Estimating the effects of removing negative features on engineering analysis

This paper provides a general framework for the quantitative estimation of the effects of removing negative features on engineering analysis, or modification sensitivity for short. There are two main applications: (i) when defeaturing models so that finite element analysis may be carried out more quickly and with lower memory requirements, and (ii) when performing iterative design based on finite element analysis. Our approach can handle large as well as small features, and features with Neumann/natural boundary conditions prescribed on them; previous methods have difficulties in handling such cases. Estimation of the modification sensitivity is achieved by reformulating it as a modeling error caused by use of different mathematical models to describe the same engineering analysis problem. Results are obtained using the dual weighted residual (DWR) method in combination with a heuristic assumption of small variation of dual solution after defeaturing. The final derived sensitivity estimator is expressed in terms of the difference of local boundary integrations over the feature boundary, which can be explicitly evaluated using solutions defined on the defeatured model. The algorithm’s performance is demonstrated using a Poisson equation. Comparisons to results obtained by previous approaches indicate it is both accurate and computationally efficient.