| Summary: | 碩士 === 國立雲林科技大學 === 工業工程與管理研究所碩士班 === 95 === Due to the light, short and multi-function development trend, particles with 1~100 nanometer (nm) grain size become new advanced materials. We usually use wet-type mechanical milling process to make nano-particles. As the grain size of nano-particles decreases, the surface area is increased and the benefit of nano-particle powders increases. So keeping the particle size and size variations small becomes an important issue. Regarding to the optimal parameter design, most engineers use Taguchi methods to investigate the parameter design. It is known that Taguchi method only useful for single quality characteristic. But this research investigates the quality characteristics of particle size and particle variations. Therefore, the purpose of this study is to find out the optimal milling parameter and to optimize the two objectives simultaneously.
First of all, study was the multi-attribution decision making methods combine multi-quality characteristics into single quality characteristic to get several better parameter set. The methods include simple additive weight method (SAW), grey correlation analysis and technique for order performance by similarity to ideal solution (TOPSIS). In order to more precisely get the global optimal solution, we use Adaptive Network- Based Fuzzy Inference System (ANFIS) to learn the result of Taguchi experiment, and use it to simulate the input-output relationship of this system and then use it as the objective function in evolution algorithm. Then, we use non-dominated sorting genetic algorithm (NSGAII) to search the value between levels of Taguchi experiment. In addition, three improved ideas are proposed in this research. First, we apply the concept of contribution rate in Taguchi method to refine the crossover and mutation ways in NSGAII algorithm. Second, we adjust the crossover and mutation ratios in the evolution of each generation by using fuzzy control theory. Third, we use the results of the multi-attribution decision making as the initial solutions in the NSGAII algorithm. After the test and verification, we propose the adaptive NSGAII to search the global optimal solutions.
Under the same population size 80 and evolutional generations 50, the coverage of NSGAII is 0.375, and the coverage of the adaptive NSGAII which we proposed is 0.238. So the adaptive NSGAII compared with NSGAII can find the better optimal solutions that are much closer to the Pareto front.
|
|---|
