融合字典学习的滑动轴承摩擦故障趋势自记忆预测  

Self-memorization model for predicting the wearing trend of sliding bearing with fusion dictionary learning

作　　者：张峻宁[1] 张培林[1] 张云强[1] 杨望灿[1] 华春蓉[2] 

机构地区：[1]军械工程学院七系,河北石家庄050003 [2]西南交通大学机械工程学院,四川成都610031

出　　处：《振动工程学报》2017年第3期519-526,共8页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(51205405;51305454)

摘　　要：将自记忆模型引入到滑动轴承的接触摩擦故障发展趋势预测中,针对一般形式的多变量自记忆模型的不足,提出一种基于字典学习的滑动轴承摩擦故障趋势的多变量自记忆预测方法。首先,构建多变量字典,利用稀疏编码筛选出影响系统演化的主要变量。然后通过误差平方和最小原则更新不同作用形式的变量字典,确定变量影响系统演化的最佳作用形式和影响系数,最终得到多变量的系统动力核函数,解决了系统动力核与多变量关系难处理问题。最后,引入自记忆函数,将所得的系统动力核反演成一个微分-差分方程,由此得到滑动轴承的多变量自记忆预测模型。应用到实例中,有效地预测了故障的发展趋势,为滑动轴承摩擦退化趋势预测提供了一种新的途径。The self-memorization theory is applied to the prediction model estimating the development trend of friction fault of the sliding bearing. The common multi-dimensional self-memorizaiton model is modified to improve its accuracy, and a wearing trend prediction method for the sliding bearing based on the self-memorization prediction model is put forward in this paper. First, the multivariate simplified dictionary is constructed, and the factors which influenc the evolution of the system are select ed by sparse coding. Then the optimal function form and the influence coefficient of the variables are determined based on the least square error principle. Finally, the system differential equation is obtained, which solves the relationship between the sys tern dynamics and the multi variables. The self-memory model for the displacement prediction of the development trend of the sliding bearing friction is established on the basis of the derived nonlinear differential equation. By introducing the sel^memory function, the dynamic core of the bearing wearing signal of the bearing system is transformed inversly into a difference equation. An example is given which shows that the modified self-memory method performed well. This method provides a new way to predict the degradation trend of the sliding bearing.

关 键 词：故障诊断 滑动轴承 摩擦故障 字典算法 自记忆原理 

分 类 号：TH165.3[机械工程—机械制造及自动化] TH133.31