A generative sampling system for profile designs with shape constraints and user evaluation. (June 2019)

• Record Type: Journal Article
• Authors: Dogan, Kemal Mert; Suzuki, Hiromasa; Gunpinar, Erkan; Kim, Myung-Soo
• Is Part Of: Computer aided design. Volume 111(2019)
• Page Start: 93
• Page End: 112
• Publication Date: 2019-06
• Subjects: Computer aided design (CAD) -- Constraint development -- Shape similarity -- Generative design; Computer-aided design -- Periodicals; Engineering design -- Data processing -- Periodicals; Computer graphics -- Periodicals; Conception technique -- Informatique -- Périodiques; Infographie -- Périodiques; Computer graphics; Engineering design -- Data processing; Periodicals; Electronic journals; 620.00420285
• DOI: 10.1016/j.cad.2019.02.002
• Languages: English
• ISSNs: 0010-4485

Abstract

Abstract: A sampling approach is proposed for deriving profiles of an existing product design using profile similarities and primitive shapes, such as circles, triangles, and ellipses, as constraints. A common approach for a designer is to generate or collect sketches to create a design space. The design space can be explored to retrieve the samples directly based on requirements or crossing over them to obtain new design variations. In this stage, the proposed method can be used to derive a large number of samples from a design image to work with. First, the user clicks on the image to define each design feature as cubic Bézier curve segment. The primitive shapes are then constructed for each segment and used as constraints, such that the control points can move only to where these shapes allow. Moreover, the design similarities are computed using the triangles based on their anisotropy ratios, which are measures of the deformation between corresponding triangles and which are used to ensure that they are highly related samples in the example design. Modified Hausdorff distances are also computed between the biarc approximations of the samples. There is diversity provided that these distances are large enough. A customized sampling algorithm that fulfills our constraints is executed several times synchronously via parallel programming to create a design space. Finally, the user-specified number of distinct samples is retrieved from the design space by minimizing the Audze–Eglais potential energy. We provide an additional tool with which the users can adjust the weights for each control point to guide the sampling process and record the chosen samples. The main goal of this study is to generate samples, which are diverse but still convey the key features of the supplied shape. Using the settings related to similarity, a designer can derive samples sticking to their initial idea to explore its better version or more creative results that they could not imagine on their own. The proposed system offers simplicity automating every step that must be done by the user. Thus even inexperienced users without needing any programming skill can vary shapes defined using a set of points in a matter of seconds, and experienced users can even focus on specific shape features weighting them without dealing any constraint or rule definition requirements. Besides, the proposed system allows partial modifications, in which users can select a specific region to modify while the rest of the shape maintains the original appearance and functionality of the exemplary design. Highlights: Diverse sampling of profile curves using similarity measures. Efficient sampling from single image using parallel chain algorithm. Comprehensive representation of constraints using primitive geometries. Easy to use prototype system for inexperienced users.