Data-driven XGBoost model for maximum stress prediction of additive manufactured lattice structures

Abstract

Lattice structures created using additive manufacturing technology inevitably produce inherent defects that seriously affect their mechanical properties. Predicting and analysing the effect of defects on the maximum stress is very important for improving the lattice structure design and process. This study mainly used the finite element method to calculate the lattice structure constitutive equation. The increase in defect type and quantity leads to difficulty in modelling and reduces calculation accuracy. We established a data-driven extreme gradient enhancement (XGBoost) with hyperparameter optimization to predict the maximum stress of the lattice structure in additive manufacturing. We used four types of defect characteristics that affect the mechanical properties—the number of layers, thick-dominated struts (oversize), thin-dominated struts (undersizing), and bend-dominated struts (waviness)—as the input parameters of the model. The hyperparameters of the basic XGBoost model were optimised according to the diversity of the inherent defect characteristics of the lattice structure, while the parameters selected by experience were replaced using the Gaussian process method in Bayesian optimization to improve the model’s generalisation ability. The prediction datasets included the type and number of defects obtained via computer tomography and the calculation results of the finite element model with the corresponding defects implanted. The root mean square error and R-squared error of the maximum stress prediction were 17.40 and 0.82, respectively, indicating the effectiveness of the model proposed in this paper. Furthermore, we discussed the influence of the four types of defects on the maximum stress, among which the thick strut defect had the greatest influence.