【作者】 张旭；

【导师】 郭晨霞；

【作者基本信息】 中北大学 ， 仪器科学与技术， 2024， 硕士

【摘要】 轴承是机械设备中一种重要零部件,具有固定机械传动、降低其运动过程中的摩擦系数的作用。轴承的健康状态对机械的稳定性至关重要。因此,准确的预测轴承的剩余使用寿命对观测机械设备的健康状态具有重要意义。近年来,针对轴承的剩余使用寿命(RUL)研究成为热点。然而,传统方法使用线性退化过程构建健康指数(HI)模型,并不能充分表达退化过程和时间的关系。为了解决上述问题,本文在HI模型构建、HI标签建立和RUL预测几个方面展开研究,具体共工作内容如下:(1)介绍了轴承的分类方法,并概述了机械维修策略的发展历程。从故障后维修到基于状态维修(Condition Based Maintenance,CBM),最后发展到剩余使用寿命(RUL)预测方法,推动维修策略发展。介绍RUL三种主要的研究方法,讨论方法的优势与缺点。选用数据驱动方法对轴承数据的剩余使用寿命预测,该方法避免了物理方法的拓展性低与经验法依赖专家经验的缺点,并对主要研究内容健康指数(HI)具体说明。(2)针对轴承数据采集特点,计算出轴承震动数据与时间的相关性。使用KMeans算法完成特征提取。设计了新的HI构建方法,使用特征提取数据表征轴承的退化程度,通过累加法完成HI的构建。以HI为阈值判别数值,将退化速度作为中间参数进行预测。该方法避免了阈值判别错误的问题,同时使用退化速度作为中间参数,减少数据与标签的分布差异,提高了预测准确性。(3)设计了基于迁移学习(Transfer learning)的退化速度预测模型,在源域数据建立数据增强模型,对轴承震动数据进行数据增强,更好的学习退化末期数据分布。建立了阈值判别模块,实现了轴承退化速度模型的设计。(4)针对以退化速度为中间变量的预测模型,本文与三种常见的模型标签进行实验对比,证明论文方法提高了预测精度。对比完备数据与非完备数据集的健康指数情况,证明了论文方法在非完备数据集的适用性。最后计算模型的百分比误差与得分函数,与文献中常见的多种方法进行对比,验证论文方法的预测效果。更多还原

【Abstract】 Bearing is an important part of mechanical equipment,which has the role of fixing mechanical transmission and reducing the friction coefficient during its movement.The health of the bearing is very important to the stability of the machine.Therefore,accurately predicting the remaining service life of bearings is of great significance for observing the health of mechanical equipment.In recent years,the research on the remaining service life(RUL)of bearings has become a hot topic.However,the traditional method uses linear regression process to construct a health index(HI)model,which cannot adequately express the relationship between regression process and time.In order to solve the above problems,this paper carries out research on HI model construction,HI label establishment and RUL prediction.The specific work is as follows:(1)The classification of bearings are introduced,and the development of mechanical maintenance strategy is summarized.From post-failure Maintenance to Condition Based Maintenance(CBM),and finally to the remaining useful life(RUL)prediction method,the maintenance strategy is promoted.Three main research methods of RUL are introduced,and the advantages and disadvantages of these methods are discussed.A data-driven method is used to predict the remaining service life of bearing data.This method avoids the disadvantages of low expansibility of physical method and dependence on expert experience of empirical method.The health index(HI)of the main research content is specified.(2)According to the characteristics of bearing data acquisition,the correlation between bearing vibration data and time is calculated.K-Means algorithm was used to complete feature extraction.A new HI construction method is designed,which uses feature extraction data to characterize the degree of bearing degradation and completes the construction of HI by summation method.HI is used as the threshold value and degradation rate is used as the intermediate parameter to predict.This method avoids the problem of threshold discrimination error,and uses degradation rate as an intermediate parameter to reduce the distribution difference between data and labels and improve the prediction accuracy.(3)A degradation velocity prediction model based on Transfer learning was designed,and a data enhancement model was established in the source domain data to enhance the bearing vibration data,so as to better learn the end-stage data distribution of degradation.The threshold discrimination module is established and the bearing degradation velocity model is designed.(4)For the prediction model with degradation rate as the intermediate variable,the experiment is compared with three common model labels,and it is proved that the method in this paper can improve the prediction accuracy.By comparing the health index of the complete data with the incomplete data set,the applicability of the method in the incomplete data set is proved.Finally,the percentage error and score function of the model are calculated,and the prediction effect of the method is verified by comparing with many common methods in the literature.更多还原