Measurements OUT B te Richard S. Figliola Donald E. BeasleyChapter 1 Basic Concepts of Measurement Methods 1.1 INTRODUCTION We make measurements everyday. For example, we routinely measure body weight ‘on a scale or read the temperature of an outdoor thermometer. Most people put little thought into the selection of instruments for these very simple measurements. After all, the direct use of the data is clear to us, the types of instruments and techniques have long been established by custom, and the outcome of these measurements is not important enough to merit much attention to features such as accuracy or alternative methods. However, when the stakes become greater, the selection of measurement equipment and techniques and the interpretation of the measured data can demand considerable attention. Just contemplate on how you might verify that a new engine is built as designed and meets the performance specifications and emissions limits required. But first things first. The primary objective in any measurement is to establish the value or the tendency of some variable. But this is based on the value or the tendency suggested by the measurement device. So just how does one establish the relationship between the real value ofa variable and that actually measured? Howcan_ ‘a measurement plan be devised so that the measurement provides the unambiguous information sought? How can a measurement system be used so that the engineer can easily interpret the measured data and be confident in their meaning? There are many considerations that have to be addressed to answer these basic but most ‘measurement questions. At the onset, we want to stress that the subject of this text is real-life ori- 4 measurement system and measurement procedures represents whose outcome will not have a unique solution. to solving a measurementt ' ' 1 1 1 1 1 1 1 1 t t 1 | ‘ Figure L1 Components of « general measurement system. e generic measurement system. A measurement variable. That physical variable becomes th nt system is a tool used for this quantification of ‘such, it is used to extend the abilities of the human senses, 1 detect and recognize different degrees of roughness, length, are limited and are not very adept at assigning specific values ‘A general template for a measurement system is illustrated in such a system consists of part or all of four general stages: ducer stage, (2) the signal-conditioning stage, (3) the output stage, B 4oHl (Wy to feedback-control stage. These stages form the bridge between the input 40 the measurement system and the system output, a quantity that is used to infer the value of the physical variable measured. The relationship between the input informa. tion, as acquired by the sensor, and the system output is established by a calibration, ‘The sensor is a physical element that uses some natural phenomenon to sense the variable being measured. ‘The transducer converts this sensed information into a detectable signal form, which might be electrical, mechanical, optical, or otherwise. ‘The goal is to convert the sensed information intoa form that can be easily quantified. For example, the liquid contained within the bulb on the common bulb ther- mometer of Figure 1.2 exchanges energy with its surroundings until the two are in thermal equilibrium, At that point they are at the same temperature. This energy ‘exchange is the input signal to this measurement system. The phenomenon of ther- mal expansion of the liquid results in its movement up and down the stem, which in this case is the output signal from which we determine temperature. The liquid in the bulb acts as the sensor. By forcing the expanding liquid into a narrow cap- illary, this measurement system transforms thermal information into a mechanieal displacement, Hence, the bulb’s internal capillary design acts as a transducer. Transducer Lt “There are many terms introduced that are common in instrumentation, A glossary of terms is located in the back of the text for your reference,12 General Measurement System 3 Output tage Discley sale Sen bub Sensor -tranecucer stage Figure L2 Components of a bulb thermometer equivalent to sensor, —- transducer, and output stages. It is worth noting that the term “transducer” is also often used in reference to a packaged device, which may contain a sensor, transducer, and even some signal conditioning elements. The context in which the term is used usually prevents any ‘Sensor selection, placement, and installation are particularly important, because the input to the measurement system is the information sensed by the sensor. Ac cordingly, the interpretation of all information passed through and indicated by the System depends on that which is actually sensed by the sensor. conditioning equipment takes the transducer signal and modifies it to a desired form. This optional intermediate stage might be used to perform tasks such $s increasing the magnitude of the signal through amplification, removing portions of the signal through some filtering technique, andlor providing mechanical or optical finkage between the transducer and the output stage, for example, converting a i ‘of a sensor into a rotational displacement of a pointer. consist of one or more devices, which are often connected in series. diameter of the thermometer capillary relative to the bulb volume the stem the liquid moves with increasin; ing temperature. It ei Jat might be a simple readout display or a marked sca, or it might ‘can record the signal for later analysis. Examples of these devices , re ‘and computer disk drives. The readout scale of the jFigire: 1:2 serves ee the ‘output stags of that measurement in proces contro, fourth‘A simple measurement system with a control stage ‘The operator fixes the set point for tempera. ‘and the furnace is activated as the local tempera, ined by the sensor within the device, rises or fal In a more sophisticated controller, a signal from a measure. ‘ ‘used as an input