In order to achieve high accuracy in hybrid fault diagnosis of rotating parts, a hybrid fault diagnosis method based on information fusion of two channels under variable working conditions is proposed in this paper. The signal includes both rolling bearing and gear vibration signal. The vibration signal of channel 1 is generated by generalized S transform, and the feature map is used as the model input of channel 1. In channel 2, the time-domain signals of the rotating parts are taken as the characteristic input, and the two-channel output layer is randomly fused with the features. By fine-tuning the parameters of the whole two-channel convolutional neural network (CNN) model, the diagnosis and identification of the mixed fault state of the rotating parts under varying working conditions are realized. The results show that the proposed method can be effectively applied to the mixed fault identification and diagnosis of rotating parts. Compared with the one-dimensional and two-dimensional convolutional neural network and other machine learning methods, the proposed method has the highest fault identification accuracy, reaching 98.18%.