The Research Of Mechanical Fluid Seal End Face Contact State Monitoring

Nowadays along with industrialization constantly processing forward, mechanical seal, as the key technology of fluid machinery, has developed rapidly. To maintain the mechanical seal face at a certain state of film thickness is the key to ensure the normal operation of mechanical seal. In the traditional eddy current method, which is a direct measurement of film thickness, do damage to the internal structure of the sealing is inevitable.Therefore, it’s not beneficial for field application and can not meet the growing demand of modern industry. And the feature of acoustic emission which detect without changing the mechanical seal structure makes a hot spot.Based on the eddy current sensor which can directly measure film thickness signal and acoustic emission sensor which is the method of indirectly measuring film thickness signal., in this paper, we set up an experimental platform of film thickness for the non-contact mechanical seal condition monitoring. Eddy current data and acoustic emission data are collected via the condition of constant pressure with variable speed and variable pressure with constant speed. Considering the limit that the acoustic emission signal can’t directly react film thickness information, we instruct indirect measurement results by the results of direct measurement and divide the acoustic emission signal into three groups according to the degree of film thickness. Then the condition monitoring research of film thickness based on acoustic emission signal is launched.The original acoustic emission signal is processed using zero homogenization. An advanced signal analysis technology, namely ensemble empirical mode decomposition (EEMD), is used for time-frequency analysis.Then time-frequency domain characteristics and related characteristics of each frequency component are extracted. Due to the acoustic emission signal is more sensitive to the environment and the signal contains a large number of random disturbance, it’s usually difficult to extract better characteristics. Using Kernel Principal Component Analysis (Kernel Principal Component Analysis, KPCA) to optimize the feature dimension reduction cannot only reduce the feature dimension reduction the amount of input and calculation, but also weak the interrelation between the nonlinear characteristics.This article uses two advanced theories for model training and identifying the final film thickness, the support vector machine (SVM) and discrete hidden markov model (DHMM). Then we compare the two methods through experiments analysis. Experiments show that DHMM model recognition rate (above 81.7%) is lower than the SVM (above 94.8%). But DHMM has the advantages like modeling simply and having a fast training speed. We can choose between the two methods depending on the emphasis of diagnosis system. Overall, both methods achieve recognition of state of mechanical seal end face film thickness. These studies have a certain practical significance for improving the state of mechanical seal end face film thickness, and also provide a good technical support for industrial application of mechanical seal end face film thickness monitoring.