The study on the optimization of the patameters in plasma welding -the application of Taguchi method and Artificial Neural Network.

• Main Authors: Li-Chang Hsu, 徐立章
• Other Authors: Sheng-Chai Chi
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/94042651842406966719

碩士 === 義守大學 === 管理科學研究所 === 87 === According to articles, experiments, and industrial applications, it has been proven that quality of Plasma Arc Welding (PAW) is much better than that of well-known TIG welding. The reason is that with the fantastic feature of highly concentrated energy, PAW performs the effect of "keyhole" which penetrates material deeply. The appearance of keyhole narrows the area of material influenced by heat during the welding process and reduces damage of material structure caused by heat stress and compression stress in the period of cooling. The welding quality of PAW is really remarkable. However, the application of PAW technology in our country is still in the development stage, not in the matured stage. Taguchi method, a popular experimental design method applied in the industry, can improve the disadvantage of full-factor experiments in the conventional experimental design. It approaches the optimization of parameter design, though the number of experiments is reduced. Nevertheless, Taguchi method is not good at handling the problems of multiple quality characteristics and quality characteristic with ordinal data. Therefore, this proposal plans to integrate gray relationship analysis method and fuzzy theory to deal with these two problems. Further, artificial neural network (ANN), a technique simulating the neural system of human beings, has the capabilities of learning, fault tolerance etc. This technique has been utilized in the broad areas such as prediction, classification of workparts, identification of sound. Thus, this research attempts to take advantages of artificial neural network and fuzzy theory to develop a fuzzy Radial Basis Function (RBF) neural network. This model may solve the following problems: (1) time-consuming of learning in back-propagation neural network, (2) fluctuation of the values of parameters during welding, and (3) fuzzy linguistic-term judgment for welding quality. Based on the results obtained from the Taguchi experiments, the developed fuzzy neural network can be trained to establish a quality prediction system for plasma arc welding. Finally, this research will apply the system to predict the interpolation outputs for other combination of welding parameters and plot three-dimensional pictures of suitability area of parameters to provide the industry in application.