Özet:
The service life performance of conventional and modified concrete subjected to harsh climatic condition environment is directly related to durability properties of concrete like abrasion, freezing and thawing cycles. These properties are critical issues that should be predicted before performing experimental test. On this basis, the basic purpose of this paper is to predict the abrasion loss, freezing and thawing properties of concrete modified with silica fume (SF) and steel fiber (SFb) by using mix design and additional properties. From this point of view, a conducted experimental study was selected as a case study. In the control concrete (CC) mixtures, Portland cement, crushed stone aggregate, and superplasticizer (SP) were used in the selected experimental study. SP in concrete mixtures was used in the amounts of 1.0%, 1.5%, and 2.0% by weight of cement, and so modified concrete was produced with and without SFb according to the target strength of C25. Furthermore, SF and SFb were used in different amounts to modify the concrete. The SF was replaced with cement in the amounts of 7.5%, 10.0%, and 15.0%. In total, 16 different mix designs were prepared with different SP and SF ratios. In addition, SFb was added to all mixtures of designed concrete at a constant amount of 65 kg/m3. Additionally, a 16-mix design was prepared with SFb. Cumulatively, 32 different mix designs were prepared for the experimental study. Tests on the fresh, hardened, and life-cycle performance properties of the concrete were conducted. As for the metaheuristic part of this study, on the basis of the available experimental data, life-cycle performance parameters of the concrete modified with SF and SFb are predicted by using single and hybrid generalized extreme learning machine methods. Eight different data sets were generated with gradually extended input data. Two different outputs were considered: abrasion resistance (AL) and freezing/thawing (FT). Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) algorithms were used to produce binary and ternary hybrid methods. Four different models were proposed as listed: single use of Generalized Extreme Learning Machine (GRELM), binary use of GRELM-PSO, and GRELM-GWO. Finally, PSO and GWO were hybridized and integrated into GRELM. Two quality indicators, namely Root Mean Square Error (RMSE) and correlation of determination (R2), were considered to see the performance of the prediction. The results showed that the proposed ternary prediction model composed of GRELM-PSO-GWO provided more accurate results in all sets from 74% to 91% by extending input parameters, even if complicated parameters are inserted in as an input to the data set. © 2022 Elsevier Ltd