隨著工業 4.0 的時代來臨，智慧製造成為近年來最熱門的產業趨勢與競爭決勝關鍵，世界各大廠商均紛紛導入智慧製造相關技術，臺灣各工具機大廠也整合人工智慧 (Artificial Intelligence, AI)、感測器 (Sensor) 及物聯網 (Internet of Things, IoT) 等技術，將設備轉型為智慧機械，發展成當今全球市場樣態，而預測性維護技術在智慧製造中扮演非常重要的角色，會將機器設備加裝的感測器所收集並儲存於雲端系統的即時資料進行分析，預測未來機台故障的可能性與時間點，避免停機問題的發生，進而提升整體設備效率。 本研究使用 A 公司塗佈機實際生產時所回傳的資料，提出一項基於門控循環單元 (Gated Recurrent Unit, GRU) 的異常檢測方法，對設備進行診斷，先將資料透過主成分分析 (Principal Components Analysis, PCA) 進行降維，取出重要特徵後再透過 GRU 來建置預測模型，最後將驗證結果繪製出來，可以觀察約能在 30.5 秒前偵測到異常，促使維修人員在設備故障前進行維護程序，預先解決潛在問題，進而提升設備正常運作時間與生產品質。 ;With the advent of Industry 4.0, smart manufacturing has become the most popular industrial trend and the key to competition in recent years. The world′s leading manufacturers have imported into technologies related to smart manufacturing, and Taiwan′s major tool manufacturers have also integrated Artificial Intelligence (AI), Sensor and Internet of Things (IoT) technologies to transform their equipment into smart machines, which have developed into the current global market pattern. The technology of predictive maintenance plays a very important role in the smart manufacturing. Through the analysis of the real-time data collected by the sensors installed in the machine equipment and stored in the cloud system, the possibility and timing of future machine failure can be predicted to avoid the occurrence of machine downtime. This will prevent downtime problems and improve overall equipment efficiency. In this study, we propose an anomaly detection method based on Gated Recurrent Unit (GRU) to diagnose the equipment by using the data returned from the actual production of coating machines in Company A. The data is reduced dimension through Principal Components Analysis (PCA) to extract important features and then GRU is used to build a predictive model. Finally, after plotting the validation results, you can observe that anomalies can be detected about 30.5 seconds earlier. To prompt service person to carry out maintenance procedures before equipment failure and solve potential problems in advance, thus improving equipment operation time and production quality.
