Taibi Ahmed, Ikhlef Nabil, Aomar Lyes, Touati Said, Baitiche Oussama, Alsabah Yousef A, Rabehi Abdelaziz, Guermoui Mawloud, Benghanem Mohamed
Electronics and Industrial Electrical Laboratory (L2EI), Univercity of Jijel, Jijel, Algeria.
Nuclear Research Center of Birine (CRNB), Bp 180, Ain Oussera, Algeria.
Sci Rep. 2025 Sep 1;15(1):32128. doi: 10.1038/s41598-025-12407-7.
Misalignment is among the most frequent mechanical faults in rotating electrical machines, often resulting in partial or complete motor failure over time. To tackle this issue, the present study proposes an innovative methodology for diagnosing misalignment faults in rotating electrical machines. The method integrates the dual-tree complex wavelet transform with a refined composite multiscale fluctuation dispersion entropy algorithm (DTCWT-RCMFDE) for feature extraction, combined with the least-squares support vector machines algorithm (LSSVM) for fault classification. Initially, the DTCWT is employed to decompose the torque signal into multiple sub-bands using range entropy (RE). Subsequently, the RCMFDE is calculated for each sub-band to construct discriminative fault feature vectors. These vectors are then used to train and test the LSSVM classifier to identify different types of misalignment faults. The proposed method was validated using experimental data, and the results demonstrate its superior diagnostic performance. Compared to existing approaches, the DTCWT-RCMFDE-LSSVM model achieved the highest classification accuracy of 98.33%, outperforming other methods such as MSE-SVM (94.1%), DTCWT-EE-PSO-SVM (96%), Multi-features-t-SNE-LSSVM (96.25%) and AR model coefficients-mRMR-SOM neural network (97.22%). These findings confirm the method's high precision in detecting both parallel and angular misalignments. This research holds significant potential for industrial applications in sectors reliant on rotating machinery such as power generation, petrochemical, nuclear, and manufacturing where early and accurate fault detection is essential to minimize downtime and enhance operational reliability.
不对中是旋转电机中最常见的机械故障之一，随着时间的推移，常常会导致电机部分或完全失效。为了解决这个问题，本研究提出了一种用于诊断旋转电机不对中故障的创新方法。该方法将双树复小波变换与改进的复合多尺度波动分散熵算法（DTCWT-RCMFDE）相结合进行特征提取，并结合最小二乘支持向量机算法（LSSVM）进行故障分类。首先，利用DTCWT通过范围熵（RE）将转矩信号分解为多个子带。随后，计算每个子带的RCMFDE以构建具有判别力的故障特征向量。然后使用这些向量训练和测试LSSVM分类器，以识别不同类型的不对中故障。所提出的方法通过实验数据进行了验证，结果表明了其卓越的诊断性能。与现有方法相比，DTCWT-RCMFDE-LSSVM模型实现了98.33%的最高分类准确率，优于其他方法，如MSE-SVM（94.1%）、DTCWT-EE-PSO-SVM（96%）、多特征-t-SNE-LSSVM（96.25%）和AR模型系数-mRMR-SOM神经网络（97.22%）。这些发现证实了该方法在检测平行不对中和角度不对中方面的高精度。这项研究在依赖旋转机械的工业应用领域具有巨大潜力，如发电、石化、核能和制造业，在这些领域中，早期准确的故障检测对于最小化停机时间和提高运行可靠性至关重要。
