A latent feature oriented dictionary learning method for closed-loop process monitoring.

Area of Science:

• Industrial automation
• Process monitoring
• Cyber-physical systems

Background:

• Industrial cyber-physical systems (ICPS) are crucial for modern industry, featuring advanced computing, communication, and control.
• Closed-loop processes in ICPS exhibit mixed stationary/nonstationary and dynamic features due to equipment degradation and changing conditions.
• Traditional open-loop monitoring methods in ICPS lead to false alarms, compromising safety and trustworthiness.

Purpose of the Study:

• To propose a latent feature oriented dictionary learning (LFDL) method for precise feature separation in ICPS closed-loop process monitoring.
• To enhance the accuracy and reliability of fault detection in dynamic industrial environments.
• To address the limitations of traditional methods in handling complex process variations.

Main Methods:

• The proposed LFDL method involves three stages for latent feature separation.
• Stage 1: Spatial separation of closed-loop process variables into stationary and nonstationary components.
• Stage 2: Temporal extraction of static and dynamic features using slow feature analysis and cointegration analysis.
• Stage 3: Global monitoring by fusing local results via dictionary learning.

Main Results:

• The LFDL method effectively separates latent features from raw data in closed-loop processes.
• The approach accurately distinguishes between normal operational condition changes and actual faults.
• Experimental validation on three-phase flow, Tennessee Eastman process, and industrial roasting processes confirms the method's effectiveness.

Conclusions:

• The developed LFDL method provides accurate and reliable monitoring for closed-loop processes in ICPS.
• Simultaneous consideration of temporal and spatial features enhances fault detection capabilities.
• The method offers a robust solution for improving the safety and trustworthiness of modern industrial systems.