摘要
根据振动与语音信号的相似性和离心泵故障信号的特点,将连续隐马尔可夫模型引入了离心泵的故障诊断中。利用自回归谱不受数据长度的限制,及自回归模型参数对状态变化规律反映敏感的特点,以信号的12阶自回归谱系数为特征矢量,将其输入到各个状态连续隐马尔可夫进行训练,来实现离心泵的故障诊断。为防止数据下溢,引入前向–后向比例因子算法求其对数似然概率,并且采用K-means算法对连续隐马尔可夫进行参数初始化。在给定的观测序列中每一种模型的优化路径通过Viterbi算法实现,用Baum-Welch算法实现参数重估。最后通过2BA-6A离心泵试验系统验证了该方法的有效性。
According to the similarity between vibration signal and speech signal, a new method with fault diagnosis of for centrifugal pump based on continuous hidden markov model （CHMM） was introduced. The autoregressive （AR） spectrum was not restricted by length of data, and AR spectrum parameters was sensitive to law of condition change. 12 rank AR spectrum coefficients of signals are considered as feature vectors of running state of centrifugal pump to train in each CHMM, fault classification can be made. Forwards-backwards algorithm was introduced to calculate log-likelihood avoiding the data to underflow and K-means algorithm was also used to initialize the parameter. In the given observation sequence, every model was optimized with Viterbi algorithm, and parameters were re-estimated with Baum-Welch algorithm. The method was tested with the experimental data collected from the 2BA-6A centrifugal pump experimental system and the result demonstrates that the model is effective to classify classical faults.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2008年第20期88-93,共6页
Proceedings of the CSEE
基金
吉林省教育厅科学技术研究项目资助(2007047)
关键词
离心泵
故障诊断
连续隐马尔可夫模型
自回归谱分析
centrifugal pump
fault diagnosis
continuous hidden Markov model
autoregressive spectrum
