Universit`a degli studi della Basilicata
Transcript
Universit`a degli studi della Basilicata
Università degli studi della Basilicata Dottorato di ricerca in “Ingegneria Industriale e dell’Innovazione” TITOLO DELLA TESI: Numerical Simulation of Multi-dimensional Hypersonic Plasma Flows Settore Scientifico-Disciplinare “Macchine a fluido (ING-IND/08) e Fisica dei reattori nucleari (ING-IND/18)” Coordinatore: Prof. Vinicio Magi Dottorando: Ra↵aele Pepe .......................................... Tutor: Prof. Aldo Bonfiglioli .......................................... Dr. Ing. Antonio D’Angola .......................................... A.A. 2013/2014 Ciclo XXVII .................................... Giudizio del collegio dei docenti Il dott. Raffaele PEPE espone il proprio lavoro concernente “Numerical Simulation Of Multi-Dimensional Hypersonic Plasma Flows”. Al termine della presentazione il Collegio si riunisce per decidere sull’ammissione del dottorando all’esame finale. Il relatore, Prof. Aldo Bonfiglioli illustra l’attività svolta dal dottorando nel triennio. Durante tutto lo svolgimento del dottorato di ricerca, sei mesi del quale trascorsi presso il von Karman Institute for Fluid Dynamics, il dott. Raffaele PEPE ha dimostrato acume e dedizione al lavoro di ricerca, predisposizione al confronto costruttivo, anche in ambiti multidisciplinari, rigore metodologico ed una notevole capacità di approfondimento dei problemi. L’attività di ricerca, rapidamente inserita in un contesto internazionale di notevole spessore, è stata correttamente impostata grazie ad un assiduo lavoro di ricerca bibliografica ed è stata sorretta costantemente da un grande impegno ed una buona capacità organizzativa. Tale attività ha portato alla sottomissione di 3 articoli su rivista internazionale, di cui uno attualmente in stampa, 4 articoli ed 1 poster presentati a conferenze internazionali ed un rapporto tecnico. I risultati sono interessanti ed analizzati con buon senso critico. Nell’esposizione il dottorando dimostra di avere ottima conoscenza delle problematiche trattate. Il Collegio dei Docenti, sentito il parere del tutor, valuta più che positivamente il lavoro svolto e delibera l’ammissione del dottorando Raffaele PEPE all’esame finale. "Our best ideas are often those that bridge between two different worlds" Marvin Minsky Acknowledgments I wish to thank various people for their contribution to this thesis work. I would like to thank my supervisors, Prof. Aldo Bonfiglioli and Dr. Antonio D’Angola for their constant support and guidance in the development of this work. I am very grateful to them for encouraging me to pursue my passion for research. During last three years, their dedication and rigor in research work has been a constant example for me. A special thank goes to Dr. Gianpiero Colonna, from IMIP-CNR and Prof. Renato Paciorri, from University of Rome for providing me the necessary insight on nonequilibrium plasma models and on the subject of hypersonics. I would like to thank both of them for their uncountable theoretical suggestions and the practical help. Thanks to Dr. Andrea Lani, from the Von Karman Institute. He was enthusiastic about the opportunity to collaborate from the first email that I sent to him. It was a real honor and pleasure for me to work with him at the VKI. I would like to thank Jesuś Garicano Mena, from VKI, for all the useful discussions and the practical help on the use of COOLFluiD. I wish him all the best for his Ph.D. I acknowledge Prof. Herman Deconinck, from VKI, for having accepted my request to spend a research period at the VKI as visiting Ph.D. student. i ii Acknowledgment Thanks to Renato, Pasquale, Marianna and Michele who shared the office with me during these years. I wish them all the best. Thanks to my bandmates Antonio, Claudio, Daniele and Donato for all the good vibes that music has given to me. Thanks to all my friends: Francesco, Antonio S. senior, Claudio, Lina, Cosimo, Antonio G., Dino, Donatella, Stefano, Anna, Angelica, Pasquale, Rocco, Alessandro, Antonio S. junior, Giunio, Maurizio...etc. Last but not least I would like to thank my family: my father Giuseppe, my mother Agnese, my brother Donato and my sister Mariateresa. Thank you for trusting in me. Abstract In the present thesis, the capabilities of eulfs, an unstructured 2D/3D solver developed for thermally and calorically perfect gas, have been extended making it capable to deal with chemical nonequilibrium plasma flows. Preliminary tests have been carried out for an ionized argon mixture flowing in a converging-diverging nozzle. 2D and 3D results obtained by using the extendend version of the eulfs code have been compared with those obtained with a well established quasi-unidimensional code developed at the IMIP-CNR of Bari, showing a good agreement. The CFD code can be coupled with a newly developed, unstructured, shockfitting algorithm which treats the discontinuities as moving boundaries that border regions of the flow-field were a smooth solution to the governing PDEs exists. The unstructured shock-fitting algorithm has been extended to deal with an ionized argon mixture to model shock waves in chemical reacting flows. Promising results have been obtained using the shock-fitting approach for a 2D hypersonic flow past the fore-body of a circular cylinder. The unstructured shock-fitting algorithm has been extended to deal with thermochemical nonequilibrium flows and thanks to its modurality, has been coupled with COOLFluiD, an in-house shockcapturing CFD solver developed at the Von Karman Institute. Results obtained in the computation of hypersonic flows past circular cylinders have been obtained for both ideal gas and dissociated Nitrogen in thermochemical nonequilibrium. iii iv Contents Acknowledgment i Abstract iii List of Figures ix List of Tables xiii List of Symbols xv 1 Introduction 1 2 Physical model 5 2.1 2.2 2.3 Governing equations . . . . . . . . . . . . . . . . . . . . . 6 2.1.1 Mixture parameters . . . . . . . . . . . . . . . . . . 6 2.1.2 Equation of state . . . . . . . . . . . . . . . . . . . 7 2.1.3 Thermodynamic model . . . . . . . . . . . . . . . . 8 2.1.4 Conservation equations . . . . . . . . . . . . . . . . 13 Chemical models . . . . . . . . . . . . . . . . . . . . . . . 17 2.2.1 Kinetic model for an ionized argon mixture . . . . . 17 2.2.2 Kinetic model for a dissociated nitrogen mixture . . 21 Electrohydrodynamic model . . . . . . . . . . . . . . . . . 22 3 Computational tools 27 v vi Table of contents 3.1 Shock-capturing solver . . . . . . . . . . . . . . . . . . . . 28 3.1.1 Fluctuation and conservative linearization . . . . . 29 3.1.2 Signals or Residual Distribution . . . . . . . . . . . 32 3.1.3 Solution of the discretised equations . . . . . . . . . 34 3.2 Shock-fitting algorithm . . . . . . . . . . . . . . . . . . . . 36 3.3 Details on the implementation . . . . . . . . . . . . . . . . 40 4 Numerical results 4.1 4.2 4.3 43 eulfs results . . . . . . . . . . . . . . . . . . . . . . . . . 43 4.1.1 Ionized argon, inviscid flow in a nozzle . . . . . . . 43 4.1.2 Ionized argon, inviscid flow over a circular cylinder 48 COOLFluiD results . . . . . . . . . . . . . . . . . . . . . . 51 4.2.1 Ideal gas, inviscid flow over a circular cylinder . . . 51 4.2.2 Ideal gas, viscous flow over a circular cylinder . . . 54 4.2.3 Dissociated nitrogen, inviscid flow over a circular cylinder . . . . . . . . . . . . . . . . . . . . . . . . 60 EHD results . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.3.1 Parallel electrodes . . . . . . . . . . . . . . . . . . . 62 4.3.2 Ionized argon, inviscid flow in a nozzle with two opposite electrodes . . . . . . . . . . . . . . . . . . 65 5 Conclusions and future work 67 A Appendix: Split Jacobian matrix 71 A.1 Jacobian transformations . . . . . . . . . . . . . . . . . . . 75 A.2 Transformations between U and Z . . . . . . . . . . . . . 77 A.2.1 U ! Z . . . . . . . . . . . . . . . . . . . . . . . . 77 A.2.2 Z ! U . . . . . . . . . . . . . . . . . . . . . . . . 77 B Appendix: Nondimensionalization 79 B.1 TCneq, external flows . . . . . . . . . . . . . . . . . . . . . 79 Table of contents vii B.2 TCneq, internal flows . . . . . . . . . . . . . . . . . . . . . 81 B.3 EHD, external flows . . . . . . . . . . . . . . . . . . . . . . 82 B.4 EHD, internal flows . . . . . . . . . . . . . . . . . . . . . . 83 B.5 Nondimensional parameter vector . . . . . . . . . . . . . . 84 B.6 Nondimensional pressure derivatives . . . . . . . . . . . . . 85 C Appendix: Rate coefficients fits 87 C.1 Rate coefficients fit . . . . . . . . . . . . . . . . . . . . . . 87 References 91 viii List of Figures 1.1 Schematic of the flowfield surrounding a space vehicle during the reentry phase (reproduced from [60]). . . . . . . . . 2 2.1 Ionization and recombination rates from the ground level as a function of the electron temperature (courtesy of G. Colonna) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.2 Electrical conductivity of argon (p = 0.013atm), reprinted from [8]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3 Scheme of the coupled circuit . . . . . . . . . . . . . . . . 26 3.1 Residual distribution concept. . . . . . . . . . . . . . . . . 28 3.2 Inward scaled normal. . . . . . . . . . . . . . . . . . . . . 33 3.3 Starting point: a) background mesh and b) shock boundary (right). . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.4 a) Cell removal. b) Local remeshing. . . . . . . . . . . . . 38 3.5 a) Normal and tangential unit vector computation. b) Interpolation of the phantom nodes. . . . . . . . . . . . . . . 39 3.6 Interpolation of mesh point jumped by the shock boundary. 39 3.7 Modularity of the shock-fitting code. . . . . . . . . . . . . 41 4.1 Flow in a converging-diverging nozzle: 2D and 3D geometries flooded by Mach number. . . . . . . . . . . . . . . . . 44 4.2 Test A: distributions along the nozzle axis. . . . . . . . . . 45 4.3 Test B: distributions along the nozzle axis. . . . . . . . . . 46 ix x Table of contents 4.4 Flow in a converging-diverging nozzle: convergence histories for test B. . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.5 Baseline mesh and “fitted” mesh. . . . . . . . . . . . . . . 48 4.6 2D flow past a circular cylinder: comparison between the shock-capturing (S-C) and shock-fitting (S-F) solutions. Both S-C and S-F solutions are first order accurate in space (N scheme). . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.7 2D flow past a circular cylinder: comparison of shockcapturing (S-C) and shock-fitting (S-F) solutions along y = 0. 50 4.8 Grids used for the inviscid flow over a circular cylinder. . . 52 4.9 Pressure contour plot: comparison between the shock-fitting (S-F) and shock-capturing (S-C) second-order-accurate solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 4.10 Pressure contour plot for the shock fitting solutions: comparison between COOLFluiD and eulfs solutions. . . . . . 54 4.11 Grid used for the viscous flow over a 1 m radius circular cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.12 Pressure and temperature contour plot: comparison between the shock-fitting and shock-capturing second-orderaccurate accurate solutions. . . . . . . . . . . . . . . . . . 57 4.13 Wall distributions. . . . . . . . . . . . . . . . . . . . . . . 57 4.14 Heat flux distribution on the wall. . . . . . . . . . . . . . . 58 4.15 Distributions along the stagnation streamline. . . . . . . . 58 4.16 Distributions along the stagnation streamline (enlargement of the boundary layer). . . . . . . . . . . . . . . . . . . . . 59 4.17 Adimensional total enthalpy. . . . . . . . . . . . . . . . . . 59 4.18 Grid used for the Nitrogen flow over a circular cylinder. . 60 4.19 COOLFluiD + SF vs. Hornung’s experimental measurements: a) Non-dimensional shock-wall distance and b) finite interference fringe patterns. ✓ is the azimuthal angle which takes value 0 at the stagnation point, /R is the shock-wall distance divided by the cilinder’s radius. . . . . 61 Table of contents 4.20 Pressure contour plot: qualitative comparison between the fitted solution obtained with COOLFluiDand the solution obtained by Wang and Zhong [76]. Hornung’s experimental measurements of the shock stando↵ distance are represented by the red circles. . . . . . . . . . . . . . . . . . . 62 4.21 Electric potential contours for the electrode channel with a constant electrical conductivity ( = 1 ⌦ 1 m 1 ): a) eulfs results and b) reprinted from [32]. . . . . . . . . . . . . . . 63 4.22 Parallel electrodes with a non constant electrical conductivity ( = ): a) mesh b) boundary conditions. . . . . . . 64 4.23 Parallel electrodes with a non constant electrical conductivity ( = ): a) electric potential contour plot (numerical vs. analytitcal solution) b) distribution of the electric potential along the y-axis (numerical vs. analytitcal solutions). 65 4.24 Flow in a converging nozzle with two parallel electrodes. . 66 4.25 Temperature distributions along the nozzle axis for di↵erent values of the generator potential. . . . . . . . . . . . . 66 C.1 Relative error associated to the e A forward rate coefficient vs. the translationa l temperature . . . . . . . . . . 89 xi xii List of Tables 2.1 Chemical species in the ionized argon mixture . . . . . . . 18 2.2 Chemical processes in the ionized argon mixture . . . . . . 18 2.3 Equilibrium constants for all processes accounted for in the present model . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.4 Chemical species in the dissociated nitrogen mixture . . . 22 2.5 Chemical processes in the dissociated nitrogen mixture . . 22 4.1 Area distribution along the nozzle. . . . . . . . . . . . . . 45 4.2 Inlet flow conditions. . . . . . . . . . . . . . . . . . . . . . 45 4.3 Freestream conditions for an ionized argon flow over a 0.05m radius cylinder. . . . . . . . . . . . . . . . . . . . . 49 4.4 Characteristics of the grids used for the inviscid hypersonic flow over a circular cylinder. . . . . . . . . . . . . . . . . . 52 4.5 Freestream conditions for a viscous flow over a 1m radius cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.6 Freestream conditions for a nitrogen flow around a 1 inch radius cylinder. . . . . . . . . . . . . . . . . . . . . . . . . 60 4.7 Boundary conditions for the EHD solver. 4.8 Inlet flow conditions. . . . . . . . . . . . . . . . . . . . . . 65 . . . . . . . . . 63 C.1 Coefficients of function f1 = log KeqI (cm 3 ) for the Ar ionization . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 C.2 Coefficients of function f2 = log kf1 (cm 3 ) for the reaction 1 (e A ionization from Ar0 ) . . . . . . . . . . . . . . . . 88 xiii xiv Table of contents C.3 Coefficients of function f2 = log kf2 (cm 3 ) for the reaction 2 (e A excitation) . . . . . . . . . . . . . . . . . . . . . . 88 C.4 Coefficients of function f2 = log kf3 (cm 3 ) for the reaction 1 (e A ionization from Ar⇤ ) . . . . . . . . . . . . . . . . 89 C.5 Rate coefficients kf (cm3 /s) for A-A processes . . . . . . . 