than 500 psi. Each strength test result be the average of two cylinders from the same sample tesed at 28 days or the specified cali The requirements stated above clearly imply the need for probability and statistics, n the quality assurance of conerete material, Similar requirements may be found for the1.4 CONCLUDING REFERENCES Ang, A: HLS, “Extended 8 AIAA. Jul 5. and DeLeon D. Referenco SUMMARY this opening chapter, we emphasized the roles and importance of probability and statisti in decision making for engineering planning and design. It is essential that these roles probability concepts be recognized in order to appreciate thei true significance, In particu ‘ig important to stress that the description of statistical information and the estimation ¢ istics, such as the means and variances, are not the only applications of probability theory. Indeed, the much more significant role of probability concepts in engineering lies n providing a unified and logical framework for the quantitative analysis of uncertainty and in the assessment of its significance in system performance, and in the formulation of isk-based trade-off studies relative to decision making, planning, and design. The examples enumerated and described in Sections 1.2 and 1.3 should s strate the pervasiveness of probability concepts in engineering planning and design, many examples presented also serve to illustrat, with real data and realistic engineer ing problems, that randomness of real-world phenomena and imperfections of engineering mnodels are facts of life, Consequently, uncertainties are unavoidable. The uncertainties ass ciated with randomness are inherent variability, called aleatory uncertainty, whereas those underlying imperfect models of reality are knowledge-based, called epistemic uncertaint Even though both types of uncertainty may be represented in statistical terms, their respec fe significances may be different. Also, as a predictive model is improved the epistemic uncertainty can be reduced; on the other and, because the inherent variability is part o nature, the aleatory uncertainty may not be reducible, Finally, it is important to correct any misconception that extensive data are required to apply probability concepts; in Fact, the usefulness and relevance of such concepts are equall significant, irrespective of the sufficiency of data or quality of information. In particular. the amount of data affects the definition of the variability (ie., aleatory uncertainty) of information; however, epistemic uncertaintis associated with imperfect models of reali generally require judgmental assessments irrespective ofthe amount of data. Probability and tistics are the conceptual and theoretical bases for modeling and analyzing uncertainty The sufficiency of data and quality of information will affect the degree of uncertainty, but should not lessen the usefulness of probability as the proper tool for the analysis of such uncertainty and for the evaluation ofits effects on engineering performance and design. In the ensuing chapters, the essential and fundamental concepts and methods of probability and statistics for the above purposes are systematically developed and expounded. ind Mi KH. Hin, MLA. and MeNames, M, “Effects iments on the Fatigue Strength of Stes! Beams” NCHRP26 > Chaptar 1. Roles of Probability and Statistics in Engineerin Forhes, WS. "A Survey of Progress it House Builing.” Bul Management Nl 14), Xp 1969, pp. 88-9) Callgan, WL, and Snodgrass, DLV, “Machine Sess Rated Lambe uggs C. Kanb, A. Hache, W, Teyseire, P and Pal F,“Remose Joncs, LR. and Bachmann, RW, “Prediction of Phosphorous and ‘Chlorophyll Levels Lakes" Journal ofthe Water a Julian, OG. “Synopsis of Fist Px smite on Katumine, NM, “Various Volume Definitions with Con Umsignalired'Itersctons” ASCE Journal of Thansporation Kothandaraman. V.. and Ewing. B.B., “A Probabilistic Analysis Dis nical Onygen Demand Relationship Kulak, GL. “Sutsical Aspects of Strength of Connection,” Proc Lan Put R. “Dynan Response of Tal Billing to Random Wind ues Tokyo, Sepember 1971 Litton LW." Compansn of Flxaral Siengt-Stifass Relat hips for Clear Wood and Sitral Grades of Lumber.” Report VP-X-30 Forest Prt Lab B.C. Canada, Devember Stream Quai.” Water Resources Research, Vo. 2, No 3, Septem Mats, My shar, 1, and Hib, K, Ditional Characteristics of Vol 50, Elsevier Science, Ld, 200 Mohan, Sf. and Gant, P, “Cast of Highway Work Zone Injurcs, Motseni, 0, Erikson, TR, and Stefan, HG. “Uppe Seam Temperatures in the Contiguovs Unite States," Jounal of ‘sent Procedutes Guide for NASA Manager and Pracitons, National Research Coun, "Science sed Judgment in Risk Asses Peat, THLD.- "Statistical Analysis of Dens Test” Jounal Hi Dix. ASCE, Vl, HW2, Novent Pract, F, and Gjrr,O-E, Patch Repir and Macrocell Activity i ‘Concrete Sites" ACE Materials Journal, Va, 99, March-Ape Rouilan,V,*Clasfcation of Rou Surfae Profiles,” Jornal Tr pp as. ‘holt Christensen, “Stochastic Modstng ofthe Diffusion Cot Todd, DK.-and Meyer, CF, “Hydeology and Geology ofthe Hono Ate" Jour Hydradics Di, ASCE, Wl Febrary 1971 United Kingdom Health and Safety Executive (HSE), "Use of Ri Gelder, HAJ, “Statistical Mets forthe Risk-Based Desig Viggini ©. Analysis and Design of Pied Foundations.” Rivist ho Viner. .6. "Recent Developments in Roadside Crash Cushions.” Jour Wi, K., Rahunijo, Hand Peng, SC. “Charcerzation of Resid Wo, F-Q. and Wang. $4, "Stasi Model for Stace of Jointed Rock Mass” Geotechnigue Vol 52, 20: ‘ang, Lang, W Ht and Np. CW, "Relslity of Axilly Lowe Diiven Pile Group” ASCE Journal of Geote ™ naincering, Vo. 127 (12), December 200Fundamentals of Probability Models 2.1 EVENTS AND PROBABILITY 21.1 Characteristics of Problems Involving Probabilities Clearly, on the basis of the discussions in Chapter 1, when we speak of probability, we are referring to the likelihood of occurrence of an event relative to other events; in other ‘words, there is (implicitly atleast) more than one possibility, because otherwise the problem would be deterministic, For quantitative purposes, therefore, probability can be considered ‘a numerical measure ofthe likelihood of occurrence of an event within an exhaustive set of all possible alternative event Accordingly, the first requirement in the formulation of a probabilistic problem is the identification ofthe set ofall possibilities (the possibiliy space) and the event of interes Probabilities are then associated with specific events within a particular possibility space It is important to emphasize this last point, as probabilities are meaningful only within a given possibility space. To illustrate the various aspects of a probabilistic problem, characterized above, consider the following engineering-oriented problems. > EXAMPLE 21 A contractor is planning the acquisition of construction equipment, including bulldozers, needed fo anew p remote area. Supy from his priot experience with similar bulldozers, he estimated that there is a 50% chance that each bulldozer can remain operational fr atleast 6 months If be purchased three bulldozers forthe new project, what is the probability that there will be only 1 bulldozer left operational ater 6 months into the project? First, we observe that athe end of 6 months, the possible number of operating bulldozers will be 0,12, 07 3;therefor, this set of aumbers constitute the possiblity space ofthe numberof al bulldozers after 6 months. However, this possibilty space isnot pertinent to the question re above, Fortis latter purpose, the possibility space must be derived from the possible statu of each bulldozer after 6 months, as follows: Denoting the cordition of each bulldozer after 6 months as 0 ral and N for nonope (000—all three bulldozers are operational (OON—first and second bulldozers are operational, whereas the tind one is nonoperational NNN-—all three bulldozers are nonoperational Noo28 > Chapter 2. Fundame > EXAMPLE 22 tals of Probability Models NNO oNo NON sof the eight possible outcomes as indicated au ‘or nonoperationa after 6 months te eight possible outcomes are equaly likely to occur. tis als ied at the end of Therefore, the pertinent possibility space cons hove. We observe also that since the condition of a bulldozer is iy likely 40 be operational worth noting that among the e ‘6 months; this means thatthe different possibilities zhi possible outcomes, only one of them can be re mutuaty exclusive (we shall say more on this point in Section Finally, among the eight possible outcomes, the realization of ONN, NON, or NNO is tantamount to the event of interest, namely, “only one bulldozer is operational.” Because each possible outcome cur, the probability ofthe event within the above possibility space is 378. is equally likely 1 In designing # let-um lane for eastbound trafic ata highway intersection, such as shown in Fig. 22, sine te required probability of 5 or more cars waiting for left turns at any ‘ofthe left-turn lane sivsrurre Tetentieneu drs; 12 [zemeecatteneteesese) Figure £22 Design ofa left-turn lane For the above purpose, suppose that over a period of 1 week, 60 observations were made at regular time intervals (during periods of heavy traffic) ofthe numberof eastbound motor vehicles waiting For eft tur at this intersection, with te Following results EUSA MT aioe Case eg ase cece eee No. of Cars No.of Observations Relative Frequency 5 4 4160 1 16 16/60 2 20 20160 5 4 4160 4 3 3160 7 0 0 Conceivably, the number of vehicles waiting for left tums, during heavy traffic hours, could be any number, however, based on the above observations, it snot likey tha there wil be seven> EXAMPLE 23 > EXAMPLE 24 21 Events and Probobility < 29 's approximately 260 + 1/60 = 3/60, The estimated probabilities based on relative frequencies are approximate because of “sampling error” which may be significant when the estimate is based on a number of observations; this i part of the episemic uncertainty that we discussed earlier in he accuracy of the estimated probabilities will improve asthe total numberof observations mple sie) increases us we shall discuss further later in Chapter 6. < The simply supported Beam AB shown in 12 Joad of 100 kg that may be placed anywhere along the span ofthe bear The reaction atthe support 4, Rj, can be any value between (0 and 100 kg depending on the position ofthe load on the beam; in this ease, therefore, any ¥ between O and 100 kg is a possible value of Ry, and thus is its possibility space An event of interest may be that the reaction i in some specitid imerval; fr example, (10 < R 20 kg) oF (R, ). Therefore if particular value of Ry is realize, the event (defined by an interval) containing this vale of R has occured, and we can speak ofthe probability that Ry will ‘oF wll no, be ina given interval, For example, if we assume thatthe 100 kg load is equally likely to placed anywhere along the beam span, then the probability thatthe value of fy will be in a given terval is proportional to the length ofthe interval; for example P(IO = Ry £20) = 10/100 =0.10 and PCR, > 60) = 40/100 = 0.40, Figure £23 A simply suppor Consider the bearing capecity of the footing foundation for a building, Suppose that from pri experience, it isthe judgment of the foundation engineer thatthe bearing capacity of a footing atthe building site has a 95% probability of at least 4000 pst (pounds per square foot). If 16 individual footings are required for the building foundation, whst isthe probability that all the footings wil ve atleast 4000 ps bearing capacity? Conversely, wha isthe probability tha atleast one ofthe 16 footings would have its bearing capacity less than 4000 pt In this case, the possibility space consists of 2!°= 65,536 sample points. Suppose that each footing has « probability of 0.95 that its bearing capacity will be a least 4000 ps. Then ifthe bearing capocities of the different footings ae statistically independent ofeach othe, the probability that ali he footings will have bearing capacities of a east 4000 ps is then (0.95) = 0.440, Tn the second question, “at least one isthe complement of none.” Therefore, the probability of at least one footing with bearing eapacity less than 440 ps is | — 0.440 = 0.560. The concept omplement of an event and statistical independence will be ¢ later in Sections 2. < probabilistic problems30 > Chapter 2. Fundamentals of Probability Models 1. Every problem is defined with reference to a specific possibility space (containing more than one possible outcome), and each event is composed of one or more outcomes within this possibility space 2. The probability of an event is a function of the probabilities of the individual comes within a given possibility space, and may be derived from the probabilities of these basic outcomes. In Sections 2.2 and 2.3, we shall present the mathematical tools pertinent o and useful for each of these purposes. 21.2 Estimating Probabilities From the erved that in estimating or calculating the probability of an event, a basis for assignin nples illustrated above, it may be 0 probability measures tothe various possibl ‘outcomes is necessary, The assignments may be based on prior conditions (such ‘on the basis of prescribed assumptions based on subjective judgments, or based on results of empirical observations, o7 In Examples 2.1 and 2.3, the probabilities of all the possible outcomes were based on th ofthe possible operational ca prior assumptions. Inthe case of Example the three bulldozers was assumed to be equally probable, each equal to 1/8 (consistent with the prior information that each bulldozer is equally likely to be oper tional or nonoperational ter 6 months), whereas, in Example 2.3 the probability that the reaction R given interval was assumed to be proportional to the interval length (co assumption that the 100-kg load is equally likely to be pl beam). However, in Example 2.2 the probability of the number of vehicles waiting for left turns is based on the corresponding observed relative frequency, which is determined from available empirical observations. Finally, in Example 2.4, the probability thatthe bearing eater than 4000 psf is based on the experience and subjective will be in a sistent with the iced anywhere along the span ofthe capacity of a footing will be judgment of the foundation engineer. It should be emphasized that we shall teat probability as 2 measure necessary and useful for solving engineering problems where more than one possible outcome or event is possible (i., problems that are nondeterministic). In particular, we shall eschew the philosophical question of the meaning of a probability measure, and limit our concern ‘only tothe utilitarian aspects of probability and its mathematical and statistical theories fo ‘modeling and analy we use the factor of safety to develop en: tor. or employing Newton's second law of motion without being ig problems under conditions of uncertainty, in the same sense that designs without worrying about the real meaning of a safety f concemed about the meaning of mass and force. The usefulness ofa calculated probability, however, will depend on the appropriateness of the basi for its determination, In this regard, we should emphasize that the validity of th a priori basis for estimating a probability depends on the reasonableness of the underlying assumptions, whereas the validity of the empirical relative frequency may depend on the amount of observational data, When data are limited, the relative frequency by itself may give an inaccurate estimate ofthe true probability, or at best an approximate estimate of th true probability A third basis for estimati jective assumptions with available empirical obs bination is the Bayes’ theorem (see Section 2.3.5), and the probability (presented in Chapter 9) probability involves the combination of intuit result is knownElements of Set Theory—Tools fr Defining Events 31 2.2 ELEMENTS OF SET THEORY—TOOLS FOR DEFINING EVENTS We can recognize that the first requirement in the formulation of a probabilistic problem is the definition of the event of interest within a particular possibility space. The basic ‘mathematical tools for this purpose are contained in the elements of the theory of sets. In this section we present the basic elements of set theory followed in Section 2.3 with the jundamentals of the theory of probability as they relate to and are useful forthe formulation probabilistic problems in e 22. Important Definitions In the terminol of set theory, the set of all possibilities in a probabilistic problem is collectively a sample space, and each of the individual possibilities is a sample point. An event is then defined as a subset of the sample spe Sample spaces may be discrete or continuous. In the discrete case, sample points are discrete entities and countable; in the continuous case, the sample space is composed of a continuum of sample points. A discrete sample space may be finite (composed of finite number of sample: infinite (.., with acount le points). The possible configurations f the three bulldozers in Example 2.1 are an example of a finite diser e sample space each of the possible configurations isa sample point, and the eight possible configuration collectively constitute the corresponding sample space. Other examples of finite sample spaces are as follow Intemational Aiport the total mimber of fights landing at O"Hare in a 24-hour day isa fini mple space and each ofthe flights isa sample point within this sample space Examples of discrete sample spaces with countably infinite number of sample points re the following + The number of flaws in given length of welding—there may be none or only a few flaws in the weld, or the number of flaws could be very large. The actual number of flaws in a weld could conceivably be infinite od of one year. An accident may possibly occur with the first car erossing the ation on atoll bridge where a traffic accident may occur, each the possible locations isa sample point, and the sample space would be the continuumhapter 2. Fundamentals of Probability Models + Ifthe bearing capacity of a clay soil deposit is between 1.5 tsf (tons per square foot) and 4.0 sf then any value within the range 1.5 to4.0 isa sample point, and the entire respective of whether a sample space is discrete or continuous, however, an eve always a subset of the appropriate sample space. Therefore, an event will always contain ne or more sample points (unless it is an impossible event which is a null set), and vent, Finally, again, when we speak of probability, we are always referring to an event within a particular sample space The following example should serve to clarify the preceding definitions and concepts > EXAMPLE25 Consider again a simply supported bear AB as shown in Fig. E2.Sa (@) 1a concentrated Joad of 100 kg ean be placed only at any ofthe I-meter interval points on the team, the sample space ofthe reaction R will bea follows {0, 10,20, 30,40, $0, 60.70, 0, $0, 100 kg} ors] 10 1m Figure E250 Beam AB (b) The sample space of Rs and Ry that is, all pars of Ry and Rp such that Ry + Ra = 100 belong to the sample space, Graphically this sample space is shown in Fig. E2.5b. PENS faees fo egret cee Fesction at A o 2 40 "00 Kg igure EZSb Sample space of Ry and Ry. Figure E25 Sample space of Ry (6) Ite 100-kg toad can be placed anywhere along the length of the beam, the sample space of R can be represented by the straight line between 0 and 100 (Fig. E2.Se), whereas the corresponding Sample space of Ry and Ry isthe diagonal line shown in Fig. E2.Sd. In Fig. 2.5, an event may be Sefined as 20 < Ry = 40), whereas, in Fig. 2.5d an event for (Ry, Ry) may be between (20, 80) and 40, 60)Osrirst Scksany Rese: "300 ko Figure £2.54. Sample space of Figere E250 Sample mace of Ri, 0c (2) Next, consider thatthe . 