Estimation of the FRP-Concrete Bond Strength with Code Formulations and Machine Learning Algorithms

Abstract The present study pertains to the bond strength and development length of FRP bars embedded in concrete. The experimental results in the literature were compared to the analytical estimates from the equations of different international codes and machine learning techniques, i.e. Gaussian Process Regression, Artificial Neural Networks, Support Vector Machines Regression, Regression Tree and Multiple Linear Regression. The comparison was realized for four different experimental methods, i.e. hinged beam, beam-end, spliced beam and pull-out, to specify the analytical equation or method with the highest agreement with the test results for each method. GPR method was found to provide the highest accuracy with a mean value of 0.95 and a standard deviation of 0.14 for the predicted-to-experimental bond strength ratio. Based on coefficient of determination, Root Mean Square Error and Mean Absolute Percentage Error statistical criteria, GRP method was followed by ANN, MLR and SVMR based on the agreement with the experimental results. Among the code equations, the bond strength equation of the ACI 440.1R-15 code resulted in highest agreement with experimental results, but the predicted values remained on the over-conservative side. The other code formulations were established to yield to estimates, nearly constant for varying test parameters and highly conservative.