To guarantee a steady supply of electricity and self-sufficiency, power transmission and transformation equipment must operate reliably. In order to guarantee the safe and dependable functioning of the self-supply power system, it is crucial to continuously check the equipment's health state and to swiftly detect and eradicate any safety threats. Any anomalies in the equipment will cause composite electromagnetic vibration parameters to deviate considerably from normal operation, which in turn reflects the equipment's health state. With the ever-improving state monitoring technology, power transmission and transformation equipment may have their operational condition determined in real-time, which can lead to the discovery of potential equipment fault risks and the provision of technical assistance for maintenance and fault prediction. The purpose of this study is to present a Convolutional Block Attention Module—Multi-scale Convolutional Neural Network (CBAM-MSCNN) fault detection model for self-supply system power transmission and transformation equipment status monitoring. This model can assess composite electromagnetic vibration characteristics efficiently. The model structure of the convolutional neural network is expanded from one channel to several channels, allowing for the extraction of more comprehensive essential characteristics from the original signals. This is achieved by using convolution kernels of various sizes and multiple channels. To help the model zero in on useful data and ignore noise, we've included a Convolutional Block Attention Module (CBAM) attention mechanism into every channel. To mitigate diagnostic mistakes brought on by a single channel's undue load, a gate control module is inserted after every channel. The model's robust generalizability and noise resistance characteristics are demonstrated by the experimental findings.