51 / 2025-03-29 17:02:23

Fault Diagnosis of Automotive Body-in-White Welding Production Line Based on Imbalanced Data

synthetic minority over-sampling technique,ResNet,fault diagnosis,bearings

In practical production scenarios, especially in the automotive white body welding production line, the sample distribution across different fault categories is often highly imbalanced, posing significant challenges for data-driven methods in handling imbalanced data, which in turn limits the fault recognition and classification capabilities. To address the imbalance issue of bearing vibration data in the welding production line, an improved Residual Networks method combining Synthetic Minority Over-sampling Technique and Squeeze-and-Excitation module is proposed. The Synthetic Minority Over-sampling Technique  balances the class distribution by generating synthetic samples, significantly improving the classification ability of minority class samples. The Squeeze-and-Excitation module introduces a channel attention mechanism that dynamically adjusts the feature channel weights, enhancing the model's focus on key features. Experimental results show that the proposed method outperforms other methods in classification metrics such as F1 Score, Precision, and Recall, particularly achieving an improvement in the classification accuracy of the minority class from 87.50% to 100%.