管制圖可以用來決定系統的狀態並偵測製程中隨時可能發生的異常情況。異常的管制圖形狀與製程變異中一些特殊的非機遇性原因有關聯,因此有效地辨識異常管制圖形狀能減少可能需要的檢查次數,並加速診斷搜尋。近年來類神經網路已經成功地應用在管制圖形狀辨識上,但大多數的研究均以原始資料作為類神經網路的輸入向量(Raw data based, RB);部分研究則是利用由原始資料所擷取出的特徵當作輸入向量(Feature data based, FB)以減少網路規模。本論文中所使用的訓練範例與測試範例均為利用蒙地卡羅模擬法產生出生產線製程數據並結合由製程數據所擷取出的統計特徵值一同當作倒傳遞類神經網路的輸入向量(Hybrid data based, HB),再利用類神經網路軟體來訓練與偵測異常管制圖形狀。本論文同時討論在常態環境與自相關環境下使用HB與RB於靜態下對異常管制圖形狀辨識做績效測試,並再對HB做動態測試模擬。靜態測試結果得知在常態環境下RB的平均辨識率為92.99%,HB的平均辨識率為95.89%,而在自相關環境下RB的平均辨識率為92.11%,HB的平均辨識率為95.12%,代表在常態環境下使用HB能擁有較佳的辨識能力,而在自相關環境下也仍然維持著良好的辨識績效。此外,由動態測試結果得知在常態環境下HB的平均辨識率為87%,而在自相關環境下HB的平均辨識率則為82%,兩者數據均明顯比靜態測試時來的要差,而自相關的動態測試則較常態環境下略差,但辨識率還可維持在80%以上,代表著使用HB在常態環境下擁有不錯的辨識能力,而在自相關環境也仍然維持著一定的辨識績效。
Control chart patterns (CCPs) can be used to determine the status of system. Unnatural CCPs can be associated with a particular set of assignable causes for process variation. In recent years, artificial neural networks (ANNs) have been successfully used in the CCP recognition task. In intelligent SPC, most of researches used raw data (RB) as input vector and the other researches have used statistical feature data extracted from raw data (FB) as input vector for reducing network size. In this thesis, we present an ANN-based approach, in which an improved hybrid training data (HB) integrates both the time series data (Raw data) and the statistical feature data (Feature data). The training data set and testing data set used in this thesis were generated by Monte-Carlo Simulation Method for production line process data. Both HB and RB will be examined in normal environment and auto-correlated environments at a static state while performing the tests of abnormal CCP recognition and then simulating HB at a dynamic state. The static test result shows that the average recognition rates of RB and HB are 92.99% and 95.89%, respectively, in normal environment. The static test results of RB and HB are 92.11% and 95.12%, respectively, in auto-correlated environment. The experiment results show that HB has better recognition performances in normal environment than in auto-correlated environment. Besides, the dynamic test result shows that the average recognition rate of HB is 87% in normal environment and 82% in auto-correlated environment. Both statistics are worse than they are in static environment and the auto-correlated results are inferior to the normal results in dynamic state. However, the auto-correlated results can still be maintained over 80% in real time on-line test. Our experiments yield better performance than previous works by using the proposed new method. Hence, it can be conclude that HB has better recognition ability in normal environment and still has well-performed ability in auto-correlated environment.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。