Motor system is the most basic and critical part of rotating machinery. Different transmission systems have different requirements in processing and manufacturing accuracy, installation and debugging, working environment and equipment maintenance. The transmission shaft system may also have the characteristics of variable speed, variable load and lasting work. Therefore, in all kinds of mechanical transmission systems, the motor is very prone to failure. In actual production, the probability of misalignment fault is as high as more than 60%, which is a high incidence of motor system fault. Therefore, it is of great significance to diagnose and troubleshoot the misalignment fault of motor system in time. In this paper, a method of diagnosing the parallel misalignment fault of servo motor based on Long Short Term Memory (LSTM) model is proposed. The speed data of healthy and faulty motors are used as the input signals of the network model to achieve the goal of predicting and diagnosing the fault signal. A series of experimental results verify the feasibility of the proposed method.