Parallel CNN three-phase Inverter Open-Circuit Fault Diagnosis based on STFT and GADF Images

To apply the advantages of deep learning in 2D image recognition to the fault diagnosis of three-phase inverters, this paper proposes a Parallel Convolutional Neural Network (CNN) model based on short-time Fourier transform (STFT) and Gramian Angular Difference Field (GADF) images. Firstly, the one-dimensional voltage signal output by the inverter is converted into a two-dimensional image using STFT and GADF methods. Then, the transformed image data is input into the dual-channel parallel CNN model, where convolution and pooling operations are carried out in the two CNN branches to extract multi-level features of the fault signal. Finally, through feature fusion and fully connected layers, effective classification and diagnosis of inverter open-circuit faults are achieved. Experimental results show that the classification accuracy of the proposed method exceeds 97% under different noise levels. Compared to other methods, the proposed approach demonstrates significant advantages in accuracy and reliability, with strong anti-interference ability and robustness when processing noisy data.