300 ky and ean be placed anywhere along ean in this ca ‘or Rp, contains all the values between O and 300 kg as presented by the line sho eas the sample space of (Ry, Ro) is epresented by thre lines shown in Fi {) Finally, ifthe Joad can be any value between 100 and 300 kg and can be placed anywhere along 1 beam, the sample space of R, oF Ry is also the straight line of Fig. E2.Se, whereas the sample space of (24, Rx) Would be the hatched arca in Fig. E25 /Evont (Fy < 100, Fg> 1 20 a00Kg Figure E251 Sample space o Figure £2.59 Sample space of (Ry. Re) The’ evéot within the sample ‘space of Pig. 2.5, (%; 3/240 kx) is ite triamgelar région sbown i ig. E2.Se, whereas the even 0; Ry > 100) isthe trapezoid inthe same figure. < > EXAMPLE 26 From historical data of floods for a river, suppose the annual maximum flood levels above iver flow range fro Sm. Ifthe annual maximum flow is measured in an inerement of 0.1m, the sample space ofthe annual flow would conta the 51 sample points (1.0, 11,12, ....4.8 4.9, 5.0 m). The event of annual flood flow exceed 8,49, 5.0m) ‘nthe other hand, if the annual masimum flow canbe any level from 1 mo 5m, hen the sample space othe annual flow would be the continuum of infinite values between 1 m and 5 m. Similarly, he event of flood flow exceeding 3.0 m will be the continuum of values between 3m and Sm.Chapter 2. Fundamentals of P ability Models Special Events We define the following special events and adopt the corresponding notations indicated below + Impossible event, denoted @, is the event with no sample point. I is, therefore, an + Certain event, denoted S, isthe event containing all the sample points in a itself space; Le. itis the sample sp of an event E contains all the sample points in $ that are + Complementary even s Figure 241A Venn diagram of sample space S The Venn Diagram A sample space and its subsets (or events) can be r igram, As illustrated in Fig. 2,1, the sample space S is represented by a rectang on) within this rectangle, and the event Eis then represented by the circle (or any closed 1 vartof the rectangle that is outs xd region isthe corresponding complementary event E. In other words, the event £ contains allt oints in $ that are outside of E his clos ‘ample points within the closed region whereas E contains all the sample Figure 22b Venn diagram with nts A, Band C Figure 22a Venn diagram with two events A and B. he cam with two (or more) events i illustrated in F practical problems, the event of interest may be a combination of several Jy other events. For instance, in Example 2.1 the event of at least nwo bulldozers in operating This event would be the combination of vo on of the tw may be of inte condition, Such an event involves the w or three bulld There are only two ways that events may be combined, or an event may be derived section. Consider two events Ey and E In set from other events; namely, by the union ot in The union of E; and E, denoted E, U Ep, is the occurrence of E oF theory, or is used in an inclusive sense, which means or/and.) This means that Ey UE another event containing all the sample points belonging to eith The Venn diagram for the union of Ey and E> would be the hatched region shown in or both Fig. 23. It follows, therefore, thatthe region outside of the hatched region within Sis the complementary event EU E, ie. the complement of the event (Ey U F2).> EXAMPLE 27 Elements of Set Thaory—Tools for Defining Events < 35 Figure 23. Venn diagram for F Figure 24 Venn diagram of EE The union of three or more events, as shown Fig, 2.2b, means the occurrence of atleast one of them, and is the subset of the sample points within the three hatched regions of the dividual events A, B, and C The intersection of two events E and Ea, denoted E) M E> or simply EE, isan event ‘epresenting the joint occurrence of Ey and E; in other words, EE isthe subset of sample paints belonging to both E; and £, The Venn diagram of EE would be the double hatched ‘egion shown in Fig. 2.4 nally, the intersection of three or more events is the occurrence of all of them and jould be the subset of sample points belonging to all three individual events. xamples of Union of Events In describing the state of supply of construction materials, if E, represents the shortage of concrete, and E3 represents the shortage of steel, then the union Ey U E is the shortage of concrete or steel, or both, In this case, the complementary event, E; U E mneans no shortage of construction material, ¢., concrete and steel are both available, whereas E; U Ey means there is no shortage of concrete or there is no shortage of stcel (observe the subtle difference). goes between Chicago and New York may be by ar, hi If the availability of each ofthese three modes of transportation is denoted, ely, as A, H, and R, the available means of transpk 1s between (New York is (A UH UR), ie., cargoes may be shipped by air orhighway Examples of Intersection of Events «+ Referring tothe first example above, the event ££ would mean the shortage of both ‘onerete and steel, and Ey E> means no shortage of both materials Referring to the second example above, the event AHR means that all three modes of transportation between Chicago and New York are available, Observe also thatthe event AH R means that only air transportation is available ‘Suppose there are two highway routes from city A 10 city B as shown in Fig. E2.7a. Let Ey represent the event that Route | is open and F that Route 2 is open to traffic. Then Ey UE: means that Route | is open or Route 2 is open; in other words, atleast one ofthe two routes is open Figure £274 Two routes from A to B.36 > Chapter 2. Fundamentals of Probability Mode ion, E,Es, means that both Routes | and 2 are open, whereas EF means that tut Route 2s closed, and E E's means that both routes ae closed perhaps because Figure E27 Route | from to from B 10 ¢ Nextconsder the three cites with Route I connecting A to and Route 2 connectin is shown in Fig. E2.7b. I Band E>, respectively, mean that Route B06 hat its not possible 1 go fom city Ato ty 2 ee seas Figure £27¢ Routes | and? from Ao B Finally, assume that there are two alternative routes from city A to city B, and only Route 3 from city Bocity C.as shown in Fig. E2-7e. In this ease, there ace two possible ways to go from city A to city (Cs namely, Ey Es EB altematively, this ean be expressed as the event (E) U Ea) Es. Observe that the event Ey Es U Ey means that there will be no transportation from city A to city C < > EXAMPLE28 ‘Consider the last case of Example 25, in which the load ranges from 100 kg to 300 kg and the sampl space ofthe two reactions (Ra, Ra) as shown in Fig. E25 If A = theeventthat Ry > 100g; and B = theevent that Ry > 100 kg re events A and B would be the respective subsets ofthe point pairs shown in Figs. E2.8a and E2. Fo Fa 00) 200) er 2001 mA 109] 0100 200 300% 0100 200 S00Kg Figure E28 Subset of event F Figure £2.84 Subset of even A.22 Elements of fools for Dofining Events 37 “Fix 0100200 300% 7 100200 300% Figure £28 The union AU & Figure £284 The intersection AB, The union A UB isthe hatched region in Fig, E28, whereas the intersection AB would be the hatched oa in Fig. E284. Note that in this example, Figs, E2.8a through E2.8d also serve as the corresponding Venn dingrams. < Mutually Exclusive Events, Two events are mutually exclusive if the occurrence of one precludes the occurrence of the other event; that is, the occurrence of both events is impossible, The corresponding subsets of sample points, therefore, will have no overlap as shown in the Venn diagram of F E Figure 25 Mually exclusive events £) and E If two events are mutually exclusive, then thei intersection is an impossible EE=6 Examples of events that are naturally mutually exclusive include the follow 1. A car making a right tum and making a left tur at a street intersection 2. Occurrence of flood and occurrence of drought of a river at a given time 3. Collapse and no damage of a building under a strong earthquake Similarly, three or more events are mutually exclusive if the occurrence of one event pre cludes the occurrence of all the other events. Examples are the follov 1. If there are three competing locations for a single airport, then the three choices for the final location of the airport are mutually exclusive. 2. In Example 2.1, the number of bulldozers that will remain operational after 6 months are mutually exclusive. 3. In Example 2.2, the number of vehicles waiting for left tums at the intersection are mutually exclusive38 > EXAMPLE29 dementals of Probability Models However, in Example 2.7, the conditions ofthe different routes are not mutually exclusive the elosing of one route does not necessarily preclude the closing of another route. Likewis in Example 2.8, the events A and B are not mutuall sive because both Ry and Re can Kcced 1UN) ke if the load is sufficiently high, say > 200 kg; also the intersection AB is nc null set Collectively Exhaustive Events Two or more events are collectively exhaustive if the union of the events make up the underlying sample space. For example, the event E and its complement £. as shown in y, EUE=S. Two construction companies @ and b are bidding for projects. Define A a Company a ompany b wins a bid. Let us sketch the Venn diagrams forth wins a bid, and B as the event that C ne project, and Company bis submiting ts own bid for anoshe 1. Company ais submiting abd projet, The Venn kd be as shown in Fig, E2.9 OM RE) a7) LN \ i Le Li fe ietkted = i> Figure E29. Sample space of A Figure £2.96 Samp and inutualyexclsive and B In this case. itis posible fr boh companies to win their respective bids as represented by the intersection of A and B, are also other bidders for is aang b are submiting bids for the same project, and the onding Venn diagram would be as showa in Fig the project. The con In this case, Company @ or b may be awarded the pro vended the project. If Company a wins its bid, then no other companies, including Company b can Tho be the winner of the award. That is, the occurrence of A precludes the occurrence of B: ths, The ewente A and Beare mutually exclusive, as shown in Fig. E2.9b where there is no overlappin region between A and B. Moreover, the complement of AUB, Le, AU B;means that one of the ther ale projec ‘3. Company a and Company b are the only companies si ompeting bid The Venn diagram for this case woul. case, since Company a and Company b are the only bidders for the single projec, In tis particul wy one of thems can be the winner, the events A and B are mutually exclusive, and are also and leerively exhaustive or AB =S. Thus, the sample space contains only the two sets A and B as shown in Fig. 2.9 AB Ki pace with A and B only Figure E239¢22 Elements of Set Theory—Tols for Defining Events < 39 > EXAMPLE 2.10 There are three possible sites, denoted as Site, Site B and Site efor the constriction of «new airport for a major city: Define the following Site ais selected forthe airport B = Site bis selected fo the sirport: and C= Site is selected for the airport It the above three sites are the only feasible sites forthe aipor, then the available alleratives for the location ofthe airport isthe union A UBUC. However, if one ofthe altemative sites is selected, then this precludes the selection ofthe two other sites therefore, the events A, 2, and C are mutually sat as shown in Fig. £2.10. Figure £210 Sample space of events A, , and C In this case, we also observe the following: ABC means that Site ais selected and not Site bor Site, AUBUC means that none ofthe thee sites is selected as location forthe airport. 222 Mathomatical Operations of Sets In Section 2.2.1, we observed that two of more sets, or events, can be combined in only two ‘ays; namely, by taking the union or taking the intersection. These two operations and tl ‘of taking the complement of an event constitute the basic operations involving sets, The notations that we have adopted to designate sets and the associated operations are as the union the intersection longs to, oF is contained in E = the complement of E With these notations, the mathematical rules governing the operations of sets are the Equality of Sets, Two sets are equal if and only if both sets contain exactly the same sample points. On this40 > Chapter 2. Fundamentals of Probebilty Models Als A 4 Furthermore AUA=A ANA=A and, forthe sample space 5, ANS=A On Complementary Sets With regard to an event E and its complement E, we observe the following: Con mutative Rule re commutative; that is, for two sets A and B. AUB=BUA Associative Rule he union and intersection of sets are also associative; that is, for thre AUB)UC = AU(BUC Also, (aByc = A(BC) Distributive Rule Finally, the union and intersection of sets are distributi A,B, and C and als, (ABUC = (AUC) (BUC We might observe that the above commuta sare similar to the same algebraic rules for numbers. In particular, the operational rules22 Elements of Set Theory— Defining Events < 41 ferning the addition and multiplication of numbers apply (with certain equivalences) to the union and intersection of sets. With the following equivalences—union for addition and inters tion (ie + and 1 x)—the rules of conventional algebra apply to operations of sets or events, Moreover, in accordance with the hierarchy perations, intersection takes precedence over union of sets, unless parenthetically herwise. Tt should be emphasized that the above equivalences are only valid in an operational sense; conventional algebraic operations such as addition and multiplication have no mean. for sets or events, Moreover, there are no equivalent operations for subtraction or sion of sets. On the other hand, there ate operations and operational rules that apply to that have no counterparts in conventional algebra, For example, fora set A AUA=A and ANASA Another case in point isthe second of the distributive rule described above, which says that (AUCKBUC) = ABUACU BCUCC Hence, the final result is OB UC) =ABUC ‘whereas in conventional algebra we would hav £ ab + Finally, another important rule that applies to sets but has no counterpart in conventiona algebra is the de Morgan's rule, as described below De Morgan's Ri This rule relates to sets and their complements. For two sets, or events, By and Ep, the de Morgan’s rule says that FUR =BnF neral validity ofthis relation can be shown with the Venn diagrams in Fig. 2.6. The unhatched region in Fig. 26a is clearly E, U Ez. The two Venn diagrams in 2,6b show, respectively, the complementary sets E and E>, the intersection of which is the ouble-hatched region in Fig. 2,6c. From Figs. .6a and 2.6¢, we see the equality ofthe two sets Ey U Fy = Ei 1 Es, thus verifying the de Morgan's rule In more general terms, the de Morgan’s rue is E, =5 Applying Eq. 2.3ato the complements of Ey, E E42 > Chapter 2. Fundamentals of Probability Models g, E, Ey Figute 26. Venn diagrams showing de Mor Tn light of Eqs. 2.3a and TBUCWAUC xamples > EXAMPLEZ11 Consider a simple chain consisting of two sry the load F if ether link breaks; thus, eet Figure £2.11 A two-link chain, For further illustrations of the de Morg ks Bx = the breakage hence, taking the complements of both sides of the ubove equation, the de Morgan’s rule E E elation: The comple TBC = ABUT (AUB)UT = ABUT ABC UAC = ABC NAC shown in Fig. E2.11. Cleary, the chain will fail to y = the beeakage of lnk 12 Elements of Sat Theory—Toos for Defining Events ~< 43 The Failure of chain = BU E No failure ofthe chain, therefore plement EU Fz. However, no failure of the chain also means that both links survive hats hich is a demonstration of the validity of the de Morgan's rule applied to an en problem > EXAMPLE 2.12 ‘The water supply for twa cities Cand D comes from the two sources A and Bas shown in Water is transported by pipelines consisting of branches 1, 2, 3, and 4. Assume that ether o wo sources, by itself is sufficient to supply the water for both cites. branch 1 By = failure of branch 2 Ey = failure of branch 3; and Bq = failure of branch 4 re-of& pipe branch means thee is serious leakage or rupture ofthe branch Source A Figure £2.12 A water supply system. Shortage of water in city C would be represented by E11 U Es and its complement EE3UE cans that there is no shortage of water in city C. Applying the de Morgan's rule, we have FEUE EE The lastevent above means that there is no failure in branch Jor branch 2 and also no failure in branch rent EE UE There 2. Similarly, shortage of water in city D would be the eno sho of wate in city Dis REU RUE = ‘which means that there is sufficient supply at the station, ie., (Ey UF), and there are no failures in both branches 3 and 4, represented by EE <44 Chapter 2. Fundamentals of Probability Models 2.3 MATHEMATICS OF PROBABILITY Inallof our discussions thus far, we have tacitly assumed that a nonnegative measure, called 1d that such probability measures possess certain properties and follow certain lity. As with other branches of mathematics, the theory based on certain fundamental assumptions, or axioms, that are not subject are as follows (see, e., Feller, 1957; Parzen, 1960; Papoulis, 1965} Axiom I; For every event F in a sample space S, there isa probability P(E)=0 24) Axiom 2: The probability of the certain event $ is PS) = 10 es) Axiom 3: Finally, for two events Ey and B that are m exclus P(E, UE) = PCE) + PUE 26 Equations 2.4 through 2.6 constitute the basic axioms of probability t sumptions and, therefore, cannot be violated. However, these axioms and thy resulting theory must be consistent with and useful for real-world problems. In this latter regard, we may observe the following + The probability of an event, P(E), is « relative measure, i, relative o other events in the same sample space. For this purpose, itis natural and convenient to assume + Because an event, convenient to normalize its probability relative to § (the certain event), as specified always defined within a prescribed sample space S, itis nerefore, on the basis of Eqs. 2.4 and 2.5, it follows that the probability of an event E is bounded between 0 and 1.0; that is, Vs P(E) < 10 with frequency standpoint, if an event Ey occurs my times among m repetitions of an ¢ stitions, in which Ey and E cannot ‘occur simultaneously (they are mutually exclusive), then Ey or BE will have occurred gard to the third axiom of Eq, 2.6, we may observe intuitively that from a relative and another event £3 occurs my times in the same m r he m repetitions of the experiment, Thence, on the basis of n = PU ures. As expected, all such asized that the mathematical theory of prob: bility provides the log should be em ;elationships and any theoretical results are based on the three basic axioms stated in Eqs,23 Mathematics of Probability ~ 45 231 The Addition Rule As an event E and its complement E are mutually exclusive, we obtain on the basis Eq, 2.6 P(EUE) = P(E) + PE but since E 5, we have, on the basis of Eq. 2.5, P(E UE) = P(S) = 1.0. He we obtain the first useful relation P(E) =1- PE) More generally, if two events Ey and Ey are not mutually exclusive, the addition rule i The general addition rule of Eq. 2.8 can be shown to follow from Eq. 2. For this purpose, we first observe from Fig. 2.7 that By UE = E Es gE, < | SE, Ey} Figure 27. Union of E) and Eye As can be seen in Fig. 2.7, £ and EyE are o Eq. 2.6 Fie) = P(E) + PE But E\E2 U E,£: =(E\U Ei)E2 = SEs = Ez; and EyE and Ey E> are clearly mutually xelusive. Hence P(E\E:) = P(E) ~ PE\E from which we obtain Eg. 2.8 > EXAMPLE2.13 A contractor i starting two new projects—jobs | and 2. There is some uncertainty on the schedul ‘ompleton of each job; atthe end of 1 year, the condition of complet A = definitely compl 18 = completion questions PRonLems 1. Deseribe the sa situations regarding the completion of both jobs | and 2 at the end of I year in ter hove: €.2, AA moans both jobs willbe definitely completed in 1 ye np space forthe 4 pertinent Venn di By iste even that job | wil efnitely be completed in 1 year, then mutually exclusive. Therefore, according amis shown in Fig. E2.13 where ll the sample points are contained in 5.