Reader feedback: Model-free multivariable control?

Oct. 21, 2015
A reader says little is yet known about MPC. Is that so?

In a recent online forum, a reader asks the experts about "model-less" multivariable control. The answers were very brief, probably because very little is yet known about this concept.

I've been exploring model-less, multivariable control for many years now. Or, to be more precise, I've been working to understand the low "glass ceiling" of model-based predictive (MPC) control performance, and in recent years that has led me to the realization that model-less multivariable control is both feasible and even potentially advantageous in terms of lower cost, greater ease of use and greater effectiveness.

A key insight along the way is that essentially all processes are multivariable, and multivariable control has always been a prominent part of industrial process operation. Before computers, it was carried out manually by operators (without models). This is still the case today in the vast majority of industrial processes.

A second important insight has been that the root causes of degraded MPC performance are structural, i.e. they're rooted in the nature of industrial processes and process operation, and are not problems with MPC theory. I think this is one reason MPC performance has been so difficult to come to grips with (because the theory is sound). Nonetheless, seeing that the limitations are structural and therefore not likely to be resolved easily or soon further compels the case to find a more practical alternative, which again casts the spotlight towards the possibility of a model-less solution.

The concept of model-less multivariable control, as I have developed it so far, relies only on the gain direction of process interactions (not detailed models). It uses preselected move rates that are based on operational experience and safety. And it uses a technique called "rate-based control" that completes move sequences in a manner that lands controlled variables on targets without overshoot or cycling (it also happens to be inherently adaptive to dynamic changes in process gain, which has been a particular problem for model-based control). To date, I have actually programmed this method and applied it to several process simulations successfully.

I hope this brief statement at least shows that the question of model-less multivariable control can't be easily answered or dismissed out of hand. It has large potential importance to the advanced control community and to the process industries. I'd be glad to answer any further questions that readers may pose.

Allan Kern, PE
[email protected]

Sponsored Recommendations

Measurement instrumentation for improving hydrogen storage and transport

Hydrogen provides a decarbonization opportunity. Learn more about maximizing the potential of hydrogen.

Get Hands-On Training in Emerson's Interactive Plant Environment

Enhance the training experience and increase retention by training hands-on in Emerson's Interactive Plant Environment. Build skills here so you have them where and when it matters...

Learn About: Micro Motion™ 4700 Config I/O Coriolis Transmitter

An Advanced Transmitter that Expands Connectivity

Learn about: Micro Motion G-Series Coriolis Flow and Density Meters

The Micro Motion G-Series is designed to help you access the benefits of Coriolis technology even when available space is limited.