Coupled effects of thermal exposure and high strain rate on CO2 emissions of concrete structures: A comparative study of AI-driven emission signatures

Abstract

This study introduces a novel AI-driven framework to predict CO2 emissions of steel fiber-reinforced concrete (SFRC) under the coupled effects of thermal exposure (200 degrees C-1200 degrees C) and strain rate (10-5/s to 102/s), conditions that simulate real-world scenarios such as fire, impact, and blast events. Unlike traditional life-cycle assessments that focus solely on material composition, this research integrates mechanical stressors to model dynamic environmental responses. A dataset of 399 scenarios, combining empirical and synthetic data, was used to train Random Forest (RF) and Extreme Gradient Boosting (XGBoost) models. XGBoost outperformed RF with a peak R2 of 0.987 and RMSE of 8.90 kg CO2e/m3, effectively capturing the nonlinear relationships among input features. SHAP (SHapley Additive exPlanations) analysis identified exposure temperature, cement content, and supplementary cementitious materials (SCMs) as the most influential variables. The study also introduces the concept of "emission signatures" distinct emission patterns triggered by thermal and mechanical stress redefining CO2 output as a function of service performance rather than static material properties. This approach bridges the gap between structural resilience and environmental responsibility, offering a scalable tool for designing low-carbon, high-performance concrete systems in infrastructure exposed to extreme conditions.