互信息规范的卷积神经网络及其在轴承故障特征提取中的应用
Convolutional Neural Network of Mutual Information Specification and Its Application in Bearing Fault Feature Extraction
王振亚 1刘韬 1伍星2
作者信息
- 1. 昆明理工大学机电工程学院 昆明 650500;云南省先进装备智能制造技术重点实验室 昆明 650500
- 2. 云南省先进装备智能制造技术重点实验室 昆明 650500;云南机电职业技术学院 昆明 650500
- 折叠
摘要
近年来,各种深度学习模型在故障诊断领域取得了突破性进展.该类方法能自动的从复杂的数据捕捉到设备在不同状态的规律,但其"不透明"特点导致了使用者对模型的不信任.从模型的拟合角度出发,提出一种通过互信息规则来规范模型输入输出之间关系的卷积神经网络模型.通过改进模型的损失,增强网络模型的特征提取能力,并将结果可视化.首先,将振动数据转换为包络谱作为模型第一层的输入;然后将用于特征提取的多层卷积视作一个系统,并设计该系统的输出与模型的输入大小一致使得模型便于理解;其次,添加注意力机制层以增强明显特征;最后,基于信号降噪前后之间的互信息规律设计了损失函数来驱动卷积系统可以提取更多故障特征频率成分.通过仿真信号与真实案例表明:所提方法的可视化结果不仅具有一定的解释性,并且在轴承的故障特征提取方面取得了一定的效果.
Abstract
From the perspective of model fitting,we propose a convolutional neural networks model that regulates the relationship between model inputs and outputs through mutual information rules.The loss of the model is improved to enhance the feature extraction ability of the bearings and visualize the results.First,the vibration data is converted into an envelope spectrum as the input of the model so that the input layer has some physical meaning;then the multilayer convolution used for feature extraction is treated as a system and the output of the system is designed to be the same size as the input of the model so that the model is easy to understand;second,the attention mechanism layer is added to enhance the apparent features;finally,a loss function is designed to drive the convolutional system based on the mutual information law between the signal before and after noise reduction to extract more frequency components of the fault features.The simulated signals and real cases show that the visualization results of the proposed method are not only interpretable,but also effective in the extraction of fault features of bearings.
关键词
卷积神经网络/故障特征提取/互信息/滚动轴承/特征可视化Key words
convolutional neural networks/fault feature extraction/mutual information/rolling bearing/feature visualization引用本文复制引用
基金项目
国家自然科学基金(52065030)
云南省重大科技专项计划(202202AC080008)
云南省教育厅重点(KKDA202001003)
出版年
2024
NETL
NSTL
万方数据