Fault Diagnosis by Multisensor Data: A Data-Driven Approach Based on Spectral Clustering and Pairwise Constraints

Area of Science:

• Signal Processing
• Machine Learning
• Manufacturing Systems

Background:

• Multisensor data in high-dimensional spaces presents challenges for traditional clustering.
• Spectral clustering, based on affinity matrices and Laplacian graphs, is effective for grouping data.
• Fault diagnosis in manufacturing relies heavily on clustering techniques.

Purpose of the Study:

• To present an enhanced spectral clustering approach for multisensor data.
• To improve the efficiency and accuracy of fault diagnosis in high-dimensional spaces.
• To incorporate pairwise constraints into spectral clustering for better performance.

Main Methods:

• Utilizing spectral clustering algorithms for data grouping.
• Constructing an affinity matrix to represent pairwise data point similarity.
• Applying spectral decomposition of the Laplacian graph.
• Augmenting the spectral clustering approach with pairwise constraints.

Main Results:

• The enhanced spectral clustering approach demonstrates efficient identification of fault scenarios.
• The method is validated through a case study on a diesel injection control system.
• Improved fault detection capabilities were observed.

Conclusions:

• The proposed enhanced spectral clustering method is effective for fault diagnosis using multisensor data.
• Pairwise constraints significantly improve the performance of spectral clustering in this context.
• The approach offers a valuable tool for fault detection in manufacturing applications.