Fault detection and monitoring using a data-driven information-based strategy: Method, theory, and application

2024-05-06

Mechanical systems and signal processing

AI Summary

To address the challenges of early fault detection, scarcity of fault samples, and stringent requirements for model deployability and decision interpretability in nonlinear, non-Gaussian industrial systems, this paper proposes a model-free, adaptive, data-driven fault monitoring method. The approach uniquely integrates entropy rate estimation with dynamic information bottleneck theory to establish an online monitoring framework grounded in uncertainty quantification and anomalous information flow tracking, enabling real-time discovery and interpretable attribution of previously unseen fault patterns. It synergistically combines sliding-window-based dynamic information analysis, nonparametric statistical testing, and lightweight online learning. Evaluated on multiple industrial benchmark datasets, the method achieves an average fault detection rate of 98.7%, an early-warning accuracy of 92.4%, and reduces mean detection delay by over 40%.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Problem

• Detects incipient faults in systems
• Addresses non-linear, non-Gaussian system challenges
• Proposes distribution-free mutual information estimator

Innovation

• Uses information-driven fault detection method
• Employs novel concept drift detector
• Applies distribution-free mutual information estimator