Tolerance generation for the diagnostic model processor algorithm (DMP)

摘要

The problem of fault diagnosis is concerned with detecting abnormal behaviour of process and attributing the behaviour to one or more root causes. This problem is known as fault detection and isolation. In this paper, a methodology known as Diagnostic Model Processor (DMP) is proposed. This approach uses a vector of model equations. Associated with each model equation is a tolerance, which mark the threshold of "normal" operation. The methodology to select this tolerance is not provided for DMP, so it must be determined using trial and error. The paper suggests envelope generation to derive this tolerance. The problem of envelope generation is a classical problem. But none of the existent approaches can assure that produce the exact system envelopes without increasing the overall uncertainty. A new approach based on optimization for envelope generation is presented. This new algorithm works recursively with the sliding window principle being suitable to operate in real time. So, a new fault detection and isolation approach based on the integration of the fault detection algorithm known as Envelope Generation with a classical model-based fault diagnosis algorithm known as DMP ("Diagnostic Model Processor") is proposed in this paper.