A Novel Quadratic Entropy Classifier for Fault Detection and Isolation with Application to the LIU-ICE Benchmark

Model-based fault diagnosis is the most powerful supervision approach when a model of the dynamic system under study is available or identifiable. When the underlying system presents complex nonlinearities or time varying components so that it is not feasible to obtain an accurate mathematical description of the system, knowledge-based fault diagnosis approaches are a viable alternative. However, these techniques require classification algorithms to understand if residuals are associated with a fault condition or not. This work presents a novel classification approach based on the concept of quadratic entropy. The proposed methodology is validated on the LiU-ICE benchmark, which requires to detect and isolate faults on a spark ignite engine used in the automotive industry. Results are compared with the current best solution of the benchmark which employs open-set classification techniques.