Information on mechanical conditions is contained in the vibration signals generated from the actuator control system of a high-voltage vacuum circuit breaker. To overcome the limitation on feature representation in the traditional fault detection methods, this paper proposes a hybrid network model combining a convolutional neural network and a long-short term memory network. The convolution layer and gate recurrent unit in the network are respectively used for the feature conversion and the extraction of local time-domain characteristics of the vibration signals. The feature extracted by the network layers is sensitive to the fault. Experiments on a 10 kV high voltage vacuum circuit breaker show that the proposed method is able to detect various types of mechanical faults. The ROC curve and PR curve show that the diagnostic effect of the proposed model is better than a SVM model. This end-to-end fault diagnosis strategy improves the diagnostic accuracy of the mechanical state of a high voltage circuit breaker by deepening feature mapping. This work is supported by the National Natural Science Foundation of China (No. 61703191), the Foundation of Liaoning Educational Committee (No. L2017LQN028) and the Scientific Research Foundation of Liaoning Shihua University (No. 2017XJJ-012).