Co-simulation framework for estimating the rotor bar currents of a cage induction motor using FEA and ANN

• Published in: 2022 International Conference on Electrical, Computer and Energy Technologies (ICECET) pp. 1 - 6
• Main Authors: Barukcic, Marinko, Varga, Toni, Stil, Vedrana Jerkovic, Bensic, Tin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.07.2022
• Subjects: artificial neural network; Artificial neural networks; cage induction motor; co-simulation; Computational modeling; Estimation; finite element analysis; Induction motors; rotor bar current; Rotors; Steady-state; Training
• DOI: 10.1109/ICECET55527.2022.9872604

The paper presents a research work on the estimation of rotor bar currents of a squirrel-cage induction motor (IM). The main objective of the research conducted is to investigate whether it is possible to estimate the values of IM rotor bar current with artificial neural network (ANN) with satisfactory accuracy. Another objective of the study is to investigate the generality of such bar current estimation for different operating conditions of the motor. For this purpose, different designs of ANN are also investigated. The method is based on the application of a finite element analysis simulation tool to determine rotor current values under transient and steady state conditions. The ANN based estimation method uses the standard measurable data of stator current and rotor speed. In the next step of the proposed method, the calculated rotor current values are used to train an artificial neural network. Based on this approach, the presented method represents a data-based estimation model. After the ANN is trained, ANN is tested on motor transients that are different from those used in learning the artificial neural network. Data from a real motor is used for the study. The three different ANN designs are examined in the study. The values of the loss function (mean square error, used in the ANN training process) are (for normalized data) 0.0013, 0.0013, and 0.0014 (during ANN training) and 0.0038, 0.0035 (ANN prediction for new input data) for the proposed designs ANN 1, ANN 2, and ANN 3.