Comparative analysis of switching power supply fault diagnosis model based on particle swarm optimization and convolutional neural network

Abstract Due to the complex topology of switching power supply, the traditional fault diagnosis method has been unable to meet the needs of engineering. A neural network model is developed to enhance the efficiency and accuracy of fault diagnosis in switching power supplies. This model utilizes the fault characteristics post wavelet packet transformation as input and predicts the corresponding fault type as output. We establish an experimental platform for forward switching power supply, construct 10 typical fault modes, collect 30 groups of output voltage signals under each fault mode, and extract the fault characteristics after wavelet packet transform as data samples. Simultaneously utilizing the PSO-BP neural network and CNN neural network enables the realization of fault diagnosis and analysis for the switching power supply. Various test outcomes have demonstrated that the fault detection ratio of the PSO-BP neural network structure exceeds 95%, with a speed of convergence achieving 60 seconds. Meanwhile, the fault detection ratio of the CNN neural network structure exceeds 90%, with a convergence speed of 31 seconds. These both demonstrate remarkable performance in fault diagnosis and reliability.