Abstract: The vibration generated by explosion-proof electrical equipment in coal mines during operation can compromise its mechanical integrity, leading to fastener loosening, component wear, and changes in the structure and vibration modes of the equipment. This can cause complex changes in signal features, resulting in confusion between normal vibration frequency and new frequency components induced by faults. As a result, the boundary between normal and fault signals becomes unclear, reducing the accuracy of traditional fault detection methods. To address this issue, an automatic vibration fault detection method for coal mine explosion-proof electrical equipment was proposed based on One-Class Support Vector Machine (OCSVM). First, the normal state features and vibration fault state features of the equipment were constructed. Based on the characteristics of OCSVM, the normal state feature sequence was set as the learning target for the decision boundary of the OCSVM kernel function. Due to the nonlinear and high-dimensional characteristics of vibration faults in explosion-proof electrical equipment, a polynomial kernel was selected as the OCSVM kernel function after comprehensive consideration. Then, grid search combined with K-fold cross-validation was used to optimize the parameters of the OCSVM, ensuring better performance. Finally, by obtaining the optimal solution of the OCSVM objective function, the optimal decision boundary was determined to realize automatic fault detection of vibration faults in coal mine explosion-proof electrical equipment. Experimental results showed that: ① When the number of iterations is 20, the OCSVM algorithm can complete convergence and achieve stability. ② In the electrical equipment signal classification experiment based on OCSVM, the use of the polynomial kernel function accurately classified samples for detection. ③ In the performance analysis of automatic vibration fault detection, the proposed method showed significantly higher accuracy across different sample sizes than infrared thermography and detection methods based on grey wolf optimization and support vector machine. Under small sample sizes, it achieved an accuracy of 98.25% with good stability.