Detecting continual anomalies in monitoring data stream based on sampling GPR algorithm

Prognostics and health management (PHM) can improve the system availability and efficiency to realize Condition Based Maintenance (CBM). As the primary and critical process of PHM, the anomaly detection can discover the abnormal condition and potential fault in time. Generally, monitoring data can reflect the operating condition of components, subsystems or systems. Thus, the monitoring data can be applied to detect the anomalies to improve the system readness and help the system health management. However, monitoring data arriving in the form of streaming gradually shows the growth in variability, velocity and volume. As a result, detecting the anomalies in data stream brings new challenges to traditional anomaly detection algorithm. In this case, this paper proposes a sampling Gaussian process regression (GPR) method for continual anomaly detection based on the specialty of streaming data, providing important information for estimating the system conditon. The effectiveness of this method is evaluated by the synthetic datasets and public data sets.