Direct Numerical Simulation Of Two-phase Reactive Turbulent Boundary Layer And Data-driven Modeling

Today,fossil fuels are still the principal source to obtain energy for human in the world.The burning of these fuels are usually organized in wall-bounded devices,accom-panied by multiphase turbulent flow.In the near-wall region,the interactions between particle,turbulence,flame and wall play important role.For example,turbulence may be modulated by particles.Flame can be bended and curved due to the strain of turbulence.Flashback and quenching of flame can occur at the wall.The wall in turn generates bound-ary layer turbulence and might be abraded by particles.These processes can have great influence on the combustion safety,efficiency and formation of pollutants.However,due to strong shear,strong coupling and strong nonlinearity of two-phase turbulent combus-tion near wall,the study of this problem is challenging,and our current understanding is far from enough.Therefore,we performed high-precision direct numerical simulations to explore the interactions of particle-turbulence-flame-wall system,and to provide guidance for practical industrial problems.Firstly,we studied the interactions between finite-size particles and isotropic turbu-lence using fully-resolved immersed boundary method.It was found that the inertial par-ticles do positive work on fluid,which enhances the turbulence.In addition,the particles also induce great dissipation thereby damping turbulence.The induced-dissipation mainly comes from the small-scale turbulence in the front and transverse part of particle.The two opposite mechanisms were both found to be related with the slip motion between particle and fluid,which can be characterized by particle Reynolds number.Later,we established quantitative relations between the two mechanisms and particle Reynolds number.It was revealed that the work done by a single particle is larger than the extra dissipation induced by it,thus the turbulence is enhanced,and vice versa.The criterion based on the above analysis can also explain the overall modulation by particles.Using trip-wire method to trigger the transition,an accurate flat plate boundary layer turbulence was obtained.Under the background of reheat stage gas turbine,we investi-gated the interactions between two-phase boundary layer turbulence and lean-premixed H₂-O₂flame.It was identified that flame propagation and auto-ignition both exist,and flame propagation mainly occurs in the buffer layer of turbulence.In the viscous sublayer,flame quenching happens due to the cooling effect of cold wall,leading to large heat flux.The range of this heat flux can be roughly captured by the head-on quenching results of one dimensional flame propagation and auto-ignition flames.Different from early knowledge about this problem,we found that large heat release exists in the extinction zone,and heat release rate has inverse correlation with the dot product of flame normal vector and wall normal vector.The flame behaves like a filter,and it damps the Reynolds stress of bound-ary layer turbulence as well as the stream-wise fluctuation in the out layer.In addition,the hairpin vortex structures are broken down,and the low-speed streaks become sparse and wide in the buffer layer.The huge heat release also changes the alignment of flame normal vector and direction of principal strain rate.When considering the inert particles in reactive boundary layer,it was found that the trend of preferential concentration in low-speed streaks is reduced compared with simulations without reaction.Due to the relative frequency of sweep event increases,the particle number density in the near wall region in-creases as well.Besides,due to the enhancement of stream-wise fluctuation in buffer layer,the particle fluctuation velocity also increases in the stream-wise direction accordingly.Finally,aiming at making full usage of the huge amount of data from direct numerical simulations,we proposed data-driven modeling concept,and it was successfully applied to the closure problem of subgrid-scale stress in large eddy simulation.The a priori validation showed that the current model is much better than the traditional model.In the a posteriori test,the data-driven model also has better performance for predicting the turbulent kinetic energy,subgrid-scale dissipation and spectral distribution of energy.What's more,there are apparent correlations between the second derivative of velocity and subgrid-scale stress.On the one hand,the present work improved our understanding about two-phase tur-bulent combustion in boundary layer.On the other hand,we generated large amounts of reliable database full of information,and proved that data-driven modeling is an effec-tive way to find correlations,which lays the foundation for larger-scale data mining and physical modeling.