Optimal induction machine parameter estimation method with artificial neural networks

İpek S. N., Taşkıran M., Bekiroğlu K. N., Ayçiçek E.

Electrical Engineering, vol.106, no.2, pp.1959-1975, 2024 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 106 Issue: 2
  • Publication Date: 2024
  • Doi Number: 10.1007/s00202-023-02049-1
  • Journal Name: Electrical Engineering
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Compendex, INSPEC, DIALNET
  • Page Numbers: pp.1959-1975
  • Keywords: Grid search, Induction machine, Machine learning, Optimization, Parameter estimation
  • Yıldız Technical University Affiliated: Yes


Induction machines are widely utilized in the industry due to their sturdy construction nature and relatively simple maintenance. For the design, control, optimization, and failure analysis procedures of induction machines, equivalent circuit parameters are required. These parameters can be estimated using a variety of techniques, such as traditional testing, experimental research, analytical methods, programming, and machine learning algorithms. However, the aforementioned methods have limitations; experimental methods are laborious and time-consuming, analytical methods require complicated and iterative computations for each machine, and programming algorithms involve writing of lengthy command sequences. The article concentrates on a high precision estimation method based on neural network algorithms to avoid complex calculations, thus, minimizing time and effort, and eliminating errors. In this context, it is aimed to provide a greater reliability compared to studies using small amounts of datasets and to investigate earlier methods, which are not studied vastly, for estimating the parameters of induction machines. In this regard, three distinct artificial neural networks were applied to a large dataset consisting of 1164 machines with power output ranging from 4 to 900 kW and belonging to 7 different manufacturers. RMSE values of 0.0125 were attained in parameter estimation, and artificial neural networks produced encouraging results.