Quality-related fault detection for dynamic process based on quality-driven long short-term memory network and...

Area of Science:

• Industrial Process Monitoring
• Machine Learning for Fault Detection
• Data Analytics in Manufacturing

Background:

• Industrial dynamic processes are complex, making fault detection challenging.
• Distinguishing quality-related faults is difficult, leading to losses.
• Existing methods struggle with uneven dynamics and feature extraction.

Purpose of the Study:

• To develop a novel fault detection method for complex industrial dynamic processes.
• To improve the accuracy of fault detection by incorporating quality variables.
• To effectively distinguish between different types of faults, including quality-related ones.

Main Methods:

• Utilized a quality-driven long short-term memory and autoencoder (QLSTM-AE) model.
• Employed a long short-term memory (LSTM) network for dynamic feature extraction.
• Incorporated quality variables in parallel and used reconstruction error (SPE) and Hotelling T² (H²) statistics for fault detection.

Main Results:

• The QLSTM-AE method demonstrated high accuracy in fault detection.
• The proposed strategy effectively distinguished between various fault types.
• Experiments on simulations and the Tennessee Eastman (TE) benchmark confirmed the method's reliability and effectiveness.

Conclusions:

• The QLSTM-AE method offers a reliable and accurate approach to fault detection in industrial dynamic processes.
• It successfully addresses challenges related to complex dynamics and quality-related fault identification.
• The method outperforms existing state-of-the-art techniques in accuracy and fault separation.