Abstract: At present, most of the research objects for fault diagnosis of switch machines are based on current and power curves. Aiming at the problem of low efficiency of manual fault diagnosis of switch machine, a fault diagnosis method based on traction force of switch machine and 1DCNN-LSTM is proposed. The traction force of the ZDJ9 switch machine is taken as the research object, and the ability of 1DCNN and LSTM to extract data space and time is integrated. The feature extraction and fault classification are combined, and the optimal value of the hyper-parameters of the model is found through multiple experiments. The confusion matrix, accuracy curve and loss function curve are given to show the accuracy of the model for fault diagnosis. The T-sne visualization is used to reflect the effectiveness of the feature extraction of the model. The experimental results show that the 1DCNN-LSTM model can effectively extract the spatio-temporal information of the original signal, and achieve end-to-end fault diagnosis with a simple structure, which meets the application requirements of the on-site maintenance switch machine.