| DC 欄位 |
值 |
語言 |
| DC.contributor | 機械工程學系 | zh_TW |
| DC.creator | 蔡宜璋 | zh_TW |
| DC.creator | I-Chang Tsai | en_US |
| dc.date.accessioned | 2021-11-12T07:39:07Z | |
| dc.date.available | 2021-11-12T07:39:07Z | |
| dc.date.issued | 2021 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=108323004 | |
| dc.contributor.department | 機械工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 射出成型過程中的收縮是造成產品缺陷的原因之一，緣於高分子熔膠的流動特性使整個充填過程變得難以掌控，而溫度與壓力等參數的設定就是改善此問題的重點對策。傳統作法為透過經驗豐富的操作人員手動調整參數的方式，經過多次的試誤法(Trial and error)來找出適合的製程參數，除了仰賴操作員的經驗，也較缺乏嚴謹性與科學根據支撐。本研究利用模具型腔內的溫度與壓力感測器來監控成型時的熔膠狀態，將數據轉換成比容的形式來呈現成品品質，再透過實驗設計法結合類神經網路來預測並改善試片的成品品質，成品品質為體積收縮率與比容非均勻性的組合。實驗設計法使用田口方法與反應曲面法，類神經網路採用倒傳遞網路、遺傳演算法基底網路與徑向基函數網路，並比較各模型間的預測能力。首先經由田口方法之訊噪比篩選出顯著因子，再利用顯著因子進行反應曲面法之全因子實驗，得到一組最佳化製程參數與比容預測值，最後利用類神經網路的模型來驗證結果。
本研究透過實驗數據來建立製程參數與成品品質之模型，結果顯示，反應曲面法預測四組驗證範例之預測值與實驗值之平均驗證誤差為1.64%，且最佳化製程參數與中心點實驗相比改善了31.2%；倒傳遞網路預測值與實驗值之平均驗證誤差為5.49%；遺傳演算法基底網路預測值與實驗值之平均驗證誤差為8.42%；徑向基函數網路預測值與實驗值之平均驗證誤差為3.25%，顯示結合實驗設計法與類神經網路能有效的預測與改善射出成型之非線性問題。 | zh_TW |
| dc.description.abstract | Shrinkage is a reason that causes product defects in injection molding process, considering the flow behavior of the polymer melt makes the filling process difficult to control, which means the setting of parameter such as temperature and pressure is the key orientation to improve the problem. Traditionally, researchers can obtain the appropriate parameters set through trial and error method. However, relying on experience might cause the lack of rigorous, the experiment still needs a method with theoretical support. In this paper, temperature and pressure sensors in the mold cavity to monitor the melt state during molding process and collect data. Convert the data into specific volume to present as the quality of the product. Product quality represents the combination of volume shrinkage and non-uniformity of specific volume. Then combine the design of experiment (DOE) and artificial neural network (ANN) to predict and improve product quality. Taguchi method (TM) and response surface methodology (RSM) were used for design of experiment. Back propagation network (BPN), genetic algorithms of neural network (GANN) and radial basis function network (RBFN) were used for Ann. Next, discuss the pros and cons between these models. First, through signal to noise ratio (SNR) from Taguchi method to select the significant factors. Then use these factors to conduct the full factor experiment of response surface methodology, obtain a set of optimal process parameters and prediction value of specific volume. Finally, compare the experimental outcome with prediction value of the ANN model for validation.
In this paper, a model which contains process parameters and product quality constructed through experiment data finally shows the validation error rate of response surface methodology prediction of four validation dataset and value of experiment is 1.64 percent in average. And the prediction value of optimal parameter is improved by 31.2 percent compared to the center point experiment; validation error rate of back propagation network prediction and value of experiment is 5.49 percent in average; validation error rate of genetic algorithms of neural network prediction and value of experiment is 5.49 percent in average; validation error rate of radial basis function network prediction and value of experiment is 3.25 percent in average. The results show that the combination of design of experiment and artificial neural network is effectively on predicting and improving the nonlinear problem in injection molding. | en_US |
| DC.subject | 收縮率 | zh_TW |
| DC.subject | 田口方法 | zh_TW |
| DC.subject | 反應曲面法 | zh_TW |
| DC.subject | 倒傳遞網路 | zh_TW |
| DC.subject | 遺傳演算法基底網路 | zh_TW |
| DC.subject | 徑向基函數網路 | zh_TW |
| DC.subject | Shrinkage | en_US |
| DC.subject | Back propagation network (BPN) | en_US |
| DC.subject | Genetic algorithms of neural network (GANN) | en_US |
| DC.subject | Radial basis function network (RBFN) | en_US |
| DC.subject | Taguchi method (TM) | en_US |
| DC.subject | Response surface methodology (RSM) | en_US |
| DC.title | 應用類神經網路與實驗設計法預測與優化射出成型試片之收縮率 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Shrinkage prediction of injection molded specimen by using artificial neural network and optimization with experimental design | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |