Design and Simulation of Short Circuit Fault Detection in DC Power Supply using Daubechies Wavelet and Convolutional Neural Networks

Power supply systems are an essential component of current infrastructure and must operate reliably. Short-circuit defects pose a danger to power supply stability and may potentially result in equipment damage and system outages. This research presents a novel approach for predicting and detecting short circuit faults in power supply systems by combining the Daubechies wavelet transform with convolutional neural networks (CNNs). The researcher will make use of MATLAB Simulink and focuses on the design and simulation with short circuit faults introduced at known points of linear power supplies, boost converters, and buck converters. The voltage and current from the simulations were exported and then fed to Python with the use of PyWavelet for mathematical analysis. The researcher will make use of the db4 transform for the Daubechies wavelet. This decomposes the signals into detail coefficients at various scales, which are then converted into 1D vectors to be flattened and used for training with CNN. Based on the data and results, this shows 94.22% accuracy and 94.67% mAP (mean average precision) at 35 epochs. The researchers concluded that this study was feasible and reliable for implementation in real-world scenarios.