摘要
传统滚动轴承故障识别算法存在特征提取与特征选择困难的问题,为此,提出了一种基于完备集成辛几何分解(CESGD)和改进多通道卷积网络(IMCCN)的滚动轴承故障识别方法。首先,在采集到的滚动轴承振动信号中,添加符号相反且幅值相等的正负白噪声对,利用辛几何分解(SGD)将轴承多传感器振动信号分解为若干辛几何模态成分(SGMCs),并进行了集成平均;利用评价指标选择较能反映轴承故障特征的SGMCs分量并重构,进而完成了对轴承振动信号的有效降噪;然后,在卷积神经网络基础上提出了IMCCN结构,并将CESGD降噪后的振动信号输入IMCCN进行自动特征学习与故障识别;最后,为验证CESGD-IMCCN模型的可行性和有效性,在轴承故障模拟实验台以及CWRU轴承数据集上对此进行了测试,并将结果与采用其它方法获得的故障识别结果进行了对比分析。研究结果表明:基于CESGD-IMCCN的模型能够对不同故障工况及故障严重程度类型的滚动轴承进行有效识别和分类,其故障识别率达到99.52%,且标准差仅为0.12;CESGD-IMCCN模型在一定程度上避免了复杂的人工特征提取过程,其故障识别准确率和稳定性较高,在泛化能力、特征提取能力和故障识别能力方面比其他组合模型更具明显优势;对于含有噪声的滚动轴承振动信号,其故障识别准确率依然较高。
Considering the difficulties of traditional rolling bearing faults identification methods in manual feature extraction,manual feature selection and fault identification of rolling bearing vibration signals,a method based on complete ensemble symplectic geometric decomposition(CESGD)and improved multi-channel convolutional network(IMCCN)was proposed.Firstly,in order to improve the signal-to-noise ratio(SNR)of bearing vibration signals based on the idea of complete ensemble empirical mode decomposition,the white noise pairs with equal amplitude and opposite signs were added to vibration signals;the symplectic geometric decomposition(SGD)was employed to decompose the bearing multi-sensor vibration signals into several symplectic geometric mode components(SGMCs)and the integrated average was performed.The SGMCs which could better reflect the fault characteristics of the rolling bearing were selected and reconstructed by evaluation index to reach the purpose of effective signals noise reduction.Then,the IMCCN was proposed on the basis of convolutional neural network(CNN),and the noise-reduced signals of CESGD were input into IMCCN for automatic feature learning and fault identification.Finally,in order to verify the feasibility and effectiveness of the CESGD-IMCCN model,it was tested on the bearing failure simulation test bench and the CWRU bearing data set,and the results were compared and analyzed with the failure recognition results obtained by other methods.The research results show that the proposed CESGD-IMCCN model can effectively identify different fault conditions and severity types of rolling bearing,the fault recognition rate reaches 99.52%and the standard deviation is only 0.12.The CESGD-IMCCN model avoids the complex manual feature extraction process to a certain extent;with high fault recognition accuracy and stability,it shows obvious advantages over other models in terms of generalization ability,and fault identification ability.For vibration signals containing noise,the fault recognition accuracy is still high.
作者
沈为清
周正平
常兆庆
SHEN Wei-qing;ZHOU Zheng-ping;CHANG Zhao-qing(School of Intelligent Engineering and Technology,Jiangsu Polytechnic of Finance and Economics,Huai'an 223003,China;College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;Jiangsu Shuguang Electro-Optics Co.,Ltd.,Yangzhou 225000,China)
出处
《机电工程》
CAS
北大核心
2021年第12期1579-1585,1598,共8页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(51775272)。
关键词
滚动轴承
故障识别
完备集成辛几何分解
改进多通道卷积网络
rolling bearing
fault identification
complete ensemble symplectic geometric decomposition(CESGD)
improved multi-channel convolutional network(IMCCN)