to an “expert system” controller that, through a int “ scheme, determines the optimum set conditions for the 13 EXPERIMENTAL TEST PLAN Suppose you wanted to answer the question, “What is the fuel use of my new car?” What might be your test plan? Two important variables to measure would be distance and fuel volume consumption. What other variables might influence your results? If your intent was to estimate the average fuel usage to expect over the course of ‘ownership, then the driving route you choose would play a big role in the results and is a variable. Obviously, driving only on highways will impose a different trend on the results than driving in the city, so you might want to randomize your route by using various types of roads. If more than one driver uses the car, then the driver becomes a variable because each individual drives somewhat differently. Certainly ‘weather and road conditions influence the results, and you might want to consider this in your plan. Thus we sce that the utility of the measured data is very much affected by variables beyond the primary ones measured. In developing your test, the question you propose to answer will be a factor in developing your test plan, and you should be careful in defining that question so as to meet your objective, Imagine how your test would differ if you were interested instead in providing values used to advertise the expected average fuel use of a model of car. Also, you need to consider just how good an answer you need. Is 2.35 Vkm or 1 mile/gal close enough? If not, then the test might require much tighter controls. Interestingly, this one example contains all the same elements of any sophisticated test, If you can conceptualize the factors influencing this test and how you will plan around them, then you are on track to handle almost any test. Before we move into the details of measure- ments, we focus here on some important concepts germane to all measurements and tests. Experimental design involves developing a measurement test plan. A test plan will draw from the following three steps”: 1. Parameter Design Plan. This is the test objective and identification of process variables and parameters and a means for their control. Ask: “What question am I trying to answer? What has to be measured?” and “What variables will affect my results?” 2. System and Tolerance Design Plan, This is the selection of a measurement technique, equipment, and test procedure based on some preconceived toler ance limits for error.’ Ask: “How will I do the measurement and how good do the results have to be?” These three strategies ar esi. The Tolerance Design Plan strategy used in this text draws on uncertainty (seasitivity) analyses Sensitivity ‘methods are common in design optimization, unilar to the bases for certain design methods [1S] used in engineering systems a ie1.3 Experimental Test Plan 5 3. Data Reduction Design Plan, Plan ahead on how wo analyze, present, and use the anticipated data. Ask: “How will | interpret the resulting data? How will Luse the data to answer my question?” Going through all three steps in the test plan before any measurements are taken is a useful trick of a successful engineer. Often Step 3 will force you to reconsider Steps 1 and 2! In this section, concepts related to Step 1 are introduced. Variables “Identify the relevant process parameters and variables.” This is the first step of experimental design measurement strategy. In addition to the targeted measured variable, there may be other variables pertinent to the measured process that affect the outcome. All known process variables should be listed and evaluated for any possible cause and effect relationships. If a change in one variable will not affect the value of another variable, the two are considered independent of each other. A variable that can be changed independently of other variables is known as an independent variable. A variable that is affected by changes in one or more other variables is known as a dependent variable. The control of variables is important. A variable is controlled if it can be held at a constant value or at some prescribed condition during a measurement. Complete control of a variable would imply that it can be held to an exact prescribed value. Such complete control of a variable is not usually possible, We will use the adjec- tive “controlled” to refer to a variable that can be held as prescribed, at least in a nominal sense. The cause and effect relationship between an independent variable and a dependent variable is found by applying a controlled value of the independent variable while measuring the dependent variable. Variables can be described as being continuous or discrete in nature. In this text, we use the term “discrete variable” to refer to one whose possible values can be enumerated. For example, the outcome of the roll of a die represents a discrete variable, as the numbers on a die are discrete (1-2-3-4-5-6). Recall the driver or route driven in the fuel usage example carlicr. Variables that describe an entity or specific jitem are discrete; examples include a test machine or instrument, a test specimen, or a test operator. Otherwise, the variable is considered to be continuous. For a variable whose value is determined from a pointer dial graduated between 1 and 6 would be continuous because the pointer could point to integers 1 through 6 or to any fractional value in between. Engineering variables such as displacement, pressure, ‘strain, or temperature are continuous by their nature. Variables that are not or cannot be controlled during measurement but that affect the value of the variable measured are called extraneous variables. Theit influence can confuse the clear relation between cause and effect in a measurement. Would© Test 1 (30.2 in. Hg) wm Test 2 (29.7 in. He) © Test 3 (30.0 in. Ha) 208 207 206 208 Tie min} Figure 1.3 Results of a boiling point test for water. Why should the data from three seemingly identical tests show such different results? Suppose an evaluation of the measurement system accuracy accounts for only 0.1°F of the test data scatter, Thus another plausible contributing factor is the effect ‘of an extraneous variable. Indeed, a close examination of the test data shows a measured variation in the barometric pressure. The pressure variation is consistent with the trend seen in the boiling point data, Because the local barometric pressure was not controlled (i.