Modeling and State Estimator Design Issues for Model-Based Monitoring Systems

Abstract

The JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL has a rich history in the areas of system modeling, control theory, and system design. In this paper, I demonstrate the primary importance of modeling to the design and performance of model-based (dynamic, observer-based) monitoring systems. The focus is on monitoring rather than diagnostics, but the results show, nevertheless, that good monitoring systems impact the design and performance of diagnostic systems. Modeling procedures, some based on results from the automated modeling literature, are developed to produce not only simple, accurate models to improve estimation accuracy, but also to utilize existing sensors to measure accessible signals, generate models for which estimation techniques exist, and finally to produce models that reduce the complexity of the diagnostic task, thus reducing the heavy reliance on empirical rule-based decision systems. In addition, for model-based monitors based on deterministic state estimators, a design procedure is described to control the shape of the transient error and the steady-state bias error of the estimates. This procedure utilizes a performance index, which accounts for many of the realistic problems encountered in implementing these systems, such as model uncertainty, round-off errors, and unknown initial conditions. Machine tool drive systems are used to motivate and illustrate the monitoring ideas and techniques. Model-based monitoring appears, despite its current limited use in machine monitoring, to hold great potential as the basis for high-performance, low-cost machine monitoring systems.