Newswise – A research paper led by researchers from the Southern Institute of Technology in China has published research papers Frontier in Mechanical Engineering2024, Volume 19, Issue 4. This study proposes a data-driven approach based on the accumulation of ensemble learning to predict the mechanical properties of Ti6Al4V alloys formed by large-scale laser powder bed fusion (LPBF).
When manufacturing TI6AL4V parts via LPBF, the process involves a number of interaction parameters, making it difficult to determine the appropriate parameters. Machine learning technology can address this problem, but a single model has a hard time capturing complex relationships. In this study, stacking models are applied that combine the advantages of multiple models to improve performance, predicting the mechanical properties of TI6AL4V alloys in the field of additives.
Tensile specimens were prepared in a specific LPBF instrument using spherical TI6AL4V powder. The range of parameters for laser power, scan speed, and hatch interval were set, and 64 parameter combinations were obtained through orthogonal array experiments. We constructed stacking models to predict tensile strength using algorithms such as ANN, ENET, KRR, GBR, and Lasso. We optimized the model using Bayesian optimization and cross-validation, and applied multiple metrics to evaluate its performance.
Pearson correlation coefficient analysis reveals that scan speed has the greatest effect on tensile strength, followed by laser power, with the hatch spacing having the lowest effect. After optimization, the optimal combination of the base model of the stacking model was determined. Training and testing results show that the stacking model exhibits higher prediction accuracy and stability than the ANN model, and better captures the complex relationship between process parameters and tensile strength.
The proposed stacking ensemble learning model provides an effective framework for predicting the tensile strength of Ti6Al4V alloys manufactured by large-scale LPBF. This study determines the extent of the impact of key parameters, examines the advantages of stacking models in multiple aspects, and provides a reliable method for predicting the mechanical properties of metal parts in the LPBF process.
Paper “Machine learning that enables predictions in the mechanical performance of Ti6Al4V manufactured by large-scale laser powder bed fusion via stacked models” Changjun Han, Hubao Yang, Dorin Yuan, Kai Li, Yang Giang Yang, Zeon Zang, Di Wang. Full Open Access Paper: https://doi.org/10.1007/S11465-024-0796-0.
