透過您的圖書館登入
IP:54.234.233.157
  • 學位論文

運用實驗設計和倒傳遞類神經網路結合模式於降低觸控玻璃製程阻抗之研究

Impedance Reduction in Touch Panel Glass Manufacturing Process by Using the Methods of DOE and Back-Propagation Network

指導教授 : 周永燦

摘要


近年來,各個產業均期望能在最短的時間內掌握影響產品產出的不確定因素,因而發展了許多最佳化的預測模型,其中以實驗設計法(DOE)和倒傳遞類神經網路法(BPN)是進行最佳化常見的兩種方法。 DOE的優點是可以掌握實驗因子的參數和預測最佳結果,但需再進行確認實驗才能驗證結果的正確性,浪費了許多時間和成本。BPN的優點是將數據區分為訓練組和測試組,建構了預測值之信賴區間,但缺點是對於實驗參數和區域的選擇,沒有考慮變數是否對品質特性是真正的顯著影響因素,也沒有探討實驗區域是否包含最佳的參數值,造成結果可能僅是區域的最佳解,而非真正的最佳化值。因此,本研究整合DOE和BPN成為一個結合模式,利用DOE來掌握實驗因子的參數和預測最佳結果,再使用BPN的測試來驗證預測值,可以消除DOE和BPN的缺點,以協助企業在最短的時間內掌握影響產品產出的因素和找到最佳參數來優化產品的品質。 研究以一個跨國企業的觸控玻璃生產製程為例,電性不佳是影響觸控玻璃產出的主要因素,因此以降低觸控玻璃的電測不良率為研究個案。首先以影響電性品質的參數和TRIZ的矛盾矩陣表建構一個關聯性表,探討參數相互間的關聯,加總各參數的關聯數量和進行排序,掌握顯著影響因素,篩選出實驗變數;然後進行二階段的DOE,先以田口方法(Taguchi method)來進行實驗,找出顯著影響產品的參數組合,再利用反應曲面法(RSM)對顯著影響產品的參數進行最佳化工程,探討出最佳的預測解;最後透過BPN的學習與測試,進行最佳化預測模式的驗證和比較。 研究結果顯示,運用二階段的DOE所得到電測良率最佳的預測值為94.76%,藉由BPN的學習和驗證得到電測良率的最佳預測值是94.245%,RSM和BPN的預測結果均優於改善前的88%,並且夠接近改善後由實證得到的電測良率94%,同時,高阻抗的不良是影響電測良率的最主要原因,其不良比重也由原來的53.2%降到改善後的4.6%,證實了DOE和BPN的結合模式是一個有效的分析模式,確實有利於產業在推動產品品質改善的績效表現。

並列摘要


A need to control the uncertainties that affect product output within a given timeframe has led to the development of numerous optimization prediction models, the most common of which are design of experiments (DOE) and Back-Propagation Network (BPN). As an example, this study used the process employed by a multinational corporation for the manufacture of touch panel glass. We began by combining parameters within the TRIZ contradiction matrix to construct a correlation table. We then examined the mutual correlation between parameters, summed the number of associations for all parameters, and ranked the parameters for the selection of experiment parameters. Finally, we applied Taguchi method to an experiment in order to identify the parameters with the greatest impact on the product and utilized response surface methodology (RSM) for optimization, the prediction of the optimal solution within the experimental area, and verification of the results with those of the DOE optimization prediction model using BPN learning. Our study results showed an improvement in yield rate from 88 % to 93.59 % using the Taguchi method. We then employed a process optimized by RSM to examine the three influential factors ascertained by RSM method, which resulted in a predicted yield rate of 94.76 %. The predicted yield rate using BPN learning was 94.245 % and empirical measurements were 94%. These results demonstrate the feasibility of using of TRIZ for the selection of experiment parameters in conjunction with two-stage DOE followed by BPN. Meanwhile, the Defect rate of impedance also decreased from 53.2% to 4.6%. Confirmation by experiment results verified the efficacy of the analysis model, which could be highly beneficial to industries seeking improvements in product quality.

參考文獻


Ahilan, C., Kumanan, S., Sivakumaran, N., Dhas Edwin Raja, J. (2013). Modeling and prediction of machining quality in CNC turning process using intelligent hybrid decision making tools, Applied Soft Computing, 13(3), 1543–1551.
Altan, M. (2010). Reducing shrinkage in injection moldings via the Taguchi, ANOVA and neural network methods, Materials & Design, 31(1), 599-604.
Andersen, B., Fagerhaug T. (2006). Root Cause Analysis: Simplified Tools and Techniques, American Society for Quality, Milwaukee.
Asafa, T. B., Tabet, N., Said, S. A. M. (2013). Taguchi method–ANN integration for predictive model of intrinsic stress in hydrogenated amorphous silicon film deposited by plasma enhanced chemical vapour deposition, Neurocomputing, 106(15), 86-94.
Asiltürk, I. and Neşeli, S. (2012). Multi response optimisation of CNC turning parameters via Taguchi method-based response surface analysis, Measurement, 45(4), 785-794.

被引用紀錄


林俊源(2015)。應用六標準差手法與TRIZ建構 射出成型製程能力改善與驗證〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840%2fcycu201500275

延伸閱讀