Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis

This paper proposes an unsupervised approach for anomaly detection in marine diesel engines using a transformer neural network based AutoEncoder (TAE) and residual analysis with Sequential Probability Ratio Test (SPRT) and Sum of Squares of Normalized Residuals (SSNR). This approach effectively capt...

Full description

Saved in:
Bibliographic Details
Main Authors: Qin Liang, Erik Vanem, Knut Erik Knutsen, Vilmar Æsøy, Houxiang Zhang
Format: Article
Language:English
Published: The Prognostics and Health Management Society 2024-10-01
Series:International Journal of Prognostics and Health Management
Subjects:
Online Access:https://papers.phmsociety.org/index.php/ijphm/article/view/3853
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849324934503858176
author Qin Liang
Erik Vanem
Knut Erik Knutsen
Vilmar Æsøy
Houxiang Zhang
author_facet Qin Liang
Erik Vanem
Knut Erik Knutsen
Vilmar Æsøy
Houxiang Zhang
author_sort Qin Liang
collection DOAJ
description This paper proposes an unsupervised approach for anomaly detection in marine diesel engines using a transformer neural network based AutoEncoder (TAE) and residual analysis with Sequential Probability Ratio Test (SPRT) and Sum of Squares of Normalized Residuals (SSNR). This approach effectively captures temporal dependencies within normal time-series data, eliminating the need for labeled failure data. To assess the performance of the proposed methodology, faulty data is collected under the same operational profile as normal training data. The TAE is trained on the normal data, after which the faulty data is tested using the trained model. Subsequently, the SPRT and SSNR methods are used to analyze residuals from the observed (input) and reconstructed (output or tested) faulty data. Deviations exceeding a predefined threshold are identified as anomalous behavior. Furthermore, this study explores various architectures of transformer neural networks and other types of neural networks to conduct a comprehensive comparative analysis of the performance of the proposed approach. Insights and recommendations derived from the performance analysis are also presented, which offers valuable information for potential users to leverage. The test results demonstrate the ability of the proposed approach to accurately and efficiently detect anomalies in marine diesel engines. Specifically, it can detect anomalies more than 1000 time steps ahead of system alarms, outperforming other tested models.
format Article
id doaj-art-4f85b42322f042acada048c0975737a2
institution Kabale University
issn 2153-2648
language English
publishDate 2024-10-01
publisher The Prognostics and Health Management Society
record_format Article
series International Journal of Prognostics and Health Management
spelling doaj-art-4f85b42322f042acada048c0975737a22025-08-20T03:48:32ZengThe Prognostics and Health Management SocietyInternational Journal of Prognostics and Health Management2153-26482024-10-01153113https://doi.org/10.36001/ijphm.2024.v15i3.3853Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual AnalysisQin Liang0https://orcid.org/0000-0002-1612-9840Erik Vanem1https://orcid.org/0000-0002-0875-0389Knut Erik Knutsen2Vilmar Æsøy3Houxiang Zhang4https://orcid.org/0000-0003-0122-0964DNV and Norwegian University of Science and TechnologyDNVDNVNorwegian University of Science and TechnologyNorwegian University of Science and TechnologyThis paper proposes an unsupervised approach for anomaly detection in marine diesel engines using a transformer neural network based AutoEncoder (TAE) and residual analysis with Sequential Probability Ratio Test (SPRT) and Sum of Squares of Normalized Residuals (SSNR). This approach effectively captures temporal dependencies within normal time-series data, eliminating the need for labeled failure data. To assess the performance of the proposed methodology, faulty data is collected under the same operational profile as normal training data. The TAE is trained on the normal data, after which the faulty data is tested using the trained model. Subsequently, the SPRT and SSNR methods are used to analyze residuals from the observed (input) and reconstructed (output or tested) faulty data. Deviations exceeding a predefined threshold are identified as anomalous behavior. Furthermore, this study explores various architectures of transformer neural networks and other types of neural networks to conduct a comprehensive comparative analysis of the performance of the proposed approach. Insights and recommendations derived from the performance analysis are also presented, which offers valuable information for potential users to leverage. The test results demonstrate the ability of the proposed approach to accurately and efficiently detect anomalies in marine diesel engines. Specifically, it can detect anomalies more than 1000 time steps ahead of system alarms, outperforming other tested models.https://papers.phmsociety.org/index.php/ijphm/article/view/3853marine diesel enginephmdeep learningmachine learning
spellingShingle Qin Liang
Erik Vanem
Knut Erik Knutsen
Vilmar Æsøy
Houxiang Zhang
Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
International Journal of Prognostics and Health Management
marine diesel engine
phm
deep learning
machine learning
title Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
title_full Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
title_fullStr Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
title_full_unstemmed Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
title_short Anomaly Detection of Marine Diesel Engines: A Novel Approach using Transformer Neural Networks for Reconstruction and Residual Analysis
title_sort anomaly detection of marine diesel engines a novel approach using transformer neural networks for reconstruction and residual analysis
topic marine diesel engine
phm
deep learning
machine learning
url https://papers.phmsociety.org/index.php/ijphm/article/view/3853
work_keys_str_mv AT qinliang anomalydetectionofmarinedieselenginesanovelapproachusingtransformerneuralnetworksforreconstructionandresidualanalysis
AT erikvanem anomalydetectionofmarinedieselenginesanovelapproachusingtransformerneuralnetworksforreconstructionandresidualanalysis
AT knuterikknutsen anomalydetectionofmarinedieselenginesanovelapproachusingtransformerneuralnetworksforreconstructionandresidualanalysis
AT vilmaræsøy anomalydetectionofmarinedieselenginesanovelapproachusingtransformerneuralnetworksforreconstructionandresidualanalysis
AT houxiangzhang anomalydetectionofmarinedieselenginesanovelapproachusingtransformerneuralnetworksforreconstructionandresidualanalysis