Evaluation of the Auto-Associative Neural Network Based Sensor Compensation in Drive Systems

The paper performs a deep analysis of the sensor drift compensation in motor drives approach presented in past publications [11-12]. In the past, the auto-associative neural networks (AANN) were found to be effective for this application. However, it is still unclear how much improvement may be obtained compared with other modeling techniques and when it is adequate to be applied. Therefore, the modeling techniques, specially the AANN, are detailed and evaluated using performance metrics. Additional experimental results in a motor drive are provided to show the compensation capability of the AANN. The feedback signals are given as the AANN inputs. The AANN then performs the auto-associative mapping of these signals so that its outputs are estimations of the sensed signals. Since the AANN exploits the physical and analytical redundancy, whenever a sensor starts to drift, the drift is compensated, and the performance of the drive system is barely affected.