Predicting concrete compressive strength using optimized deep learning and large language models.

Area of Science:

• Materials Science
• Civil Engineering
• Artificial Intelligence

Background:

• Predicting concrete compressive strength is crucial for sustainable construction.
• Complex interactions between mixture components, admixtures, and curing conditions pose challenges.
• Existing methods often struggle with the nonlinear nature of these interactions.

Purpose of the Study:

• To develop an advanced hybrid AI framework for enhanced concrete compressive strength prediction.
• To improve the accuracy and robustness of predictive models for construction materials.
• To leverage novel optimization and deep learning techniques for material science applications.

Main Methods:

• Integration of the iHow Optimization Algorithm (iHowOA) with Spatio-Temporal Graph Convolutional Networks (STGCN).
• Utilization of a large language model (LLM) for data preprocessing, including semantic validation and feature harmonization.
• Optimization of STGCN architecture using iHowOA's adaptive decision-making and knowledge acquisition capabilities.
• Graph-based modeling to capture spatial dependencies and temporal strength evolution.

Main Results:

• The proposed iHowOA-STGCN framework demonstrated superior predictive performance compared to ten other metaheuristic optimizers.
• Achieved lower prediction errors and higher correlation coefficients on a public dataset.
• Identified key relationships between cement properties, age-dependent strength gain, and physicochemical interactions.

Conclusions:

• The iHowOA-STGCN framework offers a promising data-driven decision-support tool for concrete strength prediction.
• The LLM-driven preprocessing enhances data quality and model input robustness.
• Further validation on diverse datasets is recommended to confirm generalizability and practical applicability in real-world scenarios.