Simultaneous Fault Detection and Identification in Continuous Processes via nonlinear Support Vector Machine based...

Area of Science:

• Chemical Engineering
• Machine Learning
• Process Control

Background:

• Real-time data from advanced sensors enables data-driven process monitoring.
• Machine learning, including Support Vector Machines (SVM), is increasingly used for analyzing complex process data.
• Accurate fault detection and identification are vital for safe and profitable industrial operations.

Purpose of the Study:

• To apply a novel nonlinear SVM-based feature selection algorithm for process monitoring and fault detection.
• To enhance the accuracy of fault detection models and facilitate fault diagnosis in continuous industrial processes.
• To evaluate the proposed methodology using the Tennessee Eastman process as a case study.

Main Methods:

• Development of a nonlinear, kernel-dependent SVM feature selection algorithm based on sensitivity analysis.
• Utilization of greedy algorithms for feature ranking and guiding fault diagnosis.
• Training fault-specific two-class SVM models for detecting faulty operations.

Main Results:

• The proposed SVM feature selection algorithm improved the accuracy of fault detection models.
• The methodology effectively guided fault diagnosis, identifying specific process issues.
• Performance was validated against existing approaches on the Tennessee Eastman process.

Conclusions:

• The novel SVM-based feature selection approach offers a robust method for process monitoring and fault management.
• This technique enhances diagnostic capabilities, contributing to safer and more efficient industrial operations.
• The findings demonstrate the value of advanced machine learning for real-time industrial process analysis.