3 MATHEMATICAL MODELLING PRINCIPLES 34 INTRODUCTION ‘The models addressed inthis chapter ate bused on fundamental theories or laws, such asthe conservations of mass, energy, and momentum. Of many approaches to undet- Standing physical systems, engineers tend to favor fundamental models for several reasons. One reason is the amazingly small number of principles that can be used to explain a wide range of physical systems; thus, fundamental principles simplify our view of nature. A second reason is the broad range of applicability of fundamen- {al models, which allow extrapolation (with caution) beyond regions of immediate ‘empirical experience; this enables engineers to evaluate potential changes in operat- ing conditions and equipment and to design new plants. Perhaps the most important eason for using fundamental models in process contol is the analytical expressions they provide relating key features of the physical system (flows, volumes, temper- ‘tures, and so forth) to its dynamic behavior. Since chemical engineers design the ‘ocess, these relationships can be used to design processes that are as easy 10 con- fol as possible, so that a problem created through poor process design need not be Partially solved through sophisticated control calculations. ‘The presentation in this chapter assumes that the reader has previously stud- fed the principles of modelling material and energy balances, with emphasis on Beady-state systems, Those unsure of the principles should refer to one of the many lory textbooks in the area (e.g. Felder and Rousseau, 1986; Himmelblau, 982). In this chapter, a step-by-step procedure for developing fundamental models ispresented a that emphasizes dynamic models used to analyze the transient behavior Drocesses and control systems. The procedure begins with a definition of the {nd proceeds through formulation, solution, results analysis, and validation 33 CHAPTER54 process pynaMics Analytical solutions will be restricted tothe simple integrating factor for this chapter and will be extended to Laplace transforms in the next chapter. Experience has shown that the beginning engineer is advised to follow this, procedure closely, because it provides a road map for the sequence of steps and a checklist of issues to be addressed at each step. Based on this strong recommenda- tion, the engineer who closely follows the procedure might expect a guarantee of reaching a satisfactory result. Unfortunately, no such guarantee can be given, be- cause a good model depends on the insight of the engineer as well as the procedure followed, In particular, several types of models of the same process might be used for different purposes; thus, the model formulation and solution should be matched with the problem goals. In this chapter, the modelling procedure is applied to several process examples, with each example having a goal that would be important in its ‘own right and leads to insights for the later discussions of control engineering. This approach will enable us to complete the modelling procedure, including the impor- tant step of results analysis, and leam a great deal of useful information about the relationships between design, operating conditions, and dynamic behavior. 3.2 A MODELLING PROCEDURE Modelling is a task that requires creativity and problem-solving skills. A general method is presented in Table 3.1 as an aid to learning and applying modelling skills, ‘but the engineer should feel free to adapt the procedure to the needs of particular problems. Itis worth noting thatthe steps could be divided into two categories: steps 1 to 3 (model development) and steps 4 to 6 (model solution or simulation), because several solution methods could be applied toa particular model. Al steps are grouped together here as an integrated modelling procedure, because this represents the ‘TABLE 34 Outline of fundamental modelling procedure 1, Deine goals ‘5. Analyze results ‘a Specific design decisions ‘8 Check results for correctness '. Numerical valves 1. Limiting and approximate answers . Functional relationships 2 Accuracy of numerical method 4. Requzed accuracy be, Imerpret results 2, Prepare information 1. Plot solution ‘Sketch process and identity system 2 Characteristic behavior like oscilla i. Kdemity variables of imrest, tions or extrema State ssumptions and data 3. Relate results fo data and assumptions 13, Formulate model 4 Evaluate seasivity ‘8. Conservation balances 5S. Answer “what if questions i. Constitutive equations 6, Validate model . Rationlize (combine equations and col- a, Select key values fo validation lect terms) i. Compare with experimental results 4. Check degrees of freedom . Compare with results frm more complex fe Dimensionless form model 4, Determine solution ‘a. Analytical . Numerical MATHEMATICAL MODELLING PRiNcrrLEs 5S ‘yernacular use ofthe term modeling and stresses the need for the model and solution technique to be selected in conjunction to satisfy the stated goal successfully. Also, while the procedure is presented in a linear manner from step 1 to step 6, the reality is thatthe engineer often has to iterate to solve the problem at hand. Only experience ccan teach us how to “look ahead” so that decisions at earlier steps are made in a man. ner that facilitate the execution of later steps. Each step in the procedure is discussed. inthis section and is demonstrated for a simple stirred-tank mixing process. © Define Goals Perhaps the most demanding aspect of modelling is judging the type of model needed to solve the engineering problem at hand. This judgment, summarized in the goal statement, is critical element of the modelling task. The goals should be spe sific concerning the type of information needed. A specific numerical value may be needed; for example, “At what time will the liquid in the tank overflow?" In addition to specific numerical values, the engineer would like to determine semi-quantitative information about the characteristics ofthe system's behavior, for example, “Will the level increase monotonically or will i oscillate?” Finally, the engineer would like to have further insight requiring functional relationships; for example, “How would the flow rate and tank volume influence the time that the overflow will occur?” Another important factor in seting modelling goals isthe accuracy of a model fad the effects of estimated inaccuracy on the results. This factor is perhaps not em- lisized sufficiently in engineering education—a situation that may lead tothe false sion that all models have great accuracy over large ranges. The modelling and is methods in this book consider accuracy by recognizing likely errors in as- ions and data atthe outset and tracing their effects through the modelling and EET snadsis steps. Its only through this earful analysis that we canbe assured igns will function properly in realistic situations. Example 3:1. Goat Te dynamic response of the mining tnkn igure 3.11 a sep change in the let conceaton sto be determined, slong with th a he ped a she of $pimse depend onthe volume and ow rte in ths exam, ale steno se fr utr poten ul 90% of th change out concen hao ured hereto, Specfi al othe example Stein ho nga hang the oul seam each thi composion fa FIGURE 31 ‘Continuous-flow ste tank.