Abstract: To solve the problems of an imperfect index system and insufficient intelligence level in power equipment condition detection, this paper focuses on the core fault characteristics of 500kV transformers and switchgear equipment, and constructs a multi-dimensional condition assessment index system covering oil chemical test indicators, as well as tripping coil and closing coil current detection. The scheme integrates a multi-source sensor data acquisition and preprocessing system, and adopts an improved Kalman filter and time series interpolation algorithm to optimize data quality; it designs a fault diagnosis model based on a multi-scale convolutional neural network, and integrates deep learning models, support vector machines and random forests through a decision fusion framework to improve assessment accuracy, thus forming a full-process technical solution from data acquisition to condition early warning. Simulation verification results show that the proposed scheme achieves a detection accuracy of 94.2% for transformer winding overheating faults, and the detection sensitivity for tripping coil current anomalies of switchgear equipment is significantly improved; the overall fault identification accuracy is stably above 92.5%, and the detection latency is controlled within 100ms. The cumulative weight of indicators related to transformer oil chemical tests accounts for 30% of the physical characteristic dimension, and the indicators related to coil current of switchgear equipment account for 63% of the electrical performance dimension, which effectively fills the gaps in traditional detection indicators and provides data support and technical guarantee for the precise operation and maintenance of power