A novel support vector regression (SVR) model for the prediction of splice strength of the unconfined beam specimens

摘要

Splice strength in reinforced concrete is an important parameter for the safe design of any structure which should be assessed with ease and accuracy. Analytically the assessment of splice strength is a complex problem because of the improper idealization of the stress field around the splice region. Further, for the empirical models, the assessment is still complicated due to the variable nature of materials used i.e. concrete and steel especially in case of high strength concrete beams. The focus of the study is to develop a robust model for the prediction of splice strength encompassing significant parameters in a wide range using support vector regression which is for the first time used for this assessment. Further, for the purpose of comparison, in addition to existing empirical models NMR (Nonlinear Multi-regression), and ANN (Artificial Neural Networks) models were also formulated. A data set of 267 splice beam specimens from the literature was used including the authors own generated data for the training and testing of the models. The parameters under consideration are the diameter of the bar, the compressive strength of the concrete, development length and cover to the reinforcement. The statistical analysis of the models suggests that NMR is better than the existing empirical correlations however inferior than the SVR and ANN models for the prediction of splice strength. Furthermore, SVR and ANN show comparable accuracy in predicting the splice strength of unconfined beam specimens however, SVR is found to be more efficient than ANN. The study concluded that SVR has the potential to predict the splice strength with higher accuracy in comparison to the prescriptive empirical relationships and can be used for design purposes. (C) 2020 Elsevier Ltd. All rights reserved.