Selecting the optimal stability framework for high-rise steel buildings is a critical decision that impact both economic efficiency and sustainability. This decision is not easy since there are many stability frameworks to choose from, such as tube, concrete shear-wall, and outrigger system. Designers and clients aspire to compare multiple structural designs, considering both different frameworks and variations in geometry. Unfortunately, the current process of evaluating multiple designs is time-consuming and relies on the designer's experience and rules of thumb, rather than being driven by data. To address this challenge, predictive models can be used to estimate performance in terms of structural and environmental costs based on the given geometry and framework. However, traditional curve fitting methods often fall short in accurately predicting a complex relationship. In response, this study explores the potential of artificial neural networks to accelerate the decision-making process by predicting the most efficient framework during the early design phase while maintaining sufficient accuracy.

To determine this potential, two parametric models were developed using Rhinoceros software to represent buildings with braced framed tubes and outrigger systems. These models automatically optimised beam and column dimensions by FEM of Karamba, aiming to minimise the mass of these structural elements, across a range of building widths (15 to 60 meters) and heights (48 to 300 meters). The resulting data sets of both braced framed tube and outrigger system reflected the structural and environmental costs for the different designs. Separate neural networks were modelled for each framework. These networks were trained on the different data sets. By comparing predicted structural and environmental costs, the models assisted in the selection between braced framed tubes and outrigger systems.

The artificial neural networks accurately approximated structural and environmental costs for both stability frameworks. The Mean Absolute Percentage Error (MAPE) for the braced framed tube was 14%, while the most accurate alternative curve fit, a third-order polynomial, had a MAPE of 25%. For the outrigger system, the neural network achieved an even lower MAPE of 7%, where the most accurate alternative curve fit, also a third-order polynomial, had a significantly higher error with a MAPE of 31%. The neural networks outperformed traditional curve fitting methods. Additionally, the neural network generated instant results, taking only a second compared to the parametric model’s 5 to 30 minutes. However, achieving overall time efficiency with the neural models will require approximately three months when considering both setup time and output generation time. Optimal stability varied based on specific width and height combinations: when looking at environmental costs, the braced framed tube excelled for lower (50-80m) and higher heights (200-300m), while the outrigger system was more efficient for middle heights (80-200m). As the structure’s slenderness increased, the braced framed tube regained efficiency for the middle heights.

Impact of the stability framework is defined as the relative contribution of embodied carbon of the stability framework to the total embodied carbon of the structure (including superstructure and floors). The impact of the stability framework ranged from 25% to 57% for the braced framed tube and 33% to 66% for the outrigger system, with the impact increasing with the building's height...