¢., it was not held fixed among the tests), the pressure acted as an extraneous variable adding to the differences in outcomes between the test runs. Control important variables or be prepared to solve a puzzle! Parameters A parameter is defined in this text as a functional relationship between variables. A parameter that has an effect on the behavior of the measured variable is called a control parameter, Available methods for establishing control parameters based on known process variables include similarity and dimensional analysis techniques (e.g, [1-3]), and physical laws. A control parameter is completely controlled if it ean be set and held at a constant value during a set of measurements. ‘As an example, the flow rate, Q, developed by a fan depends on rotational speed, n, and the diameter, d, of the fan, A control parameter for this group of three variables, found by similarity methods, is the fan flow coefficient, C = Q/nd*, For a given fan d is fixed (and therefore controlled), and if speed is somehow controlled, the fan flow rate associated with that speed can be measured and the flow coefficient ean be determined. Parameters can also be affected by extraneous variables. variables affect measured data can be delineated into noise ‘is a random variation of the value of the measured signalTest Plan 7 13 Experime as a consequence of the variation of the extraneous variables. Examples include variations in measured values arising from incomplete control of a variable, normal random variations in environmental conditions that affect the measured variable and/or the measurement system, or thermal noise (called Johnson noise) caused by the random temperature-induced motion of electrons within wiring. A completely controlled variable contains no noise. Interference produces undesirable deterministic trends on the measured value because of extraneous variables. One form common to electrical instruments is that of a sinusoidal wave superimposed onto a measured signal path. Examples include well-defined deterministic variations in environmental conditions, local ac power line noise (60 or 50 Hz) or fluorescent lighting arc noise (120 or 100 Hz), and electromag- netic interference (EMI) and radio-frequency (rf) interference. Hum and acoustic feedback in public address and andio systems are ready examples of interference effects that are superimposed onto a desirable signal An undesirable situation arises if the period of the interference is lor 7 than the period over which the measurement is made, because the interference will superimpose a false trend in the behavior of the measured variable. Consider the effects of noise and interference on the signal, y(r) = 2 + sin 2#. As shown in Figure 1.4, noise adds to the scatter of the signal. Interference imposes a trend onto the signal. An important goal of a measurement plan should be to per- form the measurement in a manner that will break up interference trends so that they appear as random variations in the data set. Although this will increase the scatter in the measured values of a data set, scatter can be handled by statistics It is far more important to eliminate false trends in the data set, which will other- wise lead to a misinterpretation of the measured data. Randomization methods are available that can be easily incorporated into the measurement plan and will mini- mize or eliminate interference trends. Several are discussed in the paragraphs that follow. 2+ sin (2x0) Signal + interference Signal yt) Lhd Oe 06 10 MS 20 Time (sh Figure 1.4 Effects of noise and interference superimposed on the signal y(s) = 2+ sin 2nt.y depends on x., fuel traveled. We determine y by varying these car). However, the test result can be affected ind weather and road conditions. For srhighways will impose a false (untypical) trend on our ‘so we could drive on different types of roads to break ‘the situation in which the dependent variable, y, isa function variables, Xa, Xp, However, the measurement of y can by several extraneous variables, z), where j = 1,2,..., such that (9, .th, ...32)). For the dependence of y on the independent variables to be they are varied. Although the influence of the z) variables on these tests cannot be eliminated, the possibility of their introducing a false trend on y can be minimized by the use of a proper test strategy. A random test is one such strategy. A random test is defined by a measurement matrix that sets a random order in the value of the independent variable applied. Trends normally introduced by the coupling of a relatively slow and uncontrolled variation in the extraneous variables with a sequential application in values of the independent variable applied will be broken up. This type of plan is effective for the local control of extraneous variables that change in a continuous manner. Consider Examples 1.1 and 1.2. Discrete extraneous variables can also be treated a little differently. The use of different instruments, different test operators, and different test conditions are exam- ples of discrete extraneous variables that can affect the outcome of a measurement. Randomizing a test matrix to minimize discrete influences can be done efficiently through the use of an experimental design using random blocks. A block consists of a data set of the measured variable in which the controlled variable is varied but the extraneous variable is fixed. The extraneous variable is varied between blocks. This enables