Noise – that random variation in your data signals – is seemingly everywhere. Low frequency process and high frequency signal noise are two sources that directly affect control. There’s no avoiding it. Indeed, when tuning PID controllers practitioners simply need to deal with it. As they say: It is what it is!
So exactly how does noise affect control? It’s generally understood that tuning PID controllers starts with modeling the process. By performing a step, bump or other test, the fundamental dynamic behavior of a process can be revealed. Once a change to the associated process is initiated the controller’s response is soon to follow. A trend of the two showcases their cause-and-effect relationship and provides the essentials for process modeling. That is, of course, as long as noise doesn’t obscure the relationship and impede model estimation.
Following are a few thoughts on noise, its effects on control, and how it can be addressed while tuning:
As mentioned, noise can obscure a process’ response to change and make it difficult to calculate an accurate model. That difficulty often applies equally to manual methods as it does when software tools are applied. More precisely, when tuning manually noise can make it challenging for practitioners to “eyeball” their trended step test data and to accurately estimate values for a process’ Gain, Time Constant and Dead-Time. Similarly, when tuning with the help of software, noise that’s typical of real-world applications often thwarts the tool’s modeling capabilities, resulting in sub-par coefficients.
Good and Bad
From a tuning standpoint noise isn’t all bad as it’s an important part of the overall picture. If controlling a process’ dynamic behavior is the goal, then it’s essential to 1) account for all of the behavior that the controller encounters and 2) tune the controller to respond appropriately. A previous post explained the simple wire-in, wire-out nature of the PID controller. Its purpose is to respond to Error – the difference between the Process Variable and Set Point. In order for a PID controller to regulate loop performance effectively, the underlying model must account for the good and the bad even if that means accounting for noise.
Filtering is often applied in control applications to dampen the effects of noise on a given process signal. Filters – whether external or internal filters – effectively mute noise and smooth a controller’s response to dynamic behavior. Whereas internal filters manipulate data within the controller and leave the raw signal unchanged, external filters are external to the PID controller and manipulate the data going to the controller. These filtering options involve additional hardware and/or software cost. They may make it easier to estimate model parameters, but they also hide data that could be important to the bigger picture – control.
While very few traditional PID controller tuning packages account effectively for noise, those tools equipped with the ability to model non-steady state process data (i.e. NSS Modeling) have been shown to perform well with highly variable data. In addition to accurately modeling oscillatory process behavior they effectively account for the type of variability associated with a noisy data signal. Tools equipped with NSS Modeling eliminate the need for costly filtering solutions. Additionally, they simplify the modeling and tuning procedure will delivering more accurate and consistent results compared to manual methods.
While noise is pervasive in industrial applications it doesn’t have to limit a practitioner’s ability to tune PID controllers. If your processes are subject to excessive noise, then consider tools equipped with NSS Modeling. They’ve proven to deliver quality results even when confronted with the noisy, oscillatory data that practitioners view as the “real world” and that’s saying something.
Share this post:
Our most popular posts:
Expand your knowledge with this featured publication:
While most technology investments require proof before a purchase can be made, each production facility has unique attributes for which...Learn more