摘要
针对自适应共振理论2(ART2)神经网络在分类时存在只选择输出值最大的神经元以及噪声对结果影响较大的缺点,提出一种结合小波软阈值和K均值算法的ART2神经网络分类方法;采用小波软阈值对滚动轴承故障信号进行降噪,并通过相对小波包能量体现降噪后的信号更好的信号互异性,然后运用ART2神经网络进行初步分类,将K均值算法引入ART2神经网络,对原有的算法进行修正,并与ART2神经网络分类结果进行对比。结果表明,改进的方法解决了上述的问题,提高了故障分类的准确性。
In view of the adaptive resonance theory 2 ( ART2) neural network only chose the neuron with the largest output value when classifying and noise had a great impact on the result,a classification method by using ART2 neural network combined with wavelet soft threshold and K -means algorithm was proposed for problem solving.The wavelet soft threshold was used to denoise rolling bearing fault signal noise.The denoised signal was applied to reflect the better signal heterogeneity by using the relative wavelet packet energy method.The ART2 neural network was used to perform preliminary classification.The K -means algorithm was used to modify the classification results of ART2 neural network,and the results were compared with the results of ART2 neural network.The results show that the improved method solves the above problems,which improves the accuracy of failure classification.
作者
吴肇中
郝如江
陆一鹤
WU Zhaozhong;HAO Rujiang;LU Yihe(Handan Locomotive Depot,China Railway Beijing Group Co.,Ltd.,Handan 056003,China;School of Mechanical Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)
出处
《济南大学学报（自然科学版）》
CAS
北大核心
2019年第6期537-543,共7页
Journal of University of Jinan(Science and Technology)
基金
国家自然科学基金项目(51375319)
河北省引进留学人员资助项目(CL201721)
关键词
小波软阈值
ART2神经网络
K均值算法
相对小波包能量
wavelet soft threshold
ART2 neural network
K -means algorithm
relative wavelet packet energy
