Identifying process windows of aluminum alloys in laser powder bed fusion using machine learning

The metal additive manufacturing (AM) sector is limited by the scarcity of certified materials. Developing new materials is costly and time-consuming due to the complexity of AM processes where numerous parameters impact the part’s quality. The goal of this study is using simulation and empirical data of AlSi10Mg and an alloy from the Al5000 series (AlMgty80) to train machine learning (ML) models to predict the part density in the laser powder bed fusion of metals. The models are used to predict process windows for a new alloy from the Al7000 series (AlZnty) to identify the potential of recognizing process windows prior to empirical testing. By integrating simulated data, material influences are indirectly accounted for in the dataset. Various ML algorithms were trained and evaluated using root mean squared error (RMSE), R2, and qualitative process map assessments. The results show that ML models based on AlSi10Mg and AlMgty80 data accurately predict the density for the respective materials, achieving an RMSE of around 1% and R2 close to 0.9. Predictions for AlZnty showed larger errors (RMSE 3%–7%, R2 near zero), yet the qualitative evaluation of process maps shows that high-density regions align with experimental data due to the model systematically underestimating the density. Overall, the methodology demonstrates the ability to identify process windows for novel alloys using simulations and ML models based on similar alloys, despite limitations in density prediction accuracy. Future work will involve training models including AlZnty data to iteratively enhance accuracy. Furthermore, limited data availability continues to challenge ML in AM, highlighting the need for extensive databases.