文献类型：期刊文献

英文题名：An online data-driven method for predicting crack propagation and remaining fatigue life via combining linear and nonlinear ultrasonic

作者：Zhou, Jiachen[1];Liu, Lishuai[1,2];Xu, Haiming[1];Xiang, Yanxun[1];Xuan, Fu-Zhen[1]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai Key Lab Intelligent Sensing & Detect Tech, Shanghai 200237, Peoples R China;[2]Shanghai Inst Aircraft Mech & Control, Shanghai 200092, Peoples R China

年份：2025

卷号：155

外文期刊名：ULTRASONICS

收录：;EI(收录号：20252318538618);WOS:【SCI-EXPANDED(收录号:WOS:001504064900001)】；

基金：AcknowledgmentsThis work was supported by the National Key Research and Devel-opment Program of China (Grant No. 2024YFF0619000) , the National Natural Science Foundation of China (Grant Nos. 12422415, 12327807, and 12374434) , the Shanghai Science and Technology Innovation Ac-tion Plan (No. 23JC1401600 and 24DZ2202000

语种：英文

外文关键词：Ultrasonic features; Fatigue crack; Remaining useful life; Long short-term memory (LSTM)

摘要：Ultrasonic features play a critical role in evaluating the structural integrity of metallic components, yet current approaches predominantly rely on individual ultrasonic parameters for predictive analysis. This study presents an online data-driven method that combines linear and nonlinear ultrasonic parameters through an optimized weighting function to predict crack propagation and remaining fatigue life (RFL) over the entire fatigue life from microstructural changes to macroscopic crack formation of plate structures. LSTM neural networks are employed to learn sequential features captured by various PZTs. Experimental results on 6061 aluminum plates demonstrate that the proposed method predicts crack length and RFL with average errors of 0.568 mm and 4.50 % of the total fatigue life of the structure, respectively. Comparative analysis reveals that the combined approach with the optimal weighting function outperforms predictions using individual parameters. This method shows significant robustness under varying conditions, underscoring its potential for real-time fatigue monitoring and predictive maintenance.