PREDICTION OF MECHANICAL PROPERTIES OF CAST TITANIUM BASED ON THE CHEMICAL ELEMENTS OF THE ALLOY USING A MACHINE LEARNING MODEL

Abstract

Cast titanium is widely used in advanced engineering applications due to its high strength-to-weight ratio and good corrosion resistance. However, its mechanical properties are highly sensitive to variations in the alloy's chemical composition, making conventional testing less efficient in terms of time and cost. This study aims to predict the mechanical properties of cast titanium, namely yield strength and tensile strength, based on the alloy's chemical composition using a machine learning approach. The dataset was obtained from the Materials Algorithms Project (MAP) with 100 cast titanium specimens developed through small and controlled variations in chemical composition to represent realistic manufacturing conditions. Modeling was performed using Decision Tree and Random Forest algorithms with data splitting schemes of 60:40, 70:30, and 80:20 and k-fold cross validation. Model performance evaluation was performed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²) metrics. The results of the Pearson correlation analysis showed that Al, V, and Fe-X elements have a strong positive correlation with mechanical properties, while Ti and O show a significant negative correlation, which is in line with the theory of titanium alloy metallurgy. The modeling results show that the Random Forest algorithm provides the best performance with lower prediction errors and better stability compared to Decision Tree. This study proves that the machine learning approach, especially the Random Forest algorithm, is effective in predicting the mechanical properties of cast titanium based on chemical composition, with the best performance shown by the RMSE value of 70.95 and MAE of 47.95, thus potentially supporting the design and optimization of cast titanium alloys based on data.