Performance monitoring of industrial controllers based on the predictability of controller behavior

Abstract

This paper focuses on performance assessment of industrial controllers. Instead of using process or controller models, it is based on process data collected at regular time intervals. Data analysis includes a set of tests that are reviewed in the paper and implemented in a software system. A methodology based on the concept of the predictability of controller errors is also proposed for performance monitoring. It considers the time series of the error and verifies the existence of predictable patterns beyond the control horizon in each one of the controlled variables of the process. The result of the analysis is given as a performance index. Examples using industrial data from a refinery are provided.