A Neural Network Based Initial Position Detection Method To Permanent Magnet Synchronous Machines

Initial position is one of the key points in the permanent magnet synchronous machine (PMSM) senseless control. This paper presents a neural network method with currents difference as its inputs and the rotor position angle as its outputs. 256 different 100mus-width voltage vector pulse pairs injects into stator before position calculation. The voltage vector direction distributes symmetrically around an electrical cycle. So 256 current pairs can be obtained. Because of the saturation of the stator, the positive and negative vector pulses have different current responses. 256 current difference data can be got. As there´s much noise in the data a neural network is introduced. 256 current difference data are used as the inputs of the neural network, and the electrical angle the rotor position is the outputs. A direct torque controlled (DTC) PMSM experimental system with the proposed initial position detecting method is set up. The scheme is verified by the experimental system with a 600w PMSM. The simulation and experimental results show that the scheme presented in the paper is available to get the initial position with 5.6% error and able to start a PMSM successfully