基于优化SVR模型的大跨度样本疲劳寿命预测  

Fatigue Life Prediction of Large-Span Samples Based on the Optimized SVR Model

作　　者：杨大炼[1] 刘义伦[1,2] 周维[1] 羿九火 

机构地区：[1]中南大学机电工程学院,湖南长沙410083 [2]中南大学轻合金研究院,湖南长沙410083

出　　处：《东北大学学报（自然科学版）》2015年第9期1321-1326,共6页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金资助项目(51375500);湖南省教育厅项目(2013SK2001);中央高校基本科研业务费专项资金资助项目(2013zzts037)

摘　　要：针对传统方法在大跨度、小样本情况下的疲劳寿命预测准确率不高的问题,研究基于优化SVR模型的寿命预测方法.根据大跨度样本的特点,提出有效的预处理方法、SVR模型的训练方法及参数优化准则.以LY12CZ(2A12)铝合金疲劳寿命预测为实例,分析了高斯核函数、多项式核函数及多层感知核函数对SVR模型训练误差的影响.结果表明高斯核函数更适用于SVR模型的训练,并通过细菌觅食算法对核参数γ及惩罚因子C进行优化选取,LY12CZ(2A12)铝合金疲劳寿命预测结果验证了该方法的有效性.Aiming at the issue that the prediction accuracy of fatigue life is not high by the traditional methods with large-span and small samples,a new life prediction method based on the optimized SVR model was studied.Considering the traits of large-span samples,the effective sample pretreatment method,the training method for the SVR model and the criterion for parameter optimization were put forward.Taking the life prediction of LY12CZ （2A12 ） aluminum alloy for example,the effects of the kernel functions of Gauss,polynomial and multilayer perception on the training error of the SVR model were analyzed.The results showed that the Gaussian kernel function is more suitable for SVR model training and the kernel function parameter γand the penalty factor C can be optimized by the bacterial foraging algorithm.Thus, the life prediction results verify the validity of this method.

关 键 词：大跨度 支持向量回归 疲劳 寿命预测 铝合金 

分 类 号：TG146.21[一般工业技术—材料科学与工程]