Design optimization of bio-inspired 3D printing by machine learning

In this study, the stiffener geometry was optimized using curvilinear 3D printing to enhance the buckling resistance. A bio-inspired skin/stiffener composite that mimicked spider-web structures was generated. A dataset was formulated for the regression analysis, covering buckling stresses under distinct feature values. The regression equations, crafted using a deep neural network trained on the dataset, were evaluated. The derived regression equation was subjected to sequential quadratic programming, a mathematical optimization, to determine the optimal value of the explanatory variable. This was aimed at maximizing the buckling stress-to-stiffener volume ratio, which is the objective variable. The optimized arrangement exhibited significantly improved buckling resistance, with approximately 163% higher buckling stress than conventionally designed structures with straight stiffeners of similar weight.