7.2 Graphical Fitting of First-Order Models Using Step Tests 169 In the step response equations, t is the independent variable instead of the input x used earlier, and y is the dependent variable expressed in deviation form. While K appears linearly in both response equations, 1 in (5-18) and 1, and 7) in (5-48) are contained within nonlinear functions, ruling out the use of linear regression to estimate them. Since K can be calculated directly from the steady-state data, we assume that it can be determined prior to estimation of 1 (or 7, and 12). Sometimes a transformation can be employed to modify a nonlinear model so that linear regression can be used. For example, if K is assumed to be known, the first-order model (5-18) can be rearranged to give In ( - 2) =-4 (7-10) Since In(1 — y;/KM) can be evaluated at each time 1, this model is linear in the parameter 1/7. The transformation in Eq. 7-10 leads to the fraction incomplete response method of determining first-order models discussed in the next section. However, for higher-order step response models, such as Eq. 5-48, the transformation ap- proach is not feasible. In these cases, we must use an iterative optimization method to find the least-squares estimates of the time constants [3]. There are many op- timization techniques that can be used for nonlinear regression; computer packages are available specifically for performing parameter estimation. On the other hand, there are a number of graphical correlations that can be used quickly to find approximate values of 1, and 7. The accuracy of models obtained in this way is usually sufficient for controller design. In the next two sections, we present several methods for estimating transfer function parameters based on graphical analysis. 7.2. GRAPHICAL FITTING OF FIRST-ORDER MODELS USING STEP TESTS The output response of a process to a step change in input, when plotted, is sometimes referred to as the process reaction curve. If the process of interest can be approximated by a first- or second-order linear differential equation, the model parameters can be obtained by inspection of the reaction curve. For example, recall the dynamic model for a first-order process, gy ta ty aks (7-11) where the system is initially at rest (y(0) = 0, corresponding to x = 0). If the input x is abruptly changed from zero to M at time r = 0, the step response in Eq. 5-18 results. The normalized graphical response is shown in Fig. 7.2. The response of y(t) reaches 63.2% of its final value at t = 7. The steady-state output deviation value is y(») = KM. From Eq. 5-18 or 7-11, after rearranging and evaluating the limit at ¢ = 0, the initial slope of the normalized step response is a (&) | at (7-12) a \KM) |,2o 7170 DEVELOPMENT OF EMPIRICAL DYNAMIC MODELS FROM STEP RESPONSE DATA Tangent to ie at¢ lop ——f—-————-ss— 8 0 0.632 37 ‘ Figure 7.2. Step response of a first-order system and graphical constructions used to estimate the time constant, + From the graphical interpretation in Fig. 7.2, the intercept of the tangent at 1 = 0 with y/KM = 1 occurs at t = 7. Hence, for a simple first-order system, there are two quick ways to calculate the time constant 7 from a step response plot: 1. Find the value of ¢ at which the response is 63.2% complete. 2. Construct the tangent to the initial response curve and find its intercept with the final steady-state value. Note that these procedures do not explicitly require the output variable to be expressed in deviation or normalized form. EXAMPLE 7.2 Figure 7.3 gives the response of the temperature T in a continuous stirred-tank reactor to a step change in feed flow rate w from 120 to 125 kg/min. Find an approximate first-order model for the process for these operating conditions. 128 (oo) w (kg/min) 120 wo} 160 Tho) 150 TCO 140 | I 130 oO 5 10 15 20, Time (min) Figure 7.3. Temperature response of a stirred-tank reactor for a step change in feed flow rate.7.2 Graphical Fitting ot First-Order Models Using Step Tests 171 Solution First note that Aw = M = 125 — 120 = 5 kg/min. Since AT = T(») — 1(0) = 160 — 140 = 20°C, the process gain is | Ate 20eG 1 ke ~ Aw 5 ke/min kg/min The time constant obtained from the graphical construction shown in Fig. 7.3 is 7 = 5 min. Note that this result agrees with the time when 63.2% of the response is complete, that is, T = 140 + 0.632(20) = 152.6 °C Consequently, the desired process model is EGE = Ws) Ss +1 where the gain has units of °C/kg/min Very few experimental plots of the step response show purely first-order be- havior because: 1. It is difficult to form a perfect step input. Usually process equipment, such as pumps and the control valves discussed in Chapter 9, cannot be changed in- stantaneously from one setting to another but must be ramped over a finite time. However, if the ramp time is small compared to the process time constant, a reasonably good approximation to a step input may be obtained. The true process model is neither first order nor linear. Only the simplest pro- cesses will exhibit such ideal dynamics. The output data are usually corrupted with noise, that is, they contain a random component in the measured output. Noise can