Developing a Parametric Model for process controls work comprises the
entire project (a process controls up- Estimating Process Control Costs • grade project). Defensible—it clearly highlights the design parameters used and provides key ratios and metrics for comparison Larry R. Dysert, CCC with other projects.
The PROCEP estimating system has
been in use since 1990, and has been a ABSTRACT: valuable tool. It has gained the confidence This article discusses the development of a parametric model used to prepare conceptual of the process control engineering and de- estimates for process control costs on capital projects. This discussion is based on a spe- sign community at the company, and is cific parametric model used by the Eastman Kodak Company, which is probably unique often used to help determine the most to its capital projects. However, the specific algorithms and the model are not the focus cost-effective process control schemes by —this article describes how the model was developed, and how it is being used. comparing various alternatives very early in the project’s life cycle. KEY WORDS: parametric model, conceptual estimates, process control
parametric cost model is an ex- for projects when minimal project scope DEVELOPING A
A tremely useful tool for prepar-
ing early conceptual estimates, when there is little technical data or engineering deliverables to pro- vide a basis for using more detailed esti- information is available. At Kodak, the capital estimating department identified the process controls discipline as an area that was difficult to estimate accurately during the strategic estimating process. To PARAMETRIC MODEL
The development of a parametric es-
timating model can appear to be a daunt- ing task; however, the use of modern com- mating methods. This article discusses the address this need, the estimating depart- puter technology (including popular development of a parametric model used ment decided to develop a parametric es- spreadsheet programs) can make the to prepare conceptual estimates for timating model based upon actual cost process tolerable, and much easier than it process control costs on capital projects. history and on key relationships that could would have been many years ago. The This discussion is based on a specific para- be identified between costs and specific process of developing a parametric model metric model used by the Eastman Kodak parameters of a process control system. should generally involve the following Company, which is probably unique to its The estimating application developed is steps: capital projects. However, the specific al- an Excel-based spreadsheet called the gorithms and the model are not the focus process control estimating program (PRO- • cost model scope determination; —this article describes how the model was CEP). • data collection; developed, and how it is being used. PROCEP allows estimators to create • data normalization; This article reviews the types of data process control estimates with the follow- • data analysis; collected to support the model, the analy- ing characteristics. • data application; sis of the data, the creation of a spread- • testing; and sheet model based on this analysis, and • Efficient—it produces conceptual es- • documentation. the various forms of output and presenta- timates in a matter of minutes when tion data available from the model. What given the appropriate design parame- is learned from this particular parametric ters. The following discussion focuses on model can be applied to other circum- • Consistent—it produces a consistent how these steps were accomplished in re- stances and used to create other forms of estimate format using documented gards to the PROCEP parametric estimat- parametric models applicable in other sit- key design parameters as the estimate ing system. uations. basis. • Accurate—it produces estimates with a predicted accuracy range of -30/+50 Cost Model Scope Determination A PROCESS CONTROL percent, applicable for class 5 and The first step in developing a para- ESTIMATING SYSTEM (PROCEP) class 4 estimates using AACE Inter- metric model is to establish its scope. This national’s estimate classifications. includes defining the end use of the Cost estimators are continually • Flexible—it can be used to develop model, the physical characteristics of the searching for effective ways to prepare process control estimates as a part of a model, the cost basis of the model, and its conceptual and early strategic estimates total project estimate, or where the critical components and cost drivers.