$6 Process pvNaMics Prepare Information ‘The first step is to identify the system, This is usually facilitated by sketching the process, identifying the key variables, and defining the boundaries of the system. for which the balances will be formulated. The system, or control volume, should be a volume within which the important properties do not vary with position. The assumption of a well-stirred vessel is often employed in this book because even though no such system exists in fact, many systems closely approximate this be- havior. The reader should not infer from the use of stirred-tank models in this book that more complex models are never required. Modelling of systems via partial dif- ferential equations is required for many processes in which product quality varies ‘with position; distributed models are required for many processes, such as paper and metals, Systems with no spatial variation in important variables are termed lumped- parameter systems, whereas systems with significant variation in one or more direc tions are termed distributed-parameter systems. In addition to system selection, all models require information to predict a sys- tem’s behavior. An important component of the information is the set of assumptions ‘on which the model will be based; these are selected after consideration of the phys- ical system and the accuracy required to satisfy the modelling goals. For example, the engineer usually is not concerned with the system behavior at the atomic level, and frequently not at the microscopic level. Often, but not always, the macroscopic behavior is sufficient to understand process dynamics and control. The assumptions used often involve a compromise between the goals of modelling, which may favor 0 MATHEMATICAL MODELLING Paincirtes 59) DOF = NV ~ NE 36) with DOP equal to the number of degrees of freedom, NV equal to the number of variables, and NE equal to the number of independent equations. Not every sym- bol appearing in the equations represents a variable; some are parameters that have ‘known constant values. Other symbols represent extemal variables (also called ex- ‘ogenous variables); these are variables whose values are not dependent on the be havior ofthe system being studied, External variables may be constant or vary with time in response to conditions external tothe system, such as a valve that is opened. according to a specified function (e.g., a step). The value of each external variable must be known. NV in equation (3,6) represents the number of remaining variables that depend on the behavior ofthe system and are to be evaluated through the model equations. Itis important to recognize thatthe equations used to evaluate NE must be n= dependent; additional dependent equations, although valid in that they also describe the system, are not to be considered in the degrees-of-freedom analysis, because they are redundant and provide no independent information. ‘This point is reinforced in several examples throughout the book. The three possible results in the degrees-of- freedom analysis are summarized in Table 3.2. ___ After the initial, valid model has been derived, a rationalization should be con- sidered. First, equations can sometimes be combined to simplify the overall model. ‘Also, some terms can be combined to form more meaningful groupings in the re sulting equations. Combining terms can establish the key parameters that affect the ‘behavior of the system; for example, control engineering often uses parameters like the time constant of a process, which can be affected by flows, volumes, tempert- tures, and compositions in a process. By grouping terms, many physical systems can be shown to have one of a small number of mathematical model structures, enabling ‘engineers to understand the key aspects of these physical systems quickly. This is ‘an important step in modelling and will be demonstrated through many examples, rape 32 Summary of degrees DOF = NV-NE. DOF ~ 0 The system is exactly specified, and the solution of the model can proceed DOF<0 The system is overspeifed, and in general no solution to the model exists (unless all external variables and parameters take values that fortuitously sts take values that foritousy satisfy the model equation). This is a symptom of an eer in the formation. The likely cause 1s either (1) improper designating a variabe() as parameter or external variable oF @) inchuding an extra dependent equations) inthe model. The model must be comected to achieve ze0 degrees of freedom The system is underspecifid, and an infinite number of solutions to the model it, The likely cause is either (1) impmpery designating & parameter or external SADE a vatale x) ot inti in the mae! al eatin hat determine the sens behavior. The mode mest be cried fo achieve eo degrees of ee ee ee60 Process pYNAMICS Dr e tial disadvantage is some decrease in Example 3.1. Formation, Sines problerivovesconceiratins overland ompanen me terial balances will be prepared. The overall material balance for a ae {Accumulation of mass} = {Mass in} ~ {Mass out} (G7) (PV ean ~ (OVde) = Font = FipSt 8) with p = density, Dividing by Ar and taking the limit as Ar > O gives by overflow ine fow ou nrelated oth igi eve coring 0 @ weir sation a ‘example of which is given below (Foust etal, 1980), Fiske Ji-Ly for L>Lw (3.10) G9) scofth overfow wir In is prob, with he = cont, = Vand Ly =I i taser nero a nthe et abe vero 1 sl amped wth eho tinh ak Tete, I.SS2 tn eel the tk sappsimaly Soman andthe sa cnacemiye nF a z ‘his ested aan assumption heat, wl be fr al ks with fi The next step is to formulate a material balance on component A. Since the tank iswettmoted he nan alt concen se he te = Fy—Fi=0 -.V = constant om) Acoma at) _ {mnt} [ Coa) mao) ny component A J ~ | Ain Aout J ofA UMW AVCahar “OMWAVEN) = OHWAFC MARCO GB) being its moles weigh, with Cx ting motesolme of component A and MW being alc thine beg sre bcs hess on ea econ, Die Sand kn nits Ar" Opes Possible introduce unacceptable errors into the result Solution to the equations is employed, as described ‘aumerical solutions are never exact, the error introduc ‘mall, often much less than the errors associated wit ‘essentially exact. The major drawback to numerical MaTInMATicaL MODELLING panicrrurs 61 grt « vaetCu C0) oy One might iniilly believe that another balance oa the solvent S, ould be included inthe model: MWs vats MWsF(C5o ~ Cs) only other component, with C5 the moles/volume and MWs the molecular weight. However, e is the sum of equations (3.14) and (3.15); thus, oaly two of the three equations are independent. Therefore, oly equations (3.11) and (3.14) are required forthe model and should be considered in determining the degrees of freedom. The following analysis shows that the model using only independent equations is exactly specified: Variables Caand Fy Extemal Variables: Fy and Cyy DOF = NV-NE~ 2-2-0 Equations: G.I) and 8.14) Note thatthe variable representing time must be specified to use the model for pre dicting the concentration ata particular time, ‘The model is formulated assuming that parameters do not change with time, which is not exactly correct but can be essentially true when the parameters change slowly and with small magnitude during the time considered in the dynamic mod. elling problem. What constitutes a “small” change depends on the problem, and a brief sensitivity analysis is included in the results analysis of this example to deter- ‘mine how changes in the volume and flow would affect the answer to this example. ® Mathematical Solution Determining the solution is certainly of importance, However, the engineer should realize thatthe solution is implicitly contained in the results of the Information and Formulation steps; the solution simply “figures it out.” The engineer would like to Use the solution method that gives the most insight into Iyticl solutions are preferred in most cases, because the Specific numerical values, (2) determine important fu design and operating variables and system behavior, ‘sensitivity of the result to changes in data, These res ‘We often make assumptions to enable us to obtain anal ently used approximation is linearizing nonlinear te the system. Therefore, ana -y can be used to (I) calculate inctional relationships among and (3) give insight into the sults are so highly prized that lytical solutions; the most fre rms, as covered in Section 3.4, to make analytical solutions s. In these cases, 2 numerical in Section 3.5. Although the iced can usually be made quite th the assumptions and data in tions can often be considered solutions is loss of insight. In some cases, the approximations necessary model; thus, properly calculated numerical solut62 pwocess pyNamics Example 3.1. Solution. The model in equation (3.14) is a linear, first-order ondinary differential {equation that is not separable. However, it can be transformed into a separable form bby an integrating factor, which becomes mare easily recognized when the differential ‘equation is rearranged in the standard form as follows (see Appendix B). che eee Ca = 2Cx0 de tes stem and will appear in many "The parameter + is termed the time constant of the system and will appear ‘models. The equation can be converted into separable form by multiplying