| Summary: | 碩士 === 國立成功大學 === 機械工程學系 === 89 === The Study of Applying Artificial Intelligence Methods to Product Design
Abstract
The main structure of this thesis is divided into two parts. The first to be introduced is the study of genetic algorithm in modular design. The second is the Kansei image analysis of applying neural network in product design.
The main problem of modular design is that the internal system components are lack of sensible and practical mechanism during the module-synthesizing process. The modular design which is introduced in most theories and studies only focuses on setting up related measuring-indicator; as regards the modular-differentiating process, charts are still mainly used for assisting human. Consequently, on the one hand, it is not objective, on the other hand, it is not efficient, naturally, the utility value of modular design will be reduced. This thesis is employing genetic algorithm in modular group-works. First of all, it will be based on objective and functional indicator to define the interaction between each functional part of products and will use integral genetic coding string to represent the module number in each internal system component; then under the principle of “maximize the internal module interaction; minimize the interaction of module interface”, it will combine objective function of interaction and the constraints of module interface and will utilize the genetic algorithm to perform an optimum module arrangement search. Then, by automation of design, the efficiency and utility value of modular design will be increased.
Traditionally, Kansei engineering uses statistic technology to analyze the relation between engineering design and consumer’s Kansei image. However, human thinking has special parallel and nonlinear qualities, when statistic theories transform collected data into mathematical formulas, it will easily cause some losses of information, moreover, the complex mathematical operation will not be able to keep up with rapidly changeable Kansei information, either. Accordingly, this thesis utilizes separately the self-organizing map network and back-propagation network to abstract Kansei conceptual adjectives and analyze the relation between designed categories and Kansei image. In addition to this, this thesis investigates whether Neural Network, as the main deduction mechanism for Kansei engineering, can be effectively close to human thinking patterns and can therefore establish a simple and highly efficient Kansei design procedure.
|
|---|
