predictions of optimum design quantities for reinforced concrete beam-slab system through reutilization of empirical data from constructed projects in Sri lanka

Abstract

Decisions in the Preliminary Design Stage (PDS) highly influence the final structural design deliverables. However, the PDS procedures are generally recursive, labour-intensive and time­ consuming. Expert-driven designs typically dominate at this stage to achieve structurally optimized solutions, yet the reutilization of earlier design information remains limited. Besides,as the floor system is the major structural component of a multi-storied building, addressing its embodied energy demands requires close attention to yield a sustainably optimized solution. To overcome the limitations of traditional practices in the PDS, this study employs an Artificial Neural Network (ANN), as a learning­ based method to predict design quantities, aiming for optimization of reinforced-concrete beam-slab systems of multi-storied buildings. The models were developed using the data collection from the constructed buildings in Sri Lanka,following standard machine learning protocols. The proposed best models are recorded predictions with R2 values above 0.8, demonstrating superior performance in the context of heterogeneous real-world data.The proposed ANN-based numerical model streamlines the selection of the best-optimized floor system by analysing a range of alternatives encapsulating performance, economic and environmental criteria. By capturing and reapplying expert knowledge from structural data and drawings, which have been previously evaluated and accepted by experienced engineers, the approach enhances decision-making in the PDS.