Machine learning predicts 3D printed Ti6Al4V performance

Researchers from South China University of Technology have recently published a paper in Frontiers of Mechanical Engineering, proposing a data-driven approach based on stacking ensemble learning to predict the mechanical properties of Ti6Al4V alloy formed by large-scale laser powder bed fusion (LPBF).

When manufacturing Ti6Al4V parts via LPBF, the process involves numerous interacting parameters, making it challenging to determine appropriate parameters. Machine learning techniques can address this issue, but single models struggle to capture complex relationships. Stacking models, which combine the advantages of multiple models to enhance performance, are applied in this study to predict the mechanical properties of Ti6Al4V alloy in the field of AM.

Spherical Ti6Al4V powder was used to prepare tensile specimens on specific LPBF equipment. The parameter ranges for laser power, scanning speed, and hatch spacing were set, and 64 parameter combinations were obtained through orthogonal array experiments. Algorithms such as ANN, ENet, KRR, GBR, and Lasso were employed to construct a stacking model for predicting tensile strength. Bayesian optimization and cross-validation were used to optimize the model, and multiple metrics were applied to evaluate its performance.

Pearson correlation coefficient analysis revealed that scanning speed has the greatest impact on tensile strength, followed by laser power, with hatch spacing having the smallest effect. After optimization, the optimal combination of base models for the stacking model was determined. Training and testing results show that the stacking model exhibits higher prediction accuracy and stronger stability than the ANN model, better capturing the complex relationships between process parameters and tensile strength.

The proposed stacking ensemble learning model from South China University of Technology provides an effective framework for predicting the tensile strength of Ti6Al4V alloy manufactured by large-scale LPBF. This study determines the degree of influence of key parameters, validates the advantages of the stacking model in multiple aspects, and offers a reliable method for predicting the mechanical properties of metal parts in the LPBF process.