Multi-Mode High-Precision Fault Diagnosis Control System Manage Method

The high-precision fault diagnosis management method in multi-mode control systems was studied. Three typical power conversion circuits were selected to use full convolutional network (FCN), residual network (RESNET) and self-attention network (TRANSFORMER) for fault diagnosis algorithm research, and a fault diagnosis management method verification platform was built based on this. The algorithm model was built using Python in the Pytorch environment on a computer with 16GB RAM and a 3060 graphics card. The switching power supply circuit based on Tps53355 was analyzed by Pspice sensitivity, and 6 devices with the greatest impact on the output were obtained. The Adam optimizer was used, the total number of epochs trained was 60, and the mini-batch size was fixed to 64. The simulation verification showed that this model can achieve the system engineering management requirements: the fault detection accuracy is not less than 91%, the false alarm rate of lightweight power warning is not higher than 1.05%, and under multiple faults, no less than 3 composite faults can be identified.