基于时序关联分析的旋转机械故障诊断

Fault diagnosis of rotating machinery based on time-series correlation analysis

随着大型旋转机械设备的广泛应用,高速旋转机械的故障诊断得到越来越多的关注。旋转机械的周期旋转特性导致信号间存在很强的时序关联关系,当故障发生时,故障特性会在旋转周期间逐渐传递。该研究分析滚动轴承不同类型故障、不同损伤程度振动信号时序相关特性的差异度,提出了周期记忆神经网络(periodization long short-term memory,P-LSTM)故障诊断方法。该方法首先提取旋转机械周期内数据特征,并利用记忆因子对特性在周期间的传递规律进行选择性遗忘,学习其周期间的时序相关特征,从而实现滚动轴承的故障诊断。最后利用滚动轴承多类故障数据对所提出方法进行性能分析和试验,验证了P-LSTM方法学习旋转机械周期间的时序相关特性的有效性,以及进行故障诊断的准确度。

With the application of large-scale rotating machinery,more and more attention has been paid to fault diagnosis of high-speed rotating machinery.Due to the periodic rotation characteristics of rotating machinery,there is a strong temporal correlation between signals.Fault features will gradually transfer during the week of rotation.In this paper,different types of faults and different levels of damages of rotating machinery were analysed,which refers to temporal association characteristics of vibration signal.Then Periodization long short-term memory(P-LSTM)fault diagnosis method was proposed.The method extracts features from periodization data and uses memory factors to forget some information that is of insignificance.Finally,the performance analysis and test of the proposed method were carried out based on the multi-fault data of rolling bearings,which verified the effectiveness of P-LSTM method in learning the time-series correlation characteristics of rotating machinery during the cycle,as well as the accuracy of fault diagnosis.