摘要
针对滚动轴承故障数据时频特征提取以及故障在变负载环境下难以准确诊断的问题,提出一种基于小波变换和残差神经网络的轴承故障诊断模型。残差神经网络(ResNet)最初用来解决模型网络层数过深而导致的退化问题,本文对传统残差网络进行优化在输入层使用DropBlock层对数据进行随机失活处理,并且在输出层使用全局平均池化代替全连接减少模型的运算量增加模型的泛化能力。首先对一维轴承数据进行小波变换,将其转化为二维时频图,消除手工特征的影响。然后使用本文构建的残差网络模型对同负载和变负载下的数据进行故障类型诊断。实验结果证明:本文模型对测试集的准确率不仅在同负载能达到99.69%以上而且在工况差异最大的情景下也能达到96.07%以上,具有良好的泛化性和鲁棒性。
Aiming at the problem of time-frequency feature extraction of rolling bearing fault data and the difficulty of accurately diagnosing faults un⁃der variable load environment,a bearing fault diagnosis model based on wavelet transform and residual neural network is proposed.The ResNet was originally used to solve the degradation problem caused by too many layers of the model network.In this paper,the traditional residual network is optimized to use the DropBlock layer to randomly inactivate the data at the input layer and to use it at the output layer.Global average pooling instead of full connection reduces the computational complexity of the model and increases the generalization abili⁃ty of the model.First,wavelet transform the one-dimensional bearing data and convert it into a two-dimensional time-frequency map to eliminate the influence of manual features.Then use the residual network model constructed in this paper to diagnose the fault type of the data under the same load and variable load.The experimental results prove that the accuracy of the model on the test set can reach more than 99.69%at the same load and more than 96.07%in the scenario with the largest difference in working conditions.It has good general⁃ization and robustness.
作者
杨腾
宁芊
陈炳才
YANG Teng;NING Qian;CHEN Bingcai(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065;School of Computer Science and Technology,Dalian University of Technology,Dalian 116024)
出处
《现代计算机》
2021年第15期82-88,共7页
Modern Computer
基金
新疆自治区区域协同创新专项(科技援疆计划)(No.2019E02142)
国家自然科学基金(No.61771089)。
