The design of 2D rectangular steel trusses demands a critical balance between structural performance and constructability, a core challenge in civil engineering. Using distinct cross-sectional profiles minimizes material use but elevates structural complexity, whereas standardized profiles facilitate construction simplicity at the cost of efficiency. This thesis develops a multi-objective optimization framework to navigate these trade-offs, targeting four essential objectives: mass minimization to reduce material requirements, connection degree to simplify joint configurations, symmetry to enhance aesthetics and standardization, and beam continuity to streamline assembly processes. By controlling the number of unique HEA profiles, the study delivers tailored solutions for preliminary structural design, aligning with engineering priorities and stakeholder preferences to optimize truss performance and practicality.
A computational framework employs the Tree-structured Parzen Estimator (TPE), a sample-efficient Bayesian optimization method, to efficiently explore the complex, discrete design space of truss configurations. TPE performance is rigorously validated against exhaustive search (EXS) to ensure accuracy in identifying optimal designs. Stakeholder-defined weights, implemented through weighted scalarization, enable customized trade-off analyses, though without direct stakeholder engagement. This approach supports the exploration of diverse configurations, effectively balancing performance and standardization while addressing the computational demands of large search spaces, thus providing a robust tool for 2D truss optimization.
The findings indicate that intermediate profile grouping often produces designs that balance structural performance and constructability. The multi-parallel plot, a dynamic visualization tool, potentially empowers stakeholders, including engineers and project managers to transparently explore trade-offs, pending practical validation. Despite limitations, such as untuned TPE hyperparameters and a focus on 2D trusses, this promising framework enhances transparency and adaptability in preliminary structural design. By integrating efficient optimization with intuitive visualization, the study establishes a foundation for future advancements in steel truss optimization, offering a versatile methodology with potential to inform broader structural engineering applications.