Tuning PID controllers is a multi-step process. While it is important to understand each step in the process, performing the bump test and collecting dynamic data is the most crucial step that generally dictates the outcome. This post delves into one particular type of test – the Step Test – before introducing other tests commonly used in PID controller tuning.
The purpose of any test is to capture data that can be used to model a process’ dynamic behavior. A good test will reveal the relationship between a control loop’s Process Variable (PV) and the associated Controller Output (CO). That can be accomplished either in open-loop by manually adjusting the Final Control Element (FCE) or in closed-loop by changing the Set Point (SP). This post focuses on closed-loop testing.
While the word “good” is subjective, there are some guidelines for performing a Step Test that should be followed to capture appropriate dynamic data for tuning. First, the Step Test should involve a large enough change to the SP so that it results in a clear PV response without putting either the process or plant personnel at risk. Best-practice is to change the SP by an amount that’s equal to at least 4 times the process noise band. From the above graphic, the process noise is estimated at +1 units. To ensure that the PV is moved far enough outside of this noise band, the SP is stepped from 70 to 75 units.
This Step Test produced the required data to model and tune the PID. As seen in the previous trend, the SP change caused a clear response in the PV and CO signals. This data is used to determine an dynamic model of the process. This model is then used to determine appropriate tuning parameters for the PID loop. Since the Step Test is the simplest test that may be performed, it can be analyzed by hand. Alternatively software may be used to automatically generate parameters for use with your PID controller.
While the Step Test may seem simple, there are several things to keep in mind when tuning PIDs:
- Did your test start and end with the process at a steady-state? In most cases your data must be held at steady-steady in order to produce results appropriate for modelling. This is particularly true when applying manual tuning techniques or when using most commercial software packages.
- Did you account for process noise? As mentioned earlier it is essential that the SP step performed moves the PV at least 4 times the noise band.
- Did the process experience any disturbances during your test? Disturbances that occur during the test renders the results unusable as the data reflects the disturbance as well as the process’ normal dynamics.
Different Kinds of Tests:
Other than the Step Test, there are three (3) other tests commonly used in industry.
The Bump Test can be thought of as two (2) Step Tests performed in rapid succession and in opposite directions. The SP is first stepped in one direction. Once the measured PV shows a clear response, the SP is then stepped in the opposite direction, returning to its original value.
Similarly, the Doublet Test is two (2) Bump Tests performed in rapid succession and in opposite directions. The second bump is implemented as soon as the process shows a clear response to the first bump. Note that the second bump returns the SP to its original value.
Finally, a Pseudo-Random Binary Sequence (PRBS) Test is a sequence of SP pulses – or Bump Tests – that are uniform in amplitude, alternating in direction, and of random duration. While more changes to the SP are involved, the PRBS Test allows each change to be of lesser magnitude.
Choosing the right test for your process depends on several factors. From experience, the Doublet Test provides the best data for tuning. More specifically a series of three (3) Doublet Tests frequently provides the best results. This allows a user to gather process data from nine (9) separate step tests. It also allows for three (3) different reactions to your process which allows for a fuller picture of data.
If you need additional guidance on how to perform a test or how to tune your PIDs for optimal performance, consider attending a training workshop that focuses on understanding process dynamics and the PID controller or contact us.