prediction of hypervelocity impact of nanocomposite materials using hybrid artificial intelligence and physics based modeling
- Paper number
IAC-08.A6.3.8
- Author
Prof. Richard Donovan, Montana Tech of The University of Montana, United States
- Coauthor
Prof. Leo Daniel, Massachusetts Institute of Technology, United States
- Year
2008
- Abstract
Predicting hypervelocity impact response of nanocomposite structures requires coupling material response to multi-scale (time and space) multi-modal (continuum and particulate response) constitutive behavior. The development of effective numerical tools to efficiently bridge these scales and modes will play a critical role in the predicting the response of complex nanocomposite structures. This paper will detail development of just such a tool that combines physics based analytic tools such as ANSYS/AUTODYN, Artificial Neural Network (ANN) simulations and experimental results. In this paper, a full simulation using existing material models and data for nanocomposites and traditional graphite/epoxy has been developed using AUTODYN-2D and ANN software for prediction of normal impact velocities in the range of 2km/s to 10km/s. The challenge of multi-scale/multi-modal modeling posed by these problems is to reliably relate the results of physics based analytic tools (such as finite difference approximations of the wave equation) to field results that likely include the effects of multi-scale/multi-modal heterogeneity not present in the analytic result. Different impact test have been performed and then simulated numerically using Smooth Particle Hydrodynamics (SPH) -A Lagrangian gridless technique that uses a set of interpolation points to model a continuum. ANNs provide an effective interpolation between scales and modes. In addition, ANNs are utilized to “fuse” the results of physics based computational models with experimental/field data. Despite the structural complexity of nanocomposites and the difficulty in modeling the impact deformation, preliminary results from these predictions were in good agreement when compared to experimental data. Details of the modeling techniques are discussed and comparison is made with the test data. Future developments related to the need for and development of generalized analytic tools such as these will also be discussed.
- Abstract document
- Manuscript document
IAC-08.A6.3.8.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.