some amount of local control over the discrete extraneous variable. In the fuel usage example, we might consider several blocks, each composed of a different driver (extraneous variable) driving similar routes, and averaging the results. Many strategies for randomized blocks exist, as do advanced statistical methods for data analysis (e.g., [4-6]). Consider Examples 1.3 and 1.4. EXAMPLE LI In the pressure calibration system shown in Figure 1.5, a sensor-transducer is exposed to a known pressure, p. The transducer, powered by an external supply, conver’s the sensed signal into a voltage that is measured by a voltmeter. The measurement approach is to control the applied pressure by the measured displacement of a pistoo that is used to compress a gas contained within the piston-cylinder chamber. The gas chosen closely obeys the ideal gas law, Hence, piston displacement, x, which Sets the chamber volume, V = (x x area), is easily related to chamber pressure. Identify oe ‘nt and dependent variables in the calibration and Possible extraneous KNOWN: Pressure calibration system of Figure 1.5(x; 7). Control variable V is changed. Depen, a of extraneous variables is to vary volume, control gas temperature, and measure important feature of all test plans is a strategy that minimizes the super- ition of false trends onto the data set by the extraneous variables. Since z), z», and and any inability to hold the gas temperature constant are continuous extraneous variables, their influence on pcan be randomized by a random test. This entails shuf- fling the order by which V is applied. Say that we pick six values of volume: V4, ¥j, Vs, Va, V5, and Vo, where the subscripts correspond to an increasing sequential order of the respective values of volume. Any random order will do fine. One possibility found by using the random function features of a handheld calculator is Wai VeVi: Va uM If we perform our measurements in a random order, interference trends will be broken up. EXAMPLE 13. ‘The manufacture of a particular composite material requires mixing a percentage by weight of binder with resin to produce a gel. The gel is used to impregnate a fiber to produce the composite material in a manual process called the lay-up. The strength, , of the finished material depends on the percent binder in the gel. However, the strength may also be lay-up operator dependent. Formulate a test matrix by which the strength to percent binder-gel ratio relationship under production conditions can be established. KNOWN = f (binder; operator) ASSUMPTION Strength is affected only by binder and operator FIND ‘Test matrix to randomize effects of operator SOLUTION ‘The dependent variable, «7, is to be tested against the independent variable, percent binder-gel ratio. The operator is an extraneous variable in actual production. As & _ simple test, we could test the relationship between three binder-gel ratios, A, B, andrandomizes the extraneous variable but has re. e-half over the direct use of three blocks for ratio ; either approach is fine, The above matrix is referred nclude our ability to control the independent variables in the could duplicate the Latin-square test several times to build ‘base. Such a duplication is referred to as a replication In general, the estimated value of a measured variable improves with the number of measurements. For example, a bearing manufacturer would obtain a better estimate of the mean diameter and the variation in the diameters of a batch of bearings by measuring many bearings rather than just a few. Repeated measurements made during any single test run or on a single batch are called repetitions. Repetition allows for quantifying the variation in a measured variable as it occurs during any one test or batch while the operating conditions are held under nominal control. However, Tepetition will not permit an assessment of how exact the operating conditions can be set. If the bearing manufacturer was interested in how closely the bearing mean diameter was controlled in day-in and day-out operations with a particular machine or test operator, duplicate tests run on different days would be needed. An independent duplication of a set of measurements using similar operating conditions is referred to as a replication. It allows for quantifying the variation in a measured variable as it occurs between different tests, each having the same nominal values of operating conditions. Finally, if the bearing manufacturer was interested in how closely the bearing mean diameter was controlled when different machines or different machine oper- ators were used, duplicate tests using these different configurations might hold the answer. Here, replication provides a means to randomize the interference effects of the different bearing machines or operators mentioned. These act as extraneous variables on the diameter of the bearings produced, _ Replication allows an assessment of the control on setting the ‘operating condi- tions, that is, the ability to reset the conditions to some desired value. Ultimately. replication estimates our control over the procedure used. i Consider aroom furnace thermostat. Set to some tem ac \. iperature, we can make repeated ‘measurements (repetition) of room temperature and come to a conclusion aout the average value and the Variation in room temperature at that particular thermostat ioe, Repeation Permits us (0 estimate the variation in this measured variable. tion permits an assessment of how well we can maintai room ueenrae at some thermostat setting. “Sone Bete! ime en Suppose we change the set temperature to some arbitrary vah ‘ ; but some time later return it to the original setting. We now duplicate the Read measure~ ‘ments (another repetition), The two sets of test data are replications of each other.14° Calibration 13 We might find that the average temperature in the second test differs from the first. The different averages suggest something about our