Energy absorption properties of a novel auxetic honeycomb based on deep learning technology

In this paper, a new type of auxetic honeycomb is designed, which introduces arc walls into the concave hexagonal honeycomb cells and has higher specific energy absorption. The deformation modes and energy absorption of the designed honeycomb are analyzed by using three methods including finite element, compression experiment and machine learning methods. A fully connected two-hidden layer backpropagation deep neural network (BP-DNN) is developed to predict the energy absorptions of the honeycomb with different geometric parameters. It is found that the error of the validation set is low, and the average correlation coefficient of the validation set is 99.2%, which indicates that the neural network can obtain good predictions. The sensitivity analysis of the input parameters shows that the thickness t has the highest sensitivity, and the ratio of the length of the straight wall to the height has the lowest sensitivity to the energy. In addition, the neural network developed can also predict the mechanical properties of the honeycombs outside the parameter range of the training set and the results are consistent with that of the sensitivity analysis. The DNN provides a fast and accurate method for the energy absorption of honeycombs, which is expected to speed up the optimization and design process of honeycomb structures.