隨著工業 4.0 的推行，智慧化、數位化及自動化是製造產業發展的重要趨勢，大多數的工廠也漸漸的將智慧製造導入，希望能藉此提高生產效率、降低設備故障率及能源的消耗、敏捷製造和產品改善。這也代表機械必須朝「精密化」、「智慧化」的方向發展，以具備故障預測、精度補償、自動排程等功能。機器的穩定度成了多數工廠在乎的問題，如何有效且精確地執行維護策略是多數工廠必須面對的問題，因此近年來預測性 維護逐漸被推廣出來。基於前述，本研究主要希望能在設備發生故障前發現異常，以利維護人員提前進行修護，目的在於提早修復異常、避免產出不良品及機器的停機。 本研究使用 A 公司所提供塗佈機上感測器的歷史數據進行分析，以預測性維護（Predictive maintenance）為主要目標。分別使用主成分分析（Principal Component Analysis, PCA）以及羅吉斯迴歸（Logistic regression）進行降維，再利用長短期記憶神經網路（Long Short-Term Memory, LSTM）來建置模型，以利維修人員能提前做出決策，確保生產線的順暢，進而降低損失。;With the implementation of Industry 4.0, intelligence, digitalization and automation are important trends in the development of the manufacturing industry. Most factories have gradually introduced smart manufacturing, hoping to improve production efficiency, reduce equipment failure rates and energy consumption, Agile manufacturing and product improvement. This also means that machinery and equipment must develop in the direction of "precision" and "intelligence" to have functions such as failure prediction, accuracy compensation, and automatic scheduling. The stability of the machine has become a problem that most factories care about. How to effectively and accurately implement maintenance strategies is a problem that most factories must face. Therefore, predictive maintenance has gradually been promoted in recent years. Based on the foregoing, this research mainly hopes to find abnormalities before equipment failures, so that maintenance personnel can repair them in advance. The purpose is to repair abnormalities early and avoid production of defective products and machine shutdowns. This study uses the historical data of the sensors on the coating machine provided by Company A for analysis, with predictive maintenance as the main goal. Principal Component Analysis (PCA) and Logistic regression are used fordimensionality reduction, and Long Short Term Memory (LSTM) is used to build the model to facilitate maintenance. So that personnel can make decisions in advance to ensure the smoothness of the production line, thereby reducing losses.
