Fault detection in dynamic plant-wide process by multi-block slow feature analysis and support vector data...

Area of Science:

• Process engineering
• Data analysis
• Machine learning

Background:

• Dynamic large-scale process fault detection is crucial for industrial safety and efficiency.
• Existing methods may struggle to capture complex interdependencies in process data.
• Highlighting local information and extracting different dynamics are key challenges.

Purpose of the Study:

• To develop a novel dynamic large-scale process fault detection algorithm.
• To leverage the strengths of multi-block algorithms and slow feature analysis.
• To enhance the accuracy and efficiency of fault detection in industrial processes.

Main Methods:

• Utilizing multi-block slow feature analysis for dynamic process fault detection.
• Calculating a mutual information-based relevance matrix to assess variable correlations.
• Employing K-means clustering to group variables into blocks based on relevance.
• Applying slow feature analysis within each block.
• Using support vector data description for final fault detection decisions.

Main Results:

• The proposed algorithm successfully detected faults in the Tennessee Eastman (TE) process.
• Demonstrated superior performance compared to existing fault detection methods.
• Validated the algorithm's efficiency and effectiveness in dynamic large-scale systems.

Conclusions:

• The multi-block slow feature analysis algorithm offers an effective approach for dynamic large-scale process fault detection.
• The combination of local information highlighting and dynamic feature extraction improves detection capabilities.
• The method shows significant promise for real-world industrial applications.