Study On Flow Resistance Of Porous Media Based On Variable Diameter Tube Ball Mode

Porous media is closely related to our life,and plays an extremely important role in petroleum,metallurgy,chemical industry,environmental protection,aerospace and other fields.However,the study of seepage mechanism and resistance prediction are still the hot and difficult problems in porous media transmission.In this thesis,theoretical research,statistical analysis,regression analysis and machine learning methods are used to study this problem in depth.The main tasks and results are as follows:(1)Based on the spherical and orifice models of porous media,combined with the actual porous media structure characteristics,a new structure and flow model--variable diameter tube ball model was established.Based on this model,the model formula of variable diameter tube ball is deduced by the theory of fluid mechanics.The rationality and reliability of the formula are verified by the experimental data of different Reynolds number range,and the applicability of the formula is obtained.(2)The formula derived from the modified model can be applied to the foamed metal with larger porosity.The comparison between the modified tube ball formula and Ergun equation shows that the modified tube ball formula has better applicability than Ergun equation for metal foam.(3)The collected factors affecting the flow resistance of porous media generated 45 sets of experimental data.The binary correlations between different factors and pressure drops were analyzed using SPSS software to influence the flow resistance and significant levels of variables based on important factors and relationships between Pearson’s correlation coefficient and pressure drop.Then,each variable was fitted linearly and nonlinearly with the flow resistance,and the most suitable function form between each variable and the pressure drop was found.Finally,a formula with a fitting degree of 78.3%was obtained.(4)To improve the accuracy of the prediction,45 experimental data sets were processed and modeled using machine learning algorithms in Python software.The experimental data were pretreated and corresponding heat maps were drawn.Lazy prediction method was used to fit and evaluate the pretreated experimental data,and the model with 90% fit was selected for learning.Draw a learning curve to check whether the model is over-fitting or under-fitting.The fitting model of "LGBMRegressor" was selected as the best model with a fitting degree of 97.1%,and parameters related to the model were obtained.It can be seen that the machine learning model has better predictive ability than both the traditional formula and the multivariate fitting formula,at least in the range of porous media parameters involved at present.Whether it can be extrapolated to other parameters still needs further application and verification.