Decoding the Structural Origin of Creep in Colloidal Gels by Machine Learning

When subjected to a sustained load, jammed colloidal calcium-silicate-hydrate (C-S-H) gels (the binding phase of concrete) tend to exhibit delayed viscoplastic creep deformations. However, the structural mechanism of creep in C-S-H gels (and driving force thereof) remains only partially understood. This partially arises from the fact, due to the long timescale of creep, its physical modeling has remained challenging. Here, based on a mesoscale model of C-S-H gels, we present an accelerated simulation method (based on stress perturbations and overaging) to model creep deformations in C-S-H [1]. Based on these simulations, we adopt Support Vector Machine (SVM, a supervised machine learning classification algorithm) to decode the structural mechanism of creep in C-S-H gels [2]. This allows us to "find Needles in haystacks," that is, to pinpoint the key structural features enabling grain reorganizations and creep deformations within the gel.

[1] Liu, H., Dong, S., Tang, L., Krishnan, N. A., Masoero, E., Sant, G., & Bauchy, M., Journal of Colloid and Interface Science, 542, 339-346 (2019).
[2] Liu, H., Fu, Z., Yang, K., Xu, X., & Bauchy, M., Journal of Non-Crystalline Solids: X, 4, 100036 (2019).