both sides by the integrating factor, and the resulting equation can be solved directly: v with 2 = 1 = time constant @16) ih ‘i (2 ihr Integrating factor = IF = oo (| 1a) ae det _ de*Cs) _ Cao yn 5 3.17) Mar a ome See ay = S| eas Ca = Cot le [Note that the integration was simplified by the fact that Cay is constant after the step change (ie, for > 0). The initial condition is Ca(t) = Cai at¢ = 0, which can be used to evaluate the constant of integration, J. This formulation implies tha the time ¢ is measured from the introduction of the step change. T= Cane Cra 16x = Cav + (Chine ~ Conde G18) (Cy ~ Caine) = [Cx0 ~ (Cao) (1 = e%) ‘The final equation has used the extra relationship that (Cao)ie = Caine Subst- tuting the numerical values gives Cx = 0.925 = (Cao ~ 0.925411 ~ €°#*7) ‘Two important aspects ofthe dynamic behavior can be determined from equation (G.18). Te fist isthe “speed” ofthe dynamic response, which is characterized by the time constant, r. The second is the steady-state gain, which is defined as output ACK fe gain = Ky = Soa 2 1.0 Steady-state gain = Ky = SSO = ACM ment (V) and Note that inthis example the time constant depends on the equipment (| ‘operation of the process (F), and the steady-state gain is independent ofthese design and operating variables, Results Analysis i is the solution is correct, The first phase of the results analysis is to evaluate whether the solu n at least to the extent that it satisfies the formulation. This can be partially verified by — ¥ vol’ thme canbe caleulated by seting Ca MATHEMATICAL MODELLING PRINCIPLES 63 suring that the solution obeys some limiting criteria that are more easily derived ‘than the solution itself. For example, the result + Satisfies initial and final conditions + Obeys bounds such as adiabatic reaction temperature ‘+ Contains negligible errors associated with numerical calculations + Obeys semi-quantitative expectations, such as the sign ofthe output change Next, the engineer should “interrogate” the mathematical solution to elicit the information needed to achieve the original modelling goals. Determining specific nu ‘merical values is major part of the results analysis, because engineers need to make ‘quantitative decisions on equipment size, operating conditions, and so forth, How. ever, results analysis should involve more extensive interpretation of the solution, ‘When taitngful, results should be plotied, so that key features like oscillacene ne @xtrema (maximum or minimum) will become apparent, Important features should be related to specific parameters or groups of parameters to assist in understanding the behavior. Also, the sensitivity of the result to changes in assumptions or data ‘should be evaluated. Sometimes this is referred to as what-if analysis, where the en- ‘Bieeer determines what happens if a parameter changes by a specified amount. A results analysis enables the engineer to understand the result of the formu- lation and solution steps. Example 3.1. Results analysis, The solution in equation (3.18) is an exponential curve as shown in Figure 33. The shape of the curve is monotonic, with the maximum rate of change cer ring when the inlet step change is entered. The manner in which the variable {charges from its initial to final values is influenced by the time constant (7), which in this problem is the volume divided by the flow. Thus, the same dynamic response ould be obtained for any stirred tank with values of flow and volume that give the ‘same value of the time constant. It is helpful to leam a few values of this curve, which ‘we will se so often in process control. The values forthe change in concentration for Several values of time after the step are noted inthe following table. ‘Time from step Percent of final steady-state change in output 0 = 0 - 7 62 a 865 3r 950 ar 982 The specific quantitative question posed inthe goal statement involves determin- Toa time until 90 percent of the change in outlet concentration has occurred, This : Cant + 09(C 9 ~ Ca) in equation (3.18), ‘Which on rearrangement gives cad :64 rRoceSs p¥NAMics FIGURE Dynami i St 7 Tinie of» lo @ #0 Im 12 Le me (nin) ee a sd °9 20 40 oO 80 100 120, Time (nin 33 eal for Example 31 r= mrtn( Gea!) = 2474-2.0) = 868 min (wine = Cao he isis ime from the introduction ofthe step change, which, since the step is Dhrooced a1, feomes Sob in Fae 3.3, One sl ask ow ioran he Speciation: fits ental, a sensitivity analysis shouldbe performed, For example iPie vine and fw ae not own cnc bt can chnge itn 25 pet of theirse vals, he tie cleats above sno exact. The rang forts ine an ‘evmated fom the bounds onthe parameters that influence the time constant 2.190.05), Maximum. 08) (2.30) = 62.8 min Mex {0.085)095)° Minimum 1 = — 72210028) (_9 39) = 51.4 mia inimum | = ~ ogsy.o5) Given the estimated inaccuracy in the data, one should wait atleast 62.8 (not 56.8) ‘minutes after the step to be sure that 90 percent of the concentration change has oc- curred. Validation sw te a en ts enough that the engineering task can be performed using the ae Si nee sm MATHEMATICAL MODELLING PRiNcirLES 65 world, this question must be evaluated with careful attention to the application of the model. We do not have enough background in control engineering at this point, so the sensitivity of process and control design to modelling errors must be deferred to a later point in the book; however, all methods will be based on models, so this question will be addressed frequently because of its central importance. ‘While the sensitivity analysis in step 5 could build confidence that the results are likely to be correct, a comparison with empirical data is needed to evaluate the validity of the model. One simple step is to compare the results of the model with the empirical data in a graph. If parameters are adjusted to improve the fit of the ‘model to the data, consideration should be taken of the amount the parameters must be adjusted to fit the data; adjustments that are too large raise a warning that the model may be inadequate to describe the physical system, tis important to recognize that no set of experiences can validate the model. Good comparisons only demonstrate that the model has not been invalidated by the data; another experiment could still find data that is not properly explained by the ‘model. Thus, no model can be completely validated, because this would require an infinite number of experiments to cover the full range of conditions, However, data from a few experiments can characterize the system in a limited range of operating variables. Experimental design and modelling procedures for empirical models are the topic of Chapter 6. Example 3.1. Validation. ‘The mixing tank was built, the experiment was performed, and samples of the outlet material were analyzed. The data points are plotted in Figure 3.4 slong with ‘the model prediction. By visual evaluation and considering the accuracy of exch data (Change fom initia concentration ‘Time from input te min) > empirical dat (squares) and model (ine) for Example 3.166 raacess pyNAMIcs point, ne would accept the model as “valid” (or, more accurately, not invalid) for most ‘engineering applications. “The modelling procedure presented in this section is designed to ensure that the ‘most common issues are addressed in a logical order, While the procedure is impor- tant, the decisions made by the engineer have more impact on the quality of the result than the procedure has, Since no one is prescient, the effects of early assumptions ‘and formulations may not be appropriate for the goals. Thus, a thorough analysis of the results should be performed so that the sensitivity of the conclusions to model assumptions and data is clearly understood. If the conclusion is unduly sensitive to assumptions or data, an iteration would be indicated, employing a more rigorous model or more accurate data. Thus, the procedure contains the essential opportunity for evaluation and improvement, 3.3 MODELLING EXAMPLES ‘Most people Jeam modelling by doing modelling, not observing results of others! ‘The problems at the end of the chapter, along with many solved and unsolved prob- tems in the references and resources, provide the reader with ample opportunity t0 develop modelling skills. To