A Novel Adaptive SFA-LII Based Fault Detection Method for Nonstationary Processes.

Area of Science:

• Industrial Engineering
• Process Control
• Statistical Process Monitoring

Background:

• Traditional Multivariate Statistical Process Monitoring (MSPM) methods assume stationarity, limiting their effectiveness in real-world industrial processes experiencing aging, load changes, and disturbances.
• Nonstationary characteristics in industrial processes hinder accurate fault detection using conventional MSPM techniques.

Purpose of the Study:

• To propose a novel adaptive fault detection method for industrial processes exhibiting both stationary and nonstationary variations.
• To improve the accuracy and efficiency of fault identification in dynamic industrial environments.

Main Methods:

• Nonstationary variables identified using the unit root test.
• Stationary residuals derived via Johansen cointegration analysis, combined with original stationary variables.
• Feature-level fusion using a Slow Feature Analysis (SFA) monitoring model.
• Local Information Increment (LII) average as the monitoring statistic with dynamic control limits based on fuzzy membership functions.

Main Results:

• The proposed method demonstrates superior fault detection performance for nonstationary processes.
• Achieved a higher fault detection rate compared to traditional methods.
• Exhibited lower computational complexity.

Conclusions:

• The novel adaptive fault detection method effectively addresses nonstationarity in industrial processes.
• Validated through application to the Tennessee Eastman process and electric servo mechanisms.
• Offers a robust solution for enhanced industrial process monitoring and fault diagnosis.