基于MDS和神经网络的滚动轴承故障诊断方法

Rolling Bearing Fault Diagnosis Method Based on MDS and Neural Network

针对滚动轴承的智能诊断问题,提出基于多维尺度分析(Multidimensional Scaling,简称MDS)和神经网络的滚动轴承故障诊断方法。该方法首先提取原始信号常用的时域统计指标,再将包含故障信息的统计指标进行MDS降维处理,减少后续模式识别难度,最后将降维后的统计指标作为神经网络的输入参数来判断滚动轴承的故障类型。对滚动轴承正常状态、滚动体故障、外圈故障和内圈故障四种模式下的振动信号进行分析,结果表明,运用MDS进行降维预处理的神经网络故障诊断方法比没有经过预处理的故障诊断方法有更高的故障识别效率,可以准确有效识别滚动轴承的故障类型。

A new fault diagnosis method for rolling bearings based on Multidimensional Scaling(MDS)and neural network is put forward.First of all,several time-domain statistics indexes of rolling bearings are extracted from original signals.Then,the indexes containing fault information are processed by MDS to reduce the data dimension.Finally,the low dimensional characteristic indexes are served as input parameters of neural network to identify fault patterns of the rolling bearings.The analysis results from rolling bearing signals with rolling element,inner-race and out-race faults show that the approach of neural network diagnosis based on MDS is superior to that without MDS and can identify roller’s fault patterns effectively.