| Engineering design problems generally involve several design objectives. These objectives, related to the technical and economical performance of the engineering system, are potentially conflicting in nature. Multidisciplinary Design Optimization (MDO) has grown to the point of gaining near universal recognition in its ability to lead to better designs, both in the areas of academic research and practical application. At the heart of MDO problem is a multiobjective optimization problem. Therefore, it is really valuable to research multiobjective optimization theory and technologies.This research focuses on synthesizing advanced multiobjective optimization such as Physical Programming with Artificial Neural Networks (ANN), fuzzy technology and interactive design technology to develop more effective multiobjective optimization methods: developed Fuzzy Physical Programming, which could deal with the fuzzy uncertainties in engineering design problem more effectively; developed ANN-based Interactive Physical Programming, which is applicable and effective in large scale complex engineering multiobjective optimization problems; based on the Model Parameter Vector concept and the ANN model of the designer's preference structure, this thesis developed Intelligent Interactive Multiobjective Optimization Method (IIMOM), providing a complete and effective solution for the medium and small scale multiobjective optimization problems.Established the first research group researching Physical Programming outside of USA. Researched the fundamental theories and the effectiveness of Physical Programming, followed the research work of the leading researchers in this area and learned a lot from the discussions with them. The author believes that Physical Programming is quite valuable in both theory and practical applications, and further researches and wider applications are needed.Based on Physical Programming, this paper developed Fuzzy Physical Programming so as to deal with the fuzzy uncertainties involved in multiobjective engineering design problems. Fuzzy Physical Programming integrates the objective function with the corresponding variation of a design objective and calculates its preference function value. This is the point where Fuzzy Physical Programming is superior to robust design, which considers the objective function value and the corresponding variation seperatelly.Developed ANN-based Interactive Physical Programming. The approximate model of Pareto surface at a given Pareto design is developed using ANN for design exploration. A map from Pareto designs to their corresponding evaluation value is established using an ANN model, which indicates the decision maker's local preference at the given Pareto design. The combination of the ANN model of the designer's preference and the optimization procedure in this method generates the intelligent guiding for the multiobjective optimization process, and Compromise Programming is also used to find the final solution. This method seeks a compromise between the effectiveness and computation complexity of large scale complexmultiobjective optimization problems, and it turns out to be an effective and practical method for this kind of problems.In the Standard Interactive Physical Programming procedure, the ability of Physical programming to reach each point of the Pareto surface evenly is used, and the interaction of the design procedure is conducted by modifying the boundary values of the design objectives' preference functions. The method confirms the feasibility and effectiveness of developing a true interactive method based on Physical Programming by modifying the boundary values of the design objectives' preference functions.Developed IIMOM, providing a complete and effective solution for the medium and small scale multiobjective optimization problems. In IIMOM, the general concept of model parameter vector, which refers to the parameter vector determined by the designer in the multi-objective optimization model (such as the weight vector in the weight... |
PDF Full Text Request