89 List of Symbols Acronyms B Blended scheme CDT Constrained Delaunay Triangulation CFD Computational Fluid Dynamics Cneq Chemical nonequilibrium CRD Contour Residual Distribution EEDF Electron Energy Distribution Function EHD ElectroHydroDynamics EOS Equation Of State FE Finite Element FS Fluctuation Splitting FV Finite Volume I/O Input/Output LDA Low Di↵usion Advection scheme LRD Linear Residual Distribution MHD MagnetoHydroDynamics N Narrow scheme ODE Ordinary Di↵erential Equation PDE Partial Di↵erential Equation RD Residual Distribution R-H Rankine-Hugoniot relations S-C Shock-Capturing S-F Shock-Fitting TCneq Thermochemical nonequilibrium VKI von Karman Institute xv xvi List of Symbols Roman symbols a speed of sound B magnetic field vector Cp specific heat at constant pressure Cv specific heat at constant volume Da Damköler number e internal energy per unit mass ee internal energy per unit mass of free electrons i e internal contribution to internal energy per unit mass e e electronic contribution to internal energy per unit mass r e rotational contribution to internal energy per unit mass t e translational contribution to internal energy per unit mass v e vibrational contribution to internal energy per unit mass E total energy per unit mass E electric field vector fe electron energy distribution function FB magnetohydrodynamic force g statistical weight h enthalpy per unit mass f h formation enthalpy per unit mass H total enthalpy per unit mass kb backward rate constant kf forward rate constant Keq equilibrium constant j current density vector Js di↵usion flux of the species s I current intensity L reference lenght mr reduced mass M Mach number M molar mass n normal unit vector n molar density N particles number density m/s T J/(kg K) J/(kg K) J/kg J/kg J/kg J/kg J/kg J/kg J/kg J/kg V/m N J/kg J/kg J/kg A/m2 kg/(m2 s) A m kg kg/mol mol/m3 m 3 List of Symbols Nr Ns p pe q qv Qs QJ r Rc Ṡev Ṡs t t T Te TE TR TV Tv u x Vdis Vq Wsh number of chemical reactions number of chemical species gas pressure electronic pressure global heat flux vibrational heat flux partition function of species s energy production rate due to the Joule e↵ect radius circuit resistance rate of production of the vibrational energy rate of production of the sth species tangential unit vector time traslational temperature electron temperature electronic temperature rotational temperature vibrational temperature vibro-electronic temperature fluid velocity vector (u, v, w) position vector (x, y, z) potential between the electrodes potential of the generator shock wave velocity xvii Pa Pa W/m2 W/m2 W/m3 m ⌦ W/m3 kg/(m3 s) s K K K K K K m/s m V V m/s xviii List of Symbols Greek symbols ↵ mass concentration ↵ion inization degree ✏ internal energy per unit volume " electron energy isentropic coefficient shock-wall distance electric potential residual roto-translational thermal conductivity v vibrational thermal conductivity µ viscosity 0 ⌫ stochiometric coefficients for the products 00 ⌫ stochiometric coefficients for the reagents ⌦ element area ⇠˙ chemical reaction velocity ⇢ density electric conductivity e electric conductivity tensor & cross section ⌧ stress tensor ⌧s relaxation time of the sth species ⇥ azimuthal angle J/m3 eV m V W/(kg K) W/(kg K) kg/(m s) m2 mol/(m3 s) kg/m3 ⌦/m ⌦/m m2 Pa s List of Symbols Subc d down e i s r up sh w 1 ⇤ xix and Superscripts convective fluxes di↵usive fluxes downstream states eth element ith node sth chemical species rth chemical reaction upstream states shock wave wall freestream conditions adimensional variables Constants ec elementar charge hP Planck constant kB Boltzmann constant me electron mass NA Avogadro’s number Rg universal gas constant µ0 permeability of the free space Species Ar0 argon in the ground state Ar⇤ argon in the metastable state Ar+ argon ion e electron N2 molecular nitrogen N atomic nitrogen 1.6022 ⇥ 10 19 C 1.3807 ⇥ 10 23 J/K 1.3807 ⇥ 10 23 J/K 9.11 ⇥ 10 31 kg 6.022 ⇥ 10 23 mol 1 8.314 J/(mol K) 4⇡ ⇥ 10 7 N/A2 xx Chapter 1 Introduction The accurate simulation of hypersonic flows past blunt bodies is still a challenge, despite more than 20 years of algorithmic developments on CFD solvers. In hypersonic conditions, the flowfield surrounding the blunt bodies is characterized by several complex physico-chemical phenomena such as chemical reactions, thermal relaxation, ablation of the surface, radiation, strong bow shocks, shock-shock and shock-boundary layer interaction. Fig. 1.1 illustartes a schematic of the flowfield surrounding a capsule during the reentry phase in atmosphere. A fundamental problem is the description of the ionization and dissociation kinetics, and, more in general, the chemical processes occurring in the flowfield past the bow shock. A primary role is played by excited states of atoms and molecules [19]. Excited levels, reduce the activation energy of exothermal processes speeding up the reaction processes. An important role is played by metastable states, that, because cannot decay by radiation emission, survive for long time, with a large influence on the flow properties. Another critical aspect is the accurate description of discontinuities, and in particular the descripition of strong shock waves, like the bow shock ahead of the blunt body. To simulate hypersonic flows, widespread “traditional” shock-capturing solvers often exhibit severe drawbacks, expecially when used on unstructured grids: stagnation point anomalies [39], carbuncle phenomenon and spurious oscillations [45] and a reduction of the order of accuracy within the entire shock-downstream region [18]. These drawbacks, defined as ”shock anomalies” [45] seem to be caused by numerical details of the capturing process, since numerically captured 1 2 Introduction Figure 1.1: Schematic of the flowfield surrounding a space vehicle during the reentry phase (reproduced from [60]). shock usually contains at least one computational cell forming numerical internal shock-structure, which is a pure numerical ”artefact” and is not related to the real physical internal structure of the shock wave [45, 69]. Despite unstructured codes are less e↵ective and accurate of structured ones in the simulation of hypersonic regimes, various unstructured CFD tools used in the aero-thermodynamic design and analysis of space vehicles entering planetary atmospheres have been developed, since unstructured grids o↵er greater flexibility than structured ones in tackling complex geometries allowing to automatically adapt the mesh to the local flow features. NASA’s FUN3D [3, 4], DLR’s TAU [33, 52] and LeMANS [66, 67] codes are three such examples of unstructured codes used in hypersonic reentry applications. The shock-fitting approach, which has been made popular since the mid 60s by Moretti and collaborators [55], has already proved to be immune to the shock-capturing drawbacks. Shock-fitting consists in using the Rankine-Hugoniot jump relations to explicitely track the motion of the discontinuities in a Lagrangian manner. Thanks to its ability of accurately simulating shocks on coarse grids, shock-fitting was very popular in the early computer era. With increasing computer power and because of some algorithmic difficulties that plagued the shock-fitting approach on structured grids, shock-capturing took over and is nowadays the method Introduction of choice for virtually all CFD simulations. Shock-fitting discretizations based on the so-called “boundary” variant have been in use until the mid 90s [53, 54] to simulate supersonic and hypersonic flows: only the strong bow shock was fitted and made to coincide with the upstream boundary of a structured mesh; all other shocks were captured. The “floating” variant of the shock-fitting technique, although more versatile since it allows to fit also the embedded shocks, is algorithmically complex, so that only few three-dimensional calculations have been reported in the literature [77]. Recent advances in unstructured grid generation and discretization techniques has allowed to develop an unstructured shock-fitting algorithm [57] which is algorithmically simpler than the shock-fitting algorithms traditionally used in the structured-grid setting. This unstructured version of the shock-fitting technique combines features of both the “boundary” and “floating” variants proposed in the structured grid setting: it therefore allows not only to fit the bow shock, but also the embedded shocks. Moreover, the geometrical flexibility o↵ered by the use of unstructured triangular and tetrahedral meshes allows to deal much more properly with interacting shock [42, 58] than it was possible in the structured-grid context. Shock-fitting algorithms have been used in the past to simulate chemically reacting nonequilibrium flows on structured grids, see e.g. the work by Pfitzner [63] and Paciorri et. al. [59] in the 90s. More recently, Prakash et al. [65] have developed an high order finite di↵erence shock-fitting algorithm to simulate thermochemical nonequilibrium flows on structured grids. In the present thesis, the capabilities of eulfs [10, 13], an unstructured 2D/3D solver developed for thermally and calorically perfect gas, have been extended making it capable to deal with chemical nonequilibrium plasma flows. Tests have been carried out for an ionized argon mixture, including also the argon metastable state, flowing in a converging-diverging nozzle. 2D and 3D results obtained by using the extendend version of the eulfs code have been compared with those obtained with a quasiunidimensional code developed at the IMIP-CNR of Bari [20, 21, 24]. In the present work the unstructured shock-fitting algorithm developed by Paciorri and Bonfiglioli [57] has been extended to deal with an ionized argon mixture to model shock waves in chemically reacting flows. Promising results have been obtained using the shock-fitting approach for a 2D hypersonic flow past the fore-body of a circular cylinder. 3 4 Introduction Thanks to its modularity, the unstructured shock fitting algorithm has been coupled with COOLFluiD [50, 51], an in-house CFD solver developed at the Von Karman Institute (VKI), to investigate the causes of anomalous heat flux distributions on the wall of blunt bodies immersed in hypersonic flows. Moreover the shock-fitting algorithm has been extended to deal with two temperature, multispecies thermochemical nonequilibrium flows and results have been obtained for an hypersonic dissociating nitrogen flow past the forebody of a circular cylinder. Finally a simple electrohydrodynamic model have been implemented within eulfs. In this model the ionized gas flow is coupled with an electric field controlled by considering a power supply and an external circuit resistence. Poisson’s equation and the Ohm’s law for the external circuit have been added to the set of multispecies Euler equations. Preliminary results an ionized argon mixture flowing in a subsonic nozzle with two opposite electrodes, showing the e↵ect on the flowfield variables of the ohmic heating due to the electric field. The physical models used to model both viscous and inviscid flows in thermochemical nonequilibrium are described in Chapter 2. The computational tools used in the dissertation are described in Chapter 3. The main features of the capturing solver used in conjunction with the shock-fitting technique are provided in Section 3.1, while the unstructured shock-fitting algorithm is described in details in the Section 3.2. Numerical results obtained using both codes, eulfs and COOLFluiD, are shown and discussed in the Chapter 4. Finally conclusions on the present work and future work are discussed in Chapter 5 Chapter 2 Physical model When dealing with hypersonic flows, the perfect gas model can no longer be considered, because of the very high temperatures reached after the strong shock wave. The flowfield in the shock layer is characterized by complex physiscal phenomena such as: chemical reactions, rotational, vibrational and electronic excitation, ablation, radiative heat-flux and viscous interactions. Under those conditions, thermodynamic properties as the specific heat, and transport properties, as the viscosity and thermal condictivity are not constant and vary with temperature, pressure and chemical composition. Concerning chemical reactions and internal energy relaxation, it is possible to define di↵erent regimes: frozen flow, equilibrium flow and nonequilibrium flow. The bounds of those regimes can be defined considering an important nondimensional parameter, the Damköler number. The chemical Damköler number is defined as the ratio between the characteristic time for the macroscopic processes occurring in the flow ⌧f , such as convective or di↵usive phenomena, and the chemical characteristic time ⌧c for chemical processes: Dac = ⌧f ⌧c Then the three regimes are defined as follows: 5 (2.1) 6 Chapter 1. Physical model Dac >> 1 Dac ⇡ 1 Dac << 1 Equilibriumf low N on equilibriumf low F rozenf low Defining a characteristic time ⌧t for the energy relaxation, it is possible to ⌧ define another Damköler number Dav = ⌧ft , which can be used to define regime of thermal equilibrium, nonequilibrium and freezing. In the following sections of this chapter we will present the set of governing equations used in this work to model nonequilibrium flows. 2.1 2.1.1 Governing equations Mixture parameters In this section we introduce the physical quantities which characterize the composition of a mixture. There are di↵erent ways to define the density of the sth chemical species and the global density of the mixture. Considering the species mass density ⇢s , the mixture mass density ⇢ is given by: ⇢= Ns X ⇢s (2.2) s=1 where Ns is the number of chemical species present in the mixture. The number of the moles of the species s, ns is given by the ratio of the species mass density and the molar mass Ms ns = ⇢s Ms (2.3) then the total number of moles n is given by the sum of the species number moles: Ns X n= ns (2.4) s=1 Chapter 1. Physical model 7 The particles number density of the species s, Ns is easly obtained as Ns = ns NA where NA is the Avogadro’s number. Total number density is trivially given by Eq. (2.5): N = Ns X s=1 Ns (2.5) In Eq. (2.6) we define respectively the mass fraction ↵s and the molar fraction s of the species s: ↵s = ⇢s , ⇢ s = ns Ns = n N (2.6) Considering Eq. (2.2), (2.4) and (2.6) we obtain the following constraint for the mass and the molar fraction: Ns X ↵s = 1, s=1 2.1.2 Ns X s =1 (2.7) s=1 Equation of state At high temperature the e↵ects of the intermolecular forces can be neglected, so that it is possible to consider a mixture of thermally perfect gases [2]. The equation of state (EOS ) for the chemical species s is given by Eq. (2.8) p s = ⇢s Rg Ts Ms (2.8) where ps and Ts are respectively the pressure and the kinetic temperature of the species s. If the fluid is assumed to be sufficiently collisional, the species rapidly thermalize with each other and it is possible to define a single temperature T for all the particles, Ts = T . On the contrary, in those circumstances in which ionization is high, the energy exchange between electrons and heavy particles is too slow, so that the free electrons are characterized by an electron temperature Te 6= T . The partial pressure of the electrons is given by the following equation: 8 Chapter 1. Physical model p e = ⇢e Rg Te Me (2.9) The EOS of the mixture can be easly obtained from the Dalton’s law which allows us to compute the pressure of the mixture p p= Ns X ps = s=1 X s6=e ⇢s Rg Rg T + ⇢e Te Ms Me (2.10) Finally if the electron are in thermal equilibrium with heavy particles the pressure of the mixture is given by Eq. (2.11) p= Ns X s=1 where R = 2.1.3 P Ns Rg s=1 ↵s Ms . ⇢s Rg T = ⇢RT Ms (2.11) Thermodynamic model For high temperature flows it is not possible to neglect the electronic, vibrational and rotational energy excitation, which implies that it is not any longer possible to assume that the gas is calorically perfect. The internal energy of each species can be expressed as the sum of a translational contribution ets , a contribution due to the excitation of the internal energy modes eis and the formation enthalpy hfs [73]: es = ets + eis + hfs . (2.12) If we use the rigid rotator and harmonic oscillator model to describe rotational and vibrational motion of the molecules we can completely separate the internal energy modes into three contributions: rotational (r), electronic (e) and vibrational (v): eis = ers + ees + evs . (2.13) Concerning the atomic species, the rotational and vibrational energy is zero, so that the internal energy is only due to the electronic energy, which Chapter 1. Physical model 9 is related to the electronic excitation. Electrons do not have an internal structure, so their electronic energy is zero. We obtain the following expressions for the internal energy per unit mass: es = ets + ees + evs + ers + hfs es = ets + ees + hfs Atoms ee = ete + hfe Electrons Molecules (2.14a) (2.14b) (2.14c) The expressions for the rotational, vibrational and electronic energy contributions can be obtained using the definition of the partition function as described in the quasi-classical statistical mechanics [14, 46, 73]. Assuming that each internal mode follows a Boltzmann-Maxwell distribution, it is possible to define a single temperature for each degree of freedom. These kind of models are known as multitemperature models [37, 61] For completeness, we provide the expressions of vibrational, rotational and electronic energy e and enthalpy h as given by the multitemperature models. Translational energy and enthalpy Since the translational energy is assumed to be completely excited the translational energy and enthalpy per unit mass are: 3 ets = Rs T, 2 (2.15) hts = ets + Rs T, (2.16) If the free electrons are not in thermal equilibrium with the heavy particles, the translational energy and enthalpy of the electrons depends upon the electrons temperature Te : 3 ete = Re Te , 2 (2.17) hte = ete + Re Te , (2.18) 10 Chapter 1. Physical model t The translational specific heat at constant volume Cv,s and the specific t heat at constant pressure Cp,s are given by Eq. (2.19): 3 t Cv,s = Rs , 2 5 t Cp,s = Rs 2 (2.19) Rotational energy and enthalpy Considering the molecules as a rigid rotor, the rotational energy and enthalpy per unit mass are: ers = hrs = Rs TR (2.20) where TR is the rotational temperature. The rotational specific heat at r r constant volume Cv,s and the specific heat at constant pressure Cp,s are given by Eq. (2.21): r r Cv,s = Cp,s = Rs (2.21) Vibrational energy and enthalpy Considering the molecules as harmonic oscillators, the vibrational energy and enthalpy per unit mass are: evs = hvs = exp Rs ⇥vs ⇣ v⌘ (2.22) ⇥s TV 1 where TV is the vibrational temperature and ⇥vs is the characteristic vibrational temperature of the species s. The rotational specific heat at v v constant volume Cv,s and the specific heat at constant pressure Cp,s are given by Eq. (2.23) (⇥vs /TV )2 exp v v Cv,s = Cp,s = Rs h exp ⇣ ⇥vs TV ⌘ ⇣ ⌘ ⇥vs TV i2 1 (2.23) Chapter 1. Physical model 11 Electronic energy and enthalpy The electronic energy and enthalpy per unit mass are given by Eq. (2.24): ees = hes = Rs P1 e l=1 gs,l ⇥s,l P1 l=1 gs,l exp ⇣ exp ⇣ ⇥es,l TE ⇥es,l TE ⌘ ⌘ (2.24) where TE is the electronic temperature, gs,l is the degeneracy of the electronic level l of the species s and ⇥es,l is its characteristic temperature. e The electronic specific heat at constant volume Cv,s and the specific heat e at constant pressure Cp,s are given by Eq. (2.25) e e Cv,s = Cp,s = where: @ees @TE (2.25) P1 2 l=1 gs,l ⌧s,l exp ( ⌧s,l ) = Rs P 1 @TE g exp ( ⌧s,l ) ⇥P1l=1 s,l e ⇤ P1 l=1 gs,l ⇥s,l exp ( ⌧s,l ) [ l=1 gs,l ⌧s,l exp ( ⌧s,l )] Rs P 2 TE [ 1 l=1 gs,l exp ( ⌧s,l )] @ees where ⌧s,l = ⇥es,l /TE Considering the rotational energy fully excited, the rotational temperature can be set equal to the translational temperature TR = T . Following [61] we can assume that the electronic temperature is in thermal equilibrium with the electron temperature TE = Te . A 3-temperature model is then obtained: 5 Rs T + ees (Te ) + evs (TV ) + hfs 2 3 = Rs T + ees (Te ) + hfs Atoms 2 3 = Rs Te + hfe Electrons 2 es = es es Molecules (2.26a) (2.26b) (2.26c) 12 Chapter 1. Physical model If all the species of the mixture are neutrals, further simplifications can be introduced considering the 2-temperature model [37, 61, 62, 64], in which the electronic temperature is equal to the vibrational temperature TE = TV = Tv : 5 Rs T + ees (Tv ) + evs (Tv ) + hfs 2 3 = Rs T + ees (Tv ) + hfs Atoms 2 es = es Molecules (2.27a) (2.27b) Finally if the flow is in thermal equilibrium it is possible to consider a single temperature model Tve = T . The mixture energy and enthalpy are given by: e= Ns X ↵s es , h= s=1 Ns X ↵s hs (2.28) s=1 State-to-state model In a state-to-state model, each excited internal state is convected as a single chemical species, which implies that there are no electronic and vibrational contributions to the internal energy of each individual chemical species [26, 43, 44]. For the k th excited level of the sth chemical species, the internal energy is given only by the sum of the roto-translational and formation energies: 5 Rs T + hfs,k 2 3 = Rs T + hfs,k 2 es,k = Molecules (2.29a) es,k Atoms (2.29b) There are no electronic or vibrational contributions to internal energy of the single excited level. Therefore in each point in space, the total electronic and vibrational energy for the individual chemical species can be obtained by summing up the formation enthalpy of each excited level, multiplied by the corresponding mass concentration: Chapter 1. Physical model ees = Ls,e X 13 ↵s,k hfs,k , evs = k=1 Ls,v X ↵s,k hfs,k , (2.30) k=1 where Ls,e and Ls,v are the electronic, resp. vibrational number of levels of the sth chemical species. 