Fault detection of feed water treatment process using PCA-WD with parameter optimization

Area of Science:

• Boiler systems engineering
• Process control and automation
• Data analytics and signal processing

Background:

• Feed water treatment processes (FWTPs) are critical for utility boiler reliability.
• Classical Principal Component Analysis (PCA) for FWTP fault detection suffers from noise, leading to false and missed detections.
• Existing methods require manual parameter tuning, limiting efficiency.

Purpose of the Study:

• To develop an improved fault detection algorithm for FWTPs by addressing noise issues in classical PCA.
• To enhance the reliability and accuracy of fault detection in utility boiler systems.
• To automate the parameter selection for the proposed fault detection algorithm.

Main Methods:

• A novel PCA-WD algorithm combining Wavelet Denoise (WD) with Principal Component Analysis (PCA) was developed.
• Particle Swarm Optimization (PSO) was employed to optimize the parameters of the PCA-WD algorithm.
• The optimized PCA-WD was validated using operational data from a coal-fired power plant's FWTP.

Main Results:

• The optimized PCA-WD effectively suppressed noise in T² and SPE statistics, significantly improving detection performance.
• Parameter optimization allowed for automatic selection of optimal PCA-WD parameters, reducing reliance on expert experience.
• Comparative analysis demonstrated superior performance of the optimized PCA-WD over classical PCA and Sliding Window PCA (SWPCA) in detecting various fault types.

Conclusions:

• The optimized PCA-WD algorithm offers a robust and automated solution for enhancing fault detection in FWTPs.
• This approach improves the reliability of utility boilers by minimizing false and missed fault detections.
• The study validates the effectiveness of integrating wavelet denoising and optimized PCA for industrial process monitoring.