In the actual work, the internal structure of the induction motor is complex and the operating conditions are high. In view of the above problems, a motor fault diagnosis method based on frequency domain-transformer model is proposed. The induction motor fault signal is represented by different vibration modes by fast Fourier transform and converted into a feature sequence, and the motor fault diagnosis is further realized in combination with the multi-head attention mechanism and SELU activation function in the Transformer model. In this paper, the feasibility of the method is verified and the average accuracy is 99.91%. Besides, comparing the noise resistance with MLP, CNN, AlexNet and BiLSTM, the average classification accuracy of the method is much higher than the other four models. The experimental results show that the motor fault diagnosis method combining FFT based on Transformer model not only has high diagnosis accuracy, but also has excellent noise resistance.