Online Sensor Fault Detection Using Machine Learning Algorithms on a Laboratory-Scale Batch Reactor: LSTM Approach.

Area of Science:

• Chemical Engineering
• Artificial Intelligence
• Process Control

Background:

• Batch reactors are crucial in chemical processes but susceptible to sensor faults.
• Real-time fault detection is essential for operational safety and efficiency.
• Existing methods may struggle with complex, superimposed, or sparse sensor faults.

Purpose of the Study:

• To develop and evaluate an online fault detection system for laboratory-scale batch reactors.
• To enhance the accuracy and speed of fault identification in dynamic industrial environments.
• To improve the reliability and safety of chemical process operations through intelligent predictive maintenance.

Main Methods:

• Implementation of a Convolutional Neural Network (CNN)-Squeeze and Excitation-based Improved Multi-Layer Long Short-Term Memory (CS-IMLSTM) model.
• Continuous monitoring of batch reactor parameters including temperature, coolant flow rate, and heater current.
• Integration of a channel-spatial attention mechanism to reduce noise and enhance feature significance.

Main Results:

• The CS-IMLSTM model demonstrated superior accuracy in fault detection compared to traditional LSTM and CNN-LSTM models.
• The proposed system exhibited faster adaptation capabilities for online learning.
• Effective identification of superimposed and sparse sensor faults was achieved in real-time.

Conclusions:

• The CS-IMLSTM model offers a robust solution for online fault detection in batch reactors.
• The developed system can be applied to intelligent predictive maintenance in dynamic industrial settings.
• This approach significantly enhances the safety and reliability of chemical process operations.