摘要
针对传统卷积神经网络自适应能力弱,难以适应变工况、少样本和类别不平衡等实际工业应用场景问题,提出了一种基于谱相关密度(SCD)和卷积神经网络(CNN)的轴承故障诊断SCD-CNN方法。该方法利用谱相关密度能有效提取轴承故障特征的能力,提高CNN的泛化能力。文章分析了SCD-CNN模型在变负载工况、少样本和类别不平衡(class imbalance)条件下模型的泛化性能,使用故障类型和非故障类型平衡率不同的样本数,对SCD-CNN模型进行干扰,模型在变负载工况下泛化性能良好;在少样本条件下,使用少量样本进行训练,就可获得较稳定的模型;在样本数据类别不平衡条件下,模型对少数类故障样本的识别能力较强。将SCD-CNN与STFT-CNN和CWT-CNN方法进行对比,结果表明:SCD-CNN的泛化性能良好,其性能优于STFT-CNN和CWT-CNN方法。
A SCD-CNN method for bearing fault diagnosis based on spectral correlation density(SCD) and convolutional neural network(CNN) is proposed to address the problem that the traditional convolutional neural network has a weak adaptive capacity which makes it difficult to adapt to practical industrial application scenarios,such as variable working conditions,few shot and class imbalance. In order to improve the generalization ability of CNN,spectral correlation density is used to extract bearing fault features effectively. The generalization performance of the SCD-CNN model under variable load conditions,few shot and class imbalance is analyzed. Different sample numbers of fault type and class imbalance rate are used to interfere with the SCD-CNN model. The model has better generalization performance under variable load conditions. Under the condition of few shot,a more stable model can be obtained using a small number of samples for training.Under the condition of class imbalance sample data,the SCD-CNN model has better fault identification capability for a few fault samples. By comparing SCD-CNN with STFT-CNN and CWT-CNN,the experimental results show that SCD-CNN has good generalization performance,which is better than that of STFT-CNN and CWT-CNN.
作者
李辉
徐伟烝
LI Hui;XU Weizheng(School of Mechanical Engineering,Tianjin University of Technology and Education,Tianjin 300222,China)
出处
《天津职业技术师范大学学报》
2022年第2期7-14,共8页
Journal of Tianjin University of Technology and Education
基金
国家自然科学基金资助项目(51375319)
芜湖市科技计划项目(2021jc1-6).
关键词
故障诊断
谱相关密度
卷积神经网络
短时傅里叶变换
连续小波变换
轴承
fault diagnosis
spectral correlation density(SCD)
convolutional neural network(CNN)
short-time Fourier transform(STFT)
continuous wavelet transform(CWT)
bearing
