PSI - Issue 66

Umberto De Maio et al. / Procedia Structural Integrity 66 (2024) 502–510 Author name / Structural Integrity Procedia 00 (2025) 000–000

508

7

Fig. 4. A convolutional neural network (CNN) architecture used for classification tasks, particularly for analyzing the first mode shape in structural analysis or vibration studies.

In Fig. 4, the results obtained from the numerical simulations and reported in terms of normalized effective stress versus strains along the vertical direction, highlighted a significant improvement in mechanical performance. Specifically, the critical buckling strain increased from 0.0328 percent to 0.0543, corresponding to an increase of about 65%, while the critical buckling load showed an increase of 34.6%. Definitively, the improved performances demonstrate the effectiveness of the new optimized geometry and highlight the great capabilities of integrating CNN and genetic algorithms to improve the performance of bioinspired materials. 4. Conclusions Inspired by the deep-sea glass sponge Euplectella aspergillum , this work aimed to develop and optimize a resilient lattice microstructure to improve its mechanical stability under uniaxial compression along the vertical direction. An advanced optimization algorithm was proposed in a computational framework that combined COMSOL Multiphysics models with MATLAB codes. An opportunely trained Convolutional Neural network was used in parallel with a genetic algorithm procedure to effectively predict the class of the instability modes (local, global, or combined), penalizing the microstructural configurations giving the onset of the global instability. The optimization procedure was performed by varying two dimensionless parameters related to the ratios between the thickness of the circular and elliptical features and the thickness of the square frame, maintaining a fixed value of solid volume fraction for each investigated combination of parameters. The new proposed design of the glass-sponge microstructure showed significant mechanical improvements, such as a 40% increase in buckling load capacity and a 70% increase in ultimate buckling deformation, demonstrating the effectiveness of the bioinspired material in creating advanced microstructural metamaterials with superior mechanical performance. The study also highlighted the great potential of combining genetic algorithms and neural networks as powerful tools for optimizing complex engineering structures, particularly in the context of bioinspired materials. Innovative design conceptualization offers a new perspective on the application of bioinspired architectures, providing a solid foundation for future research into high-performance and lightweight materials with enhanced stability under extreme conditions.