Vous êtes sur la page 1sur 14

1

Process Controls Laboratory Report


Modern process control is based upon the principle of automatic feedback loops which continually measure process variables and adjust the controller output to achieve desired set point values. In the majority of process feedback controllers, proportionalintegral-derivative (PID) controllers are used. Mathematically, the dependent, ideal PID controller equation can be expressed as:

(1)

where u(t) is the controller output signal (CO), ubias is the controller bias, e(t) is the controller error, and y(t) is the measured process variable (PV) [1]. This equation

includes the three different types of controllers; P-only control which includes only the proportional term, PI control which includes both the proportional and integral terms, and PID control which includes both aforementioned terms plus the derivative term. To optimize a PID controlled process, controller tuning may be required. PID controller tuning refers to the adjustment of the controller gain, Kc, reset time, I, and derivative time, D. A set of correlations used for tuning the P-only, PI and PID controllers include the ITAE (Integral Time Averaged Error) and IMC (Internal Model Control) correlations (Appendix C) [1]. Using these correlations, a range from conservative to aggressive tuning parameters can be calculated and used in the controller tuning equation. To assess the performance characteristics of each controller, such terms as rise time, offset, peak overshoot, decay ratio and settling time are considered. A brief description of these terms is included in Appendix B.

One widely used method to deduce the initial values for tuning variables is called the First Order Plus Dead Time (FOPDT) model. The equation for the time domain form for the FOPDT model is: (2) By fitting an FOPDT model to process data, values for the process gain, KP, process time constant, P, and process dead time, P, can be determined. The process gain describes how the process variable changes with controller output. The process time constant is a measure of how fast the process variable responds and is the time it takes the process variable response to reach 63.2% of its total change in response to a change in controller output. The process dead time is the amount of time that passes from the start of a step change in controller output to the time the process variable shows a clear response. The equations to calculate these FOPDT parameters are included in Appendix C. These FOPDT variables describe the dynamic behavior of the process and are important because they are used in the ITAE and IMC correlations to compute the controller tuning values [1]. In this experiment, the dynamic behavior of a liquid level control process was analyzed and FOPDT based controller tuning parameters determined to assess the performance of P, PI and PID controllers. The apparatus consisted of two draining tanks, 4 in. diameter and 24 in. length, in vertical series above a 12 gallon holding tank. Water was drawn from the holding tank and pumped into the top tank. A controller was connected to the lower tank level transmitter and the pump. The controller could be

placed in both manual and automatic mode to control the process. The process gain, reset rate (1/I) and derivative time could also be input into the controller to have it function in P, PI or PID mode. The diagram of the apparatus is shown in Figure 1. To collect and analyze the process data, the software program Loop Pro was used.

Figure 1: Experimental apparatus for liquid level control experiment.

The experimental procedure consisted of first determining the dynamic nature of the process by performing step tests in manual mode. Following this, an FOPDT model was fit to the process by performing a doublet bump test. From this model, initial tuning parameters were determined for a moderately-aggressive controller. Set point tracking performance of moderately-aggressive P, PI and PID controllers were then examined by altering the appropriate tuning parameters. Finally, the disturbance rejection performance of the PI controller was analyzed by introducing a 500mL aliquot of water to each of the top and center tanks. For further details on the procedure see reference [2].

To convey the cause and effect relationship of the process, a control loop block diagram as shown in Figure 2 can be used to describe the liquid level controlled, closed loop operation.

Figure 2: Control loop block diagram for liquid level controlled draining tanks in series.

As most real world processes are non-linear in nature, the draining tanks experiment was examined to assess its dynamic behavior. Three successive step tests were performed in 10% increments for a controller output from 20 to 50% (Figure 3) while in manual (open loop) mode. For each step, the process gain, time constant and dead time were calculated by Loop Pro utilizing the equations in Appendix C. The resulting values are shown in Table 1. An example of how the KP, P and P can be calculated graphically is included in Appendix C.

