Green Concrete Autonomous Designer: Single and Multiobjective Optimization of Mechanical Performance, Economic Viability, and Carbon Footprint

Abstract This paper presents an autonomous machine learning (ML) approach to optimizing green concrete mixtures by balancing mechanical performance, economic viability, and carbon footprint through single and multiobjective optimization. This approach builds upon a comprehensive database of physical tests that were analyzed via Bayesian optimization and non-dominated sorting genetic algorithm (NSGA)-II and NSGA-III algorithms. Then, the effectiveness of this approach was examined through several case studies across normal-strength, high-strength, and ultrahigh-performance concrete. Based on an analysis of over 2,300 real concrete mixes, our findings indicate that traditionally designed mixtures tend to have higher <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>CO</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:math> emissions and monetary costs, which suggests substantial inefficiencies in traditionally adopted baseline mixes. On a more positive note, the developed approach successfully overcomes such inefficiencies by attaining an average reduction of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>CO</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:math> and material cost that can be approximated at 40% and 30%, respectively (while maintaining similar strength performance). Finally, the proposed approach was packaged into a web interface to allow engineers and concrete fabricators with little-to-no coding experience to arrive at optimal mix designs. The study highlights the potential for ML-based optimization not only as a theoretical tool to create green concrete but also as a scalable tool for practical (mass-scale) applications.