基于DHMM的机械密封端面膜厚识别技术的研究

维持机械密封端面间一定的膜厚是保证机械密封正常运行的关键,利用声发射技术监测得到的反映机械密封膜厚状态的信号往往信噪比很低,对其工作状态进行分类存在一定的困难。提出一种基于声发射信号利用总体经验模式分解(EEMD)和离散隐马尔可夫模型(DHMM)识别的机械密封端面膜厚识别技术。首先对声发射信号进行分帧处理,运用EEMD方法对信号进行时频分析,对分解出的子频分量分别提取时域和频域特征,再由核主成分分析法对特征参数进行优化降维,利用简化后的特征参数矢量训练各个机械密封端面膜厚状态的DHMM,最后由训练好的DHMM实现机械密封端面膜厚状态的识别,从而实现机械密封端面接触状态的监测。试验研究表明:该方法能够快速有效地判断出膜厚状态,并且需要的训练样本少,训练速度快,对实现机械密封端面接触状态的智能化在线监测具有重要的意义。

Mechanical Seal End Face Film Thickness State Recognition Based on DHMM

To maintain the mechanical seal end face at a certain state of film thickness is the key to ensure the normal operation of mechanical seal. The signal acquired by Acoustic Emission( AE) method to monitor the mechanical seals usually has a low SNR signal,which makes the classification of the working condition of mechanical seals difficult. A new method of mechanical seal end face film thickness state recognition was proposed based on the Acoustic Emission signals,and the Ensemble Empirical Mode Decomposition( EEMD) and Discrete Hidden Markov Model( DHMM) were introduced into the mechanical seal work condition monitoring analysis. First,the AE signals were decomposed by EEMD after they were divided into some equal frames. Then time domain characteristics and frequency domain characteristics of each frequency component were extracted. Secondly,the processing of the characteristic parameter reduction was optimized by Kernel Principal Component Analysis( KPCA). The simplified characteristic parameter vectors were used to train the DHMM of each film thickness state. Finally,mechanical seal end face contact state monitoring could be achieved by trained DHMM. Studies have shown that this method can identify the film thickness state of mechanical seal end face effectively and quickly with less samples needed and fast training rate. It is important to the development of intelligent online monitoring of mechanical seal end face contact state.