Evaluation of Statistical and Machine Learning Methods on Predicting One-way Shear Strength of Reinforced Concrete Beams

Abstract

It is a well-known fact that predicting shear strength of concrete beams is a very complex topic. The traditional approach is to assume that the shear strength is the sum of the concrete contribution (Vc) and the shear reinforcement contribution (Vs). In recent years, several proposals including Bentz & Collins (2017), Cladera et al. (2017), Frosch et al. (2017), Li et al. (2017), Park & Choi (2017) and Reineck (2017) were published based on different failure mechanisms. The current code ACI318-19 (2019) also updated the one-way shear design specifications with a new set of empirical equations based on Kuchma’s research (1998). However, none of these methods can fully agree with each other either. Therefore, methods which are not directly related with failure mechanisms will be valuable to provide alternative ways to predict shear strength of concrete beams as comparisons. Recently, two database including shear database without shear reinforcement (Reineck, Kuchma, & Kim, 2003) and shear database with shear reinforcement (Reineck, Bentz, & Fitik, 2014)were collected and reviewed. With these two databases, several statistical and machine learning (ML) methods can be applied and evaluated as alternative methods to compare with traditional theoretical methods. In this thesis, three theoretical methods (ACI318-19, Frosch et al. and Li et al.) were chosen to be validated by these two databases. Selected statistic (multiple linear regression, LASSO, LARS) and machine learning (random forest, neural network, and support vector machine) methods were applied on the databases to evaluate the efficiency and accuracy of these methods on predicting shear strength. This study is the first study attempting to apply the statistical/ML methods on these databases to predict the shear strength of concrete beams. A sensitivity analysis was conducted on all these different methods at the end of this thesis to evaluate the stability of these methods.