Fault Detection and Diagnosis Using Combined Autoencoder and Long Short-Term Memory Network

Area of Science:

• Industrial Process Safety
• Machine Learning for Time Series Analysis

Background:

• Accurate fault detection and diagnosis are crucial for industrial process safety.
• Handling rare events in multivariate time series data presents significant challenges.

Purpose of the Study:

• To propose an integrated learning approach for joint fault detection and diagnosis of rare events.
• To enhance system safety by accurately identifying and classifying faults in industrial processes.

Main Methods:

• An autoencoder is employed for anomaly detection to identify rare fault events.
• A Long Short-Term Memory (LSTM) network is utilized for classifying different fault types.
• The approach combines autoencoder's nonlinear representation learning with LSTM's time series analysis capabilities.

Main Results:

• The integrated approach accurately detects deviations from normal operational behavior.
• It effectively identifies specific types of faults within a useful time frame.
• Performance was validated against a deep convolutional neural network on the Tennessee Eastman process.

Conclusions:

• The proposed integrated learning approach offers a robust solution for fault detection and diagnosis.
• This method enhances industrial process safety by improving the reliability of rare event identification and classification.