Fitting a Parametric Model to a Cloud of Points via Optimization Methods

摘要

Computer Aided Design (CAD) is a powerful tool for designing parametric geometry. However, many CAD models of current configurations are constructed in previous generations of CAD systems, which represent the configuration simply as a collection of surfaces instead of as a parametrized solid model. But since many modern analysis techniques take advantage of a parametrization, one often has to re-engineer the configuration into a parametric model. The objective here is to generate an efficient, robust, and accurate method for fitting parametric models to a cloud of points. The process uses a gradient-based optimization technique, which is applied to the whole cloud, without the need to segment or classify the points in the cloud a priori..;First, for the points associated with any component, a variant of the Levenberg-Marquardt gradient-based optimization method (ILM) is used to find the set of model parameters that minimizes the least-square errors between the model and the points. The efficiency of the ILM algorithm is greatly improved through the use of analytic geometric sensitivities and sparse matrix techniques. Second, for cases in which one does not know a priori the correspondences between points in the cloud and the geometry model's components, an efficient initialization and classification algorithm is introduced. While this technique works well once the configuration is close enough, it occasionally fails when the initial parametrized configuration is too far from the cloud of points. To circumvent this problem, the objective function is modified, which has yielded good results for all cases tested.;This technique is applied to a series of increasingly complex configurations. The final configuration represents a full transport aircraft configuration, with a wing, fuselage, empennage, and engines. Although only applied to aerospace applications, the technique is general enough to be applicable in any domain for which basic parametrized models are available.