Abstract: The global energy transition has continuously increased the demand for clean and sustainable energy. Lithium- ion batteries, with their high energy density, long cycle life, and efficient charge–discharge performance, have been widely applied in electric vehicles and renewable energy storage systems. However, short-circuit faults pose a serious threat to the safe operation of batteries, particularly in the early stage of short circuits, which are difficult to detect using traditional methods. To address this issue, this paper proposes a micro-short-circuit fault diagnosis method based on aggregated features of two-dimensional wavelet transform. The method does not rely on battery models or labeled fault data for training, and it enables real-time and accurate diagnosis even under noise interference. Specifically, Gaussian smoothing filtering (GSF) is first applied to the collected voltage data to suppress noise and enhance data usability. Then, horizontal and vertical detail components are extracted through two-dimensional wavelet decomposition, and correlation-coefficient-based aggregated features are constructed for fault identification. Experimental results demonstrate that the proposed method can effectively detect short-circuit faults under different operating conditions. The anomaly score increases significantly when a fault occurs, while remaining at a low level during normal operation, thereby achieving efficient diagnosis with a low false alarm rate. The method shows strong generalization ability, offering a new approach to enhancing the safety of battery systems.