Prediction of SEM-X-ray images' data of cement-based materials using artificial neural network algorithm

Recent advances of computational capabilities have motivated the development of more sophisticated models to simulate cement-based hydration. However, the input parameters for such models, obtained from SEM-X-ray image analyses, are quite complicated and hinder their versatile application. This paper addresses the utilization of the artificial neural networks (ANNs) to predict the SEM-X-ray images' data of cement-based materials (surface area fraction and the cement phases' correlation functions). ANNs have been used to correlate these data, already obtained for 21 types of cement, to basic cement data (cement compounds and fineness). Two approaches have been proposed; the ANN, and the ANN-regression method. Comparisons have shown that the ANN proves effectiveness in predicting the surface area fraction, while the ANN-regression is more computationally suitable for the correlation functions. Results have shown good agreement between the proposed techniques and the actual data with respect to hydration products, degree of hydration, and simulated images.