Research On Computer-aided Modeling Technique Based On Intelligent Computing

As a modern optimization method, the intelligent computation method has drawn much attention of researchers on theory and application. Typically, this type of method named metaheuristics, which includes evolutionary algorithm, swarm intelligence, neural networks, etc. The main characteristic of these algorithms is that they all construct their algorithm process by simulating the natural biological process and activity. For instance, they simulated the movement of the swarm animals or the rule of the evolutionary in the nature world. Take the evolutionary computation for example, the algorithm constains severl elements: the searching subject is population based; the solution space is mapped into the search space; the individuals of the poplution will be evaluated by the fitness function; and the new population is generated through the operator toward them. Compared to the triditonal optimization algorithm, this class of technique often acts with problems with high complexity nonlinearity, parallel and versatility. The main domain of application of these methods contains function optimization, machine learning, etc.Recently, with the deeply research on these algorithms'basic theory and the much more extensive application, more domains use this type of method as the effective tools. Intelligent computing based computer-aided design is the certain merging domain which has been the research hot spot for years. The process of design is the process of searching the optimal solution with some constraints. And the process is just like the optimization problem so that the intelligent computation can be used to solve the problems within it.Functionally, in the field of design, usage of the intelligent method could fall into two main categories: engineering optimization and design scheme generation with creativity. The first one, the engineering optimization problem, offen has the detailed problem description that includes the clear fitness function and optimal solution and value. The key aspects of these problems include how to design the effective algorithm to solve the large amount of data and increase algorithm effectiveness. The second one, design searching problem, offen has the fitness function more subjective and has flexible problem representation. The major function of this problem is to search the design scheme space to assist the designer to find the optimal solution without any fixed rules. Since the fitness function is hard to determine, how to set the convergence stardand is the key problem. Because of this reason, a new method is proposed name interactive evolutionary algorithm in which the human factor could directly set the fitness function. Since every individual of the solution must be evaluated by the human, the algorithm speed could be much lower than normal, thus make this method faile to handle large volume data.In the stages of design process, the way of representation of the solid object in computer is also an open problem. There are various ways of storing and editing the solid object in digital method. As the development and dependence of the the digitalization, many high abstracted approaches are well developed. For example, modeling with hand sketches, skeleton based modeling and fast modeling in an effective interactive context.This paper focus on the research of the intelligent computing based shape conceptual design. The study object is a kind of shapes of which the appearance can be represented by their skeletons. Firstly, an encoding method is proposed to represent the skeleton. Secondly, apply the genetic algorithm to the variation of the skeleton individual and construct the appropriate fitness fuction to control the skeleton's shape feature. Thirdly, cluster the searching space to increase the efficiency of the algorithm.The main content of this research includes:(1) Proposing a design-toward skeleton encoding method using the radius and angle value. The skeleton data, as a representation method of the solid shape, are usually used in 3d animation, pattern recognition and fast modeling domain. It is the main shape base of the implicit surface modeling method. However, there is little research on the skeleton itself as the main design object. According to the feature of the skeleton representation, the skeleton can be expressed by its length of line segment and angles of the joints. Through encoding the angle of the skeleton, a skeleton can be expressed simply and effectively. In addition, the location of the control points are easy to obtain and the angle limit is also simple to set. Since the lengths of the line segment are fixed, the angles can totally express the appearance feature of the original shape.(2) Applying the genetic algorithm to the assistant design problem using the skeleton encoding. To obtain the appropriate deisgn solution, the total skeleton appearance can be controlled by the setting the suitable fitness function. The function can be customized from various perspectives. One kind is setting the range of the angle of joints, through which the skeleton's certain joints could hold the value while others can vary. Another kind is controlling the locations of the skeleton's points, making them separate for a certain distance or meeting certain distribution. Experiment result show that the method can generate design schemes effectively.(3) By clustering the feasible solution set, the method can further increase the efficiency. This problem is a kind of optimization with multipl optimal solutions. There could be many or infinite optimal solutions named feasible solutions in a searching space with a certain fitness function. Therefore, the design solutions generated by the algorithm could repeat thus produce same design schemes. To avoid this phenomenon, a cluster operation toward the generated solutions is proposed. By the unsupervised clustering of the set, the algorithm can produce limited number of design solutions. Through this way, the system could provide more flexible support to assist the designer.Innovative viewpoints of this dissertation are summarized as follows:(1) An angle and radius based skeleton representation method toward design is proposed. By using this approach, skeleton can not only be represented effecitively, but the encoding sytle with that tends to be more sutible for the conceptual design using evolutionary computing method.(2) A method of genetic algorithm based skeleton design is proposed. Whithin this method, interactive evolutionary algorithm is improved and customized, and several constraining methods are proposed to help designers directly control the geometry feature of the skeleton. This method could generate design schemes customized by the human with high diversity and quantity.(3) Using the clustering method to the feasible solution set could improve the efficiency of the method. By analyzing the similar solutions which are created by the previous steps, a k-means clustering is used to reduce the volume of the feasible schemes, thus the number of human operations is also reduced and the whole algorithm's efficiency is increased.