冷陰極管之最佳塗佈製程研究

The Study on Optimal Coating Proces for Cold Cathode Fluorescent Lamp

莊凱超(Kai-Chao Cheng)

指導教授 ： 張耀仁

中原大學/工學院/機械工程研究所/碩士(2009年)

https://doi.org/10.6840/cycu200901552

摘要

國內各製作冷陰極管的科技大廠，為開發色差均勻度更好與更高燈管亮度為導向，對於其研究開發的實驗方法，通常是使用傳統因數實驗設計方法來進行，以找出該研究期望的燈管品質特性。但是，其實驗設計取決於各因數的水準，往往會造成因數實驗次數多，不易找到最佳的參數特性。在目前製造系統漸趨複雜下；同時，廠商在激烈競爭壓力下，對於製程系統穩定性的要求日漸提高，一旦生產過程中的一個參數設定不正確時卻又無法即時發現，生產產品出現問題時而導致產線被迫停線，將造成嚴重的損失，所以製程系統之穩定性變成相當重要的課題。
本研究主要以冷陰極管的塗佈製程為研究，由於類神經網路模式中參數的選擇對求解的成效影響甚大，考慮各項參數比例組合對品質特性的影響，並將製程因數水準固定，應用田口實驗之直交表法與灰關聯分析法配合著類神經網路之方法建模，再利用智慧型基因演算法取得最佳參數問題之研究，其優點是引入階層化的架構，將繁雜的實驗設計簡單化，使輸出的結果更具目標值，大幅縮短配方測試時間，找出合適之參數設定，再驗證經由實驗設計後的求解成效，優於未經實驗設計的求解成效。

關鍵字

冷陰極管

並列摘要

To develop lamps with better uniformity of color difference and brightness, major domestic manufacturers for cold cathode fluorescent lamp（CCFL）, frequently conduct the factorial analysis method, a method commonly used to isolate the optimum mix for quality CCFL characters.
However, the quality inconsistency for each factor selected usually leads directly to redundant test runs and failed attempts researching the optimum mix. As the manufacturing systems today become more complicated and competition intensified in the marketplace, the stability requirements of the manufacturing systems are of the utmost importance to quality-conscious manufacturers. That is, if they are not able to immediately identify and fix incorrect parameters in the production processes, these manufacturers will incur huge losses as the defect products come out of the production line, and they have to shut down the entire production line just to look for causes.
This paper addresses the drawbacks of the factorial analysis method and proposes a composite analysis model to isolate the optimum mix for quality CCFL characters in the coating process. This model first adopts the artificial neural network approach. The artificial neural network approach is especially effective for identifying parameters, measuring their mixed effects for quality characters, and ensuring that the selected factors do not co-vary with one another. This composite analysis model also includes the orthogonal array and grey relational analysis used in Taguchi methods and intelligent genetic algorithm.
The advantages of this model include: a layered structure that can simplify the experimental designs to produce results more practical in the real situations; and greatly shortening test times for formulating the optimum parameter mix. With the simulated result from the coating process, this study also shows that the model can indeed produce the desired mix better and faster than that generated by the factorial analysis method.

並列關鍵字

Cold Cathode Fluorescent Lamp

參考文獻

architecture of feedback artificial neural network” ,Physica, pp.574-594.

10、D. Bailey and D. Thompson, (1990), “Developing neural-network

Approach for Fault Diagnosis ” , pp.1001-1014.

for Human Face Detection” , Pattern Recognition , pp 781-787.

15、江孟龍（2002），“多響應系統之製程最佳參數化”，私立中原大學機械