Abstract: Addressing the limitations of traditional power cable fault monitoring methods—such as reliance on manual feature extraction, weak interference resistance, and low fault identification accuracy—this paper proposes an intelligent fault monitoring solution based on Convolutional Neural Networks (CNN). By establishing a multimodal sensing system, the approach enables the acquisition and preprocessing of fault signals including partial discharges, vibrations, and temperature data. An enhanced CNN model incorporating multi-scale convolutional and attention mechanisms achieves adaptive extraction and deep enhancement of high-dimensional fault features. Laboratory simulations and field measurements demonstrate that the model achieves an average recognition accuracy of 93.3% for typical cable faults, significantly outperforming traditional SVM and BP neural network algorithms. Its lightweight design meets real-time monitoring requirements at edge devices, providing intelligent technical support for the safe operation and maintenance of power cables throughout their entire lifecycle.