Online Monitoring and Fault Diagnosis for High-Dimensional Stream with Application in Electron Probe X-Ray...

Area of Science:

• Data Science
• Statistical Monitoring
• Signal Processing

Background:

• High-dimensional data streams present challenges for anomaly detection due to sparsity.
• Existing methods struggle with timely and accurate identification of subtle changes.

Purpose of the Study:

• To develop an innovative two-stage framework for monitoring and diagnosing high-dimensional data streams with sparse changes.
• To enhance the accuracy and speed of anomaly detection and fault diagnosis in complex data.

Main Methods:

• Stage 1: Online monitoring using exponentially weighted moving average (EWMA) statistics, extreme value theory, and multiple hypothesis testing for change point detection.
• Stage 2: Fault diagnosis mechanism for pinpointing abnormal components upon anomaly detection.
• Validation through extensive numerical simulations and electron probe X-ray microanalysis (EPXMA) applications.

Main Results:

• Rapid anomaly detection, often within 1-2 sampling intervals post-change.
• Near 100% detection power with type-I error rates around 5%.
• Fault diagnosis accuracy of 99.1% in 200-dimensional streams, outperforming PCA-based methods by 28.0% in precision.

Conclusions:

• The proposed framework significantly advances the management of high-dimensional sparse-change data streams.
• Demonstrates superior performance in both anomaly detection and precise fault localization compared to existing techniques.
• Effective for real-world applications requiring robust data stream monitoring and analysis.