Temperature control of a laminator prototype
One of our clients is developing a novel laminator and ordered a british manufacturer to make a prototype. The picture above does not show this laminator, but another one to get an impression of how it looks like. An essential part of the laminator is sufficiently accurate temperature control of products inside the laminator. The prototype was not able to maintain temperature within specifications. Using the PID Tuner and consultancy of DotX, the control problem was analysed and resolved. Please note that all references to specific details, like numbers, have been removed in this post, due to agreements between DotX and our client.
The laminator prototype was equiped with standard Omron temperature control hardware modules (CJ1W-TC002). These modules contain PID algorithms, but the code cannot be changed. As a result, one cannot analyse the control challenge properly, and there are no options to improve PID control with add-ons.
Hardware changes
The laminator prototype was equipped with standard Omron Temperature Control Modules (CJ1W-TC002). These modules contain PID algorithms, but the controller code cannot be changed. As a result, one cannot analyse the control challenge properly, and there are no options to improve PID control (by adding other algorithms), should it be necessary.The laminator prototype was already equipped by an Omron PLC for handling product transport, communicating with the Human Machine Interface (HMI), etcetera. We therefore advised our client to remove the non-programmable Temperature Control Modules and move the temperature control tasks to the Omron PLC, and they agreed. This required some minor rewiring and programming (executed by DotX). The hardware was now suitable for analysis.
We then used the PID Tuner in on-line mode to perform step experiments (to create data for identifying the process models). Since there were three temperatures to be controlled (by three adjacent heaters), we expected cross effects or 'interactions' . Therefore, we created one project in the PID Tuner with all three PID loops. This allows the PID Tuner to simulate the cross effects, and tune the PID controllers, taking interactions into account. Figure 1 shows the results of the open loop identification experiment.
Figure 1: Open loop response to step change on Manipulated Variable (heat input)
Analysis
We then fit a model to the responses (see blue line in Figure 1) and tried PID tuning. The interactions appeared to be sufficiently small to be ignored. We then proceeded to tune the PID controllers and tested them in closed loop. It is a good practice to perform a closed loop verification experiment. The verification step is now standard included in the PID Tuner.Figure 2 shows the result of a verification experiment.
Figure 2: Closed loop response to setpoint change, for verification. Since the simulated response matches the measured response, this confirms that the model of the process seems to be suitable (for control design), and that the PID controller responds as simulated.
First production tests
The PID controllers were tuned well, and the closed-loop response behaved as expected (near optimal) during a test where the setpoint changed, without being perturbed. The next step was to test the controllers during normal production. During normal production, the temperature is perturbed at regular intervals, when a new product enters the laminator. The temperature then first drops and then recovers, due to the PID controllers that increase the heat input.Figure 4 shows the result of the first test during production. Clearly, the temperature deviates, on average, from setpoint. How is this possible?
Figure 4: First production test: temperature deviates, on average, from setpoint.
Removing offset
The problem of an offset was not expected by us, since the PID uses Integral action. However, the PID algorithms usually contain something known as “integral windup”. This means that the integration is stopped as soon as the MV (i.e. the PID output) hits the minimum value, usually 0%, or its maximum value 100%. The integral wind up avoids the integration part to exceed these limits. If you remove these limits, or if your would remove the integral windup, the MV could “windup” to a large negative (-10000) or positive number, and the PID then becomes useless. On the other hand, when hitting a limit, its integral action is not functioning anymore, and on average, the PV can deviate from setpoint.To resolve this issue, the easiest solution is to change the limits such that the PID output does not hit its limits during normal production. Some more advanced changes to the PID algorithm have been proposed in literature, but in many practical cases, your PID algorithm will be a fixed ‘block’ that you cannot change. We changed the minimum MV to -10%. This way, our PID algorithm is ‘fooled’ and thinks it can go to -10%, while the actual MV will never drop below 0%. Anti-windup kicks in only if the MV drops below this -10%.
Figure 5 shows the temperatures and setpoint after this change to the MV limits, during production. The temperature now shows no offset, on average.
Figure 5: Successful production test.