Comparison of NDT Data Fusion for Concrete Strength using Decision Tree and Artificial Neural Network

Abstract

Fusion of Non-Destructive Test (NDT) data results in more accurate estimation of concrete strength when compared to any single NDT data. Estimation of concrete strength from NDT results assumes importance for health assessment and evaluation of existing concrete buildings, particularly those near the end of their design life. Application of machine learning tools and response surface method has found popularity in recent years for this purpose. In this study, universally popular Artificial Neural Network (ANN) and relatively un-explored Decision Tree (DT) are applied to estimate concrete strength from rebound number and ultrasonic pulse velocity data collected from literature, in single and combined forms. A ranking system based on ratios of multiple performance measures was demonstrated for cases where different models are adjudged better considering different performance measures. From the results, it was concluded that fusion of NDT data resulted in better accuracy, for both ANN and DT. Comparing the selected performance measures as well as the ranks of the two machine learning tools, ANN models were found to perform better as compared to the DT models. The narrow range of multiple performance metrics obtained for three different data divisions (into modelling and evaluation sets) in all cases imparted confidence in the robustness of the approach of model development adopted in this study.