【摘要】 为解决离心泵叶轮在初生空化故障下动应力及疲劳寿命预测难以实现快速、准确预测的问题，基于Workbench对故障进行非定常流固耦合数值模拟，在此基础上，使用贝叶斯优化-支持向量回归（Bayesian Optimization-Support Vector Regression, BO-SVR）对叶轮内流场压力与结构动应力响应进行训练和预测，结合Miner线性累计损伤理论和材料S-N曲线对叶轮疲劳寿命进行分析。将BO-SVR模型与XGBoost、LightGBM和随机森林3种常用预测模型进行对比，R~2分别提高了1.114%、1.369%、1.319%，并与nCode DesignLife软件的疲劳寿命分析结果进行校验，误差为4.981%，相较于其他模型误差分别减少了0.299%、1.035%、0.78%，且耗时仅为有限元法的1.077%，结果表明，所提方法具有优越的非线性捕捉能力，在保证预测精度的同时显著提高了计算效率。更多还原

【Abstract】 To address the challenge of achieving rapid and accurate prediction of dynamic stress and fatigue life in centrifugal pump impellers under incipient cavitation, this study proposes an approach based on Bayesian Optimization-Support Vector Regression(BO-SVR). A transient fluid-structure interaction simulation of incipient cavitation was first conducted using Workbench. Based on the simulation data, the BO-SVR model was employed to establish the relationship between the flow field pressure and the structural dynamic stress response. The fatigue life of the impeller was then analyzed by integrating Miner’s linear cumulative damage theory with the material S-N curve. The performance of the BO-SVR model was compared against three commonly used machine learning models: Extreme Gradient Boosting(XGBoost), Light Gradient Boosting Machine(LightGBM), and Random Forest(RF). The results demonstrated that the BO-SVR model achieved superior predictive accuracy, with R~2 values increasing by 1.114%, 1.369%, and 1.319%, respectively. Furthermore, validation against fatigue life analysis results from nCode DesignLife software showed a minimal error of only 4.981% for the BO-SVR model.This error was reduced by 0.299%, 1.035%, and 0.78% compared to the other models, while the computational time required was merely 1.077% of that required by the finite element method. These findings indicate that the proposed method possesses excellent capabilities for capturing nonlinear relationships. It significantly enhances computational efficiency while maintaining high prediction accuracy, providing a reliable solution for efficiently predicting the transient dynamic stress response and fatigue life of centrifugal pump impellers under incipient cavitation conditions.更多还原