基于多轴载荷相位差的神经网络预测钛合金疲劳寿命  

Multi-axial fatigue life prediction of titanium alloy based on neural network of load phase difference

作　　者：郑战光[1] 张剑 孙腾 谢昌吉 黄增 ZHENG Zhan-guang;ZHANG Jian;SUN Teng;XIE Chang-ji;HUANG Zeng(School of Mechanical Engineering,Guangxi University,Nanning 530004,China;School of Mechanical and Marine Engineering,Beibu Gulf University,Qinzhou 535011,China;Guangxi Vocational and Technical College of Mechatronics,Nanning 530007,China)

机构地区：[1]广西大学机械工程学院,南宁530004 [2]北部湾大学机械与船舶海洋工程学院,钦州535011 [3]广西机电职业技术学院,南宁530007

出　　处：《中国有色金属学报》2023年第3期781-791,共11页The Chinese Journal of Nonferrous Metals

基　　金：国家自然科学基金资助项目(52265018,51675110);广西自然科学基金资助项目(2021GXNSFAA220119)。

摘　　要：钛合金是航空、航天等领域重要的结构材料,面临着多轴疲劳寿命预测困难与繁琐等问题。为此,本文考虑了对称正弦波加载条件下的相位差与加载路径非比例度的关系,构建一种以相位差、正应变幅值和切应变幅值作为输入变量,以疲劳寿命作为输出变量的多轴疲劳寿命预测的神经网络模型;分别以18组纯钛、14组BT9钛合金和33组TC4钛合金多轴疲劳试验数据进行训练,用另外的2组纯钛、2组BT9钛合金和5组TC4钛合金数据进行测试;最后与等效应变模型、最大切应变模型、临界面模型进行比较。结果表明:与基于实验数据的唯象预测方法相比,基于多轴载荷相位差的神经网络方法在预测钛合金疲劳寿命方面更高效、更准确、更具普适性。Titanium alloy is an important structural material in aeronautics,astronautics and other fields,so it is often subjected to multi-axial fatigue loads.At present,multi-axis fatigue life prediction is very difficult and time-consuming.Therefore,an efficient and universal method of multi-axis fatigue life prediction was proposed based on neural network.In the proposed method,the relationship between the phase difference and the non-proportional degree of the symmetrical sine wave loading was used to describe multi-axial loading path,so phase difference,normal strain amplitude and shear strain amplitude were taken as the input units of neural network and fatigue life was taken as the output unit of it.Then,multi-axial fatigue experimental data of 18 groups of pure titanium,14 groups of BT9 titanium alloy and 33 groups of TC4 titanium alloy were used to train the neural network,and the others of 2 groups of pure titanium,2 groups of BT9 titanium alloy and 5 groups of TC4 titanium alloy were tested.Finally,the proposed method was compared with the equivalent strain model,the maximum shear strain model and the critical plane model.The results demonstrate that the proposed neural network method is more efficient,accurate and universal than the phenomenological prediction method based on the experimental data.

关 键 词：钛合金 相位差 神经网络 多轴疲劳寿命 加载路径 

分 类 号：TH114[机械工程—机械设计及理论]