Data-driven surrogate model-assisted topology optimization of GLARE fiber metal laminates for superior tensile and flexural performance

Abstract

This study aims to develop and validate an integrated experimental–computational framework that combines physical testing, topology optimization, and a data-driven surrogate model to enhance the mechanical performance of glass laminate aluminum reinforced epoxy (GLARE) fiber metal laminates. The methodological approach comprises three core components. First, we fabricated and mechanically characterized GLARE laminates containing 1, 2, 4, 6, and 8 glass fiber–reinforced polymer (GRP) layers under both tensile (ASTM D3039) and flexural (ASTM D790) loading. Second, we applied a solid isotropic material with penalization (SIMP)-based topology optimization to a cantilever beam subjected to combined tensile and bending loads, allowing us to interpret the underlying experimental trends. Third, we developed a quadratic polynomial regression surrogate model trained exclusively on our experimental data for rapid property prediction. Our principal findings reveal a non-monotonic tensile response, where the strength increases from 70.8 MPa to a peak of 170 MPa at 6 layers—representing an approximate 140% improvement—followed by a reduction to 136 MPa at 8 layers due to interfacial delamination. In contrast, the flexural strength increases monotonically and substantially, reaching 322 MPa at 8 layers, which is an approximate 225% improvement. The topology optimization produces a diagonal truss-like material layout with discrete load paths. This layout provides a clear physical explanation for both the continuous flexural enhancement and the existence of an optimal tensile configuration. Furthermore, the surrogate model achieves high predictive accuracy, keeping errors below 6% for tensile strength and below 5% for flexural strength. The unique contribution of this work is the unified integration of delamination-aware experimental data, dual-loading SIMP topology optimization, and a computationally efficient quadratic surrogate model—a combination not previously reported for GLARE laminates. Ultimately, this framework establishes a direct, quantifiable link between experimental characterization, topology optimization, and data-driven modeling, offering an effective tool for advanced composite engineering.