Anomaly Detection in Asset Degradation Process Using Variational Autoencoder and Explanations

Abstract

Development of predictive maintenance (PdM) solutions is one of the key aspects of Industry 4.0. In recent years, more attention has been paid to data-driven techniques, which use machine learning to monitor the health of an industrial asset. The major issue in the implementation of PdM models is a lack of good quality labelled data. In the paper we present how unsupervised learning using a variational autoencoder may be used to monitor the wear of rolls in a hot strip mill, a part of a steel-making site. As an additional benchmark we use a simulated turbofan engine data set provided by NASA. We also use explainability methods in order to understand the model’s predictions. The results show that the variational autoencoder slightly outperforms the base autoencoder architecture in anomaly detection tasks. However, its performance on the real use-case does not make it a production-ready solution for industry and should be a matter of further research. Furthermore, the information obtained from the explainability model can increase the reliability of the proposed artificial intelligence-based solution.

Explainable Artificial Intelligence Model for Predictive Maintenance in Smart Agricultural Facilities

“…Prognostics predict future system performance, while diagnostics identify current issues through performance analysis. Among the reviewed literature, eleven studies focused on prognostics (Garouani et al, 2022;Gashi et al, 2023;Ghasemkhani et al, 2023;Hajgató et al, 2022;Hermansa et al, 2021;Jakubowski et al, 2021;Kononov et al, 2023;Kuzlu et al, 2020;Serradilla et al, 2021;Upasane et al, 2021;Wu et al, 2021), three on anomaly detection (Choi et al, 2022;Langone et al, 2020;Mey & Neufeld, 2022), and two on both prognostics and diagnostics (Serradilla et al, 2021;Steurtewagen & Van den Poel, 2021). Notably, none focused solely on diagnostics, revealing a significant research gap.…”

“…Ensemble Learning (EL) offered diverse algorithmic solutions and found utility in the prognostics realm of manufacturing industries (Garouani et al, 2022). Despite their complexity, other methods like Balanced K-Star, Multi-Layer Perceptron (MLP), Extreme Learning Machine (ELM), and Transfer Learning (TL) provide alternative approaches, as do Deep Convolutional AutoEncoders (Hajgató et al, 2022;Huang et al, 2006;Jakubowski et al, 2021;Serradilla et al, 2021). Diagnostics in PdM have been less explored, with only a few studies touching on it (Serradilla et al, 2021).…”

Prediction of Technical State of Mechanical Systems Based on Interpretive Neural Network Model

“…For example, autoencoders have been successfully used for feature extraction [ 5 ], multi-sensory data fusion [ 6 ], fault diagnosis [ 7 , 8 ], and anomaly detection [ 9 ] because they do not require prior knowledge of the data and can compress and fuse multi-sensory data. Autoencoders are not as efficient at reconstruction compared to generative adversarial networks (GANs) [ 10 ], which allows GANs to solve the class imbalance problem [ 11 , 12 ].…”

“…However, the interpretability of neural network models is still considered only in very few publications that include a mention of PdM [ 34 ]. Even among them, the issue has been raised only for models of RUL prediction [ 35 , 36 ] and anomaly detection [ 9 ]. It was also observed that the analysis of the contribution of features was often limited to the calculation of SHAP values only, without any hypotheses being put forward and confirmed.…”

An Unsupervised Machine Learning Approach for Monitoring Data Fusion and Health Indicator Construction

“…With the development of autoencoder methods, derivative methods such as the denoising autoencoder [ 25 ], sparse autoencoder [ 26 ], and stacked autoencoder [ 27 ] have been invented. In the field of PHM research, the autoencoder has also been gradually applied to anomaly detection [ 28 , 29 ], RUL prediction [ 30 ], fault diagnosis [ 31 , 32 ], and health status assessment [ 33 ].…”