Cost Engineering Vol. 43/No. 2 FEBRUARY 2001 31
For the PROCEP system, the end use better identified, and the parametric • cost scope. of the system is to prepare strategic and model was revised. conceptual estimates of process control The original data collection study For the PROCEP model, the data costs, including process control hardware, began in 1990, and collected information were required to be normalized for escala- instruments, installation, software, engi- on 40 projects dating back to 1987. Since tion to bring all costs up to the same base neering and design, and debugging. It is its inception, the PROCEP model has year. All of the project data was from the intended to be a reasonably simple spread- been updated with additional project data Kodak Park Rochester site, so no adjust- sheet model that could be used with min- in 1994, 1995, and 1998. The database of ments were required for location. Site imal project scope information to produce project information currently holds data conditions and system specifications were estimates with an intended accuracy range on approximately 100 projects. elements identified on the data input of up to -30/+50 percent. It should pro- The data obtained by the input forms sheets, and were not normalized. Instead, duce a standard estimate report and also are entered into a Excel spreadsheet that these items were analyzed to determine display key metrics for comparisons with includes the following cost information: their effect and used as key parameters of other projects. the resulting parametric model. Normaliz- The model should be based on actual • major hardware and purchased soft- ing for cost scope implies making adjust- costs from completed projects and reflect ware costs; ments for unusual costs that a specific an organization’s engineering practices • instrumentation purchase costs; project may incur but that a “typical” proj- and technology. It should be applicable • installation costs (labor and bulk ma- ect would not. A comments section on the for the variety of different process tech- terials—separated for hardware and data input forms was used to identify these nologies that are involved with Kodak instrumentation, if available); occurrences, and the data were adjusted to projects (machine process, chemical • engineering and design costs; exclude the costs for unusual circum- process, building automation, etc.). The • project administration costs; stances. model should be based on current-year • software development costs; and costs or provide a mechanism to escalate • debugging costs. to current-year costs. Data Analysis The model should be based upon key The following scope information and The next step in the development of a process control parameters that can be de- process control system characteristics are parametric model is data analysis. There fined with reasonable accuracy early in also collected: are many diverse methods and techniques project scope development, and should that can be used in data analysis, and they allow an estimator to easily adjust the de- • type of process control system (gener- are too complex to delve into here. Typi- rived costs for specific complexity, or other al type as well as hardware manufac- cally, data analysis consists of performing factors affecting a specific project. turer/model information); regression analysis of costs versus selected • type of process controlled (chemical, design parameters to determine the key machine, HVAC, etc.); costs drivers for the model. Most spread- Data Collection • installation conditions (hazardous, sheet applications now provide regression Data collection and development for corrosive, etc.); analysis and simulation functions, which a parametric model require a significant • Input/output (I/O) count (analog and are reasonably simple to use. effort. The quality of the resulting para- digital or new and existing); and For PROCEP, a number of regression metric model can be no better than the • types of I/O (pressure, temperature, analysis cases (linear and nonlinear) were quality of the data it is based upon. Both analytic, etc.). run against the data to determine the al- cost and scope information must be iden- gorithms that eventually comprised the tified and collected. For the PROCEP parametric model. The various relation- model, it was decided to collect data from Data Normalization ships (cost versus design parameters) were recent projects where the cost and scope After the data have been collected, examined for “best fit” by looking for the information for process controls could be the next step in the process of developing highest R-squared value. R2 has the tech- identified readily and clearly. The collec- a parametric model is to normalize the nical sounding statistical name of coeffi- tion involved both complete projects data before the data analysis stage. Nor- cient of determination, and is commonly (where process controls is a part of the malizing the data refers to making adjust- used as a measure of the goodness of fit of overall project scope) and projects that ments to the data to account for differ- a regression equation. In simple terms, it is were primarily process control projects ences between the actual basis for each one measure of how well the regression (upgrades or replacements of process con- project and a desired standard basis of data equation explains the variability of the trol systems). to be used for the parametric model. Typ- data. The algorithms resulting from the re- The type of data to be collected was ically, data normalization implies making gression analysis are applied to the input decided upon in cooperation with the adjustments for: data sets to determine on a project-by-proj- process control engineering community. A ect basis how well the regression algo- data input form was developed to aid in • escalation; rithm predicted the actual costs. collecting the data. Over time, this form • general location; Regression analysis can be a time- has been revised as the data needs were • site conditions; consuming process (especially with the • system specifications; and simple regression tools of a spreadsheet
32 Cost Engineering Vol. 43/No. 2 FEBRUARY 2001
program), as experiments are made to dis- • estimate title information; examination of the model to look for the cover the best-fit algorithms. As an algo- • total digital and analog I/O to be ad- obvious relationships that you expect to rithm is discovered that appears to provide dressed by the controller hardware; see. If the relationships from the model good results, it must be tested to ensure • total digital and analog I/O to be pur- appear to be reasonable, run additional that it properly explains the data. Ad- chased by the project; tests for statistical significance and to veri- vanced statistical tools can quicken the • total control valves to be purchased; fy that the model is providing answers process but can be more difficult to use. • total digital and analog I/O to be in- within an acceptable range of error. Sometimes an estimator will find that er- stalled; One quick check to run is to test the ratic or outlying data points need to be re- • hardware type (distributed, program- regression results directly against the input moved from the regression runs in order to mable logic controller, etc.); data to see the