When tuning PID controllers, what sample rate is required?

Sample Rates may differ from one loop to the next...

The Importance of Accurate Data Doesn't Change!

When it comes to data collecting requirements and controller tuning, practitioners frequently follow simple recommendations. When analysing the dynamics of a PID control loop, these "rules of thumb" ensure that there is sufficient data resolution. The capacity of a process engineer to understand the dynamics and tweak for better control can be harmed by a lack of reliable data. Any tuning process will be hit-or-miss at best if the sample rate is too sluggish. So, what are those rules, exactly?

From simple "rules of thumb" to scientific methodologies, there are a variety of approaches (e.g. Nyquist-Shannon sampling theorem). On the simpler end of the spectrum, sampling rates of 1 second or quicker for Flow loops and 1-2 seconds for Pressure loops are usual. Changes in Controller Output are immediately detected by such loops. Many Temperature and Level loops, which have slower dynamics, follow closely after with rates of 2-5 seconds.

The rules are a little more extensive when examining sample rate from a scientific or mathematical standpoint, and they are outside the scope of this blog post. Nonetheless, it's considered best practise to base the sample rate on frequency or time-based components of the process data. The usage of data that is 5-10 times quicker than the Process Time Constant or 3-5 times faster than the Process Dead-Time is suggested by basic formulas.

Other issues should be kept in mind in addition to these approaches:

Aliasing

Aliasing is a trick of the eye. It arises when the sample rate is insufficient to accurately portray the dynamics of a process. When this type of data is collected and plotted, it can produce results that are completely wrong and deceptive. Using this type of data to generate controller tuning parameters might be ineffective and even dangerous.

Compression

Compression is a data storage technique that helps manufacturers keep their IT costs down. One method is to remove data on a set schedule. Another aspect of change is the removal of data that does not change. Compression, regardless of the technology used, can cause aliasing by reducing the data resolution required for process modelling and PID controller tuning.

Software

Controller tuning software requires "excellent data" in order to produce good results. As a result, it's no surprise that the value of software drops as the sample rate slows. When working with bad data, software can model the Process Gain, but it can't produce meaningful numbers for the Process Time Constant or Process Dead-Time. In such cases, traditional trial-and-error testing to determine functional values for the corresponding controller tuning coefficients is usually required.

PID control loop tuning and monitoring plant-wide control loop performance are both aided by guidelines for applying an effective sample rate. Practitioners are effectively flying blind and limited in their capacity to maintain safe and profitable operations if the data lacks sufficient resolution. Most process control courses include best-practices for data collecting and analysis as part of the curriculum.