The 4th Asian Concrete Federation Symposium on Emerging Technologies for Structural Longevity (ACF2023- ETSL)

14 / 2022-02-28 18:54:15

Prediction of Recycled Aggregate Concrete Creep using Back Propagation Neural Network and Support Vector Machine

Creep model, Machine learning, Recycled aggregate concrete.

Promoting recycled aggregate concrete (RAC) can effectively reduce carbon emissions throughout the life cycle of the construction industry and increase the rate of waste concrete recycling. Due to the significantly higher creep of RAC than that of natural aggregate concrete (NAC), the accurate prediction for long-term RAC creep deformation is vital. Hence, this study investigates the predictive model for RAC. In this study, a creep database of RAC with 106 groups of 1309 experimental data points considering 15 influencing parameters has been set up. Furthermore, the back propagation neural network (BPNN) model and the support vector machine (SVM) model have been adopted to process data and predict. A comparative analysis of the accuracy of the prediction between the existing RAC creep model, the BPNN model, and the SVM model. The results show that，for the prediction of RAC creep, the BPNN model and SVM model are much more accurate than the existing RAC creep model. Finally, the extended parameters of the model were analyzed based on the BPNN model to further clarify the effects of the recycled coarse aggregate (RCA) replacement rate, the RCA residual mortar content, and the RAC strength on the creep properties of RAC.