2.1.4 Conservation equations In this work we have considered multidimensional reacting flows under the conditions of thermal equilbrium, with a single temperature, and in thermal nonequilibrium with two di↵erent temperatures. Hereafter we denote with thermochemical nonequilibrium model (TCneq) the 2temperature model and with chemical nonequilibrium model (Cneq) the 1-temperature model . Conservation equation for the thermochemical nonequilbrium model are given by the continuity equations for each chemical species, the momentum equation, the total energy equation and the vibrational energy equation [37]. Species continuity equations The species continuity equations reads: @⇢s + r · (⇢s u) = S⇢s @t r · Js (2.31) where Js is the di↵usion flux associated to the chemical species s and S⇢s is the mass production term given by the Law of Mass Action Eq. (2.32): S ⇢s = M s Nr X 00 ⌫sr 0 ⌫sr ⇠˙r , (2.32) r=1 where ⇠˙r is the velocity for the rth chemical reaction: ◆⌫sr Ns ✓ Y ⇢ s ⇠˙r = kf r Ms s=1 0 ◆⌫sr Ns ✓ Y ⇢s kbr . M s s=1 00 (2.33) 14 Chapter 1. Physical model In Eqs. (2.32 - 2.33) Ns is the number of chemical species, Nr the number of chemical reactions, kf r and kbr resp. the forward and backward reaction 0 00 rates and ⌫sr and ⌫sr resp. the stoichiometric coefficients of the products and the reagents. The global continuity equation is retrieved summing up all the species continuity equations: @⇢ + r · (⇢u) = 0 @t (2.34) since: Ns X S⇢s = 0, s=1 Ns X Js = 0. (2.35) s=1 Momentum equations The momentum conservation equation is given by the Navier-Stokes equation: @⇢u + r · (⇢uu) + rp = r · ⌧ (2.36) @t where ⌧ is the viscous stress tensor. Under the hypothesis of negligible bulk viscosity e↵ects the viscous stress tensor is given by: ⌧ = µ ru + (ru)t 2 r·u 3 (2.37) where µ is the mixture coefficient of viscosity. Total energy equation The total energy conservation equation is given by: @⇢E + r · (⇢Hu) = r · (⌧ · u) @t r·q (2.38) where E = e + u · u/2 is the total internal energy per unit mass and H Chapter 1. Physical model 15 is the total enthalpy expressed as H = E + p/⇢. In Eq. (2.38) q is the global heat flux, given by: q= where and ductivity. v v rT Ns X rTv Js hfs (2.39) s=1 are the roto-translational and vibrational thermal con- Vibrational energy equation The vibrational energy equation PNs v express the conservation of the mixture v vibrational energy e = s=1 es . @⇢ev + r · (⇢ev u) = @t r · q v + Se v (2.40) v In Eq. (2.40) qv = rTv is the vibrational heat flux while Sev is the vibrational energy source term. In absence of ionzation the vibrational energy source term accounts only for the energy exchange between the roto-translational and the vibrational modes: Se v = Ns X s=1 ⇢s (evs (T ) evs (Tv )) ⌧s , (2.41) where evs is the specific vibrational energy and ⌧s is the relaxation time of the sth species. Vectorial form of the conservation equations Conservation equations (2.31),(2.36),(2.38),(2.40) can be rewritten in a vectorial form as follows: @U + r · Fc = r · F d + S @t (2.42) 16 Chapter 1. Physical model where U stands for the vector of conserved variables: U= ⇥ ⇢i , ⇢u, ⇢E, ⇢ev ⇤t . (2.43) The components of the convective flux tensor Fc are given by: Fc = ⇥ ⇢i u, ⇢uu + pId⇥d , ⇢uH, ⇢uev ⇤t , (2.44) In Eq. (2.45) Id⇥d is the identity matrix of order d , where d is equal to 2 for two-dimensional (2D) flows and 3 for three-dimensional (3D) flows. The di↵usive flux tensor Fd is given by: Fd = ⇥ Js , ⌧, ⌧ · u q, qv ⇤t , (2.45) S is a vector collecting the source terms due to chemistry and thermal relaxation process. S= ⇥ S⇢s , 0t , 0, Sev ⇤ . (2.46) Considering an inviscid flow, the di↵usive terms are neglected so the vectorial form of the conservation equations 2.47 is reduced to the following expression: @U + r · Fc = S @t (2.47) When a chemical reacting mixture is in thermal equilibrium, the flow can be characterized by only a single temperature and the vibrational energy equation can be neglected. Finally for a single inert gas, governing equations are given only by the global continuity equation, the conservation equation of momentum and the conservation equation of total energy. Eq. (2.47), can be rewritten in a quasi-linear form as follows: @U + A · rU = S @t (2.48) Chapter 1. Physical model where the Jacobian matrix of the inviscid fluxes is defined as A = @Fc /@U. The expressions of the Fci fluxes and of the corresponding Jacobian matrices Ai are reported in Appendix A, see Eqs. (A.1) and (A.2). 2.2 Chemical models In this work two di↵erent kind of reacting mixture have been considered: • An ionized argon mixture containing only monoatomic species and electrons. • A dissociated nitrogen mixture containing only neutral atoms and molecules. In the following subsections the two mixtures are described in detail with emphasis on the kinetic process considered. 2.2.1 Kinetic model for an ionized argon mixture In this work we have considered a quasi-neutral argon plasma, i.e. the molar density of the positive particles is assumed to be equal to the molar density of the negative particles. A precise description of an argon plasma in non-equilibrium would require a detailed collisional-radiative model. This implies that not only the chemical reactions between the various species should be accounted for, but also all possible transitions involving the atomic electronic excited levels [75]. Furthermore, the energy distribution of the free electrons may follow a non-Maxwellian distribution function [20, 22, 23, 27]. Due to the extremely high computational cost, this type of modelling cannot be used in the context of a multidimensional CFD approach. For an argon plasma, a good compromise is to consider a reduced number of electronic excited levels for the atomic species and to use a Maxwellian electronic energy distribution function (EEDF) for the electrons, which amounts to define a single temperature for the electrons (Te ). Moreover, we have made the hypothesis of thermal equilibrium between the electrons 17 18 Chapter 1. Physical model and the heavy particles, which amounts to use a single temperature model with Te = T . In our model we take into account only three chemical species: the neutral atoms Ar, the positive ions Ar+ and the electrons e . Following [22], we consider a two-levels system for the neutral atom, with the ground state Ar0 and the 4s metastable state Ar⇤ , while we consider only the ground state for the positive ion Ar+ . The chemical species are reported in Tab. 2.1: we include metastable argon atoms since ionization of the excited atoms is caused by collision with particles of lower energies, so that the role of excitation and ionization from the metastable state cannot be neglected. Moreover, electron induced rate coefficients can be deeply a↵ected by the presence of metastable states resulting in complex structures in the EEDF [16]. Table 2.1: Chemical species in the ionized argon mixture Chemical Species Ground Metastable Positive Ion Electron Symbol Ar0 Ar⇤ Ar+ e Formation energy (eV) 0.0 11.55 15.76 0.0 Statistical weight 1 6 1 0 As shown in Tab. 2.2, the chemical processes we account for are: ionizationrecombination and electronic excitation-de-excitation induced by collision with electrons (r = 1 3) and atoms (r = 4 6). Photoionization and photo-recombination are not included in the model, since their contribution is negligible compared to electronic and atomic processes [78]. Table 2.2: Chemical processes in the ionized argon mixture r 1 2 3 4 5 6 Reaction Ar0 + e ⌦ Ar+ + e + e Ar0 + e ⌦ Ar⇤ + e ⇤ Ar + e ⌦ Ar+ + e + e Ar0 + Ar0 ⌦ Ar+ + e + Ar0 Ar0 + Ar0 ⌦ Ar⇤ + Ar0 ⇤ Ar + Ar0 ⌦ Ar+ + e + Ar0 Description e-A ionization e-A excitation e-A ionization A-A ionization A-A excitation A-A ionization The forward rate coefficients of the electron-atom (e A) processes have been computed for di↵erent values of the electron temperature, integrating the cross section over the corresponding EEDF: Chapter 1. Physical model kr = Z 19 1 &(")⌫e (")fe (")d". (2.49) 0 For a Maxwellian EEDF, Eq. (2.49) becomes: !1/2 Z 1 1 8kB T kr = (")"e (kB T )2 ⇡mr 0 "/kB T d". (2.50) These rate coefficients have been fitted to reduce the computational cost and are reported in tables C.2, C.3 and C.4, in Appendix C. The fitted forward and backward rate coefficients for the electron ionization from the ground level are plotted in Fig C.1 versus the electron temperature. The forward rate coefficients for the atom-atom (A been taken from [5, 75]: p kf r = br T ("ij + 2kB T )e "ij /kB T A) processes have (2.51) where "ij is the di↵erence between the formation energies of the chemical species involved in the reaction and br is a coefficient which depends on the chemical process. Both "ij and br are reported in Tab. C.5, in Appendix C. As far as the atom impact processes are concerned, the backward rate coefficients have been obtained using the detailed balance principle: kbr = kf r . Kr (2.52) The equilibrium constants Kr that appear in Eq. (2.52) are easy to compute once the global ionization equilibrium constant KeqI is known [14]: + KeqI [Ar+ ][e ] QAr Qe = = . [Ar] QAr (2.53) In Eq. (2.53) the concentrations are those at equilibrium and the equilibrium constant KeqI has been directly computed from the complete parti- 20 Chapter 1. Physical model Figure 2.1: Ionization and recombination rates from the ground level as a function of the electron temperature (courtesy of G. Colonna) tion functions of Ar and Ar+ and then fitted as a function of the temperature; the fit function and the coefficients are reported in Appendix C, in Tab. C.1 [15]. For the two-levels system described below, the Ar partition function is: QAr = g0 + g⇤ e "⇤ /kB T (2.54) where g0 and g⇤ are the statistical weights of the ground and metastable species. The Ar+ partition function is equal to the contribution of the ground state only: + QAr = g0+ (2.55) Chapter 1. Physical model 21 and the electron partition function is: Qe ✓ 2⇡me kB Te =2 h2P ◆ 32 . (2.56) To compute the backward coefficients rate of the atom-induced processes, we need to know the expressions of the equilibrium constants for both the ionization and the excitation processes. The equilibrium constants for the ionization processes from the ith excited level can be computed using the following equation: QAr "i /kB T Kr = KeqI e , gi (2.57) while for the excitation processes between the ith and j th levels, the equilibrium constant reads: Kr = gj e gi "ij /kB T . (2.58) In Tab. 2.3 we report the expressions of the equilibrium constants of all the processes accounted for in the present model. Table 2.3: Equilibrium constants for all processes accounted for in the present model 2.2.2 r 1,4 2,5 Equilibrium constant Ar Kr = KeqI Qg0 Kr = gg⇤0 e "0⇤ /kB T 3,6 Kr = KeqI Qg⇤ e"⇤ /kB T Ar Kinetic model for a dissociated nitrogen mixture For the dissociated nitrogen the two temperature model has been used, in which the vibrational energy is not in equilibrium with the rototranslational energy. The dissociated nitrogen mixture is constituted by 22 Chapter 1. Physical model only two neutrals species: the molecular nitrogen and the monoatomic nitrogen. Details about the chemical species are shown in Tab. 2.4. Table 2.4: Chemical species in the dissociated nitrogen mixture Chemical Species Molecule Atom Symbol N2 N Formation energy (kJ/mol) 0.0 470.818 Vibrational temperature (K) 3392.7 The set of chemical reactions includes the atom-molecule (A M ) dissociation and the molecule-molecule (M M ) dissociation (see Tab. 2.5). Table 2.5: Chemical processes in the dissociated nitrogen mixture r 1 2 Reaction N2 + N ⌦ 3N N2 + N2 ⌦ 2N + N2 Description A-M dissociation M-M dissociation The Park model [61] has been considered to model the chemical reaction rates. In this model the forward kf and backward kb rate constants are expressed using the Arrhenius Law : ⌘r kf r = Ar T e ✓r T (2.59) where Ar and ⌘r are some semi-empirical coefficient and ✓r is the activation energy temperature for the reaction r. T in the Park model [37, 40, 61] is an avereged temperature T = T q TV1 q , with q = 0.5. The backward rates are computed using the detailed balance (see Eq.( 2.52) and the equilibrium constants Kr are expressed using the curve fit proposed by Park [37, 61]. 2.3 Electrohydrodynamic model In this thesis the nonequilibrium model has been extended to the case in which an external electromagnetic field is applied. We have investigated the e↵ects of an external electromagnetic field on an inviscid chemical reacting mixture in thermal equilibrium (Cneq). The conservative equations are given by the continuity equations for all the chemical species, the Chapter 1. Physical model 23 momentum equation with the Lorentz force, FB as source term and the total energy equation with the Joule heating, QJ as source term [8, 9, 34]: @⇢i + r · (⇢i u) = Si (2.60) @t @(⇢u) + r · (⇢uu) + rp = FB (2.61) @t @(⇢E) + r · (⇢uH) = QJ (2.62) @t The Lorentz force is given by the vectorial product of the current density j and the magnetic field B. FB = j ⇥ B (2.63) The Joule heating can be computed in each point of the domain as the scalar product of the current density and the electric field E. QJ = j · E (2.64) The electromagnetic variables, j, E and B can be computed coupling the Maxwell equations to the conservation equations [34]. Some simplifications can be made in the cases of our interest since the magnetic Reynolds number, Rem , is assumed to be small [9]: Rem = uL µ0 (2.65) In Eq. (2.65) u is the streamwise velocity of the flow, L is the reference length, is the electrical conductivity and µ0 is the permeability of the free space. When Rem << 1, the induced magnetic field can be neglected and only the external magnetic applied field is present in the flow, then the current density can be determined solving the current continuity equation: r · j = r · [e · (E + u ⇥ B)] = 0 (2.66) where e is the electrical conductivity tensor. In Eq. (2.66) the current density has been expressed using the generalized Ohm’s law, j = e · 24 Chapter 1. Physical model (E + u ⇥ B). The electric field can be expressed as the gradient of an electrical potential as E = r . Whithout an external magnetic field, the Hall e↵ect is absent, then the electrical conductivity tensor reduces to a scalar and Eq. (2.66) reduces to the following Laplace equation: r · ( E) = r · ( r ) = 0 (2.67) where is the scalar electric condutivity. The Ohm’s law is given by the following equation: j= E (2.68) The electrical conductivity of the mixture depends on thermodynamic and electromagnetic variables. As pointed out by Bisek et al. [9], di↵erent electrical conductivity models can lead to very di↵erent results, in the frame of MHD hypersonic flows on a blunt body. The experimental data obtained for the electrical conductivity of an Argon plasma are plotted versus temperature in Fig. 2.2. In the present work, the electrical con- Figure 2.2: Electrical conductivity of argon (p = 0.013atm), reprinted from [8]. Chapter 1. Physical model 25 ductivity has been modeled using the semy-analytic model developed by Chapman and