Motor Fault Diagnosis Based on Generative Adversarial Network Using Hyperchaotic Sequences and Mixed-Dimensional Network

Fault is extremely destructive in industrial process, and imbalanced data greatly affect the accuracy of fault diagnosis. Many methods have been proposed to deal with imbalanced data, but the concern for improving the performance of fault diagnostic networks is not enough. Therefore, novel modified conditional generative adversarial network (MCGAN) based on memristive hyperchaotic sequences and mixed-dimensional convolutional neural network (MCNN) is proposed. The 2-D data are obtained by fast Fourier transform and piecewise reconstruction of vibration signals. A novel tanh-input-type memristive hyperchaotic map is utilized to obtain chaos-based random noises. MCGAN can generate synthetic samples for augmenting the fault sample and reducing the imbalanced rate, and chaos-based random noises are used as the noise variable of MCGAN to generate high-quality synthetic samples. By cascading convolution layers with different dimensions, the lightweight MCNN is designed to improve accuracy of motor fault diagnosis. Experiments are implemented using the Case Western Reserve University and practical laboratory platform. The results show that the accuracy of the proposed method is higher than that of some diagnostic networks under imbalanced data.