assist the reader in applying the procedure toa variety of problems, this section includes a few mote solved example problems with solutions. Inall examples, steps 1 to 5 are performed, but validation is not. Example 32 Isothermal CSTR. The dynamic response of a continuous-flow, stored. tank chemical reactor (CSTR) wil be determined in this example and compared ‘withthe stired-tank mixer in Example 3.1 Goal, Determine the dynamic response of a CSTR to a step inthe inlet concentra- tion. Also, the reactant concentration should never go above 0.85 mole/m’, Han alarm» sounds when the concentration reaches 0.83 mole/m?, would a person have enough time to respond? What would a correct response be? Information. ‘The process isthe same as shown in Figure 3.1, and therefore, the system js the liquid in the tank. The important variable is the reactant concentration in the “Assumptions. The same as for the stirred-tank mixer. Data. ‘The fow, volume, and inlet concentrations (before and after the step) are the same a5 forthe stired-tank mixer in Example 3.1, and 1. ‘The chemical reaction is first-order, ry = —KCy with k = 0.040 min’ 2. ‘The heat of reaction is negligible, and no heat is transferred tothe surroundings. Formulation. Based on the model of the stirred-tank mixer, the overall material al~ ‘ance again yields Fy = Fi = F. To determine the concentration of reactant, a com ponent material balance is required, which is different from that ofthe mixing tank because there is a (negative) generation of component A as a result of the chemicel reaction. MATHEMATICAL MODELLING pRiNcirtes 67 (ome| (Ore ]-(oxmn] [ae] ow (MWaVCa evar ~ (MWaVCa), = (MWaFCao ~ MWa FC, ~ MW,VKCa)at 620) ‘Again, dividing by MW(A1) and taking the limit as Ar» 0 gives tr Ne, Fe v Ge Zex= ECa0 withthe time constant > = —" 2 di v' Fev 00 ‘The degrees-of freedom analysis yield he jes one equation, one variable (Ca), two external aes (dC a fram (Va), Siem aa is equal tothe number of equations, the degrees of freedom are zero, an exactly specified. ar so and the model is Solution. Equation (321) 3.21) isa monseprabe linear ortinny differentia equation which can be solved by application of the integrating factor: ” “nations = $ $ ri FO oa tt = FOOT gt gy 8.22) Fr, Cam eg tle cant he ‘Te nal condon gives the ine 1.0925 mole’ concent of 0925 aon at the sine fete p= 0. Teil easter omer ine Sead om he aan equton (3.21) with dC (Cra = FE pECanhn ie (0.085 m BS + (2.17004) 7°5 = O46 ‘The constant of integration can be evaluated to be p= FuiCanden = Cao), -FACHs) F+Vk F+VR This can be substituted in equation (3.22) 0 give yg = Fla _ FSCO) are " FeVE” Feve® a ij 623) (Cade + FE Gy lCa0 ~ (Cand — €*) This can be rearranged with K, FAF + Vk). give Ca~(Crdais = KpACao(t = 2%) ACs = (0.503,(0.925)(1 oy68 speed” e. Note that inthis ex- sin, the me constant determines the “speed” ofthe response, Noted ‘tan he i an depens othe een (Ve operation (and he ‘hemical action (E), and tha by comparing equation (3.16) and (3.21) the time con aca eeepc reactor salays shorter than the tre constant forthe mike, ate the same values for F and V, The steady-state gn is ‘mole/m* Ky pag ~ °° notehnt “Thus the steady sate gain inthis example depends on equipment design and operating condtons aby, sa a tafe espe sl beaten ey content tert netcrmmst cme Face 35h Te model forthe response ar the alarm value has been reached, 29.6 i! en a] Eos | Eos f Alam | Sos | 2) z a to. B04 7 7 ar ce ow 2 0 4 SO GO 7 BO o © 2 30 40 SO 70 “Time ai) ‘Time ni) ee — || : ie ‘| Cd Le eet °o a 8 ro roe ® AIGURE-S = FIGURE 35 co) whoa an the sla ae) wh sion the la ak MATHEMATICAL MODELLING PRINCIPLES 69) F Co Ca ‘| yw FIGURE 3.6 ‘Two CSTR in series. ‘minutes, is of the same form as equation (3.23), with the same time constant and sain, Example 33 Two isothermal CSTR reactors. A problem similar wo the single CSTR in Example 3.2 is presented, with the only difference that two series reuctors ae included as shown in Figure 3.6. Each tank is one-half the volume af the tank in Example 3.2 Goal. The same as that of Example 3.2, with the important concentration being inthe second reactor, Determine the time when this concentration exceeds 0.85 maleien” Information. The systems are the liquid in each tank. The data i the same as in E ample 3.2, except that Vi = Va = 1,05 mi Formulation. Again, duc tothe assumptions forthe overflow tanks, the volumes inthe ‘wo tanks ean be taken to be constant, and all flows are constant and equal. The value ‘of the concentration in the second tank is desired, bu it depends on the concentration in the first tank, Therefore, the component balances on both tanks are formulated. 14858 « RG 64) = Vike am ViEO8 & PCqs ~ Cua) ~ Vike 625 ‘The result is two linear ordinary differential equations, which in general must be solved ‘simultaneously. Note thatthe two equations could be combined into a single second order differential equation; thus, the system is second-order. Solution, In tis ease, the balance on the fist tank does not involve the concentration {nthe second tank and ths canbe solved independently from the equation representing the second reactor. (More general methods for solving simultaneous linea differential equations, using Laplace transforms, ar presented in the next chapter) The solution forthe frst balance canbe seen to be exactly the same form asthe result for Example 3.2, equation 3.23). Th analytical expression fr the concentration atthe outlet of the first tank can be substituted into equation (3,25) to give F (FCs _ FMC ( + VE FS VE ) a‘Since the volumes and flows for the two reactors are identical, ae 3 ko OF + Vik G2 Asin aon 25 ante ody png te iene fcr rr =eo([ a) = =) aCe) _ Felt (Fox _ dt v \FevE fF (fla | era F308 | arena] fal condition is (C2) Which is determined by solving the steady-state mode!, or eau = Elegtyg) amet a FY 1 = Caadain ~ (Coat Substituting this value gives the final expression forthe second reactor concentration: ne uag — F EMC pet + Cuahan = ee Cra = (Cradina — FER te + Cards — Cardinal 330) “The data can be substituted into this equation to give Caz = 0.828 ~ 0.050165 + (0.414 — 0.828) Results analysis. ‘The shape of the transient of the concentration in the second of two reactors in Figure 3.7 is very different from the transient for one reactor in Figure 53.3, The second-order response for this example has a sigmoidal or “S" shape, with a derivative that goes through a maximum at an inflection point and reduces to zero atthe ‘new steady state. Also, the total conversion of reactant is different from Example 3.2, although the total reactor volume isthe same in both cases. The increased conversion in the two-reactor system is due tothe higher concentration ofthe reactant inthe fist reactor. In fact, the concentration of the second reactor does not reach the alarm ot Timiting values after the step change for the parameters specified, although the close approach to the alarm value indicates that a slight change could lead to an alarm, “The action upon exceeding the alarm limit in the second reactor would not be ‘as easily determined for this process, since equation (3.25) shows that decreasing the inlet concentration othe first reactor does not ensure thatthe derivative of the second reactor's concentration will be negative, The system has “momentum,” which makes it more difficult to influence the eutput of the second reactor immediately. MATHEMATICAL MODELLING PaiscirLes 7 a 7 as Bo 5 BT um 4 z é 5 dos _ 2 0 a na “ime nin Tain a — ar a) Time (nin FIGURE 37 Dynamic esponses for Example 3.3, Example 34 On-off room eating. The heating ofa dwelling with an on-off eater Ws discussed ia Scton 12 The temperature was controled by a feedback system, bd Semi-quanttative arguments led tthe conclusion thatthe temperature would os cillate, In this section, a very simple model of the system is formulated and solved. Goat, Determine he dynamic response ofthe oom temperature, Also, ensure thatthe famace doesnot have to switch on or off more frequently than once per 3 minutes o low the combustion zone to be purged of gases before reignition Information. The system is taken to be the air inside the dwelling, A sketch of the system is given in Figure 1.2. The important variable: poem in geen in Figur ;portant variables are the room temperature and Assumptions. 