Consistency-based fault isolation for hybrid dynamic models

Rapid detection and isolation of faults in modern systems improves safety, performance and reduces operational costs drastically. To date, expert-systems have been the method of choice, but the field of model-based diagnosis (MBD) has emerged which offers many benefits. A model of the correct structure and behavior of the device allows the diagnostic program to detect unexpected failures, or faults which have never before been seen. The heart of model-based diagnosis is the model, which has traditionally taken the form of a qualitative logic program. These languages are not familiar to most developers who are used to quantitative dynamic, or hybrid dynamic models. This thesis presents a novel method of consistency-based diagnosis (one form of MBD), which allows the use of any model type for which an appropriate calculation engine can be written. This new technique computes the same result as any other method built upon consistency-based diagnosis given perfect knowledge, but degrades gracefully in the presence of uncertainty, which can be caused by noise or modelling errors. As a proof of concept, a quantitative hybrid dynamic model was created for the fluid power circuit of a single-degree of freedom hydraulic test bench and a calculation engine based upon hcc (hybrid concurrent constraint programming) was developed. The model and engine were used to diagnose various failures on the physical test bench with approximately 85% accuracy. The ability to use any model type, and specifically, quantitative hybrid dynamic models, represents an important contribution to the field of model-base diagnosis.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat