If you’re a controls or process engineer, then you’ve probably spent more time than you care to admit tuning PID loops, I certainly have! I shudder remembering some of the different methods I tried – a mix of guess and test procedures passed along or picked up from one job to the next. Fortunately, I was introduced to proper techniques before I damaged anything…or anyone!
More often than not Process Control is taught at university as part of the Chemical Engineering curriculum. While a majority of engineers graduate with another degree such as Electrical or Mechanical, a good portion of those same students eventually encounter a PID controller. That one controls course can make a world of difference as it can outline a range of methods for performing effective tests, calculating a process model, and tuning for a specific control objective. These methods tend to be very calculated and not real-world process focused. One approach that always strikes me as peculiar is Zeigler-Nichols (ZN).
It’s curious how the ZN method pops up routinely in discussions held among and between process engineers as though it’s their go-to approach for just about every loop. For good reasons – some very obvious – it should not be the go-to method no matter how simple the steps the results are rarely good enough. Consider these three (3) reasons as to why that’s the case:
- Quarter Decay Challenge
More often than not practitioners refer to the ZN method in terms of its objective outcome – ¼ decay. Essentially the ZN method starts with overshoot, meaning it accepts overshoot as the starting point. Only then does it seek to reduce the level of overshoot, decreasing it by 25% with each cycle.
For many processes overshoot simply isn’t acceptable. As one example: consider bioreactor processes where the cultured organisms are susceptible to significant spikes in temperature. ZN does not fulfill the requirement for disturbance rejection associated with those and other sensitive processes. Too many times I have seen quarter decay in a process response: It is no longer good enough as any overshoot is wasting a combination energy, time and materials and wastes it every response no matter how small!
- Trial and Error Costs
When calculating model parameters using ZN it’s necessary to first identify the inflection point along a response curve. Using that point a tangent line representing the maximum slope of the Process Variable is used to estimate the Process Gain (KP) value. Then the associated Controller Gain (KC) should be increased methodically so as to induce nearly unstable oscillatory behavior.
Given the current state of the art with process and software it seems counterintuitive that practitioners should rely on a trial and error practice such as this. When a method sets the expectation that multiple attempts should be expected, that’s when a technician or engineer should start researching for other options. That should be especially true when slow processes are involved as trial and error can be costly
- Instability or Insanity
Let’s revisit that last one - in order to tune a loop the ZN method directs practitioners to bring their processes to the edge of instability. With an understanding of the tipping point between stability and instability those same practitioners should then adjust the KP value so that oscillations steadily decay and the process eventually reaches a steady-state.
Let me simply ask: Whatever could go wrong? While the method seeks to establish the breaking point between a stable and an unstable process, putting a process – or even the whole plant – at risk isn’t an acceptable practice. What can go wrong will and…. Normally does! Leaving a process just off its stability tipping point needs good justification and understanding of where it sits in the overall process, not just because it was “in the method”
Do a quick Internet search for articles about ZN and you’re likely to find that most begin with warnings. Don’t apply the ZN method if the process tends to oscillate when the loop is operated in manual. Don’t apply the ZN method for temperature loops or others that can be considered unstable. Don’t apply ZN if the associated final control element is subject to Stiction. What you probably won’t find is an article advocating for the use of ZN. Attempting to apply a method that raises so many red flags should be used with extreme caution with an understanding and consideration of its impacts.
In previous posts I’ve written about the importance of good data when tuning PIDs and I’ve outlined a simple, repeatable method for tuning PIDs. Before you damage something – or someone – consider this approach!