基于自回归-隐马尔可夫模型的离心泵故障诊断方法研究

Research on Fault Diagnosis Methods for Centrifugal Pump Based on Autoregressive and Hidden Markov Model

根据离心泵故障振动信号的特点,提出了一种结合自回归(autoregressive,AR)谱分析与隐马尔可夫模型(HMM)的离心泵故障诊断方法。利用AR谱不受数据长度限制和AR模型参数对状态变化规律反映敏感的特点,以振动信号做自回归变换后的AR谱系数作为特征向量,将其输入到各个状态HMM进行训练,其中输出概率最大的状态即是离心泵的运行状态,从而实现离心泵的故障诊断。最后通过2BA-6A离心泵实验系统验证了该方法的有效性。

According to the characteristics of fault vibration signal, a new method for centrifugal pump based on autoregressive （AR） spectrum and hidden Markov model （HMM） was produced. The AR spectrum is not restricted by length of data, and AR spectrum parameters are sensitive for law of condition change. AR transformation was made to vibration signals, and then AR spectrum coefficients were got, which can be utilized as feature vectors of running state of centrifugal pump to train in each HMM, fault classification can be made according to the maximum-likelihood probability. The method was tested with the experimental data collected from the 2BA--6A centrifugal pump experimental system and the results demonstrate that the model is effective to classify classical faults.