ability to set and control the temperature in the room. Replication permits the assessment of how well we can duplicate a set of conditions. ‘Concomitant Methods A good strategy is to incorporate the use of concomitant methods in a measurement plan, The goal is to obtain two or more estimates for the result, each based on a different method, which can be compared as a check for agreement. This may affect the experimental design in that additional variables may have to be measured. or the different method could be an analysis that estimates an expected value of the measurement. As an example, suppose we needed to establish the volume of a cylindrical rod of known material. We could simply measure the diameter and length ofthe rod to compute this. Alternatively, we could measure the weight of the rod and compute volume based on the specific weight of the material. The second method complements the first and provides an important check on the adequacy of the first estimate. 14 CALIBRATION ‘The relationship between the value of the input to the measurement system and the system's indicated output value is established during a calibration of the mea- surement system. A calibration is the act of applying a known value of input to mncasurement system for the purpose of observing the system output, The known value used for the calibration is called the standard. Static Calibration “The most common type of calibration is known as a static calibration. In this proce- dace. known value is input to the system under ealibration and the system Output fs recorded. The term “static” refers to a calibration procedure in which the values vor the variables involved remain constant; that is, they do not change with time. Ip or ic calibrations, only the magnitudes of the known input and the measured output14 Calibration 15 ‘The proper procedure for calibration is to apply known inputs ranging from the minimum to the maximum values for which the measurement system is to be used. ‘These limits define the operating range of the system. The input operating range is defined as extending from nnn tO Xnax« This range defines its input span, expressed as 1) = Xmax — Yuin (2) Similarly, the output operating range is specified from mun tO Jina The OME SPAM» or full-scale operating range (FSO), is expressed as (1.3) {tis important to avoid extrapolation beyond the range of known calibration during measurement since the behavior of the measurement system is uncharted 1n these regions, As such, the range of calibration should be carefully selected. Accuracy ‘The accuracy of a system can be estimated during calibration. If the input value of calibration is known exactly, then it ean be called the true value. The accuracy of a measurement system refers to its ability to indicate a true value exactly. Accuracy is related to absolute error. Absolute error, ¢, is defined as the difference between the true value applied to a measurement system and the indicated value of the system: ¢ = true value ~ indicated value (14) from which the percent accuracy is found by te lel A= (1 gatas) *1 as) By definition, iecuracy can be determined only when the true value is known, such as during a calibration. An alternative form of calibration curve is the deviation plot, such as shown in Figure |. Such a curve plots the difference or deviation between a true or 5 ~ Value, y",and the indicated value, y, versus the indicated value, Deviation isa ‘extremely useful when the differences between the true and the indi ae ae atl et trons ncrescalbraton plots They areofea equ oe ete reicod to tbe moses poms agers during calibration from a temperature-sensing thermocoupk fr the nominal expected values obtained from reference tables, :005 01 O15 020 ‘y, Measured output mV] Figure 1.7 Calibration curve in the form of a deviation plot for a temperature sensor. 025030 If the indicated measured value differs from the true value, the measured value is said to contain a bias error. The average error in a series of repeated calibration measurements defines the bias error. The bias error is the difference between the average value and the true value. Both precision and bias errors affect the measure of a system’s accuracy. ‘The concepts of accuracy and of bias and precision errors in measurements can be illustrated by the throw of darts. Consider the dart board of Figure 1.8, where the goal is to throw the darts into the bull’s-eye. For this analogy, the bull’s-eye represents the true value and each throw represents a measurement value. In Figure 1.8(a), the thrower displays good precision (ie., low precision error) in that each throw repeatedly hits the same spot on the board, but the thrower is not accurate in that the dart misses the bull’s-eye each time. This thrower is precise, but we see that low precision error alone is not a measure of accuracy. The error in each throw can ‘be computed from the distance between the bull’s-eye and each dart, The average value of the error yields the bias error. This thrower has a bias to the left of the target. If the bias could be reduced, then this thrower’s accuracy would improve. In Figure 1.8(6), the thrower displays high accuracy and high repeatability, hittingThe precision and bias errors of a measurement system are the result of several interacting errors inherent to the measurement system, the calibration procedure, and the standard used to provide the known value. These errors can be delineated and quantified as elemental errors through the use of particular calibration procedures and data reduction techniques. An example is given for a typical pressure transducer in Table 1.1. Sequential Test A sequential testapplies a sequential variation in the input value over the desired input range. This may be accomplished by increasing the input value (upscale direction) or by decreasing the input value (downscale direction) over the full input range. Hysteresis: ‘The sequential testis an effective diagnostic technique for identifying and quantifying a hysteresis error in a measurement system. Hysteresis error refers to differences it the values found between going upscale and downscale in a sequential test. The effect of hysteresis in a calibration curve is illustrated in Figure 1.10(a). For a particular input value, the hysteresis error is found from the difference in the upscale and downscale output value by ¢n = ())upscale ~ (Y)downsale: Hysteresis is usually specified for a measurement system in terms of the maximum hysteresis error found in the calibration, €h,... 