Data-driven shape-topology optimization method for curved shells

Due to the extreme lightweight requirements of surface components in compact design space, this paper proposed a data-driven shape-topology optimization method of curved shells, which consists of three stages, namely the offline stage, the online stage and the update stage. First, in the offline stage, the Latin hypercube sampling method is used to extract the sample points from the design space, and the mesh deformation technique is used for modeling to obtain the mesh model corresponding to the sample points. Then, topology optimization was carried out on the mesh models to obtain the optimized strain energy. Based on the sample data obtained in the above steps, the radial basis function surrogate model is trained, where the shape design variable is the input and the strain energy after topology optimization is the output. In the online stage, optimization is carried out based on the surrogate model obtained in the offline stage, and the covariance matrix adaptive evolution strategy is adopted to improve the optimization efficiency. In the update stage, the real response of optimization results of the surrogate model is calculated and added to the sample dataset to update the surrogate model. Finally, the algorithm is verified by a simply supported beam and a spacecraft cabin door. The results show that compared with the topology optimization with the fixed shape, the strain energy obtained by the proposed method can be reduced by 20.08% and 37.93%, respectively, indicating that the proposed method has better design capability.