Data-driven multiphysical microstructure engineering of polymer composites using convolutional vision transformers and conditional generative adversarial inverse design frameworks

Design of polymer composite microstructures with optimal mechanical, thermal, and electrical properties remains a critical challenge in materials science due to the complexity of their multiphysical interactions. Conventional physics-based simulations and empirical models are computationally expensive and limited in scalability and interpretability, while recent AI-based approaches often focus on single-property optimization without enabling inverse design. To overcome these limitations, this study proposes a data-driven dual-framework combining a Convolutional Vision Transformer (CvT) for forward multiphysical property prediction and a Conditional Generative Adversarial Network (cGAN) for inverse microstructure generation guided by property-consistency feedback. The framework is implemented using Python with TensorFlow and PyTorch, trained on an extensive dataset comprising 18,000 real and 42,000 synthetic polymer composite samples, with mechanical, thermal, and electrical labels obtained through FEM-based simulations. The CvT accurately captures spatial–property relationships, achieving R² = 0.973 and RMSE = 0.058, while the cGAN efficiently generates realistic microstructures with SSIM = 0.982, ensuring high morphological fidelity. Compared with existing CNN- and GAN-based models, the proposed approach demonstrates an accuracy improvement of 4.2% and reduces design convergence time by 40%, offering both computational efficiency and physical interpretability. The proposed framework, though modular, establishes a novel direction for real-time, property-driven polymer composite design, creating an intelligent platform that bridges predictive learning and generative synthesis. This study provides a transformative step toward accelerating material discovery and adaptive manufacturing, inspiring reviewers and researchers to explore the integration of advanced AI paradigms in practical composite engineering.