48 a percentage of full-scale output range: eben = x 100 ao re such as that indicated in Table 1,1, Hysteresis occurs when the output of a me dependencies can be brought about through some realistic system limitations _as friction or viscous damping in moving parts or residual charge in electrical come” ‘nents, Some hysteresis isnormal for any system and affects the precision of the syste"dicted output value based on a linear relation alsystems, truly linear behavior is only approximate, nt device specifications usually provide a statement he static calibration curve for the device. The relation value y(x) is a measure of the nonlinear behavior of . ex(x) = yx) — y(") as @ €1(x) is the linearity error that arises in describing the actual system behavior by equation (1.7). Such behavior is illustrated in Figure 1.10(b), in which 2 linear curve has been fit through a calibration data set. For a measurement system that is essentially linear in behavior, the extent of possible nonlinearity in a measurement device is often specified in terms of the maximum expected linearity error for the calibration as a percentage of full-scale output range: %e Lae = Sta x 100 a9) ‘This is how the linearity error for the pressure transducer in Table 1-1 was estimated Statistical methods of quantifying such data scatter about a line are discussed in Chapter 4. Sensitivity and Zero Errors ‘The scatter in the data measured during a calibration affects the precision in the estimate of the slope of the calibration curve. Asshown for the linear calibration curve in Figure 1.10(c), if we fix the zero intercept at zero (a zero output from the system for zero input), then the scatter in the data leads to a precision error in estimating the slope of the calibration curve. The sensitivity error, ex, is a statistical measure of the precision error in the estimate of the slope of the calibration curve (as discussed in Chapter 4). The static sensitivity of a device is sometimes temperature dependeat and this is often specified. In Table 1.1, the sensitivity error reflects calibration results al a constant reference ambient temperature, whereas the thermal sensitivity rtf was found by calibration at different temperatures. If the zero intercept is not fixed but the sensitivity is constant, then drifting of the zero intercept introduces a vertical shift of the calibration curve, as shows © Figure 1.10(d), This shift of the zero intercept of the calibration curve is know? ‘as the zero error, ¢, of the measurement system. The zero error can usually 0 reduced by periodically adjusting the output from the measurement system undef ‘4 zero input condition. However, some random variation in the zero intercept * ‘common, particularly with electronic and digital equipment subjected to temperat™t variations (e.g., thermal zero drift in Table 1.1), Instrument Repeatability ‘The ability of a measurement system to indicate the same value upon repeated Pt n ee Ga ee oh Berns oe eared ree ns of repeatability are based on multiple calibration tests (repli ‘within a given lab on the particular unit. Repeatability, as shown # a‘When we use a known value as the input to a measurement system during a cal bration, it becomes the standard on which the calibration is based. Obviously, the actual primary standards are impractical as standards for normal calibration use, but they serve as a reference for exactness. It would not be reasonable to travel 19 France to calibrate an ordinary laboratory scale by using the primary standard for mass (nor would it be permitted!). So for practical reasons, there exists a hierarchy ‘of secondary standards that attempt to duplicate the primary standards. Secondary standards provide reasonable approximations to the primary standards but can be accessed more readily for calibration purposes. However, an amount of uncertainty must be accepted with the use of standards that are replicas of primary standards At the top of the standards hierarchy, just below the primary standard, are the national reference standards maintained by designated standards laboratories throughout the world. In the United States, NIST maintains primary and secondary standards and recommends standard procedures for the calibration of measurement systems. Each subsequent level of the hierarchy is derived by calibration against the standard at the previous higher level. Table 1.3 lists an example of such a lineage for standards from a primary standard at a national lab down to a working standard. such as might be used in a typical laboratory to calibrate a measurement system. As ‘one moves down through this lineage, the degree of exactness by which a standard approximates the primary standard from which it is derived deteriorates. That is. increasing elements of error are introduced into the standard as one goes from one generation of standard to the next. As a common example, an institution might maintain its own local standard, which it uses to calibrate the everyday measurement devices found in the individual laboratories throughout the institution. Calibration of the local standard might be against an NIST transfer standard. NIST will periodically calibrate its own transfer standard. This is illustrated for a temperature standard traceability hierarchy in Table 1.4. Therefore, the uncertainty in the approximation of the known value increases as one moves