物理導向AI驅動跨材料複材應用之零樣本參數精準導航

在淨零碳排的趨勢下，具備輕量化與高強度特性的複合材料成為航太與運具等產業的發展核心。 然而，複材製程參數開發長期受限於高昂的CAE（Computer-aided Engineering）軟體門檻與繁瑣試誤流程。 當導入新材料時，傳統仰賴大量模擬與實驗的方法，常導致開發週期過長且成本高昂。本研究提出「物理導向零樣本整體式學習」（Physics-guided Zero-Shot Ensemble Learning）AI 技術，能利用既有材料知識推論新材料參數，解決跨材料特性預測問題，成功減少超過50% 的模擬與試驗次數。同時透過遷移學習（Transfer Learning），以少量真實數據進行模型補償，使模型預測值與真實強度誤差降至 3% 以下，大幅提升複材製程優化效率與精度。

Amid the global trend toward net-zero carbon emissions, composite materials with lightweight and high-strength characteristics have become a key development focus in the aerospace, green energy, and transportation industries. However, the development of composite manufacturing parameters has long been constrained by high CAE software barriers and time-consuming trial-and-error processes. When introducing new materials, traditional approaches relying on extensive simulations and physical experiments often lead to prolonged development cycles and high costs. This study proposes an innovative AI technology, Physics-guided Zero-Shot Ensemble Learning, which leverages existing material knowledge to infer parameters for new materials, effectively addressing the challenge of cross-material property prediction. This approach successfully reduces more than 50% of the simulation and experimental efforts required for new material development. Furthermore, Transfer Learning is employed to bridge the gap between simulation and real-world manufacturing. By using a small amount of physical data for model compensation, the prediction error for material strength is reduced to below 3%. The proposed AI solution demonstrates that intelligent software can effectively replace traditional experience-based methods, enabling efficient and high-precision process optimization in composite manufacturing.