> EXAMPLE 2.14 Figure £213. Sample space with and E ‘Similarly, if Bp isthe even that job 2 will defintely be completed in| yea 2. Assuming that each state of completion of both jobs is equally likely athe end of 1 yea (i.e. ea ample point has a probability of V9), w ib will definitely be completed atthe end of 1 year? nthise wentofinterestis the union E) UE thatthe imersection E, E3 > (AA\: therefore, according to Eq, 28 PCE UE) z 2.13 that (E1 U Ea) 3 (AA.AB, AC, BA, CA) its probability is also We can also observe from F E 3 (AB.AC, BA, CA) Its probability, therefore, is 49 a fon yielded the results shown in Table E2. By = more than 2 vehicles waiting for left tums Ey = at most 4 vehicles waiting fr left turns The different number of vehicles waiting for left turns atthe intersection are obviously mutually cies in Table F2.14 to represent the corresponding exclusive events. Then, ding the relative frequebility < 47 TABLE E214 Observations of Vehicles Weiting for Left Turns of Vehicles Waiting No. of Observation 4 Also, the intersection ith corresponding probabil en, according to Eq. 2.8 0 57 17_ 60 In this ease, we observe that the un F; contains al he possible number of vehicle for left turns, Le, the entire sample space. Hence, its probability is 1.0, > EXAMPLE 2.15 in Example 2.8, two events associated with the reactions at A and Bare defined as A = (Ry > 100kg) B 100s) ‘These are represented by the respective subsets in the sample space of Fig E215. For this illustration, we may assume thatthe sample point are all equally likely to occur. Ths implies that th Figure £2.15 Venn diagram of events.48 > Chaptor2 Fu > EXAMPLE 2.16 tals of Probability Models Total area of samy (300)? — (100)71 = 40.000 Then, refering to Fig. £2.15, we ha Similarly, PCB POAB) = “0000 ~ 3 Tpeetah be PAUB)=h+4—2 2a 8 18 From Fig. £2.15, we also obtain the area of the unio A UB) = 40,000 — +(100)* = 35,000 from which we also obtain P(A U B) = 290 = f ms 40,000 = 8 < Extending the addition rule, Eq. 2.8, to thre Ex. Ex have P(E\ UE U Es) = PIE UES) UE P(E\ U Ex) + PUES) ~ P(E, UEDDE oo P(E\) + PUEx) + P(Es) — PELE Pe Pi. PEL ExED The above procedure of the addition rule may be extended to the union of any number by applying the de Morgan's rule, as follows: E E)=1-7 ) P(E\Ex--E If the m events are all mutually exclusive, however, extension of Axiom 3, Bg 2.6, yields Te major artine industry i subject wo labor strikes by the pilots, mechanics, and the or by two oF more Tabor groups. Using the following notations, B = strike by the mes = srike by the ight attendant etcrmine the probability ofa labor strike in the major rine industry in the next 3 years. Assume dl of strikes by the three individual groups: P(A) = 0.03; P(B)= 0052 Mathematics of Probability 49 PEC) =0.05, and that strikes by the diferent labor groups are starsically independent (s 2.32 and 2.3.3), which means according to Eq, 2.15, P(AB) = P(A)P(B): PLAC) = PLA)PIC,—-P(BC) = PEEP and PLABC) = P(A)P(B)PCC SOLUTION A strike in the industry will occur in the next 3 years if any one oF more ofthe thee labor groups go.on strike during this period: therefore we are interested in the union of A, B, and C ‘whose probability according to Eq. 29 is P(AU BUC) = 0.05 + 0.03 + 0,05 ~ (0.05 x 0.03) ~ (005 x 0.05) The solution may also be obtained (more conveniently) with Fi. 2.10 as follows: P(AUBUC)=1-PABO) Note that A, B, and C are also statistically independent; ie, P(A BC) = P(A)PCB)P(C). Thus, we also obtain PAUBUC) = 1 ~(095 «097 x 0.95) = 0.1246 < 232 Conditional Probability here ions whe bability of an event may depend on the occurrence hnonoceurrence) of another event. If this dependence exists, the relevant probability is condi rabability. For this purpose, We shal use the notation P(E\\E2) = the probability of Ex assuming the oceurence of Ez, or simp he probability of By given E> In the Venn diagram of Fig. 2.8, we may observe intuitively the followin Te conditional probability P(E\|E2) may be interpreted asthe likelihood of realizing effectively interested in the event Ey sample point of Ei that is in E. In other word! herefore, the conditional probability pertain within the “reconstituted sample space” E to the sample points of £y relative (0 those of E and thus must be normalized with respect to Ey; hence, with the appropriate normalization, we obtain the conditional probabil PEELED), PIE) = 2. I may be well to emphasize that the conditional probability, as defined in Eq, 2.1 merely a generalization of the (unconditional) probabilit event, When we speak of EE, Figure 28 Venn diagram for E give50 > Chapter2. Fundamental > EXAMPLE 2.17 Probability Models the probability ofan event E, itis implicitly eonditione ‘on the sample space $, To be more plicit, P(E) should be expressed as PUES) But since BS = E and P(S) = 1.0. PUEIS) = PE) Inother words, conditioning on the sample space S is presumed to be understood; howeve Wealsoobserve the fol PEE PCE\|E2) + PEE) + 1 FE Pie: U EES P(EY) PE) Therefore P(E\|Es) = 1 — P(E\IED) 12) which is a generalization of Eg. 2.7. It is important to recognize that in Eq. 2.12 the conditioning event Eis the same reconstituted samp! .. £1) and For this reason, one must make sure, when af ing Eq, 2.12, thatthe event ( its complement refer to the same conditioning event E Observe, for example, the following PCEAIE PEE PORES #1 ~ PE\IED) There are two highways from City A to City B as shown in Fig. £2.17 Rowte 1 ison fat trrain, whereas Route 2s a scenic route that goes through mountainous te During severe winter sea ns, one or both routes may be closed to trafic because of heavy snowfall Route Route Figure £2.17. Highway routes from A to B.23 Mathematics of Probabilty $1 4 = Route Lis open to trafic Bz = Route 2s open to traffic Between the two routes, Route 2s obviously more i ths. Moreover, the condition of Route 1 durin 1nd te probability that both routes willbe open is Then, if Rowte 2s open during a snow storm, the probability that Route 1 is also open is, accord PEVE:) _ 040 _ 9 59 (On the other han ssed during a severe snow storm, the probability that Row 1 is. . Per) ae PEE) = 1- PEE) =1- PEVUED) PE\) + PEs) = PCEVED)) Therefore, we obtain on PEE) = 0 4 Route 1 is open given that Route 2s closed i P(eE\\E) = 1-020 = 0.70 < > EXAMPLE2.18 that mo poaching a certain intersection are tice as likely to go st than to make ar ims are only half likely aright tw As vehicle approaches the intersection, the possible directions may be define as fellows By = straight abead Ey = turing righ Ey = wming lef52 > Chapter 2. Fundamentals of Probability Models ind the respective probabilities willbe 4 1 PIE) PCE) = 2; and PUES) = 5 AL te intersection, ifa vehicle is definitely making a tur, the probability that it wil be aright turn is (observe that the three alternative directions are mutually exclusive) PLE(E:UEs)) _ PEL Eo P(EUE) ~ PUUES) PUES) PE) + PU (On the other hand, if the vehicle is definitely making a tun atthe intersection, the probability that it By) = 1 ~ PUSIELUE Statistical Independence Ifthe occurrence, or nonoccurrence, of an event does not affect the probability of occurrence ‘of another event, the 1wo events are statistical 1. In other words, the probabil ‘of occurrence of one event does not depend on the occurrence or nonoceurrence of another event, Therefore, if two events Ey and istically independent PEE) = PLE PCE E,) = PLE 2.13) It might be prudent to point out here the difference between events that are independent versus those that are mutually exclusive. The difference is profound. and the should be no confusion—statistical independence between two events refers to the proba mutually exclusive when their joint occurrence is impossible, as the occurrence of one event precludes the occurrence of t other. In other words, P(E>|E,) = Dif E) and E are mutually exclusive. Finally statistical independence of two or more events pertains to the probability of the joint event, whereas 233 The Multiplication Rule From Eg, 2.11, itf nat the probability of the joint event BE is PE\E2) = PCEWIEs)P(ED) PUEVE:) = P(ENE P(E) ith Eq, 2.13, if Ey and Ey are statistically independent events the above mu> EXAMPLE2.19 23 Mathematics of Probability < $3 For three events, th PEE) P(E | Es) PUES) events are statistically independen We would expect that if two events Ey and By are independent, t ments Ej and Ey would also be statistically i PCE We might emphasize that all the mathematical rules pertaining to the probabilities of vents apply equally to the conditional probabilities of events that are conditioned on the particular, we may abserve the followin P(E, U ExlA) = P(E\|A) + PURI P(E\E2\A) = PCE\|E:|A) CEDIA) and if Fy and Ey are statistically independent events, given event A. Ea|A) = PEA) P(ES|A Consider again the chain system of Example 2.11 consisting of two links as shown in Fig, £2.19 subjected to a force F'= 300 kg, F=300 kg C—O 300 kg Figure £2.19 A 1wo-link chain Ifthe fracture strength ofa link is less than 300 kg, it wil fail by facture. Suppose tha the probability ofthis happening o either of the two Tinks is 0.05, Clearly, the chain wil fail if one or both of the two links should fail by fracture, To determine the probability of failure of the hain, define By = fracture of link J Then, P(E) = PE 1 probability of failure of the chain is P(E) + PCE) — PELE = 0105-4 0.05 — PCEIE DPE = 0.10 — 0.0SPCESIES4 > Chapter 2. Fundamo > EXAMPLE220 > EXAMPLE 221 Js of Probability Models We hatte solution requires the conditional probability PEE), which isa anti ‘mutual dependence between B} and E If thereisno dependence or they are statistically independent PAE: |E\) = P(E) =005. In this ease, the probability of failure ofthe ein is P(E\U Ex) = 0.10 0.05 x 0.05 = 0.0975 her hand, i there is complete or total dependence between Ey and Ea, which means that link fractures the other will also fracture, then P(E) = .0. In such a failure ofthe chain becon P(E; U E:) = 0.10 = 005 x 1.0 = 0.05 In this later case, we see that the failure prot probability of a single ink, Therefore, we ean state thatthe probability of failure of ridaion ofa tall building may fail because of inadequate bearing capacity or by excess n failure, and assame PC) = 0.00 The settlement. Let and S represent the respective modes of founda (5) =0.008, and P1B|S) =0.10, whichis the conditional probability of beain that there is excessive settlement. Then, the p P(BUS) = P(B) + PCS) — PCBS PB) + P(S)~ PCBIS)PCS) = 0.001 + 0.008 — (0.1)(0.008) 0.0082 However, the probability thatthe building will experience excessive settlement but without bearin PIS NB) = PCBS) PIS) = (= PEaIIPLS! In this problem, we might observe thet the conditional probability P(B|S) cannot be larger than 1/8 The reason for this assertion isa follows: P(BIS)P(S) = PLSB) PCB) 0.001 P(Bi8) = OE mosiay = Lv(siB) Because P(S|B) < 1.0, the vale of P(A|S) is Himited to-a maximum of 18 < {wo rivers, a and b flow through the neighborhood ofa paper mill that is allowed to dispose its waste sissolved oxygen, DO, level in the water downstream is Let into both rivers. The degree of pollution of A = the event that River is found to have unacceptable level of pollution 1B = the event that River b is found to have unacceptable level of pollution Based on the records of the respective DO levels inthe two rivers tested over the past year it was> EXAMPLE 222 P(A) = 20% and -P(B) = 33% P(AB) = 10% (On a given day, the probability that at least one of the rivers will have an unac level of pollution is P(AUB) =0.20+0.33 0.10 = 0.43 If River ais tested to hav the probability that River b will also have PAB) _ 0.10 Poa EC) Pay = 020 whereas the probability that River @ wil have an unacceptable pollution level assuming that River & PLAB) _ 0.10 AIB) = oo L030 P@) ~ 0:33 SOLUTION ‘This means that either one of the two rivers is polluted and the ether isnot; thus, the p(AB UAB) = PCB\A)P(A) + PCAIB)PCB) 0.20 +0.70 x 033 = 0.33 < The electrical power for a city is supplied by two generating plants—Plant a and Plant b. Each of the plants has sufficient capacity to supply the average daily power requirement of the entire city However, during peak hours of a day both plans are needed; ot be brownouts in parts ofthe ity. Der A = failur B = failure of Plant b nd assume Pia) = 0.05 P(B) = 0.07 P(AB) = 0.01 If one of the two units should fail on a given day, what isthe probability of failure ofthe other unit SOLUTION ‘The probability of ty unit fils est. For P(AB) _ 001 BI ispeay ans Te56 > Chapter 2. Fundamentals of Probability Models P(AB) _ 001 Praia AB) _ ON 0.14 Related questions and answers: Whats the probability of a brownout in thecity ona given day’? Brownouts will occur when ofe or both ofthe plants fil therefore, the pertinent probability is P(AU B) = P(A) + P(B)~ P(AB) = 008 + 0.07 0.0 ou If there i a brownout within the city during the peak hours ofa day, the probability that itis caused solely by the failure of Plan ais PLAB(AY BD BAUABB) __PCAB P(ABIAU B) pees SAE sf PAU A) PAUB PGBIAYP(A) _ (1 — 0.20005 Similacy, if there is a brownout, the probability thai is caused solely by the failure of Pant b is PAB) _ PCAIB)PCB POU ie ioe 1 On Finally the probability that a brownovt will be caused by the failures ofboth plants would be PIAB(AU BY _ P(ABAUABB) __P(AB AB\AUB) = Bye PAUB) P(AUB) ~ PCAUB) — 991 _ O09 201 90 > EXAMPLE 223 Suppose that before a section (say 1/10 km in length) of a newly constructed highway pavement 's accepted by the State Department of Highways, the thickness of a 25-em pavement is inspected or specification compliance by taking ukrasonie readings at every 1/10-km point of the constructs pavement, as shown in Fig. F2.23, Fach 1/10-km section willbe accepted if the measured thickness ted or a penalty will be imposed E 0.10 km + aren essrsazat | Q d Utrasonic racings it Figure £223. Inspection of constructed pavement rom past experience, 90% of constructed highway works by the same contractor were found >be in compliance with specifications, The ultrasonic thickness determination i only 80% reliable23 Mathematics of Prob Use the following notations: G = constructed thickness of pavement is at Teast 23 em PxGlA) = 0.80 and als PGA) = 081 This may also be interpreted to mean that an ultrasonic reading could have an error of 20%, which PGA) = PIGIA) = 02 Based on the contractor's past construction record, we may 2 1 9056 of his constructed pavement will have accep Th bee accepted by the Highway Department is, probability that a particular 1/10-km section ofthe pavement is well constructed and will PGA) = PUGIA)PCA) The more pertinent an practical questions ar perhaps the followin 1. What isthe probability that a well constructed section of the pavement will be accepted by th Highway Department on the bass ofthe ultrasonics test? 2. Conversely, what is the probability tha a poorly constructed section will be rejected: SOLUTIONS The probability that a well-constructed section ofthe pavement will be accept Since G = G(AUA) = GA U GA, where GA and GA are mutually exclusive PUG) = P(GA) + PGA) = P(GIA)P(A) + PGA) pai = PGP PIG) 10 a < 234 The Theorem of Total ProbabilityChapter 2. Fi > EXAMPLE 224 3s of Probability Models Ej.i=1.2,...nand the probability of A will depend on which of the £,'shas occurred. O} such an occasion, the probability of A would be composed of the conditional probabilities riditioned on each of the £;’s) and weighted by the respective probabilities of the Such problems require the theorem of roral probabi Jefore formally presenting the mathematical theorem, let us examine the following -cample to illustrate the essential elements of the theorem. The flooding ofa riverin the spring season will depend onthe accumulation of snow inthe mountains during the post winter season. The accumulation of snow may’be described as heavy. normal, and Tight. Clearly ifthe stow accumulation in the mountains is heavy, the probability of Rooding inthe Flooding of course, may also be caused by rainfall F = occurrence of flood H = heavy accumulation N = normal accumula L Gnctuding no P(E IH) = 0.90; Then the probability of flooding inthe river during P(F) = PUF|H) PUD + P =o41 in Example 2.24, we can observe the followit 1. The accumulations of snow in the past mutually exclusive 2. The probabilities of the three levels of pio 1.0. Therefore, the three events Hl, N, and L are namely Ej, E Eq. That is, Ey UE same sample space § as shown in Fig. 29, lows: A= A(E\ UE = AE\UA\ where AE},AE3,....AEy are also mutual Pw) = sin the spring. With the following notations ing inthe accumulation of snow =040; PUPIL) =0.10 050; Pu) =0.30 PIN)P(N) + PCPIL)P( 0.50 +0.10 x 0.30 < inter, namel normal, and ligh snow accumulation, namely, HN, and Ladd mutually exclusive and also collectively ex Eq = §. Then, if A is an event in the we derive the theorem of total probability as E ysive as can be seen in the Venn diagram> EXAMPLE 225 babilty < $9 AE, AE, AE, AE, E | & \& Figure 29. Intersect 1nd by virtue of the multiplication rule, Eq. 2.14, we obtain the theorem of P(A) = PALE) PCE) + PLIES) PLE POALE,) PCE Again, in applying the abov. probability theorem, it is important to observe that the conditioning events Ey must be mutually exclusive and collective exhaustive Hurricanes along the Gulf of Mexico and the eastern seaboard ofthe United States occur every yea hurricanes are classified into five categories from C1 through ‘mostly in the summer and fll. Th C5: tobe classified as a hurricane, the wind speed must be atleast 75 miles per hour (120 km). The ‘wit the categories; fr example, a Category C5 frequencies of hurricanes, of eourse, would dec hurricane, with sustained wind >150 mph (242 kav), would very ‘We might observe first thatthe five categories of hurricanes, C1, C2, C3, C4, C5, are mutual 1 winds (denoted CO) are also exclusive, a it is reasonable to assume that no £80. ries pls nontr all possible hurticanes; thus, these five cae collectively exhaustive. Assume that annually there can be at most one hurricane striking a particular areain the southern coast of Louisiana along the Gulf of Mexico, and the annual occurrence probabilities ofthe different hurricane categories areas follow P(CI)=035; (C2 =025; PIC =0.14, PICA) =005; PCS) = 0.01 tural damage to an engineered building inthe reference area can be expected to occur depending onthe eategory of hurricane that the building will be subjected to, Suppose the conditional probabilite of damage to the building are xs follows: P(DICI) = 005; P(DIC)=0.10;-—-PEDIC3) = 025; (DIC) = 0.60; P(DICS)= 1.00; and PUDICO) hence the annual probebilty of wind damage tothe building would be PUDICA) PIC) + PCDICS) PICS) PAD) = PDICI)P(C1)+ PCD|C2)P(C2)+ PLDICI) PLC = 0.05 % 0.35 +0.10 x 0.25 + 0.25 x 0.14-+ 0.60 x 0.05 + 1.00 x 0.01 = 0.1175 to the building is about Therefore, annually the probability of hurricane wind damag We might observe from the above caleulaions that the greatest contributions to the annus damage probability are from hurricanes of Categories 3 and 4, P(DIC3)P(C3) = 00085 and PUDICHP cour under a Category 5 huricanc, P rnnually P(CS) = 0.01, which means that it might occur (on the average) only about once in e 100 years. 