arise from normal operation of the process, for example, inadequate mixing that produces eddies of higher and lower concentration (or temperature), or it can result from electronic instru- mentation. If noise is completely random, a first-order response plot may still be drawn that fits the output data well in some time-averaged sense. However, nonrandom noise, such as linear drift arising in the measurement instrumen- tation, can cause problems in the analysis. Another input (disturbance) may change during the step test without the op- erator’s knowledge. r » = Hence, normally there will be some departure from the curve in Fig. 7.2. To account for higher-order dynamics that are neglected in a simple first-order model, a time-delay term can be included. This additional dynamic element will improve the agreement between model and experimental responses. The fitting of a first-order plus time-delay model, Ke a G(s) (7-13) to the actual step response requires the following steps, as shown in Fig. 7.4 [3]: 1. The process gain K for the model is found by calculating the ratio of the change in the steady-state value of y to the size of the step change M in x.172 DEVELOPMENT OF EMPIRICAL DYNAMIC MODELS FROM STEP RESPONSE DATA Inflection point Figure 7.4. Graphical analysis of the process reaction curve to obtain parameters of a first-order plus time delay model. 2. A tangent is drawn at the point of inflection of the step response; the intersection of the tangent line and the time axis (where y = 0) is the time delay. 3. If the tangent is extended to intersect the steady-state response line (where y = KM), the point of intersection corresponds to time t = @ + 1. Therefore t can be found by subtracting @ from the point of intersection. The tangent method presented above for obtaining the time constant suffers from using only a single point to estimate the time constant. Use of multiple points may provide a better estimate. Returning to Eq. 7-10, note that introducing @ and rearranging gives the expression In ps . v@) since y(~), the final steady-state value, equals KM. In (7-14), y(%) — y,; can be interpreted as the incomplete response; dividing by y(*) yields the fraction incom- plete response. A semilog plot of [y(%) — yi]/y(®) vs. (f — 8) will then yield a straight line with slope = —1/z, from which an average value of the time constant is obtained. An equation equivalent to (7-14) for the variables of Example 7.2 (not in deviation form) would be In [F2=78] oe ed (7-15) (7-14) T(*) — TO) The major disadvantage of the time-delay estimation method in Fig. 7.4 is that it is difficult to find the point of inflection, due to measurement noise (errors) and small-scale recorder charts. The method of Sundaresan and Krishnaswamy [4] avoids use of the point of inflection construction entirely to estimate the dead time. They proposed that two times (1, and 2) be estimated from a step response curve, corresponding to 35.3 and 85.3% response times, respectively. The time delay and time constant are then estimated from the following equations: © = 1.3%, — 0.29% = 0.67(ty — th) 7 (7-16) A78 Fiting Second-Order Models Using Step Tests 173 These values of + and @ approximately minimize the difference between the mea- sured response and the model, in a least-squares sense [4]. Note that, using actual step response data, the parameters K, 7, and @ calculated for a first-order model approximation can vary considerably depending on the operating conditions of the process, the size of the input step change, and the direction of the change. These variations usually can be attributed to process nonlinearities. The methods considered above yield first-order models, only. If a first-order model is not sufficiently accurate, a second-order model (two time constants) can be estimated, as discussed in the next section. It is difficult to identify more than two time constants using graphical techniques. 7.3 FITTING SECOND-ORDER MODELS USING STEP TESTS Normally a better approximation to the step response can be obtained by fitting a second-order model to the data. Figure 7.5 shows the range of shapes that can occur for the step response assuming an overdamped model with no numerator dynamics. The transfer function used in Fig. 7.5, _ K ~ (as + Is + 1) includes two limiting cases: t,/t, = 0, where the system becomes first order, and 1/7, = 1, the critically damped case. The larger of the two time constants, 7;, is called the dominant time constant. The S-shaped response becomes more pro- nounced as the ratio of r)/7, becomes closer to one. Time constants for second-order systems which include time delays can be estimated using several graphical methods. Harriott [5] and Oldenbourg and Sar- torius (6] provide estimation procedures for the parameters of the following model: Ke~* (ms + I(ms + 1) Gis) (6-39) Gs) = (7-17) Thus, these methods apply only to overdamped systems. A method due to Smith {7] is more general since it yields a model of the form Keeu Gs) = = “<—_ 9 = 3 ts 1 (7-18) which includes both overdamped and underdamped systems. 1.0 08 0 1 2 3 ellry + 72) Figure 7.5. Step response for several overdamped second-order systems.