| 摘要: | 在現今科技的快速演進與工業 4.0 的發展推動下，全世界的製造業逐漸以「高生產效率」、「高穩定度機具設備」、「高產品品質」為目標進行發展，使得導入智慧製造相關技術，成為全球各大廠商目前最競爭的致勝關鍵與熱門的議題。整合運用物聯網（Internet of Things, IoT）、雲端運算（Cloud Computing）、人工智慧（Artificial Intelligence,AI）、感測器（Sensor）等先進的技術方式，使得設備轉型智能化，讓機台得以達到預測性維護之目的，發覺異常並盡速進行維修與保養，進而降低故障頻率的發生。
本研究使用 A 公司塗佈機之生產真實數據資料，基於預診斷與健康管理（Prognostics and Health Management, PHM）為架構，透過長短期記憶（Long Short Term Memory, LSTM）神經網路方法建置預測模型，由歷史數據進行分析與預測塗佈機之機況異常，並運用三種不同優化器-Adam、Adamax 及 RMSprop，進行模型效能比較，選定最佳模型完成模型驗證分析，將其預測結果繪製呈現，本研究之實驗結果可使機台於 38 秒前偵測出異常，準確率為 99.87%、精確率為 95.29%、召回率為 100%，F1-Score 作為最終整體綜合評價指標，高達 97.59%，以此作為機台設備健康狀況指標，實現預測性維護策略，得以盡早完成檢修保養計畫，並有效確保機台持續運轉，進行產品產出提高生產效率，大幅降低停機時間和延長設備壽命。;With the rapid advancement of technology and the development of Industry 4.0, the global manufacturing industry is progressively focusing on "high production efficiency," "high equipment stability," and "high product quality" as key development goals. Implementing smart manufacturing technologies has become a highly competitive and popular topic among major manufacturers worldwide. The integration of advanced technologies such as the Internet of Things (IoT), cloud computing, artificial intelligence (AI), and sensors enables the transformation of equipment into intelligent systems. This facilitates predictive maintenance, allowing for the timely detection of abnormalities and prompt repairs and maintenance, thereby reducing the frequency of equipment failures.
This study utilizes real production data from Company A′s coating machine. Based on the framework of Prognostics and Health Management (PHM), a predictive model is built using the Long Short-Term Memory (LSTM) neural network method. Historical data is analyzed to predict anomalies in the coating machine, and using three different optimizers－Adam, Adamax, and RMSprop are employed to compare model performance. The optimal model is selected through validation analysis, and its predicted results are presented graphically.
The experimental results of this study enable the detection of abnormalities in the machine 38 seconds in advance, with an accuracy rate of 99.87%, precision rate of 95.29%, recall rate of 100%, and an impressive F1-Score of 97.59% as the overall evaluation metric. These results serve as indicators of the machine′s equipment health condition, enabling the implementation of predictive maintenance strategies. This ensures timely completion of inspection and maintenance plans, effectively ensuring continuous machine operation, improving production efficiency, significantly reducing downtime, and extending equipment lifespan. |
