Quality-related fault detection for dynamic process based on quality-driven LSTM and autoencoder

Fault detection always plays an important role in industrial dynamic processes since it can avoid production loss in time. However, since industrial dynamic process becomes more and more coupled and complex, which brings uneven dynamics hidden in collected data, extracting dynamic features effectively faces lots of challenges. In addition, it is a tricky business to distinguish whether the fault that occurs is quality-related or not, resulting in unnecessary repairing and large losses. In order to deal with these issues, this paper comes up with a novel fault detection method based on quality-driven long short-term memory and autoencoder (QLSTM-AE). Specifically, LSTM network is first used to extract dynamic features, and quality variables are embedded in parallel to learn the quality-related features. Then, a fault detection strategy based on reconstruction error statistic SPE and the quality monitoring statistic H
²
is designed, which can distinguish various types of faults to realize accurate monitoring for dynamic processes. Finally, Tennessee Eastman (TE) benchmark process demonstrate the reliability and effectiveness of the proposed novel method, which indicates it has higher accuracy and can separate different faults efficiently compared to some state-of-the-art methods.