| Summary: | 碩士 === 國立臺灣大學 === 機械工程學研究所 === 88 === The purpose of this paper is to develop a new approach to stabilize the machining conditions of WEDM by means of an on-line algorithm for estimating the thickness of the work-piece. Although there are some on-line mathematical models for estimating the thickness of the work-piece, they are all too complicated to use, especially there is no rule to choose the coefficient for the models. Because the mathematical model needs a lot of experiments to decide its coefficients and that costs a lot of time, the Neural Network is introduced in the study. The Neural Network is known for its self-learning ability and its robustness to input noises. Through repeat learning with the reduced error method, the Neural Network is able to estimate the thickness of the work-piece without too many experiments. With the noise tolerability, the Neural Network is not affected by the noises in its learning process. These two characteristics make the estimation work simple and regularized.
In the paper, the response of the servo feed control system to the change of the work-piece thickness is studied first in terms of the servo voltage''s influences over the sparking frequency, the table feed and the electrical discharge condition. After all the factors relating to all kinds of electrical discharge machining phenomenon and all the causes making the discharge unstable are figured out, the learning samples for the Neural Network are obtained. After the learning process is finished, the Neural Network is able to perform on-line estimation of the work-piece thickness with the accuracy 1.2mm and identify the change of the thickness with 2 seconds. At the last stage, based on the thickness estimated on-line, the optimal electrical parameters are calculated and fed back to the servo control system, in order to reduce the unstable phenomenon in the EDM process.
|
|---|
