基于深度学习的滚动轴承早期微弱故障声发射信号特征提取算法研究

项目名称： 基于深度学习的滚动轴承早期微弱故障声发射信号特征提取算法研究

项目编号： No.51505234

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 机械、仪表工业

项目作者： 赵晓平

作者单位： 南京信息工程大学

项目金额： 19万元

中文摘要： 滚动轴承早期微弱故障信号具有潜在性和动态响应的微弱性等特点，特征提取是限制早期故障诊断发展的关键问题。针对这一问题，本项目拟从故障机理的研究和影响特征提取的关键因素入手，开展如下主要研究：1) 探究故障源特性、声波传播和衰减机理及传感器的布置等参量对声信号的影响，构建故障源、噪声和采集到的声信号之间的关系模型。2）探索粒子滤波原理，构建观测路径相似性的自适应粒子滤波器，抑制背景噪声对特征提取的干扰；3）研究深度学习算法，结合滚动轴承声发射信号的特点，构建滚动轴承早期微弱故障声信号的含有多隐层的构架模型，提高特征提取方法的性能；4）根据研究内容1）、2）和3）所得到的声信号特征参量，建立滚动轴承早期微弱故障特征提取框架。本课题将通过仿真实验与实物试验相结合的方式进行验证，预计为机械设备早期微弱故障诊断提供理论基础和应用支撑。

中文关键词： 早期微弱故障；故障机理；特征提取；深度学习；粒子滤波

英文摘要： Rolling bearing early weak fault signal characteristics and dynamic response of the weak potential, feature extraction is the key problem of the development of early fault diagnosis. In order to solve this problem, this project carry out the following research: 1) to explore the influence of parameters to acoustic mission, such as the source of trouble characteristics, acoustic wave propagation and attenuation mechanism and the distribution of sensor, to build the relationship between the fault source, noise and acoustic emission. 2) to explore the principle of particle filter, build path similarity of particle filter, and suppress background noise interference of feature extraction; 3) research on deep learning algorithms, combining with the characteristics of rolling bearing acoustic emission signal, to build a many hidden layer structure model and improve the performance of feature extraction method; 4) according to the research content 1), (2) and (3) of acoustic signal characteristic parameters, establish rolling bearing in the early fault feature extraction framework. This topic will be through the way of combining simulation and physical experiment validate, early weak fault diagnosis for mechanical equipment is expected to provide theoretical basis and application support.

英文关键词： Early weak fault;Failure mechanism;Feature extraction;Deep learning;Particle filter