A robust model selection framework for fault detection and system health monitoring with limited failure examples: Heterogeneous data fusion and formal sensitivity bounds

Roberto Rocchetta, Qi Gao, Dimitrios Mavroeidis, Milan Petkovic

Abstract: Highlights•A robust model selection of failure detection and system health monitoring models.•Fusion of time-to-failure survival data, system structure model, and sensors monitoring data.•Robust SVM fault classifiers thanks to new sensitivity/specificity bounds.•The number of support vectors defines the complexity of the classifier and error bounds.•Lack of failure data uncertainty affecting the fault detector is formally quantified.