Hybrid artificial neural network with biogeography-based optimization to assess the role of cement fineness on ecological footprint and mechanical properties of cement mortar expose to freezing/thawing

Abstract

The role of cement plants on negative environmental impacts is known. Therefore, investigating the type and amount of cement used in the mix design can significantly reduce the cement consumption and, consequently, reduce the pollution resulting from its excessive production. Also, despite the efforts by researchers to suggest the prediction model of mechanical properties of cementitious materials, there is no reliable model that simultaneously predicts them under freezing and thawing (FT) condition because of the lack of attention to the effective parameters. To fulfil these purposes, initially, an extensive experimental work considering 54 mix designs (810 specimens) was prepared, and the impact of cement fineness in the form of three cement strength classes (CSC 32.5, 42.5 and 52.5 MPa) and amount of cement consumption (Sand/Cement of 2.5, 2.75 and 3.0) on porosity, flexural (Ff) and compressive strength (Fc) under five different FT cycles (0, 50, 100, 150 and 200), and also cement paste texture by SEM and XRD analysis were investigated. Then, a new multi-objective model was developed based on hybrid artificial neural network (ANN) with biogeography-based optimization (BBO) to improve prediction accuracy.Results demonstrated that specimens with fine cement (CSC 52.5 MPa) experienced decreasing the porosity up to 55% and increasing the load carrying capacity about 60% Ff and 80% Fc compared with coarse cement ones (CSC 32.5 MPa) at 200 FT cycles. Also, SEM images and XRD analysis illustrated that the finer cement can improve the homogeneity of the cement paste, and thereby it became denser texture. Comparison of results with previous studies confirmed the S/C of 2.75 as optimum proportion in mix design. Results of sensitivity analysis on models by performance indicators revealed that the best performance for MOANNII-BBO model compared with other models due to considering the CSC and amount of cement consumption as input parameters. Furthermore, the analysis of influencing input parameters indicated the sensitivity of 20% CSC and 18% amount of cement to the model performance. The findings of this work can bring notable benefits for the range of issues involved.