The effects of graphite and copper electrode on EDMed surface of SKD11 cold tool steel by using Taguchi method and artificial neural network

• Main Authors: HUANG, SHUN-WEI, 黃舜韋
• Other Authors: TAI,TZU-YAO
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/29w8yg

碩士 === 南臺科技大學 === 機械工程系 === 105 === In this study, the L16 Taguchi method was used to perform the discharge processing on SKD11 with graphite and red copper as electrode tools. The effects of different processing parameters and the electrode tools of different materials were observed with material removal rate, electrode tool wear rate, surface roughness and white layer thickness. In view of the processing results observed in the experiment, due to the use of Taguchi permutations and combinations, it is necessary to calculate the signal-to-noise ratio, that is, the S / N ratio, to discuss the machining results of the performance, and finally calculate the percentage of contribution to find the highest impact factor of the processing parameters. In the result of the electric discharge processing, the electrode consumption rate and the material removal rate are both the discharge current and the continuous discharge time as the maximum influence factor, and the higher the discharge current and the discharge duration, the higher the electrode consumption rate and the material removal rate. While the surface roughness in the graphite processing part of the continuous discharge time as the main factor, red copper processing is the discharge current. Also in the re-casting layer thickness, graphite processing and copper processing are both discharge current and discharge duration are the maximum impact factor. In addition, in order to prove whether the experiment has sufficient accuracy and experimental reproducibility, the verification experiment is carried out and the difference between the experimental value and the actual experimental value is calculated. The final calculation results show that the gap value is smaller than the trust interval value. The experiments in this study do have experimental reproducibility. Finally, this study uses the inverted neural network to simulate the experimental results and compare the predicted data with the experimental data. It is found that as long as give a sufficient amount of the data for training and verification, can make the neural network model has good convergence results, and obtain good prediction ability. The prediction results show that the error is within 5%, which both graphite and copper in the hidden layer neurons and the number of the best predictor set is not the same, that between the two will be due to the material properties vary, but doesn't change the neural network suitable for discharge processing results of prediction.