The Monitoring Research On The Open Process Of Hydrodynamic Mechanical Seals Based On Acoustic Emission Characteristics

Mechanical seal is one of the most common sealing methods in rotating machines. It is widely used in areas of aerospace, nuclear power and petroleum & chemical industry because its high reliability, less leakage, long working life and strong applicability. So the performance of mechanical seals would directly impact the equipment on safety, efficiency and cost during the manufacturing process. Consequently, it is necessary to take real-time condition monitoring of mechanical seals and obtain useful characteristic information, which can be used to analysis the end surface friction state of mechanical seals. Therefore, the failure of seals could be found in time and fixed, avoid the waste of resources and costs causing by premature replacement of seals, or accidents causing by replacing seals too late.By designing the monitoring experiment of the open process of mechanical seals, chosen the sensors of eddy and acoustic emission to monitor the information of film thickness and got the eddy and acoustic emission signals via the whole open process. The corresponding relationship was built between changes of signals, end surface contact state and friction state through the analysis of eddy signals. Divided the open process into three friction states:dry friction state, mixed friction state and fluid friction state. Signals collected in different state performed significant differences. After analyzing the acoustic emission signals, found the correspondence between acoustic emission signals changes and friction states. Then reduced the signal noise by Wavelet Packet Analysis. Concluded that high-frequency signals contained more information about mechanical seals by a contrast experiment conducted in sealed and unsealed environment. Then conducted time-frequency domain feature extraction of high-frequency acoustic emission signals based on apposite characteristic index related to end surface friction characteristics.Selected effective characteristic of acoustic emission signals and normalized them, which possessed good recognition to those three friction states. Then used them as input vectors to Elman neural networks and built four-layer network model with feedback layer. After training the network by test samples, recognized the test samples and gained good results. Chose different training samples and testing samples to build different networks to carry out mode recognition and found they were all good at classifying the different friction states. It is proved that chosen acoustic emission characteristic could well recognize the end surface state of open process of mechanical seals. Optimized the neural network with an improved particle swarm optimization by adding inertia weight factor and matrix, which increased the operational efficiency of the algorithm while increasing its convergence rate. Finally, contrasted the outputs of neural networks before and after optimization, found that the PSO algorithm had obvious good results in many aspects such as training speed and accuracy, convergence rate and state recognition accuracy.