Enhancing Geopolymer Lightweight Concrete Performance through AI and Deep Learning-Based Mix Design Optimization

Abstract

This study introduces a hybrid Artificial Intelligence (AI) framework that integrates Deep Learning (DL) and Natural Language Processing (NLP) to predict Lightweight Concrete (LWC) performance. LWC, a critical material in sustainable construction, is valued for its lightweight nature, thermal insulation, and energy efficiency. However, the complex interdependencies among its properties, including density, compressive strength, and thermal conductivity, present significant challenges for accurate modelling and optimisation. The proposed framework addresses these challenges by combining NLP for automated extraction of material properties from unstructured sources, such as research articles, with DL for predictive analytics. The integration of experimental data with NLP-extracted insights forms a comprehensive dataset, enabling precise performance predictions. The hybrid AI model outperformed standalone methods, achieving higher accuracy, reduced error rates, and meaningful insights through SHAP-based feature importance analysis, which highlighted density and compressive strength as key predictors. These findings demonstrate the framework’s potential to bridge data gaps, enhance the optimisation of LWC properties, and facilitate its application in sustainable construction practices. By advancing AI-driven solutions in material science, this framework offers a scalable and innovative approach to addressing challenges in construction engineering and promoting sustainability.