The accurate detection of arc faults is critical for ensuring system safety and reliability, particularly in complex electrical networks. In this study, we propose a deep learning model that leverages advanced preprocessing techniques, including time-series data transformation into images, to enhance the detection capabilities. By utilizing methods such as Continuous Wavelet Transform (CWT) and deep convolutional neural networks (CNNs), our approach captures both temporal and frequency-domain features of voltage signals. We applied this model to real-world voltage data, focusing on abnormal event detection such as arc faults. The results demonstrate the model's ability to accurately classify normal and abnormal voltage patterns with significant precision, showcasing the potential for enhanced fault detection in power systems. Our findings suggest that this method could contribute to improved fault prediction and maintenance in various industrial applications.