Damage-Coupled Physics-Informed Neural Networks for Predicting Long-Term Creep Strain Evolution in Lightweight Aerospace Alloys

Lightweight alloys in aerospace precision structures undergo slow but cumulative creep deformation during long-term storage, wherein strain accumulation over years can compromise dimensional stability and operational reliability. However, continuum damage mechanics (CDM) constitutive models, while physically grounded, require extensive parameter calibration and exhibit degraded accuracy during the primary creep stage. Meanwhile, purely data-driven approaches are impractical for the sparse datasets typical of accelerated creep testing, wherein as few as 14 data points may be available per condition. Although physics-informed neural networks (PINNs) have shown promise in computational mechanics, existing PINN-based creep studies predict only scalar life quantities rather than the full strain–time curve (Formula presented.), and none embed damage evolution equations as differential constraints. This study proposes a damage-coupled PINN framework (termed DC-PINN) that predicts the complete creep strain evolution (Formula presented.) by embedding CDM damage evolution ordinary differential equations (ODEs) as hierarchical differential constraints within the learning process. The framework couples the predicted strain rate (Formula presented.) with the damage state (Formula presented.) through material-specific constitutive ODEs, supplemented by monotonicity enforcement and boundary conditions. Alloy-specific formulations are developed for 2A12-T4 aluminum (Arrhenius kinetics, no damage) and ZM6 magnesium (Sandström dislocation model with Ostwald-ripening-driven grain coarsening damage). Validated on 13 experimental conditions spanning both alloys (50–100 °C, 20–60 MPa, 14–100 points per condition), DC-PINN achieves (Formula presented.) for 2A12-T4 and (Formula presented.) for ZM6 across all tested conditions. Ablation studies show that the total physics-driven (Formula presented.) improvement is 5.8 times larger for the data-sparse ZM6 (14–34 points) than for the data-rich 2A12-T4 (∼100 points), with the CDM damage coupling alone accounting for 22% of the improvement in ZM6. To the best of our knowledge, this represents the first integration of CDM damage evolution ODEs as differential constraints within PINNs for creep strain modeling, providing a physically consistent and data-efficient tool for the storage life assessment of aerospace structures.