percent error for each of avoid distortions in the results. It’s also im- • redundant hardware/spare capacity the inputs. This lets you quickly deter- portant to recognize that many cost rela- information; and mine the range of error, and interpreting tionships are nonlinear, and therefore one • type of process being controlled. the results can help you to determine or more input variables will be raised to a problems with the algorithm. power. Experimentation is needed with In addition, there are a number of After all of the individual algorithms both with the variables you are testing complexity adjustments that a user can have been developed and assembled into a against and the exponential powers used make based on his/her understanding of complete cost model, it is important to test for the variables. Regression analysis tends the specific project requirements, cost the model as a whole against new data to be a continuing trial-and-error process model, and experience. The spreadsheet (data not used in the development of the until the proper results are obtained that uses built-in Excel protection features to model). Consult a statistical text for more appear to explain the data. allow the user to only enter information in information about testing regression re- For the PROCEP model, the follow- the data entry cells. All other cells, includ- sults and cost models. ing data relationships resulted in regres- ing calculations and lookup table infor- sion relationships that were used in the mation, are protected from being modi- parametric model: fied or overwritten. Documentation The spreadsheet contains macros to The resulting cost model and estimat- • hardware costs versus I/O count ad- allow users to easily print the results of an ing application must be documented thor- dressed by the process control system; estimate in a consistent format. The oughly. A user manual for the estimating • field instrumentation costs versus I/O model also calculates and displays various application should be prepared that shows count purchased; key ratios between costs that can be used the steps involved in preparing an estimate • field installation costs versus I/O to benchmark the results and to compare using the cost model, and clearly describes count installed; cost trends over time. the required inputs to the cost model. The • software development costs versus I/O data used to create the cost model should count addressed by the system; be documented, including a discussion of • engineering costs versus field capital Testing how the data were adjusted or normalized cost; and One of the most important steps in for use in the data analysis stage. It is usu- • debugging costs versus field capital developing a cost model is to test its accu- ally desirable to make the actual regres- cost. racy and validity. When using regression sion data sets available, along with the re- analysis, one of the key indicators of how sulting regression and test results. All as- well a resulting algorithm explains the sumptions and allowances designed into Data Application data is a term called R2. This is the coeffi- the cost model should be documented, as The data application stage involves cient of determination, and it provides a should any exclusions. The range of ap- establishing the user interface and presen- measure of how well the resulting algo- plicable input values, and the limitations tation form for the parametric cost model. rithm predicts the calculated costs. An R2 of the model’s algorithms, also should be Using the mathematical and statistical al- value of 1 indicates a perfect fit, while an explained. gorithms developed in the data analysis R2 value of .89 indicates an 89 percent For the PROCEP model, the instruc- stage, the various inputs to the cost model confidence that the regression equation tions for use are contained in a user guide. are identified, and an interface is devel- explains the variability in cost (based on Additional information is contained in a oped to provide the user with an easy and the data used in the development of the separate resource manual and includes straightforward way to enter this informa- model). However, a high R2 value by itself the following: tion. Electronic spreadsheets provide an does not imply that the relationships be- excellent mechanism to accept user input, tween the data inputs and the resulting • original data collection forms; calculate costs based upon algorithms, cost are statistically significant. • tables showing the regressed data sets and display the resulting output. Once a regression analysis has been and regression results; As mentioned previously, PROCEP is performed, and an algorithm with a rea- • lists of the resulting algorithms used an Excel spreadsheet application. A data sonably high R2 value has been obtained, in the PROCEP cost model; entry form was generated to accept the an estimator still needs to examine the al- • graphical representations of the algo- user input that includes: gorithm to ensure that it makes common rithms plotted along with the results sense. In other words, perform a cursory
Cost Engineering Vol. 43/No. 2 FEBRUARY 2001 33
can help you to determine problems with the algorithm; and • summary charts that show the overall relationships derived from the cost model.
arametric cost models can be a
P valuable resource when prepar-
ing early conceptual estimates. Our experience at Kodak has demonstrated that effective parametric es- timating systems can be developed using basic skills in estimating, mathematics, statistics, and spreadsheet software. It is important to understand that the quality of the results can be no better than the qual- ity of the input data, and great care should be taken during the data collection stage to gather appropriate and accurate project scope and cost data. Using the statistical analysis and modeling capabilities of mod- ern spreadsheet (and/or statistical) soft- ware, cost models can be created that pro- vide a consistent and accurate basis for conceptual estimates. AACE Member Larry R. Dysert, CCC, is a project estimator for the Eastman Kodak Company in Rochester, Plus Program NY. Larry has been an active member of AACE International, How would you like to have your annual membership in A ACE free, and is currently serving or order a new book from the A ACE publications catalog as Chair of the Technical Board. In the (or other A ACE products) absolutely free? past, Larry has held several positions in the local Genesee Valley Section of AACE, in- For each new member of A ACE you sign up, you will be issued a cluding serving as President, Vice Presi- credit coupon good for A ACE products; it will be redeemable for any dent, and Programs Director. Prior to join- ing Kodak, Larry managed his own con- and all A ACE goods and services up to sulting firm for 3 years, and worked for 12 months from the date of issuance. many years with a large A&E firm. Larry has 21 years of experience in project esti- Check your dues bill for a form, or contact us mating and project management for a wide at 800-858-COST or 304-296-8444 variety of industries. He has spoken at many national [US] and international con- ferences relating to estimating and project management topics.◆ Start Planning Now!! Future AACE International Annual Meetings will be held in the following locations—