Prediction of Equivalent Strength of Hydrated Cement Paste Based on Neural Networks

To optimize material performance and ensure the safety of engineering structures, it is essential to investigate the mechanical properties of cement hydration models with complex structures. This study aims to investigate the influence of the water-to-cement ratio and phase volume fractions on the equivalent mechanical properties of cement paste, particularly focusing on how these parameters influence the behavior of the material. A data-driven model is proposed to predict the mechanical performance of hydrated cement structures. Three-dimensional structural slices of Portland hydrated cement paste were created by utilizing the HYMOSTRUC 3D software. Subsequently, an automated batch-processing script coded in Python was applied to transform these slices into ABAQUS models. Tensile simulations were performed to determine the equivalent elastic modulus and equivalent strength of the structures. Based on the simulation results, a backpropagation prediction model was developed using a data-driven approach. Hyperparameter optimization of the model was performed using K-fold cross-validation to improve its generalization capability. Consequently, the trained neural network model demonstrates high accuracy in predicting the mechanical properties of hydrated cement structures. This approach not only ensures reliable predictions but also significantly reduces the complexity associated with traditional microscale material analysis methods. Overall, this study offers an efficient and robust solution for performance prediction of cement-based materials.