부산대학교 도서관

The author present an artificial intelligent (AI)-based deep generative model that demonstrate how to generate design options of mechanical systems, which are not only suitable for specific working conditions but also optimized for engineering performance. In current study, (1) a structural generative residual netowork (SG-Resnet) model is developed to establish the non-linear mapping between the working conditions and the external dimensions of the reducer, the main hyperparameters influencing the prediction ability and learning rate of the SG-Resnet are analyzed. (2) The mixed population non dominated sorting genetic algorithm-II (MP-NSGA-II) is proposed, and used to obtain pareto optimal solutions of the internal dimensions of the reducer. Experiments are performed to validate the positive effect of the structural generative model on the stiffness of the reducer. This research provides a novel method for reducer design and lays a solid foundation for the development of sequential engineering software for integrated rotate vector (RV) reducer.