Identification based fault detection: Residual selection and optimal filter

Abstract In this work, an identification based fault detection method is proposed. The idea is to identify a dynamic process model from test data and to generate residuals using the identified model for fault detection. The method intends to improve fault detection performance while taking disturbance and model error into account. To this end, a fault detection performance index is introduced in a statistical framework. Then it is shown that the output error residual is more suitable for fault detection than the prediction error residual. Further an optimal detection filter maximizing the performance index is developed. Practical issues for implementing the detection filter are also addressed. Finally, the proposed method is illustrated through a numerical example and Tennessee Eastman process.