文献类型：期刊文献

英文题名：Efficient Reduced-Order modeling for steam turbine rotors using TCN-enhanced SVD

作者：Cai, Xiao-Bing[1];Gu, Hang-Hang[1];Mu, Wei-Wei[1];Yan, Jian-Jun[2];Zhang, Xian-Cheng[3]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Shanghai Key Lab Intelligent Sensing & Detect Tech, Shanghai 200237, Peoples R China;[3]East China Univ Sci & Technol, Key Lab Pressure Syst & Safety, Minist Educ, Shanghai 200237, Peoples R China

年份：2025

卷号：270

外文期刊名：APPLIED THERMAL ENGINEERING

收录：;EI(收录号：20251118055733);WOS:【SCI-EXPANDED(收录号:WOS:001448063500001)】；

基金：This work was supported financially by the National Key Research and Development Program of China (No.2021YFB3702204) , National Natural Science Foundation of China (Nos.51725503, 52005185) .

语种：英文

外文关键词：Turbine rotor; Reduced-order model; Singular value decomposition; Temporal convolutional network; K -nearest neighbor

摘要：Accurate and efficient reduced-order modeling of transient stress and temperature fields is crucial for digital twin applications in steam turbine rotors. These fields are essential for assessing creep-fatigue damage, lifespan estimation, and overall reliability of the rotor, particularly at dangerous locations where high stress and temperature gradients occur. However, full-scale finite element analysis is computationally expensive, and existing reduced-order models often fail to maintain accuracy in these critical regions, limiting their applicability in longterm structural integrity assessments. To address these challenges, this study proposes an efficient reduced-order modeling framework that integrates Singular Value Decomposition for global model reduction with a Temporal Convolutional Network to enhance prediction accuracy at dangerous locations. A dataset of 50 reduced-order models is generated to cover different operating conditions comprehensively. The Local Outlier Factor algorithm evaluates the similarity of a new operating condition to the existing dataset. If it falls within an acceptable range, an interpolation approach based on the K-nearest neighbor method, weighted by Euclidean distance, is applied to obtain an accurate prediction. Experimental results demonstrate that the proposed method significantly reduces computational costs while accurately predicting transient stress and temperature fields. The framework achieves a root mean square error of 4.003973 and a coefficient of determination of 0.997983, effectively capturing stress variations at dangerous locations. These findings highlight the potential of the proposed approach for real-time monitoring, early fault detection, and lifetime assessment of turbine rotors, enhancing the accuracy of creep-fatigue analysis and improving reliability in industrial power generation systems.