Engineered Cementitious Composites (ECC) has exceptional ductility and strain-hardening behavior, yet its inverse design remains a significant challenge due to the inherent “one-to-many” ambiguity. This study proposes a physics-informed generative framework, AdaIN-WGAN, for the performance-driven and multi-objective optimization of ECC. The model interprets material performance targets as “styles” to be modulated onto the mixture design space. A pre-trained Transformer-based predictor is incorporated as a physical consistency constraint. Results demonstrate that the model effectively decouples fundamental mixture logic from performance-specific characteristics. The AdaIN-WGAN achieves a 32% to 67% reduction in generator loss compared to traditional embedding and transfer learning models. Experimental validation of 96 dog-bone specimens confirmed the proposed framework generates ECC mixtures with material properties that closely align with design targets. The generative covariance is lowered to one-fourth of baseline levels, ensuring high confidence in structural reliability analysis. Finally, a cradle-to-gate Life Cycle Assessment (LCA) shows that the framework can facilitate significant decarbonization of ECC, reducing the carbon emissions by up to 59% for low-ductility ECC and 31% for high-ductility ECC.
X et al. (Fri,) studied this question.