Research Information System
International Journal of Computational Materials Science and Engineering, 2025 (ESCI)
Prediction models are commonly trained using datasets derived from open-source material data or detailed experimental studies to estimate material properties. However, preparing such datasets often requires time-consuming and labor-intensive processes. The CALPHAD (CALculation of PHAse Diagrams) methodology, on the other hand, enables the simulation of the physical, mechanical, and thermodynamic properties of materials for specific alloy compositions through phase transformation calculations and thermodynamic data sources. In this study, data obtained from aluminum alloys simulated using the CALPHAD-based JMatPro software were utilized as a dataset source to train a prediction model developed with Explainable Artificial Intelligence and Artificial Neural Networks. Accordingly, the thermo-physical properties (density, thermal conductivity, linear expansion, average expansion coefficient, Young's modulus, bulk modulus, shear modulus, and Poisson's ratio) of 225 different 6xxx series (Al-Mg-Si) aluminum alloys in the temperature range 25-155∘C were simulated, generating a dataset of 6075 rows. Of this dataset, 80% was used for model training, while the remaining 20% was used for validation and testing. The analyses revealed that the proposed model successfully predicted compositions with the desired thermo-physical properties at specific temperatures. The predicted composition was simulated under the same conditions as those used to generate the dataset in JMatPro software, and the obtained values showed a 98.57% accuracy rate compared to the target values. In conclusion, the dataset generated from aluminum alloys simulated using the CALPHAD methodology provides a reliable training source for AI-based material analysis. The proposed approach offers significant advantages in terms of speed and accuracy for material design and development processes.