Generative AI for cementitious microstructure synthesis
This paper presents my first attempt at using generative AI to synthesize microstructures of cementitious materials. A Denoising Diffusion Probabilistic Model (DDPM) was employed to generate synthetic BSE images of OPC with a water-to-cement ratio (w/c) of 0.4. The generated images closely resemble real BSE images in both statistical and microstructural analyses. I have recently extended this work to the conditional generation of three-dimensional hydrating cementitious microstructures with varying w/c ratios, Blaine values, and curing ages. Stay tuned for further updates!
Mar 14, 2025
A Mini-TSTM for Efficient Stress Test
A Mini-TSTM is devised to allow for efficient stress test, which further enanbles comprehensive numerical study on the stress evolution, particularly on the creep-relaxation conversion in the simulation.
Jan 1, 2024
Active Learning for Stress Evolution Prediction
An active ensemble learning is constructed to efficiently predict the early-age stress evolution based on a themo-chemo-mechanical model, light gradient boosting model, and gaussian process.
Dec 1, 2022