AI-Enabled Performance-Based Structural Optimization of Steel Truss and Mezzanine Systems for Industrial Facilities

Abstract

This study investigated AI-enabled, performance-based structural optimization of steel truss and mezzanine systems used in industrial facilities through a quantitative computational benchmarking design. Multiple optimization algorithms were evaluated under identical structural modeling assumptions, load combinations, and constraint conditions to determine their influence on material efficiency, serviceability control, stability adequacy, vibration response, and computational performance. The final analytic dataset consisted of 468 valid optimization runs derived from 16 standardized structural configurations, equally representing truss and mezzanine systems. Results demonstrated an overall mean weight reduction of 18.6% relative to baseline designs, with truss systems achieving higher reductions (20.1%) compared to mezzanine systems (17.0%). Differential Evolution and Hybrid Metaheuristic approaches produced significantly greater material savings than the reference Genetic Algorithm, with mean improvements exceeding 3.0 percentage points (p < 0.001). Dimensionality exerted the strongest influence on computational cost, as high-dimensional cases increased runtime by an average of 5.41 minutes and significantly raised convergence iterations (p < 0.001). The runtime regression model explained 56% of variance, while the weight reduction model explained 41% of variance. Feasibility performance remained stable across complexity levels, with an overall feasibility rate of 91.6%. No statistically significant differences were observed among algorithms for overall structural adequacy once feasible convergence was achieved, indicating consistent compliance with strength, serviceability, and stability requirements. Reliability analysis confirmed strong internal consistency for composite structural adequacy measures (α = 0.88) and acceptable consistency for optimization efficiency (α = 0.74), supporting inferential modeling validity. The findings demonstrated that algorithm selection and dimensional complexity significantly affected material efficiency and computational performance, while structural compliance remained consistently maintained. The study provided a statistically grounded benchmarking framework for evaluating AI-enabled optimization strategies in industrial steel structural systems.