Opening the Black Box: An Explainable, Few-shot AI4E Framework Informed by Physics and Expert Knowledge for Materials Engineering

Physics-informed AI improves both accuracy and trustworthiness: By embedding physical laws and expert knowledge directly into the AI framework, the researchers achieved 88% accuracy in predicting hot-cracking tendency in aircraft superalloy welding while producing interpretable equations that engineers can understand and verify, rather than opaque black-box predictions.

AI can work effectively with limited data when designed thoughtfully: The framework successfully operated with only 32 experimental samples by using physics-based synthetic data augmentation—a critical finding for safety-sensitive industries like aerospace where collecting large datasets is expensive and time-consuming.

Explainability enables practical engineering insights: The resulting interpretable equations revealed how thermal, geometric, and metallurgical factors couple together to cause cracking, advancing engineers' understanding of the welding process beyond simple predictions—knowledge that can inform process optimization and design decisions.

Responsible AI for high-stakes applications requires multiple safeguards: The framework incorporated hard physical constraints, calibrated noise injection based on real process variabilities, and preservation of parameter relationships—demonstrating that trustworthy AI in safety-critical sectors like aerospace requires deliberate architectural choices, not just accurate predictions.

Interpretable models serve multiple purposes simultaneously: The constitutive equation functions as both a predictive tool and an optimization instrument, showing that explainable AI can be practically useful rather than merely a compliance requirement.