Gearbox Fault Diagnosis Based on a Dilated Convolutional Neural Network

As a critical transmission component widely used in industrial applications, the gearbox has significant engineering value in fault diagnosis research. To address the problem of gearbox fault diagnosis, a diagnostic method based on a one-dimensional dilated convolutional neural network (1DD-CNN) is proposed. First, the fault mechanisms of the gearbox are analyzed. Based on this analysis, a dilated convolutional network tailored for gearbox fault diagnosis is designed and validated using the PHM-Beijing 2024 dataset, achieving an accuracy exceeding 99%, which demonstrates its outstanding diagnostic performance. Furthermore, the proposed model is compared with other models such as ResNet and AlexNet. The results show that the 1DD-CNN model achieves higher accuracy and better performance. Under varying signal-to-noise ratios (SNR), the 1DD-CNN consistently achieves accuracy above 98%, outperforming ResNet, AlexNet, and other baseline models. Its noise resistance and diagnostic capability are significantly superior to those of traditional methods. The model achieves a diagnostic accuracy of 99.36% on the test set, indicating strong generalization and robustness. The proposed method exhibits clear advantages in feature extraction and noise robustness, offering new insights into gearbox fault diagnosis and demonstrating substantial theoretical and practical value.