Prediction of concrete compressive strength using support vector machine regression and non-destructive testing

Performance assessment of existing building structures, especially concrete compressive strength assessment, is a crucial aspect of engineering construction for most industrialized countries. Non-destructive testing (NDT) techniques are commonly employed to assess the compressive strength of concrete structures. However, existing methods for predicting concrete compressive strength using NDT techniques and machine learning methods do not take into account the concrete mix proportion design. This study proposes an effective method to predict concrete compressive strength by combining NDT tests with different mix proportion designs and curing ages. Specifically, support vector machine regression (SVR) and back propagation neural network (BPNN) models are established. Furthermore, various machine learning evaluation indexes are utilized to assess the model performance. To construct and validate the prediction models, a total of 180 datasets containing concrete specimens with different mix proportion designs and curing ages are collected from the existing research literature. The prediction results show that the coefficients of determination (R2) of the SVR and the BPNN prediction models for the test set of concrete compressive strength are 86.0 % and 86.7 % without considering the concrete mix proportion design. The R2 of the prediction results of the SVR model is higher than 95 % when considering the effects of concrete mix proportion design and curing age. The R2 of the BPNN prediction model ranged between 92 % and 97 %. All the evaluation indexes of the SVR model for predicting the compressive strength of concrete are better than those of the BPNN model. Consequently, the SVR model can be utilized to accurately evaluate concrete compressive strength during the structural performance assessment of existing buildings.