A Deep Quality Monitoring Network for Quality-Related Incipient Faults

Area of Science:

• Chemical Engineering
• Process Control
• Data Science

Background:

• Quality-related process monitoring has advanced, but detecting incipient faults remains challenging.
• Existing methods like Partial Least Squares (PLS) primarily focus on larger fault magnitudes, neglecting early deviations.

Purpose of the Study:

• To develop a novel Deep Quality Monitoring Network (DQMNet) for effective detection of quality-related incipient faults.
• To address the limitations of current methods in identifying subtle, early-stage process anomalies.

Main Methods:

• DQMNet architecture: feature input layer, feature extraction layers, and output layer.
• Feature extraction using base detectors, singular values (SVs) of sliding-window patches, and Principal Component Analysis (PCA).
• Bayesian inference for constructing statistics from quality-related/unrelated feature matrices.

Main Results:

• Demonstrated superiority of DQMNet through numerical simulations.
• Validated DQMNet's effectiveness using benchmark data from the Tennessee Eastman Process (TEP).
• Successfully detected incipient faults that are typically missed by conventional methods.

Conclusions:

• DQMNet offers a robust solution for quality-related incipient fault detection.
• The proposed network enhances process monitoring by identifying subtle deviations.
• DQMNet shows significant potential for improving industrial process safety and efficiency.