Sensorless Detection of Angular Displacement of a Stepping Motor by Neural Network.

Abstract

A stepping motor is frequently used as an actuator of position control mainly in a low-cost system. In such case, the motor is usually driven by open loop control. For requirement of more accurate performance in positioning, closed loop control may be adopted. However, a position sensor is needed for feedback and the advantage of simple and low-cost construction is reduced by existence of the sensor. To realize a closed loop system without spoilage of this advantage, a technique for sensorless detection of position must be employed.The authors developed a new method for sensorless detection of angular displacement of a stepping motor. In this method, relationship between waveforms of angular displacement of the rotor and the current of the stator winding in a step response is previously learned by feedforward neural network, and a waveform of angular displacement is estimated at each step by the neural network using the information of the electrical current data only.Applying this method to an actual system by using the neural network which has learned previously using the step response data with various inertial loads, waveforms of angular displacement are estimated sufficiently in the cases driving condition is not different from that of the learned data. However, in the case with different condition, accuracy of estimation is not kept. This shows that the data for learning must be obtained in many cases of different step responses with a wide range of driving condition.