管制圖非隨機性模型的辨認是實施統計製程管制成功與否的重要關鍵,本 論文提出 應用類神經網路來偵測管制圖中的非隨機性模型,研究中所辨 認的非隨機性模型包 括趨勢模型(包含上升、下降)、系統性模型、週期 性模型、混合性模型以及平均值 跳動模型(包含向上、向下)。 雖然已經有許多學者提出運用類神經網路於辨認管制圖非隨機性模型的研 究,但在 大多數的研究中是以原始數據或區間分數轉換後的數據作為類 神經網路的輸入訊號 ,本研究中則是從製程數據中擷取出特徵,轉換成 為類神經網路的輸入訊號,以獲 得更佳的效果。 本研究中使用正確辨別率與第一次偵測到目標模型的平均連串長度作為評 估系統的 指標,利用模擬法所得之結果顯示本研究所提出的辨認系統能 夠正確且迅速的辨認 出管制圖中的非隨機性模型。
Control chart patterns recognition is an important aspect of statistical process control (SPC). This paper provides a neural network-based pattern recognizer for the analysis of control patterns. This pattern recognizer looks for the following five nonrandom patterns: trend, systematic, cycle, mixture and shift. In the past, various neural network-based approaches have been developed for the analysis of control chart patterns. In this research, we also develop a neural network-based pattern recignizer. However, the features extracted from process data were used as the network input instead of raw data. The performance of the developed neural networks was evaluated by estimating the average run length and rate of correct classification (ROCC). Simulation results show that the pattern recognizer can recognize the control chart patterns correctly and fastly.