1.030. Observe also that even though damage to the building will certainly oc (DIC) = 1.00, the occurrence of such husricanes is very rare <> EXAMPLE 226 tals of Probability M Figure E2.26 shows the eastbound directions of wo interstate highways f, and fz merging into another highway Zs Interstates Fy and 3 have the same traffic capacities; however, the rush-hour tafe volu on i about twice that on so that during rush hours the p respectively s Ey and Ea. are as follows: Figure £226 Eastbound interstate hi Also, when one route has excessive trafic, the chance of excessive traffic on the ether route can We might wish also to determine the probability of traffic congestion onthe third route Is, PUE 1. First et us assume thatthe capacity of Fy i the same as that of fy or J, and that when fy and fa are both carrying less than ther respective trafic capacities, there is a 20% probability that will experience excessive trallc:ie., PCEs|E \E:) = 0.20 We would expect thatthe relevant probability will depend on the trafic conditions on fy and ; Ia, which may be E\E>, EE, ExEs, or EE observe that these four joint events are mutually exclusive and collectively exhaustive. Their respective probabilities are then PUE\E:) = PCEs|Es)P(Es) = 0.40 0.20 = 0.08 PUE\E2) = PCEy\E,)P(E\) = (1 — 0.8090.10 = 0.02 P(E E:) = PCE \E)P(E:) = (1 ~ 0.4090.20 = 0.1 P(E 2) = 1-008 0.02 0.12 = 0.780 c ongested when the traffic on Jor Fa or both are excessive, Then leaty, the traffic on Fy will be sbtin the total probability of traffic congestion on Js tobe (Es) = PCEsIEy Ea) P(E\E:) + PCEs\E\ Ea) PCE\Es) + POEs E 2. Next, assume thatthe trafic eapacity of fy is twice that of Fy or Iz In this ease, if only one of he two toutes fy or da has excessive trafic, the probability of congestion on Jy will be 25%, ie PUEs\EvEs) = PCE capacity of fs will be exceeded is 95%, meaning P(Es|F\F:) = 0.95 ,) = 0.25. If both routes have excessive trafic, the probability tha the> EXAMPLE 2.27 23 Mathematics of Pr ity «61 In this latter case, the probability of traffic congestion on Fy is then P(Ey) = 095 x 0.08 + 0.25 x 0.02 +025 x 078 =o < In the next example, we will consider a problem involving double conditional events, in association with the application of the total probabil The town of Urbana is in the midwestern United St 1 county of Champaign, Ilinos, which is located in the plains area of High-wind storms ca occur in tis ar 50 mph (96 kav), and occasionally such storm: ‘with maximum wind speeds in spawn torsades creating even hig the annual probability of structural damage to residential houses in Urbana, and base on historical data were able to determi storm the probability that it will be accompanied by tornadoes is 0.25. The probability occurring inthe county of Champaign wil hat tornado Strike the town of Urbana is 0.15. The probability of severe dan to houses in Urbana du rads) is 0.05; however, when tom town of Urbana, the probability of s wind storm (inthe absence of 3 county but do not ste the amage 0 houses is 0.10, w reas ifthe tomadoes nthe probability of severe damage to one of more houses in Urbana will be a With the above information, we define the following: F = severe damage to houses in Urbana $= occurrence of storm in Champaign County T = occurrence 5 in Champaign County during storm HY = a tornado striking the town of Urbana nd al P(S)=0.50; PUT|S)=0.25; PLIST) = 0.15 events are mutually exclusive and collectively exhaustive: STH, STH, ST,ST, whereas ST = ¢ nulls st ST Figure £227 Venn diagrara ofS, 7. H, and F> EXAMPLE 228 ding probabilities: D(STH) = P(H\ST)P(ST) = PCH\ST)PUT|S)P(S) 15 x 0.25 «0.50 = 001875 P(STH) = PCHIST)P(ST) = POHISTPCTIS) POS) PST) = PITIS)PUS 0.25\0.50) = 0.3750 PGT 0.01875 — 0.10625 — 0.3750 = 0.500% and the probabilities of damage to houses depending onthe above different conditions are FST) = 0.05; P(FISTH) = 1.00; PCE\STH) =0.10 Therefore the annul probability of wind storm damage to houses in Urbana would be F|STH)P(STH) + PUF\ST)P(ST) PurisTHy PST PUES T)PST 1.00 x 0.01875 +0.10 x 0.10625 + 0.05 x 0.3750 + 0 x 0.500 A tower may be subjected to earthquake loads which could be of high intensity (event 0) or ofl uation (event £). Ii estimated that ifthe oad has long dura is high is 0.7. Also, ifthe load has high intensity. dhere is 204% probability that it will be of shor nthe probability tha its intensity ‘duration. Finaly, the probability of having & long duration earthquake load is The designer estimated thatthe probability of failure when the tower is subjected to a short duration igh intensity earthquake is 0.05, whereas, his probability is doubled if the earthquake iso sng duration but low intensity. Also, hes certain thatthe tower wil faiif subjected to an earthquake (a) Are the events and £ mutually exclusive? (@) Ate the events H and L statistically independent? (6) Are the events H and L collectively exbaustiv Finally, calculate the probability of failure of this tower when subjected to an earthquake SOLUTIONS From the problem statements, we observe the following probabilities: PCH|L) = 07 PUIIL) = 0.3: ako, PUL|H)=0.2 and P(L|H)=0.8. Furthermore, P(L)=0.3 and P(E) = 0.7, Note that PU) is not give, bt it can be found from = PILE /P(LIH) PCH\L)P(L)/ PULA Also the conditional damage probabilities ar PUFLHE) = 0.05, PCFIL) =0.1, PUFIHL) = 1, and PUPA) =0 (a) Since PUH\L) =0.7 £0, H can happen given that L occurs. So H and L are not mutually23 Mathomaties of Probability = 63 () PUL i =1 (LIM): hence, Lis more likely to occur given the occurrence of H, so H and L are not statist (0.80, but the unconditional probability PCL) = 0.3 ally independ PH UL) = PUD) + PL) ~ PL = PUM) + PCL) — PL) PH) Which is les than 1.0, Hence, Hand L are not collectively exhaustive (@) Lat F denote te failure ofthe tower. The total probability of Fis obtained by summing the com mely HL, HL, HL, H L, which ae collectively exhaustive Hence, by applying the theorem of toca probability, PCF) = PCFIHL)POHL) + PCE\HE) PHT) + PCPA. : PCF|HL) PCL) PCL) + PCF | DPE) PU) + PCL) PHL) PCL) + 0 x PCT 0.10303 PCL) + PUR Dy T 235 The Bayes’ Theorem In deriving the theorem of total probability, Bq. 2.19, the probability ofthe event A depends ‘on which of the conditioning events 2. has occurred. On the other hand, fe may be interested in the probability of a particular E; given the occurrence of A. In sense, this is the “inverse” probability, which is given by the Bayes’ theorem, which may be derived as follows Applying Eq. 2.14 to the joint event AE, we have PCAJE)PUE:) = PUEVIA) P(A (ALE PCE Pula) = # 2.20) which is the Bayes’ theorem. In Eq, 2.20, if P(A) i& expanded using the total probabilit theorem, Eq, 2.20 become AE))PUE Yo Pale PE) > EXAMPLE 223 tes forthe construction of a reinforced concrete biking are supplied by two companies, «10 deliver 600 truck lols a day and 400 | ruck toads a day from Company b. From prior experience, itis expected that 3% ofthe material fr | Company a wll be substandard, whereas 1% ofthe material from Company b are substandard Define the following events: A = Aggregates supplied by Company a B = Agaregates supplied by Company b } E = Aggregates ae substandardrapter 2. Fundeme > EXAMPLE 2.30 als of Probability Models E|A)=0.0% and E) = PCE|A)P(A) + PUEIB) PCB) Howeve herem, Bg, 2.208, a CELA PAI P(A) + PCE|B)P(B) : ee tae ell to observe the difference between P(EIA) and P(AIE. The former is the proportion o aes from Company a tet are substandard, whereas the laters the probability that aggregal found to be substandard are from Company a. The diference is clearly significant < The Bayes’ theorem provides a valuable and useful tool for revising or updatin calculated probability as additional data or information becomes available. The followin examples will serve to illustrate this concept, including how prior information (which ma be based on subjective judgments) can be combined with test results to update a calculated probabil In onder rial used ina reinforced concre cylinders are collected at random from concrete mixes delivered to the construction site by a mixing plant, Past records of concrete fom the same plat show that satisfactory quality ‘ensure thatthe concrete delivered on site is of good quality the engineer require ich day be tested (after 7 days of curing) for minimum compres ‘one eylinder among those coll sive strength The test method isnot perfect—its reliability is only 90%, meaning the probability that tests 0.10. Define the following events G = good quality concrete T = aconerete cylin the test According tothe available information, we have PG)=080; and PUIG) = 090; PUTIG) = 0.10 is updated as follows iran PUIG) 0.90 x 0.80 ; PUTIG)PG) = PUT|GPCG) — 090 080+ 0.10 x 0 Therefore, with positive test result, the probability of good-quality concrete used in the constrctio s increased from 80% to 924 Concluding Summary 65 Now, suppose the engineer isnot satistied with just testing one cylinder, and requines that a second eylinder bet IF the second cylinder tested also gave a positive result, the probability that aan PCIE) i TIG)PG) + PCI PG 90 x 0.973 + 0.10 x 0087 7; = first cylinder is tested positive PUTTIG)PC POGITIT) = PU BIGPIG) + PO TIG)P n this second calculation, we implicitly assumed tha the 1wo tes are statistically independent; i P(TTIG) = PCTIG)PCEIG) a PUTAGPO PG|NT) = —— PURTAG PG) + PC 5 = 950x010 x 080-+0.10x 090x020 = "*? 7 aay 24 CONCLUDING SUMMARY : This chapter presents the basic fundamentals of the mathematics of probability useful fo modeling engineering and physical problems, The principles are all presented with example illustrating the modeling of physical problems of relevance to engineering. In particular, we learned that in the formulation and solution of a problem involving probability, two things are paramount: (1) the definition of the possibility space and the Jentification of the event within this space; and (2) the evaluation of the probability of he event. The relevant mathematical bases and tools useful for these purpo he clementary theory of sets and the basic theory of probability. The basic elements of both of these theories have been developed in nonabstract terms that should be more transparent . and more easily comprehensible to engineers and physical scientist Simple elements of set theory are used to define an event and its complement, and the operational rules of sets and subsets govern the combination of events, which cons of th \d interseetion of two or more events to form another event. Similarly based on three simple assumptions (or axioms), the mathematical theory of probabil provides the operational rules for developing logical relationships among the probabil e of different events within a given possibility space, which are basically the addition ri probabi n.and the Bayes” theorem66 vdementels of Probability Models : fundamentals necessary for corre Chapters PROBLEMS 2.1 Suppose the travel time between tw ites A and B by siris 6 or7 hrf the flight is nonstop; however, if there i on stop, the travel time would be 9, 10, oF 11 hr. A nonstop ight between A and B would cost $1200, whereas with one si iis only $550, Then, between cites B and C, all Sights a onstop requiring 2or 3 hours ata cost of $300, For a passenger ishing to travel from city Ato city C (a) What is the possibility space or simple space of his travel times from A to B? From A 10 C (b) Whats the sample space of his travel cost from A to C (€) IFT travel time from city A tocity C, and S=cost oft from A to, what is the sample space of T and S? 2.2 The setlement ofa bridge pier, say Pier 1s estima 1 between 2 and 5 em, Similarly the settlement of an adj cent pier, Pier 2, is also estimated to be between 4 and 10 em. Ther be a possibilty of differential settlements petween these two adjacent (a) What would be the sample space of this differential set {b) If the differential setlements inthe imple spa0e a ually Tikly, what would be the probability thatthe differential seitlement will he between 3 and 5 mi? 2.3 The direction of the prevail ac particular buildin tis between due East (0 =0 North (@ = 90"), wind speed V can be any value between 0 and 20 {a) Sketch the sample space for wind speed and direction V > 35 kph Ey = (1S kph < V = 45 kph Identify the events Ey, Es, Es, smd FE) within the sample space of Part (a, () Use new sketches to identify the following events: ENE; B=E\UE E,ME:nE Are the events A and! mutually exclusive? How about between the events A and C 24 A cylindrical tank is used 0 store water fr town as shown in the figure below. On any given day, the water supply iow perday may fill the tank with an additional 6,7, or 8 ft of wat The daily demand or consumption ofthe water for te town will applyin Jraw down the water level in the tank by an amount equivalen 105, 6,0r7 feof th he tank Cylindrical water tank, (8) What are the possible combinations of inflow and outflow ‘of water inthe tank in a given da ) Ifthe water le € ror the bottom atthe tart ofa day, what are the posible water levels inthe tank a ne end of the da {6) IF the amounts of inflow water For the tank re both xual likey, what ability that there woud be at least 9 ft of water remaining inthe tank atthe end A 20-f cantilever beam is shown inthe figure ad W, =200 Ib. or W or both may be aplied atthe mid Point B orat the end ofthe beam C. The bending moment induced {the fixed support A, My, will depend on the magnitudes ofthe Isat B and C 20-4 cantilever beam (a) Determine the sample space o (b) Define the following event Ey = (My > 5.000816) 1,000 = My < 12,000 fet Ae the events and Es mutually exchasive? Explaineee ee un 2 : 5 least 2 years the corresponding probability fr lane in route B (a) Determine the probability that route A will require major resurfacing in the next2 years, Do the same for route B. Assume atthe other lane of the same route will also need resurfacing 27. From past experience it is known that, onthe average, 10 of welds performed by a particular welder are defective. If welder is required todo three welds in ada 8) What is the probability that none ofthe welds will be dee (©) Whatis the probability that exactly two of the welds will be (6) What isthe probability that al the welds fora day are defee Its assumed that the condition of each weld is independent the conditions ofthe other weld. 2.8 On a given day, casting of concrete structural elements ata Pro 7 from a premixed conerete supplier. However, it is not always sertain that these sources of material will be available, Furth ‘more, whenever it rains atthe ste, casting anno be performed Ona given day, define the following evens £; = production of concrete material atthe jb sites Feasible Ey = supply of premixed concrete is avila ih the folowing respective probabilities: P(E) =08; PLE) =07; UE) = 085 nd P(EsiF:) = 04 whereas and are statistically independent of E fa) Identity the following events i tems of E}, Es, and E casting of conerete elements can be performed (i) B= casting of concrete elements cannot be performed on (b) Determine the probability of the event B. (6) If preduction of concrete material a the job site isnot feas Se, what the probability that eating of concrete elements ei sil be performed on a given day 2.9 construction firm purchased 3 tractors from a certain com pany. At the end ofthe fifth year, let Es, , Es denote, respec tively, the events that tractors no. 1, 2, and 3 are still in goo perational condition (a) Define the fllowing events atthe end of the Sth year, i terms of E, Es, and E;, and ther respective complements A = only tractorno. 1 isin good condition BB = exactly one tractors in good condition C = atleast one cto is in good condition, (@) Past experience indicates that the chance of a given tae ‘manufactared by this company having a useful ite longer tha 5 year (i. in good condition athe end ofthe Sth yea 660 one tractor needs tobe replaced (not in good operational con Sitio) atthe end of he Sth year, the probability of replacement for one ofthe other two tractors is 60s: i wo tractors need d the probability of eplacement ofthe remaining o salute the probabilities ofthe evens A,B, and C 2.10 A contractor has two subcontractors for his excavatior work. Experience shows that in 60% of the time, subcon avaiable 80% of the time, Also, the coatractor is ale 1 get a least one ofthese two subcontactors 90% ofthe time fa) What isthe probability thatboth subcontractors willbe avail able todo the next job (b) IF the contractor learned that subcontractor A sila for the job, what isthe probability thatthe other subcontractor68 > Chapter 2. Fundamentals of Probability () Suppose £, den ontrato @ Ate Ex and E (i) Are Eq ad Ey collect 2.11. An underground site is being considered forthe storage of hazardous waste, Within the next 100 years, there isa 1% chance thatthe hazardous material could leak outside ofthe storage con xinmeat. Twoadjacent towns, A and B, rely on ground water fo het water supply. The water fo each town will be contaminated if there is a leakage in the waste storage and if there exists a ‘ould allow the contaminant to move freely and quickly a im of sand exists between X and A XandB 1 cvent of leakage from the storage is independent ier the th 1 the presence of seams of sand. Con period o 100 years. inated river. as shown reflow (fh and. f 2 Towns A, By and Clie alr sling), The an towns A, Band flooding are 02,0 respectively. The events of flooding in each of the to By and Car wed eat, the probability that town B will also be loded flooded in given year, the probability that town A will ls flooded that year is increased t0 0.8, However, if town C does hotexpetience flooding ina given year the probability that both towne A and B will also not suffer any flooding in that year is (ay What A AD a) 243 sucessful cpletontaconstnctonrie dp the supp mals and bor wll a te Weather cond (a) Coot water ant lest bor or mater ar adequately al (Bat eater ut both bor and materi eae ve Det u ers Suppose ML) =0.