Aimed at the problems of insufficient development of nondestructive testing,difficult in evaluation of signal characteristics and lack of intelligent characteristics of friction state discrimination in the field of liquid film seal condition monitoring,a friction state recognition method of liquid film seal based on acoustic emission time-frequency analysis and convolution neural network (CNN)was proposed.In this method,acoustic emission nondestructive testing technology was applied to the friction state monitoring of liquid film seals,convolutional neural network was used as the means of autonomous decision-making of friction state of liquid film seals,and the time-frequency information of acoustic emission signals was used as the feature input of convolution neural network.The influence of three time-frequency analysis methods of short-time Fourier transform,S transform and wavelet transform on the recognition performance of convolution neural network was investigated.The results show that for the acoustic emission signal of liquid film seal,the optimal order of the combination of three time-frequency analysis methods and convolutional neural network is short-time Fourier transform,S transform,wavelet transform.The recognition method of friction state recognition of liquid film seal based on acoustic emission time-frequency analysis and convolution neural network has high accuracy,and can achieve good recognition results compared with other recognition methods.