Sometimes it can feel like PID tuning is a cross between engineering, mathematics and the Dark Arts. There are some definitive engineering and mathematical steps to take when tuning a loop (for the first time or for the ninetieth), but there is also a lot to be said for experience and understanding how to tweak things based on the loop makeup and response you are seeing.
From an integrator perspective, we typically only interact with the tuning parameters of a loop during the start-up and; once the system commissioning is complete and unless we are called back in to “re-adjust”, what happens to those loop tuning parameters is unknown and falls into the category of mysterious. We always wonder, do operators turn loops into manual (or even off) to control the system themselves? (Happens more often than people care to admit.) Do different operators feel they have a better handle over the process than others and frequently tweak the loop to their ‘right’ numbers? (Again, more than we care to mention). Forget about manual human intervention for a minute, do those loops responsiveness lessen over time due to degrading equipment and real world conditions? (They certainly do.) Finally, does the initial system process design always capture all of the different phases and modes of operation where alternate PID tuning parameters would be better than the originals? (Rarely is the design perfect.)
In our experience, we often see loops that are set up and tuned initially and then forgotten about or ignored unless something “major” happens. If “nothing major happens”, the lack of attention to those important parameters creates a breeding ground for quality issues, significant efficiency losses and operational inconsistency to fester. Given the current advancements in data and analytics technology, we are in luck and have options to solve this issue. With data capture and storage so readily available, it is easy to gather information on your loops to provide better insight into how they are performing and where you could get improvement.
At Avanceon, we are using machine learning to work with our customers to evaluate their process-critical loops performance over time. This evaluation is yielding great insight to improve operations and know what is happening with the system when no one is watching. So, what do you see, with how you manage your process control loops? Do you see variance? Do you have competing theories on the ‘best’ way to operate the process within your organization? Drop us a line and let us know.