Fault detection for chemical processes based on non-stationarity sensitive cointegration analysis

Area of Science:

• Chemical Engineering
• Process Monitoring
• Data Analysis

Background:

• Chemical processes exhibit time-varying conditions, leading to non-stationary characteristics.
• Conventional monitoring methods struggle to capture these non-stationary variations, posing challenges for abnormality detection.
• Abnormalities in non-stationary processes can manifest as shifts in stationary variables.

Purpose of the Study:

• To propose a novel monitoring method sensitive to non-stationary variations in chemical processes.
• To enhance the detection of abnormalities in dynamic and time-varying industrial settings.
• To develop a robust process monitoring framework for complex chemical systems.

Main Methods:

• Utilized Augmented Dickey-Fuller test to identify essential non-stationary variables.
• Selected non-stationarity sensitive variables from faulty data for enhanced abnormality detection.
• Established cointegration analysis models for both variable types to analyze dynamic equilibrium.
• Employed Bayesian inference for combining monitoring results.

Main Results:

• The proposed method successfully identified non-stationary variations in process data.
• Demonstrated improved sensitivity to abnormalities compared to conventional approaches.
• Validated effectiveness on the Tennessee Eastman process and a vinyl acetate monomer plant model.

Conclusions:

• The non-stationarity sensitive cointegration analysis is effective for monitoring chemical processes with time-varying conditions.
• The method enhances the capability to detect abnormalities by focusing on sensitive variables.
• The approach offers a feasible and high-performance solution for complex industrial process monitoring.