加工製程產線日趨複雜且產能的需求提升，為降低生產設備維護成本及提高生產效率，預防甚至預測的產線維護策略逐漸成為主流；為達到產線設備壽命評估的需求，需仰賴巨量數據的資訊探勘與建模；從近5年工業4.0發展趨勢，異質機台聯網與物聯網系統架構的整合亦為趨勢。本研究的內容為建構整合監測系統與資訊串流功能。系統具備多物理量同步擷取功能，達到加工製程中多樣態資訊收集需求，納入開放平台通訊統一架構(Open Platform Communication Unified Architecture, OPC UA)與Firebase雲端開發平台，強化異質聯網能力，亦具備資料前處理模組，可以即時檢視資料狀態；而智慧預測模組可掛載離線訓練所獲得的預測模型，即時做出決策判斷。而研究中，以型鋼加工鋸帶切削數據進行數據再分析，對所發展系統作驗證；數據經段落選擇、清理、特徵提取等步驟，並以相關係數與方差擴張因子篩選特徵，使用非監督式聚類演算法?? 自組織映射(Self-Organizing Map)訓練預測模型，搭配最小量化誤差(Minimum Quantization Error)建立鋸帶老化衰退指標；最終引入整合監測系統確認系統有效性。使用SOM訓練模型對不同品牌做預測，可有效在驗證鋸帶#1壽命達97%、驗證鋸帶#4 壽命達95.5%、測試鋸帶#3壽命達83.3%、測試鋸帶#5壽命達97.6%時提出預警。;As production lines become more complicated and high production capacity is demanding, preventive even predictive maintenance strategies are becoming more and more mainstream in order to reduce maintenance costs and improve production efficiency. To achieve the prognostic needs of production equipment, we rely on the information exploration and modeling of big data. With the development trend of Industry 4.0, the integration of heterogeneous machine network and Internet of Things system architecture is gradually gaining importance. Based on the above, this study first constructs an integrated monitoring system and data streaming function. The system is equipped with multi-physics signal synchronized acquiring function to meet the needs of multi-mode information collection in the process. Incorporate Open Platform Communication Unified Architecture (OPC UA) and Firebase cloud development platform to strengthen heterogeneous networking capabilities. Develop data pre-processing module for real-time overview of data status, intelligence predict module can mount predict models from offline training to make real-time predictive warning. Afterward, data re-analysis by band saw cutting data in the steel machining center for system verification. The analysis consist of segment selection, data cleaning, screening redundant features by correlation coefficient and Variance Inflation Factor (VIF). Moreover, Apply the unsupervised learning algorithm—Self-Organizing Map (SOM) to train the prediction model, and use Minimum Quantization Error (MQE) to build up the aging and degradation assessment of saw bands. Finally, integrate the model into the integrated monitoring system to confirm system effectiveness. The result show that it is effective in providing early warning when the life of saw band #1 is 97%, the life of saw band #4 is 95.5%, the life of saw band #3 is 83.3%, and the life of saw band #5 is 97.6%.
