| Liquid Film Seal is an important component of rotating machinery such as aero-engine,centrifugal pump and compressor.Because of its low leakage,low friction coefficient and high reliability,it is widely used in aerospace,petrochemical and other fields.In order to master the friction state and abnormal information before the failure or failure of sealing equipment,it is necessary to study the condition monitoring of liquid film seal.Aiming at the important development direction of nondestructive monitoring of liquid film seal,this paper uses acoustic emission technology as testing means,and constructs the research system of liquid film seal condition monitoring technology through signal feature extraction,experimental research,pattern recognition and other methods.Considering that the AE signal is easily affected by noise and the feature signal is difficult to extract when the AE technology is applied to the condition monitoring of liquid film seal face,a signal processing method based on SVD-AVMD is proposed.Firstly,the influence of random strong noise in the signal is eliminated by singular value decomposition,and then the noise reduction signal is obtained in order to obtain the optimal modal component,the variational modal decomposition is used to decompose the denoised signal under the condition that the significance level between the modal components is greater than the threshold.The test results show that the ability of SVD-AVMD to capture the center frequency of each modal component and the recovery effect of each modal component are significantly better than the simple variational mode decomposition,which can filter the background noise and retain the effective information to the greatest extent.Based on the shear strain energy theory and CEB contact model,the theoretical mechanism of acoustic emission of liquid film seal end face is studied.The correctness of the theoretical mechanism is verified by experiments.The acoustic emission signal which can characterize the friction state of the seal is obtained by SVD-AVMD signal processing method,and the purpose of judging the friction state of liquid film seal end face through the root mean square of acoustic emission signal is achieved.In order to give more intelligent features to the state monitoring of liquid film seal,the time-frequency characteristics of acoustic emission data in different friction states are taken as the input samples of convolution neural network,and the influence of network parameters and samples constructed by different time-frequency analysis methods on the recognition performance of convolution neural network is analyzed.A convolution neural network model suitable for the friction state recognition of liquid film seal is constructed.The experimental results show that the recognition effect of the combination of short-time Fourier transform and convolution neural network is the best,followed by S-transform and wavelet transform,and the recognition accuracy of the convolution neural network model is significantly higher than other conventional recognition methods.Considering the engineering applicability and functional completeness,the program and method in this paper are connected and visualized to develop a liquid film sealing condition monitoring software,which integrates the functions of acoustic emission signal acquisition and processing,convolution neural network identification model building and friction condition identification,and realizes the real-time non-destructive monitoring of liquid film sealing condition. |
