Transfer Learning Through Deep Learning: Application to Topology Optimization of Electric Motor
Jo Asanuma, Shuhei Doi, Hajime Igarashi
Abstract
This article proposes the use of transfer learning for the deep neural network to reduce the computing cost of the topology optimization of electric motors based on a genetic algorithm (GA). The average torque and torque ripple values are shown to be accurately inferred by the transfer learning with small learning data. The individuals on the Pareto front are only evaluated by the finite-element method, while others are fast evaluated only by convolutional neural networks (CNNs). The proposed method makes it possible to reduce the computing cost to less than 15% of the conventional topology optimization method.
Topics & Concepts
Computer scienceTopology (electrical circuits)Topology optimizationTransfer of learningConvolutional neural networkTorque rippleArtificial neural networkDeep learningNetwork topologyTorqueGenetic algorithmFinite element methodArtificial intelligenceMachine learningInduction motorDirect torque controlPhysicsMathematicsVoltageCombinatoricsThermodynamicsQuantum mechanicsOperating systemTopology Optimization in EngineeringAdvanced Multi-Objective Optimization AlgorithmsMetaheuristic Optimization Algorithms Research