| 摘要: | 隨著現代科技的進步，全世界的各大企業正逐步地往工業4.0前進，伴隨著人工智慧這項技術已逐漸成熟，不論是製造業、金融業、能源業等不同領域，皆為了穩定高效的生產效率、提高產品良率及機台的低故障發生率，紛紛投入大量的研發費為了導入智慧製造相關的技術。
本研究運用A公司提供的塗佈機運轉數據，採納了長短期記憶網路（ Long Short-Term Memory, LSTM ）和單類支持向量機（ One-Class Support Vector Machines, OCSVM ）兩種方法的混合模型，以半監督式學習的方式建立，旨在實現預測性維護方法中的預診斷與健康管理（ Prognostics and Health Management, PHM ）。準確度為85.83%，精確度為46.82%，召回率為98.97%，F2-Score為80.9%，表明LSTM-OCSVM方法在異常偵測上的優越性。且預測模型相較於實際張力異常紀錄能提前約500秒檢測到異常徵兆，以利機台負責人員能於故障前做出決策，確保產線順暢，並降低因機台異常導致產品良率下降及機台壽命減短等問題。
;As modern technology advances, major companies worldwide are gradually moving towards Industry 4.0. Along with the gradual maturation of artificial intelligence technology, various sectors such as manufacturing, finance, and energy are all investing heavily in research and development to implement smart manufacturing technologies. This is aimed at achieving stable and efficient production, improving product quality, and reducing machine failure rates.
This study uses operational data from coating machines provided by Company A, adopting a hybrid model of Long Short-Term Memory (LSTM) and One-Class Support Vector Machines (OCSVM) methods, established in a semi-supervised learning manner. The aim is to achieve prognostics and health management (PHM) in predictive maintenance methods. The accuracy is 85.83%, precision is 46.82%, recall is 98.97%, and the F2-Score is 80.9%, indicating the superiority of the LSTM-OCSVM method in anomaly detection. The predictive model can detect abnormal signs about 500 seconds earlier compared to actual tension abnormal records, allowing machine operators to make decisions before a failure occurs, ensuring smooth production lines, and reducing issues such as decreased product yield and shortened machine lifespan caused by machine anomalies. |
