Energy absorption prediction for lattice structure based on D2 shape distribution and machine learning

Although lattice structure (LS) has the advantages of light weight, energy absorption, and high specific strength, exhibiting different mechanical properties with different structural forms. Thus, the vast design space gives a great challenge for properties prediction of LS. In this research, 210 unit cells with different shapes were designed and their D2 vectors describing the shape were extracted. A deep neural network method based on D2 distribution was employed to predict energy absorption effects. Moreover, to validate the transferability of the method, three new unit cells were designed and fabricated by additive manufacturing for experiments. The results show that the proposed method can well predict the energy absorption effect with similar to 13 % error and the performance rank even of novel unit cells beyond the dataset. A good correlation between experimental values and predictions demonstrates the effectiveness of the method. In addition, through investigation of size effect for lattice structure, it is found that the energy absorption effect has a slow increase with the size factor, and their performance rank does not vary with the change of the size factor. This study could contribute to accelerating the design process of LS for specific applications.