Semantic model difference identification for aerospace sheet metal part CAD models

Abstract

Many product lifecycle management’s processes like information reuse can take a lot of advantage from a semantic model difference identification (MDI). A semantic MDI can be defined as a solution that identifies and represents the differences between two compared models in terms of meaningful engineering information – for instance: this hole was moved. Semantic difference identification is especially challenging due to the variety of modeling solutions, the non-uniqueness of modeling sequences and the use of low-level information in engineering communications. This work proposes an MDI method that identifies and represents the differences between 3D Computer-Aided Design (CAD) models based on engineering semantics through features. Brake- and hydro-formed aerospace structural sheet metal parts are used as the application domain in which to propose and illustrate the method. Because engineering communications of 3D models are commonly in low-level information i.e., B-rep models, an automated feature recognition (AFR) method is first needed. An AFR solution elevates semanticity of engineering communication information without being reliant on the suboptimal modeling solutions. Although structural sheet metal parts form a significant portion of airplanes, there is no specialized AFR method dedicated to them. Despite the presence of several AFR methods for sheet metal parts, none of them are tuned to recognize the design semantics of the aerospace industry. This work proposes the first AFR method to recognize aerospace sheet metal features and design semantics to elevate the level of abstraction of the information from 3D STEP models. It starts with preprocessing the 3D STEP model in order to classify the topological elements of the B-rep models and create relevant novel face sets and subtypes of faces, face boundaries and edges. Then, rule-based steps are used to recognize aerospace sheet metal features. The extracted features are described by their geometry, their relationship with other features and their pertinent parameters. Once the features are recognized, model difference identification is performed. This work’s MDI method consists mainly of a pose registration stage and a difference identification stage. The pose registration method exploits the fact that all the features of a part serve specific functions, some of which are fundamental to the part’s essential functionality, and that they are intertwined with the design intent of the part, which is particularly true for aerospace sheet metal parts. This provides the opportunity to semantically register feature-based 3D CAD models according to the unique purpose of the features in this specific domain of application. Difference identification is approached by primarily identifying and segregating the commonality between the compared 3D CAD models and then identifying the differences. The differences between 3D CAD models are classified as added, removed or differed features. The proposed MDI method describes a way to fully pose-register 3D CAD models and identify their differences semantically based solely on their features, and, by extension, their design intent. Both the AFR and MDI methods are implemented to be validated. The prototypes are modified implementation of the original methods and the differences are outlined and pointed out. To validate the AFR method and verify its correct implementation, a collection of 26 real-world aerospace structural sheet metal parts was used to create CAD models that were subsequently converted to STEP models. The results show perfect accuracy and confirm that AFR works for this domain of application. To validate the MDI method and verify its correct implementation, 3 of the real-world ASM parts used in testing AFR prototype were modified to reflect the possible differences that could occur between similar parts in real-world scenarios. The results show perfect accuracy and confirm there is great potential for further development of MDI algorithms for feature-based models of parts from specialized domains of application.

Date	23 Feb 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Louis Rivest (Supervisor)