於統計製程管制(SPC)之中，管制圖被廣泛應用於偵測製程屬性上的偏移。而大部分的機台廣泛依靠sensor收集生產過程所產生的製程資料，而隨著工業持續進步，資料收集的間隔已經大幅所短，此型態的製程資料存在序列相關的特性，可視為時間序列資料。 X ? Control chart與R Control chart對於製程具有資料獨立性與常態性的假設的限制。因此應用於低時間間隔情境下所獲取的製程資料，易產生誤判。因此，我們提出一項以傳統管制圖偵測時間序列資料的管制圖建立機制，此方法提出一套完整的資料離散方法與抽樣概念，解決傳統管制圖的限制條件。 本次研究基於A公司所提供之塗佈機生產資料進行應用，塗佈機設置大量感測器偵測多項屬性資料，以PID(proportional–integral–derivative)控制器調節機台的運轉，系統接收資訊並隨時修正運轉狀態。但是，當發生突發狀況，導致機器運轉的狀況超出系統允許及時修正的範圍，將大幅度降低系統修正的效率，而重大異常的停機事件與事後檢修亦造成相當可觀的成本。 本研究可視為一項先導研究，根據上述提到的問題，我們將以新型態的管制圖建立流程，發展一套針對塗布機生產環境之異常偵測方法，以提供工程師更多判斷異常資訊的依據。本方法亦可對應多種參數設定環境下所之生產資料，與同時監控多個區域以及具備良好的彈性。 ;In Statistical Process Control (SPC), control charts are widely used to detect deviations in process attributes. Most of the machines widely rely on sensors to collect data during the process. With the continuous progress of the industry, the data collection interval has been greatly shortened. This type of process data has serial correlation, which can be regarded as time series data. The X ? Control chart and R Control chart has limitations on the assumptions of data independence and normality. Therefore, it may cause misjudgments when applied to the process data obtained in the context of low time intervals. We propose a control chart establishment mechanism that uses traditional control charts to detect time series data. This method proposes a complete set of data discrete methods and sampling concepts to solve the limitations of traditional control charts. This research is based on the process data of the coating machine provided by company A. The coating machine is equipped with a large number of sensors to detect multiple attribute data, and the Proportional–Integral–Derivative(PID) controller is used to adjust the operation of the machine, and the system receives Information and revise the operating status at any time. However, when an emergency occurs that causes the machine to operate beyond the range allowed by the system to be corrected in time, the efficiency of system correction will be greatly reduced, and major abnormal shutdowns and post-repairs will also cause considerable costs. This study can be regarded as a pilot study. Based on the problems mentioned above, we will develop a set of abnormal detection methods for the production environment of the coating machine with a new type of control chart establishment process, so as to provide engineers with more information about abnormalities. More judgments on the basis of abnormal information. This method can also correspond to the production materials in a variety of parameter setting environments, monitor multiple areas at the same time, and have good flexibility.
