摘要
针对由于受噪声干扰导致采用传统方法所提取的故障特征易缺失以及在复杂的工况环境下故障无法完全分类识别的问题,提出一种改进RegNet的滚动轴承故障诊断方法。首先,利用连续小波变换(Continuous Wavelet Transform,CWT)将一维振动信号转换为二维时频图,并将其作为模型的输入;其次,构建含有注意力机制的改进调节残余网络模型(Improved-Regulated Residual Networks,IRegNet):利用3×3卷积核对特征进行预处理,在保持空间分辨率的同时减少输入模型参数,提高模型的训练效率;将ConvLSTM作为调节器记录时空信息,与残差结构并行,使其在专注学习自身特征的同时捕获更长期的时空依赖信息;融入混合注意力机制(Convolutional Block Attention Module,CBAM)兼顾通道和空间特征信息,增强特征提取能力。试验结果表明,相较于其他模型,所提方法精度更高,在复杂的实验条件下分类效果更好。
Aiming at the problems that the fault features extracted by traditional methods are easily missing due to noise interference and faults unable to be completely classified and identified in complex working conditions,an improved rolling bearing fault diagnosis method based on RegNet is proposed.Firstly,continuous wavelet transform is used to convert one-dimensional vibration signals into two-dimensional time-frequency diagrams,which is used as the input of the model.Secondly,an improved regulated residual network model(IRegNet)containing an attention mechanism is constructed.A 3×3 convolution kernel is used to pre-process the features to reduce the input of model parameters while maintaining the spatial resolution and improving the training efficiency of the model.Finally,ConvLSTM is used as a regulator to record time-spatial information in parallel with the residual structure,so as to enable it to capture longer-term time-spatial dependency information while focusing on learning its own features;the Convolutional Block Attention Module hybrid attention mechanism is integrated to take into account both channel and spatial feature information and enhance the feature extraction ability.The experimental results show that compared with other models,the proposed method has higher accuracy and also achieves better classification results under complex experimental conditions.
作者
赵翼帆
孙士保
王国强
石念峰
赵一鸣
杨向兰
ZHAO Yifan;SUN Shibao;WANG Guoqiang;SHI Nianfeng;ZHAO Yiming;YANG Xianglan(School of Information Engineering,Henan University of Science and Technology,Luoyang 471000,Henan,China;School of Computer and Information Engineering,Luoyang Institute of Technology,Luoyang 471000,Henan,China;Henan Key Laboratory of Green Building Materials Manufacturing and Intelligent Equipment,Luoyang Institute of Technology,Luoyang 471000,Henan,China)
出处
《噪声与振动控制》
北大核心
2026年第1期157-163,共7页
Noise and Vibration Control
基金
河南省自然科学基金项目(232102210065)
龙门实验室前沿探索课题项目(LMQYTSKT034)。
关键词
故障诊断
滚动轴承
深度残差网络
特征提取
卷积递归神经网络
fault diagnosis
rolling bearing
deep residual network
feature extraction
convolutional recurrent neural network