Figure 3: Step tests to examine the dynamic behavior of the draining tanks process. Table 1: FOPDT Model Parameters
Controller Step 20-30% 30-40% 40-50% Kp (in/%) 0.33 0.42 0.45 p (s) 45.0 60.9 79.0 p (s) 14.0 16.6 15.9

If this process were linear, then the process gain, time constant and dead time for each successive step test would be equivalent. It is shown however, that this process is non-linear as these values vary as the CO changes. Due to the non-linearity of this process, the best approach to determine initial tuning parameters would be to perform a doublet test which steps both above and below the process design level of operation (DLO) to account for the dynamic behavior. Therefore the process was stepped 10% CO around the DLO of 37% (10 in.) and a moderately- aggressive FOPDT model was fit to the data as shown in Figure 4. Utilizing Loop Pros Design Tools, the P-only, PI and PID tuning parameters for the FOPDT model were determined and are shown in Table 2. Loop Pro utilizes the ITAE and IMC tuning correlations given in Appendix C to compute each tuning parameter.

Figure 4: FOPDT model fit to doublet test for moderately aggressive controller. Table 2: Moderately aggressive tuning parameters from FOPDT model.
Controller P-only PI PID KC 1.97 1.57 2.34 I (s) 45.74 58.06 D (s) 9.71

The performance of each of the three types of controllers varies due to the differing components of controller equation (1). In P-only control, the only adjustable tuning parameter is KC as the proportional term is the only term in the corresponding controller equation. The proportional term adds or subtracts from CObias based on how far the PV is from the SP at any instant of time. The setpoint tracking of a P-only controller was examined by changing KC to its original value and also doubling it as shown in Figure 5.

Initial Case

Figure 5: Setpoint tracking of P-only controller by changing tuning parameter KC: 0.98, 1.97, 3.93.

The advantage of P-only control is that there is only one tuning parameter to adjust and therefore the best tuning values are obtained rather quickly, however as Figure 5 demonstrates, the disadvantage to P-only control is that it permits offset. To minimize offset, KC may be increased, however this results in greater oscillatory behavior. Offset for KC values of 0.98, 1.97 and 3.93 were 1.7 in., 1.2 in., and 0.2 in. respectively. In PI control, there is both the proportional and integral term of equation (1). The integral term considers the error (e(t) = PV-SP) over time and based on the tuning parameters for KC and I, will continually add or subtract from CObias and cause the controller output to change. The advantage to PI control is that it eliminates the offset present in P-only control by minimizing the integrated area of error over time. To assess the effect changing the two tuning parameters has on a PI controller performance, both KC and I were halved and doubled as shown in Figure 6.

Initial Case

Figure 6: PV SP tracking of PI controller by changing tuning parameters KC: 0.787, 1.574, 3.148 and I: 22.87s, 45.74s, 91.48s. KC increases along the positive y-axis and I increases along the positive x-axis.

From Figure 6 it can be determined that the greatest oscillatory behavior occurs when KC is at its largest value (3.148) and I is at its smallest value (22.87s). In this process, using these tuning parameters actually resulted in increased magnitude of oscillations over time and thus an unstable system. Either lowering I, or increasing KC from the initial value resulted in a greater peak overshoot, larger settling time and larger decay ratio (Appendix C). However it can be seen that halving I has more of an effect than doubling KC as the decay ratio is larger with 60% versus 43%, the settling time is longer with 675s versus 440s, and the peak overshoot is larger with 83% versus 70% (while holding the other variable at the initial value). In PID control all three terms in equation (1) are utilized. The function of the derivative term is to determine the rate of change of the error (slope) and then add or subtract to CObias and thus influence the controller output. A rapidly changing error will have a larger derivative and therefore a larger effect on controller output. The derivative term will therefore work to decrease the oscillatory behavior in the process variable. To assess the effect of changing derivative time, a comparison of the tuning parameter D was made for the PID controller by halving and doubling the initial value of 9.71s (Figure 7).

Initial Case

Figure 7: PV SP tracking of PID controller changing tuning parameter D: 0.081s, 0.162s, 0.324s.

From Figure 7 it can be determined that increasing the derivative time results in less oscillatory behavior of the process variable however there is also an increased noise in the controller output. Increasing D also increase rise time, settling time, and decreases peak overshoot as shown in Table 4.
Table 4: Performance comparison of PID tuning parameter derivative time.
Performance Parameter Rise Time (s) Settling Time (s) Peak Overshoot (%) D (0.081s) 35 105 37.5 D (0.162s) 43 128 32.5 D (0.324s) 60 170 20

In normal operation, disturbances to a process may occur. To assess the disturbance rejection performance of this experiment, a 500mL aliquot of water was added to the lower tank while operating a PI controller in moderate, moderately aggressive and aggressive modes (Figure 8). The tuning parameters used for the moderate PI controller and aggressive PI controller can be found in Appendix C.

