Hybrid Topology Optimization of a Concrete Structure via Finite Element Analysis and Deep Learning Surrogates

Abstract

The cement industry significantly contributes to global CO2 emissions, making material efficiency in concrete structures a crucial sustainability goal. This study addresses the challenge of excessive cement usage in traditional concrete design by optimizing a cast-in-place concrete bench. A density-based topology optimization framework was implemented in ANSYS Mechanical and enhanced with a deep-learning surrogate model to accelerate computational performance. The optimization aimed to minimize the structural mass while satisfying serviceability and strength constraints, including limits on displacement and compressive stress under realistic public-use loading conditions. The topology optimization converged after 62 iterations, achieving a 46% reduction in mass (from 258.3 kg to 139.4 kg) while maintaining a maximum deflection below 2 mm and a maximum compressive stress of 15.5 MPa, within the allowable limit for C20/25 concrete. The deep-learning surrogate model achieved strong predictive accuracy (IoU = 0.75, Dice = 0.73) and reduced computation time by over 105× compared to the full finite element optimization. The optimized geometry was reconstructed and rendered using Blender for visualization. These results highlight the potential of combining topology optimization and machine learning to reduce material use, enhance structural efficiency, and support sustainable practices in concrete construction.