Kenol Jules

Fast Multidisciplinary Design Optimization via Taguchi Methods and Soft Computing

It is difficult to perform multidisciplinary design optimization using traditional searchbased optimization techniques due to possible conflicts among objectives from different disciplines,thetimeconsumingsearch,andthepossibilityofdivergence.Toovercomethedifficulties, this paper presents simulation-based design optimization techniques using Taguchi methods and soft computing (i.e. fuzzy logic and neural networks). An aircraft engine cycle design optimization with four conflicting design objectives is used to validate the presented approach. The result shows significant performance improvement in optimizing single and multiple design objectives. HE emerging field of Multidisciplinary Design Optimization (MDO) seeks to improve design methodology to rapidly and efficiently explore multiple-dimension design spaces with the goal of increasing system performance significantly, thereby reducing end-product cost substantially. Search-based and simulation-based are the two major system design approaches. The former is traditional and mathematical, and has existed for a long time. The optimum solution has to do with the selected starting point, and the optimization method used. A possibility of divergence in solution seeking is a major drawback in this approach. In contrast, the simulation-based approach uses the analysis and evaluation of a candidate solution, and the assessment of the degree to which the candidate satisfies the requirement. This optimum design tool uses the simulation-based approach. 1 With this new approach, the optimum solution can be obtained in real time. The traditional search-based optimization is a typical example of hard computing. In hard computing, the prime desiderata are precision, certainty and vigor. In contrast, in soft computing the principal notion is that precision and certainty carry a cost; and that computation, reasoning, and decision-making should exploit (whenever possible) the tolerance for imprecision, uncertainty, approximate reasoning, and partial truth for obtaining low cost solutions. Fuzzy logic and neural networks, the two major soft computing techniques, have very contrasting application requirements. Fuzzy systems are appropriate if sufficient expert knowledge about the process is available, while neural systems are useful if sufficient data are available or measurable. Furthermore, neural networks possess the ability to learn the input-output relationship. A trained neural network provides instantly input-to-output mapping with reasonably good accuracy, but without knowledge representation. Fuzzy logic, on the other hand, possesses the ability for knowledge representation and inference, but has no capabilities for automated learning. Thus, fuzzy logic and neural networks compensate each other in terms of information processing.

Optimized multidisciplinary system design for aircraft and propulsion systems

The optimized system design methodology for multiple disciplines is presented. Taguchi techniques, fuzzy logic and neural networks are combined to develop a user-friendly design tool, which can perform fastoptimized multidisciplinary system designs and analyses. Taguchi techniques are used to identify the input factors or parameters that are significant to the output of a system. From the Taguchi analysis, parameter significance index is obtained. This index is then used in a fuzzy inference system to find the optimum solution with multiple design objectives. Then, a compromised optimum solution is obtained at the multidisciplinary system level where the design objectives of the system as a whole are considered. Neural Networks are built to provide fast mappings for analysis and design. The application of this design methodology to aircraft and propulsion systems show significant improvement in optimizing the design objectives, and in predicting the system output in real time within reasonable accuracy.