%; UG) 06 and independent of Mand G. Ifthe we sd, whereas (a) Formulate pletion in terms of G.L and M. tb) Determine 1 cessfully comple isthe prob had been inade 2.4 The high ported for a Type of Accident (B) Run ino) Strck by Train sin Time of Day (D) x 0 Occuzrene Night (N 20 Suppose there are 1000 rail-highway grade-crossings i province XY. ‘2) What isthe probability that an accident will occur ata given() I an accident is reported to have oceurred in the dayt nat isthe probability that i is a “struck by train” accident? al and 80% of the “struck by train” accidents are fat is the probability that the next accident will be fata (4) Suppose D = event next accident occurs in tl A = event thatthe next accident is a “run into (i) Are D and R mutually exclusive? Justi Are D and R statistically independent? Justify. (Ans no 215 The promising alternative energy sources curently unde developmentare fuel cel technology and large-scale solarenerey power. The probabilities tht these two sources will be success ly developed and commercially viablein the next 15 years are respectively, 0.70 and 0.85. The successful development of thes two energy sources are statistically independent, Determine the follow (a) The probability that there will be energy supplied by these itemative sources inthe next 15 years b) The probability that only one of the two alternative ene 216 An examination ofthe 10-year record of rainy day for a town reveals the following 1, 30 of the days ae rainy days. 2. There is 4 50% chance that a rainy day will be followed ‘3 There is a 20% chance that two consecutive rainy days will be followed by a third rainy day A house is scheduled for painting starting next Monday for a Period of 3 da 1 E, = Monday isa rainy day E a nyd , = Wednesday i ruin xpress the ev responding to the three probabil es indicated above; ie 1,2, and 3, n terms of E, Es, E b) What is the probability that it wll rain on both Monday and day during the painting pe {d) What isthe probability that chere will be atleast one rainy day during the 3-day painting period 217 Mr. X who works in the office buikling D is selected for servation ina study ofthe parking problem ona college cam 's Fach day, assume that Mr, X will check the parking lots A and Cin tat sequence, ax shown in the igure below, and will park his car as soon as he finds an empty space. Assume also there are only these three parking lots available and no street Problems < 69 bilities of getting a space on each weekday moming in los A. ind Care 0.20, 0.15, and 0.80, respectively. However, i lt A is full, the probability that Mr. X will find a space in lt B is only 1.05. Also, if buh fos A andl Bae full, Mr. X will only have probability of 40% of getting a parking space in lt C Determine the follwing: (a) The probability that Mr, X will not find fee parking weekday mom (b) The probability that Mr. X will be able to park his ca smpus on a weekday morning (©) ITM X successfully parked his caro campusona weekday orig, what i the probability that his parking is free’ Mex ‘ate a i iA Bao (8 Blog. A study of eampus parking 2.18 A building may fail by excessive setlement of the foun dation oF by collapse of the superstructure. Over the life ofthe fof the foundation, the probability of superstructure collapse wil (2) What isthe probability that building failure will occur over its ite () If building failure should occur daring its life, what isthe obability that both Failure modes will occu (€) Whatis the probability that only one ofthe wv failure modes will occur over the life ofthe building 19 The automobile brake system consists of the following ponents: the master cylinder, the wheel cylinders, and the ke pads. Failure of any one or more ofthese components will the brake system, Within a period the master cylinder, wheel eylinders, and brake pads are 0.02 9.05, and 0.50, respectively, The probability that both the mast cylinder and the whee! cylinders will al within the same brak independent of the nd wheel cylinder a) Whatis the probability that only the wheel cylinders will fail within 4 years or 50,000 miles’ (&) Whats the probability thar the brake system will fil with 4 years or $0,000 mie (©) When there is failure inthe brake system, what isthe prob ilty that only one ofthe thre components failed70 > Chapter 2. Fun rnd denote te events of excess amentas of Probability Models PE (b) Whats fe) What cho (@) Mahe ps PuE P(Ex|Ey) PUESIEs F2), PCE Hin Start out with the rid * ability that the subsequent winter will no lished a) What that the summer (©) What is the probability that there will beat least one hot obability of 040%; and (ii) 2.22 A metropolitan city derives is clectrieal powe erating plants A, B, and C. Each plant hus a ge megawatt). The chances that th pacity of 50 MW Plants wil be shut down (for periodic maintenance Toad, socidents, 4c.) in any given week are, respective! EXE?) The operations of plants A and B are related in > overload, whereas the operation of plant C i in a severe storm, lightning knocked out the other to pl (a) During What isthe chan (b) What isthe put dvr dy dependent of thatthe city will suffer a complete blackout caused by the storm! habit that the city will ha (6) Whatis the probability thatthe available power supply tou tity is less than or equal to 100 MW in a given w 2.28 The probability thatastrong earthquake will cause da (a) What isthe probability thatthe structure can survive three () What is th thatthe siucture will be damaged A team of two engin and B, was assigned to check The two work simaltaneously but sep ineer A spttin Set of computations ately anc independently. The probability of e a given entor is O8, whereas that for B is 0.9 that it was discovered by A alone? (Ans. 0.082) (6) Suppose there isan alternative team consisting of three en ers C;, Co, and Cs. each of whom works separately and independently and has a probability of potting a given ror, Would this team of three engincers be selected instead. Please justly. (A {@) Im part (a, suppose there were two tions. What i the probability that both errors will Bes by the eam of two engincers? Assume that the events ing between any two errors ae statistically independent. (Ans. 2.25 A self-standing antenna diskis suppor atice tue (b) The probability that nodes will occur during a wind stom () Ifthe disk is damaged during a wind storm, what isthe prob 2.26 The probability ofa severe fre (denoted event F) occurSensitivity, it may also cause fase alarms with a probability of PIA|F)=0.1. Assume that there is no possibility for more than (a) Listthe setof mutually exclusive and collectively exhaustive (@) Calculate the probabilities for each of the event fisted in (6) What isthe probability that the alarm system will be trig (4) What is the probability of @ real fre given that the alarm D) during one’s lifetime is very small, with A tal. Fortunately, modem medical science has provided a diagnos le testF to detect the presence ofthe disease: however, the testis rot always corect. IF person has the disease, there is only 85 robabiity that the test willbe positive... PUT|D) = 0.85. Al ere isa small probability of 2% thatthe test willbe positiveeven when a patient does not have the disease, ie, (7|D) =0.02. (a) Identity the set of mutually exclusive and collectively ex: (b) Calculate the probabilities foreach of the events identified (€) Whats te probability of a positive tes result (@) If a person's test result is positive, what is the probabil that he/she has the disea 2.28 A car is traveling from point M to point Q according to the route indicated in the figure below, The driver has to pass through thee intersections, namely at A, B, and C where tai hs are installed. Information on the traffic lights relative to the driver is as follow 1, The light at A is equally likely to be either red or green The ight a Bis equally likely tobe either sve, ifthe driver encounters green at A, he will meet a green light at witha probability of 0.80. 3. The driver will encounter a red or green or lft turn only at lights at C ae statistically independent of those at A and B, and let tums at € are permitted only when the left turn onty signal ison : w| | N A = a signal at Bis green; and Ges = left tur onty signal is om at C, define the following events ts between Mand N. Problems < 71 y= stopped by signal at C from M10 Q stopped by red signal once between M and N. (b) Compute the probability that the driver willbe stopped by traffic signals atleast once traveling from M to Q. (6) Compute the probability thatthe driver wil be stopped by traffic signals at most once going from M to N. rom previous records on winter weather for town, the probability tha it snows on a given day is 0.2. Records also re ‘eal that low temperature (say less than 0 F) occurs on $6 of th days, whereas windy days ooeur 109 ofthe time. I temperatute is low, the probability of snow on that day is inereased to 30 Let S,C, and W denote the events of snow, low temperature and win, respectively ona given day, The event W may be assumed obe statistically independent ofS and C (a) What i the probability of having a doomsday (i... snow Jow temperature, and wind all occurring onthe same da)? (Ans (b) Suppose a construction project cannot progress if itis wind 1 COM, but itis not affected by stow. What isthe probability tha construction work will e stopped on a given day? (Ans. () Onaday without snow, what is the probability that construc jon work cannot proceed” (Ans. 0.139 (@) Wha is the probability ofa nice winter day (i.e, no snow. 10 gusty wind, and no low temperature)? () Suppose U denotes the event of uncomfortable wind chill condition ona given day. If either Cor W (but not both) oecur the probability of U is $0%, whereas, ifboth Cand W occur the probability of U is 10%, and U will not accu i both C and W to not occur, What isthe probability ofthe event U on a given 2.30 Lead and bacteria are the 1wo common sources of com tamination in a water distribution system, Suppose 4% of the (a) Determine the probability that a water distribution (b) Ifa system is indeed contaminated, what is the probabil 1 itis caused by lead ony? (Ans, 0652 2.31 The 15-ft diameter tank shown below resis on @ concrete hase. When there is water inthe tank, it will be a the 10-8 20.f level. and the probability of either level is 40. The weight the tank is 100,000 Tb and the weight ofthe water is 11,000Chapter 2. Fundamentals of Probability Models (a) If Cis equaly Hikely to be 0,10 or 0.20, identity the sample pace ofthe fictional resistance to horizontal sliding ofthe tank (b) Ie the total maximum horizontal force during a wind storm 5,000 Ib, what i the probability thatthe tank will be dis placed (slide) horizontally during the storm? The value of C is Statistically independent ofthe teal weight () Suppose that the total maximum horizont durin 1 tank n to prevent the contaminants leaking from = soil stratum resulting from rainfall infiltration. layer of ‘thetic material called geomembrane was also placed above he lay material to provide additional protection agains leakage ontaminants, Nevertheless, the quality of workmanship dr clay might have been compacted poorly second, the geomem brane might have holes punctured by sharp stones that were n Jetected ding inspection. Moreover, extremely heavy rainfall ould happen during the operation ofthe landfil, which could vduce excessive pore pressure on the geomembranelclay layers, —_. ‘Sal trun agineer inthis case believes that Teakage will happen “dur ing extremely heavy rainfall, and ether the clay was not we pacted or there were holes in the geomembrane” (event I Leakage could also occur “under ondinary rainfalls (Le, withou Let W = event of well-compacted clay ad this occurs wth 90% probability H = event of geomembrane containing holes; ad this is 30% likely E = event of extremely heavy rainfall; and the lihood is 20 Te quality of construction has no effect on the future amount of rainfall. However, if the geomembrane contained holes probability ofa wellcompacted clay is reduced to 60 (4) Express event [in fers ofthe symbols defined above. Ri peat for event I {b) Determine the probability of event I. Repeat for event Ih (el) Ate Wand H mutually exclusive? Are they statistical (2) Ave the events [and I] mutually exclusive? Are they col lectively exhaustive? Ans. yes, no) (@) Determine the probability of leakage for this landfill eon 2.38. A community i supply 0 he coming winter. Suppose there are three major sources of e ergy forthe community, namely electrical power, natural gas, und oil Let E, G, and O denote, respectively the shortages of these sources of energy forthe next winter, Also, iis estimate thatthe respective probabilities oF these shortages areas follows: shortage of gas may be assumed 10 b of shortages in cil and electric 2) What isthe probability that there wll be shortages inal the three sources of energy next winter? {) What is the probability that there will be shortages in atleast ne of the following sources next winter: gas and electric 2.34 An acral reconnaissance system consists of three remote components over a period of 10 years are, respectively (0.03, and 0.02, whereas operating a ultrahigh altitudes, the cor responding failure probabilities would be 0.07,0.08, and0.03. In {607% ofthe time, the reconnaissance system will be operating at ormal altitudes, and 40% ofthe time twill be used at ultrahigh failure probability of component B will be twiee itsprobability. On the other hand, th ‘omponent Cis silure of compo Midwest ci ‘ofthe winters inthis ity Moreover, 40% ofthe 2.38: Inany given year, the winter in ‘ean be cold (a) Are the events C and W statistically independent? Justi b) What is the probability of an unpleasant winter in given {e) Whats the probability that the winter in any given yeur w 2.36. city commuter can take one ofthe ble routes AB, oF to work, On & given weekday mornin rush hours, thechancesthat routes A, and willhave allie congestion, H, are, respectively, 605, 605, and 40%, Routes A and Bare close by each other such that ithe trafic is congested in one route the hance of congestion inthe other route increases to 85%, whereas he condition in route C is unaffected (i... independent) by the raffe conitions in routes and B. Als, ifthe traffic in all the ate, L, for work is OU, otherwise it will be 30 (a) What isthe probability that exactly one ofthe three routes il be congested on a weekday mornin () What is the probability thatthe commuter will be late For work on a given weekday moming (e) If i is known that the tafe in route A is congested, hoy £37 The damage ina structure after an earthquake can be cl 1 (L) or heavy (H), For a new undamaged hat it will suffer ight and heavy dam structure, the probability ges after an earthquake is 0.2 and ier two earthquakes? Assume that no repair arthquake. (Ans. 0.188) heavily damaged (b) Ia structure is indeed heavily damaged after two earth quakes, what isthe probability tha the structure was either un Jamaged oighty damaged before the secondearthquake? (Ars hr carthguake, what is the p ring thre earthquakes? (Ans. 238 1 tity may be low (L), average (A), oF high (I) w robabilties of 0,60, 0.30, and 0.10. The various suppliers of Problems «73 tithe demand is average (A) or high (2), the supply ma (¢) What isthe probability of shortage of cement in at leas, pnseeutive months 2.39 Pr project ste. Suppose that on ste, Due to etrors in the fabrication process of the panels it is amber of defective wal panels i a shipment When the wall panels are delivered, they are inspected by he supervisor at the constraction site, The supervisor will accept the entre shipment if at most one panel is found to be detective otherwise the shipment is rejected. {a) Determine the probability that «shipment will be accepted (b) What is 1 probability that exactly one shipment wil bs sting of 5 working days? {6) The inspection procedre, however, is aot perfect it is onl 80% likely that a defective panel will be corey identified 2.40 in the labor market, thete is only @ 50. 50 chance that he can successfully launch reais predicts that the relative likelihoods of good, normal, and tad weather conditions in the next 2 months are 1:2 (a) What is the probability that the completed on schedul wire probabil project was not completed on schedule, what isthe 2AL A county is examining its energy situation for the nex a its principal sources of energy. Enengy expert in the country testimated that there is @ 40% probability that gas supplyChapter 2. Fundamentals of Probability Models supply is 206. However. iil supply slow, the probability that other loading ‘288 supply will also be ow is increased to 50 The demand for energy in the next 10 years is also pr pers estimated that the probabilities of Tow, _ditionsreative tothe fatigue stresses in be 03,06, of load and lane postion Five that a vehicle isi lane B, what isthe prob 110.1, respectively. Depending on the level of energy demand ¢b) itis fully loaded (event crisis" might occur aecording to the following table (©) What is the probability of the oocurrence of critical faigu stresses in the girde Tow fd) Ira critical stress is observed, what i the probability tht it Ene Occurrence of as caused by a vehicle in lane A’ Demand Oil and Gas Supply nergy Crisis ion Sais ae 2d There tno pose meso eof eine Noma’ /yArleai gana Ves (UC ea any so alas, whe ow ae sieeiosret nen scat Xe ey without warning. and bending faire which is preceded pale large defection, Experience indicates that 5% of al failures o RIC beams ate by the shear mode andthe est by bending mod (a) If there is normal energy demand inthe next 10 years, Wht Lghoratory tess show that 80% ofall shear failures exhibit d the probability ofan “energy crisis agonal eracks a the end ofa beam prio o failure whereas onl Gb) What is the probability of an “energy crisis” in the next 108 of bending failures show such diagonal cracks. 0 year (8) What isthe probability that an R/C beam will show diagonal i racks before failure ) Altera severe earthquake, upon inspection ofan RIC bu ofthe beams 2.42 Inan offshore area in the Gulf of Mexico the pro spetvely, 20% and 5%. The probability of morethantwo #88, a9 engineer found cracks atthe end of eng Fern rr pmucton pao ni the ailing. Shouk! he recommend immediate repute If the hurricane occurs only once in a given year, the annual Shear failure is likely to occur o that failure probabi xceed probability that both substrctres will not sufer any damage 78 Eo oy dp 9 MUPENSS 345 ate gls war ot mnt i teeta , vo one-way sees, Suppose 85% ofthe vehicles will decelerat Ta is uate tiie cfc (ated Oe nO re eal cele tne buncane ina given year Assume the damages othe WO feet of nove who aeelerted wl eentaly ran substructures are statistically independent, W! ould be this: pang ong rar arE eee igh, and only 2% ofthe “indcisve™ drivers will ef probability if wo hurricanes shoul occur in the given year Typ What is the probability thatthe platform vail suffer some nS ht hurricane damage next year’ cae tion, what isthe probability tha it will un te rea ig (orf petaily train RAAGHaeswcctrvod tye? NON NURS ae Uh ar wed tat, it a 2.43 The common fve-axle semitaler trucks dominate the probability thatthe driver had acelerated ra tock taf overa bridge that has been selected Torstrac- (e) The likelihood of an accident resin fom a veil red light (referred to asa problem vehicle) i studied as tural monitoring and testing. For simplicity, the heavy vehicles it Tal ful (HD, or fulslouded (F).Only follows, Suppose in 60% of the time, vehicles are waiting onthe aoe caenlcatatime can be bservedon the bridge at any instant, other stretat the start oftheir green ight cycle realy to cross the rae reams be in ether of two lanes, lane A or lane B. The likeli intersection. Most of these drivers, say 80%, are cautious before je times greter than in lane they entered the intersection, whereas the res are ot cautious B. The lane in which a truck crosses the bri {a problem vehicle in the intersection of the load it carries a cautious drier can avoid the problem vehicle 95% ofthe time ‘Of concern are stresses in a girder under lane A. Ifa fully whereas 20% of the noncautious drivers will collide with the loaded truck, F. crosses the bridge in lane A, a critical fatigue problem vehicle. What isthe probability that x problem vei stress is certain to occur, whereas ifthe same truck were ro cross will lead to an accent {d) Suppone the annual trafic flow in one of these on Tane B, the likelihood ofa eitcal stress is reduced to 605 100,000 vehicles and 5% of them would encounter The likelihood that ahalf-loaded truck, H, will produce acriical street A, but only 10% when in ane B.'No amber signal light. Estimateinthe intersection that would be traced back to vehicles running hrougha red light on that ste 2.46. major construction company his three branches A, B, J.C operating in different parts of the country, The chan: the branches willbe profitable in any given year are 70 re related such that if one makes a profit the probability of the other brunch also making a profit inreases t 90%, whereas branch Cisindependent of both A and B, Atthe end ofeach year, ifatleastrwo branches ae profitable, the chance that employee 8) Wha isthe probability that exactly two branches will make () Determine the pr that the company employees wil (6) Ifit is known for sure that branch A will end this year inthe ed (i.e, not mak pro), how would this fat change the probability of pat ( 247 A contractor finds that some of his building projects in lve difficult foundation conditions, and others are considered easy His curren projects rein three counties lino; na in Champaign, Ford, and Iroquois, The contractor has statistic 1m previous jobs in these counties which enable him to mak the following observations with confidene 1. The probability of any buikling project picked at random will havea difficult Foundation problem i 23 2. 