Abstract: To solve the problem that real-time online quantitative evaluation of rolling bearing health status of rotating equipment in industrial enterprises,based on the monotonicity principle,the mean value of multi-scale correlation number and multi-scale spectral distance index were selected as the health index.By using multi-scale analysis,improved l1   trend filtering and fuzzy C-means clustering analysis,a data-driven real-time health evaluation model and the evaluation criteria knowledge base of rolling bearing were constructed.Referring to the basic principle of transfer learning,a mapping method of real-time data to the coordinate system corresponding to the constructed knowledge base of health assessment criteria was proposed.The health assessment model trained by the experimental data of "run to failure" can be applied to engineering practice.Taking the second group of bearing experimental data of the University of Cincinnati Intelligent Maintenance System (IMS) center as the model training data,the knowledge base of bearing health assessment criteria was established.The first group of experimental data of IMS center and the vibration acceleration data of P3409A centrifugal pump bearing in hydrocracking unit of a petrochemical company in China were used as the model validation data to verify the health assessment model of rolling bearing.The results showed that the model had generalization and could effectively represent the health state of rolling bearing.The evaluation process did not rely on the prior knowledge of external experts and did not need the historical fault data of the bearing to be evaluated,which had important engineering application value for realizing the "unmanned" intelligent operation and maintenance of rotating equipment.

Key words: health assessment, multi-scale analysis, improved l1 trend filtering, fuzzy C-means clustering, evaluation criteria, rolling bearings

摘要： 为解决工业企业旋转设备滚动轴承健康状态难以实时在线量化评估的难题,基于单调性原则筛选多尺度互相关系数均值和多尺度谱距离指标均值作为健康指数,采用多尺度分析、改进的l1趋势滤波、模糊C均值聚类分析方法,构建了一种数据驱动的滚动轴承实时健康状态评估模型和评估准则知识库;借鉴迁移学习基本原理,提出了一种实时待测数据向构建的健康状态评估准则知识库对应的坐标体系下的数据映射方法;应用“运转到坏”的实验数据训练健康状态评估模型并实现工程实践运用。以辛辛那提大学智能维护系统(IMS)中心第二组轴承实验数据为模型训练数据,建立轴承健康状态评估准则知识库,应用IMS中心第一组实验数据和中国某石化公司加氢裂化装置P3409A离心泵轴承“运转到坏”的振动加速度数据为模型验证数据,对构建的滚动轴承健康状态评估模型进行验证。验证结果表明,该模型具有泛化性且能够有效表征滚动轴承健康状态,评估过程不依赖外部专家先验知识、不需要待评估轴承历史故障数据,对于实现旋转设备“无人化”智能运维具有重要的工程应用价值。

关键词: 健康状态评估, 多尺度分析, 改进l1趋势滤波, 模糊C均值聚类, 评估准则, 滚动轴承