Electromechanical actuators (EMA) are becoming more and more widely used. As the core technology of EMA, servo control technology determines their performance. In this paper, an active disturbance rejection control (ADRC) method with an improved extended state observer (ESO) is proposed to design a cascade controller of EMA based on permanent magnet synchronous motor (PMSM). The mathematical model of PMSM in a two-phase rotating coordinate system is established, then it is decoupled by an $i_d=0$ current control method to realize the vector control of the motor. In a three closed-loop vector control system, a PID controller including current loop, speed loop and position loop is designed. To solve the problems caused by measurement noise, the filter link and system are modeled as a whole, and an improved ESO is constructed. On this basis, an ADRC controller of the speed loop and position loop of PMSM is designed and simulated based on Simulink. Based on the physical test platform, a load step test and load disturbance test of ADRC are completed. The results show that, in comparison to the PID method, the ADRC method shortens the response time by 25% on average, and reduces the overshoot by 60% on average. So, it can be concluded that ADRC has good static and dynamic performance, which has a good guiding role for engineering practice.

In this paper, an inter-turn short-circuit fault of a permanent magnet synchronous motor in an electromechanical actuator is analyzed, and a fault diagnosis method based on transfer learning with a VGG16 convolution network is proposed. First, a 2D finite element model of an inter-turn short circuit fault of a permanent magnet synchronous motor was established in ANSOFT Maxwell, and then a simulation experiment analysis was completed. A three-phase current was chosen as a fault characteristic signal. Second, a fault diagnosis method with a VGG16 deep convolutional neural network and based on transfer learning was designed, and the fine tuning of the hyperparameters of the fault diagnosis model was completed by using grid search and cross verification methods. Finally, based on the transfer learning VGG16 model established in this paper, the inter-turn short circuit fault of a permanent magnetic synchronous machine (PMSM) was diagnosed and verified. The experimental results showed that the proposed convolutional network model based on transfer learning can identify faults effectively and accurately, and has a good engineering guidance significance.