Flow Pattern Identification Of Gas-liquid Two Phase Flow In Rod Bundle Channel Based On Entroy Analysis

Gas-liquid two phase flow phenomenon exists widely in all kinds of industrial heat exchange equipment,and the gas liquid two phase flow in the rod bundle channel is common in the fuel assembly of nuclear power plant.Flow pattern has an important influence on the operation and safety of gas-liquid two phase flow equipment,and the flow pattern and flow pattern identification method are different from that of common circular tube.In this paper 7×7rod bundle channel acquisition of the two-phase flow pressure difference signal,through multi-scale entropy and multi-scale entropy marginal spectrum two time-frequency entropy method,the entropy rate under small scale had revealed the flow characteristics of flow signals from the microscopic details,the entropy rateunder large scale had fininshed the flow pattern identification from the macroscopic;Then,the joint distribution of multi-scale entropy rate,multi-scale marginal spectral entropy rate and spectral entropy value as the feature entropy were optimal combination of flow pattern identification based on artificial neural network.The multi scale entropy method was used to identify the flow pattern,andthe numerical distribution of multi-scale entropy rate under 104 flow conditions were calculated.We found that the overall recognition rate of 97.11%,only 3 bubble-churn flow not in the regional division,because the bubble-churn flow belongs to the transitional flow characteristic is not obvious,not easy to distinguish,the main pattern of bubble flow and churn flow recognition rate could reach100%.The multi-scale marginal spectral entropy was used to identify the flow pattern,calculatingthe numerical distribution of multi-scale entropy rate under 104 flow conditions,by using multiple scale marginal spectrum entropy generation rate and spectral entropy mean joint distribution can accurately distinguish 4 kinds of flow.To the bubble-churn flow identification had a good effect,the recognition rate is up to 100%.For pattern recognition feature entropy combined with artificial neural network recognition,multi-scale marginal spectrum entropy rate and spectral entropy mean as the characteristic value,recognition rate was significantly higher than the multi-scale entropy rate as the characteristic value.Error identification is bubbly-churn flow and annular flow,bubble flow and churn flow recognition rate of 100%;Identification rate of support vector machine as the recognition model is significantly higher than that of BP neural network;Multi scale entropy rate as the feature is the recognition speed slightly faster than the multiple scale marginal spectrum joint distribution entropy and spectral entropy mean recognition speed,but the advantage is notobvious;The joint distribution of the marginal spectrum entropy rate and the spectral entropy mean as the characteristic value and the support vector machine(SVM)combined with the identification flow pattern was fast and accurate.Multi-scale entropy could be used to reveal the dynamic characteristics in small scale,and the multiscale marginal spectrum entropy could reveal the dynamic characteristics of different flow regimes from the frequency domain.The validity of the two flow pattern identification methods of multi-scale entropy and multiscale marginal spectrum entropy was demonstrated.