在當今工業和科技不斷進步的背景下，設備健康監測(PHM)和預測剩餘失效時間（RUL）成為了工業管理中至關重要的部分。準確地預測設備的剩餘失效時間有助於提高設備的可靠性、降低維護成本，並優化生產計劃。然而，傳統的 RUL 預測方法在面對複雜多變的時間序列數據時面臨一些挑戰。這包括對設備狀態變化的準確捕捉以及長期依賴關係的建模。在大數據和機器學習的時代，如何利用網路上多元的資料做進一步的研究是現今熱門的課題。本研究旨在解決傳統 Transformer 的注意力機制在長序列預測上難以發現可靠的時序依賴的問題，提高運算效率及記憶體使用的優化；提升對局部序列特徵的提取能力；降低異常值對時間序列的影響，建立有效的預測模型幫助企業降低風險，提高設備維護效率，減少停機時間。為達成這些目標，本研究提出了一個混合模型，結合了 Autoformer 模型、STL 時序分解方法和時序卷積網路。這些方法的結合將有助於更準確地預測設備的剩餘失效時間，提高設備管理效率和生產效率。 ;In the context of today′s industrial and technological advances, equipment health monitoring (PHM) and predicted remaining life (RUL) have become a critical part of industrial management. Accurately predicting the remaining life of equipment can help improve equipment reliability, reduce maintenance costs, and optimize production schedules. However, traditional RUL prediction methods face a number of challenges when dealing with complex and variable time-series data. These include accurately capturing changes in equipment state and modeling long-term dependencies. In the era of big data and machine learning, how to utilize the multifarious data on the Internet for further research is a hot topic nowadays. In this study, we aim to solve the problem that the traditional attention mechanism of Transformer is difficult to find reliable time series dependencies in long sequence prediction, to improve the computational efficiency and optimize the memory usage, to enhance the ability of extracting local sequence features, to reduce the impact of anomalies on time series, to build an effective prediction model to help enterprises to reduce the risk, to improve the efficiency of equipment maintenance,and to reduce the downtime. To achieve these goals, this study proposes a hybrid model that combines the Autoformer model, the STL time-series decomposition method, and the time-series convolutional network. The combination of these methods will help to predict the remaining life of equipment more accurately and improve the efficiency of equipment management and productivity.
