An integrated framework for statistical change detection in running status of industrial machinery under transient conditions.

Early detection of changes in machine running status from sensor signals attracts increasing attention for the monitoring and assessment of complex industrial machineries under transient conditions. This paper presents a detection method that integrates one-class SVM with a pre-defined Autoregressive Integrated Moving Average (ARIMA) regression process. Meanwhile, an automatic cyclic-analysis method is also developed as a preprocessing to suppress temporal non-stationarity in condition signal before feeding it to the monitoring process. As such, a novel framework of continuous monitoring of condition signal is finally presented to inspect whether an unexpected running status change occurs or not during continuous machine operations. The proposed framework is applied to three representative condition monitoring applications: external loading condition monitoring, bearing health condition assessment, and rotational speed condition monitoring. Comparisons with existing methods are also provided, where the proposed method demonstrates its significant improvements over others.