基于数据驱动的增材制造铝合金的疲劳寿命预测

Data-driven Fatigue Life Prediction of Additively Manufactured Aluminum Alloys

增材制造金属材料的疲劳损伤及寿命预测问题是当前研究的热点.论文以增材制造AlSi10Mg为典型应用对象,采用数据驱动方法开展疲劳寿命预测,考虑到其疲劳试验数据有限,采用经过试验验证的可靠的理论模型和数值计算方法来获取足够的疲劳数据,以弥补试验数据的不足.首先,提出了基于缺陷特征参数的疲劳损伤模型,其次,建立了理论模型的数值实现方法,并将数值计算结果与试验结果进行对比,验证了所提方法的可靠性.然后,开展数据驱动模型的训练与预测,采用K最近邻的数据驱动算法预测了增材制造AlSi10Mg的疲劳寿命,最后,深入分析了疲劳寿命随增材制造内部缺陷、疲劳载荷的变化规律,研究了数据驱动模型的训练数据量及模型参数对预测精度的影响.

The fatigue damage and life prediction of additively manufactured metal materials are a hot topic of current research.In practical applications,fatigue damage is a typical failure mode of additively manufactured metallic materials and structures,which are often subjected to cyclic loading.Therefore,in order to improve the safety and reliability of additively manufactured metallic components in service,it is necessary to study their fatigue damage and life prediction methods.In this paper,the fatigue life prediction is carried out using a data-driven approach with the typical application of additively manufactured Al-Si10Mg.Considering the limited fatigue test data,a reliable theoretical model and a numerical method verified by experiments are used to obtain sufficient fatigue data.First,based on the damage mechanics theory,a fatigue damage model based on the defect characteristic parameters is proposed.The model can reasonably reflect the influence of internal defects on the damage evolution and fatigue life of the additively manufactured metal by introducing the parameters of defect size,ellipse aspect ratio,and the shortest distance from the defect center to the surface.The material parameters of the damage-coupled elastic-plastic constitutive model and the defect characteristics-based fatigue damage model are calibrated in conjunction with the static tensile and fatigue tests of the additively manufactured AlSi10Mg.Second,the numerical implementation of the theoretical model is established,and the numerical results are compared with the test results to verify the reliability of the proposed method.For different directions of the additive process,the fatigue lives of additively manufactured AlSi10Mg under different cyclic loads are computed,and the results are used as a database for the data-driven model along with the experimental data.Then,the training and prediction of the data-driven model are carried out,and the fatigue life of the additively manufactured AlSi10Mg is predicted by the K-nearest neighbor(KNN)algorithm.Finally,the laws of variation of fatigue life with the internal defects occurring in the additive manufacturing process and fatigue loading are analyzed in depth,and the influences of the number of training data and parameters of data-driven model on the prediction accuracy are studied.