Research On Quantitative Estimation Of The Impacts Of Defeaturing To Nonlinear Engineering Analysis

In order to improve the efficiency and quality of mesh generation to meet the demand for high-end engineering simulation, defeaturing, or detail removal, is usually involved in the process of seamless CAD/CAE integration. It is still an open question to get a simplified geometry via defeaturing without unduly affecting analysis results.The key to get analysis-reliable simplified geometry is to quantitatively characterize the importance of suppressing a single feature to engineering simulation results (referred to as defeaturing error). Thus way, the importance of the feature can be estimated. In this thesis, novel approaches to quantitatively estimating the importance of suppressing negative features to engineering analysis is proposed and implemented for some nonlinear elliptic equations, respectively the nonlinear Poisson issues and Navier-Stokes equation. This proposed approach can’be applied to very general cases:the features to be suppressed may lie within the model’s interior or along its boundary, and may be constrained with either Neumann or Dirichlet boundary conditions.The defeaturing error is measured via changes of local quantities of engineering interest, or called goal-oriented error, which is more general in engineering. The error is estimated in a posteriori way without using solution of the original complex model. The final error estimator is mainly achieved by using the classic adjoint theory. Through rigorous mathematical derivations, the defeaturing error is first reformulated into a local quantity defined over the boundary of the feature to be suppressed, via Taylor expansion and Green’s theorem. However, the derived error expression still involves unknown terms, which are overcome using heuristic approaches via a further local computation around the features to be suppressed. The overall framework is demonstrated via two classes of typical nonlinear equations:nonlinear Poisson equation and Navier-Stokes equation. Compared with the former, the latter equation involves vector derivations and is more complex.In the end, the performance and effectivity of these approaches are tested on2D and3D internal and boundary features, with either Neumann or Dirichlet boundary conditions. The numerical results show that our approaches are feasible to estimating the defeaturing error, and effectively measure the importance of feature to engineering analysis.