1, The ar in the room is well mixed. 2, No transfer of material to or from the dwelling occurs. ‘3. The heat transferred depends only on rere tantered depends only on the temperature diference between the oom ‘4. No heat is transferred from the floor or ceiling. §. Effects of kinetic and potential energies are negligible.n Data. : 1. The heat capacity of the air Cy is 0.17 eal(g°C), density is 1190 gia 2, The overall heat transfer coeficient, UA = 45 % 10° eal(°C hn). 13. The size ofthe dwelling is Sm by 5 m by 3 m high. 44. The furnace heating capacity Qs is either 0 (off) or 1.5 % 10° (on) cal. The furnace heating switches instantaneously atthe values of 17°C (on) and 23°C (of. 66. The inital room temperature is 20°C and the initial furnace status is “oft 7. The outside temperature T, is 10°C. termine the tem- Formulation. ‘The system is defined as the air inside the house. To determine t ‘an energy balance should be formulated and since no material is transferred, toma ‘required. The application of the energy balance in equation (3.5) ‘no material balance is to this system gives £ ~@-@+0-W, aan Ft transfer, the fol “The shaft work is zero. From principles of thermodynamics and heat transfer, the sions ean be used fora system with negligible accumulation of potential Towing expres and kinetic energy ae on 3.32) é 7 UAC ~ 7.) + Qn 63 a NG 7 with 0 when > 23"C 1510" when? < 17°C unchanged when 17< T<23°C togive an pvcySh = UAT 1.) + Os 33) ero since the model has two equations. “The degrees of freedom for tis formulation i 2 the model two variables (T and Qj), four parameters (UA, Cy, Vand p),and one external variable (7), Thus, the system is exactly specified with equation (3.33), when the status of the hoeating has been defined by equation (3.32). Solution, Reatranging equation (3.33) gives the following nonseparable linear ordi- nary differential equation, which can be solved by application of the inte aT, Lp _ UAT, + Os _ Ye. ply = UAT Oe with s = SE ada” > Vp UA T ~The = Tat ~ Tel — 6%) 335) where = time from step in Qu 1 = time constant = 0.34 br inal value of Tas t—> = = Ts + ONUA 10°C when Qy = 0 43°C when Qy = 15x 10° the value of T when a step in Qy occurs Tina Tet Room temperate Hester tats MATHEMATICAL MODELLING pRuscirLES 73 Time the) nat 002 0s 6 O81 Time tn) MIGURE38 Dynamic response for Example 34 Result ana. Fis the numerical resis determined and pled in Figure 38 From the inal condition with he fuace off the lempetatu dares seeding 1 equation (3.38) unl he switch value of 17°C reached. Them th ores hen ing Bein insiataneosly (Oy changes fom Oo 1.5% 10), and ie test fsa wtho monn the engrish procedure is repeued she oon temperature flo aperiodic efecto Bete and 23°C. = verti teion be 7 ‘The analyal solution provides insight imo how to ale the Behavior ofthe system, The tne constant is popotna othe mae nthe ru whieh ears sonable Also ts inenly proportional othehet waster ounce eee the et arse he more quickly the stem reaches an equim wis soa ings: therefore insulating the house will decrease UA and incense the ti cont Finally the ine constant does oot depend cn ie heating byt ones whic ee forcing fanetion ofthe system threes nctesing the capa of te eae wl ital ie ost i at te re oe su nal of the modeling exercise have been sso. The tempore has teen determined a fncn of tin, and ihe stebing ey of tones ak been determine toe over Smits ia longer thane ns nt Howe switch could ocut mich ser deta sen change muti trp oe disurbance suchas door ving opened. which woulallow ara exchanger ery and cold iz. Therefore speci afty system woul be nlued to tose ta the fornace would not be estate ut ast tine pra ata of, Biking beating an ai condoning have bees stated Int vly and mre Accurate dait and mods are available (MQuison and Parker, 1988), Ale woe74. process pyNamics extensions to this simple example are suggested in question 3.9 the end ofthe chapter {adding capacitance, changing UA, and including ventilation), “This example isthe fist quantitative analysis of a continuous feedback control system, The simplicity of the model and the on-off control approach facilitated the Solution while retaining the essential characteristics of the behavior. For most indus trial processes, the oscillations associated with on-off control are unacceptable, and imore complex feedback control approaches, introduced in Part II, are required 10 achieve acceptable dynamic performance. 3.4 LINEARIZATION ‘The models in the previous sections were easily solved because they involved linear equations, which were a natural result of the conservation balances and constitu. tive relationships for the specific physical systems. However, the conservation and constitutive equations are nonlinear for most systems, and general methods for de: veloping analytical solutions for nonlinear models are not available. An alternative is numerical simulation, covered Jater in this chapter, which can provide accurate Solutions for specific numerical values but usually offers much less understanding Fortunately, methods exist for obtaining approximate linearized solutions to nonlin: ear systems, and experience over decades has demonstrated that linearized methods ff control systems analysis provide very useful results for many (but not all) realis tic processes. Therefore, this section introduces the important method for developing approximate linear models. First, the concept of lineurity needs to be formally defined. This will be done using the concept of an operator, which transforms an input variable into an output variable. ‘An operator $ is linear if it satisfies the properties of additivity and proportionality, ‘whieh are included in the following superposition, where x, are variables and a and b are constants ax, + brs) = a(x) + 6a) G36) We can test any term in a model using equation (3.36) to determine whether itis linear. A few examples are given in the following table. Function ‘Check for linearity Is check satisfied! B(x) = ke K(axy + bx) 2 kaxy + kbxr Yes. F(x) = kx! Ray + bxy)!? 2 kari)!” + kom)? No Next, it is worthwhile considering the dynamic behavior of a process, such as the stirred-tank heat exchanger shown in Figure 3.9, subject to changes in the feed temperature and cooling fluid flow rate. For a linear system, the result of the two changes isthe sum of the results from each change individually. The responses to step ‘changes in the feed temperature (atr = 5) and cooling medium flow rate (att = 20) MATIEMATICAL MODELLING PRINCIPLES 75 F Te ' Tom Tewe 5 : FIGURE39) Sred tank with heat exchanger are shown in Figure 3.10. The responses in parts a and are the eff r 1 : arts a and b are the effects of each dlstbance individually, andthe response in part cis the total effect, which is linear process is the sum of the two individual effects. Note that the true physical system experiences only the response in Figure 3.10c; the individ- ual responses are the linear predictions for each input change. (The model for (Change in fom Ty ° » © @ . uo @ Time (in) Time ni) ® fai ‘Toa hangs in 7 Time (in) © Micune 3.10 sf the linear sysem in Figure 3.9 to two input changes76 pwocess pysamics this system will be derived in Example 3.7.) This concept, as an approximation to real nonlinear processes, is used often in analyzing process control systems. ‘A linearized model can be developed by approximating each nonlinear term with its linear approximation, A nonlinear term can be approximated by @ Taylor ‘series expansion to the nth order about a point if derivatives up to nth order exist at the point; the general expressions for functions of one and two variables are given in Table 3.3. "The term R is the remainder and depends on the order of the series. A few examples of nonlinear terms that commonly occur in process models, along with their linear approximations about x,, are the following: Fx) = al? + 5x7) FO) = 7 + “The accuracy of the linearization can be estimated by comparing the magnitude of the remainder, R' tothe linear term. For a linear Taylor series approximation in one variable, ae 2ax| “The accuracy of a sample linearization is depicted in Figure 3.11. From this figure and equation (337), it canbe seen that the