Extreme fine-tuning and explainable AI model for non-destructive prediction of concrete compressive strength, the case of ConcreteXAI dataset

Abstract

This groundbreaking study introduces a novel approach employing Extreme Fine-Tuning (XFT) combined with Explainable Artificial Intelligence (XAI) for the accurate, non destructive prediction of concrete compressive strength. By analyzing a state-of-the-art dataset containing 18,480 data points, this study developed a deep neural network that, through extensive hyperparameter optimization, achieves unprecedented prediction accuracy of approximately 98.7%. The novelty of this research lies in the sophisticated integration of XFT and XAI techniques, which not only significantly enhances prediction accuracy but also provides insightful explanations of the model’s decision-making process, shedding light on the factors influencing concrete strength. This dual focus on accuracy and explainability represents a significant advancement in the application of Artificial Intelligence (AI) in material science and civil engineering, offering a powerful tool for researchers and practitioners.This study culminates in a model that outperforms existing methodologies in predicting concrete compressive strength, with an accuracy superior to 98.5% in both instances, testing and validation. By integrating XAI into this approach, we have also opened new avenues for understanding the complex relationships between concrete composition and its mechanical properties. This study marks a substantial step forward in the non-destructive evaluation of construction materials. It sets a new benchmark for transparency and interpretability in AI models within the engineering domain.