1/3 of the projets are in Ford County 3. 25 of his projects are in Iroquois County and involve dit 4. OF all his p rd County, 50S imoive difficult oundation comalitions 5. The events of a project in Champaign County and the fou tion condition aze statistically independent a) What is the probability of having the next project in Ford County that has easy foundation conditions () What isthe probability of having the next project in Cham: (©) If we know only that his next projet will have easy foun: tion conditions, what isthe probability that it i in Iroquois 2248. The three major loadings ona nuclear power plant, as far ax safety is concemed, ae those due to severe earthquakes (F), loss coolant accident (L), and thermal transients (T). Fora typi cal plant, 1s of occurrence of E, L, and T in a given year are respectively, 0.0001, 0.0002, and 0.00015. Also, severe arthquakes sometimes cause L duet pipe breaks; itisestimated that the chance of L increases to 10% if severe earthquakes oe ithe same year. Tisassumed tobe independent ofboth E and L a) What is the probability tha in a given yea, all three types tb) IFat least one of the major loadings occurs, the plant has to be shut down fora period of time and the utility company w eeeeeeee eee ee ere ceeeeereee ve cece eee cre Problems < 75 Suffer some loss of revenue due to power outage. What is th obability ofthis loss in a given year (6) Irthe wility company incurs this loss of revenue in a given Year, what is the probability that severe earthquakes occurred 2.49 Ata construction project, the amount of material used ponding probabilities of 0,60 and 0.40, If the amount of ma al eoquted in a day is 100 units, the probability of shortage o material is 010, whereas if the amount of material required is units, the probability of shortage of materials 0.30, a) What isthe probability of shortage of material in a given () If there is a shortage of material in a given day, what is th y thatthe amount of material required that day is A space vehicle is designed to land on Mars, Assume that und condition on Mars iseither hard or soft. IFhard ground encountered during landing, the vehicle will be suceesslly landed with probability 0.9, whereas, if soft ground is encoun only 05. Based onthe available information, itis judged that the (a) What is the probability of a successful andi before landing. It will penetrate inta soft ground with probabil were observed to penetrate into the grou, what is the proba bility thatthe ground is hard? (Ans. 0.4) () What isthe probability ofa successful landing if the stick penetrated the ground? (Ans. 0.66) 251 Cement and reinforcing see (rebars) are essential mat onstucting a reinforced conerete building. During co tion, the probabilities of encountering shortages of thes terials (e-., caused by sirikes at the factories) are, respec tively, 0.10 and 0.05, However, if cements avaiable, the proba bility of shortage of rebar is reduced to one-half ofthe original probabil (a) What isthe probability that there will be shortage of ithe ported by tnucks remaining 40% by probability of 0 ry by tucks is 076 > Chapter 2. Fundamentals of Probability Models (@) Whats the probability that materials tothe construction site willbe delivered on schedule (@) If there is delay inthe transportation of construction mat Water for a cit A and Source B. During the summ thatthe supply from Source A will be below normal is ihe corresponding probability for Source B is 0.15. How B will also be below normal during the same summer season is The probability of water shortage inthe city wll obviously lepend on the supplies from the two sources. In particular, i ‘only Source A is below normal supply, the probability of w stage is 0.20, whereas if only Source B is below normal the corresponding probability of shor (4) If water shortage should occur in the city, what is the prob itty that it was caused by below-normal supplies from both what i the probability there was normal supply from Source A (1) Field work—18 wea ability that the field work wil be completed on schedule is 0.90 (2) Computations Two independent computers are available to perform the required ealeulations. Each computer has a reli bility of 70% (ie, the probability of working i 0.70, If only fone of the computers is working, te probability of completing 1 computations on time is 0.60, whereas, if bath computers “orking, this probability inereases to 0,90. Furthermore if bot ofthe computers are not working, the engineer mus perform his calculations using desk esleulators which are 100% reliable, but n schedule (b) What is the probability thar the computations will be com 2.54 Ata rock quarry, the onto a truck sequal like numb trucks in a queue siderably: data f The time required 10 load the queue size. (a) It there are two trucks the quary, what isthe probe (b) Before arriving atthe g ofthe queue upon artval, w arr, and thas not knowing the siz hat isthe probability that the waitin ime ofthe truck wil be less than S minutes’ trucks. Suppose the bridge sneineachlane, The event two trucks are present sim Suppose 10% oft load limit) and the event nt between trucks. The when both trucks ae ‘only one is everloaded and loaded fa) What isthe probabil porting two truck of possible damageto the bridge when racks are overloaded (ic. above legal iwerloading is statistically indepen ‘damage probability ofthe bridge i overloaded; the probability is 8% if trucks) (int: Determine istBB Return to pat (a). Suppose the coumy boa anal ‘quency of trucks such that the fraction of overloaded trucks en brid 2.56 A geol 4 underneath a site could Tikelihood that anor ‘One plan cals forthe use of geophysical techniques. If anomaly s presen, such techniques will have 50% probabil fof detecting the anomaly; otherwise, no signal wil be registered (a) IF the geophysical technique is used but it failed to detect ny anomaly, what isthe probability that the occurrence of a nomaly is still possible atthe site? (Ans. 0.176 () At this point, a more discriminating plan is used such that the probability of detecting an anomaly is as high as 80% if an anomaly is present. Suppone this new plan also did not detect i) How confident isthe engincer now about his claim thatthe 7 i) foundation system will be built at he site. The engineer mates thatthe foundation should be 99.99% safe if there sno anomaly. However, if an anomaly exist, the reliability of the foundation is reduced to 80%, What i the probability failure of this foundation system? Ans: 0.008) (i) Suppose failure of the foundation system could bring a loss of one milion dollars, whereas survival of the foundation 287 Past records show that batch of mixed concrete supplied by a certain manufacturer can be of good quality (G). average ality (4), bad quality (B), with respective probabilities of 130, 0.60, and 0.10. Suppose thatthe probability of failure of reinforced concrete component would be 0.001, 0.01, oF 0.1 de pending on ality of concrete is ood (A), or bad (B), {@) What would be the probability of failure ofa reinforced con crete siuctural component cast with abatch of concrete supplied by the manufactur () A test may be performed to give more information on the 1 passed the test Problems <7 (i) What is the probability that it will be of good quali i) In this ese, i, concrete passed the test, what would be the probability of fulure of a siactual component east rom this batch of coner 2.58 The maximum intensity of the next earthquake in a eit may be classified (for simplicity) as low (L), medium (M), or jh (H) with relative likelihoods of 15:41, Suppose also th buiklings may be divided in poorly constructed () city are known to be poorly constructed for earthquake resis spectively. However, a well-constructed building will survive low-intensity earthquake, although tmay be damaged when sub of 0.05 oF 0.20, respective (a) Whats the prot be damaged during the next earthquake? (b) What proportion ofthe buildings inthis ety willbe damage during the next earthquake (6) Ifa building in the city is damaged after an earthquake, what isthe probability thatthe building was poorly constructed 2.59 A transit system consists of one-way trains running be a stations as shown in the igure below. The distance conceming tion of passeng Destnai Onin 1 2 3 4 2 06 0 3 1 3 0s 0. oa orexampl riginating from Station | will get off at Station 2, 3, probabilities 01,03 and 0.8, respectively thermo in of tips originating from Stations 3, and 4 are 0:25, 0.15, 0.35, and 0.25, respect Intersttor ‘tance in miles oS (a) Wha78 > Chapter 2. Fundamentals of Probability Models (b) What i the expected trp length for a passenger boarding at ion 1 (Note: “Expected value of X= Baik py where pi senger trips will excoed 10 miles {€) What fraction ofthe passe he taint Station 3 originated from Station 1? (Ans: 2.60 Three machines A, B, and C produce, respectively, 6 pust records, the percentages of defective outputs ofthese ma Hines are, y. 24% 3%, and 456, An item is selected a random and is found to be defective (a) What isthe probability thatthe panicular defective tem was roxiuced by machine A? (Ans. 0-48 (y) What isthe probability that the defective item was produced 2.61 An engineering Company E is submitting bids for two projects A and By the prababiiies of winning are estimated tobe, respectively, 0:50 and 0.30, Als, if the eompany wins i. his chance of winning the other bid is reduced to one-half ofthe orginal probabil {a) What isthe probability of Company E winning at least on fits ids 0) If Company £ wins at last of one of its bids, what is the probability that it will befor project A and not project B {e) If Company Fis awarded only one project, whats the pr ability that it will be projet A Furthermore, on the basis of past performance, itis estimated that the peobabilty of Company £ completing project A within project, the probability of on-time completion of project A fd). What isthe probability that project A will be completed on (6) Ifthe project A is completed on time, what i the probability 2.62 Defective construction project can be caused by either poorag ‘kmanship such as selection ind prading of material, pouring curing) or both The quality o rates is also affected by the quality of workmanship, and (On a given project, the probability of poor aggregates i (0.20. The probabil regates are poor is 0.30, The probability of poor a 1 poor work regates if there is poor (a) Whats the probability of poor workmanship onthe project! (b) What i the probability of atleast one of the causes of de rete occurting onthe project (¢) Determine the probability that only one ofthe two possible {@) Inthe above project, if theres only poor aggregates and not poor workmanship, Tr ther i only poor workmanship but good aggregates, the cor responding probabil ure present, the probability of defective concrete would be 0.80 if none of the causes is present, the corresponding probabil i 00S. Determine the probability of defective coperete on the (e) lthere isd te onthe project, what isthe prob ability that 2.63 The structure shown in the figure below could be subjec fo setlement problems, The likelihood of having a setiement tem (event A) depends on the subsoil condition, in partic Tar whether weak zone exist in the subsoil or not. If there i Small weak rone (event S), the probability of A is 0.25 ifthe ‘weak zone is lage (event L), the probability of A becomes 0.6 iscnly 0.05. Based on their experiences withthe geology ofthe neighborhood andthe soil information from the preliminary sit is 470% chance of no weak zane in the stratum under! sr if there is a weak zone, it would be twice as Tikly to be small than la > Large or small? (a) What isthe probability that the structure will have a ment problem? (Ans. 0.1 {b) Suppose an alitonal boring ean be performed a the sitet thermore information about the presence of the weak materia The engineers judge that Ifa large weak zone exist, itis 8 likely tha the boring wil encounter it: this encounter probabi ity drops to 30% fora small weak zone. Obviously, the boring will not encounter any weak material if the weak zane does not exist at all Suppose the additional boring failed to encounter any Ans 0.16 vil ‘Whats the probability of small weak zone Tn this case, what isthe pr tractare 1 settlement problem? Ans. 264 Adam seismically active are a Neste Dam of),rl ss il bei Two regions, A and B, can be identified in rhguakes in either ares could eause dma Jam, Earthquakes occur independently betw B. Suppose the annual probabilities of earthquake occurrences in regions A and B are 0.01 and 0.02, respectively, Moreover, the hance of (wo or more earhxjuakes occurring annually in eat (b) If an earthquake occurred in A (but notin B), the likelihood st notin A), the likelinood of damage is only 0.1. Further: or, if earthquakes occurred in both regions, the dam 9 nave a 50-50 chance of damage, What isthe probability that (@) Suppose the dam can be relocated close to the center of regi A, such that earthquakes in region B will not cause any dams tothe dam, However, the likelihood of damage due to a earth w location if the objective is to minimize the probability of incurring damage? Please substantiate your answer (4) Would the decision in part (c) ehange if the new sit also susceptible to: (i) a landslide caused by severe rain Storms with an annual probability of 0.002, and (i) a 0.001 nnual probability of subsidence due to poor supporting sub oil structures? Explain. Assume that th be dam J during landslide or subsidence, Als, the events of dara ised by earthquake, landslide, and subsidence are statistical depende 2465 Three oil companies, A, B, and C, are exploring for oi ea. The probabilities tha they will discover oll ae, re spectively. 040.0. 0. IF B discovers ol, the proba hat A will also discover ol is inereased by 20%, whereas es affect the chance of € discovering oil (i... C is in endent of B). Moreover, assume that C i aso independent ib) If ils discovered inthe area, what fs the probability that wll be discovered by Company € (| What isthe probability that only one of the three compan 2466 Delays the siport are common phenomena forairtrave +. The likelihood of delay often depends onthe weather cond ion and time ofthe day. The following information is available Ten the morning (4%), lights are always on time if good weather prevails; however, during bad weather, hal of the fights 4, Bad weather is more likely in the morning : the mornings ae associated with bad weather, but only 10% of Pa hours are subject to bad weathe Assume only ovo kinds of weather, namely, good or ba > = Dela = Bad weathe Answer the followin (a) What fraction of the lights at this ieport willbe delayed (Observe that thisisthe same asthe probability tha given Might willbe delayed. (Ans. 0.282) (@) Ifa ight is delayed, what isthe probability that itis caused by bad weather? (Ans. 0.330) (@) What fraction of the morning Nights at this apr wil b delayed? (ans. 104%) pany can be the resultof one ofthe following independent cause 1, Malfunction ofthe machinery, which occurs 5% ofthe pr ction im 2. Carclessness of workers, which occurs 8% ofthe production Defective articles are produced only by these twocauses. Also, if ‘only the machinery malfunctions, the probability ofa defect tice is 0.10, whereas whenever there is only carelessness of workers the probability of a defective ace is 0.20. Howeves 2468 Leakage of contaminated material is suspected from a ge has occurred, The location of two wells, Asn B ae shown or Lanafilt, If teakage has occurred, it will be observed by Well A with 80 probability, whereas Well Bis 90% likey to detect the leak Assume that either well wil not register any contaminant if there sno leakage from the indi. Before the well reinstalled, th80 > Chapter 2. Fundamentals of Probability Models mngineer believes that there is « 70% chance that leakage has {@) Suppose Well A has been installed and no contaminants were has occurred? (A Assume thatthe (b) Suppose both wells have been install Fident isthe engineer in concluding that no leakage has oc aed? (An (6) Ifthe cost oF instal Well isthe same as Well B, and budget is sufficient forthe installation of only one well, which a given community, the probability of is drinking water con unt of pollutant is 0.1, wher REFERENCES pollutant problems from drinking the water? Assume tha person's resis: tance to pollutant B is innate, which i independent of the even of having excessive pollutant in the drinking water 2.70 A contractor submits bids to three highway jobs and 0W0 building jobs The probability of winning.each job is 0.6. Assume ‘one job? (Ans. 0.08 b) Whats the probability thatthe contractor will win atleast a jobs? Ans 0.913) {€) What isthe probability that he or she will win exactly on highway job, but none ofthe building jobs? (Ans. 2.093 Analytical Models of Random Phenomena 3.1 RANDOM VARIABLES AND PROBABILITY DISTRIBUTION In Chapter 2, we developed the fundamental concepts and tools for defining the oceur rences of random phenomena and the determination of the associated probabilities. In this 3.1.1 Random Events and Random Variables A random variable isa mathematical vehicle for representin leterministic variable that can assume a definite value, the value o values. If X is defined as a nis an event, wher x variable may be defined within a range of po variable, then X = x, or X x, re range of possible values of X. Intuitively, a random variable may be defined mathematically funetion that transforms (or maps) the 7 sts in & possibility space into the number cal sciences, many random phenomena of interest are associated with the numerical outcomes of some physical quantities. In several of the e amples we illustrated earlier in Chapter 2, the nun somplete a pr mean flow level are all outcomes in numerical terms. In the other examples, we also illus bulldozers that remain ope and the flood of a river after 6 months, the time requited to problems in which th in, the degree of completion ot Therefore, the possible out andom phenomenon can be represented bj americal values, ether naturally ned artificially. In any case, an outcome or event the value of range of values of a function, which is ealled a may then be identified th om variable. We with lowere all denote a random variable with ter, and its s. The value, of range of st82 > Chaptor3. Analytical Models of Random Phenomena EUE SX mapoing Figure 31. Mapping of events in sample space Sint real lin represents the possible states of a chain (as described in the paragraph above), then X = failure of the chain. In other words, a random variable is a mathematical device for mth, in terms of the random variable identifying events in numerical terms. Hene speak More formally, a rancdom variable may be considered as a mathematical function or rule an event as (X = a), oF (X a), and : E\E,=(e EXAMPLE 31 1 Random Variables and Probability Distribution 85 ofa random variable must necessarily satisfy the axioms of probability theory, as described carlier in Sect. 2.3.1. Accordingly, the