accuracy of the Hinear approxima: tion is relatively better when (I) the second-order derivative has a small magnitude {there is little curvature) and (2) the region about the base points small. The success- ful application of linearization to process control systems is typically justified by the small region of operation of a process when under contol. Although the uncontrolled System might operate over a large region because of disturbances in input variables. the controlled process variables should operate overa much smaller range, where the Tinear approximation often is adequate, Note that the accuracy of the linearization ‘would in general depend on the normal operating point x; Several modelling examples of linearized models are now given, with the lin- carized results compared with the nonlinear results. In all cases, the models will be R Gx? with € between xand x, (3.37) TABLE 33 Taylor series for functions of one and two variables Function of one variable about x, uF] Lael F(x) = F(x) + § xoat sy en aPtR 3.38) coy = Fans ME ox nds G8 ¢ ‘ Feta vals abot 5 ar] j sss) ~ Fata) + ax ssn) ~ Fai SEL LaF * aaa, oF Infomation. The pees The pes espment and fw arte sme as shown in Figure ‘The important variable is the reactant concentration in the reactor. ec 5 Assumpions, The ame an Example 32. MATHEMATICAL MODELLING PRINeIrLES 7 0 02 04 o6 os 1 12 14 16 18 2 independent variable FIGURE 3.1 ‘Comparison of nonlinear Function y ‘ion about x, = 1 1.32 +3) with its near approxima designated with a prime (’). Deviation variable: (x ~ x,) = x! ‘with x, = initial steady-state value A deviation variable simply translates the total variable by a constant value, and the total value of the variable is easily recovered by adding the initial steady-state value 2, to its deviation value, «'. The use of deviation variables is not necessary and provides no advantage at this point in our analysis. However, expressing a model in deviation variables will be shown in Chapter 4 to provide a significant simplification in the analysis of dynamic systems; therefore, we Wi in iynamic systems; » We will begin (o use them here for linear or linearized systems. mantoue tember eral sample 35 Isothermal CSTR. The soon th cs PAT. The tution othe single tmic CSTR problem in Example 32s now peente ora seconde hemialeacion, Goal, Determine the transient response of the tank concentration {in the inlet concentration for the nonlinear and linearized models. response toa step8 Data, The same as in Example 3.2 except the chemical reaction rate is second-order, with ry = —KCZ and k = 0.5 [(mote/m?)min} Fomulaton,Teforsan the eqns adams dees of tomate tovaneasin Etonic32 cca eam moe ease soem ite second pve ies « Rie 60- VIC a Che CR + 2Cq(Cy~ Cm) ea Recall that Cx is evaluated by sting the derivative to zero in equation (3.40) and solving for Ce. with Co having its inital vale before the input perturbation because the linearization is about the ntl steady tat. The approximation is now substituted inthe process mode! VEE — F(Cyy ~ Ca) ~ [WACK + VEC IC Cad] G42) “The model can be expressed in deviation variables by fst repeating the linearized model equation (3.42), wich s vali for any time, at the steady-state pont, when the ‘ariale fs equal tits steady state vale aC, eer VEGE = FeCane— Cu) WACK, + 2VKC Cae Cas) G43) “Ten equation (5.43) can be subir fom equation 2) to give the equation in deviation variables: VEh = F(Ciy — Ch) — 2VECWCK oss) a “The resulting model is a first-order, linear ordinary diferenial equation which can be rearranged into the standard frm, acy a v wih? = FEBVEER, ica 5) Solution, Since the input forcing function is again a simple step, the analytical solution ‘ean be derived by a straightforward application ofthe integrating factor. a ) MH) = ACioK (1 _(__F 7 ol earaes F FRIES 46) with Kp = ‘The data can be substituted into this expression to give Cx = (0.925)0.146)(1 - 8) Results Analysis. ‘The linearized solution fom equation (3.46) i plotted in Figure 3.12, in comparison with the solution to the original nonlinear differential equation, equation MATHEMATICAL MODELLING pRisciPLEs 79) Deviation variables ower Time (nia) TT. i iy o 2 4 6 © 0 2 4 Time (in) NnguRE 3.2 Dynamic responses for Fxample 4S 4h ne na be sei gv go equate deo ‘a nporanavatage of the linsized ohon sin he analytic reaton ships For example the ine coats and gins of the tee sar sna: flow stined tank poceses “mite iea aco apd Iineaaed melo onions reato—ae summarized in Table 24 These rel cn be ted to et hat pe, Ces equipment design an proces operating conan alee danas opus Cary, the raya! polins provide fet eal of tee! inna one tlaonsip betwen design and operating conditions an ymamie hoes TABLE 34 ‘Summary of linear or linearized models for single stirred-tank ee Tsthesysem ‘Tine constant Steady-state gin Physial system linear? o ee Biape3.1(CsTiing) —Yer—WFSSS~S~«w _ oo with ‘Yes. VAF + Vk) FAF + VR) — ee ‘Example 3.5 (CSTR with No VAP + 2VKCy,) doa}. v FAP +2VkC x) second-order linearized model) (linearized model)80 L 1 ricuresi3 Level in draining tank for Example 36 Example 3.6 Tank draining. The level and flow through a partially opened restric tion out of the tank system in Figure 3.13 is considered in this example, Goal. Determine a model for this system. Evaluate the accuracies of the linearized solutions for small (10 m'Mnr) and large (60 mh) step changes in the inlet flow rate Information, ‘The system i te liquid in the tank, and the important variables are the level and flow out Assumptions, 1, The density is constant. 2, The cross-sectional area of the tank, A, does not change with height. Data. 1. The initial steady-state conditions are () ows = Fy = Fi = 100 m'fa and Gi level = L = 7.0m. 2. The cross-sectional area is 7 Formulation, The level depends on the (otal amount of Tiguid in the tank; thus, the ‘conservation equation selected isan overall material balance on the system. ‘This single balance does not provide enough information, because there are two unknowns, F and L. Thus, the number of degrees of freedom (1) indicates that an- ‘other equation is required. An additional equation can be provided to determine F without adding new variables, through a momentum balance on the liquid in the exit Pipe. In essence, another subproblem is defined to formulate this balance. The major assumptions for this subproblem are that 1. The system is at quasi-steady state, since the dynamics ofthe pipe flow will be fast with respect to the dynamics ofthe level 2. The total pressure drop is due to the restriction. ‘3. Conventional macroscopic flow equations, using relationships for friction factors and restrictions, can relate the flow tothe pressure driving force (Foust et al, 1980, Bird, Stewart, and Lightfoot, 1960) With these assumptions, which relate the flow out tothe lig balance becomes i level in the tank, the Fy = EMP. + ph~ Py = kel? 3.48) with P, constant. The system with equations (3.47) and (3.48) and with two variables, Fy and Lis exactly specified. After the equations are combined, the system can be described by a single first-order differential equation: AGE = Fo~ bel 8.49) ‘The basic balance now has a nonlinear term, which can be linearized TPS = 195 4051; — 1) 50) ‘This expression can be used to replace the nonlinear term. The resulting equation, after subtracting the linearized balance at steady-state conditions and noting thatthe input is a constant step (ie., Fj = AFo).is, AE = Fy —105kesL;°5 @s1) Solution. ‘The linearized differential equation can be rearranged and solved as before. du lt ry trl = pam with a a A OT OSE siving the solution (352) _ AR a ‘The initial condition is that L’ = Oat r = 0, with time measured from the input step; thus, = =7AFiA, Substitution gives TAFo, v= Ra L flew 53) vw ) 354) © ARK e) with K, A” O5kp1;°5 Forthis example, Fu foe “os “mos Li = O14aF (1 — eM) = 378 7 = 098 hr Results Analysis, The solution of the linearized model indicates an exponential re- sponse to step change. The results for the small and large step changes in flow in ‘ate plotted in Figures 3.14a and 14, respectively. The solution to the approximate lin- ‘satized model is quite accurate forthe small step; however, it is inaccurate fora large Step, even predicting an impossible negative level at the final steady state. The general {tend that the linearized model should be more accurate for a small than fora large step ‘conforms to the previous discussion of the Taylor series. Also, the large variation of the level, which forthe larger input step is not maintained close to its initial condition as shown in Figure 3.14, suggests thatthe linear solution might not be very accurate82. rwocess pyvamics vasiables Level (OR operas sence ses es seas) Time i) 5 ° 2 i * 0 “0 Te Devin me z a BL a ' Time du) ig 50 (Ogos geet sence ses en. seas Time (he » FIGURE 314 [Dynamic rexponses for Example 3.6: (a fora small input change inearized and nonlinear essentially the same curve); (b) fora large input change, MATHEWATICAL MODELLING FaINeNLES 83 Example 37 Stirred-tank heat exchanger. To provide another simple example of an energy balance, the stirred-tank heat exchanger in Figure 3.9 is considered, Goal. The dynamic response of the tank temperature toa step change in the coolant flow isto be determined, Information. ‘The system isthe liquid in the tank, Assumptions. 