This study presents a deep learning framework for rapid prediction of mechanical properties–elastic modulus E and Poisson’s ratio ν –and phase volume fraction in high-elasticity Al-Si-Ni casting alloys directly from X-ray diffraction (XRD) data. A total of 421 experimentally measured datasets were constructed, each comprising XRD pattern, chemical composition, casting cooling rate, and mechanical properties, while phase volume fraction was obtained from thermodynamic calculation. Chemical composition is used as an additional input for elastic modulus prediction, with cooling rate further included for Poisson’s ratio, whereas phase volume fraction is predicted solely from XRD data. To overcome limited amount of data, a variance-preserving denoising diffusion probabilistic models (DDPM)-based augmentation method expands the number of training data tenfold, improving prediction accuracy while maximally preserving peak information and mass fractions for critical phases. Convolutional ResNet-based regressors achieves high predictive performance, yielding mean absolute errors of 0.91 GPa for elastic modulus, 0.0050 for Poisson’s ratio, and 0.014 vol.% for phase fractions. A multi-task integrated model enables simultaneous prediction of both mechanical properties with competitive accuracy. Quantitative explainable AI analysis reveals key diffraction peaks and confirms that the model learns physically interpretable representations. • 421 datasets included composition, cooling rate, XRD, E , ν , and phase fractions. • DDPM forward diffusion augmented XRD tenfold, improving prediction R² by 5–20%. • ResNet models predicted E , ν , and phase volume fraction with high accuracies. • A multi-head regressor jointly predicted E and ν with good performance. • Explainable AI showed that predictions used physically meaningful features.
Oh et al. (Sun,) studied this question.