基于特征融合与DBN的矿用通风机滚动轴承故障诊断

Fault diagnosis of rolling bearing of mine ventilator based on characteristic fusion and DBN

针对现有矿用通风机滚动轴承故障诊断方法仅提取时频分量特征和采用浅层网络结构,导致故障诊断精度不高的问题,提出了一种基于多域特征融合与深度置信网络(DBN)的矿用通风机滚动轴承故障诊断方法。该方法首先对轴承振动信号进行小波包降噪处理,对降噪后的轴承振动信号进行时域特征、频域特征、IMF能量特征提取,得到相对全面的高维特征集;然后通过基于类内、类间标准差的特征筛选方法剔除对分类无效及效果不明显的特征,筛选出高效特征;最后采用核主成分分析(KPCA)对高维筛选特征进行降维融合,消除特征间冗余,将融合特征输入至DBN中完成故障诊断。实验结果表明,相比于基于特征单一和浅层网络的诊断方法,基于多域特征融合与DBN的矿用通风机滚动轴承故障诊断方法平均准确率最高,平均诊断时间最少,对于不同损伤故障数据表现出良好的稳定性和泛化能力。

The existing mine ventilator rolling bearing fault diagnosis method only extracts time-frequency component characteristics and adopts shallow network structure,thus causing low fault diagnosis accuracy.In order to solve this problem,a fault diagnosis method of rolling bearing of mine ventilator based on multi-domain characteristic fusion and deep belief network(DBN)is proposed.Firstly,the method performs wavelet packet noise reduction on the bearing vibration signal,and extracts time domain characteristics,frequency domain characteristics and IMF energy characteristics from the bearing vibration signal after noise reduction to obtain a relatively comprehensive set of high-dimensional characteristic set.Secondly,the characteristic selection method based on intra-class and inter-class standard deviation is used to eliminate the characteristics that are not effective for classification and characteristics with no obvious effect so as to filter out high-efficiency characteristics.Finally,kernel principal component analysis(KPCA)is used to reduce and fuse the high-dimensional screening characteristics,eliminate the redundancy between characteristics,and input the fused characteristics into DBN to complete the fault diagnosis.The experimental results show that compared with the diagnosis method based on single characteristic and shallow network,the average accuracy of mine ventilator rolling bearing fault diagnosis method based on multi-domain characteristic fusion and DBN has average accuracy and less average diagnosis time showing good stability and generalization ability for different damage fault data.