Using artificial neural networks for predicting the elastic modulus of recycled aggregate concrete

• ANN models are proposed to predict E values of recycled aggregate concrete.
• ANN model is enhanced by incorporating related data sets into their training.
• ANN models perform better than correlation regression models.
• ANN can model the properties of concrete made with RAs from different sources.

This paper is an extension of the previous study to further explore the applicability of artificial neural networks (ANNs) in modeling the elastic modulus (Ec) of recycled aggregate concrete (RAC). In this study, ANNs-I is firstly constructed by using 324 data sets collected from 21 international published literatures, which are randomly divided into three groups as the training, testing and validation sets, respectively. Then ANNs-II with 16 more data sets of the authors’ own experimental results added to the learning database of ANNs-I is established to examine whether the performance of ANN can be further improved. The predicted results are compared with the experimentally determined results and that modeled by conventional regression analysis. The constructed ANNs-I and ANNs-II are also applied to other experimental data sets obtained from the authors and a third party published literature to test its applicability to recycled aggregate (RA) taken from different sources. The results show that the constructed ANN models can well predict the elastic modulus of concrete made with RA derived from different sources.