## Tuning Tips – How to Improve Your Results

I have read several posts on LinkedIn where the writers state that tuning rules don’t work. Well, I politely argue that it is not the rules that don’t work. You have to know how to apply the rules properly and what to expect from them. It’s not rocket science, but if you miss a piece, your calculated tuning settings might not work. So I provide this checklist with tips to give you a better understanding of what’s involved with tuning a controller. Hopefully it will improve your results.

**Valve Performance**

Is the control valve working properly? (See article on control valve problems.) Dead band can severely affect your step-test results. Stiction and positioner overshoot can cause oscillations regardless of how well the controller is tuned.

Execute specific tests that check for dead band, stiction, and positioner overshoot in the control valve.

**Step-test Procedure**

Begin with a steady process variable and make a step-change large enough that the process variable’s response is clearly visible above the noise/disturbance level. A good rule-of-thumb is that the process variable must move five times as much as the peak-to-peak noise/disturbance level. You have to make measurements, and if the signal-to-noise ratio is too low (step too small) major errors could be made. If you can’t get the process variable steady enough you may have problems elsewhere that should be addressed first. There is a procedure for step-testing at this blog.

Also, do multiple step-tests so that you can compare the calculations from different step tests with each other. If you do just one step-test, you won’t know if a disturbance affected the process during your test. For most loops I do four step tests. I have had to do more than a dozen step tests on some volatile processes to get good average values for process response. Take the time, and get good results.

**Step-Test Measurements**

Make sure you know how to measure process gain, dead time and time constant from a process response curve. Do this for each step test. An Excel spreadsheet could help if you don’t have tuning software. Compare the numbers from each step test, remove outliers, take the average of the remaining values.

**Time Base**

All the popular tuning rules assume that you are making time measurements in the same units as those used by your controller. Are your controller’s integral and derivative time settings in minutes or seconds? Convert your measured values to match you integral and derivative time units, if necessary.

**Scaling**

All tuning rules assume the process variable and controller output measurements are normalized. That means that changes in process variable have to be divided by the span of the measurement device, and changes in controller output have to be divided by the span of the controller output. The latter is normally 0% to 100%, but on some DCSs the controller output range can be different for outer loops in cascade control to match the range of the process variable.

**Tuning Rule, Robustness, and Control Objective**

The Ziegler-Nichols tuning rules were designed to provide ¼-amplitude decay, which is undesirable for most processes. The control loop is also not very robust – it can easily go unstable if the process gain or dead time increases. These two problems can easily be solved by dividing the calculated controller gain by two. Note that the Ziegler-Nichols tuning rules result in sluggish loops if the process dead time is longer than the time constant. See my comments on Z-N tuning rules for more detail.

The Cohen-Coon rules were also designed to provide ¼-amplitude decay, and have the same robustness problem as the Ziegler-Nichols rules. These can easily be solved by dividing the calculated controller gain by two.

Minimum IAE tuning rules give something close to ¼-amplitude decay, and have the same robustness problem as the Ziegler-Nichols and Cohen-Coon rules. The solution is to divide the calculated controller gain by two.

The Chien-Hrones-Reswick (CHR) tuning rules come in two sets, one for 20% overshoot (not recommended for most processes because of overshoot and low robustness) and a 0% overshoot rule (which is more robust and okay to use). Note that the CHR tuning rules result in sluggish loop response if the process dead time is longer than the time constant.

The Lambda and Internal Model Control (IMC) tuning rules give very stable response (robust control loops), and no overshoot if applied correctly. But loops with long time constants respond sluggishly to disturbances. See article on Lambda tuning for more detail.

High-end tuning software should allow you to select your tuning objective, and calculate tuning settings accordingly. High-end tuning software should also warn you when calculated settings will result in a control loop with low robustness.

**Controller Algorithm**

Some tuning rules (like Ziegler-Nichols) have been developed for interactive PID algorithms, while others (like minimum IAE) have been developed for noninteractive algorithms. There are conversions available to go from PID settings on one type to the other. Note that if you don’t use derivative (most people don’t), there is no difference between interactive and noninteractive algorithms. See this article on controller algorithms.

A few DCSs and PLCs have parallel controller algorithms, and you have to convert your calculated integral and derivative settings for use on a parallel algorithm.

**Integral’s Unit of Measure**

All popular tuning rules assume your controller’s integral setting is in units of time (as in minutes or seconds), and not the inverse (as in repeats per minute or repeats per second). Invert the calculated integral time if necessary.

**Controller Gain’s Unit of Measure**

All popular tuning rules assume your controller has controller gain, and not proportional band. Does your controller use proportional band or gain? Convert your calculated controller gain to proportional band if necessary.

**Process Linearity**

Tuning rules assume your process gain, dead time, and time constant don’t ever change. If you have a process where these numbers change significantly (increase more than 50% or decrease more than 33%), this could severely affect loop response and stability. You should consider linearizing the final control element or measurement, or implement controller gain scheduling. (And you can schedule integral and derivative time too – it’s still called gain scheduling).

**Tools**

Your effectiveness and results will not only depend on what you know about controllers, tuning, and the process you are controlling, but also on what tools you are using.

Tuning software automatically identifies the process model, calculates controller settings that match the controller algorithm and its units of measure, and provides simulations of expected loop response after tuning. This can be well worth the money you pay for a software tool.

Manual tuning (not using software) can be greatly simplified by using an Excel spreadsheet to help calculate process model parameters and controller settings. If you are not using software, take the time and compile a spreadsheet to help you.

Let me know if you have questions or need more information.

Stay tuned!

Jacques Smuts – Author of the book Process Control for Practitioners