Genetic sampling for surrogate-assisted topology optimization in electromagnetic riveting device design

Abstract

Electromagnetic riveting (EMR) has been widely used in aerospace field, but the riveting gun is heavy and its cost is high. In this paper, the reduction of cost and weight for rivet gun were investigated from the optimization of concentric-ring slave coil. For the lack of accuracy of the traditional genetic algorithm coupled with artificial neuron network (GA-ANN), an improved computational framework was proposed. In the framework, a genetic-drop sampling strategy enhanced the accuracy of GA-ANN and helped ANN to generalize to the optimal. The prototype was manufactured in a case study to verify the riveting quality of the optimized EMR gun. Besides, coils of different sizes proved the generalization ability. The proposed method has the potential to significantly reduce the prediction error by 62% and effectively decrease the number of numerical model calculations by 73%. Optimization for EMR gun led to a 51% reduction in weight, a 73% increase in material utilization and little effect on riveting quality. Meanwhile, the algorithm was still accurate on coils of different sizes. The proposed computational framework could improve efficiency in the design of EMR gun in industry.