基于隐马尔科夫模型的旋转机械故障诊断新方法的研究

It is very important to assure rotating machines to run under health condition. However, the technology of condition monitoring and fault diagnosis of rotating machines is awaiting to be improved, considering many existing problems in it. Theoritically, the vibratory signals collected from a machine under up-speed and down-speed contain plenty of information capable of manifesting the running condition of the machine, which is especially valuable for fault diagnosis. Aiming at such difficulties as too much information, nonstationarity and bad repeatitation of sympotom in vibration of rotating machine under up-speed and down-speed, it is necessary to find the appropriate methods for feature extraction and fault patterns recognition. HMM is a kind of statistic model for time series analysis, and poweful in classification of fault patterns expressed by time series. The target of this research is to introduce HMM into fault diagnosis of machines, and to develop the new method for fault diagnosis of rotating machines. In recent two years, our team has devoted to explore the theory and method of HMM as dynamic pattern recognition and its application to fault diagnosis of rotating machines, developed the software of fault diagnosis based on HMM and verified this method by means of the Bently rotor kit. Our reserch is very significant for development of fault diagnosis of rotating machines.

针对旋转机械升降速过程振动信号信息量大、非平稳、重复性差的特点，引入隐马尔科夫模型作为建模与识别工具，进行旋转机械运行状态和故障识别的研究。隐马尔科夫模型具有强大的动态时序模式分类能力，有望与神经网络、遗传算法等结合在升降速过程的故障诊断中得到成功应用，对于促进旋转机械故障诊断技术的进步具有重要理论意义和使用价值。