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

This article presents an online fault detection system for a laboratory-scale batch reactor (BR) using the Convolutional Neural Network (CNN)-Squeeze and Excitation-based Improved Multi-Layer Long Short-Term Memory (CS-IMLSTM) model. To identify superimposed and sparse sensor faults in real time, the system continuously monitors the BR parameters, such as reactor temperature, coolant flow rate, and heater current. To reduce noise and dynamic fluctuations, the CS-IMLSTM integrates a channel-spatial attention mechanism and enhances feature significance. The performance of the proposed model is compared with LSTM and CNN-LSTM models. The results indicate that the CS-IMLSTM demonstrates improved accuracy, faster adaptation of online learning, and effectiveness in identifying abnormal circumstances compared with LSTM and CNN-LSTM models. The proposed approach can be used for intelligent predictive maintenance in dynamic industrial environments to enhance the reliability and safety of chemical process operations.