| 摘要: | 工業4.0作為新一階段的工業革命，以智慧製造為核心，結合大數據(Big Data)、物聯網(Internet of Things, IoT)、人工智慧(Artificial Intelligence, AI)和雲端計算(Cloud Computing)等新興技術，顯著推動維護策略的革新。上述技術的應用使得預測性維護策略成為可能，能於設備故障發生前提前識別異常並安排維護計畫，有效降低停機時間及維修成本。透過實時監測及數據分析，智慧製造不僅提升生產效率，也提高設備運行的穩定性和安全性，使企業能夠在高度競爭的市場中保持優勢。
本研究採用A公司提供的塗佈機歷史生產數據，以預測性維護策略(Predictive Maintenance, PdM)為架構，建構卷積神經網路(Convolutional Neural Network, CNN)和長短期記憶神經網路(Long Short-Term Memory, LSTM)之預測模型，並進行數據分析及機台異常偵測。研究結果顯示，CNN-LSTM模型搭配PReLU激勵函數和RMSProp優化器時，模型表現最佳，其準確率達99.82%，精確率達94.81%，F_1-Score為97.34%，與僅使用LSTM之模型相比，CNN-LSTM模型在準確率、精確率及F_1-Score方面皆有顯著提升。此外，CNN-LSTM模型可提前25秒預測到機台主速度的異常徵兆並減速，此有助於操作人員提前進行維護，進而降低損失並縮短停機時間，達到預測性維護之目的。;Industry 4.0 represents a new phase of the industrial revolution, centered on smart manufacturing and integrating emerging technologies such as the Internet of Things(IoT), Cloud Computing, Big Data, and Artificial Intelligence(AI). These advancements significantly drive the innovation of maintenance strategies. The application of these technologies enables Predictive Maintenance strategies, which can identify anomalies and schedule maintenance plans before equipment failures occur, effectively reducing downtime and maintenance costs. Through real-time monitoring and data analysis, smart manufacturing not only enhances production efficiency but also improves the stability and safety of equipment operations, allowing enterprises to maintain a competitive edge in a highly competitive market.
This study utilizes historical production data from Company A′s coating machine, employing a Predictive Maintenance(PdM) framework to construct predictive models using Convolutional Neural Networks(CNN) and Long Short-Term Memory(LSTM) networks for data analysis and anomaly detection. The results indicate that the CNN-LSTM model, when combined with the PReLU activation function and RMSProp optimizer, performs optimally with an accuracy of 99.82%, a precision of 94.81%, and a F_1-Score of 97.34%. Compared to the model using only LSTM, the CNN-LSTM model shows significant improvements in accuracy, precision, and F_1-Score. Additionally, the CNN-LSTM model can predict abnormal signs in the main speed of the machine 25 seconds in advance and slow it down, which helps operators perform maintenance proactively, thereby reducing losses and minimizing downtime, achieving the goal of Predictive Maintenance. |
