Partial Inverse Design of High-Performance Concrete Using Cooperative Neural Networks for Constraint-Aware Mix Generation

The researchers proposed a framework based on collaborative neural networks for the partial inverse design of high-performance concrete (HPC). This framework integrates a completion model and a proxy strength predictor, enabling the generation of effective and performance-consistent hybrid designs through single forward propagation without the need for re-training for different constraint scenarios. Compared with benchmark models such as autoencoders and Bayesian inference, this method improved the R-squared value between the predicted and actual strengths of the mixed materials to 0.84 to 0.89, and reduced the strength consistency mean square error by approximately 42% and 60%, respectively. The research results confirmed the novelty, accuracy, and computational efficiency of artificial intelligence in the generation of high-performance concrete mixtures.