A comparative study on parameters estimation of squirrel cage induction motors using neural networks with unmemorized training

Abstract

Induction machines are often preferred in industrial applications at present. Therefore, it is an important problem to know the electrical parameters of induction machines correctly. Electrical parameters of induction motors can be obtained experimentally by performing DC test, no-load rotor test, and locked rotor test. Furthermore, equivalent circuit parameters of the induction machines can be estimated with high accuracy using the data from machine manufacturers. In this study, equivalent circuit parameters of squirrel-cage induction motors have been successfully estimated by using Feed Forward Neural Network (FFNN) for single-cage and double-cage models. Although there is a feed forward neural network study in the literature, the training process of the neural network structures was carried out with unmemorized training. In addition to FFNN, an unmemorized Elman Neural Network (ENN) structure has been proposed to solve this problem. The proposed methods were compared with the literature for both models, and their performances were examined. The obtained FFNN results suggest that the proposed method performed better results than both adaptive neuro fuzzy inference system (ANFIS) and artificial neural network (ANN) for the single-cage model. In the double-cage model, FFNN performed better than ANN but relatively weaker than ANFIS. The results of the ENN structure are close to ANFIS for both single-cage and double-cage models.

Keywords

• Induction motor
• Feed forward neural network
• Elman recurrent neural network
• Equivalent circuit
• Single-cage model
• Double-cage model