Cowling for weakly ionized gas [9]: = 3.34 ⇥ 10 12 ↵ion p ⌦ & T 1 cm 1 (2.69) where ↵ion is the ionization degree defined by the ratio of the number of ionized particles and heavy particles: ↵ion = nion ntot nion (2.70) and &[cm2 ] is the collision cross section of the gas. The collision cross section is taken to be the total collision cross section for argon-argon collisions using the hard sphere model, with a diameter of d = 4.04 ⇥ 10 10 m [7], then & = 5 ⇥ 10 17 cm2 . A more accurate model would require to include the e↵ect of chemical nonequilibrium and the e↵ect of electric field. External circuit coupling The electrodes have been considered connected with an external electrical circuit as shown in the Fig 2.3. The electric potential on the electrodes is given by Ohms law for the external circuit: Vdis (t) = Vg Rc I (t) (2.71) where Vdis is the discharge voltage, Vg the power supply potential, Rc the circuit resistence and I the electric current in the circuit. The electric current can be computed as the surface integral of scalar product of current density vector and the surface vector over the elecrodes. As the surface integral of the scalar product of the current I= Z electrodes j · dA (2.72) At each time step the discharge voltage changes according to the Eq. (2.71), and the electric potential of one of the electrodes is set equal to ref + 26 Chapter 1. Physical model Rc# V g# +" #" I# electrodes" Vdis# Figure 2.3: Scheme of the coupled circuit Vdis (t), where ref is a reference value for the electric potential (generally ref is set equal to zero). Chapter 3 Computational tools In this chapter the numerical methods used to the discretize the governing equations will be described. The chemical nonequilibrium model has been implemented within eulfs [10, 13], an unstructured 2D/3D solver developed for thermally and calorically perfect gas. The CFD code can be coupled with a newly developed, unstructured, shock-fitting algorithm which treats the discontinuities as moving boundaries that border regions of the flow-field were a smooth solution to the governing PDEs exists. The unstructured shock-fitting algorithm has been extended to deal with an ionised argon mixture to model shock waves in chemical reacting flows and with The shock-capturing discretization will be presented first, since it is also used in the shock-fitting approach to solve the governing PDEs in the smooth regions of the flow-field. One of the key features of the unstructured shock-fitting algorithm is its modularity: the shock-capturing code used to discretise the governing PDEs (2.47) in the smooth regions of the flowfield is used as a “black box” by the shock-fitting algorithm. The only constraint is that the unstructured shock-capturing solver must feature a vertex-centred storage of the unknowns. In [12, 42, 57, 58] the unstructured shock-fitting algorithm had been coupled with the in-house eulfs shock-capturing code. In the present work, the shock-capturing solver has been coupled, with a very limited coding e↵ort, with the VKI-developed COOLFluiD [50, 51] solver. This was 27 28 Computational tools done both for demonstrating the modularity of the unstructured shockfitting algorithm, but also because COOLFluiD features a wider range of modelling capabilities than eulfs. 3.1 Shock-capturing solver Both codes, eulfs and COOLFluiD, use Residual Distribution schemes (RD) 1 for the spatial discretization, so we briefly introduce the main characteristics of the RD discretization for the governing PDEs (2.47) [1, 28, 71]. The computational domain ⌦ ⇢ Rd is tessellated into triangles in the 2D space, and tetrahedra in 3D. A dual tessellation is also defined, which consists in the so-called median dual cells. Both the primal and dual tessellations are shown in Fig. 3.1 for the 2D case: Ci is the median dual cell centred about gridpoint i and ⌦e is the triangle e. The dependent variables are stored at the vertices of the computational mesh and are assumed to vary linearly and continuously in space, just as in iso-P1 Finite Elements. 2 3 1 e 4 5 (a) The flux balance of cell ⌦e is scattered among its vertices. (b) Grid-point i gathers the fractions of cell residuals from the surrounding cells. Figure 3.1: Residual distribution concept. We use the term residual (or fluctuation) to refer to the sum of the surface integral of the inviscid and di↵usive fluxes through the control 1 RD schemes are also known under the name of Fluctuation Splitting schemes (FS). Computational tools 29 volume boundary and the volume integral of the source term. Observe that the residual vanishes whenever U is a steady solution of Eq. (2.47). At the discrete level, both cell and nodal residuals can be defined: for cell residuals the control volume coincides with the simplicial element (triangle/tetrahedron) whereas for nodal residuals the control volume is the median dual cell. When using simplicial meshes, as we do with FS schemes, the number of cells exceeds the number of gridpoints: by a factor close to 2 in two dimensions and ranging between 5 and 6 in three dimensions; therefore, since the FS discretization is vertex-centred, at steady-state it will only be possible to drive the nodal residuals to zero, not the cell residuals. With FS schemes, rather than calculating the nodal residual by numerical quadrature along the boundary of the median dual cell and over its volume, as would be done with conventional Finite Volume schemes, the nodal residual is obtained by collecting fractions of the cell residuals of the triangles/tetrahedra that meet at mesh-point i. This is accomplished using a main loop over all triangular/tetrahedral elements in the mesh: within each element the cell fluctuation (or residual) is computed and then split into signals which are sent to the vertices of the cell (see Fig. 3.1(a)). At the end of the loop, the nodal residual has been assembled within each gridpoint by collecting the signals scattered by the elements that surround that gridpoint (see Fig. 3.1(b)). 3.1.1 Fluctuation and conservative linearization Given an element e with volume ⌦e and boundary surface @⌦e , the fluctuation (or residual ) e is defined as follows: I I Z e c d e,d e,s = F · ndS F · ndS SdV = e,c (3.1) @⌦e @⌦e ⌦e where e,c and e,d are resp. the net inviscid flux and the di↵usive flux balance across @⌦e and e,s the volumetric integral of the source term. The net inviscid flux over each triangular/tetrahedral element is evaluated by means of a conservative linearization that allows to compute the inviscid flux balance over a cell using the quasi-linear form (2.48) of the Euler system: 30 Computational tools e,c = I @⌦e F · n dS = Āi ✓ ◆ @U |⌦e |. @xi (3.2) The problem stated by Eq. (3.2) amounts to define proper cell averages of both the Jacobian matrices Āi and of the gradient of the conserved variables rU. Conservation is essential to guarantee correct shock-capturing, whereas working with the quasi-linear form of the equations allows to give an upwind flavour to the discretization. When dealing with a single-component, calorically and thermally perfect gas, this can be accomplished [29] by using Roe’s parameter vector: Z= p ⇢ (1, H, u)T (3.3) as the dependent variable featuring piece-wise linear variation over each simplicial element. Indeed, since both the conserved variables U and the components Fi of the inviscid flux vector are quadratic function of the components of the vector Z, the cell fluctuation can be computed exactly as follows: e,c I Z ✓ = F · n dS = @⌦e ◆✓ ◆ ⌦e Z ✓ @Fi @Z dV = @Z @xi ⌦e @Fi @xi ◆ dV = (3.4a) ✓ ◆✓ ◆ @Z @Fi |⌦e | (3.4b) @xi @Z Z=Z where we have taken advantage of the fact that rZ is cell-wise constant and the value of @Fi /@Z, averaged over a cell, equals its analytical expression evaluated in the arithmetic average: ! X 1 Z= Zj d + 1 j2e since @Fi /@Z is also linear in the components of Z. The cell averaged Jacobian matrices and the gradient of the conserved variables in Eq. (3.2) can be written as in [29]: Āi = Ai Z rU = ✓ @U @Z ◆ Z=Z rZ. Computational tools 31 The conservative linearization can be extended in a straightforward manner to a multi-species fluid in chemical non-equilibrium only in the restrictive case in which each species has the same number of degrees of freedom (hence the same adiabatic index = Cp /Cv ) and behaves as a calorically perfect gas. In this circumstance, it has been shown by Degrez et al. [30] by using the following definition for Roe’s parameter vector: p Z = ⇢(↵i , H, u), (3.5) the vector of the conservative variables and of the inviscid fluxes are homogeneous functions of second degree in Z, just as in the perfect gas case. By contrast, when dealing with an arbitrary mixture of thermally perfect gases or a single calorically imperfect gas, pressure is not any longer a quadratic function of Z and the issue of finding an averaged state becomes considerably more complicated. As far as FS schemes are concerned, a Roe-type conservative linearization has only been obtained in the case of arbitrary fluids in chemical non-equilibrium [30]. A di↵erent approach, named Conservative Residual Distribution (CRD) [25], which consists in calculating the flux integral by means of numerical quadrature, thus avoiding the conservative linearization, is used when dealing with thermochemical non-equilibrium flows; details can be found in Refs. [47– 49]. The availability of a conservative linearization is also a key ingredient to generalise the scalar FS schemes to non-linear hyperbolic systems of equations. Although the individual Jacobian matrices Ai cannot be simultaneously made diagonal in more that one space dimension, hyperbolicity of the Euler system implies that matrix Ān = Āi ni , n = ni ei being an arbitrary unit vector, has real eigenvalues and a complete set of linearly independent eigenvectors, so that it can be diagonalised as follows: ¯n Ln Ān Rn = ⇤ (3.6) 1 where Ln and Rn = Ln are the matrices of left and right eigenvectors ¯ and ⇤n the diagonal matrix of the corresponding eigenvalues. Detailed expressions of the various matrices that appear in Eq. (3.6) are given in Appendix A, see Eqs. (A.6), (A.7), (A.9) and (A.10). The volumetric integral of the chemical source term is computed using a quadrature rule [30]. The simplest possible choice is the one-point 32 Computational tools quadrature rule: e,s = Z ⌦e S dV = S (xec ) ⌦e = S Z ⌦e . (3.7) where xec is the centroid of the element e. 3.1.2 Signals or Residual Distribution Once the cell fluctuation has been computed according to Eq. (3.2), it is e,d scattered among the d + 1 vertices of the cell by sending signals e,c i , i and e,s which must fulfil the following conservation constraints: i e,c e,d e,s = = = d+1 X i=1 d+1 X i=1 d+1 X e,c i , (3.8a) e,d i , (3.8b) e,s i . (3.8c) i=1 Eq. (3.8a) guarantees global conservation in space in the sense that the inviscid flux through the boundary of the computational domain is exactly recovered once the inviscid component of the fluctuations of all cells in the mesh have been summed up. Di↵erent criteria can be devised to split the cell fluctuation into signals. Among them, there are schemes that depend linearly upon the solution (when solving a linear PDE) and are either monotonicity preserving, but limited to first order of accuracy, as the case of the N scheme, or, if second order accurate, may lead to oscillatory behaviour in the neighbourhood of a captured discontinuity, which is the case of the LDA scheme [72]. Within cell e, N and LDA schemes send the following signals: e,N i = K+ i (Ui e,LDA i U ), = where: U± = d+1 X j=1 ± j Uj and ± j = d+1 X `=1 K± ` ! + i e (3.9) 1 K± j . (3.10) Computational tools 33 and Ki is the nodal upwind parameter, defined as: 1 Ki = Āni Si d (3.11) In Eq. (3.11) Āni = Ā · ni is the Jacobian matrix projected along the direction of the unit inward normal to the face opposite to node i. Si is the area of the face opposite to node i as it is shown in Fig. 3.2. Si ni i Figure 3.2: Inward scaled normal. Using Eq. (3.6) we obtain: 1 ¯ ± L n Si K± = Rn i ⇤ ni i i d (3.12) ¯ ± is defined as: where the generic element of the diagonal matrix ⇤ n ± k = 1 ( 2 k ± | k |) . is one of the m = d + Ns + 1 eigenvalues of matrix Ān . Non-linear schemes that capture the discontinuities monotonically and preserve second order of accuracy in smooth regions of the flow-field can be constructed by using a solution-dependent weighting function which blends the linear N and LDA schemes in such a way that the former scheme is activated only in the neighbourhood of the captured discontinuities whereas the latter is used elsewhere. k 34 Computational tools e,B i e,N i = ⇥e + (I ⇥e ) e,LDA i e (3.13) These kind of non-linear schemes (the matrix ⇥ depends upon the local solution) are referred to as B (blended) schemes. The CRD variants of the RD schemes are identified by a c character appended to the name of the scheme; for example the CRD variant of the LDA scheme is named LDAc. The di↵usive contribution to the cell residual are distributed with the Galerkin method. When a mesh made of iso-P1 (simplicial) elements is used, the di↵usive cell residual sended to the node j reads: d i = Z ⌦e Ni r · Fd dV (3.14) The fluctuation associated with the source term is split into signals using the LDA weighting, i.e. e,s + e,s (3.15) i i = The motivation for using an upstream weighted distribution of the source terms lies in the analogy between FS and Petrov-Galerkin FE schemes [17]. 3.1.3 Solution of the discretised equations Once the cell fluctuation has been computed and scattered among the vertices of each cell, the nodal residual in gridpoint i is obtained by collecting all the signals scattered from its neighbouring elements. Ri (U) = = I c @C X⇣i i3e F · n dS + e,c i + I d F · n dS + @Ci ⌘ e,d e,s . i + i Z S dV (3.16a) Ci (3.16b) Equation (3.16) is the discretized version of the spatial terms in the governing conservation equations, where the control volume Ci is the median dual cell. Computational tools Steady solutions of the governing conservation laws, have been calculated by solving the following pseudo-unsteady system of equations: dU V = R (U) (3.17) d⌧ introducing the pseudo-time ⌧ . Since accuracy in pseudo-time is irrelevant, the mass matrix has been lumped into the diagonal matrix V whose entries are the areas/volumes of the median dual cells and a first-order accurate, two time levels Finite Di↵erence (FD) formula: dU U = + O ( ⌧) U = Un+1 Un (3.18) d⌧ ⌧ is used to approximate the pseudo-time derivative in the l.h.s. of Eq. (3.17). An explicit scheme is obtained by evaluating the nodal residual in Eq. (3.17) at the known pseudo-time level n. Explicit time stepping is used when the code is operating in shock-fitting mode (described in Sect. 3.2), whereas implicit time stepping is used when the code works in shock-capturing mode. An implicit scheme is obtained if the residual R in Eq. (3.17) is evaluated at the unknown pseudo-time level n + 1. expanding R about pseudotime level n and introducing the non uniform pseudo-time-step ⌧n , one obtains the following sparse system of linear equations: 1 @R V J (Un ) U = R (Un ) , J= (3.19) ⌧n @U to be solved at each pseudo-time-step until the required convergence of R is obtained. Two distinct approaches are currently available for calculating the Jacobian matrix (J) of the residual: i) one in which some approximations are introduced in convective terms and chemical source terms are neglected in the calculation of the Jacobian and ii) an FD approximation of the “exact” Jacobian which accounts for all terms in the equations incuring only into the truncation error of the one-sided FD formula. Even if the analytical approximation of the Jacobian is much cheaper to compute than its FD approximation, it is less robust. Moreover the FD Jacobian approximation allows, in the limit of infinite ⌧n to recover Newton’s root-finding 35 36 Computational tools algorithm, which is known to yield quadratic convergence when the initial guess is close to the steady solution. However, unless a reasonably good initial guess is available, the implicit scheme built upon the FD Jacobian approximation may quickly diverge. A typical run is therefore split in two steps. During the first step, the approximate, analytical Jacobian is used to start the calculation from scratch until the residual has dropped some orders of magnitude; in this first phase the pseudo-time-step length is not allowed to increase beyond values corresponding to a CFL number of the order of one hundred. Then, the solution is re-started using the FD Jacobian approximation and the pseudo-time-step is allowed to grow un-boundedly so that quadratic convergence is eventually recovered over the last pseudo-time-steps. At each inner iteration, The linear system (3.19) is solved using the suite of iterative, preconditioned Krylov subspace algorithms available in the PETSc library [6]. Further details regarding the Newton-Krylov algorithm can be found in Ref. [11]. 