The Impact of Using Mostly AI Synthetic Data in Improving Predictions of the Mechanical Properties of Low Alloy Steel

Machine learning methods enable industry professionals in the metallurgy sector to rapidly, accurately, cost-effectively, and environmentally responsibly design the mechanical properties of low-alloy steel to meet specific application needs. However, the success of these methods highly relies on data quality and a substantial amount of data. In this study, we evaluate the use of synthetic data in predicting the mechanical properties of low-alloy steel using six machine learning algorithms. Synthetic data is generated using the Mostly AI synthetic data platform, which includes chemical composition, temperature, and the mechanical properties of low-alloy steel. Each model is trained with three different training dataset combinations: (1) using synthetic data, (2) using only experimental data, and (3) using a combined dataset that integrates synthetic and experimental data. Each model is evaluated with three evaluation metrics: MAE, RMSE, and R-squared. The research findings demonstrate that employing a combined dataset with the Decision Tree (DT) algorithm yields significantly better performance compared to models using only experimental or synthetic data. DT achieves an MAE of 8.87, RMSE of 20.38, and an R-squared of 0.98 for YS prediction. For TS, DT attains an MAE of 9.06, RMSE of 20.96, and an R-squared of 0.97. These findings suggest that utilizing synthetic data in modeling the prediction of low-alloy steel’s mechanical properties through machine learning methods can substantially enhance model performance.