function must be nonnegative and the probabilities fords, if F(x) is the CDF of X, then it must satisfy the following conditions: @ Fy) = and Fx(20) ii) Fy(x) 0, forall values of x, and is nondecreasing with Conversely, any funetion possessing all the above properties can be a bona fide CDF By virtue of these properties and Eqs. 3.2 through 3.5, the PMF and PDF are nor functions of x, whereas all the probability masses of a PMF add up to 1.0, and the total are under a PDF is equal to 1.0. mixed random variables. We observe that we ean write Eq. Similarly, fora discrete X, we have a EXAMPLE 32 Figure Eth PMF of X ° awe 3 4% Figure Ee CDF of X. < hely to be placed anywhere alon 1 ofthe beam of 10 m, then the PDF of the load position X is uniformly distributed in 0 < x < 10; thats, here ¢ is @ constant. AS the area under the PDF must be equal to 1.0, the constant © = 1/10 Graphically, the PDF is shown in Fig. E The corresponding CDF of X would b F 7 O 10 Graphically the above CDF is shown in Fig. Eeee eee ee ee ee ae Distribution «87 1 Random Variables and Probability e110 Figure E328 PDF of X Fun} ° Figure £326 CDF of X ition ofthe load willbe between 2 m and 5 mon the beam is PQ EXAMPLE 33 The useful life, T (in hours) of wel in which isa constant, Graphically this PDF88 > Chapter Analytical Models of Random Phenomena Mean ofT y 250800 Figure E332 Exponential PDF of useful lite 7: L L 080 ~+100~«180~—«200 Figure E33b CDF of useful life 7: The corresponding CDF would be F(t) = fe 16% as shown graphically in Fig. F330, < 31.3. Main Descriptors of a Random Variable Te probabilistic properties of a random variable would be completely described if the form its CDF or PDF (for a continuous random vai or its PME (in the case of a discrete random variable) and its associated parameters, are ecified. In practice, the exact form of the probability distribution function may not be known; in such cases, a riptors can be useful and sometimes may be nec Moreover, even when the distribution function is kno useful, because they contain information on the properties of the random variable tht a first importance in many practical applications. Also, the parameters of a known distribution function may be derived as functions of these principal quantities, or in some cases may be neters themselves (see Chapte Central Values As there isa range of possible values of a random variable, som ve ofthis range.3.1 Random Vera 1 and Probability istbution < 89 he different values of the random variable robability densities, the “weighted aver associated with different probabilities or 120” (i.e., weighted by the respective probability measures) would be of special interest: this weighted average is the mean va pected value of the random variable. Therefore, ifX isa discrete random variable with PME, py(), its mean value, denoted E(X), is EX) = Ds whereas ifX isa continuous random variable with PDF, fy), its mean value is BX) = | xfetondx (3.76) (Other quantities that may be used to designate the central value ihe mode, of modal value, and the median. The mode, which we shall denote as X. is the most probable value of a random variable; ie. it is the value of the random variable with the largest probability or the highest probability density. The median of a random variable X, which we shall denote as sp, is the value at which the CDF is 50% and thus lar ally probable; that is, and In general, the mean, mode, and median of a random variable are different, especially if the underlying PDF is skewed (nonsymmetric). However, ifthe PDF is symmetric and unimodal Je mode), these three quantities will be the same Mathematical Expectation bed in Eq, 3.7, can be generalized for a function of X. Given a function g(X), its expected valoe Eg(X)]. can be obtained as a generalization of Eq. 3.7 as £1g20] = Veeppxd (3.9) if X is discrete, whereas ifX is continuous, Elg(X)] | g(x) fxr ide (3.96) Incither case, E[g(X)] is known as the mathematical expectation of g(X) and is the weighte average of the function 2(X) Measures of Dispersion Again, as there is range of possible values of a random variable associated with different probabilities or probability measures, py(x)) or /x(1),some measure of dispersion is needed to indicate how widely or narrowly the values of the random variable are dispersed. Of recial interest is a quantity that gives a measure of how closely or widely the ate are clustered around a central value. Intuitively, such a quantity must be a function e or belo of the deviations from the central value. However, whethe ‘a deviation is abo the central value should be imsiaterial; therefore, the function should be an even function of the deviations.90 > Chapter3, Analytical Models of Random Phenomene > EXAMPLE 34 we mean value, then a suitable measure of the the deviations are defined relative to Jispersion is the variance. Fora discrete random variable X with PMF py (1) the variance fis hted average of the squared OraOk) = Oyen? Therefore, according to a. 39, fora continuous X with PDF wart) = [c= wo? feed 3.108) BUX?) — 2uy BOO + wy? = EO?) — 2? + The same result would be obtained by expanding Eq. 3.10a. Thus, a useful relation for the VartX) = EX Gin In Eq, 3.11, the term E(X?) is known as the mean square value of X Dimensionally, a more convenient measure oF dispersion is the square root of the atiance, which is the standard deviation 0: that is, ox = VVariX) G.2 We might observe that based solely on the value of the variance or standard deviation, it may be difficult to state the degree of dispersion; for this purpose, a measure of the dispersion relative to the central value would be appropriate, In other words, whether the dispersion is is often a preferred and convenient nondimensional measure of dispersion or variabili In Example 3.1 the PMF of the numberof operating bulldozers after 6 months shown in Fig. E316 1g bulldozers after 6 months as Cn this basis, we obtain the expected number of operatn uy = E(X) = 0108) + 18) + 278) +308) 1.50 As the random variable is discrete, the mean vale of 1.5 is not necessarily a possible value; inthis we, we may only conclude tha the mean number of operating bulldozers is between 1 and 2 at the end of 6 The corresponding variance is Var(X) = (0 = 1.5)°C/8) + (1 = 1.5798) + 2 — 15°) +6 ~1.5)7(18)> EXAMPLE35 31 Random Variables and Probably Distribution < 9 The corresponding standard deviation is, therefore which means that the degree of dispersion is over SO% of the mean value, a relatively large In Example 33, the usefl fife, T, of welding machines is a random variable with an exponent ur = E(t) = [rhea Performing the integration by parts, we obtain yer = 1 erefore, the parameter} ofthe exponential distribution i the reciprocal ofthe mean value ie = UE Tn this case, the mode is zero, whereas the median life fy is obtained as follows:92 > Chapter3. Anavt > EXAMPLE36 is of Random Phenomena pleted on schedule, If this record prevails, fn the next six projects can be des TX is the number of projects completed on schedule amon; 4.3.30) px=1=(°) oor Renae, the factor ( saa | | f 0.037 * ojos | ae pe Figure E35 PMF of X, no. of projects completed on schedul 04 = 0.03686 + 2(0.13830) + 3.27640) + (0.31110) + 5(0.18660) + 610.0466 Therefore, the expectéd number of projects among six that can be completed on schedule by th company js between three and four, The corresponding variance is31 Random Variables and Probsbilty Distribution < 93, 6 : (3.60) oo roar + -3607 06,04) 6-01 0 2607 get 08704" + 83607 ze 06704) 3.0) 090.47 + 6-307 (060.4) ai + 6-3) a 9 + 0.0985 + 0.0198 + 0.1626 + 0.2684 uu UL _ 9.308 ighest probability of 0.311 < Tn this case, X= 4 hus the probable numberof projects that will be complete on schedule i four Measure of Skewness Another useful and important propert of its PDF or PMF, and the associated E(X ~ xP = Dla — en)? peter) The above third moment would be zero if the PDF or PMF of the random variable is symmetric about the mean value jy; otherwise, it may be positive or negative. It will be alues of X above jx ate more widely J moment will be ne positive if the tive if the relat alues below j1y. Conversely, this th sion is reversed, Therefore, the skewness of the PDF or PME o in accordance with the sign of the third central n andom variable may be desig ment E(X — px) the magnitude of this third mome These properties are illustrated in F nless measure of skewness is the sk ee of skewness,94. > Chapter. Analytical Models of Random Phonomen: Positive skewness Negative skewness 4 siemens ig /\ POF of x2 POF of x2 Figure 33 Properties The Kurtosis Finally, another useful property is the fourth central moment of a random v as the kurtosis. Extending what we have presented above for the second and third cent ‘moments, the kurtosis is Ee = wat = Dei = H0)'v or discrete X and BX = nyt = fpr! fata continvous X Physically, the kurtosis is a measure ofthe peakednes ofthe underlying PDF of X 's with Geometrical Properties ind the variance of a random variable are analogous, respectively, to the nce and the central moment of unit area, To see this, we consider Figure 34° An irepular unit a31 Random Variables and Probability Distribution 9S Comparing Eq. 3.7b with Eg. 3.15, mm variable, Extending this terminology, we may defin > EXAMPLE 37 > EXAMPLE3896 > Chapter Anaytical Models of Random Phenomena ponential distribution is oy = str 1 the standard deviation of the to Eq. 3.14, 18 0 = 20, < We recall from Example ‘Therefore, the shewnes ficient ofthis distibution, accord 3.2 USEFUL PROBABILITY DISTRIBUTIONS Theoretically, any function possessing all the properties described earlier in Sect. 3. nt the probability distribution of a random variable. For practical purposes ssed to repre ind continuous distribution functions that are however, there are a number of both discret jally useful because of one or more of the following reasons: 1. The function is the result of an underlying physical process a ved on the basis of certain physically reasonable assumptions. ‘The function is th 3, It is widely known, and the necessary probability and statistical information (in probability tables) are widely available Several of these probability distribution functions are presented, and their special properties are described and illustrated in this section. 3.21 The Gaussian (or Normal) Distribution ‘The best known and most widely used probability distribution is undoubtedly the Gaussian distribution. Ms PDF for a continuous distribution, which is also known as the non andom variable X, is given by for the Gaussian dis of the tribution, these parameters are also the mean random variable X. A popular and convenient short notation for this distribution is N(y4, 0), wich we shall also adopt in specifying the normal PDF. “The significance of the two parameters, j. and 0, may be observed graphically from and 30, whereas in is held constant at 60 and o varies with10. Fig. 35. In Fig, 35a, 3.5b. yz varies with values of 30, 45, and 60 for a constant « = 10. Figure 35 Gaussian PDFs (a) with varyin32 Useful Probability Distributions. < 97 The Standard Normal Distribution 4 Gaussian probability distribution with the parameters e = 0 and ot = 1.0 is known as the dard normal distribution and is denoted appropriately as N(O, |). [ts PDF is accordingly iy - Becau wide usage, a special notation (3) is commonly used to designate t imulative distribution function (CDP) of the standard no Ssi denote the shaded area p under the PDF as (sp). Figure 36 CDF of N(0)). =O) These notations will be adopted throughout this boo In Fig. 3.7, we can see thatthe areas (or probabilities) covered within 1, 2, and £3 i.e, = number of standard deviations about the mean je =O of the standard normal distribution are, respectively, equal ti 95.4%, and 99.7%. The same probabilities are also valid for any general normal distributions, ie., NG, ), with the same + number he CDF of N(0, 1), ie., (8), is tabulated widely as tables of normal probabilii ich as Table A.1 of Appendix A. It is important to observe that the probabilities in Table A.1 start at x = 0, and are given only for positive values of the random variable, By virt of symmetry of the standard normal PDF about zero, the probabilities for negative values of the variate ean be obtained as os ms 3.19) Furthermore, values of s corresponding to probabilities p < 0.5 are negative and may be tained using Table A. as op) =—0-' — py 6.20, fen in Appendix A.1 is to ealeulate the pro Gaussian distribution, That is, the probabilities of any normal distribution can be evaluated readily using (s) as follows. Suppose a normal variate X with distribution N(ju, 2); the probability of (a < X EXAMPLE 39 > EXAMPLE 3.10 32 Usoful Probability Distributions «99 7 packsb) : ww Figure 38 Area of Nii, ) bot The ds of 1.2 millon gallons per rom a community during storm is. «normal random variable estimated to have a mea y (mgd) anda standard deviation of 0.4 mgd; i. (1.2, 04) mgd, If the lesigned with a maximum drainage capacity of 1.5 med, what isthe underlying wing during a storm that is assumed in the des storm drain system is probability of Flooding the community willoccur when the drains oad exceeds the capacity ofthe drainage the probability of flooding is Pix lps 1-80.75) =1-0.7784 busined (0.75) = 0.7734 from Table A.1 in Appendix A terest are the following: (i) The probability thatthe drainage during storm willbe between 1.0 mgd and 1.6 md, which is 16-1 (2.0 Pao Chapter, Analytical Models of Random Phenomena 1, for. typical driver who drives 10,000 miles per year, the probability of him/her having Pox < 10,000) = o( 12:00= 15.00) _ (1.33) 10.9082 which means that the probability of accident of an individual driver is around 95% annually. oF the probability of no accident is 91%. If the driver has driven 8000 miles ina given year without encoun tring any accident, what is the probability of histher having an accident forthe remainder ofthat year? In this case, we have a conditional probability as follows: $,000 — 15,000 P x < 10,00 °%-9(—3550_) P(X < 10,000 | X > 8,000) = 00) AGREES 1-o( HEE) _ 0.092 (1.87) _ 0.092 1.9692) _ 0.061 To == 096; 0.9692 Therefore, the probability of accident-fee driving for the remainder ofthe year would be around 94 < > EXAMPLE 3.11 In the fabrication of steel beams and columns by a manufacturer, there are unavoidable variablities eel members. Suppose ina building construction, the inthe dimensions (for example, length) of th the actual lengths be erection of the beams and columns forthe building frame would require tl s with a probability or reliability of 99.74% «ofthe production process? be assumed to be normally distributed What isthe required precision, in terms of an allowable with zero mean (denoting no systematic bias in the fabrication process) a bility of 99.76 is equivalent to £30"s representing the p shown in Fig. 3.7 for the normal PDE. Therefore, to sat sy the required reliability, the tolerance E rs <8 <5)=0(2—°) -o(—*) =20(2) -1 = 0997 °1(0.9985) = 2.97 sired precision, 2, ofthe fabrication process is < 322 The Lognormal Distributi The logarithm stribution. Ifa random variable X has a lognormal distribution, its PDFUsoful Probability Distributions. < 101 Figure 39° The lognormal PDFs for various value where i= F(In X) and ¢ = /Var(InX), are the parameters of the distribution, which means that these parameters are, respectively, also the mean and standard deviation of In X. The above PDF is illustrated graphically in Fig. 3.9 for various values ofthe parameter ¢. Observ that this distribution is strictly for positive values of the variate X We may observe from Fig. 3.9 that as the parameter ¢ increases, the positive skewness Chapter 4, we shall show that if X is lognormal with parameters the PDF also in In Example 4, and ¢, then InX is normal with mean 4 and standard deviation ¢: i.e., N(Q, ¢). Because of this logarithmic relationship with the normal distribution, probabilities associated with the | fi normal probabilities, such as Table A. 1. We show this with the following: On the basis of iqs. 3.3 and 3.22, the probability that X will assume values in an interval (a,b) is Le 1(*=4)] normal distribution can be determined conveniently also from the table of standard which is analogous to E and the calculations involve the same standard normal probabilities that can be obtained from Table A.1 In view of the convenient facility For calculating the probabilities ofa lognormal random 23, and also because the values of the random variable must variable, as indicated in Eq, be pos fined only for x > 0), the lognormal distribution is especially useful in those applications where the values of the variate are known, from physical consideration, to be strictly positive; this includes, for example, the strength and fatigue life of material, thseful Probability Distributions < 103 We should note that if the e.0.v, oF X, By, is not large, say <0.30, In(1 + 3x?) In riable. By the definition of Eq, 3.8, the median x», is PUX 56. Th »-10.50) =0 Hence, in terms of the median of X, the parameter hein G2 Equating the tight sides of Eqs. 3.24b and 3.27, and using Eq. 3.25, we obtain the relation between the mean and the median of a lognormal random variable as Vi+8, G2 Which means that the mean value of a lognormal variate is invariably larger than the co sponding median; i > EXAMPLE 3.12 In Example 39, if the distribution of storm drainage from the community is a lognormal random First, we oblain the parameters 4 and ¢ ofthe lagnormal distribution as follows: From Eq. 3.25, And with Eq, 3.248 a= 1n1.20~ 1(0.3247 = 0.130 Therefore, the probability of flooding becomes 1 60.85) = 1 — 0.8023 P(X > 1.50) = 1 = PX = 1.50) =1- 0(104 > chapter > EXAMPLE3.13 lytical Models of Randam Phenomena MOmgdand 1.6 med a 1.6 (into -0.130) pao 3666) _ PAT > 5.66) (232 Usetul Probabiity Distributions «10S Therefore, the pi without breaking scheduled regular maiotenanee. It may be observed thatthe probability ofthe equipment operating for 5.66 months from the beginning is PUT > 5.66) = (0423, which is less than 50% < robubility is 614% (better than a 50% chance) thatthe equipment will remain operational down for 2 months beyond its 323 The Bernoulli Sequence and the Binomial Distribution In many engineering applications, there ate often problems involving the occurrence o 1 “trials.” For exam: recurrence of an event, which is unpredictable, in a sequence of disc ple, in allocating a fleet of construction equipment for a project, the anticipated co (of every piece of equipment in the fleet over the duration of the project would have some bearing on the determination ofthe required fleet size, whereas in planning the flood control system for a river basin, the annual maximum flow of the river over a sequence of years would be important in determining the design flood level. In these cases, the operational conditions of every piece of equipment, and the annual maximum flow of the river relat to a specified flood level constitute the respective trials. We should point out that in th nonoceurrence of an event— EXAMPLE 3.14 where 1 and p are the parameters, and is known as the binomial coefficient. The corresponding CDF is rin 3 ("Jota = pyr (3.30a) tao \k Values of the CDFs of Eq. 3.30a are tabulated in Table A.2 for specified integer values of n and given probability p. We can get a better understanding of the basis of Eq. 3.30 by observing the following: By virtue of statistical independence, the probability of realizing a particular sequence of exactly x occurrences and (n — x) nonoccurrences of an event among n trials is p\(1 — p)"™. However, the x occurrences of the event can be permuted among the n trials, so that the : . (tt we f number of sequences with x occurrences 1s ( te for example, if there are x malfunctions x " . FS Fi . E n ri among a fleet of n pieces of equipment, the x malfunctions may occur in ( ) different = sequences among the n machines. Thus, we obtain Eq. 3.30. Five road graders are used in the construction of a highway project. The operational life 7 of each grader is a lognormal random variable with a mean life of 1500 hr and a cov of 30% (see Fig. E3.14). Assuming statistical independence among the conditions of the machines, the probability that two of the five machines will malfunction in less than 900 hr of operation can be evaluated as follows. 