On the use of Machine Learning to Defeature CAD Models for Simulation

Abstract

International audienceNumerical simulations play more and more important role in product development cycles and are increasingly complex, realistic and varied. CAD models must be adapted to each simulation case to ensure the quality and reliability of the results. The defeaturing is one of the key steps for preparing digital model to a simulation. It requires a great skill and a deep expertise to foresee which features have to be preserved and which features can be simplified. This expertise is often not well developed and strongly depends of the simulation context. In this paper, we propose an approach that uses machine learning techniques to identify rules driving the defeaturing step. The expertise knowledge is supposed to be embedded in a set of configurations that form the basis to develop the processes and find the rules. For this, we propose a method to define the appropriate data models used as inputs and outputs of the learning techniques

Capturing simulation intent in an ontology: CAD and CAE integration application

“…The review by (Thakur, Banerjee, and Gupta 2009) presents a complete list of simplification techniques employed to simplify CAD models. We can add to this review the work of (Danglade, Pernot, and Véron 2014;Danglade et al 2017) which uses artificial intelligence techniques to defeature CAD models for simulation and to estimate the impact of model simplification on the analysis results. AI technologies are promising but still face challenges in the number of available examples to learn from, in the number of different requirements posed by the various simulation domains and in the efficient processing of complex CAD shapes.…”

A Tool for Idealisation of Cad Models

“…Hamdi et al (2012) proposed a hybrid method based on a combination of the elimination details and merging faces. Danglade et al (2013) proposed an approach that uses machine learning techniques for identification of detailed features to be suppressed. Lee (2005) proposed the multi-resolution models of a solid model simplified at various levels of details (LOD).…”

Second‐order defeaturing error estimation for multiple boundary features

“…However, heuristic estimators have been proposed for positive features, and other estimators for negative boundary features [11,13,15,21]. Ferrandes et al [21] gave an overall framework for adaptively generating a simplified model suitable for analysis based on heuristic error estimators, while further approaches to selecting features for removal have been based solely on geometric criteria [7,22,23] or machine learning [24,25].…”