現況製造業各站製程中使用較先進量測技術進行多點量測作業,在多點量測作業中會發生量測數據產生件內和件間的變異,若是將同一件產品上不同位置的量測值視為一組合理樣本,利用傳統的 SPC 進行監控,而計算出的管制界限則會較窄,造成增加型 I 誤差判讀。 由於件內多點量測位置,會受到多個獨立來源所影響,故本研究主要探討利用獨立成分分析將製程的變異來源進行分析。分別以監控原始觀測值的I-MR-R/S管制圖,以及ICA為基礎的I-MR 和Hotelling T2管制圖,並且以平均連串長度做為製程監控的績效指標。 模擬範例中,研究結果發現當出現平均數偏移以及平均數呈線性趨勢遞增之干擾項時,以ICA監控為基的多變量Hotelling T2管制圖監控績效最佳,其次為I-MR管制圖,而原始觀測值的I-MR-R/S管制圖績效最差。而在實際案例化學/物理氣相沉積薄膜(CVD,PVD)、電性阻抗值檢測站(TEG)製程中,Hotelling T2管制圖相較於I-MR、I-MR-R/S更快偵測到製程變化狀況。
The use of advanced manufacturing technology for multi-point measurement, multi-point measurements in the data generated will occur within-part and between-part of the variation, if the same product in different locations on the measurement value as a group reasonable sample, the use of traditional SPC to monitor and calculate the control limits will be narrow, resulting in increased type I error of interpretation. As a within-part the multi-point measurements will be affected by a number ofindependent sources, this study focuses on the use of independent component analysis of the sources of process variation analysis. Original observations were to monitor the I-MR-R/S control chart, and ICA-based I-MR and Hotelling T2 control chart, and the average length as a series of performance indicators to monitor the process. In the simulation example, the study found that the mean shift when there is a linear trend and the average increase of the interference items to ICA-based multi-variable control Hotelling T2 control chart monitoring the best performance,followed by I-MR (IC) control chart, and the original observations of the I-MR-R control chart performance is the worst. The real case, Hotelling T2 control chart compared to the I-MR(IC), I-MR-R/S (X)to quickly detect changes in process conditions.