AI-infused characteristics prediction and multi-objective design of ultra-high performance concrete (UHPC): From pore structures to macro-performance

Performance prediction and mixture design are important for the engineering applications of ultra-high performance concrete (UHPC). This study introduces an approach for both predicting and optimizing the design of UHPC. Initially, low-field nuclear magnetic resonance (H-1 NMR) was introduced to characterize the UHPC pore structure. Subsequently, D-optimal and artificial intelligence (AI) models were synergistically employed to achieve the precise design of UHPC. The random forest (RF) and Bayesian optimization convolutional neural networks (BO-CNN) models were employed to predict UHPC performance, with transparent delineation of their hyperparameters in decision trees and Gaussian processes. The results indicated that both models demonstrated proven accuracy (R-2 > 0.90), but the BO-CNN model showed superior precision with an R-2 value exceeding 0.95. Furthermore, genetic algorithm (GA) was employed to optimize the balance between strength and porosity, enhancing UHPC structural performance. The multi-objective design using NSGA-II resulted in five groups in the solution space, achieving the dual goals of maximizing strength and minimizing porosity. Finally, the CO2 emission and cost associated with UHPC, which demonstrated the effectiveness and accuracy of the proposed approach. Overall, this research contributes to a deeper understanding of UHPC mechanical characteristics and porosity, and promotes the AI applications in the field.