In the context of digital transformation, it is essential to ensure the safe operation of electrical equipment. In order to solve the problem of low accuracy of existing electrical equipment fault detection algorithms in diagnosing unknown faults, this study collects industrial field data to construct a dataset, and develops a fault identification model integrating convolutional neural network and long short-term memory network based on deep learning framework. Experiments show that the model has an average accuracy of 98.5% in the detection of five main fault types, which is nearly 10% higher than that of the traditional method, and the recognition rate of subtle faults is over 96%, with good generalization and robustness. The study also analyzes the impact of noise and optimizes the hyperparameters, which is expected to promote the upgrade of intelligent operation and maintenance in the manufacturing industry.