Utilizing Optimized Machine Learning Techniques to Predict the Compressive Strength of Concrete through Non-Destructive Testing Methodologies

Abstract

Examining the concrete quality in its original location and optimizing machine learning models for precise forecasting of concrete compressive strength(fc) is crucial. Current research advocates the fine tuning of hyperparameters within machine learning methodologies in tandem with non-destructive testing techniques to forecast the compressive strength of concrete. This study aimsto incorporate age as a crucial factor by utilizing data spanning from 3 days to 365 days. This approach enhances the study’s applicability for real-time forecasting purposes. In the methodology of this current research, three machine learning (ML) models— specifically, Multi-Linear Regression (MLR), Decision Tree Regressor (DTR), and Random Forest Regressor (RFR)—are introduced within the context of age as a significant factor influencing measurements obtained from the Rebound Hammer (RN) and Ultra Sonic Pulse Velocity (UPV). These ML models were sequentially applied, followed by a meticulous process of hyperparameter finetuning conducted through grid search Cross-Validation (CV). To gain insights into the predictive results, the study also employed SHapley Additive exPlanations (SHAP) for interpretation purposes. The results of this study reveal the development of an empirical relationship using Multi- Linear Regression, which yielded an R2 value of 0.88. Furthermore, the evaluation showed that Random Forest Regression outperformed other models with an R2 value of 0.95 in the training and 0.92 in the testing datasets. These models hold promise for facilitating decisions about qualitative analyses based on UPV and Rebound Hammer measurements relative to the age of the concrete. Rigorous validation of the models was conducted through standard cross-validation techniques. The research has created and validated hyper tunned machine learning models with the help of grid search cross-validation function, with Random Forest Regression being the most effective. These models can potentially guide decisions regarding qualitative analyses using UPV and Rebound Hammer measurements concerning concrete age. They provide a valuable tool for on-site assessments in construction and structural evaluations. The primary objective of the research is to introduce age as a significant feature. To achieve this, data ranging from 3 days to 365 days was integrated. This inclusion aims to enhance real-time decision-making in construction processes, facilitating actions like the prompt removal of formwork in high-speed construction projects.