針對(duì)這些問題,華東理工大學(xué)朱明亮教授、軒福貞教授等人提出了數(shù)據(jù)-物理模型驅(qū)動(dòng)的超高周疲勞壽命預(yù)測框架(圖1),通過選擇結(jié)構(gòu)材料的小樣本疲勞壽命數(shù)據(jù),使用Z參量壽命模型進(jìn)行數(shù)據(jù)擴(kuò)展,引入多種機(jī)器學(xué)習(xí)算法和物理模型,采用多種材料對(duì)數(shù)據(jù)-物理模型的預(yù)測能力進(jìn)行比較與驗(yàn)證。系列研究成果以“On micro-defect induced cracking in very high cycle fatigue regime"“A data-physics integrated approach to life prediction in very high cycle fatigue regime"和“Data-driven approach to very high cycle fatigue life prediction"為題先后發(fā)表在Fatigue Fract. Eng. Mater. Struct. 2022; 45: 3393、Int. J. Fatigue 2023; 176: 107917和Eng. Fract. Mech. 2023; 292: 109630上。研究發(fā)現(xiàn),訓(xùn)練集越大,機(jī)器學(xué)習(xí)方法預(yù)測材料疲勞壽命準(zhǔn)確率越高,數(shù)據(jù)與物理模型的融合可顯著提升預(yù)測準(zhǔn)確度,為小樣本數(shù)據(jù)下的超高周疲勞壽命預(yù)測提供了解決方案。基于Z參量模型和人工神經(jīng)網(wǎng)絡(luò)搭建的Z-PINN模型對(duì)15Cr鋼、FV520B-I鋼和GCr15鋼的超高周疲勞壽命預(yù)測準(zhǔn)確率分別為78.9%、89.3%和94.3%(圖2)。