摘要
以往开发的绝大多数故障诊断算法基于数据的平稳性假设,没有考虑机械某一运转周期内的时间相关细节特征。本文针对特定对象强调了非平稳模型用于信号分析的必要性,讨论了模型的时变,耐不变算法及相应的特征提取、工况判断过程。本文引入正交变换,一方面实现了数据的大规模压缩,另一方面完成了代表正常工况的母体模型的建立;其次,借助于模式识别理论的相似性判据得到对多个特征定量监测的方法。一关于柱塞泵振动监测的实例说明了方法的应用过程。
Most previously developed algorithms used for machinery diagnostics and condition monitoring are based on the assumption of signal stationarity. Few attentions are paid to the timedependent details of vibration signals. In this paper, nonstationary modelling of vibration signals are proposed. Time variant AR model and corresponding processes of feature extraction and condition monitoring are investigated in detail. Then, orthogonal transform is introduced for both feature compression and the establishment of statistical model of the template that represents the normal condition. Lastly, the monitoring measure is determined by means of similarity analysis of pattern recognition theory. An example for the vibration monitoring of hydraulic piston pump shows the application of the proposed method. The method proposed is very suitable to the fault diagnistics and working condition monitoring of the machinery of which the vibration signals are with a quasi-periodic characteristics.
基金
教委博士点基金
浙江大学流体传动及控制开放实验室基金
关键词
机械系统
故障诊断
振动信号
Fault diagnostics, Condition Monitoring, Hydraulic piston pump
