Structural Design Optimization of Steel Beams Using Random Forest and Multi-Criteria Analysis

Abstract

The increasing demand for structurally efficient and cost-effective construction has encouraged engineers to explore advanced optimization techniques for structural elements, particularly I-beams, which are widely used in bridges, buildings, and industrial applications due to their high load-bearing efficiency. This thesis proposes a comprehensive methodology for the structural optimization of simply supported universal I-beams by integrating classical engineering calculations with machine learning, specifically using the Random Forest algorithm. A dataset of 96 universal I-beams manufactured in the UK is analyzed, containing essential parameters such as cross-sectional dimensions, mass per meter, and cost. For each beam, the plastic moment capacity and deflection under a 10-meter three-point loading scenario were computed using well-established structural formulas. These calculations were verified in both Python and Excel to ensure consistency and accuracy. A scoring system was developed to evaluate beam performance by integrating plastic moment, deflection, and cost, allowing objective comparison and ranking. The Random Forest model was trained on the dataset to predict plastic moment and deflection for hypothetical beam configurations outside the original database, generating new, optimized design options. Results demonstrate that this combined approach successfully identifies high-performing beams and captures the non-linear relationships between geometry and mechanical performance. The methodology offers a practical decision-support tool for engineers seeking efficient and economical structural solutions.