Artificial Neural Network (ANN) Based Meta-Model Design Optimization of Direct-Drive Wheel Hub Motor for an Electric Scooter

Abstract

Direct-drive electric motor or In-Wheel Hub Motor (WHM) is gradually becoming popular as a new solution for electric propulsion. By meas of direct drive, the conventional transmission of belt or chain sprocket is not required to deliver the high torque requirement at the wheel. This paper studied metamodel machine leasning (MOP-ANN) implementation to predict the relationship model of geometries input of 42 slot/40 pole WHM with corresponding output performance. The multi-layer perceptron (MLP) of the neural network structure was trained to reveal the underlying pattern within the data sample. The sample data was obtained from a sensitivity analysis drive from FEA. The optimum geometry of the motor was directly solved by implementing a multiobjective optimization (MOO) method of genetic algorithm. Finnaly, the optimum geometry of the MOP-ANN based result is validated by FEA.