Geri Dön

Ecient generative methods for assisting aestheticdesign in conceptual stage

コンセプト段階での意匠設計支援のための効率的なジェネラティブデザイン手法

  1. Tez No: 608097
  2. Yazar: KEMAL MERT DOĞAN
  3. Danışmanlar: PROF. DR. SUZUKİ
  4. Tez Türü: Doktora
  5. Konular: Endüstri Ürünleri Tasarımı, Industrial Design
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2019
  8. Dil: İngilizce
  9. Üniversite: The University of Tokyo (Tokyo Daigaku)
  10. Enstitü: Yurtdışı Enstitü
  11. Ana Bilim Dalı: Belirtilmemiş.
  12. Bilim Dalı: Belirtilmemiş.
  13. Sayfa Sayısı: 158

Özet

Özet yok.

Özet (Çeviri)

Common product design processes usually start with problem definition, then generate and evaluate solutions for the defined problem, and finally detail and test the promising solutions until deciding on a final design. The conceptual design is an early stage of the design process, where possible solutions to the design problem are generated in abstract forms. The solutions are evaluated based on engineering and industrial design requirements, which a product must satisfy. To reach aesthetically plausible shapes for target customers without compromising engineering requirements or causing extra costs, many design samples usually need to be generated until finding an optimal solution if there is not any guidance for designers to use in addition to their limited imagination. In the first part of this thesis, design parameters are screened to find important ones for an aesthetic objective. Therefore, the desired design solutions can eciently be created as there will be a smaller number of design parameters to work with. In the second part, a generative sampling method is proposed to derive many samples from a single exemplary design using computer power with minimum possible user e ort. Using this method, the initial idea can be varied quickly, which is useful for those who need to create and examine many design solutions to find better alternative or get inspiration from them. The first approach uses a design concept from an earlier study called adjective-based design in which adjectives are used to describe product designs that are represented by geometric design parameters. In this earlier study, human preferences are collected by a survey method and converted into mathematical models, where variables are the design parameters, using a machine learning method to be used for quantitative judgments of aesthetics for a new generated design. Nevertheless, some design parameters were absent in the mathematical models although they were in the learning process due to nonexistent or weak correlations between them and participant responses. Getting rid of such irrelevant parameters, the reliability of the models can be enhanced as well as increasing sampling eciency thanks to the knowledge of the relevant parameters to the objectives. In Chapter 2, the survey method is supported by visual evaluations using eye tracking technology so that the design parameters can be screened based on their attractiveness to the participants. The main advantage of eye tracking in a survey is that the collected human perception data is more likely to be accurate and objective compared to the conventional surveys since the evaluation is based on survey participants' attention rather than applying solely questionnaires that ii cannot tell much about reasons of participants' choices. Eye tracking tools such as area of interest (AOI), gaze plot, and heat map are used to evaluate the attractiveness of the design parameters. Finally, a regression analysis method is used to find relations between attractive design parameters and the adjectives based on the gaze data. Aesthetic design ideas are mostly developed by sketches on paper rather than using the generative power of computer environments due to time taking requirements. The second approach is a generative sampling method that derives design samples from an example design by auto constraining and using some conditions extracted from its profile curve to minimize the e orts needed from users. As a first step, a composite profile curve of an example design is defined using a number of control points that should be enough to represent each shape feature by a cubic B´ezier curve segment. Primitive shapes such as triangles and circles are then constructed for each segment and used as constraints to prevent generating infeasible shapes. Moreover, profile similarities are computed using the triangles based on their anisotropy ratios to preserve key shape features of the original profile and modified Hausdor distances between the control points to make sure the modification is made suciently while a new profile is generated for the sake of diversity. A customized sampling algorithm that fulfills the segment constraints and the similarity requirements is executed several times synchronously via parallel programming to create numerous samples. Using the settings related to similarity, it is possible to derive samples sticking to an initial idea to explore its better version or more creative results that they could not be imagined even by designers. The proposed method is carried on with a software that o ers simplicity automating required steps to generate samples. An additional tool is also provided in this software with which weights can be adjusted for each control point to guide the sampling process and the chosen samples with their choices can be recorded to be analyzed later. This system also includes a sample management interface where additional geometric constraints can be defined to filter out undesired samples respect to these constraints. To sum up, two approaches are studied to increase the eciency of generating aesthetic ideas. The first approach aims to identify important design parameters for specific aesthetic objectives based on customer feelings so that new designs can be generated with fewer parameters that have direct influences on the target aesthetic objectives. The second approach is for generating many design samples with as minimum as possible e ort, time and experience requirements for various profile curves. Finally, problem-specific geometric constraints can be utilized through the sample management system to filter out undesired solutions from the generated samples.

Benzer Tezler

  1. A space subdivision framework for virtual reality and haptic applications

    Sanal gerçeklik ve haptik uygulamalar için sekizli-ağaç tabanlı bölümleme altyapısı

    ENGİN DENİZ DİKTAŞ

    Yüksek Lisans

    İngilizce

    İngilizce

    2006

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolBoğaziçi Üniversitesi

    Sistem ve Kontrol Mühendisliği Ana Bilim Dalı

    DR. ALİ VAHİT ŞAHİNER

  2. Development of strategies for reducing the worst-case message response times on the controller area network

    Denetleyici alan ağı (CAN) üzerindeki en kötü durumdaki mesaj tepki sürelerini azaltmak için stratejiler geliştirme

    VAKKAS ÇELİK

    Yüksek Lisans

    İngilizce

    İngilizce

    2012

    Elektrik ve Elektronik MühendisliğiOrta Doğu Teknik Üniversitesi

    Elektrik ve Elektronik Mühendisliği Bölümü

    DOÇ. DR. ŞENAN ECE SCHMİDT

    YRD. DOÇ. DR. KLAUS SCHMİDT

  3. Haptic rendering of continuous parametric models

    Sürekli parametrik modellerde haptik gerçekleme

    MELDA ULUSOY

    Yüksek Lisans

    İngilizce

    İngilizce

    2010

    Makine MühendisliğiSabancı Üniversitesi

    Mekatronik Mühendisliği Ana Bilim Dalı

    YRD. DOÇ. DR. VOLKAN PATOĞLU