管制圖是品質改善活動中常用的方法之一,當數據形態為計數值時一般會採用計數值管制圖,如p管制圖或c管制圖以管制不合格率或不合格點數的變異。傳統的計數值管制圖是以二分法來判斷產品品質是否為合格,但在很多狀況下產品品質並不會在適用與報廢間突然轉換,反而會有許多中間等級的產品。因此若沒有充分利用中間等級的資訊,將降低計數值管制圖之成效。 為改善傳統二分法的缺失,本研究將以印刷電路板 (PCB) 製程為例,建立以模糊理論中的語意項目來表示每件電路板品質特性之模糊p管制圖。在本研究中,是以類神經網路為基礎,發展一個模糊計數值資料平均值變化之偵測系統。PCB品質是以模糊語意變數來表示,類神經網路是以模糊集合代表值為輸入,用來管制製程平均值是否發生偏移。類神經網路之效益是以平均連串長度 (ARL) 來評估,模擬之結果顯示,本研究發展之類神經網路對於製程平均值之偵測較模糊p管制法為優。 在實務使用上,當只取兩個語言項目來表達產品品質時,模糊p管制圖與傳統的p管制圖在使用上效益相同。所以本研究也將以大台北地區某一PCB公司製程為例,針對以語意項目為基礎的管制程序,如樣本記錄方法、管制界限的計算與管制方法做完整說明,期望能提供業界一個可參考之模式。
When quality-related characteristics are measurable on attributes, attribute control charts are often used. Traditional attribute control charts determine the product quality by the binary classification. This is not appropriate in situations where product quality does not change abruptly from perfect to worthless, and there might be a number of intermediate levels. Without fully utilizing the intermediate information, the use of attribute control charts usually causes poorer results. In this research, we propose a neural network-based quality control procedure to improve the performance of the fuzzy control charts. The primary focus is on the detection of changes in the process mean shift. The inputs of the neural network are linguistic variables. The performance of the neural network has been evaluated based on the average run length (ARL). Simulation results show that the proposed neural network approach is better than the fuzzy p-chart. Data from printed circuit board (PCB) process are used to demonstrate the application of the neural network approach.