1, The tank is well insulated, so that negligible heat is transferred tothe surroundings. 2. The accumulation of energy in the tank walls and cooling coil is negligible com: pared wit the accumulation in the liquid 3. The tank is well mixed, 4, Physical properties are constant, 5, The system is initially a steady state Data. F = 0,085 m/min; V calg°C) To = 150°C; To 10° p/m’ 21 ms T, = 854°C; p = 10° sim; Cy = 1 25°C: Fax = 0.50 m/min; Cpe = 1 callg"Ch pe = Formulation, Overall material and energy balances on the system are required to de- termine the flow and temperature from the tank. The averall material halance is the same as for the mixing tank, with the result that the level is approximately constant and Fy = Fy = F.Forthis system, the kinetic and potential energy accumulation terms are zero, and their input and output terms cancel if they are not zero, The energy balance is as follows: aE ad (Ho} ~ {Hh} + ~ W, 355) ‘Also, iis assumed (and could be verified by calculations) that the shaft work is neg- ligible. Now, the goal is to express the intemal energy and enthalpy in measurable variables. This can be done using the following thermodynamic relationships (Smith and Van Ness, 1987): dé fae pVC.aTIdt = pVCyaT/At 6.56) Hy = OCpFiIT) ~ Tet] as1) Note that the heat capacity at constant volume is approximated a the heat capacity at Constant pressure, which is acceptable for this liquid system. Substituting the elation- Ships in equations (3.56) and (3.57) into (3.55) gives a PVCySL = PCpFUTo ~ Trt} ~ 171 ~ Taal) + @ 3.58) ‘This isthe basic energy balance on the tank, which is one equation with two variables, T and Q. To complete the model, the heat transferred must be related tothe tank tem. erature and the extemal variables (coolant flow and temperature), Thus, a subproblem,84 involving the energy balance onthe liquid inthe cooling cols is now defined and solved (Douglas, 1972). The assumptions are 1, The coil liquid is at quasi-steady state, 2. The coolant physical properties are constant 13. The driving force for heat transfer can be approximated as the average between the inlet and outlet, ‘With these assumptions, the energy balance on the cooling coil is 59) Boke “The subscript c refers tothe coolant fluid. Now, two constitutive relationships are em ployed to complete the model, The heat transferred can be expressed as (T= Ton) + 0 = -vaare = -va(™ =) aw 2 } ‘The heat transfer coefficient would depend on both film coefficients and the wall re- sistance, For many designs the outer film resistance in the stirred tank and the wall resistance would be sinall compared with the inner film resistance; thus, UA = hi A. “The inner film coefficient can be related to the low by an empirical relationship ofthe form (Foust etal, 1980) UA= af? asp Equations (3.59) to (3.61) can be combined to eliminate Tyga and UA to give the following expression for the heat transferced eT) G62) “This solution to the subproblem expresses the heat transferred in terms of the speci fied, extemal variables (F- and Tyg) and the tank temperature, which is the dependent variable to be determined. Equation (3.62) can be substituted into equation (3.58) to give the final model for the strred-tank exchanger. rac at ve C,pF (ls ~T) 5-7 ~Tin) B68) Fet 2pCre “The degroes-of-feedom analysis results in one variable (7), one equation (3.63) four external variables (ag To, and Fare assumed constant, and F can change with time), and seven parameters, Thus, the model is exactly specified. However, the mode! is nonlinear duet the variable F raised to the powers band B + 1 and to the product of variables F. and T. Therefore the heat transfer term mus be Fnesrized though & “Taylor series in two variables. = Q.+ KT -T,) + KrF Fos) G64) MATHEMATICAL MODELLING PRINCIPLES 85 nak ~ Te) Q, =| 65) ‘The lnear approximation can be used sed to replace the nonear term, and again ‘the equation can be expressed in deviation variables: “ ar’ a VCyp' BC p(—T') + KyT! + Krk 3.66) Solution. The resulting approximate model is a linear first-order ordi is a linear first-order ordinary differential equation that can be solved by applying the integrating factor “ aT Ly Kr F Kr \' “(6-%5)' aon a, ar 2” ~ Voc, Fora step change in the coolant flow rate at = Oand 70 i. » with ), the solution is given Kye Voc, r= 0 eo) = AF.K, [Ie] (3.68) Forthis example, b = 0.5 and a = 1.41 x 108 cali “C) : / call(min °C), and the linearized coef cients can calculated to be Ke = ~3.97 x 10°((aliminf/fn min), Ky = -909 10 (calminJ*C). The steady-state gain and ime constant canbe determined to be Kn °c F k, =- ra(E Voc, = amin ( Results anatis. Te sltion gives an exponential eatoship between tine snd harbors. Te spite nate on pn Fie 313 og with he ston fo the nonlinear mode ore magni hese shane cei aproinatn pena oda bes Kap Tita inated apanatin roves lop bwesn ta ent reponse and paces design and operation For example since Rr 0 eaten (8.67 demonstrates that the ine constant forthe hea exchange isalways smaller than the tne constant fr the same stimed tank wibuat is excunge or hich = Example 38 Flow manipulation. As explained brity in Chapter 1. process con- trl requires manipulated variable that can be adjusted independently by a person or mation system. Possible manipulated variables include motor speed and electrical Power, bt the manipulated variable inthe majority of process control system is valve opening, whic influences the ow of gas, gi or sly. Therefore, itis worth- le briefly considering a model forthe effect of valve opening on flow. A simplified86. rnocess pywawics MATHEMATICAL MODELLING PRINCIPLES 87 Deviation ‘Thus, linear or linearized models involving low ean be expressed asa function of valve variables position using equation (3.70). This is the expression used for many ofthe models in the next few chapters. More detail on the industrial flow systems will be presented in ‘Chapters 7 (automated valve design) and 16 (variable characteristic and pressures). a ‘The examples in ths section have demonstrated the ease with which lineariza- tion ean be applied to dynamic process models. As shown in equation (3.37), the second-order term in the Taylor series gives insight into the accuracy of the Hinear approximation. However, there is no simple manner for evaluating whether a linear approximation is appropriate, since the sensitivity of the modelling results depends ‘on the formulation, input variables, parameters, and, perhaps most importantly, the goals of the modelling task. An analytical method for estimating the effects ofthe second-order terms inthe Taylor series on the results of the dynamie model is avail able (Douglas, 1972); however, it requires more effort than the numerical solution of the original nonlinear equations. Therefore, the analytical method using higher- arder tems in the Taylor series is nt often used, although it might find application fora model solved frequently. One quick check on the accuracy of the linearized model is to compare the final values, as time goes to infinity, of the nonlinear and linearized models. If they differ by too much, with this value specific to the problem, then the linearized model os sas asa would