Time Series Extended Finite-State Machine-Based Relevance Vector Machine Multi-Fault Prediction

Fault prediction means to detect faults that can occur in the future. While most studies focus on predicting one fault at a time, multi-fault prediction is more practical for industrial processes as multiple faults can cause much more damage than a single one. A time series extended finite-state machine (TS-EFSM)-based relevance vector machine (RVM) approach is proposed for multi-fault prediction. Time lags and correlation coefficients between the process variables and process states are determined. Then, a variable and a state dependence diagram based on the correlation coefficients is established with the EFSM. Furthermore, the RVM is applied to identify parameters for the sake of better prediction accuracy and shorter testing times. With the prediction parameters, faults can be predicted using the aforementioned TS-EFSM state transitions.