Applying Model Predictive Control in Non-Refinery Settings

MPC Was Built for Refineries — Applying It Elsewhere Is a Different Problem

Model Predictive Control (MPC) has decades of track record in refineries, where processes tend to run continuously, near steady state, with mature instrumentation and dedicated engineering teams to sustain the models. Outside the refinery — in mining, ethanol production, upgraders, petrochemicals, and general manufacturing — those conditions often don't hold, and applying the same MPC playbook without adjusting for that tends to produce disappointing results.

What Makes Non-Refinery Settings Different

A few recurring challenges show up across these industries: more frequent grade or product changes and batch transitions, which mean the process rarely sits at the steady operating point classical MPC models assume; higher inherent process variability — in mining, for example, feedstock composition itself is a moving target (see our Mining Advanced Control case study); less mature historian and instrumentation infrastructure to build reliable models from, particularly in facilities that haven't invested in advanced control before; smaller technical teams available to sustain and retune the models over time; and different economic drivers — a mining operation optimizing for throughput and fuel use, for example, cares about a different objective function than a refinery optimizing for yield against a fixed feedstock.

How This Gets Addressed in Practice

The projects summarized in our Mining Advanced Control, Upgrader Advanced Control, and Ethanol Plant Monitoring case studies each work through a version of this problem — building models around AI-informed advisory modes that account for feedstock variability, layering soft sensors in where instrumentation is thinner, and designing systems operations teams can actually sustain rather than a model that only the original implementation engineer understands. That's the throughline of Capstone's approach to non-refinery MPC: adapt the methodology to the process, not the other way around.

This page summarizes an academic seminar presentation on the specific challenges and approaches to applying MPC outside refinery settings; the full presentation is available in the downloadable resource below.

Who This Is For

Relevant for process engineers and operations leaders in mining, ethanol, upgrading, or general manufacturing evaluating whether MPC can work in a process that doesn't look like a textbook refinery unit — which, in practice, describes most non-refinery processes.