一种增量式半监督VPMCD齿轮故障在线诊断方法

A novel incremental semi-supervised VPMCD gear fault on-line diagnosis method

针对齿轮故障诊断中难以获得大量故障样本的问题及实时在线诊断的需求,提出了一种基于增量式半监督多变量预测模型(Incremental Semi-supervised Variable Predictive Model based Class Discriminate,ISVPMCD)的齿轮故障在线检测方法。首先使用VPMCD方法给少量的已知样本建立初始预测模型,接着利用VPMCD方法中的判据给未标识样本赋予初始伪标识,然后通过互相关准则筛选出伪标识样本,最后利用伪标识样本和已知样本作为训练样本更新初始预测模型,使得更新的预测模型能兼顾整个样本集的信息,从而可以有效地解决小样本的故障诊断问题,另外,由于该方法在实时更新新样本的过程中不需要再次建立判别模型,从而缩短了分类时间,为实时在线诊断提供了新的思路。对UCI标准数据以及齿轮实测数据的分析结果表明,适合于小样本的ISVPMCD模式识别方法可以更快更准确地识别齿轮工作状态和故障类型。

Aiming at the problem that getting a large amount of fault samples is difficult and the demand of real-time online diagnosis for gear fault diagnosis,a novel incremental semi-supervised variable predictive mode-based class discriminate （ISVPMCD）gear fault on-line detection method was put forward here.Firstly,the VPMCD approach was used to establish an initial prediction model for a small number of labeled samples.Secondly,the criterion of VPMCD was used to provide initial pseudo labels for unlabeled samples.Thirdly,the pseudo labeled samples were screened with the cross-correlation rule.Finally,the pseudo labeled samples and labeled samples were taken as the training samples to update the initial prediction model,so that the global information of the whole sample set could be considered,and the problem of fault diagnosis of a small set of samples could be solved effectively.In addition,the method did not need to establish a discriminant model in the process of real-time updating new samples,it shortened the time of classification and offered a new way for real-time online diagnosis.The analysis results of the UCI standard data and the test data of gears showed that the ISVPMCD pattern recognition method being suitable for small samples can be used to identify the gear working state and fault type much more quickly and accurately.