3.2 Shock-fitting algorithm The unstructured shock-fitting algorithm that has been recently developed by Paciorri and Bonfiglioli [12, 42, 57, 58] consists of two key ingredients: i) a local re-meshing technique that constructs a time-dependent mesh in which the fitted discontinuities are internal boundaries of zero thickness and ii) an algorithm for solving the Rankine-Hugoniot jump relations that provides the Lagrangian velocity of the discontinuity and an updated set of dependent variables within the downstream side of the fitted shock. More precisely, in two space dimensions the fitted shock fronts are made of polygonal curves, i.e. a connected series of line segments (which we call the shock edges) that join the shock points. The downstream state and the shock speed are computed according to the Rankine-Hugoniot jump relations and the fitted shock is allowed to move throughout a background triangular mesh that covers the entire computational domain. At each time level t the starting point is given by the solution on a background mesh and a polyline describing the shock (see Fig. 3.3). The Computational tools 37 (a) (b) Figure 3.3: Starting point: a) background mesh and b) shock boundary (right). position of the shock nodes are totally independent by the location of gridpoints on the background mesh. At each node of the shock two states are associated: one for the upstream state and one for the downstream state. The process of updating the mesh and solution to time level t + t can be split into di↵erent steps that will be described in details: 1. Cell removal. The first step consists in finding and removing the cells around the shock front. All cells of the background mesh that are crossed by the shock boundary are removed with the gridpoints too close to the shock front. These mesh nodes are identified as“phantom” nodes (dashed circles in Fig. 3.4(a)). 2. Local re-meshing. After the cell removal, the remaining part of the background mesh is split into two separated regions. A constrained Delaunay triangulation (CDT) is performed to remesh the hole (see Fig. 3.4(b)). 3. Computation of the normal and tangent vectors. The normal n and tangent ⌧ unit vectors are computed for each shock node (Fig. 3.5(a)). These vectors in a generic shock node are computed by means of finite-di↵erence formulas which involve the coordinates of the shock point itself and the neighbouring shock points. 38 Computational tools (a) (b) Figure 3.4: a) Cell removal. b) Local remeshing. 4. Solution update using the S-C solver. The solution is updated by the S-C code to time level t+ t on the modified mesh by treating the shock front as an internal boundary. The shock downstream states are not accurately updated except for the characteristic rising from the downstream region (Rd ): Rdt+ t = at+ d t 1 ut+ d t ·n (3.20) 2 where ad and ud are the values of the sound and flow velocity of the downstream state of the shock nodes. The shock is treated as an internal boundary by the S-C solver. + 5. Shock calculation using the R-H relations. The upstream state of each shock node have been correctly updated during the S-C solver computation, whereas only the Rd quantity has been correctly updated within the shock- downstream state. The exact downstream state and shock speed Wsh of each shock point are computed by solving a system of five algebraic non-linear equations given by the four R-H relations, the Eq. (3.20), where the value Rd has been computed by the S-C solver. 6. Interpolation of the phantom nodes. After the exact update of the downstream state the solution is updated also on the phantom nodes which belong to the background mesh.This update is Computational tools (a) 39 (b) Figure 3.5: a) Normal and tangential unit vector computation. b) Interpolation of the phantom nodes. performed finding the cell containing the phantom node and interpolating the phantom node state using the states of the cell vertices (see Fig. 3.5(b)) 7. Shock displacement. The position of the shock nodes at time t + t is computed considering the following first order integration formula: t+ t xt+ t = xt + Wsh n t (3.21) Figure 3.6: Interpolation of mesh point jumped by the shock boundary. The shock-fitting algorithm, originally developed for an ideal gas, has been extended to Cneq and TCneq flows. The generalization to a reactive 40 Computational tools mixture has required a minimum e↵ort, since modeling the bow shock as a partly dispersed shock wave [73], the concentrations of chemical species are kept frozen while passing through the bow shock. When dealing with thermochemical non-equilibrium, the specific vibrational energy is kept constant through the bow shock. 3.3 Details on the implementation The shock-fitting algorithm has been implemented in a modular way, so that the CFD solver, the remeshing tool are used as black boxes as shown in Fig. 3.7. Comunication between the CFD solver and the remeshing tool is performed by means of I/O files, then two converters are needed in order to exchange the right data between the codes. Since the shock-fitting code uses the adimensional Roe parameter vector Z⇤ (see Appendix A), a variable conversion is performed into the converters. Local re-meshing around the shock is accomplished using public domain software: triangle [68] in 2D and tetgen [70] and yams [31] in 3D. Computational tools Figure 3.7: Modularity of the shock-fitting code. 41 42 Chapter 4 Numerical results In this chapter the numerical results will be presented. Results obtained considering the ionized argon model implemented within the CFD code eulfs are reported in Sect. 4.1. The results obtained using the CFD code COOLFluiD for perfect gas and dissociating nitrogen flows are presented in Sect. 4.2. Finally the results obtained considering the electrohydrodynamic interaction are shown in Sect. 4.3. 4.1 4.1.1 eulfs results Ionized argon, inviscid flow in a nozzle In order to validate the chemical model implemented within the eulfs code, results obtained from the 2D and 3D simulation of the flow through a converging-diverging nozzle have been compared with those obtained using a Q1D code developed at IMIP-CNR [20, 21, 24]. In this reference code, the inviscid Q1D conservation equations are solved using a space-marching algorithm and are coupled with state-to-state kinetics and Boltzmann equation for free electron transport and the rates of electron collision processes are obtained by integrating the electron impact cross section over the EEDF. Therefore, in order to make the code-to-code validation meaningful, it has been necessary to downgrade the chemical model available in the IMIP-CNR code to make it identical to the one, described in Sect. 2.2.1, that has been implemented within the eulfs 43 44 Numerical results CFD code. The analytical area distribution of the converging-diverging nozzle is given in Tab. 4.1 whereas Tab. 4.2 shows the reservoir test conditions. In the 2D/3D simulations the total temperature, total pressure and flow angles have been specified along the subsonic inlet section and the metastable argon concentration has been set equal to that obtained considering a Boltzmann distribution for the electronic levels. The flow is supersonic at the outflow section and thus requires no boundary condition there. Figure 4.1 shows the 2D and 3D geometries with Mach iso-contours superimposed: the 2D grid is made of 4439 grid-points and 8447 triangles and the 3D grid is made of 28581 grid-points and 155667 tetrahedral cells. The 2D/3D simulations use the second order accurate LDA scheme for the spatial discretization. Y Z X M 6.5 6 5.5 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 Figure 4.1: Flow in a converging-diverging nozzle: 2D and 3D geometries flooded by Mach number. Comparison between eulfs and Q1D IMIP code have been made for two di↵erent thermodynamic inlet conditions (see Tab. 4.2). Figures 4.2 and 4.3 show a comparison of the area weighted distributions along the nozzle axis; data include: the molar fractions of ions and metastable argon. The results obtained from the 2D and 3D simulations have been area weighted in order to be compared with the Q1D result Numerical results 45 Table 4.1: Area distribution along the nozzle. x 0.3 < x < 0.0 0.0 < x < 0.5 area A(x) = 0.010 + 0.6x2 A(x) = 0.010 + 0.39 x2 + 0.26 x3 Table 4.2: Inlet flow conditions. Physical quantity Total pressure Total temperature Ionisation degree Test A 1 ⇥ 105 Pa 5000K 1 ⇥ 10 7 Test B 1 ⇥ 105 Pa 8000K 1 ⇥ 10 5 Mach number Temperature 7 5000 Q1D IMIP EulFS 3D EulFS 2D 4000 Q1D IMIP EulFS 3D EulFS 2D 6 Mach number 5 T (K) 3000 2000 4 3 2 1000 1 0 -0.4 -0.2 0 0.2 0 -0.4 0.4 -0.2 0 0.2 (a) Temperature. (b) Mach number. Ar+ Ar* 1.5e+11 2e+07 Q1D IMIP EulFS 3D EulFS 2D Q1D IMIP EulFS 3D EulFS 2D particles number density, 1/m^3 N (1/m^3) 1e+11 5e+10 0 0.4 x (m) x (m) -0.2 0 0.2 x (m) (c) Argon ions: Ar+ . 0.4 1.5e+07 1e+07 5e+06 0 -0.2 0 x, m 0.2 (d) Metastable argon: Ar⇤ . Figure 4.2: Test A: distributions along the nozzle axis. 0.4 46 Numerical results Mach number 7 8000 5 Mach number 6000 Temperature, K Q1D IMIP EulFS 3D EulFS 2D 6 Q1D IMIP EulFS 2D EulFS 3D 4000 4 3 2 2000 1 0 -0.2 0 x, m 0.2 0 -0.4 0.4 -0.2 0 (a) Temperature. 0.4 (b) Mach number. Ar+ Ar* 1e+12 3e+11 Q1D IMIP EulFS 2D EulFS 3D 6e+11 4e+11 Q1D IMIP EulFS 2D EulFS 3D 2.5e+11 Particles number density, 1/m^3 8e+11 Particles number density, 1/m^3 0.2 x (m) 2e+11 1.5e+11 1e+11 2e+11 5e+10 0 -0.2 0 x, m 0.2 (c) Argon ions: Ar+ . 0.4 0 -0.2 0 x, m 0.2 (d) Metastable argon: Ar⇤ . Figure 4.3: Test B: distributions along the nozzle axis. 0.4 Numerical results 47 of the IMIP-CNR code. As shown in Fig. 4.2(a) and 4.3(a), comparisons for temperature along the axis x, are satisfactory. Mach number distributions (Fig. 4.2(b) and 4.3(b)) show some di↵erences close to the outlet boundary probably due to non-Q1D e↵ects in the diverging duct. The 2D and 3D results are superimposed, despite the fact that the 2D grid is somewhat finer than the 3D one. The comparison between the 2D/3D calculations and the reference Q1D calculation can also be considered satisfactory, given the di↵erent numerical models, inlet boundary conditions and grid resolutions being used and confirms the correct implementation of the chemical model. The flow in test A can be considered frozen, since at low temperature, the characteristic time of chemistry is much greater than the characteristic time of fluid dynamics. In test B the flow is in non-equilibrium conditions. Sample convergence histories for the 2D calculation are shown in Fig. 4.4: the L2 -norm of the nodal residual is plotted against the CPU seconds for each conservative variable. The convergence history shown in Fig. 4.4(a) has been obtained using the analytical Jacobian approximation and a maximum CFL number equal to 100: it can be seen that all residuals converge steadily towards machine zero. In the convergence history shown in Fig. 4.4(b) the first 50 iterations were run using the analytical Jacobian approximation, then the FD Jacobian approximation was put in place and the maximum CFL number set to infinity. By doing so, only 5 Newton steps (marked by ⇥ in Fig. 4.4(b)) are needed to drive the residuals to machine zero. Picard Newton (a) Approximate, analytical Jacobian. (b) Approximate, followed by FD Jacobian. Figure 4.4: Flow in a converging-diverging nozzle: convergence histories for test B. 48 Numerical results 4.1.2 Ionized argon, inviscid flow over a circular cylinder Hypersonic flows over blunt bodies are characterised by strong nonequilibrium conditions. For an hypersonic flow in argon, ionization and electronic excitation may significantly a↵ect the macroscopic physical quantities, such as temperature or Mach number, within the entire flow field surrounding the body. In this work we have considered the 2D hypersonic flow past the fore-body of a circular cylinder in order to conduct a comparative assessment of the predictive capabilities of the alternative shock-capturing and shock-fitting options available in the unstructured solver. The grid used for the shock-capturing calculation and also as the background triangulation in the shock-fitting calculation is made of 38018 nodes and 19348 elements; the shock-fitting grid at steady-state, which di↵ers from the background triangulation only in the neighbourhood of the fitted shock (see Fig. 4.5), has 19613 gridpoints and 38190 triangles. The grid has been refined near the wall and in particular in the stagnation region, since chemical activity is very important in this region. In all 0.15 0.1 0.02 0.05 y (m) y (m) 0.01 0 0 -0.01 -0.05 -0.02 0.05 0.06 0.07 0.08 0.09 x (m) -0.1 -0.15 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 x (m) Figure 4.5: Baseline mesh and “fitted” mesh. Numerical results 49 shock-capturing calculations presented herein, the spatial discretization relies upon the first order accurate N scheme. Shock-fitting calculations rely upon the first order N scheme and the second order non-monotone LDA scheme. Flow conditions are given in Tab. 4.3: due to the low freestream temperature, the shock-upstream flow is in equilibrium conditions and made only of neutral argon. Table 4.3: Freestream conditions for an ionized argon flow over a 0.05m radius cylinder. Physical quantity M1 p1 T1 u1 Radius Free-stream Value 11.2 543.95 Pa 298.7 K 3536.28 m/s 0.05 m Figure 4.6 shows the static temperature (Fig. 4.6(a)) and Mach number (Fig. 4.6(b)) iso-contours computed by means of shock-capturing (upper half of the frames) and shock-fitting (lower half). 0.15 0.15 0.1 0.1 -0.05 0.02 0.05 10.5 9.5 8.5 7.5 6.5 5.5 4.5 3.5 2.5 1.5 0.5 0 0 -0.01 -0.05 -0.02 0.05 0.06 0.07 0.08 0.09 0.01 y (m) 0.01 y (m) 0 0.02 y (m) y (m) Mach T (K) 12500 11000 9500 8000 6500 5000 3500 2000 500 0.05 0 -0.01 -0.02 0.05 x (m) 0.06 0.07 0.08 0.09 x (m) -0.1 -0.1 -0.15 -0.15 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0 0.05 0.1 0.15 0.2 0.25 x (m) x (m) (a) Temperature (K) (b) Mach number. 0.3 0.35 Figure 4.6: 2D flow past a circular cylinder: comparison between the shock-capturing (S-C) and shock-fitting (S-F) solutions. Both S-C and S-F solutions are first order accurate in space (N scheme). It is clear from the comparison between the two sets of calculations that shock-fitting gives a much more realistic shock-thickness than does shockcapturing, without the need to adapt the mesh in the shock-normal di- 50 Numerical results rection, as would be the case if mesh adaptation was used in conjunction with the shock-capturing solver. The remarkable thickness of the X 12000 0.01 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X 10000 0 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X -0.01 X X (H - Href)/Href T (K) 8000 X 6000 X -0.02 -0.03 -0.04 X X 4000 -0.05 X X 2000 Shock-Capturing (N) Shock-Fitting (N) Shock-Fitting (LDA) X X X X X X 0 0.05 0.06 Shock-Capturing (N) Shock-Fitting (N) Shock-Fitting (LDA) X -0.06 0.07 0.08 -0.07 0.05 0.09 0.055 0.06 0.065 x (m) 10 -6 X X X X X X X X X X X X X X X X X X X X X X X X 10-3 10-4 X X X X X X X X -7 X X X 10-8 X X X 10 -9 0.085 X X X 10 0.08 (b) Nondimensional enthalpy. Molar fractions (Ar+) Molar fractions (Ar*) X X -5 0.075 x (m) (a) Temperature (K) 10 0.07 X X X 10 -5 10 -6 X X X 10-7 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X 10-8 X Shock-Capturing (N) Shock-Fitting (N) Shock-Fitting (LDA) X 10 -9 X Shock-Capturing (N) Shock-Fitting (N) Shock-Fitting (LDA) X X X X X 10 -10 0.05 X X X X X X 0.055 0.06 0.065 0.07 0.075 0.08 x (m) (c) Ar⇤ molar fraction. 0.085 0.09 10 -10 0.05 X X X X X X 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 x (m) (d) Ar+ molar fraction. Figure 4.7: 2D flow past a circular cylinder: comparison of shock-capturing (S-C) and shock-fitting (S-F) solutions along y = 0. captured bow-shock is particularly evident in Fig. 4.7(a), where symbols denote the intersections of the stagnation streamline (the x-axis) with the triangular grid. As we can see in Fig. 4.7(b), shock-fitting allows to conserve total enthalpy across the captured shock. Even though the shock-capturing solution recovers the “exact” post-shock state, all those flow states that are inside the shock are a mere numerical artifact that also a↵ect the chemistry, as we shall describe later. Moreover, Numerical results the better description of the bow shock provided by shock-fitting has a clear, beneficial impact also on the smooth flow within the entire shock layer: both the temperature and Mach iso-contours within the shockdownstream region are smoother in the shock-fitting solution than they are in the shock-capturing one, see Fig. 4.6. Finally, we observe that the chemical activity in the flow-field is the consequence of the temperature rise across the bow shock; this is evident from Fig. 4.7(c) and 4.7(d), which show the molar concentrations of metastable argon and argon ions along the stagnation streamline, for both sets of calculations. It is interesting to observe how the di↵erent modeling practices, capturing versus fitting, have also an impact upon the non-equilibrium chemistry within the shock-downstream region. Indeed, since the captured shock is remarkably thick, in the shock-capturing calculation chemical reactions are activated already inside the shock, whereas they occur just behind the fitted shock in the shock-fitting solution. Due to the memory e↵ect of the non-equilibrium chemical model, the “artificial” chemical concentrations that are created inside the captured shock are also felt downstream of the shock-wave, even beyond the spatial location, see Fig. 4.7(c) and 4.7(d), were both the shock-capturing and shock-fitting solutions have reached the same post-shock temperature. Therefore, the molar concentrations computed by means of shock-capturing and shock-fitting di↵er behind the shock and the di↵erences will eventually vanish only further downstream. Concerning the shock-fitting calculations, first order and second order accurate solutions di↵er slightly only in the post-shock region. Di↵erences in molar concentrations between the shock-capturing and shock-fitting solutions just downstream of the bow shock are due to the very di↵erent descriptions of the shock wave given by the two di↵erent modeling options. 