一种融合多传感器数据的数模联动机械剩余寿命预测方法  

A Model-data-fusion Remaining Useful Life Prediction Method with Multi-sensor Fusion for Machinery

作　　者：李乃鹏 蔡潇 雷亚国 徐鹏程 王文廷 王彪 LI Naipeng;CAI Xiao;LEI Yaguo;XU Pengcheng;WANG Wenting;WANG Biao(Key Laboratory of Education Ministry for Modern Design and Rotor-Bearing System,Xi'an Jiaotong University,Xi'an 710049)

机构地区：[1]西安交通大学现代设计及转子轴承系统教育部重点实验室,西安710049

出　　处：《机械工程学报》2021年第20期29-37,46,共10页Journal of Mechanical Engineering

基　　金：国家自然科学基金(52005387,52025056);NSFC-浙江两化融合联合基金(U1709208);中国博士后科学基金(2020M673380)资助项目。

摘　　要：随着传感和信息技术的发展,各式各样的传感器获取了机械装备海量的监测数据,让剩余寿命预测有"据"可依,推动机械剩余寿命预测进入了大数据时代。但由于数据类型多样、量大面广,如何利用丰富的多传感器数据,从中快速挖掘健康状态退化信息,指导寿命预测,成为大数据时代下机械寿命预测的全新挑战。基于模型的寿命预测方法大多仅针对单一监测数据进行建模分析,无法有效利用丰富的大数据资源。数据驱动的方法则过分依赖训练数据,缺乏必要的经验指引,方法的可解释性差。为了有效利用多传感器数据指导寿命预测,从数模联动的思路出发,建立了一种融合多传感器数据的数模联动寿命预测方法。采用一种通用的Wiener过程模型对健康状态退化过程进行描述,分别建立多源观测函数和多源映射函数对状态与数据之间的因果关系和关联关系进行描述,采用粒子滤波算法将多传感器数据与模型进行动态匹配,预测剩余寿命。在提出方法的统一框架指导下,选取三种特定模型对铣刀剩余寿命进行预测,验证了提出方法的有效性。With the development of sensing and information technology, massive data have been collected from machinery by various sensors for remaining useful life(RUL) prediction, which has promoted mechanical RUL prediction entering the big data era.Meanwhile, it is still a big challenge to quickly mine degradation information from massive multi-sensor data to guide RUL prediction.Most model-based methods only focus on a one-dimensional degradation data, thus cannot effectively utilize the rich data resources.Data-driven methods are highly dependent on monitoring data. They do not rely on any experimental knowledge. Thus, they suffer from the problem of poor interpretability. To fuse multi-sensor signals for RUL prediction, a model-data-fusion RUL prediction method with multi-sensor fusion is established based on the basic idea of model and data fusion. In this method, the degradation process is expressed by a generalized Wiener process model. The relationship between states and data are interpreted from the aspects of causality and correlation using multivariate measurement function and multivariate mapping function. Multi-sensor data are fused to update the model in real time using the particle filtering algorithm for RUL prediction. Under the basic framework of the proposed method, three specific models are selected to predict the RUL of milling cutters. The prediction results verify the effectiveness of the proposed method.

关 键 词：机械装备 剩余寿命预测 大数据 数模联动 多传感器融合 

分 类 号：TH17[机械工程—机械制造及自动化]