Physics-informed graph transformer for surface hardness prediction under data-limited conditions

Hardness is a critical surface integrity metric governing the performance of machined components. Due to experimental constraints, hardness characterization typically relies on sparse measurements, limiting spatial resolution and generalization across machining conditions. This work addresses this limitation by introducing a physics-informed graph transformer for hardness prediction in milling, using process parameters and in-situ power signals. To mitigate data scarcity, power signals are synthesized, and a graph representation is constructed to embed machining physics and measurement relationships. Data analysis demonstrates robust and generalizable prediction under limited reference data, establishing a soft sensing framework for scalable characterization of hardness from milling operations.