Research Projects

Concrete structures exposed to fire often suffer significant loss of strength, stiffness, and durability due to the combined effects of elevated temperature, moisture migration, and mechanical loading. A particularly dangerous manifestation is explosive spalling, where surface layers of concrete abruptly detach under fire, posing severe structural and safety risks. Existing computational models based on thermo-hygro-mechanical (THM) frameworks often face limitations due to high computational costs, sensitivity to material parameters, and lack of integration with evolving AI tools. The central innovation of this research lies in developing a hybrid computational framework that combines physics-based finite element modelling with AI/ML-based surrogate modelling to simulate concrete behaviour under fire conditions with higher efficiency and accuracy. The proposed project aims to build a coupled multi-surface plastic-damage model integrating pressure-sensitive plasticity, isotropic damage evolution, and transient creep laws. A spectral decomposition-based stress integration algorithm will enable the decoupled treatment of tension and compression, improving numerical robustness in multiaxial states. Simultaneously, machine learning models—such as deep neural networks and Gaussian process regressors—will be trained on high-fidelity FE simulation outputs to learn mappings for complex sub-models like pore pressure evolution and thermal creep. These AI-based surrogates will be embedded directly into the finite element solver using a hybrid C++–Python API. This will drastically reduce solution time, enabling scalable simulation of large-scale structural elements subjected to realistic fire scenarios. The project will proceed through: - Development and calibration of the THM constitutive model. - Generation of training data from detailed FE simulations. - Training and deployment of surrogate models for selected sub-physics. - Validation using experimental data (e.g., Gernay, Tenchev, Davie et al.). - Parametric studies and full-scale simulations of RC members under ISO 834 and Hydrocarbon fire curves. The proposed approach offers a step-change in modeling capability by combining interpretability and physical fidelity with speed and flexibility. If successful, it will: - Provide insights into spalling mechanisms under realistic fire loading. - Reduce the dependency on extensive physical testing. - Improve predictive reliability and reduce computational demand in fire safety engineering. - Contribute open-access software tools (subject to institutional clearance). - Advance AI applications in computational mechanics and structural fire engineering. The research addresses critical gaps in performance-based fire design and aligns with national and international priorities in disaster-resilient infrastructure and AI-integrated simulation technologies.