4.2 4.2.1 COOLFluiD results Ideal gas, inviscid flow over a circular cylinder An hypersonic inviscid flow past a circular cylinder at freestream Mach number, M1 = 20, has been considered as a first test case. Results for this same test case had already been obtained by Paciorri and Bonfiglioli 51 52 Numerical results [57] using the eulfs solver. Two di↵erent grids with increasing spatial resolution have been used to asses the improvements in accuracy of the shock-fitting code. The two meshes shown in Fig. 4.8 while the number of triangles and nodes are reported in Tab. 4.4. The shock-fitted grids di↵er from the corresponding background ones because of the addition of the shock points; the background grids have also been used for the calculations in which the shock is captured. All the results obtained for the ideal inviscid gas flows have been carried out by means of an explicit backward Euler time integration scheme. Table 4.4: Characteristics of the grids used for the inviscid hypersonic flow over a circular cylinder. coarse grid Nodes Elements 351 610 411 654 Background grid Shock-fitted grid 3 fine grid Nodes Elements 5261 10151 5415 10229 3 2 2 0.6 0.6 0.4 0.4 1 1 0 y(m) 0 0.2 y(m) y(m) y(m) 0.2 0 -0.2 -1 -0.2 -1 -0.4 -0.6 1 1.2 1.4 1.6 1.8 0 1 2 3 4 -0.4 -0.6 2 x(m) -2 -3 0 1 1.2 1.4 5 6 -3 1.6 1.8 2 x(m) -2 0 1 2 3 4 x(m) x(m) (a) Coarse grid (b) Fine grid 5 6 Figure 4.8: Grids used for the inviscid flow over a circular cylinder. The comparison between the shock-capturing and shock-fitting results is shown in Fig. 4.9, where pressure contour plots are shown for both the coarse and the fine grid solutions. The non-linear B scheme has been used to obtain a nominally second-order-accurate solution when the shock is captured, whereas the linear LDA scheme has been used in Numerical results 53 3 3 2 2 0.6 0.6 0 S-F y(m) y(m) 0.2 S-C S-C 0 S-F -0.2 -1 -0.4 -0.6 1 1.2 1.4 -3 1.6 1.8 0 1 2 3 4 0.2 S-C 0 S-F S-C 0 S-F -0.2 -1 -0.4 -0.6 2 x(m) -2 1 1.2 1.4 6 -3 1.6 1.8 2 x(m) -2 5 p(Pa) 5E+07 4.6E+07 4.2E+07 3.8E+07 3.4E+07 3E+07 2.6E+07 2.2E+07 1.8E+07 1.4E+07 1E+07 6E+06 2E+06 0.4 1 y(m) 0.4 1 y(m) p(Pa) 5E+07 4.6E+07 4.2E+07 3.8E+07 3.4E+07 3E+07 2.6E+07 2.2E+07 1.8E+07 1.4E+07 1E+07 6E+06 2E+06 0 1 2 3 4 x(m) x(m) (a) Coarse grid (b) Fine grid 5 6 Figure 4.9: Pressure contour plot: comparison between the shock-fitting (S-F) and shock-capturing (S-C) second-order-accurate solutions. the shock-fitting calculation. The use of a linear scheme in the shockfitting calculation is made possible by the fact that, when all shocks are fitted, the LDA scheme is used only in smooth regions of the flow field. Figure 4.9 clearly shows that, using grids of comparable resolution, shockfitting not only avoids smearing the shock, but it also allows to obtain a much cleaner solution within the entire shock layer characterized by a low numerical error also on the coarse mesh. It is worth to observe that the shock-capturing solution computed on the coarse mesh features a strong asymmetry with respect to the stagnation point, whereas the asymmetry is completely absent in the shock-fitting solution computed on the corresponding coarse mesh. Finally, the results obtained for the this same testcase by coupling the two di↵erent CFD solvers, COOLFluiD and eulfs, with the same shockfitting code, have been compared in Fig. 4.10, where pressure iso-contour lines are displayed for both the fist-order-accurate N and second-orderaccurate LDA schemes. Not surprisingly, the results are superimposed since the two codes implement the same FS discretization schemes. 54 Numerical results 0.4 y(m) 0.2 ___ N - COOLFluiD _ _ N - EulFS ___ LDA - COOLFluiD _ _ LDA - EulFS 0 -0.2 -0.4 0.8 1 1.2 1.4 1.6 1.8 2 x(m) Figure 4.10: Pressure contour plot for the shock fitting solutions: comparison between COOLFluiD and eulfs solutions. 4.2.2 Ideal gas, viscous flow over a circular cylinder The second configuration considered is an hypersonic flow over a 1 m radius circular cylinder. This problem has been proposed by NASA Langley aerothermodynamic team [35, 36, 38] in order to asses the capabilities of unstructured Finite Volume schemes to correctly predict the heat flux wall distribution for hypersonic flows on blunt bodies. Tab. 4.5 summarizes the freestream conditions used for this testcase. The grid used is a semi-unstructured mesh obtained by cutting into two triangles each quadrilateral of a structured grid. The triangulation is symmetric along the stagnation streamline and there is a slight mesh refinement around the shock region, as shown in Fig. 4.11. The background mesh is made of 6993 nodes and 13640 elements, while the shock-fitting grid has 7181 nodes and 13826 elements. Pressure and temperature contour plots are shown in Fig. 4.12 for both shock-capturing nad shock-fitting second order results. The B scheme has been used for the fitted solution while the LDA scheme has been used for the captured solution. As in the inviscid testcase shown before, shock-fitting provides a better accuracy in the shock-layer than the shock-capturing. This is particular evident if we look at the temperature Numerical results 55 Table 4.5: Freestream conditions for a viscous flow over a 1m radius cylinder. Physical quantity M1 p1 T1 Tw Radius Free-stream Value 17.5 57.65 Pa 200.0 K 500.0 K 1.0 m 2 0.4 1 0 y(m) y(m) 0.2 0 -0.2 -1 -0.4 1 1.2 1.4 1.6 x(m) -2 0 1 2 3 4 5 x(m) Figure 4.11: Grid used for the viscous flow over a 1 m radius circular cylinder. 56 Numerical results contour plot in the stagnation region (see Fig. 4.12(b)). The wall distribution of the pressure coefficient is displayed in Fig.4.13(a) while the skin friction coefficient is displayed in Fig. 4.13(b). In all plots reporting the distribution of the wall quantities, the abscissa is the azimuthal angle, which takes values zero at the stagnation point. The various curves shown include the S-C solution obtained with COOLFluiD on the semiunstructured mesh showed in Fig. 4.11, the S-F solution and the reference solution obtained by Peter Gno↵o using FV NASA code LAURA (FV) on a structured grid [38]. The comparison between the S-F and the LAURA results is satisfactory as far as the pressure coefficient is concerned, see Fig. 4.13(a). However, when looking at the skin friction coecient, Fig. 4.13(b), the di↵erencies between S-C and S-F computation are not significant. As it is possible see in Fig. 4.13(a), the comparison between the S-F results and the LAURA results is satisfactory for the pressure coefficient, however when looking at the skin friction coefficient 4.13(b) the di↵erencies between S-C and S-F computation are not significant. The heat flux distribution on the wall is displayed in Fig. 4.14; we can see that even though the S-F results does not shown the spike present in the S-C results, di↵erences of the order of 10 Temperature and pressure distributions along the stagnation streamline are shown in Fig. 4.15. It is possible to notice that the S-F results are in good agreement with the LAURA results in the shock-layer. The S-C results are not monotone and show a peak in the post-shock region for both pressure and temperature. In particular, the S-C pressure distribution over-predicts the reference one in all the shock layer region. Pressure and temperature distributions within the boundary layer are shown in Fig. 4.16. It is clear that post-shock conditions influence the solution in the boundary layer. Finally taking a look to the adimensional total enthalpy (see Fig. 4.17) it is possible to notice the capability of the S-F algorithm to preserve total enthalpy through the shock. Numerical results 57 p (Pa) 2 0.4 1 0.2 S-F S-C 0 S-C 0 S-F S-F -0.2 -1 -0.4 S-F -0.4 1 1.2 1.4 1.6 2 3 4 1 -2 x(m) 1 S-C 0 -0.2 -1 -2 0 0.4 0.2 S-C 0 y(m) y(m) 1 T (K) 12500 11500 10500 9500 8500 7500 6500 5500 4500 3500 2500 1500 500 y(m) 2 y(m) 23000 21500 20000 18500 17000 15500 14000 12500 11000 9500 8000 6500 5000 3500 2000 500 5 1.2 1.4 1.6 x(m) 0 1 2 3 4 x(m) x(m) (a) Pressure (b) Temperature 5 Figure 4.12: Pressure and temperature contour plot: comparison between the shockfitting and shock-capturing second-order-accurate accurate solutions. 2 0.006 COOLFluiD S-C COOLFluiD S-F LAURA 0.0055 0.005 1.5 0.0045 0.004 Cf Cp 0.0035 1 0.003 0.0025 0.002 0.5 0.0015 0.001 COOLFluiD S-C COOLFluiD S-F LAURA 0 -50 0 0.0005 50 0 -50 0 50 θ(°) θ(°) (a) Pressure coefficient (b) Skin friction coefficient Figure 4.13: Wall distributions. 58 Numerical results 80 70 qW (W/cm2) 60 50 40 30 20 COOLFluiD S-C COOLFluiD S-F LAURA 10 -50 0 50 θ(°) Figure 4.14: Heat flux distribution on the wall. 15000 22500 12500 20000 17500 10000 p (Pa) T (K) 15000 COOLFluiD S-C COOLFluiD S-F LAURA 7500 COOLFluiD S-C COOLFluiD S-F LAURA 12500 10000 5000 7500 5000 2500 2500 0 0.2 0.4 0 0 0.2 0.4 x (m) x (m) (a) Temperature (b) Pressure Figure 4.15: Distributions along the stagnation streamline. Numerical results 59 15000 25000 12500 24000 COOLFluiD S-C COOLFluiD S-F LAURA 23000 p (Pa) T (K) 10000 7500 22000 5000 COOLFluiD S-C COOLFluiD S-F LAURA 21000 2500 0 -5 10 10 -4 10 -3 10 20000 -5 10 -2 10 -4 10 -3 x (m) x (m) (a) Temperature (b) Pressure 10 -2 Figure 4.16: Distributions along the stagnation streamline (enlargement of the boundary layer). 1.1 H 2 0.9 0.2 S-C 0 S-F y(m) y(m) 1 1 0.4 S-C 0 S-F 0.8 COOLFluiD S-C COOLFluiD S-F 0.7 -0.2 -1 H/Href 1 0.99 0.98 0.97 0.96 0.95 0.94 0.93 0.92 0.91 0.9 0.89 0.88 0.87 0.86 0.85 0.6 -0.4 1 -2 1.2 1.4 1.6 x(m) 0 1 2 3 4 5 0.5 0 0.2 0.4 x(m) x (m) (a) Contour plot (b) Distribution along the stagnation line Figure 4.17: Adimensional total enthalpy. 60 Numerical results 4.2.3 Dissociated nitrogen, inviscid flow over a circular cylinder The nonequilibrium shock-fitting algorithm has been tested by reference to the flow conditions of Hornung’s experiment [41], listed in Tab. 4.6. The grid used is obtained by a 120⇥120 structured mesh cutting into two triangles each quadrilateral. The triangulation is symmetric in the stagnation region, as shown in Fig. 4.18. The background mesh is made of 14884 nodes and 29282 elements, while the shock-fitting grid at steadystate, which di↵ers from the background triangulation only in the neighbourhood of the fitted shock, has 15039 nodes and 29371 elements. 0.02 0.002 y(m) 0.004 y(m) 0.04 0 0 -0.002 -0.02 -0.004 -0.018 -0.016 -0.04 -0.014 x(m) -0.02 0 0.02 0.04 0.06 0.08 x(m) Figure 4.18: Grid used for the Nitrogen flow over a circular cylinder. Table 4.6: Freestream conditions for a nitrogen flow around a 1 inch radius cylinder. Physical quantity M1 p1 T1 Tv,1 ↵N,1 Radius Free-stream Value 6.13 2908.0 Pa 1833.0 K 1833.0 K 0.07 0.0254 m The numerical simulations were carried out for inviscid flow, since the Numerical results 61 purpose of this study is to asses the capability of the unstructured shockfitting technique to correctly predict the shock position on blunt bodies in the case of thermochemical nonequilibrium flows. Thermochemical properties have been provided by the library MUTATION [60], while the 2-temperature model of Park [61, 62] has been used. Since the PDEs governing dissociating nitrogen can be very sti↵, it has been necessary to use the implicit backward Euler scheme to obtain the steady-state solution. As shown in Fig. 4.19, the shock-wall distance and the interference fringe patterns agree reasonably well with the experimental results obtained by Hornung [41]. 1.2 1 COOLFluiD S-F (Nc) COOLFluiD S-F (LDAc) Hornung’s experiment ∆/R 0.8 0.6 0.4 0.2 0 0 20 40 60 80 θ(°) (a) (b) Figure 4.19: COOLFluiD + SF vs. Hornung’s experimental measurements: a) Nondimensional shock-wall distance and b) finite interference fringe patterns. ✓ is the azimuthal angle which takes value 0 at the stagnation point, /R is the shock-wall distance divided by the cilinder’s radius. In Fig. 4.20 the pressure contour plot computed using the first-orderaccurate Nc scheme is compared with the one obtained by Wang and Zhong [76] using a third-order-accurate finite di↵erence shock fitting scheme on structured meshes. The good agreement between the two solutions obtained by the two di↵erent schemes with di↵erent order of accuracy proves that both solutions are grid-independent, i.e. characterized by a very low numerical error. 62 Numerical results 0.02 y(m) p 140000 130000 120000 110000 100000 90000 80000 70000 60000 50000 40000 30000 20000 10000 0 -0.02 -0.02 0 0.02 0.04 x(m) (a) COOLFluiD + SF (b) Reproduced from [76] Figure 4.20: Pressure contour plot: qualitative comparison between the fitted solution obtained with COOLFluiDand the solution obtained by Wang and Zhong [76]. Hornung’s experimental measurements of the shock stando↵ distance are represented by the red circles. 4.3 4.3.1 EHD results Parallel electrodes In order to validate the EHD solver, a 2D channel with two parallel electrodes has been considered. The channel walls are 1 m apart and the electrodes are 0.5 m wide, so that the domain simulated has a length of 1 m in both the x and y-directions. In this test the electrical conductivity has been kept constant ( = 1 ⌦ 1 m 1 ) therefore Laplace’s equation (r2 = 0) has been solved in spite of Eq. (4.3). The electrodes have a specified potential: one volt on the upper electrode and zero on the lower; Neumann boundary conditions are applied along the remaining @ boundaries so that the normal component of the gradient is zero, @n = 0. Tab. 4.7 lists the boundary conditions used for the EHD solver. For this testcase, the electric potential is symmetric about the center of the electrode as seen in Fig. 4.21, where Fig. 4.21(a) has been obtained using a uniform grid made of 7931 nodes and 15606 elements and Fig. 4.21(b) is obtained by Gaitonde [32]. The comparison between the Numerical results 63 Table 4.7: Boundary conditions for the EHD solver. CFD boundary conditions Far field Subsonic inlet Supersonic inlet Subsonic outlet Supersonic outlet Wall (insulated) Wall (electrode) EHD boundary conditions Neumann Neumann Neumann Neumann Neumann Neumann Dirichlet Condition @ /@n = 0 @ /@n = 0 @ /@n = 0 @ /@n = 0 @ /@n = 0 @ /@n = 0 = specif ied two solution shows a good agreement. 0.85 1 0.95 0.75 0.85 0.75 0.8 0.75 0.95 0.85 0.75 0.65 0.55 0.45 0.35 0.25 0.15 0.05 0.65 0.65 0.6 0.55 Y 0.55 0.45 0.45 0.4 0.35 0.35 0.25 0.2 0.2 5 0.15 0.05 0 -0.4 -0.2 0 0.2 0.4 X (a) (b) Figure 4.21: Electric potential contours for the electrode channel with a constant electrical conductivity ( = 1 ⌦ 1 m 1 ): a) eulfs results and b) reprinted from [32]. A second validation test has been performed, simulating two parallel electrodes separated by a distance of one meter along the y-axis with a nonuniform electrical conductivity. As in the previous test, the elecrodes have a specified potential such that the top electrode plate is equal to one volt and the bottom is set to zero. Fig. 4.22(a) illustrates the domain with a nonuniform mesh used in the simulation, which is made of 22335 nodes and 44091 elements. Considering = it is possible to obtain an analytical solution for the Eq. (2.67) as it follows. r·( r )=0 (4.1) 64 Numerical results r = ✓ k 1 k2 , ◆ (4.2) @ @ k1 k2 dx + dy = dx + dy @x @y Z Z Z d = k1 dx + k2 dy d = 1 2 2 = (4.3) (4.4) = k 1 x + k2 y + k 3 (4.5) p 2 (k1 x + k2 y + k3 ) (4.6) where k1 , k2 and k3 are trhee constants. Considering the set of boundary conditions shown in Fig. 4.22(b) we obtain the following analytical solution p = y (4.7) 1 1 Dirichlet (φ = 1V) 0.8 0.8 0.6 Neumann (∂φ/∂n = 0) Y Y 0.6 0.4 0.4 0.2 0.2 0 0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 -0.6 Dirichlet (φ = 0V) -0.4 -0.2 0 0.2 X X (a) (b) 0.4 0.6 Figure 4.22: Parallel electrodes with a non constant electrical conductivity ( = ): a) mesh b) boundary conditions. A comparison between the analytical and the numerical electrical potential contour plots is shown in In Fig. 4.23(a). In Fig. 4.23(b), the Numerical results 65 theoretical and computed distributions of the electrical potential along the y-axis are shown for a constant ( = 1) and a variable ( = ) electrical conductivity model. Numerical 1 Analytical 1 Numerical (σ = φ ) Analytical (σ = φ ) Numerical (σ =1 ) Analytical (σ =1 ) φ 0.8 Y 0.6 0.4 0.2 0 -0.4 -0.2 0 0.2 0.8 0.6 Y (m) 0.95 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0.4 0.2 0 0.4 0 0.2 0.4 X φ (a) (b) 0.6 0.8 1 Figure 4.23: Parallel electrodes with a non constant electrical conductivity ( = ): a) electric potential contour plot (numerical vs. analytitcal solution) b) distribution of the electric potential along the y-axis (numerical vs. analytitcal solutions). 