Ensemble learning based compressive strength prediction of concrete structures through real-time non-destructive...

Area of Science:

• Civil Engineering
• Materials Science
• Computational Intelligence

Background:

• Accurate prediction of concrete compressive strength (CS) is crucial for structural integrity.
• Non-destructive testing (NDT) methods like rebound hammer (RH) and ultrasonic pulse velocity (UPV) offer efficient assessment.
• Traditional analytical models often lack the accuracy required for complex material behavior.

Purpose of the Study:

• To conduct a comparative analysis of ensemble learning algorithms for predicting concrete CS using NDT data.
• To evaluate the performance of six popular algorithms: Adaboost, CatBoost, Gradient Boosting Tree (GBT), Random Forest (RF), Stacking, and Extreme Gradient Boosting (XGB).
• To assess models based on RH, UPV, and combined RH and UPV data.

Main Methods:

• Development of prediction models using ensemble learning algorithms on a dataset of 721 concrete samples.
• Data sourced from laboratory, in-situ testing, and existing literature.
• Comparative evaluation of model performance using metrics such as Mean Absolute Percentage Error (MAPE).

Main Results:

• The Extreme Gradient Boosting (XGB) model exhibited the highest performance across various NDT-based prediction categories.
• Specific models like RH-M5, UPV-M6, and C-M6 (combined UPV and RH) showed top performance.
• XGB achieved significantly lower MAPE values compared to Adaboost, CatBoost, GBT, RF, and Stacking.

Conclusions:

• Ensemble learning, particularly XGB, offers a highly accurate and dependable approach for predicting concrete CS from NDT data.
• The findings suggest XGB outperforms other soft computing techniques and traditional predictive models in this context.
• This research provides a robust computational intelligence framework for concrete quality assessment.