down the hierarchy. It follows, then, that since the calibration determines the relationship between the input value and the output value, the accuracy of the calibration will depend in part on the accuracy ol the standard. But if typical working standards contain some error, how is accuracy ‘Table 1.4 Temperature Standard Traceability (Example) nny Example) Se bor Method re) Fixed thermodynamic points 0 Platinum resistance thermometer Platinum resistance thermometer +£0.01 ‘Working instrument Glass bulb thermometer 40.1 “Typical instrument bias and precision errors,1.6 Presenting Data 27 ever determined? At best, accuracy can only be estimated, and the confidence in cea will depend on the quality of the standard and calibration techniques ‘Test Standards The term “standard” can also be applied in other ways. Test standards refer to well defined test procedures, technical terminology, methods to construct test specimens or test devices, and/or methods for data reduction. The goal of a test standard is to provide consistency in the conduct and reporting of a certain type of measurement between test facilities. Similarly, rest codes refer to procedures for the manufacture, installation, calibration, performance specification, and safe operation of equipment. Diverse examples of test standards and codes are illustrated in readily available documents ({e.g., 11-14]) from professional societies, such as the American Society of Mechanical Engineers (ASME) and the American Society of Testing and Mate- rials (ASTM). For example, ASME Power Test Code 19.5 provides detailed designs and operation procedures for flow meters, whereas ASTM Test Standard F558-88 provides detailed procedures for evaluating vacuum cleaner cleaning effectiveness and controls the language for product performance claims. Test standards and codes are legal instruments, which must be observed by engineers. 1.6 PRESENTING DATA Since we use several plotting formats throughout this text to present data, it is best to introduce these formats here. Data presentation conveys significant information about the relationship between variables. Software is readily available to assist in providing high-quality plots, or plots can be generated manually by using graph paper. Several forms of plotting formats (or graph paper) are discussed as follows, Rectangular Coordinate Format fia id format, both the ordinate and the abscissa have uniformly sized eee atncarele The he most common format used for construct- ing plots and establishing the form of the relationship between the independent and variable. ‘Semilog Coordinate Format format, one coordinate has a linear scale and one coordinate has a Ie, Plotting values on a logarithmic scale is the same as performing ‘on those values; for example, plotting y vs. x with a log- Velox teas axte.Lagucii ales ‘spans more order of magnitude. be convenient when the data ory = Syetobieeert Dre :‘of Signals a digital signal can be accomplshey a type of solid-state device that come binary number system representation. The lin ‘that corresponds to a range of voltages creates ee aa hnology that rel ‘example, ‘compact disk player is built around technology that relies ony, ; valerate continantsly available signal, such as music from a microphone, inj, a digital form. The digital information is stored on a compact disk and later reaq i, digital form by a laser playback system [1]. However, to serve as input to a traditiong) stereo amplifier, and since speakers and the human ear are analog devices, the digi information is converted back into a continuous voltage signal for playback, Signal Waveforms In addition to the classification of signals as analog, discrete time, or digital, some description of the waveform associated with a signal is useful. Signals may be char acterized as either static or dynamic. A static signal does not vary with time. The diameter of a shaft is an example. However, this idea of a completely static sig nal requires some discussion. Many physical variables change slowly enough in time, compared to the process with which they interact, that for all practical purposes these signals may be considered static in time. For example, the voltage across the termi- nals of a battery is approximately constant over its useful life, Or consider measuring temperature by using an outdoor thermometer; since the outdoor temperature does not change significantly in a matter of minutes, this input signal may be considered static when compared to our time period of interest. A mathematical representation of a static signal is given by a constant, as indicated in Table 2.1. In contrast, oftea we are interested in how the measured variable changes with time. This leads us 0 consider time-varying signals further. __ A dynamic signal is defined as a time-dependent signal. In general, dynam signal waveforms, y(1), may be classified as shown in Table 2.1. A deterministic sig is one that varies in time in a predictable manner, such as a sine wave, a step functi0®™ a ‘ ee pees - shore hone 2.5. A signal is steady periodic if the ae Ha ae ea ne Tepeats at regular intervals in time. Examples ‘ould include the motion of an ideal pendulum, or Deterministic variables Nondeterministic variable oe)24 Signal Amplitude and Frequency 47 ‘The values of C, ¢ nd ¢b* are found from the following trigonometric identities: ACOs wt + B sinwt = V/A? + B? cos(wt - 6) Acos of + B sinwt = V/A? + B?sin(wt + 6") Qu) b= tan-1 2 o A soy ak aur y=t-6 Paiva # the maximum and minimum displacements from the equilibrium position a e value C, is the amplitude ofthe oscillation. The concepts of amplitude and frequency are essential for the description of time-dependent signals. Frequency Analysis Many signals that result from the measurement of dynamic variables at nondeter- ministic in nature and have a continuously varying rate of change, These signals, having complex waveforms, present difficulties in the selection of a