統計製程管制(statistical process control, SPC)常被用來有效地維持產品的品質特徵(quality characteristic)在一個穩定而可接受的水準上。管制圖上所出現的異常形狀(pattern)通常和某些造成製程失控(out-of-control)的特定原因(assignable cause)有關,因此及早而準確地偵測到管制圖上出現的異常形狀,除了可以當成製程失控的早期訊號外,也可以縮小製程失控原因的可能範圍,從而可幫助品管人員更有效率地找到使製程失控的真正原因,因此管制圖形狀(control chart pattern, CCP)的辨識與分析是SPC的一個重要課題。近年來,有很多學者嘗試以類神經網路(artificial neural network, NN)作為辨識CCP之有效工具,結果證明NN在辨識CCP的速度與精度上優於其他的技術,但大部份的學者皆採用倒傳遞網路(back propagation network, BPN),BPN本身是靜態網路,所以並不適合應用在與時間相關的問題上,這個問題使得BPN為基的CCP辨識器在辨識相似的CCP時,產生極大的辨識誤差。本研究使用時間稽延類神經網路(time delay neural network, TDNN)來解決這個問題。本研究採用模擬的方式,並以辨識率及平均串連長度(average run length, ARL)作為績效評估指標。模擬結果顯示,相似CCP之間的誤辨識問題可被此以TDNN為基的模式有效解決。與傳統的辨識工具相比,TDNN有較佳的ARL表現,並可分辨出CCP的種類。
Most of the research in applying neural networks on SPC has used traditional back propagation networks (BPNs) that cannot deal with patterns that vary over time in an on-line CCP recognition scheme. This paper utilizes a time delay neural network (TDNN) based model to address this problem. The TDNN, with its special architecture, can represent relationships between patterns in a time sequence, and is therefore suitable to be trained with dynamic patterns that change over time. When compared with traditional BPNs, the TDNN model has better performance in both recognition accuracy and speed.