PID tuning tips
In this post we provide you with pragmatic PID tuning tips
by Jan Schuurmans, updated at 14-6-2019
Estimated reading time: 1 minute
Once you have obtained the step experiment data you can fit a model to these data, and tune the PID. The first tip is that we advice you to check if you know for certain how the PID in the control computer behaves. Does it not use scaling? Has the PID in the control computer been programmed in standard form? It is quite hard to find an answer to these questions, but there is an easy way to check this.
In the PID Tuner, go to the verifcation step, and perform a verifcation experiment with the PID tuned with 'old ' parameters. For instance, perform a step experiment on the setpoint, or on the PMV (perturbation on manipulated variable). So, do not use the 'new' tuning parameters of the PID Tuner yet.
Then, generate a report and check if the simulated response agrees with the measured response. If it does, like in the graph shown below (showing an actual response matching the simulated one) you can safely apply the 'new' PID tuning as computed by the PID Tuner. If it does not (at all) you will have to find out the cause of the difference. So far, we have seen the following causes:
The other causes, mentioned above (PID block operating at erroneous sample period, and inaccurate logging) can sometimes be regognised as 'block-like' signals. In some PLCs, like Allen Bradley, you can log signals in the PLC at very high frequency and accuracy. It is not a bad idea to use these signals as a check at least once.
Estimated reading time: 1 minute
PID tuning tips
In this post we provide you with some tips to tune PID loops programmed in some control computer (PLC, DCS, or whatever). We assume you have performed a step experiment on the process that needs to be controlled by this PID that you want to tune. Furthermore, we assume that the PID has been tuned already, but you want to check if the tuning is ok, or can be improved.Once you have obtained the step experiment data you can fit a model to these data, and tune the PID. The first tip is that we advice you to check if you know for certain how the PID in the control computer behaves. Does it not use scaling? Has the PID in the control computer been programmed in standard form? It is quite hard to find an answer to these questions, but there is an easy way to check this.
In the PID Tuner, go to the verifcation step, and perform a verifcation experiment with the PID tuned with 'old ' parameters. For instance, perform a step experiment on the setpoint, or on the PMV (perturbation on manipulated variable). So, do not use the 'new' tuning parameters of the PID Tuner yet.
Then, generate a report and check if the simulated response agrees with the measured response. If it does, like in the graph shown below (showing an actual response matching the simulated one) you can safely apply the 'new' PID tuning as computed by the PID Tuner. If it does not (at all) you will have to find out the cause of the difference. So far, we have seen the following causes:
- Scaling in the PID logic
- PID block operating at erroneous sample period
- inaccurate logging of PID signals
- System's dead time is shorter than 0.1 sec
The other causes, mentioned above (PID block operating at erroneous sample period, and inaccurate logging) can sometimes be regognised as 'block-like' signals. In some PLCs, like Allen Bradley, you can log signals in the PLC at very high frequency and accuracy. It is not a bad idea to use these signals as a check at least once.
