Faults are common and potentially dangerous problems in low-voltage power system networks, e.g., distribution networks. Upon discovery, they must be dealt with immediately to prevent further damage to the distribution system and quickly return electricity to consumers. This paper proposes a hybrid fault diagnosis method combining a signal processing technique with deep learning models. Faults are applied in a four-node test distribution feeder developed using MATLAB and Simulink. Different measurement noise levels are implemented, along with varying load and fault parameters. The short-time Fourier transform (STFT) is used on the feeder’s current signals as a feature extraction tool. These features are then used to train deep learning models to detect, classify, and locate faults. Various parameters of the models are varied to find the optimum ones, which are used to obtain the results. The findings show that the model performed exceptionally well in fault detection and classification and satisfactorily in fault location.