Experimental Study And Intelligent Prediction Of Two-phase Flow Regimes In The Rod Bundle Channel With Mixing Vane Spacer Grids

Researches on two-phase flow regime and its transition characteristics in a real-size fuel assembly play a significant role in improving the understanding of two-phase flow in the reactor core.Meanwhile,to achieve accurate prediction of two-phase flow regimes is the premise of reasonable selection for calculation correlation of thermal parameters to carry out reactor safety analysis.However,due to various of limitations derived from experimental conditions and the poor quality of two-phase flow data obtained in the past,the definition and classification of flow regime has not yet reached a unified conclusion.The conventional flow regime transition theoretical model still has a series of problems such as inherent losses in the range of application and prediction accuracy,and poor versatility.Therefore,this paper attempts to innovate the test facility and measurement scheme,carry out the visualization experiments on two-phase flow regime at the upstream and downstream of the mixing vane spacer grid in a 5×5 rod bundle,obtain high-quality two-phase flow data,complete the definition and classification of the twophase flow regime,and draw flow regime maps.The influence of air or water flow rates,Reynolds number and the structure of prototype grid on flow regime and its transition characteristics were discussed in detail.Some machine learning algorithms were compared and integrated to establish an intelligent prediction model with optimized,achieving flow regime identification in rod bundle channels with higher accuracy and efficiency.In order to obtain high-quality flow data mentioned above,the whole visual test section consists of 25 transparent fluorinated ethylene propylene(FEP)tubes of 9.5 mm in diameter and 12.6 mm pitch to simulate fuel rods.The simultaneously scheme of opposite lighting and dual high-speed cameras with high-resolution was considered to capture high-quality videos of different flows at the two visual measurement ports.By realizing the plane focusing of different subchannels,it is observed that there are five typical flow patterns in the rod bundle channel:Finely dispersed bubbly flow,Bubbly flow,Cap-bubbly flow,Cap-turbulent flow and Churnturbulent flow,and there are different flow regimes among different subchannels.The flow phenomena such as bubble tearing,collision coalescence,growth,mutual migration between sub-channels,lateral movement,backflow and periodic pulsating flow,bubble stagnation and cover in the back-pressure zone of the grid were also described more detailedly.In terms of flow regime transition characteristics,the increase of the superficial air velocity and dryness both mainly affects the bubble’s diameter,number and density,that is,the increase of the above parameters will promote the flow regime transformation in turn.The turbulence effect brought by the increase of liquid Reynolds number has the opposite effect on the flow regime and its transition characteristics,which makes the gradient of transition boundary between flow regimes positive.The uneven phase distribution caused by the tearing effect of the mixing vane spacer grid,especially the mixing fins,will delay the transition among flow regimes discussed above.Compared with the flow regimes’ transition criterion from Ishii,Venkateswararao,Liu and Hibiki theoretical models,all the flow regimes in the experimental results migrate to the lower air velocity region,and the actual critical value of the dryness should be smaller.From the perspective of data-driven,the air-water superficial velocity,Reynolds number,dryness and their corresponding arithmetic expressions were used as flow regime feature vectors.A support vector machine(SVM)flow regime intelligent identification model optimized by principal component analysis(PCA)and genetic algorithm(GA)was proposed.The results show that SVM optimized by PCA and GA significantly improves the predication accuracy and efficiency.In this data-driven mode,both the training and prediction accuracy reaches over 99%,and the time consumption is only 5.74 s,which can meet the requirements of flow regime online identification.