Alqunun Khalid, Bechiri Mohammed Bachir, Naoui Mohamed, Khechekhouche Abderrahmane, Marouani Ismail, Guesmi Tawfik, Alshammari Badr M, AlGhadhban Amer, Allal Abderrahim
Department of Electrical Engineering, College of Engineering, University of Ha'il, 2240, Ha'il, Saudi Arabia.
Laboratory of New Technologies and Local Development, University of El Oued, 39000, El Oued, Algeria.
Sci Rep. 2025 May 28;15(1):18739. doi: 10.1038/s41598-025-02439-4.
This study introduces an innovative method for addressing the bearing fault detection problem in rotating machinery. The proposed approach integrates multi-feature extraction, advanced feature selection, and state-of-the-art classification techniques using convolutional neural network (CNN) models. Leveraging the comprehensive Fault Bearing Dataset from Case Western Reserve University (CWRU), continuous wavelet transforms (CWT) and CNNs are utilized for feature extraction. The methodology also incorporates machine learning model tuning through Tree-Structured Parzen Estimators (TPE) for optimal hyperparameter adjustment, ensuring high-performance classification. Experimental results, based on the ResNet-50-SVM hybrid model, showed the effectiveness of the proposed approach in achieving an impressive accuracy of 95.51%. This confirms that the proposed methodology represents a significant advancement in bearing fault detection, providing an effective solution for predictive and preventive maintenance in industrial applications.
本研究介绍了一种用于解决旋转机械中轴承故障检测问题的创新方法。所提出的方法集成了多特征提取、先进的特征选择以及使用卷积神经网络（CNN）模型的最新分类技术。利用凯斯西储大学（CWRU）的综合故障轴承数据集，通过连续小波变换（CWT）和CNN进行特征提取。该方法还通过树结构帕曾估计器（TPE）进行机器学习模型调优，以实现最佳超参数调整，确保高性能分类。基于ResNet-50-SVM混合模型的实验结果表明，所提出的方法在实现95.51%的令人印象深刻的准确率方面是有效的。这证实了所提出的方法在轴承故障检测方面代表了一项重大进展，为工业应用中的预测性和预防性维护提供了有效的解决方案。
