This thesis deals with current control and system identification of a permanent magnet synchronous machine. To learn a data-driven controller for current control, a model of the motor is required. Such model can be constructed with the utilization of neural ordinary differential equations. Through this approach, expert knowledge can be used in combination with learnable parameters to identify the motor parameters. With the help of this model, a neural network can then be trained to control the current. The proposed approach will be compared with both model-based and model-free methods for current control.