Constructability-based multi-objective optimization with machine learning-enhanced meta-heuristics for reinforcing bar design in rectangular concrete columns

Abstract Optimization of reinforcing bar (rebar) design represents a preponderant factor in reducing material usage and wastes for reinforced concrete (RC) structures. The assessment of constructability of such rebar designs is crucial to improve their practicality and reduce construction costs, which makes the problem multi-objective (MO). However, when applying optimization methods for the design of rebar in RC structures, little attention has been paid on columns, in comparison to beams and slabs. Meta-heuristic algorithms (MA) have been the ones mostly deployed for these types of elements, which have proven to be of high computational cost. Additionally, an existing gap in the literature as to how to relate the design and construction stage of rebar in RC structures through constructability analysis is evident. In this regard, research has been focused mainly at the building level but not at the element level. This works presents a novel algorithmic framework using Machine Learning (ML)-enhanced meta-heuristics for the optimal design of rebar on rectangular RC columns. To assess the constructability of the resulting rebar layouts a Buildability Score (BS) model at the element level is proposed. The complexity analysis of rebar design under the constructability restrictions, through combinatorial optimization (CO), is used to assess the global time efficiency of the framework. The Non-Sorting Genetic Algorithm II (NSGA-II) was deployed for showcase and five different ML algorithms were used to enhance it, namely the k -NN classifier, SVM regression, ANN, Gauss Process (GP) regression, and Tree Ensembles (TE), where the latter three showed the best performance.