• Generative AI-driven SIMP topology optimization achieves 70% mass reduction in aerospace drones • Density-based FEA yields 83.94% compliance reduction while maintaining safety factor of 2.0 • 16 optimized designs validated across three materials and manufacturing constraint scenarios • Machine learning regression models confirm optimization accuracy with R² values exceeding 0.95 • Aluminum alloys achieve optimal mass efficiency (0.08 kg) for weight-critical applications The paper introduces a topology optimization framework for the structural design of aerospace drone components, where density, based topology optimization is integrated with exploratory machine learning for post, optimization performance analysis. Through density-based topology optimization using SIMP methodology, 16 design results were created for the combination of three materials (Aluminum 6061, Cobalt Chrome, Aluminum AlSi10Mg) and three manufacturing methods (3-axis milling, additive manufacturing, unrestricted). The optimization resulted in a reduction of compliance by 83.94% while the volume was reduced by 70% with a safety factor of 2.0 being maintained. The von Mises stress (13.38-15.71 MPa) and displacement (0.03-0.11 mm) of all designs were very close to each other and within the acceptable aerospace limits. In order to assess the practicability of surrogate, based performance prediction, six machine learning regression models (Linear Regression, Random Forest, SVR, Gradient Boosting, XGBoost, and Decision Tree) were used in an exploratory way; however, the findings serve only as a demonstration of the concept since the dataset size was limited. Aluminum alloys achieved the top mass efficiency (0.08 kg) for weight, critical applications, while Cobalt Chrome provided high safety factors (37.30, 43.44) for heavily stressed parts. The results show a significant weight reduction for the single static load case considered. On the other hand, the paper also acknowledges that multi, load and dynamic scenarios are required for real aerospace validation.
R et al. (Sun,) studied this question.