be deemed to be of insufficient accuracy. Ifthe final values are close enough, ° Time (min) the dynamic responses could still differ and would have to be evaluated. Also, values ofthe time constants and gain at the initial and final conditions can be determined if they are sigmiticantly different, the linearized model is not likely to provide ade- quate accuracy. The reader will be assisted in making these decisions by numerous Nea Linear 250 eos ea eas Time (in) 06 o4 Coolant flow os-— a0 FIGURE 345 Dynamic response fr Example 37. system is shown in Figure 3.16, which is described bythe following macroscopic e ergy balance (Foust etal, 1980; Hutchinson, 1976). F=Gy [Poo 3.69) voe where C, = inherent valve characteristic = valve stem position, elated to percent open FF = volumetric fow rte ‘The valve stem position is changed by a person, as with a faucet, or by an automated system. The inherent valve characteristic depends in general on the tem poston also, the pressures inthe pipe would depend onthe flow and, thus, the stem position For the presen, the characteristic and pressures will be considered to be approximately Constant n that ese the flow isa linear function of the valve sem positon: 7 3.70) Ky with k, FIGURE 316 Simplified schematic of flow through valve examples in this book that evaluate linearized control methods applied to nonlinear processes. ‘The more complete approach for checking accuracy is to compare results from the linearized and full nonlinear models, with the nonlinear model solved using nu- ‘merical methods, as discussed in the next section. Fundamental models can require ‘considerable engineering effort to develop and solve for complex processes, s0 g *pproach is usually reserved for processes that are poorly understood or known to be highly nonlinear. In practice, engineers often lear by experience which processes in their plants can be analyzed using linearized models. Again, this experience indicates that in the majerity of cases, linear models are ‘sdequate for process control. An additional advantage of approximate linear models the insight they provide into how process parameters and operating conditions Affect the transient response, 35 NUMERICAL SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS fret are stations in which accurate solution of the nonlinear equations ar re hited. Since most systems of nonlinear algebraic and differential equations cannot Re solved analytically, approximate solutions are determined using numerical meth: ls. Many numerical solution methods are available, anda thorough coverage ofthe88 roces oases pic would require complete book (for example, Carnahan et, 1969, and Maron srud Lopez, 1991), However afew of the simplest numerical methods for solving Srainary differential equations will be introduced here, and they wil be adequate, i not the most efficent for most of the problems inthis book, ‘Numerical methods do not find analytical solution ike the expressions in the previous sections they provide ast of points that are “close” 10 the te solution Br the differential equation. The general concept for numerical solutions ist use an inital value (or values) of variable and an approximation of the derivative over a Ainge step to determine the variable after the sep. For example the solution tothe differential equation ay a can be approximate from series approximation to give with y em) fests With At = tho1 ~ tu by a linear Taylor eee dy yer tLe dir = vt FOU DAE “Te procedure in equation @.72) isthe Euler numeri integration method (Cama- ham eta, 1969), This procedure canbe repeated for any number of ime steps to eld the approximate ston over aime interval ee pmericl mets an nc higher ner term in the Taylor sete prove the accuracy. Te obvious method would be to determin higher-order terms Inthe Taylor seis in equation Q.12); however, his would require algebraic manip- ulations that are generally avoided, although they coud be practical with computer algebra A manner hasbeen developed to achieve the equivalent accuracy by eval ating the ist ervtive term at several poins within the ste. The results presented here without derivation; the derivation is available in most textbooks on numerical analysis (Maron and Lopez, 1991. There are many forms ofthe solution al of which tre refored to as Runge-Kutta methods, The follwing equations are one common form ofthe Runge-Kutta fourth-order method | ere em sees = 9+ Mm + 2m 42m + mal 61 with im = font m= fort Sma + m= f+ Sma mg = fe + Sims, 11 + AD) ‘All numerical methods introduce an error at each step, due to the loss of the higher-order terms in the Taylor series, and these errors accumulate as the integration proceeds. Since the accumulated error depends on how well the function is approxi: ‘mated, the Euler and Runge-Kutta methods have different accumulated errors. The Euler accumulated error is proportional to the step size; the Runge-Kutta error in. ‘ i MATHEMATICAL MODELLING perscinLEs 89 equation (3.73) depends on the step size tothe fourth power. Thus, the Euler method requires a smaller step size for the same accuracy as Runge-Kutta; this is partially offset by fewer calculations per step required for the Euler method. Since the errors from both methods increase with increasing step size, a very small step size might be selected for good accuracy, but a very small step size has two disadvantages, First, it requires a large number of steps and, therefore, long computing times to complete the entire simulation. Second, the use of too small a step size results in a very small change in y, perhaps s0 small as to be lost due to round-off. Therefore, an intermediate range of step sizes exists, in which the approximate numerical solution typically provides the best accuracy. ‘The engineer must choose the step size Ar to be the proper size to provide adequate accuracy. The proper step size is relative to the dynamics of the solution; thus, a key parameter is Ai/r, with 7 being the smallest time constant appearing ina linear(ized) model. As a very rough initial estimate, this parameter could be taken to be approximately 0.01. Then, solutions can be determined at different step sizes; the region in which the solution does not change significantly, as compared with the accuracy needed to achieve the modelling goal, indicates the proper range of step size. There are numerical methods that monitor the error during the problem solution and adjust the step size during the solution to achieve a specified accuracy (Maron and Lopez, 1991). Some higher-order systems have time constants that differ greatly (e.g.. 7) = 1 and 72 = 5000); these systems are referred to as stiff. When explicit numerical ‘methods such as Fuler and Runge-Kutta are used for these systems, the step size ‘ust be small relative to the smallest time constant for good accuracy (and stability), but the total interval must be sufficient for the longest time constant to respond. Thus, the total number of time steps can be extremely large, and computer resources fan be exorbitant. One solution method is to approximate part of the system as a 4 Guasi-steady state; this was done in several of the previous examples in this chapter, {Such as Example 3.7, where the coolant energy balance was modelled asa steady- | State process, When this is not possible, the explicit numerical methods described -# above are not appropriate, and implicit numerical methods, which involve iterative ‘ calculations at each step, are recommended (Maron and Lopez, 1991). 4 | Bither the Euler or the Runge-Kutta method should be sufficient for the prob- lems encountered in this book, but not forall realistic process control simulations. % Recommendations on algoritiim selection are available in the references already 4, Moted, and various techniques have been evaluated (Envright and Hull, 1976). The % fumerical methods ate demonstrated by application to examples Example 3.9 Isothermal CSTR. In Example 3.5 a model of an isothermal CSTR With a second-order chemical reaction was derived and an approximate linear model ‘was solved. The nonlinear model cannot be solved analyticelly; therefore, « numerical solution is presented, The Euler method can be used, which involves the solution ofthe following equation at each step, i