Compressive strength prediction of hollow concrete masonry blocks using artificial intelligence algorithms

Abstract

In this study, artificial intelligence algorithms are proposed for estimating the compressive strength of hollow concrete block masonry prisms, including neural networks (ANN), combinatorial methods of group data handling (GMDH-Combi), and gene expression programming (GEP). To train and test the proposed models, 102 samples of hollow concrete prisms from previous research works were collected. The height-to-width ratio of hollow concrete prisms and the compressive strength of mortar and concrete blocks were considered as inputs. In order to evaluate the validity and predictability of the proposed models, they were compared with empirical models and models presented in codes and standards. Among the suggested and existing models, the ANN model with an R-value of 0.950 and MAPE error value of 6.921% had the best performance, which with a more complicated equation, can be used in the scientific aspect. In contrast, the other two proposed models (GMDH-Combi and GEP) with acceptable performance and accuracy levels and more simple closed-form equations can be utilized in practical aspects. Based on the parametric analysis, the proposed models were highly efficient and accurate. Moreover, the sensitivity analysis results showed that in all three proposed models of ANN, GMDH-Combi, and GEP, the compressive strength of concrete blocks was the most effective input parameter in the compressive strength estimation of hollow concrete prisms.