Deep residual networks-based intelligent fault diagnosis method of planetary gearboxes in cloud environments

With the widespread use of deep learning (DL) related techniques, many end-to-end planetary gearbox fault diagnosis models have been proposed. The input samples of DL models are often composed of time-domain signals, frequency-domain signals, or time–frequency domain signals from multiple rotation periods. So, redundant information is unavoidable, which may cause difficulty in the extraction of subtle fault features by the DL model and affect the accuracy of fault diagnosis. In addition, DL-based intelligent fault diagnosis (IFD) methods are developed based on a large amount of data. It is difficult to realize real-time fault diagnosis by relying on local equipment with limited computing power. Therefore, this paper proposes a deep residual networks-based IFD method of planetary gearboxes in cloud environments. A cloud-based IFD design is proposed to use the super computing power of cloud computing to solve the related issues caused by the insufficient computing power of local equipment. This method takes the wavelet time–frequency images of vibration signals as the network input. In order to obtain high accuracy, a method specialized for the selection of wavelet basis function (WBF) is developed based on the difference between the original signals and the reconstructed wavelet signals. Channel attention depth residual networks (CADRN) are used as the diagnosis model, and the channel attention module (CAM) is applied to enhance important fault features and improve the diagnosis accuracy. An ablation study and comparative experiments with three mainstream methods were carried out on a real-world dataset of planetary gearboxes, and the proposed method achieved more than 99% average accuracy rate, which verified the effectiveness of the proposed method.