Modified canonical variate analysis based on dynamic kernel decomposition for dynamic nonlinear process quality...

Area of Science:

• Process Systems Engineering
• Industrial Monitoring and Control
• Data Analytics and Machine Learning

Background:

• Efficient process monitoring is vital for ensuring operational safety and enhancing product quality in industrial settings.
• Traditional methods often struggle with the complexities of dynamic nonlinear processes, necessitating advanced analytical techniques.
• Existing canonical variate analysis (CVA) and kernel-based methods have limitations in capturing intricate nonlinear relationships.

Purpose of the Study:

• To propose a novel dynamic kernel decomposition based canonical variate analysis (DKDCVA) approach for dynamic nonlinear process quality monitoring.
• To establish a partial-correlation nonlinear model that maximizes extracted feature information between input and output variables.
• To develop a robust monitoring technique capable of handling non-Gaussian process variables and ensuring process safety.

Main Methods:

• Developed a modified canonical variate analysis incorporating dynamic kernel decomposition (DKDCVA).
• Employed singular value decomposition for orthogonal decomposition of the dynamic nonlinear model into quality-related and independent subspaces.
• Utilized Hankel matrices and kernel density estimation to construct statistical metrics and determine control limits for non-Gaussian data.

Main Results:

• The DKDCVA approach successfully establishes a partial-correlation nonlinear model, maximizing feature extraction.
• Orthogonal decomposition effectively separates quality-related information, enabling targeted monitoring.
• Kernel density estimation provides accurate control limits, outperforming traditional methods for non-Gaussian variables.

Conclusions:

• The proposed DKDCVA method is a superior technique for monitoring abnormal operations in dynamic nonlinear processes.
• Experimental validation on numerical, Tennessee Eastman, and hot strip mill processes confirms the effectiveness and robustness of DKDCVA.
• DKDCVA enhances process safety and product quality by providing reliable abnormal operation detection.