4 (0) Lognormal: jt; = 1500 ED 6; = 0.30 900} ; ; a = 0 | ~ 0 900 t, hr 1 2 SA |S Operational Life in hours Machine No. Figure £3.14 Operational life of road graders. The parameters of the lognormal distribution are: ¢ ~ 0.30; and 4 = In 1500 — § (0.3? = 7.27. Then, the probability that a machine will malfunction within 900 hr (see Fig. E3.14) is In 900 — 7.27 = P(T < 900) = Bee sy) o( 0.30 ) = (—1.56) = 0.0594 For the five machines taken collectively, the actual operational lives of the different machines may conceivably be as shown in Fig. 3.14; ice., as illustrated in the figure, machines No. 1 and 4 have operational lives less than 900 hr, whereas machines No. 2, 3, and 5 have operational lives longer than 900 hr. The corresponding probability of this exact sequence is p’( — p)°. But the two malfunctioning machines may happen to any two of the five machines; therefore, the number of32 Useful Probability Distributions. < 107 sequences with two malfunctioning machines among the five is $1213! = 10. Consequently, if X is the numberof road graders malfunctioning in 900 hr P(X = 2) =10(0.0594)7(1 — 0.0594) = 0.0294 Also, the probability of malfunction among the five graders (i, there will be malfunctions in one or more michines) would be PIX = 1) = 1— POX =0) = 1 — 0.9406) = 0.26 ‘whereas the probability that there willbe no more than two machines malfunctioning within 900 hr would be pox <2)=9)(* Joosovv 406)" = (0.9406 + 5(0.0594)(0.9406)* + 10(0.0594)°(0.9406 0.2325 + 0.0294 = 0.9981 This last result involves the CDF of the binomial distribution, which is tabulated in Table A.2 for limited values ofthe para with 1 = 5,x=2, and p = 0.05, we obtain a value 0.9988 from this table < In spite of its simplicity, the Bernoulli model is quite useful in many engin problems that can be modeled a spective Bernoulli sequences include the following: building may or may nor be damaged annu: Hh of these cases, if the situation is repeated, the resulting series may be modeled as a Bernoulli sequence We mig dividual trials must be discrete and statistically independent, In spite of this requirement i emphasize that in modeling problems with the Bernoulli sequence, the in Bernoulli sequence. For example, time and space problems, which are generally continu may be modeled with the Bernoulli sequence by diseretizing time (or space) into appropriat intervals and admitting only two possibilities within each interval; what happens in each s then a Bernoulli sequence. We illustrate this with the following example. > EXAMPLE 3.15 The annual rainfall (accumulated generally during the winter and spring) of each year in Ora County, California, is a Gaussian random variable witha mean of 15 in, and a standard deviation of 4ins Le. MUS, 4). Suppose the current water policy ofthe county is such that ifthe annual rainfall is less than 7 in fora given year, water rationing wil be required during the summer and fall of thattical Models of Random Phenomen: County in any Assuming X isthe annual rainfall, the probability of water rationing in Oran pa EXAMPLE 3.16 3.2 Useful Probability Distbutions < 109 Recurrence Time and Return Period Ina time (or space) problem that is appropriately dise nto time (or space) inter T =N, and can be modeled as.a Bemoulli sequence, the numberof time We observe that if the discretized time intervals in the sequence are tribution of the recurrence time is equal to that of the first occurrence time. Therefore, the recurrence time in a Bernoulli sequence is also governed by the Eg. 3.31. The mean recurrence time, which is popularly known in engineering as the (average) r nd therefore, TF =£(7) = pil Generally, as q p) nite series within the above parentheses yields {1 — q)? = lip? Hence, we obtain the return period which means that the mean recurrence time between two consecutive occurrences of an event is equal to the reciprocal of the probability ofthe event within one time interva It is well {o emphasize that the retum period is only an average duration between consecutive occurrences of an event and should not be construed as the actual time between the occurrences; the actual time is 7, which is a random variable design wind may be expected to occur once every 50 In this case, the probability of encountering the 50-yr wind velocity in any | yris p = 1/50-= 0.02, Then, babilty that a newly completed building in the region will be subjected to the wind velocity forthe first time on the fifth year aftr its completion is P(T = 5) = (0.020.98)' = 0.018 whereas the probability dt the frst such wind ¥ building would be Por <5) = 0.029098)! 0.02 +0.0196 + 0.0192 + 00188 + 0.0188 ‘We might point out that this later event (the ist occurence ofthe wind velocity within Sys) isthe event of at least on in S years; thus the desired probability may also be calculated a above is quite different from the event of experiencing exactly one SO-yr wind in S yr; the probability = (0.98) = 0096. However, th Ins ne gen by the nial pba wich would te (°) a110 © Chapter, Analytical Models of Random Phenome > EXAMPLE3.17 A fixed offshore platform, shown in Fig. E3.17, is designed fora wave height of 8 m above the mean | Figure E217 An offshore platform. The probability thatthe platform wil be subjected to the design wave eight within the return petiod is therefore PCH > 8, in20 yr) =1~ PU Te probability thatthe frst exceedance of the d T — (0.05 + 0.0875 + 0,0451) = 1 ~ 0.1426 0.8574 the probability that such af PT =50T >3)_ PCT =5) _ 0.05(095) pr =siT 3) - = = 0.088 Pr >) Par=3)~ O8S7S < Observe next bility of no event occurring within its return period 732 Usetul Probebiity Distributions 111 T. Expanding the above l Furthermore, for large 7. or small p, we Pino cee Poccurrence in T) = | ~ 0.3679 = 0.6321 In other words, for a rare event that is defined as one with a Tong return period, 7 th probability ofthe event's occurring within its return period is always 0.632, This result is a 1seful approximation even for return periods that are not very nstance, for T = 20 ime intervals, such as in Example 3.16, the probability is P(no occurrence inT) = 1— (1-55) 0.6415 = 0.359 which shows thatthe error inthe above exponential approximation is less thar 325 The Negative Binomial Distribution The geomettic PMF is the probability law rence of an event in a Bernoulli sequence. The number of tim units, until the frst oveut subsequent occurrence of the same event is governed by the neg. the event in a series of Bernoulli trials, then The basis of Eq. 3.33 is as follows: Ifthe kth occurrence of an event is realized atthe nth trial, there must be exactly (k — 1) occurrences of the event in the prior (n ~ 1) trials and at the mth trial the event also occurs. Thus, from the binomial law, we obtain the probability on will be subjected tothe design wind > EXAMPLE 3.18 In Example 3.16, the probability that the build forthe third time on the tenth year is, according nets=10)= (12-1) 0027 99% = (3) 02" «999 = 7240,00001)0.8681) = 0.0005112 > Chapter 3. Analytical Models of Random Phenomena whereas the probability that the chied design wind will ccur within S years would b ciaetaet pas <5) =3°("~!) 027 oss oy (098) + (4) ory «assy (0.000008) + 3(0.000008)(0.98) + 120.000008)0:96040) = 0.00012. > EXAMPLE3.19 {Aste cable built up of a amber of independent wires ax shown in Fig. 3.19. Occasionally, the ble is subjected to high overloads; on such occasions the probability of fracture of ove of the wires 1s 0.05, andthe failure of two or more wires during a single overioad is unlike cane ase eee iad Figure E319 stl cabl lft cable must be replaced when the third wire fails the probability thatthe cable can withstand PACEW oesialh ea using Eq. 333, the required probability is PCT > 6) =1— PUT, <6) =1~ PUT (0.057 0.958 ~(3) 0s" 095)- (4) 005" «95 = 1-000185 = 0.9982 < 326 The Poisson Process and the Poisson Distribution rences of events at any point in time and/or space. For example, earthquakes could strike at any time and anywhere over a seismically active region in the world; fatigue cracks may secur anywhere along a continuous weld; and traffic accidents could happen at any tim on a given highway. Conceivably, such space-time problems maj be modeled also with and possibilitis) within each nterval, hus constituting a Bernoulli trial. However, i the event can randomly occur at an} the Bemoulli sequence, by dividing the time or space into appropriate small int stant of time (or int in space), it may occur more than once in any given time or space interval. In such cases, the occurrences of the event may be more appropriate ‘modeled with a Poisson process or Poisson sequence Formally, the Poisson process is based on the following assumptions 1. An event ean occur at random and at any instant of time or any point in spa 2, The occurrence(s) of an event in a given time (or space) interval is statistically independent of that in any other nonoverlapping interval42 Useful Probability Distributions. < 113 3. The probability of and can be given by vAt, where vis the mean occurrence rate of the event (assumed srrence ofan event in a small interval Atis proportional to At, 4. The probability of two or more occurrences in Ar is negligible (of higher orders On the basis of these assumptions, the number of statistically independent « an event in (time or space) is governed by the Poisson PME: that is, if X, is the number of PX == + =041 3.34) per unit time (or space) interval, It follows then that the mean number of occurrences in E(X,) = vt furth derivation of Eq, 3.34 is given in Appendix C. It may be emphasized that the process underlying Eq, 3.34 is @ counting process with a constant parameter v Ther more, it can be shown that the variance of X; is also vt. A n imilarites and differences between the Bernoulli sequence and the Poissc I. if’ problem involves distinct entities, such as the number of operational -e-time problems, either model may be applicable; in fact, we can show, as illustrate interval is decreased. For this purpose, consider the following. From previous statistical data of traffic counts, an average of 60 cars per hour was “observed to make left tums at a given intersection, Then, suppose We are interested in the probability of exactly 10 cars making left turns atthe intersection in a 10-min interval As an approximation, we may first divide the I-hr duration into 120 intervals of ‘each, such that the probability of a leit wen (L.72) in any 30-see interval would be 0/120 = 0.5. Then, allowing no more than one L.7: in any 30-see interval, the problem is reduced to the binomial probability of the occurrence of 10 L.T. among the maximum possible of 20 L.7: in the 10-min interval, in which the probability of a 7. in each 30-sec interval is 0.5. Thus, é interval. Then, the probability of an L.7: in each interval is p = 601360 = 0. PCO L.7- in 10min) = (@)o 1667)" (0.83 (JE) 0-3) x} P(x occurrences in114 > Chapter. Anaiytical Models of Random Ph > EXAMPLE 320, resocoancncsind = tia (*) (2) (1-2) . ile) Urs) nin na att \* We can observe hat 1-4) Sea an all the other tems approach 1.0 as n+ oe, except the ter 2! Therefore, i the Tmt, 5 1-> oo, we a which the Poison dstibution of Bg. 3.34, with 2 = vf. On this basis, with v= 1 minute, the probability of 10 L.7: in 10 min is then PX = 10) = eh = 09,125 an average number of four rainstorms per year. Assuming thatthe occurrence tbe modeled with the Poisson process, the probability that there would not be any rainstorms nex @ety Px, =0)= ons whereas the probability of four rainstorms next year would be P(x, = 4) = SM ot — 0.195 ‘We note from this last result that although the average yearly oceurren for mote rainstorms in the next yea is = 1 0.018 0.074 = 0.908> EXAMPLE321 32 Useful Probability Distributions. < 115 1 74 8 0.030 2 146 9 0013 19s f 0.005 $ 19s ul 0.002 156 2 001 6 104 B 0.000 The corresponding PMF is shown graphically in Fig. E320 if We ores ee win © Figure E320. PME of no. of rainstonms in a yea ing left turns atthe In designing the lef-turn bay ata state highway intersection, the vehicles intersection may be modeled as a Poison process, Ite cycle time ofthe trafic light for lft wns is ‘and the design criterion requires a left-turn lane that will be sufficient 96% ofthe time (which ‘on in some states i the United States), the lane dst As stated above, the mean rate of left turns atthe intersection is v = 100/60 per minut us suppose the design length ofthe left-turn lane is & car lengths. Then, during Jn rogues thatthe probability of no more than cars wali PX ni SO = By trial-and 1k =3, PX <3)= 4 (2) mm sos itk = 4, P(X, < 4) =0.968 Therefore, a left-turn bay of four car lengths atthe intersection is sufficient to satisfy the deChapter 3. Analytical Models of Random Phe > EXAMPLE 3.22 mn where tomado wind Force must bes for the past 20 years, the me A structure i located in a region is once every 10 years, Assume thatthe occurrence of toma 2 i wwe ma hat the limit ofthe infinite series inside the brackets isthe exponential e™ Hence, the probability of damage in $0 yean probability against tomado wind? In this cas, w Pip) = 0.05 tea ‘0 =008, Ths standard, that each structures wind in $0 years. Then the probability the next 50 years would be evaluated as follow hat the damages between structures are statistically independent, the solution invol X as the numberof dam: vax EXAMPLE3.23 A. major steel pipeline is used to transport crude ol from an oil production platform to a refiner xe of 100 km. Even though the entire pipeline is inspected once a year and repaired on, Assume that from past inspection records the average distance between locations of such corrosion is determined to be 0.15 ki. In this ease, ifthe occurrence of corrosions along the pipeline is modeled asa Poisson process with a mean occurrence rate of » = 0.15/km, the probability that there will be 10 locations of damagin = 10) = OAS x 1007" -a1sn10 — 9.049 whereas the ps P(Xwo = 8) = 1 PX In any one ofthe corrosion sites, there may be one or more cracks that could initiate Fracture file Ifthe probability ofthis event occuring a corrosion site is 0.001, the probability of fracture failure ine between inspection and repair would be (denote F for facture along the entire 100-km pi me) yy) 2 0)= 1° PF Kw =P x 7 a SS cama 64 a nw a ge ae ee ea gua Se the probability of seh large earthquakes in Southera California inthe next 15 as follows. ist, we observe that the annual probability of occurrence of large earthquake (0.04. Then, pux= )=1-rux=0=1- ("8 004/096) = 04sitical Models of Random Phenomen: large earthquakes in Southem California were modeled 2/50 = 0.04 per yer, and Altematively, if the occurre Poisson process, we would fist determi bility of such large earthquakes in the next 15 years then becomes (04 x1 etetts = 0.451 ven site can be measured in tems of an earthquake of M > 6, the ground nl shaking ata ation = 980 cmvsec"), Suppose that durin rma distribution with a median of 0.20 (0.30¢, the probability that the building shaking intensity ¥ at a particular building site has a 1 and a cos. of 0.25. Ir the seismic eapacity of a building i wil suffer damage during an earthquake of magnitude M > 6 would be P(D|M > 6) = PUY > 0.308) = 1 ~ PU $030 1-0 33) =1-0.047 = 0053 will not suffer damage from large eat xt 20 years the probability that the build for occurrences of large earthquakes) would be gan 04x 20 PDin20 years) = 1094 eo 080 a gun 80 080) = conf -a9en2 4 asi77 2 + oom OR ‘ = 0958 < We should emphasize that in both the Bernoulli sequence and the Poisson process, the ccurrences of an event between trials (inthe case ofthe Bernoulli model) and between intervals (inthe Poisson model) are statistically independent, More generally, the occurrence {a given event in one trial (or interval) may affect the occurrence or nonoccurrence of the ‘same event in subsequent trials (or intervals) In other words, the probability of eccurrene probabilities. If this condition the resulting model isa Markov chain (or Markov process) the essential principles ed in Ang and Tang, Vol. 2 (198: arkov chain are descr 32.7 The Exponential Distribution bility of the recurrence 4). On the is described by the geometric distribution (see Sect ould be described by the expon i in the case of a Poisson process, we observe that if 7) isthe time tll the first occurrence of an event, then (T; > #) means that there is no occurrence of the event in (0,1); therefore> EXAMPLE 325 82 Useful Probabiity Distributions << 119 according to Eq, PUT > 1) = P(X, =0) =e Because the occurrences of an event in nonoverlapping intervals are statistically indepen lent, 7 is also the recurrence time between two consecutive occurrences of the same event The CDF of 71, therefore, is the exponential distribution, F(t) = POT e (35 and its PDF is F Ifthe mean occurrence rate, , is constant, the mean recurrence time, £(T;),or return pe fora Poisson process can be shown to be EM) =1 G37 This may be compared with the corresponding return period of Lip for the Bernoulli se quence, Eq. 3.32, However, for events with small occurrence rate v,I/v'~ I/p, We can sh this by observing that in @ Poisson process with occurrence rate v, the probability of an ent occurring in a unit time interval (i.e, = 1) i Hence, for small nerefore, for rare events, i., events with small mean occurrence rates or long return periods, the Bernoulli and the Poisson models should give approximately the same results, According to Benjamin (1968), the historical record of earthquakes in San Francisco from 1836 to .961 shows that there were 16 earthquakes with ground motion intensity in MM-scale of VI or highe: in the San Francisco-Bay Area can be assumed 0 sthquake will occur with Whe occure Poisson process, the probability that the next hi mean occurrence rate of high-intensity earthquakes in the region is = 0.128 quake per yeut PCT, $2) =1 0%"? = 0.226 above is equivalent to the probability of the occurrence of such high-intensity earthquakes (one ‘or more) in the next two years, With the Poisson model, tis latter probability would be Pik: = 1)=1= PO: 10) =e n, this probability may also be evaluated with the Poisson disributioModels of Random Phenomena The reiwm period of an intensity VI earthquake in San Francisco, according to Bq, 3.37, is therefore In general, the probability of o time tis given by the CCF of Tin the present c Ph 4) = ern — 0.8574 the mean value of X would EX) =a4 oe EX a)= and its standard deviation is. = WA > EXAMPLE 3.26 ‘Suppose that four identical diesel engines are used to generate backup electrical p ine may be modeled with the shifted exponential distribution, with a rated of x; the resulting distribut CDE startin sudden loss of outside electrical power. The operational if T of each In this ease, the reliability of the emergency backup system would clearly be of interest. ‘example, the probability that at least two ofthe four diesel engines will start automatically during an m can be determined as follow n within 4 years is