Development of a prediction model using fully connected neural networks in the analysis of composite structures under bird strike

Hasilci Z., BOĞOÇLU M. E., DALKILIÇ A. S., KAYRAN A.

JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY, vol.36, no.2, pp.709-722, 2022 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 36 Issue: 2
  • Publication Date: 2022
  • Doi Number: 10.1007/s12206-022-0119-5
  • Journal Name: JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Compendex, INSPEC, Metadex, Civil Engineering Abstracts
  • Page Numbers: pp.709-722
  • Keywords: Bird strike, Composite design, Deep learning, FCNNs (fully connected neural networks), SPH (smoothed particle hydrodynamics), Stacking sequence, STACKING-SEQUENCE, GENETIC ALGORITHM, IMPACT, SIMULATION, DESIGN
  • Yıldız Technical University Affiliated: Yes

Abstract

Bird strike is one of the most hazardous issues facing global aviation. In the present study, a hybrid methodology is developed utilizing, automated data generation and fully connected neural networks to practically and reliably obtain the global deformation of composite structures subject to bird strike. The validation of the proposed numerical bird strike model is accomplished by making comparisons with the available experimental data from three different resources in the literature on the chicken and gelatine strike tests to a rigid plate, and strike test against an aircraft composite vertical leading edge. For three different bird velocities, 9402 input files are created by an automatic data generator considering all possible stacking sequence combinations in accordance with the composite design guidelines. The global deformation of composite laminates caused by bird strike is estimated via the fully connected neural networks established. Results of the present study show that with the use of fully connected neural networks, global deformation of the composite laminate can be estimated reliably and preliminary design of the composite laminate can be performed very fast compared to performing nonlinear finite element analysis involving bird strike. In conclusion, the fully connected neural network model is found to be an alternative for additional LS-Dyna simulations in the optimization process.