Inverse Issue Study Of Bird-Impact To Aircraft Windshield Based On Neural Network Method

With the development of the aircraft performance at low altitude and high speed, as well as the marked progress of eco-environmental protection, the issue of bird impact to aircraft has been concerned more and more. In the research of bird-impact to aircraft windshield, the experiment is always the ultimate and the most effective method. But the existing data of bird impact are highly disperse, so that they do less help for the design of windshield and also cost more to experimental work.In this thesis, a combined method of finite element numerical simulation with neural network is studied to obtain the dynamic response, the impetus forces and impact parameters of windshield during bird impact. The thesis presents A wavelet-dynamic-delay-feedback (WDDF) type of neural network and makes in-deep studies of the WDDF theoretical framework, learning theory and method, improved training effectiveness, power to extend, and technical implementation. The presented neural network takes strain-time data as input at two locations on the back of struck body, and gives highly precise outputs of the impact force-time data, impact kinetic energy and the coordinates of impact position. Also the network shows many advantages of high training efficiency, robustness and anti-jamming energy, as well as the extensive applicability to the academic study and practical requirement of bird impact issue.In the positive-issue research of bird impact, the thesis carefully explores the basic theories, finite element numerical method and key techniques of the solution process in LS-DYNA3D software system for contact-impact issue of viscoelastic bodies at large deformation. The well-chosen cases examine the accuracy of solutions in LS-DYNA3D for large deformation impact issues. Under the proof work, the incremental constitutive equation of bilinear and nonlinear viscoelastic materials are derived, coded and verified in LS-DYNA3D. On the basis of the above mentioned theoretical and case-computational work, the element types, structural dimensions, materials, boundary conditions and calculative effects of the finite element modeling (FEM) for bird-impact windshield are all carefully studied, and some practical experiments of bird-impact to windshield are made. It follows that the FEM established in this thesis is well agreed with the experimental results and also shows that it is a reliable and highly accurate model with satisfying the requirement of trainings to the WDDF neural network and engineering windshield design.The thesis systematically describes the detailed experiments of bird impact, including the major test principles and properties of the instruments used in the experimental system. The dynamic data of displacement, strain and impetus force acquired from the experimental collection system are also well analyzed formeasuring precision and errors. Based on the underlying work and experimental features, the thesis proposes a simple, practical and well accurate impetus resultant force fitting method, weighted factor fitting method. A line-off data processing software system for bird-impact experiment, named as CADABIE, is well designed under the principle of software engineering, and has been applied, with high praise, in bird-impact laboratory in the factory numbered 320 and Esfahan bird-impact laboratory of IRAN.According to all the research of this thesis, an improved and promising project of bird impact experiment combined with the WDDF neural network can be suggested, which will be high efficiency.