The invention provides a motor fault diagnosis method for deep learning network of data fusion. A neural network includes a data compression network, a feature extraction network and a classificationnetwork. A determination and training methods comprises the steps of (1) collecting current signals of A and B phases of a motor and a vibration signal of a motor end bearing, performing data standardization, obtaining a spectrum sequence by Hilbert-Huang transform, and establishing a data set of the neural network, (2) establishing a deep neural network, determining a network structure and initializing parameters, (3) inputting a training set into the neural network, calculating loss functions of different neural networks, and updating neural network parameters by using a loss value, and (4)inputting the data of a test set into a neural network, calculating an accurate rate, and repeating the step (3) until the accurate rate satisfies a requirement, wherein the neural network can map inputted current and vibration data to a feature plane after training, and a classification network can predict whether the motor is failed according to a fault state corresponding to a region where thenetwork is located.