Low Error Rate Induction Machine Parameter Estimation with Recurrent Neural Network

Induction machines are widely preferred in plants due to their uncomplicated structure and low maintenance requirements. In order to achieve effective control over the operations of these machines, it is crucial to possess accurate information about their parameters. The estimation of these parameters can be accomplished through the utilization of artificial neural networks. Nevertheless, the majority of studies undertaken for parameter estimation were inadequate in accurately representing the network architecture's performance or achieving the desired precision. This was mostly due to the low amount of available data and the reliance on data from a single experimental setting. This study evaluates a recurrent neural network with a concise and flexible structure to address data insufficiency and the reliance on a singular experimental setting. This evaluation involves using a substantial dataset and optimizing the network parameters to achieve the most efficient network structure. Upon completion of the study, the proposed approach demonstrated promising results with high correlation levels and minimal error rates.

I. Introduction

Induction machines are highly favored for their cost-effective designs, making them the primary source of mechanical power in the industry [1]. It has the capability to be utilized in both industrial and domestic equipment by feeding directly from a single-phase AC network, a three-phase AC network, or a variable voltage/frequency converter [2]. There are types such as squirrel cage, in which aluminum rods are placed in the rotor openings to act as windings and terminated with rings at both ends [3]. Although fully reflecting the structure of these machines is a difficult task, the approximate behavior of the machine may be represented by equivalent circuits, and knowledge of these circuit characteristics is required to provide effective control and monitoring of the machine [4]. In this direction, the parameters can be established using traditional testing, such as blocked rotor and no load, conducted in accordance with technical requirements. In addition to this, several neural network and algorithm studies have been presented for parameter estimation in recent years.