(202e) Automation of Knowledge Work in Advanced Process Control

Multivariable Predictive Control (MPC) has been an active area in both academic research and industrial applications for over 25 years. It still creates a tremendous value in the process industries by optimizing operations by increasing throughput, maximizing yields and reducing energy consumption. Although many advanced algorithms have been developed over the years in the area of model identification optimization and control, the methodology for the deployment of the MPC projects has not changed significantly and many of the MPC project steps are still manual and tedious, requiring significant effort and time. As the chemical processes and operation conditions continuously change, the process models need to be frequently re-identified and the controller tuning parameters need to be re-tuned in order to sustain the economic benefits of the MPC application.

Automation of knowledge work has been identified as one of the most impactful and disruptive initiatives with similar economic impact to the Industrial Internet of Things (IIOT). In this presentation we will discuss how automation of knowledge work applies to MPC in order to automate and simplify complex analysis and workflows and thus reduce barriers to deployment. More specifically we will look into automation of knowledge work for online Model Quality Analysis, Seed Model Generation, Automatic Data Slicing, Automatic Constrained Model Identification, Colinearity analysis and repair etc.

One particular area that we will focus on is automating the complex process of tuning the steady-state MPC optimizer via a Sequential Multi-Objective Optimization (SMO) that enables the user to simply specify the operational objectives of the process. An industrial example will be presented to demonstrate that via this approach the LP tuning remains consistent and does not need to be revisited when the controller models are updated.