Intelligent Fault Diagnosis for Chemical Processes Using Deep Learning Multimodel Fusion

Area of Science:

• Chemical Engineering
• Artificial Intelligence
• Process Control

Background:

• Deep learning methods are prevalent in chemical process fault diagnosis.
• Current single-network or sequential stacking approaches have limitations in accuracy and speed.
• Existing methods struggle to fully capture complex process dynamics.

Purpose of the Study:

• To propose a novel deep learning multimodel fusion method for chemical process fault diagnosis.
• To overcome the limitations of single deep learning networks in fault diagnosis.
• To enhance both the accuracy and speed of fault diagnosis in chemical processes.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) networks for temporal feature extraction.
• Employed Convolutional Neural Networks (CNNs) for spatial feature extraction.
• Fused features from LSTM and CNN, followed by Multilayer Perceptron (MLP) for compression and final diagnosis.

Main Results:

• The proposed multimodel fusion method demonstrated superior performance compared to existing deep learning techniques.
• Successfully integrated temporal (LSTM) and spatial (CNN) feature extraction capabilities.
• Achieved significant improvements in fault diagnosis accuracy and speed on TE chemical and coking furnace processes.

Conclusions:

• Deep learning multimodel fusion, integrating LSTM and CNN, offers enhanced fault diagnosis capabilities for chemical processes.
• The method effectively captures both temporal and spatial characteristics, outperforming simpler sequential stacking approaches.
• This approach represents a significant advancement for reliable and efficient fault diagnosis in industrial chemical applications.