Fault Diagnosis in a Heat Exchanger using Process History Based-Methods

摘要

A comparison of fault diagnosis systems based on Dynamic Principal ComponentrnAnalysis (DPCA) method and Artificial Neural Networks (ANN) under the samernexperimental data is presented. Both approaches are process history based methodsrnwhich do not assume any form of model structure, and rely only on process historicalrndata. The comparative analysis shows the online performance of both approaches whenrnsensors and/or actuators fail. Robustness, quick detection, isolability capacity, falsernalarm rates and multiple faults identifiability are considered for this experimentalrncomparison. An industrial heat exchanger was the experimental system. ANN showedrninstantaneous detection for actuator faults; however, with greater (22%) false alarm rate.rnANN can isolate multiple faults; whereas, DPCA did not show this property, butrnrequired a minor training effort.