A Boosted Tree Machine Learning Alternative to Predictive Evaluation of Nondestructive Concrete Compressive Strength

摘要

This study investigates the use of tree-based machine learning methods in concrete non-destructive tests (NDT) evaluation. The study encompassed different phases. The first involved the use of destructive and non-destructive mechanisms to assess concrete strength on cube specimens. The second phase examined site assessment of selected structures using popular NDT tools. The third phase implemented a tree-based machine learning approach to characterize a relationship between concrete properties and destructive compressive strength for cubes and selected structures. It established that ultrasonic speed and rebound number were adequate to predict compressive strength. Variable importance plots from boosted tree learning suggested a hierarchy of parameter importance that challenges Pearson's correlation coefficients. In order to establish the effectiveness of tree-methods, analyses show that classical regression struggled to attain a variance score of 0.43 during training while boosted tree doubled this score during testing on unseen validation set. The results present a case for tree-based analysis in concrete NDT evaluation.