Industrial Process Fault Detection with Multi-View Feature Fusion Attention Mechanism Assisted Autoencoder Network

Fault detection technology plays a crucial role in ensuring the safety and efficiency of industrial processes. Industrial processes are subject to changing operating conditions, causing their characteristics to evolve over time and exhibit dynamic, non-Gaussian, and non-linear behaviors in the data....

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE International Conference on Software Quality, Reliability and Security Companion (QRS-C) (Online) S. 1113 - 1122
Hauptverfasser: Tong, Yifan, Wang, Zhaojing, Yan, Xiaoyun, Hu, Xinrong, Li, Lijun
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.07.2024
Schlagworte:
Accuracy
attention mechanism
Attention mechanisms
Data mining
Data models
dynamic feature
Fault detection
Feature extraction
Independent component analysis
non-Gaussian feature
non-linear feature
Security
Software quality
Software reliability
ISSN:2693-9371
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Fault detection technology plays a crucial role in ensuring the safety and efficiency of industrial processes. Industrial processes are subject to changing operating conditions, causing their characteristics to evolve over time and exhibit dynamic, non-Gaussian, and non-linear behaviors in the data. Unfortunately, existing research in process monitoring often fails to address these properties simultaneously, leading to unsatisfactory fault detection outcomes. To address this gap, a novel fault detection framework is introduced in this study, leveraging a Multi-View Feature Fusion At-tention Mechanism Assisted Autoencoder Network (MFA-AE). To capture the dynamic and non-Gaussian features of industrial processes comprehensively, slow feature analysis and independent component analysis are initially employed to extract features separately from the raw data. Subsequently, an attention mechanism is integrated to fuse these extracted features with the raw data, enhancing the information available for subsequent modeling. These integrated features are then passed through the Autoencoder (AE) network, where non-linear features are further explored, and a fault detection model is constructed based on the residuals of the AE network. To assess the effectiveness and feasibility of the MFA-AE approach, experiments are conducted using the Tennessee Eastman process. The results affirm the superiority of the proposed method over state-of-the-art techniques in terms of detection accuracy, demonstrating the promising potential of the MFA-AE scheme for enhancing fault detection in industrial processes.
ISSN:2693-9371
DOI:10.1109/QRS-C63300.2024.00147