混合密度连续HMM在旋转机械启动过程故障诊断中的应用  

Faults Diagnosis of the Running-up Process of a Rotary Machine Based on HMM with Mixture Probability Densities

作　　者：邵强[1,2] 冯长建[2] 管丽娜[2] 邵诚[1] 

机构地区：[1]大连理工大学先进控制技术研究所,大连116024 [2]大连民族学院机械系,大连116600

出　　处：《机械科学与技术》2009年第11期1439-1443,共5页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(50405023)资助

摘　　要：故障类型的振动信号往往表现为非平稳的特征,这些信号经过短时分割并提取AR系数,从而表现为一有序的AR系数矢量的观测矢量。论文根据混合密度连续HMM(CDHMM)的动态统计模式识别的基本理论,把这些观测矢量由几个高斯混合概率密度函数的线性组合进行模拟,从而对每种故障的动态模式建立起的CDHMM,并根据模型的输出概率进行故障识别尝试。Vibration signals of a rotary machine in the running-up process contain important information related to the healthy state. Vibration signals of fault types usually have non-stationary features. These signals are segmented to a series of signals of short time and extracted by AR model, therefore a sequential observation vectors of AR coefficients are formed. Based on the theory of continuous hidden Markov models with mixture probability densities（CDHMM）, these observation vectors are considered as the combination of several Gauss probability density functions, so multitype simulation faults in the running-up of a rotary machine are molded by CDHMMs, and fault identification is carried out by the output probabilities of CDHMMs. Experiments verified that the proposed method is effective.

关 键 词：CDHMM 故障诊断 振动信号 旋转机械 

分 类 号：TN911.6[电子电信—通信与信息系统] TH165.3[电子电信—信息与通信工程]