EfficientNetV2 Pump Anomaly Detection Method Based on Improved CBAM Attention Mechanism

Fuyue Sun
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:404-413, 2025.

Abstract

In the process of industrial production, pump as the core equipment, its stable operation directly affects the safety and production efficiency of the factory. However, in practical applications, pumps often face various abnormal conditions, which may lead to equipment failure and even safety accidents in serious cases. To solve this problem, we design a new anomaly detection method, which improves the EfficientNetV2 network: The original Attention Module is replaced with the optimized CBAM module, its channel attention is retained to capture the cross-channel dependencies, and the Simple Attention Module (SimAM) is introduced into the spatial attention part to effectively reduce the computational complexity and enhance the sensitivity of the model to local details and global context information. In order to better deal with the problem of data imbalance, we use MixUp data augmentation and Label Smoothing regularization strategy in the training process, and choose BCEWithLogitsLoss as the loss function. In the pre-training phase, the pump-related audio modules in the MIMII dataset are used for weight initialization, which is subsequently fine-tuned on the pump anomaly binary classification task. Experimental results show that the proposed model improves the classification accuracy by 0.76% compared with the original EfficientNetV2-Small network under the same data set and evaluation metrics, which verifies the effectiveness and superiority of the architecture optimization in pump anomaly detection.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-sun25a, title = {EfficientNetV2 Pump Anomaly Detection Method Based on Improved CBAM Attention Mechanism}, author = {Sun, Fuyue}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {404--413}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/sun25a/sun25a.pdf}, url = {https://proceedings.mlr.press/v278/sun25a.html}, abstract = {In the process of industrial production, pump as the core equipment, its stable operation directly affects the safety and production efficiency of the factory. However, in practical applications, pumps often face various abnormal conditions, which may lead to equipment failure and even safety accidents in serious cases. To solve this problem, we design a new anomaly detection method, which improves the EfficientNetV2 network: The original Attention Module is replaced with the optimized CBAM module, its channel attention is retained to capture the cross-channel dependencies, and the Simple Attention Module (SimAM) is introduced into the spatial attention part to effectively reduce the computational complexity and enhance the sensitivity of the model to local details and global context information. In order to better deal with the problem of data imbalance, we use MixUp data augmentation and Label Smoothing regularization strategy in the training process, and choose BCEWithLogitsLoss as the loss function. In the pre-training phase, the pump-related audio modules in the MIMII dataset are used for weight initialization, which is subsequently fine-tuned on the pump anomaly binary classification task. Experimental results show that the proposed model improves the classification accuracy by 0.76% compared with the original EfficientNetV2-Small network under the same data set and evaluation metrics, which verifies the effectiveness and superiority of the architecture optimization in pump anomaly detection. } }
Endnote
%0 Conference Paper %T EfficientNetV2 Pump Anomaly Detection Method Based on Improved CBAM Attention Mechanism %A Fuyue Sun %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-sun25a %I PMLR %P 404--413 %U https://proceedings.mlr.press/v278/sun25a.html %V 278 %X In the process of industrial production, pump as the core equipment, its stable operation directly affects the safety and production efficiency of the factory. However, in practical applications, pumps often face various abnormal conditions, which may lead to equipment failure and even safety accidents in serious cases. To solve this problem, we design a new anomaly detection method, which improves the EfficientNetV2 network: The original Attention Module is replaced with the optimized CBAM module, its channel attention is retained to capture the cross-channel dependencies, and the Simple Attention Module (SimAM) is introduced into the spatial attention part to effectively reduce the computational complexity and enhance the sensitivity of the model to local details and global context information. In order to better deal with the problem of data imbalance, we use MixUp data augmentation and Label Smoothing regularization strategy in the training process, and choose BCEWithLogitsLoss as the loss function. In the pre-training phase, the pump-related audio modules in the MIMII dataset are used for weight initialization, which is subsequently fine-tuned on the pump anomaly binary classification task. Experimental results show that the proposed model improves the classification accuracy by 0.76% compared with the original EfficientNetV2-Small network under the same data set and evaluation metrics, which verifies the effectiveness and superiority of the architecture optimization in pump anomaly detection.
APA
Sun, F.. (2025). EfficientNetV2 Pump Anomaly Detection Method Based on Improved CBAM Attention Mechanism. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:404-413 Available from https://proceedings.mlr.press/v278/sun25a.html.

Related Material