Simulink Control Design
This example shows how to design an array of PID controllers for a nonlinear plant in Simulink that operates over a wide range of operating points.
Opening the Plant Model
The plant is a continuous stirred tank reactor (CSTR) that operates over a wide range of operating points. A single PID controller can effectively use the coolant temperature to regulate the output concentration around a small operating range that the PID controller is designed for. But since the plant is a strongly nonlinear system, control performance degrades if operating point changes significantly. The closed-loop system can even become unstable.
mdl = 'scdcstrctrlplant'; open_system(mdl)
For background, see Seborg, D.E. et al., "Process Dynamics and Control", 2nd Ed., 2004, Wiley, pp.34-36.
Introduction to Gain Scheduling
A common approach to solve the nonlinear control problem is using gain scheduling with linear controllers. Generally speaking designing a gain scheduling control system takes four steps:
Obtain a plant model for each operating region. The usual practice is to linearize the plant at several equilibrium operating points.
Design a family of linear controllers such as PID for the plant models obtained in the previous step.
Implement a scheduling mechanism such that the controller coefficients such as PID gains are changed based on the values of the scheduling variables. Smooth (bumpless) transfer between controllers is required to minimize disturbance to plant operation.
Assess control performance with simulation.
For more background reading on gain scheduling, see a survey paper from W. J. Rugh and J. S. Shamma: "Research on gain scheduling", Automatica, Issue 36, 2000, pp.1401-1425.
In this example, we focus on designing a family of PID controllers for the CSTR plant described in step 1 and 2.
Obtaining Linear Plant Models for Multiple Operating Points
The output concentration C is used to identify different operating regions. The CSTR plant can operate at any conversion rate between low conversion rate (C=9) and high conversion rate (C=2). In this example, divide the whole operating range into 8 regions represented by C = 2, 3, 4, 5, 6, 7, 8 and 9.
In the following loop, first compute equilibrium operating points with the
findop command. Then linearize the plant at each operating point with the
% Obtain default operating point op = operspec(mdl); % Set the value of output concentration C to be known op.Outputs.Known = true; % Specify operating regions C = [2 3 4 5 6 7 8 9]; % Initialize an array of state space systems Plants = rss(1,1,1,8); for ct = 1:length(C) % Compute equilibrium operating point corresponding to the value of C op.Outputs.y = C(ct); opoint = findop(mdl,op,findopOptions('DisplayReport','off')); % Linearize plant at this operating point Plants(:,:,ct) = linearize(mdl, opoint); end
Since the CSTR plant is nonlinear, we expect different characteristics among the linear models. For example, plant models with high and low conversion rates are stable, while the others are not.
ans = 1 1 0 0 0 0 1 1
Designing PID Controllers for the Plant Models
To design multiple PID controllers in batch, we can use the
pidtune command. The following command will generate an array of PID controllers in parallel form. The desired open loop crossover frequency is at 1 rad/sec and the phase margin is the default value of 60 degrees.
% Design controllers Controllers = pidtune(Plants,'pidf',pidtuneOptions('Crossover',1)); % Display controller for C=4 Controllers(:,:,4)
ans = 1 s Kp + Ki * --- + Kd * -------- s Tf*s+1 with Kp = -12.4, Ki = -1.74, Kd = -16, Tf = 0.00875 Continuous-time PIDF controller in parallel form.
Plot the closed loop responses for step set-point tracking as below:
% Construct closed-loop systems clsys = feedback(Plants*Controllers,1); % Plot closed-loop responses figure; hold on for ct = 1:length(C) % Select a system from the LTI array sys = clsys(:,:,ct); sys.Name = ['C=',num2str(C(ct))]; sys.InputName = 'Reference'; % Plot step response stepplot(sys,20); end legend('show','location','southeast')
All the closed loops are stable but the overshoots of the loops with unstable plants (C=4, 5, 6, and 7) are too large. To improve the results, increase the target open loop bandwidth to 10 rad/sec.
% Design controllers for unstable plant models Controllers = pidtune(Plants,'pidf',10); % Display controller for C=4 Controllers(:,:,4)
ans = 1 s Kp + Ki * --- + Kd * -------- s Tf*s+1 with Kp = -283, Ki = -151, Kd = -128, Tf = 0.0183 Continuous-time PIDF controller in parallel form.
Plot the closed-loop step responses for the new controllers.
% Construct closed-loop systems clsys = feedback(Plants*Controllers,1); % Plot closed-loop responses figure; hold on for ct = 1:length(C) % Select a system from the LTI array sys = clsys(:,:,ct); set(sys,'Name',['C=',num2str(C(ct))],'InputName','Reference'); % Plot step response stepplot(sys,20); end legend('show','location','southeast')
All the closed loop responses are satisfactory now. For comparison, examine the response when you use the same controller at all operating points. Create another set of closed-loop systems, where each one uses the C = 2 controller.
clsys_flat = feedback(Plants*Controllers(:,:,1),1); figure; stepplot(clsys,clsys_flat,20) legend('C-dependent Controllers','Single Controller')
The array of PID controllers designed separately for each concentration gives considerably better performance than a single controller.
However, the closed-loop responses shown above are computed based on linear approximations of the full nonlinear system. To validate the design, implement the scheduling mechanism in your model using the PID Controller block.
Close the model.