A Fault Detection Method Based on CPSO-Improved KICA.

Area of Science:

• Signal Processing
• Machine Learning
• Industrial Process Monitoring

Background:

• Traditional Kernel Independent Component Analysis (KICA) suffers from randomness in kernel parameter selection, impacting performance.
• Existing methods like Weighted KICA (WKICA) may not sufficiently address parameter optimization challenges.
• The Tennessee Eastman Process (TEP) is a standard benchmark for evaluating process monitoring and fault detection algorithms.

Purpose of the Study:

• To propose a novel CPSO-KICA algorithm for optimizing kernel parameters in KICA.
• To enhance the accuracy and reliability of fault detection in industrial processes.
• To overcome the limitations of local optima in traditional Particle Swarm Optimization (PSO).

Main Methods:

• Developed the Chaotic Particle Swarm Optimization-Kernel Independent Component Analysis (CPSO-KICA) algorithm.
• Utilized maximum entropy of extracted independent components as the fitness function for PSO parameter optimization.
• Incorporated a chaotic algorithm (CO) to prevent PSO from converging to local optima.
• Benchmarked CPSO-KICA against Weighted KICA (WKICA) and PSO-KICA using the Tennessee Eastman Process (TEP).

Main Results:

• The CPSO-KICA algorithm successfully determined optimal kernel parameters for KICA.
• Demonstrated superior performance compared to WKICA and PSO-KICA on the TEP benchmark.
• Achieved significant improvements in fault detection rates (FDR) and reductions in false alarm rates (FAR) and detection latency (DL).

Conclusions:

• The proposed CPSO-KICA algorithm effectively addresses the kernel parameter selection randomness in KICA.
• CPSO-KICA offers enhanced accuracy and efficiency for fault detection in industrial monitoring systems.
• This approach provides a robust solution for identifying process anomalies with improved reliability.