Auto-Associative Neural Network Based Sensor Drift Compensation in Indirect Vector Controlled Drive System

Abstract

The paper proposes an auto-associative neural network (AANN) based sensor drift compensation in an indirect vector-controlled induction motor drive. The feedback signals from the phase current sensors are given as the AANN input. The AANN then performs the auto-associative mapping of these signals so that its output is an estimate of the sensed signals. Since the AANN exploits the physical and analytical redundancy, whenever a sensor starts to drift, the drift is compensated at the output, and the performance of the drive system is barely affected. The paper describes the drive system, gives a brief overview of the AANN, presents the technical approach, and then gives some performance of the system demonstrating validity of the approach. Although current sensors are considered only in the paper, the same approach can be applied to voltage, speed, torque, flux, or any other type sensor.