Green mix design of rubbercrete using machine learning-based ensemble model and constrained multi-objective optimization

Abstract

Although the use of waste rubber (WR) in concrete can alleviate some negatives effect on sustainability, it can decrease the compressive strength (C.S) of the produced rubbercrete. Predicting the C.S of rubbercrete using reliable and comprehensible methods would allow practical use of these materials in construction projects. Besides, other aspects of concrete mix design, including environment and economics should be considered in the concrete mix design to achieve an eco-friendly and cost-effective concrete. For this purpose, a green mix design model is proposed to estimate the constituents of rubbercrete using the machine learning-based ensemble model (as a combination of M5P tree and multi-gene expression programming (MGEP) algorithms) as well as constrained multi-objective grey wolf optimizer. To do so, four main goals are sought, including the C.S, cost, CO2 emission of rubbercrete, and the amount of WR consumption in the mixture. Generally, seven optimization problems with two, three, and four objectives were designed. To model the C.S. of rubbercrete, a comprehensive database with 712 data tuples was collected from the 30-international peer-reviewed papers. The amounts of rubbercrete constituents and concrete age are considered as the input attribute, and the C.S of rubbercrete is the output attribute. Comparing the error metrics of the developed models shows that the proposed ensemble model outperforms the conventional M5P tree and MGEP models by 13.7% and 5.5%, respectively. Moreover, the total numbers of 182 optimal mix designs are obtained for all defined scenarios. The results reveal the opportunity to obtain the optimal mix designs of rubbercrete with the WR to the natural aggregate ratio of about 2%–6%.