A Data-Efficient Physics-Informed Neural Network Framework for Reliable Irradiation Hardening Assessment in Nanocrystalline materials

This paper proposes a novel Physics-Informed Neural Network (PINN) framework for assessing ion irradiation hardening in nanocrystalline materials. A key innovation lies in the deep integration of irradiation hardening mechanisms into the neural network via a composite loss function, which imposes​ physical laws as hard constraints. During network training, key parameters of the physical model are simultaneously optimized, enabling adaptive matching to the microstructural characteristics of different materials. This integration ensures​ that the model’s predictions are​ both consistent with experimental data and physically plausible. Compared with conventional methods, this framework significantly reduces data dependence, overcomes the limitations of pure physical models under ill-posed conditions, and addresses the poor generalizability of purely data-driven approaches. Validation through multi-condition irradiation experiments on Ni-Mo-Cr alloys with two different grain sizes demonstrates that the model achieves a prediction error of less than 5% for the nanocrystalline alloy and maintains an error of approximately 10% for conventional alloy, with robust stability in extrapolation tests. Its evaluation accuracy and robustness are significantly superior to those of traditional physical models and purely data-driven methods. This study provides a physically interpretable, data-efficient, and highly generalizable paradigm for the assessment of irradiation hardening.