DOHO: A Data-Oriented Hybrid Optimization Technique for Squirrel Cage Induction Motors

Ipek S. N., Bekiroglu N., Taşkıran M.

Arabian Journal for Science and Engineering, cilt.51, sa.17, ss.1-33, 2026 (Scopus)

Özet

Electric machine design optimization is challenging due to conflicting objectives and complex, nonlinear relationships among parameters. Conventional optimization methods suffer from poor generalization capability due to limited data availability or incur high computational costs. This study presents a hybrid methodology that integrates Artificial Neural Network and Snake Optimization Algorithm (SO) for the design optimization of low-voltage squirrel cage induction motors. In the first stage, a Recurrent Neural Network is trained on 1164 motor dataset, whose 11 input features are sourced from manufacturer catalogs and whose 28 mechanical output features are derived through analytical computations, spanning a power range from 4 kW to 900 kW to predict the initial design parameters. In the second stage, these predictions serve as initial vectors for the SO Algorithm, which are optimized toward efficiency objectives through the adaptive constraint strategy. This hybrid framework does not require experimental procedures or high computational expenditure. It offers high generalizability, owing to its ability to learn from a diverse dataset. The hybrid pattern, which couples a deep learning predictor with a metaheuristic refinement stage under engineering feasibility constraints, can recur across domains where labeled data are sparse and physical models are available. This methodology combines the rapid prediction capability of machine learning with the global search ability of metaheuristic optimization, enabling fast convergence to feasible optimal design parameters. In scenarios where access to critical design data is limited, this approach provides a reliable starting point for processes such as drive operations and fault diagnosis, thereby contributing to industrial applications.