Gearbox fault diagnosis based on multi-channel orthogonal attention network

Xiang Wu, Yeze Fu, Qi Wu, Zhenyu Wen, Hui Dong, Dan Zhang

Abstract: The gearbox is a key component of industrial equipment transmission systems, and research on fault diagnosis methods is crucial for enhancing its safety and reliability. However, traditional methods often fail to fully account for the redundancy and complementarity of data from multiple sensors. To address this issue, this paper proposes a gearbox fault diagnosis method based on a multi-channel orthogonal attention network. First, a multi-channel convolutional neural network architecture is introduced, where each sensor channel is assigned an independent convolutional layer to ensure that the channels do not interfere with each other. This approach enhances the accuracy of feature extraction from multi-sensor data and improves the model’s adaptability. Second, a multi-channel convolutional neural network model based on a residual orthogonal attention mechanism is designed. This mechanism dynamically adjusts the sensor channel weights to reduce data redundancy and enhance complementarity, allowing the model to focus more on critical fault features. Finally, the proposed method is validated and proven to be effective and superior through experiments on a gearbox dataset.

External IDs:dblp:journals/sivp/WuFWWDZ25