Inverse design of cement and cement paste composition for targeted tensile strength using sequential neural networks and hydration based micromechanical modelling
Overview & Synopsis
The fracture performance of hydrated cement paste is a property of its evolving microstructure, which in turn is governed by the chemical composition of the initial binder system and the kinetics and thermodynamics of the hydration process. In this study, a novel data-driven meta-model, referred to as Cement-to-Tensile (C2T), is proposed to predict and optimize the tensile strength of HCP without resorting to computationally expensive simulations. The C2T framework is constructed by coupling two surrogate models in a cascade architecture: a Long Short-Term Memory (LSTM) network trained to simulate the evolution of microstructure from cement binder composition, and a regression model, that predicts tensile strength based on microstructural phase composition, including hydration products, unhydrated phases, and porosity. The model was trained on a dataset generated by an experimentally validated micromechanical model. The LSTM model was trained on a high-fidelity, numerically simulated dataset containing 39,074 unique data record, encompassing 16 Portland cement compositions under varying water-to-cement ratios and hydration times. The predictive performance of the C2T model was rigorously evaluated and subsequently deployed as the objective function in a two-step optimization framework to identify cement compositions achieving target tensile strengths. Results highlight the model's ability to capture the synergistic effects of composition, hydration, and microstructure evolution, offering a powerful tool for the inverse design of cementitious materials. This model lays the foundation for multi-objective design schemes that could take into account mechanical properties, durability criteria, and environmental impact in order to accelerate the design of substituted cementitious materials. The practicality of the C2T model lies in its capacity to suggest optimal compositions through data-driven optimization, thereby accelerating and facilitating the development of advanced cementitious materials. Moreover, it extends the conventional predictive scope of data-driven models, which are predominantly limited to forward predictions, by enabling inverse design strategies for tailored material performance.
hal.science)
← / → for pages • + / - for zoom • 0 to fit width
Powered by Mozilla PDF.js v3.11
Document Information
| Document Title | Inverse design of cement and cement paste composition for targeted tensile strength using sequential neural networks and hydration based micromechanical modelling |
|---|---|
| Author / Organization | Jinane Murr, Syed Yasir Alam, Frederic Grondin |
| Subject & Discipline | Multidisciplinary Academic Open Research |
| Categories | Science & Mathematics |
| Publication Year | 2026 |
| License / Rights | Open Access / Creative Commons |
| Resource Verification | Listed PDF (Verification Pending) |
| Original Host Website |
HAL Open Science
(hal.science)
|
| File Format | application/pdf (Portable Document Format) |
| Language | EN |
| Catalog ID | DOC-199749 |
| Original Source Link | Visit Source Page on hal.science » |