Advantages of RPC and limits of model-based control
Part 3: Rate-predictive control, an alternative algorithm to proportional-integral-derivative (PID) and model-based control, can provide single-loop control. See three RPC advantages and two model-based control limitations.
Rate-predictive control (RPC) is a patented control algorithm designed as an alternative to industry standard proportional-integral-derivative (PID) for single-loop control. It is also used as the internal control algorithm for model-less multivariable control (XMC).
Three rate-predictive control advantages
RPC has three specific advantages over PID and model-based control:
1. RPC (like PID, but unlike model-based control) is a feedback control algorithm. Feedback remains the process industry’s first choice for almost all control loops, because of its ease of deployment, low maintenance and high rate of success and reliability. The timing of feedback control is always perfect because it responds as the process responds, meaning there are no model-based timing issues. In the vast majority of applications, feedback control rejects process disturbances without unacceptable levels of deviation.
2. RPC is predictive in a different way than model-based control. RPC looks at the ongoing rate-of-change of the controlled variable and predicts its implicit future value, which is the current value plus the rate-of-change times the process settling time. The settling time is RPC’s main tuning parameter, which is easy to tune and has a forgiving margin for error, like PID integral time. The predictive method of RPC makes it more responsive to disturbances and more stable as control returns to setpoint.
3. RPC is adaptive to changes in process response. Among hundreds of U.S. patents for process control, RPC is the only one with the claim of being inherently adaptive (think naturally self-tuning). For example, if process gain changes, then the process response changes accordingly, and so does RPC’s prediction and controller response. It’s simple and elegant and also profound for an industry where retuning and model-maintenance have always been as much the rule as the exception. Successful adaptive control has long been the grail of process control.
Two limits of model-based control
Model-based control, which is synonymous with feedforward, is often considered superior to feedback because it has the potential to reject disturbances with minimal deviation. Even so, the widespread adoption of model-based control over the last few decades, primarily in the form of model-based multivariable control, has revealed limitations of model-based control in practice.
1. Reliable model-based control depends on reliable models. Where process responses vary, so they no longer match the models built into the controller, then model-based control performance degrades and may compound disturbances. In other words, the promise of model-based control to improve performance also carries the risk of poorer and less reliable performance. Decades of experience have shown process models are less reliable and shorter-lived than originally expected. This means nearly continuous model-maintenance is necessary to mitigate this risk. This is the main reason auto-tuning has fallen short of industry expectations and why even continuous adaptive modeling cannot overcome this limitation of model-based control.
2. Return to feedback control. Due to the first reason, model-based control technology has pursued numerous adaptations, and a proliferation of esoteric configuration and tuning parameters, to help improve stable and reliable performance in the face of model mismatch. But to the extent that model-based control can tolerate model error, it reverts to feedback control! This raises the question, why spend so much time and money on models and model maintenance, only to fall back on feedback control? Wouldn’t it make more sense to begin with feedback control and then apply feedforward selectively, only where it is actually necessary and warranted? The answer to this question has always been yes for single-loop control; in retrospect it should be yes for multivariable control, too. Most processes require, at most, a handful of important models for effective multivariable control and optimization, not dozens or hundreds.
In this light, a further inherent advantage of RPC and XMC is they are not model-based; so model-related activities such as plant testing, model-identification, and model maintenance, are largely eliminated. RPC and XMC incorporate feedforward selectively based on traditional proven advanced regulatory control (ARC) methodologies.
Allan Kern, P.E., is owner and president of APC Performance LLC. Edited by Mark T. Hoske, content manager, Control Engineering, CFE Media, firstname.lastname@example.org.
KEYWORDS: Advanced process control, rate-predictive control
Rate-predictive control (RPC) is a patented control algorithm that is an alternative to proportional-integral-derivative (PID) single loop control.
RPC advantages include that it is a feedback control algorithm and predictive.
Model-based control requires reliable models.
Can you close more control loops with a more reliable control algorithm?
See links for part 1 on rate-predictive control and part 2 on multivariable control benefits below.