摘要
由于铁路轴承的易损性和故障信号提取的复杂性,文中应用了多方法融合的诊断法对铁路轴承进行故障诊断,并对基于多方法融合的BP和RBF两种网络进行了比较。首先,对采集到得信号进行FIR降噪,再对降噪后的信号进行小波包分解,构造特征向量,以此为故障样本对BP和RBF网络进行训练,实现智能化故障诊断,实验结果表明文中提出的方法能很好地诊断出轴承故障类型,但多方法融合的RBF的泛化能力优于BP网络,同时,在训练时间上,RBF网络也要优于BP网络,这为机械故障诊断提供理论依据。
Since the railway beating＇s vulnerability and the fault signal extraction＇s complexity, multi-method fusion techniques for fault diagnosis were applied, BP and RBF networks based on multi-method fusion techniques were compared. First, the signal is processed through the signal de-noising based on FIR. Then, three-layer wavelet packet is adopted to decompose the de-noising signal of rolling bearing, and the wavelet packet energy eigenvector is construc- ted. Taking those wavelet packet energy eigenvectors as fault samples. BP and RBF neural network sare trained. Final- ly the intelligent fault diagnosis is realized. The experiment result shows that the trained BP and RBF neural network based on multi-method can diagnose the kind of rolling hearing faults, However, The result indicates that the generalization capability of RBF is superior to that of BP. Meanwhile, In the training time, RBF is also superior to that of BP network . It provides the theoretical foundation for machine fault diagnosis.
出处
《机械设计与研究》
CSCD
北大核心
2010年第3期70-73,共4页
Machine Design And Research
基金
国家863计划资助项目(2007AA11Z247)
山西省青年基金(2007021023)
中国博士后基金资助项目(2009045290)
