Abstract To study the topology optimization of carbon fiber composite helmets, a method of optimizing the three-dimensional (hereinafter referred to as 3D) model based on a deep learning network after dimensionality reduction is proposed. In this method, the 3D shell model is expanded and tiled into two-dimensional (hereinafter referred to as 2D) grids based on the one-step inverse forming method. The variable density method (Isotropic Material Penalty Density method) is used to collect the data set required for deep learning training, and the U-Net neural network is built to complete the training. To obtain the input tensor of the neural network, the mechanical analysis of the 3D model is carried out, and the physical properties are imported into the network for learning. In this way, the 2D optimization results of the helmet model are obtained and then reflected in the 3D model according to the unit topology relationship to obtain the final optimization results. Then, the ordinary helmet and the optimized helmet are simulated and compared under the same quality to test their mechanical properties. The results show that the stiffness performance and energy absorption effect of the optimized helmet are stronger when the quality of the two helmets is the same.
Read full abstract