Artificial intelligence (AI) assisted reactive molecular dynamics (MD) simulations of cement hydration

Description

Cement hydration plays a significant role in determining the mechanical properties of cementitious materials. Indeed, hydrated cement paste, which is the key product of cement hydration, connects various constituents in concrete including coarse and fine aggregates, admixtures and additives together. With a thorough understanding of cement hydration, advanced material development of cementitious materials from the fundamental atomistic perspective becomes feasible and the strength of hydrated cement paste can be improved by several multiples. Traditional approaches of understanding cement hydration and its related products are mainly based on experimental and theoretical analyses, in which the fundamental steps and reaction mechanism during the cement hydration process can hardly be clarified. Such knowledge of cement hydration should be investigated using advanced technology that can provide detailed information from nanoscale. In recent decades, reactive molecular dynamics (MD) simulation has been adopted as a powerful tool to describe various chemical reactions involving bond breaking and formation. However, conventional reactive forcefield (ReaxFF) may not be capable of describing the entire cement hydration process accurately in view of the complexity of the involved materials and there is a great need for further refinement and optimization of ReaxFF parameters in cementitious materials. With the recent development of artificial intelligence (AI), a predictive model of cement hydration based on the integration of AI and ReaxFF is believed to be promising for addressing the existing limitation. The objective of this research is to investigate the hydration process of dicalcium silicate, which is more abundant than tricalcium aluminate and controls a larger number of engineering properties of cements, using reactive MD simulations equipped with AI technique, together with experimental validations. Here, a comprehensive database for the hydrated product of dicalcium silicate will firstly be developed, which will be used in the AI algorithm. Then, the selected AI algorithm will be programmed using C++ and Python such that it becomes compatible with the codes of MD simulations. Afterwards, the featured parameters will be trained and a correct predictive model will be derived with experimental validations using X-ray diffraction, fourier-transform infrared spectroscopy, scanning electron microscopy and nanoindentation. Synergizing the knowledge developed in these research tasks, this proposed research will provide fundamental knowledge of cement chemistry and it is envisioned that material engineers can easily predict the performance of cementitious materials using this AI assisted reactive MD simulation approach, which leads to more sustainable applications of cementitious materials in new construction and retrofitting industries.