A novel framework of change-point detection for machine monitoring

The need for automatic machine monitoring has been well known in industries for many years. Although it has been widely accepted that a change in the structural property can indicate the fault in rotating machinery components (e.g., bearing and gears), automatic algorithms for this task are still in progress. In this paper, we propose a novel framework for change-point detection in machine monitoring. The framework includes two phases: (1) anomaly measure: on the basis of an automatic regression (AR) model, a new computation method is proposed to measure anomalies in a given time series which does not require any reference data from other measurement(s); (2) change detection: a new statistical test is employed by using martingale for detecting a potential change in the series which can be operated in an unsupervised and self-conducted manner. Experimental results on testing data captured in real scenarios demonstrated the effectiveness and the realizability of the proposed framework for change-point detection in machine monitoring, which suggests that our framework can be directly applicable in many real-world applications. •A new method of periodicity estimation is proposed based on Dynamic Time Warping.•A new computation method is proposed to measure anomalies in a given time series.•The potential change is detected by a new statistical test based on martingale.