Hybrid AI Models For Predicting Lightweight Concrete Performance: Integrating Deep Learning And NLP For Material Property Extraction

Abstract

Lightweight concrete (LWC) has become a buzzword in modern construction due to its unique properties, such as low density, thermal insulation, and sufficient strength, making it ideal for applications in tall buildings, bridges, and marine structures. However, predicting its performance remains a significant challenge due to the complexity of its mix design and the influence of external factors such as curing conditions. Traditional empirical methods and standalone AI models fail to leverage the vast amount of unstructured textual data available in standards, research papers, and technical reports. This results in suboptimal performance of predictions. This study investigates into a hybrid artificial intelligence (AI) framework that integrates natural language processing (NLP) and deep learning to address these limitations. The NLP module uses material properties such as water-to-cement ratio, aggregate size, and curing conditions from experimental data, to form comprehensive input datasets. A deep learning model, utilising convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), predicts critical performance parameters, including compressive strength, thermal conductivity, and durability. The hybrid model achieved significant improvements, with 91.2% accuracy for compressive strength predictions and 89.4% accuracy for durability, outperforming both traditional regression and standalone deep learning approaches. The study also highlights the practical implications of this approach, such as cost reduction in testing, optimisation of mix designs, and enhanced sustainability. By bridging the gap between computational intelligence and civil engineering practices, this hybrid AI framework sets a precedent for data-driven innovations in LWC design, paving the way for efficient and resilient construction solutions.