Moderate

Moderately-Aggressive

Aggressive

Figure 8: Moderate, moderately aggressive and aggressive response to unmeasured disturbance.

From the figure above it can be seen that as the controller becomes more aggressive oscillatory behavior increases, settling time decreases, the decay ratio increases, while peak overshoot remains generally the same in response to the disturbance. As the moderate PI controller exhibits the best behavior with regards to setpoint tracking, its response to further

10

disturbance rejection was tested by adding a 500mL aliquot to the upper tank. This response is shown in Figure 9 below.

Based upon the moderate PI controller response to the second disturbance, the magnitude of the oscillations decreases, the settling time improves and the decay ratio decreases. This would be expected to occur as the level transmitter is located on the lower tank and therefore a change in the lower tank level would be immediately seen by the controller whereas adding water to the upper tank does not significantly immediately increase the flowrate into the lower tank. In this study of P-only, PI and PID controllers on the draining tank process, there were found to be certain advantages and disadvantages to each controller. The P-only control exhibited offset in the process variable upon reaching steady state. In PI control, integral action eliminated the offset but resulted in more oscillatory behavior in the process variable. In PID control derivative action worked to eliminate the oscillations however any noise present in the PV signal was reflected and amplified in the controller output signal which would result in mechanical wear of final control elements.

Figure 9: Moderate PI controller response to upper tank disturbance.

11

In choosing the best performing controller it must be noted that best performance is subjective, meaning that some processes may desire a PV response with no overshoot, others may be able to tolerate overshoot and prefer faster rise times. For a process that desires fast rise time with the minimal amount of oscillatory behavior and overshoot it would be suggested to use a moderate to moderately aggressive PI controller. This would results in good setpoint tracking, no CO noise, minimal oscillatory behavior, fast rise and settling times along with good disturbance rejection. For this experiment this would be a PI controller with KC, I values of 3.148 and 91.48s or 1.575 and 91.48s or 1.574 and 45.74s or similar values within these ranges. Appendix A: References Hidden

Appendix B: Nomenclature CObias: The value of the controller output signal when in manual mode, causes the PV to settle at the DLO such that error = 0 when the disturbances are at their normal or expected values. Decay ratio: The size of the second peak above the new steady state divided by the size of the first peak above the same steady state level. Offset: Most notably associated with P-only controllers, is the difference from the SP to where the PV settles out at a steady state value. Peak overshoot: The size of the first peak above the new SP divided by the size of the SP step.

12

Rise time:

The time from when the CO begins to step to when the PV to reaches the setpoint (SP).

Settling time:

The time at which the PV reaches 5% of the total change in the process variable (PV).

Appendix C: Data and Sample Calculations ITAE (P-only) and IMC Tuning Correlations

Graphical determination of FOPDT parameters. Step test from 20-30% CO output.

13

PV63.2 = PVinitial + 0.632 (PV) = 3.5+0.632(3.15) = 5.49 in.

Performance Characteristics of Changing Tuning Parameters of PI Controller DR: decay ratio, ST: settling time, PO: peak overshoot, RT: rise time KC: 3.148 I: 22.87s UNSTABLE SYSTEM KC: 3.148 I: 45.74s DR: 43% ST: 440s PO: 70% RT: 30s KC: 1.574 I: 45.74s DR: 21% ST: 235s PO: 48% RT: 45s KC: 0.787 I: 45.74s DR: 0% ST: 300s PO: 23% RT: 75s KC: 3.148 I: 91.48s DR: 27% ST: 232s PO: 44% RT: 38s KC: 1.574 I: 91.48s DR: 11% ST: 174s PO: 23% RT: 50s KC: 0.787 I: 91.48s DR: 0% ST: 110s PO: 0% RT: 115s

KC: 1.574 I: 22.87s DR: 60% ST: 675s PO: 83% RT: 34s KC: 0.787 I: 22.87s DR: 10% ST: 323s PO: 49% RT: 57s

Note: above values are estimated determined from graphical analysis

14

Summary of Tuning Parameters


Tuning Moderate (Conservative) Controller ModeratelyAggressive Controller Aggressive Controller Controller P-only PI PID P-only PI PID P-only PI PID KC 1.24 0.59 0.80 1.97 1.57 2.34 3.12 3.11 5.57 I (s) 45.74 58.06 45.74 58.06 45.74 58.06 D (s) 9.71 9.71 9.71

Appendix D: Lab Notebook

Vous aimerez peut-être aussi