Fault Identification of Chemical Processes Based on k-NN Variable Contribution and CNN Data Reconstruction Methods

Area of Science:

• Chemical Engineering
• Process Control
• Data Science

Background:

• Traditional fault detection methods struggle with complex chemical process data (nonlinearity, non-Gaussian, multi-operating modes).
• Existing k-Nearest Neighbor (k-NN) methods primarily focus on fault detection, with limited application in fault identification.

Purpose of the Study:

• To develop an accurate fault identification method for large-scale chemical processes.
• To address the limitations of traditional methods in handling complex process data characteristics.

Main Methods:

• A novel fault identification approach combining k-NN variable contribution analysis and Convolutional Neural Network (CNN) data reconstruction.
• Creation of a fault-symptom table to map faults to abnormal variables.
• Utilizing k-NN for variable contribution index calculation and CNN for reconstructing faulty variables.

Main Results:

• The proposed k-NN variable contribution method effectively identifies abnormal variables, validated by contribution decomposition theory.
• The CNN data reconstruction technique successfully identifies all faulty variables within each sample.
• The method demonstrated reliability and validity on a numerical example and the Continuous Stirred Tank Reactor system.

Conclusions:

• The integrated k-NN and CNN approach offers a robust solution for fault identification in chemical processes.
• This method enhances the ability to detect and pinpoint specific faults, crucial for process safety and operational efficiency.