Model-plant mismatch detection for a plant under Model Predictive Control: A grinding mill circuit case study

Abstract

This articles investigates two different techniques of identifying model-plant mismatch for a grinding mill circuit under model predictive control. A previous attempt at model-plant mismatch detection for a grinding mill, in the form of a partial cross correlation analysis, is used as a benchmark for model-plant mismatch detection and degraded sub-model isolation. This is followed by an investigation of the plant model ratio technique applied to the same system. The plant model ratio technique is able to isolate the sub-model containing a mismatch as well as detect the specific parameter in a first-order-plus-time-delay model responsible for the mismatch. A simulation study is used to quantify and compare the results between the two model-plant mismatch detection methodologies. The results indicate plant model ratio accurately and timeously detects mismatches in sub-models. This allows for system reidentification or controller adaption to ensure optimal process performance. The advantage above partial cross correlation is the parameter diagnosis within the degraded sub-model coupled with the mismatch direction.