To study the dynamic characteristics of gas-liquid two-phase flow

and to solve the traditional problem that the eigenvalues extracted are so few that is not representative and result in low recognition rate

V-cone flow-meter and dynamic differential pressure sensor were used to obtain fluctuating signals of gas-liquid two-phase flow in different flow patterns

time-frequency analysis of obtained dynamic signals were performed by adaptive optimal kernel(AOK) algorithm

to convert the one-dimensional time-domain signals to three-dimensional time-frequency spectra so as to describe clearly the flow status of gas-liquid two-phase flow in the pipeline. Time-frequency spectra of different flow patterns were learned by the convolution neural network (CNN) to extract the corresponding eigenvalue automatically

then were practiced by Softmax classifier to achieve flow pattern recognition. By experiment and analysis for several common flow patterns

adopting time-frequency spectrum and deep learning method of convolution neural network was found to identify the gas-liquid two-phase flow pattern can overcome the shortcomings of the traditional flow pattern recognition method for few eigenvalue extracted

and describe appropriately the dynamic characteristics of gas-liquid two-phase flow. This method allows further investigation of more types of flow patterns as well as porosity and more.