measurement system and in the interpretation of an output signal. However, itis possible 1 sept rate a complex signal, or any signal for that matter, into a number of sine and cosine functions. In other words, any complex signal can be thought of as being made uP of sines and cosines of differing periods and amplitudes, which are added together in an infinite trignometric series. This representation of a signal as a series of sines tind cosines is called a Fourier series. Once a signal is broken down into a series of periodic functions, the importance of each frequency can be easily determined. This Pformation about frequency content allows the proper choice of a measurement system, and a precise interpretation of output signals. In theory, Fourier analysis allows essentially all mathematical functions of prac- tical interest to be represented by an infinite series of sines and cosines? ‘The following definitions relate to Fourier analysis: 1. A function y(¢) is a periodic function if there is some positive number T such that yt tT) = yO) ‘The period for y(0) is T. If both y(t) and y2(0) have period T, then ay, + by2 also has a period of T (a and b are constants). 2. A trigonometric series is given by Ag + Ay cos t+ Bi sint + Ap cos 2 + Bp sin2t ++-- where A, and By are the coefficients of the series. Asaphysical (instead of mathematical) example of the frequency content of a signal, consider stringed musical instruments such as BIS and violins, When a string is caused to vibrate by plucking OF bowing, the sound is @ result of the motion of the string and the resonance of the instrument, (The concept of resonance will be ex- plored in Chapter 3.) The musical pitch for such an instrument is the lowest frequency a, " ed as a Fourier serie if the function is plecewise continuous over 2A rid ac eat ee Fe ee aL al Din Ean en er the function is discontinuousVien Lid Prspcesy crm spectrum for 2 string. ‘one-fifth of the distance from 2 fixed end. [From N. H. Fletcher and'T. D. Rowing. The Physics of Musical Instruments. Copyright © 1991 by Springer-Verlag, New York. Used by permission.) With y(1) known, the coefficients A, and B, are to be determined. For Ao to be determined, equation (2.12) is integrated from —z to +7: [irda auf a5 (A f° oomas a. f° sam) (2.13) Since ff ccm =0 and [soma = equation (2.13) yields : =E fa (2.14) ‘The coefficient A,, may be determined by multiplying equation (2.12) by cos mt and integrating from —1 to a. The resulting expression for A, is Annie y(0) cos meds (2.15) Similarly, multiplying equation (2.12) by sin mt and integrating from —z to 7 yields By. Thus, for a function y(1) with a period 2m, the coefficients of the trigonomettic series representing y(1) are given by the Euler formulas: Ag= sz 5 f yijde Ans 4 Te y(t) cos meds (2.16) hat gacal y(0) sin neds25 Fourier Transform and the Frequency Spectrum 57 . ne 0 120240360 480600720 840. Frequency (He) Figure 2.18 Frequency content of the function y(1) = |120sin 12011. are even, the result is 120/-2 | 2 An = — = : eager taser) G29 ‘The Fourier series for the function |120 sin 12071] is 76.4 ~ 50.93 cos 240m — 10.10 cos 480nt — 4.37 cos 720mt Figure 2.18 shows the frequency content of the rectified signal, based on a Fourier series expansion. COMMENT The frequency content of a signal is determined by examining the amplitude of the various frequency components that are present in the signal. For a periodic mathematical function, these frequency contributions can be illustrated by expanding the function in a Fourier series and plotting the amplitudes of the contributing sine and cosine terms. ; : : ® On the companion software disk, several programs exist (FourCoef, FunSpect, Dataspect) to help the reader explore the concept of the superposition of simple Periodic signals to create a more complex signal. The software allows you to create your own functions and to explore their frequency and amplitude content, 28 FOURIER TRANSFORM AND THE FREQUENCY SPECTRUM. ious discussion ier analysis demonstrates that an arbitrary, but enon vee ‘as a series of sines and cosines known as a Fourier Series. The cmeunace i Fourier series specify the amplitudes of the sines and cosines, each having a specific frequency. Unfortunately, in most practical measure- it the input signal may not be known in functional form. There- Fr esate an vot Fourier analysis demonstrates that any function can beProblems 65 PROBLEMS @Y Define the term signal as it relates to measurement systems, Provide two exampl and dynamic input signals to measurement systems, 22 List the important characteristics of signals and define each, 23 Determine the average and rms values for the function of static V(t) = 30 + 2eos bar over the time periods (a) 0~0.1 s, (b) 0.4-0.5 s, (c) 0-0.33 s, and (d) 0-208. Comment on the nature and meaning of the results in terms of an analysis of dynamic signals. 24 The following values are obtained by sampling two time arying signals once every 0.4 s: Sn t nw »w®™ ¢ nO 2 00 0 0 04 1176 1529 24 -11.76 08 19.02 24.73 28 -19.02 12 1902 2473 32 -19.02 16 11.76 1529 36 -1176 -1529 20 i200. Op 40-0: 0 Determine the mean and the rms values for this discrete data. Discuss the significance of the rms value in distinguishing these signals. 2.5 A moving average is an averaging technique that can be applied to an analog, discrete time, or digital signal. A moving average is based on the concept of windowing, as illustrated in Figure 2.21. That portion of the signal that lies inside the window is averaged and the average values plotted as a function of time as the window moves across the signal. A 10-point moving average of the signal in Figure 2.21 is plotted in Figure a. Discuss the effects of employing a moving average on the signal depicted in Figure 2.21 b. Develop a computer-based algorithm for computing a moving average; determine the effect Of the width of the averaging window on the signal described by y(0) = sin Se + cos e