基于PC-Kriging模型与主动学习的齿轮热传递误差可靠性分析

为提高齿轮热传递误差可靠性分析的计算效率和精度，提出了一种高效的基于PC-Kriging代理模型与主动学习函数LIF相结合的可靠性分析方法.采用多项式混沌展开(polynomial-chaos-expansion，PCE)替代传统Kriging模型的回归基函数来增强预测模型的全局近似精度，并利用Kriging模型来捕捉预测模型局部特征的能力.采用最小角回归(LAR)构建回归基函数的最优多项式数量集，同时用Akaike信息准则(AIC)来确定最优的截断集合.并采用一种主动学习函数LIF选择每次迭代的最佳样本点以提高模型收敛效率.通过齿轮热传递误差算例表明:与传统的Kriging代理模型相比，所提出方法在保证精度的同时可以极大地减少预测模型可靠性分析中的学习次数.

Reliability Analysis of Gear Heat Transfer Error Based on PC-Kriging Model and Active Learning

To improve the computational efficiency and accuracy in the reliability analysis of gear heat transfer error， an efficient reliability analysis method combining PC-Kriging and active learning function LIF is proposed. Polynomial-chaos-expansion (PCE) is adopted to replace the regression basis function of the traditional Kriging model to enhance its global approximation accuracy and its ability to capture local features. The least-angle regression (LAR) is used to construct the optimal polynomial quantity set of the regression basis function， and the Akaike information criterion (AIC) is utilized to determine the optimal truncated set. Furthermore， the active learning function LIF is employed to select the optimal sample during each iteration to improve the convergence efficiency of the PC-Kriging model. The application to gear heat transfer error shows that compared with the traditional Kriging model， the proposed method can significantly reduce the number of performance function evaluations while ensuring accuracy in the reliability analysis.