CAJ | 학술논문

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。
침대전통방법재대과도、소양본정황하적피로수명예측준학솔불고적문제,연구기우우화 SVR모형적수명예측방법。근거대과도양본적특점,제출유효적예처리방법、SVR 모형적훈련방법급삼수우화준칙。이 LY12CZ(2A12)려합금피로수명예측위실례,분석료고사핵함수、다항식핵함수급다층감지핵함수대SVR 모형훈련오차적영향。결과표명고사핵함수경괄용우 SVR 모형적훈련,병통과세균멱식산법대핵삼수γ급징벌인자 C 진행우화선취,LY12CZ(2A12)려합금피로수명예측결과험증료해방법적유효성。
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.