01挑戰The challenge
沖床是金屬成形產線的核心設備。非預期停機會打亂生產排程；過度保養則增加成本與能源浪費。工廠需要即時掌握設備狀態，在對的時間做對的保養。Stamping presses are the heart of metal-forming lines. Unplanned stops disrupt production schedules, while over-maintenance wastes money and energy. Plants need to see machine condition in real time and maintain at the right moment.
02我們的做法Our approach
- 在沖床 STI-110 上搭載多種類型感測器，持續擷取運轉資料Fitted the STI-110 press with multiple sensor types for continuous data capture
- 以 PHM 方法建置故障預測模型，並比較多種遷移學習方法，讓模型能套用到不同沖床Built PHM fault-prediction models and compared transfer-learning methods so the models generalize across presses
- 結合 IT 技術建置健康度儀表板，即時監控間隙、油質、油量等健康指標，並給出保養建議Built a health dashboard tracking clearance, oil quality and oil level, with maintenance recommendations
03成果Results
- 即時監控工廠生產狀況，提供適時的維護保養策略Real-time visibility of production status, with timely maintenance strategies
- 每年可降低 556.8–883.2 公斤 CO₂ 排放Cuts 556.8–883.2 kg of CO₂ per year
- 獲 2023 年第 16 屆工具機「研究發展創新產品」競賽佳作獎Honorable mention at the 16th Machine Tool R&D Innovation Product Awards (2023)
Hwang, P.-W.; Chang, Y.-J.*; Tsai, H.-C.; Tu, Y.-T.; Yang, H.-P. Comparison and Optimization of Generalized Stamping Machine Fault Diagnosis Models Using Various Transfer Learning Methodologies. Sensors 2025, 25(6), 1779 (2025.3.13). https://doi.org/10.3390/s25061779 (SCI, IF = 3.4, Rank: 122/353 = 34.56%, Q2 in the WoS category of Engineering, Electrical & Electronic.)