Diagnostic Symptom Generation Aggregating Model Validity and Residual Evaluation

FDI commonly relies on an analytical model of the process to be diagnosed. This model is supposed to represent the process behavior in a particular mode for a given operating set point. Residuals are computed with this model and some measures acquired online from the process. They are compared to thresholds in order to provide the diagnostic decision. However, the model that has been used to compute a residual can be not suitable. For instance, this may be due to the identification procedure itself that has been performed with data not fully relevant. Another reason may be that the operating set point has changed, leading to a model that is not valid at this new set point. Thus, the diagnostic decision may produce a false alarm. Therefore, it is highly desirable to be able to measure the validity of the model used to provide the residuals. The aim of this paper is to propose an evaluation of this model validity indicator based upon fuzzy aggregation of several partial criteria. The model validity is then aggregated with the diagnostic decision in order to provide a symptom sensitive both to the model validity and to the residual threshold exceeding. Examples will illustrate the proposed method.