This research explores the integration of machine learning to support sustainable design in data centres by incorporating reclaimed steel elements into automated structural design workflows. The study begins with a literature review addressing the impact of steel reuse, the availability of reclaimed sections, strategies for automated, stock-constrained design and the applications of machine learning within the structural design process. Three reclaimed steel element databases are considered and three cross-section selection methods are evaluated and validated, establishing the optimal basis for a machine learning application within the design process. The cross-section selection method is used to gather data on generated
design configurations for braced structures and moment-tight frame structures. Per element, the element location, length, type and profile are recorded for four separate design choice configurations. The data collected acts as a basis for a machine learning application for cross-section prediction.

Central to this research is the development and training of a Recurrent Neural Network (RNN) for sequential classification, aimed at predicting cross-sections based on design parameters. Four models were trained, each tailored to different combinations of design choices and reclaimed steel profiles. The second step in this workflow is the reclaimed steel integration which is an optimization model that integrates the RNN’s predictions to refine grid-spaces in response to available reclaimed steel databases. Finally, the models and the resulting designs were validated to assess performance and applicability.

The trained RNN models reached a test accuracy ranging between 82% to 93% and the optimized steel integration resulted in design configurations which are slightly over-dimensioned, but all have an overall steel utilisation between 0.2 and 0.8. These findings underscore the feasibility of applying machine learning within the structural design domain to promote reclaimed steel integration at the early stages of the design process. A practical workflow is established that creates the possibility for adaptation to other building design typologies, making reuse-oriented design more accessible and efficient.