Gaussian Derivative Change-point Detection for Early Warnings of Industrial System Failures

Abstract

An early warning of future system failure is essential for conducting predictive maintenance and enhancing system availability. This paper introduces a three-step framework for assessing system health to predict imminent system breakdowns. First, the Gaussian Derivative Change-Point Detection (GDCPD) algorithm is proposed for detecting changes in the high-dimensional feature space. GDCPD conducts a multivariate Change-Point Detection (CPD) by implementing Gaussian derivative processes for identifying change locations on critical system features, as these changes eventually will lead to system failure. To assess the significance of these changes, Weighted Mahalanobis Distance (WMD) is applied in both offline and online analyses. In the offline setting, WMD helps establish a threshold that determines significant system variations, while in the online setting, it facilitates real-time monitoring, issuing alarms for potential future system breakdowns. Utilizing the insights gained from the GDCPD and monitoring scheme, Long Short-Term Memory (LSTM) network is then employed to estimate the Remaining Useful Life (RUL) of the system. The experimental study of a real-world system demonstrates the effectiveness of the proposed methodology in accurately forecasting system failures well before they occur. By integrating CPD with real-time monitoring and RUL prediction, this methodology significantly advances system health monitoring and early warning capabilities.

Figures (10)

• Figure 1: The flowchart of the proposed early warning system for monitoring and predicting system failures. The system health and RUL are analyzed through two phases: the offline diagnostic phase and the online prognostic phase. In the offline phase, training data is processed via GDCPD to identify important features and change points on these features. They are used to set the threshold for monitoring statistics and to build LSTM predictive models. For online prognostics, a sliding window mechanism is used to detect early alarms, along with the trained LSTM models for RUL prediction. This structured approach ensures dynamic, real-time system health monitoring and failure prediction.
• Figure 2: Sensor Signals of Paper Manufacturing Machine. The top figure illustrates the signal time series from one of 61 sensors, with red dots marking the times of system breakdowns. The figure below zooms in the first five breakdowns. Note that after each breakdown the system will be repaired and restarted. The repair periods are not included in the figure.
• Figure 3: LSTM Architecture.
• Figure 4: ARD Length-scales of Features.
• Figure 5: Selected Features with Small Length-Scale. These selected features were identified by the ARD mechanism of multi-input GP classification model. The red dots indicate the times of system breakdowns, while the feature values over a one-hour period preceding system breakdown are plotted by blue line segments. Notably, certain features show abrupt changes or significant fluctuations prior to breakdowns, which suggests a potential predictive relationship between feature changes and system failures.