Key Performance Indicators Based Fault Detection and Isolation Using Data-Driven Approaches

This brief considers the key performance indicators (KPIs) based fault detection and isolation problems for unknown systems in the data-driven framework. To cope with the difficulty in obtaining real-time KPIs, a KPIs predictor is first constructed by utilizing the subspace-aided approach, which makes them available online based on the measurable input and output variables. Then the statistical detection technique is applied to detect the KPIs relevant faults by evaluating the prediction residuals. Towards the isolation purpose, a generalized likelihood ratio based reconstruction algorithm is proposed, through which the multi-dimensional faulty variables that are responsible for the KPIs anomalies are isolated. Finally, simulation studies on a ship propulsion system are given to demonstrate the effectiveness of the developed detection and isolation methods.