Research And Application Of Multimode Fault Early Warning For Switching Power Supply In Chemical Plant

As an indispensable part of all kinds of electrical equipment,the health of switching power supply is related to the reliability of equipment operation,and also to the production safety of the whole chemical plant.Due to the complex process flow of chemical enterprises,the use of a large number of electrical equipment,the damage of switching power supply is also a common fault.According to the actual measurement,the failure of the filter capacitor in the switching power supply will inject harmonic signals into the power grid,which will affect the overall power supply quality of the power grid.In serious cases,it will cause abnormal operation of other electrical equipment in the power grid,resulting in chain reaction.Therefore,research on the early warning of switching power failure and Diagnosis is of great significance.The change of raw material ratio,load,temperature and other factors lead to the multi-modal working condition in the chemical production process,which leads to the inaccuracy of single neural network model in the fault diagnosis of switching power supply.In view of this situation,taking industrial switching power supply as an example,this paper proposes a multi-modal depth neural network,which uses hierarchical depth neural network.First,the first level network carries out modal identification for multi-modal data,and then establishes a depth neural network diagnosis model for each mode on the second level network,so as to improve the fault diagnosis accuracy of switching power supply under multi-modal conditions.Firstly,this paper introduces the basic principle of switching power supply,analyzes the failure model and failure mode of common failure electronic components,expounds the algorithm core idea of BP neural network,and analyzes the convolutional neural networks(CNN)and combined with the idea of residual learning,a kind of deep residual network(Res Net)with the depth up to dozens of layers is finally constructed.In the training of network model,super parameter optimization algorithm,regularization algorithm and dropout algorithm are introduced to improve the robustness of network model.Finally,the fault diagnosis model is compressed and optimized,so that it can be transplanted to the Zynq So C development platform,so that the fault diagnosis has practical application value,and build the switch power fault detection system and friendly interface.