4.3.2 Ionized argon, inviscid flow in a nozzle with two opposite electrodes Preliminary results have been obtained for a subsonic nozzle with two parallel electrodes on the surface. Tab. 4.8 shows the reservoir test conditions while the electrodes configuration and the geometry of the nozzle are shown in Fig. 4.24(a). The mesh used is made of 26174 nodes and 51228 elements. Fig. 4.24(b) shows the contour plot of the nondimensional electrical potential. Table 4.8: Inlet flow conditions. Physical quantity Total pressure Total temperature Ionisation degree Value 1 ⇥ 105 Pa 5000 K 1 ⇥ 10 7 Fig. 4.25 shows the area weighted distributions along the nozzle axist of the temperature, for di↵erent values of the generator potential, Vg . It is 66 Numerical results 0.15 0.04 0.1 0.02 0 Y Y 0.05 1.92 1.82 1.72 1.62 1.52 1.42 1.32 1.22 1.12 1.02 0 -0.05 -0.02 -0.1 -0.04 -0.15 -0.3 -0.2 -0.1 0 -0.3 -0.28 -0.26 -0.24 -0.22 -0.2 X X (a) Nozzle geometry and electrodes configuration. (b) Contour plot of the nondimensional electrical potential. Figure 4.24: Flow in a converging nozzle with two parallel electrodes. possible to see the e↵ect on the flowfield of the ohmic heating due to the electric potential imposed on the elctrodes. 5200 5000 4800 T (K) 4600 4400 4200 4000 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 x (m) Figure 4.25: Temperature distributions along the nozzle axis for di↵erent values of the generator potential. Chapter 5 Conclusion and future work In this work, the investigation of a chemical non-equilibrium plasma has been carried out by using eulfs, an in-house developed, shock-capturing CFD code which uses compact-stencil, fluctuation splitting schemes on unstructured grids made of triangular and tetrahedral elements. In the code, the inviscid conservative equations have been succesfully coupled with a kinetic model for argon plasma where argon metastable has been considered as an independent species. In presence of shock waves, the CFD code can either capture the shock or be coupled with a newly developed, unstructured, shock-fitting algorithm which treats the discontinuities as moving boundaries that border regions of the flow-field were a smooth solution to the governing PDEs exists. When operating in shockfitting mode, the shock-capturing code is only used to solve the smooth regions of the flow field and it is left to the shock-fitting algorithm to enforce the Rankine-Hugoniot jump relations and preserve the species concentrations through the fitted shocks. The shock-fitting algorithm has been developed in a modular way then it has been possible also to couple it with the VKI’s COOLFluiD code. Thanks to the vaste choice of thermochemical models o↵ered by COOLFluiD, it has been possible to extend the proposed shock-fitting technique to thermochemical non-equilibrium flows. Results obtained in the computation of the flowfield over circular cylinders immersed in perfect gas, chemical and thermochemical nonequilibrium hypersonic flow, show that 67 68 Conclusion and future work shock-fitting not only allows a better modelling of the bow shock providing a good estimate of the position and the shape of the discontinuity, but also improves the solution quality within the entire shock-layer. Finally, the eulfs CFD solver has been successfully coupled with a Poisson solver; this will pave the way for simulations of non equilibrium plasma in which the electromagnetic interaction is also accounted for. Future work may include the addition of an equation to calculate the electron temperature, which is necessary to model discharges, overcoming a limit of the present model that equalizes the electron gas temperature with that of heavy species. Interesting future applications of this model could be the investigation of plasma-flow interaction in DBDs, a device proposed as actuator in high speed flow. Other possible applications can be the expansion of the LIBS plasma in di↵erent environment, such as high pressure gas or in liquid. In all these cases, because of the small size of the plasma an accurate characterization of the plasma require accurate determination of the shock wave shape. Conclusions and future work Publications related to the thesis The results presented in this dissertation have led to the following publications in international conference and journals: Journal articles • R. Pepe, A. Bonfiglioli, A. D’Angola, G. Colonna, R. Paciorri, Shock-Fitting Versus Shock-Capturing Modeling of Strong Shocks in Nonequilibrium Plasmas. IEEE Transactions on Plasma Science, Volume 42, Issue 8, Part 1, 2526-2527, 2014. • J. Garicano Mena, R. Pepe, A. Lani, H. Deconinck, Assessment of Heat Flux Prediction Capabilities of Residual Distribution Method: Application to Atmospheric Entry Problems, under review in Communications in Computational Physics. • R. Pepe, A. Bonfiglioli, A. D’Angola, G. Colonna, R. Paciorri, An unstructured shock-fitting solver for hypersonic plasma flows in chemical non-equilibrium, under review in Computer Physics Communications. Conference Papers • R. Pepe, A. Bonfiglioli, R. Paciorri, A. Lani, J. G. Mena, C. F. Olliver-Gooch, Towards a modular approach for unstructured shockfitting. 11th World Congress on Computational Mechanics, ECCOMAS 2014. 20-25 July 2014, Barcelona, Spain. • J. G. Mena, R. Pepe, A. Lani, H. Deconinck, Assessment of Residual Distribution Method Heat Flux Prediction Capabilities: Application to Atmospheric Entry Problems. 52nd Aerospace Sciences Meeting, AIAA. 13-17 January 2014, National Harbor, U.S.A. • R. Pepe, G. Colonna, A. Bonfiglioli, A. D’Angola, R. Paciorri, A selfconsistent unstructured solver for weakly ionized gases. XXXI 69 70 Conclusion and future work International conference on phenomena in ionized gases, 14-19 July 2013, Granada, Spain. (Poster) • R. Pepe, A. Bonfiglioli, A. D’Angola, G. Colonna, R. Paciorri, An unstructured solver for argon plasma flows with reduced state-tostate kinetics. 44th AIAA Plasmadynamics and Lasers Conference, AIAA. 24-27 June 2013, San Diego, U.S.A. • M. Onofri, R. Paciorri, R. Pepe, A. Bonfiglioli, A. D’Angola, G. Colonna, A new computational technique for re-entry flow calculations based upon a shock-fitting technique for unstructured grids. The 4th International ARA Days, The Atmospheric Reentry Association. 27-29 May 2013, Arcachon, France. Technical reports • R. Pepe, An unstructured shock-fitting solver for hypersonic flows. Von Karman Institue for Fluid Dynamics Project Report, Chausse de Waterloo, 72 B-1640, Rhode-Saint-Gense, Belgium. Appendix A Split Jacobian matrix In this Appendix the derivation of the split Jacobian matrices introduced in Chapter 3 and used to compute the generalised inflow parametershas been described. The Cartesian x, y and z components of flux vector Fc = Fci ei read: 0 B B Fc1 = B B @ ⇢i u ⇢uH ⇢u2 + p ⇢uv ⇢uw 1 C C C, C A 0 B B Fc2 = B B @ ⇢i v ⇢vH ⇢uv ⇢v 2 + p ⇢vw 1 C C C, C A 0 B B Fc3 = B B @ ⇢i w ⇢wH ⇢uw ⇢vw ⇢w2 + p 1 C C C (A.1) C A In Eq. (A.1) we have used u,v and w to denote the components of the fluid velocity along the Cartesian axes and ei is the unit vector of the Cartesian axis i. The Jacobian matrices matrices Ai = @Fci /@U, already introduced in Eq. (2.48), read: 0 B B A1 = B B @ ( ij ↵i )u 0 ↵i 0 0 u(⇧⇢j H) u(1 + ⇧⇢E ) H + u⇧⇢u u⇧⇢v u⇧⇢w ⇧⇢j u2 ⇧⇢E ⇧⇢u + 2u ⇧⇢v ⇧⇢w uv 0 v u 0 uw 0 w 0 u 71 1 C C C C A (A.2a) 72 0 ( ij ↵i )v 0 0 ↵i 0 B v(⇧⇢ H) v(1 + ⇧⇢E ) v⇧⇢u H + v⇧⇢v v⇧⇢w j B uv 0 v u 0 A2 = B B @ ⇧⇢j v 2 ⇧⇢E ⇧⇢u ⇧⇢v + 2v ⇧⇢w vw 0 0 w v 0 B B A3 = B B @ 1 C C C C A ( ij ↵i )w 0 0 0 ↵i w(⇧⇢j H) w(1 + ⇧⇢E ) w⇧⇢u w⇧⇢v H + w⇧⇢w uw 0 w 0 u vw 0 0 w v 2 ⇧⇢j w ⇧⇢E ⇧⇢u ⇧⇢v ⇧⇢w + 2w (A.2b) 1 C C C (A.2c) C A Partial derivatives ⇧⇢j , ⇧⇢E , etc. of the functional p = ⇧(⇢i , ⇢E, ⇢u) with respect to the conservative variables have been introduced in the matrices defined by Eq. (A.2). Observe that p = ⇧(⇢i , ⇢E, ⇢u) is not a thermodynamic relation [56] because pressure is a function of only Ns + 1 independent thermodynamic variables, for a gas in chemical non-equilibrium. If we choose the densities ⇢i of the chemical species and the energy per unit of volume ✏ = ⇢e as the set of independent thermodynamic variables [74] and denote with i , resp. , the partial derivatives of the thermodynamic pressure w.r.t. this set of independent variables: = i ✓ @p @⇢i ◆ , ",⇢j ,j6=i = ✓ @p @" ◆ (A.3) ⇢i ,i=1...Ns the following relations between the two sets of partial derivatives can be easily established: ⇧ ⇢i = @p = @⇢i i + u·u , 2 ⇧⇢E = @p = , @⇢E ⇧⇢u = @p = @⇢u u. (A.4) In the case of a mixture of chemical species having the same number of degrees of freedom, such as the argon plasma described in the Sect. 2.2.1, the thermodynamic pressure derivatives are constant and, since all species behave as a monoatomic atom, equal to: i 5 = hfi , 3 5 = . 3 (A.5) Split Jacobian matrix 73 Let us now compute the following matrix: 0 1 ( ij ↵i )un 0 ↵ i nt A Hnt + un ⇧⇢u t An = Ai ni = @ un (⇧⇢j H) (1 + ⇧⇢E )un ⇧⇢j n un u ⇧⇢E n ⇧⇢u n + un + un Id⇥d (A.6) where n = ni ei is an arbitrary unit vector, un = u · n and terms like un denote a dyadic tensor. Matrix An has the following eigenvalues: 0 B B B B ⇤n = B B B B @ un 1 ... un un un + a un a un C C C C C C C C A (A.7) where a is the sound speed, given by: a 2 = Ns X u · u) ↵i ⇧⇢i + ⇧⇢E (H i=1 = (1 + ⇧⇢E ) (A.8a) p ⇢ (A.8b) Each eigenvalue is associated with a right and a left eigenvector. Grouping the right eigenvectors by column and the left eigenvectors by row, the following matrices are obtained: 0 B u·u Rn = B @ ij u 0 ⇧⇢ j ⇧⇢E ⇢u · s ⇢ ⇢s h ⇢ ↵ai H ah +u·n i u ⇢ a +n i ⇢ h ⇢ ↵ai H ah ⇢ u a u·n i n i 0 1 C ⇢u · t C A ⇢t (A.9) 74 0 Ln = (Rn ) 1 B B B B B =B B B B @ ↵ i ⇧⇢ j a2 u·s ⇢ ij 1 ⇧⇢ j 2⇢ a 1 ⇧⇢ j 2⇢ a ⇧ ↵i a⇢E 2 0 u·n ⇧⇢E 2⇢a +u·n ⇧⇢E 2⇢a u·t ⇢ 0 1 2⇢ 1 ⇧ t ↵i a⇢E 2 u t s ⇢ ⇧⇢E t a u nt 1 t 2⇢ n + ⇧⇢E t a u tt ⇢ C C C C C C C C C A (A.10) where n, s e t are a triad of orthogonal unit vectors. Once the eigenvalues and the left and right eigenvectors are known, it is possible to factorise the matrix An as follows: A n = R n ⇤n L n (A.11) which, in turn, allow to compute the positive and negative parts of An : ± A± n = R n ⇤n L n (A.12) Lengthy algebra reveals that: 0 1 a± a± a± i,j i,N +1 i,N +2 ± ± A @ a± A± N +1,j aN +1,N +1 aN +1,N +2 n = ± ± a± N +2,j aN +2,N +1 AN +2,N +2 with i, j = 1, ..., Ns and the individual entries read: ✓ ± ◆ ± ± ± ⇧ + un ⇢ j + + ± ± ± ai,j = i,j 0 + ↵i 2 0 a 2 2 a ✓ ± ◆ ± ↵ ⇧ + i ⇢E + ± a± 0 i,N +1 = 2 a 2 ✓ ± ◆ ± ± ± t ↵ u + i + + ± ± ai,N +2 = ⇧E + nt 0 a a 2 2 ✓ ◆✓ ± ◆ ⇣ ⌘u ± ± ⇧ ⇢j H n + ++ ± 2 ± aN +1,j = un 0 + ⇧⇢j H 2 a 2 a 2 ✓ ± ◆ ± ± ± H un ++ + ± ± ± aN +1,N +1 = 0 + ⇧⇢E 2 + 0 a 2 2 a (A.13) (A.14a) (A.14b) (A.14c) ± (A.14d) (A.14e) Split Jacobian matrix a± N +1,N +2 = a± N +2,j = ✓ ✓ u n nt ⇧⇢E H t u a2 ⇧⇢j u a2 ◆✓ un n ✓ ◆✓ ± + + 2 ± + + 2 ± + ± 2 ✓ ◆✓ ± ⇧⇢E ++ = nn uu 2 a 2 a± N +2,N +1 A± N +2,N +2 75 ⇧⇢E = a ± + ± ◆ ◆ H u n ± nt ⇧E ut 0 + 2 a a (A.14f) ◆ ◆ ± ± ✓⇧ u ⇢ n j ± + + n u 0 2 a a (A.14g) ◆ ± ± ± u + + n (A.14h) 0 a 2 ◆ ✓ ◆ ± ± 1 ⇧⇢E + ± un nu + ± 0 + 0 Id⇥d a a 2 (A.14i) ± ±✓ ± + where: 0 A.1 = un , ± = 0 ± a. (A.15) Jacobian transformations In this section we report the Jacobian transformations between the conserved variables U and the parameter vector Z: 0 1 zi + ij z⇢ 0 0t @U @ @p @p A zH @z z⇢ @z@pH = @zu j @Z zu 0 z⇢ I 0 1 1 zi 0 0t ij z⇢ 2 z⇢2 @Z B ⇧⇢ j zu 1 zH 1 =B z⇢ 2 z⇢2 z⇢ (1 + ⇧⇢E ) z⇢2 ⇧⇢E @ @U 1 zu 1 0 2 z2 z⇢ I ⇢ Knowing the expression of p(Z): p(Z) = z⇢ 1 + ⇧⇢E Ns X j=1 ⇧⇢j zj + ⇧⇢E zH + ⇧⇢u · zu we can write the pressure deivatives z⇢ = P Ns s=1 zs and (A.16) 1 C C A (A.17) ! @p @p @zj , @zH (A.18) , @p @zu : 76 @p 1 = ⇧⇢s z⇢ + @zs 1 + ⇧⇢E Ns X j=1 ⇧⇢j zj + ⇧⇢E zH + ⇧⇢u · zu ! (A.19) @p ⇧⇢E = z⇢ @zH 1 + ⇧⇢E (A.20) @p ⇧⇢u = z⇢ @zu 1 + ⇧⇢E (A.21) Here we report Jacobian transformations between the flux vectors F, G, H and the parameter vector Z 0 zu ij B 0 @F B @p =B @zj B @Z @ 0 0 0 zv ij B 0 @G B =B 0 @p @Z B @ @z j 0 0 B @H B =B @Z B @ 0 zu 0 0 zi 0 zH 0 @p @p 2zu + @zu @zv zv zu zw 0 0 zv 0 0 0 zv @p @zH @p @zH @p @zu 0 0 zw ij 0 0 zw 0 0 0 0 @p @zj @p @zH zi zH zu @p 2zv + @z v zw 0 0 0 0 zw 0 0 zw @p @zu @p @zv 1 0 0 C @p C C @zw C 0 A (A.22) 1 0 0 C C 0 C @p C A @z (A.23) zu w zv zi zH zu zv @p 2zw + @z w 1 C C C C A (A.24) Split Jacobian matrix A.2 77 Transformations between U and Z The transformations between the conservative variables U = (U⇢i , U⇢E , U⇢u ) and the parameter vector Z = (Z⇢i , ZH , Zu ) for a mixture of monoatomic and thermally perfect gas, are reported in this section. A.2.1 U!Z ZH = p U⇢ U⇢ Z ⇢i = q P i Ns U⇢ =p i U⇢ i=1 U⇢i U⇢E ✓ ◆ Ns 1 U⇢u · U⇢u X f + U⇢i hi 2U⇢ i=1 U⇢u Zu = p U⇢ A.2.2 (A.25) (A.26) (A.27) Z!U U⇢i = Z⇢i Ns X j=1 Z ⇢j = Z ⇢i Z ⇢ (A.28) 78 U⇢E = Z⇢ U⇢E = Z⇢ ZH ✓ ZH + ( Z⇢ 1 + ⇧⇢E N s Zu · Zu X 1) + Z⇢i hfi 2Z⇢ i=1 " Ns X j=1 ◆ ⇧⇢j Z⇢j + ⇧⇢E ZH + ⇧⇢u · Zu U⇢u = Zu Z⇢ (A.29) # (A.30) (A.31) Appendix B Nondimensionalization In this Appendix details on the nondimensional form of governing equations are provided. Two di↵erent kinds of nondimensionalization are considered for internal and external flows. B.1 TCneq, external flows The following choice of reference variables are considered: Freestream density Freestream velocity magnitude Freestream temperature Freestream viscosity Reference length ⇢1 , q1 , T1 , µ1 , L. (B.1) Denoting the non-dimensional variables by an asterisk, one obtains: x⇤ = x/L 2 p⇤ = p/ ⇢1 q1 2 e⇤ = e/q1 where: ⇤ ⇢ = Ns X s=1 t⇤ = t/ (L/q1 ) u⇤ = u/q1 ⇢⇤s = ⇢s /⇢1 T ⇤ = T /T1 2 e⇤v = ev /q1 ⇢⇤s Ns X ⇢ ⇢s = = ⇢ ⇢1 s=1 1 79 (B.2) (B.3) 80 Nondimensionalization Replacing definitions B.2 in the in the conservation equations, one obtains: ⇢1 q1 @⇢⇤s + r⇤ · (⇢⇤s u⇤ ) = Ss ⇤ L @t 2 ⇢1 q 1 @(⇢⇤ u⇤ ) + r⇤ · (⇢⇤ u⇤ u⇤ ) + rp⇤ = 0 ⇤ L @t 3 ⇢1 q 1 @(⇢⇤ E ⇤ ) + r⇤ · (⇢⇤ u⇤ H ⇤ ) = 0 ⇤ L @t 3 ⇢1 q 1 @(⇢⇤ e⇤v ) + r⇤ · (⇢⇤ u⇤ e⇤v ) = Sev ⇤ L @t (B.4) (B.5) (B.6) (B.7) Dividing Equations. (B.4) to (B.7) by the factor appearing in parenthesis on their left hands side, their non-dimensional form is obtained: @⇢⇤s L ⇤ ⇤ ⇤ + r · (⇢ u ) = S s s @t⇤ ⇢1 q 1 (B.8) @(⇢⇤ u⇤ ) + r⇤ · (⇢⇤ u⇤ u⇤ ) + rp⇤ = 0 ⇤ @t (B.9) @(⇢⇤ E ⇤ ) + r⇤ · (⇢⇤ u⇤ H ⇤ ) = 0 ⇤ @t (B.10) @(⇢⇤ e⇤v ) L ⇤ ⇤ ⇤ ⇤ + r · (⇢ u e ) = S e v v 3 @t⇤ ⇢1 q 1 (B.11) The non-dimensional equation of state reads: N N s s X T1 X ⇤ ⇤ p = 2 ⇢s R s T = ⇢⇤s Rs⇤ T ⇤ q1 s=1 s=1 ⇤ (B.12) Nondimensionalization 81 where Rs⇤ = B.2 T1 Rs 2 q1 (B.13) TCneq, internal flows The following reference variables are considered: Inlet total pressure Inlet total temperature Inlet total density Inlet velocity magnitude Inlet total viscosity Reference length p01 , T10 , 0 ⇢ p1 , R1 T10 , µ01 , C. (B.14) which gives rise to the following set of non-dimensional variables: ⇣p ⌘ p ⇤ ⇤ 0 x = x/C t = t R1 T1 /C u⇤ = u/ R1 T10 (B.15) p⇤ = p/p01 ⇢⇤s = ⇢s /⇢01 T ⇤ = T /T10 e⇤ = e/R1 T10 e⇤v = ev /R1 T10 The adimensional conservation equations of the single chemical species are @⇢⇤s C ⇤ ⇤ ⇤ p + r · (⇢ u ) = S (B.16) s s @t⇤ ⇢01 R1 T10 and the adimensional vibrational energy is: @(⇢⇤ e⇤v ) C ⇤ ⇤ ⇤ ⇤ + r · (⇢ u e ) = S e v v 3/2 @t⇤ ⇢0 (R1 T 0 ) 1 (B.17) 1 The non-dimensional equation of state reads: N N N s s s X ⇢01 T10 X 1 X ⇤ ⇤ ⇤ ⇤ ⇤ p = 0 ⇢ Ri T = ⇢ Rs T = ⇢⇤s Rs⇤ T ⇤ p1 s=1 s R1 s=1 s s=1 where Rs⇤ ⇢01 T10 = 0 Rs p1 (B.18) (B.19) 82 Nondimensionalization B.3 EHD, external flows The following choice of reference variables has been considered for EHD external flows: Freestream density Freestream velocity magnitude Freestream temperature Reference length Reference electric potential ⇢1 , q1 , T1 , L, (B.20) ref . Denoting the non-dimensional variables by an asterisk, we obtain: x⇤ = x/L 2 p⇤ = p/ ⇢1 q1 2 e⇤ = e/q1 t⇤ = t/ (L/q1 ) u⇤ = u/q1 ⇢⇤s = ⇢s /⇢1 T ⇤ = T /T1 ⇤ = / ref E⇤ = EL/ ref (B.21) ⇢1 q1 @⇢⇤s + r⇤ · (⇢⇤s u⇤ ) = Ss ⇤ L @t (B.22) 2 ⇢1 q 1 @(⇢⇤ u⇤ ) + r⇤ · (⇢⇤ u⇤ u⇤ ) + rp⇤ = 0 ⇤ L @t (B.23) 2 3 ⇢1 q 1 @(⇢⇤ E ⇤ ) ⇤ ⇤ ⇤ ⇤ + r · (⇢ u H ) = ref E⇤ · E⇤ ⇤ 2 L @t L (B.24) Conservative equations in non-dimensional form are given by the following epressions: @⇢⇤s L + r⇤ · (⇢⇤s u⇤ ) = Ss ⇤ @t ⇢1 q 1 (B.25) @(⇢⇤ u⇤ ) + r⇤ · (⇢⇤ u⇤ u⇤ ) + rp⇤ = 0 ⇤ @t (B.26) 2 @(⇢⇤ E ⇤ ) ref E⇤ · E⇤ + r⇤ · (⇢⇤ u⇤ H ⇤ ) = ⇤ 3 @t L⇢1 q1 (B.27) Nondimensionalization 83 In Eq. (B.27), it possible to identify a nondimensional expression for the electrical conductivity: 2 ⇤ = ref 3 L⇢1 q1 (B.28) Subsituting Eq. (B.28) in Eq. (B.27), one obtains: @(⇢⇤ E ⇤ ) + r⇤ · (⇢⇤ u⇤ H ⇤ ) = ⇤ @t ⇤ E⇤ · E⇤ (B.29) Considering Eq. (B.28), the Poisson’s equation becomes: 3 ⇢1 q 1 r⇤ · [ ⇤ r⇤ ⇤ ] = 0 ref L (B.30) r⇤ · [ ⇤ r⇤ ⇤ ] = 0 (B.31) then: B.4 EHD, internal flows The following set of reference variables is introduced: Inlet total pressure Inlet total temperature Inlet total density Inlet velocity magnitude Reference length Reference electric potential p01 , T10 , 0 ⇢ p1 , R1 T10 , C, ref . (B.32) which gives rise to the following set of non-dimensional variables: ⇤ x = x/C p⇤ = p/p01 e⇤ = e/R1 T10 ⇣p ⌘ p 0 t =t R1 T1 /C u⇤ = u/ R1 T10 ⇢⇤i = ⇢s /⇢01 T ⇤ = T /T10 ⇤ = / ref E⇤ = ECref / ref ⇤ (B.33) 84 Nondimensionalization Then the conservation equations of the single chemical species are @⇢⇤s C ⇤ ⇤ ⇤ p + r · (⇢ u ) = S s s @t⇤ ⇢01 R1 T10 (B.34) and the conservation of the energy equation is: 2 @(⇢⇤ E ⇤ ) ref + r⇤ · (⇢⇤ u⇤ H ⇤ ) = E⇤ · E⇤ 0 0 3/2 @t⇤ C⇢1 (R1 T1 ) (B.35) Introducing the non-dimensional electrical conductivity, one obtains: 2 ⇤ = ref C⇢01 (R1 T10 )3/2 (B.36) @(⇢⇤ E ⇤ ) + r⇤ · (⇢⇤ u⇤ H ⇤ ) = ⇤ E⇤ · E⇤ (B.37) ⇤ @t Similarly to external flows the non-dimensional Laplace’s equation reads: ⇢01 (R1 T10 )3/2 ⇤ r · [ ⇤ r⇤ ⇤ ] = 0 ref C (B.38) r⇤ · [ ⇤ r⇤ ⇤ ] = 0 (B.39) then: B.5 Nondimensional parameter vector The adimensional parameter vector Z⇤ is given by the following expressions: • Cneq • Tcneq Z⇤ = Z⇤ = where ↵s⇤ p p ⇢⇤ (↵s⇤ , u⇤ , H ⇤ )t (B.40) ⇢⇤ (↵s⇤ , u⇤ , H ⇤ , e⇤v )t (B.41) ⇢⇤s ⇢s /⇢ref ⇢s = ⇤= = = ↵s ⇢ ⇢/⇢ref ⇢ (B.42) Nondimensionalization B.6 85 Nondimensional pressure derivatives In this section, a procedure to determine the pressure derivatives in nondimensional form is given. P Ns ⇤ ⇤ ⇢s R s ⇤ = PNs=1 (B.43) s ⇤C ⇤ ⇢ s=1 s Vs ⇤ s = Rs⇤ T ⇤ ⇤ e⇤s (B.44) 2 First the pressure derivative k ⇤ is expressed. Defining Rsref = q1 /T1 for 0 0 0 external flows and Rsref = p1 /⇢1 T1 = R1 for external flow, one obtains: P Ns ⇤ Rs s=1 ↵s Rsref ⇤ = P Ns ⇤ C Vs s=1 ↵s Rs ref P Ns = PNs=1 s ↵ s Rs s=1 ↵s CVs = (B.45) In the Eq. [B.45] the relation ↵i⇤ = ↵i is considered. The following non-dimensional form is used for external flows: ⇤ s = Rs T1 T 2 T q1 1 es s = 2 2 q1 q1 (B.46) es s = 0 R1 T 1 R1 T10 (B.47) and for internal flows: ⇤ s = Rs T R1 T10 86 Appendix C Rate coefficients fits C.1 Rate coefficients fit In this Appendix numerical fits used to approximate the rate coefficients for the electron-atom processes are reported. The following fitting functions have been used: f 2 = m1 + m2 f 2 = m1 + m2 1000 Te ! m3 1000 Te + m4 exp !m3 ✓ + m4 exp Te m5 ◆ ✓ Te m5 + m6 exp ◆ " ✓ (C.1) Te m7 ◆2 # (C.2) The coefficients mi , reported in the following Tabs. C.1-C.5, have been obtained using the least square method. In the following tables the coefficients with the associated standard errors are reported for the equilibrium constant and the forward rate coefficients. In table C.1 the coefficients mi for the equilibrium constant are reported. In Tables C.2, C.3 and C.4 the coefficients used for the fit of kf of e A processes are reported. In Tab. C.5 the coefficients of Eq. (2.51) for the computation of the forward rate coefficients of A A processes are reported. In Eq. (2.51) T is in K while the products kB T are in eV. 87 88 Table C.1: Coefficients of function f1 = log KeqI (cm 3 ) for the Ar ionization Coefficient m1 m2 m3 m4 m5 Value 52.222 -188.2 0.99254 16.486 319.23 Error 0.0.035015 0.099176 0.00015501 0.49331 8.5891 Table C.2: Coefficients of function f2 = log kf1 (cm 3 ) for the reaction 1 (e ionization from Ar0 ) Coefficient m1 m2 m3 m4 m5 m6 m7 Value -18.8 -176.7 1.018 -9.464 1545.0 -2.242 9690.0 Coefficient standard error 0.0225 0.2488 0.00079 0.61466 58.61 0.27119 114.67 Table C.3: Coefficients of function f2 = log kf2 (cm 3 ) for the reaction 2 (e excitation) Coefficient m1 m2 m3 m4 m5 m6 m7 Value -18.98 -133.6 1.003 -2.114 8409.0 -0.6737 2047.0 A Coefficient standard error 0.01023 0.12937 0.00056 0.07367 96.81 0.07929 71.71 A Rate coefficients fits 89 Table C.4: Coefficients of function f2 = log kf3 (cm 3 ) for the reaction 1 (e ionization from Ar⇤ ) Coefficient m1 m2 m3 m4 m5 m6 m7 Value -15.39 -49.61 0.994 -1.02 6865.0 0.2212 9272.0 A Coefficient standard error 0.00115 0.01293 0.000159 0.02424 45.19 0.00979 105.53 Table C.5: Rate coefficients kf (cm3 /s) for A-A processes Reaction 4 5 6 ✓ cm3 br s K1/2 eV 3.7755755 ⇥ 10 1.1436529 ⇥ 10 3.3010664 ⇥ 10 ◆ 17 17 16 "ij (eV ) 15.76 11.55 4.21 0.3 0.2 Relative error (%) 0.1 0 -0.1 -0.2 -0.3 kf1 kf2 kf3 -0.4 -0.5 10000 20000 30000 T(K) Figure C.1: Relative error associated to the e translationa l temperature 40000 50000 A forward rate coefficient vs. the 90 References [1] R. Abgrall. Residual distribution schemes: Current status and future trends. Computers & Fluids, 35(7):641 – 669, 2006. Special Issue Dedicated to Professor Stanley G. Rubin on the Occasion of his 65th Birthday. [2] J. D. Anderson. Hypersonic and High-Temperature Gas Dynamics. AIAA Education Series. AIAA, 2 edition, 2006. [3] W. K. Anderson and D. L. Bonhaus. An implicit upwind algorithm for computing turbulent flows on unstructured grids. Computers Fluids, 23(1):1 – 21, 1994. [4] W. K. Anderson, R. D. Rausch, and D. L. Bonhaus. Implicit/multigrid algorithms for incompressible turbulent flows on unstructured grids. Journal of Computational Physics, 128(2):391 – 408, 1996. [5] J. Bacri and A. M. Gomes. Influence of atom-atom collisions on thermal equilibrium in argon arc discharges at atmospheric pressure. Journal of Physics D: Applied Physics, 11(16):2185, 1978. [6] S. Balay, J. Brown, K. Buschelman, W. D. Gropp, D. Kaushik, M. G. Knepley, L. Curfman McInnes, B. F. Smith, and H. Zhang. PETSc Web page, 2012. http://www.mcs.anl.gov/petsc. [7] G. A. Bird. Molecular Gas Dynamics and the Direct Simulation of Gas Flows. Number v. 1 in Molecular Gas Dynamics and the Direct Simulation of Gas Flows. Clarendon Press, 1994. [8] N. J. Bisek. Numerical Study of plasma-assisted aerodynamic control for hypersonic vehicles. PhD thesis, University of Michigan, 2010. 91 92 References [9] N. J. Bisek, I. D. Boyd, and J. Poggie. Numerical study of magnetoaerodynamic flow around a hemisphere. Journal of Spacecraft and Rockets, 47(5):816 – 827, 2010. [10] A. Bonfiglioli. Fluctuation splitting schemes for the compressible and incompressible euler and navier-stokes equations. International Journal of Computational Fluid Dynamics, 14(1):21–39, 2000. [11] A. Bonfiglioli, S. Campobasso, B. Carpentieri, and M. Bollhöfer. A parallel 3d unstructured implicit rans solver for compressible and incompressible cfd simulations. In Roman Wyrzykowski, Jack Dongarra, Konrad Karczewski, and Jerzy Wasniewski, editors, Parallel Processing and Applied Mathematics, volume 7204 of Lecture Notes in Computer Science, pages 313–322. Springer Berlin / Heidelberg, 2012. [12] A. Bonfiglioli, M. Grottadaurea, R. Paciorri, and F. Sabetta. An unstructured, three-dimensional, shock-fitting solver for hypersonic flows. Computers & Fluids, 73(0):162–174, 2013. [13] A. Bonfiglioli and R. Paciorri. A mass-matrix formulation of unsteady fluctuation splitting schemes consistent with roes parameter vector. International Journal of Computational Fluid Dynamics, 27(4-5):210–227, 2013. [14] M. Capitelli, G. Colonna, and A. D’Angola. Fundamental Aspects of Plasma Chemical Physics, volume XVII of Springer Series on Atomic, Optical, and Plasma Physics. Springer, 2012. [15] M. Capitelli, G. Colonna, D. Giordano, L. Marra↵a, A. Casavola, P. Minelli, D. Pagano, L. D. Pietanza, F. Taccogna, and B. Warmbein. Tables of internal partition functions and thermodynamic properties of high-temperature mars-atmosphere species from 50k to 50000k. ESA Scientific Technical Review, 246, October 2005. [16] G. Capriati, G. Colonna, C. Gorse, and M. Capitelli. A parametric study of electron energy distribution functions and rate and transport coefficients in nonequilibrium helium plasmas. Plasma Chemistry and Plasma Processing, 12(3):237–260, 1992. References [17] J. C. Carette, H. Deconinck, H. Paillere, and P. L. Roe. Multidimensional upwinding: Its relation to finite elements. International Journal for Numerical Methods in Fluids, 20(8-9):935–955, 1995. [18] Mark H. Carpenter and Jay H. Casper. Accuracy of Shock Capturing in Two Spatial Dimensions. AIAA Journal, 37(9):1072–1079, September 1999. [19] G. Colonna and M. Capitelli. Electron and vibrational kinetics in the boundary layer of hypersonic flow. Journal of Thermophysics and Heat Transfer, 10(3):406–412, July 1996. [20] G. Colonna and M. Capitelli. The influence of atomic and molecular metastable states in high-enthalpy nozzle expansion nitrogen flows. Journal of Physics D: Applied Physics, 34(12):1812, 2001. [21] G. Colonna and M. Capitelli. Self-consistent model of chemical, vibrational, electron kinetics in nozzle expansion. Journal of Thermophysics and Heat Transfer, 15(3):308–316, July 2001. [22] G. Colonna and M. Capitelli. The e↵ects of electric and magnetic fields on high enthalpy plasma flows. In Fluid Dynamics and Colocated Conferences. American Institute of Aeronautics and Astronautics, June 2003. AIAA 2003-4036. [23] G. Colonna and M. Capitelli. Boltzmann and master equations for magnetohydrodynamics in weakly ionized gases. Journal of Thermophysics and Heat Transfer, 22(3):414–423, July 2008. [24] G. Colonna, M. Tuttafesta, and D. Giordano. Numerical methods to solve euler equations in one-dimensional steady nozzle flow. Computer Physics Communications, 138(3):213 – 221, 2001. [25] A. Csik, M. Ricchiuto, and H. Deconinck. A conservative formulation of the multidimensional upwind residual distribution schemes for general nonlinear conservation laws. Journal of Computational Physics, 179(1):286 – 312, 2002. [26] D. D’Ambrosio, G. Colonna, and M. Capitelli. Numerical prediction of non-equilibrium flows in hypersonic nozzles: State-to-state kinetics versus macroscopic models. In Fluid Dynamics and Co-located 93 94 References Conferences. American Institute of Aeronautics and Astronautics, June 2003. AIAA 2003-3549. [27] A. D’Angola, G. Coppa, M. Capitelli, C. Gorse, and G. Colonna. An efficient energy-conserving numerical model for the electron energy distribution function in the presence of electronelectron collisions. Computer Physics Communications, 181(7):1204 – 1211, 2010. [28] H. Deconinck, H. Paillère, R. Struijs, and P.L. Roe. Multidimensional upwind schemes based on fluctuation-splitting for systems of conservation laws. Computational Mechanics, 11(5/6):323–340, 1993. [29] H. Deconinck, P.L. Roe, and R. Struijs. A Multi-dimensional Generalization of Roe’s Flux Di↵erence Splitter for the Euler Equations. Computers and Fluids, 22(2/3):215–222, 1993. [30] G. Degrez and E. van der Weide. Upwind residual distribution schemes for chemical non-equilibrium flows. In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, November 1999. AIAA 99-3366. [31] P. J. Frey. Yams: a fully automatic adaptive isotropic surface remeshing procedure. Technical report, nstitut National de Recherche en Informatique et en Automatique (I.N.R.I.A.), 2001. http://www.ann.jussieu.fr/ frey/software.html. [32] D. V. Gaitonde. A high-order implicit procedure for the 3-d electric field in complex magnetogasdynamic simulations. Computers Fluids, 33(3):345 – 374, 2004. [33] Thomas Gerhold, O Friedrich, J Evans, and M Galle. Calculation of complex three-dimensional configurations employing the dlr-taucode. In Aerospace Sciences Meetings. American Institute of Aeronautics and Astronautics, June 1997. AIAA 1997-167. [34] D. Giordano. Hypersonic-flow governing equations with electromagnetic fields. In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, 2002. AIAA 20022165. References [35] P. Gno↵o. Computational Fluid Dynamic Technology for Hypersonic Applications. 2003. AIAA/ICAS International Air and Space Symposium and Exposition: The Next 100 Years; Dayton, OH; July 14-17. [36] P. Gno↵o. Simulation of stagnation region heating in hypersonic flow on tetrahedral grids (invited). In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, June 2007. AIAA 2007-3960. [37] P. Gno↵o, R. N. Gupta, J. L. Shinn, United States. National Aeronautics, and Space Administration. Office of Management. Conservation equations and physical models for hypersonic air flows in thermal and chemical nonequilibrium. NASA technical paper. National Aeronautics and Space Administration, Office of Management, Scientific and Technical Information Division, 1989. [38] P. Gno↵o and J. A. White. Computational Aerothermodynamic Simulation Issues on Unstructured Grids. 2004. AIAA-CP-2004-2371. [39] Peter Gno↵o. Semi-analytic reconstruction of flux in finite volume formulations. In Aerospace Sciences Meetings. American Institute of Aeronautics and Astronautics, January 2006. AIAA 2006-1090. [40] R. N. Gupta, K.-P. Lee, R. A. Thompson, J. M. Yos, and National aeronautics. A review of reaction rates and thermodynamic and transport properties for an 11-species air model for chemical and thermal nonequilibrium calculations to 30000 K 1MF. Number N90-27064/NASA-RP-1232. 1990. [41] H. G. Hornung. Non-equilibrium dissociating nitrogen flow over spheres and circular cylinder. Journal of Fluid Mechanics, 53(1):149– 176, 1972. [42] M. S. Ivanov, A. Bonfiglioli, R. Paciorri, and F. Sabetta. Computation of weak steady shock reflections by means of an unstructured shock-fitting solver. Shock Waves, 20(4):271–284, 2010. 95 96 References [43] M. G. Kapper and J. L. Cambier. Ionizing shocks in argon. part i: Collisional-radiative model and steady-state structure. Journal of Applied Physics, 109(11):113308–113308–14, 2011. [44] M. G. Kapper and J. L. Cambier. Ionizing shocks in argon. part ii: Transient and multi-dimensional e↵ects. Journal of Applied Physics, 109(11):113309–113309–11, 2011. [45] K. Kitamura, P. Roe, and F. Ismail. Evaluation of euler fluxes for hypersonic flow computations. AIAA Journal, 47(1):44–53, January 2009. [46] L. D. Landau and E. M. Lifshitz. Statistical Physics, Vol. 1, volume 24. Oxford: Pergamon, 1980. [47] A. Lani and H. Deconinck. Conservative residual distribution method for hypersonic flows in thermochemical nonequilibrium. In Aerospace Sciences Meetings. American Institute of Aeronautics and Astronautics, January 2009. AIAA 2009-460. [48] A. Lani, J. Garicano Mena, and H. Deconinck. A residual distribution method for symmetrized systems in thermochemical nonequilibrium. In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, June 2011. AIAA 20113546. [49] A. Lani, M. Panesi, and H. Deconinck. Conservative residual distribution method for viscous double cone flows in thermochemical nonequilibrium. Communications in Computational Physics, 13(2):479–501, 2013. [50] A. Lani, T. Quintino, D. Kimpe, H. Deconinck, S. Vandewalle, and S. Poedts. The coolfluid framework: Design solutions for high performance object oriented scientific computing software. In VaidyS. Sunderam, GeertDick van Albada, PeterM.A. Sloot, and JackJ. Dongarra, editors, Computational Science ICCS 2005, volume 3514 of Lecture Notes in Computer Science, pages 279–286. Springer Berlin Heidelberg, 2005. References [51] A. Lani, N. Villedie, K. Bensassi, L. Koloszar, M. Vymazal, S. M. Yalim, and M. Panesi. Coolfluid: an open computational platform for multi-physics simulation and research. In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, June 2013. AIAA 2013-2589. [52] Andreas Mack and Volker Hannemann. Validation of the unstructured dlr-tau-code for hypersonic flows. In Fluid Dynamics and Colocated Conferences. American Institute of Aeronautics and Astronautics, June 2002. AIAA 2002-3111. [53] S. Menne. Computation of non-winged vehicle aerodynamics in the low supersonic range. In J. J. Hunt, editor, Aerothermodynamics for space vehicles, volume 367 of ESA Special Publication, page 73, 1995. [54] S. Menne, C. Weiland, and M. Pfitzner. Computation of threedimensional hypersonic flows in chemical nonequilibrium. Journal of Aircraft, 31(3):623–630, May 1994. [55] G. Moretti and M. Abbett. A time-dependent computational method for blunt body flows. AIAA Journal, 4:2136–41, 1966. [56] L. Mottura, L. Vigevano, and M. Zaccanti. An evaluation of roe’s scheme generalizations for equilibrium real gas flows. Journal of Computational Physics, 138(2):354 – 399, 1997. [57] R. Paciorri and A. Bonfiglioli. A shock-fitting technique for 2d unstructured grids. Computers & Fluids, 38(3):715 – 726, 2009. [58] R. Paciorri and A. Bonfiglioli. Shock interaction computations on unstructured, two-dimensional grids using a shock-fitting technique. Journal of Computational Physics, 230(8):3155 – 3177, 2011. [59] Renato Paciorri, Filippo Sabetta, and Bernardo Favini. On the role of vibrational excitation in hypersonic flow computations. Meccanica, 33(4):331–347, 1998. [60] M. Panesi. Physical models for nonequilibrium plasma flow. PhD thesis, Universitá degli studi di Pisa, 2009. 97 98 References [61] C. Park. Nonequilibrium hypersonic aerothermodynamics. J. Wiley and sons, New York, Chichester, 1989. [62] C. Park, J. T. Howe, R. L. Ja↵e, and G. V. Candler. Review of chemical-kinetic problems of future nasa missions. ii - mars entries. Journal of Thermophysics and Heat Transfer, 8(1):9–23, 2001. [63] M. Pfitzner. A 3-d non-equilibrium shock-fitting algorithm using effective rankine-hugoniot relations. In Fluid Dynamics and Co-located Conferences, pages –. American Institute of Aeronautics and Astronautics, June 1991. AIAA 91-1467. [64] R. K. Prabhu, United States., and Langley Research Center. An implementation of a chemical and thermal nonequilibrium flow solver on unstructured meshes and application to blunt bodies [microform] / Ramadas K. Prabhu. National Aeronautics and Space Administration, Langley Research Center Hampton, Va, 1994. [65] A. Prakash, N. Parsons, X. Wang, and X. Zhong. High-order shockfitting methods for direct numerical simulation of hypersonic flow with chemical and thermal nonequilibrium. Journal of Computational Physics, 230(23):8474 – 8507, 2011. [66] L. C. Scalabrin. Numerical Simulation of Weakly Ionized Hypersonic Flow Over Reentry Capsules. PhD thesis, The University of Michigan, 2007. [67] L. C. Scalabrin and I. Boyd. Development of an unstructured navierstokes solver for hypersonic nonequilibrium aerothermodynamics. In Fluid Dynamics and Co-located Conferences. American Institute of Aeronautics and Astronautics, June 2005. AIAA 2005-5203. [68] J. R. Shewchuk. Triangle: Engineering a 2D Quality Mesh Generator and Delaunay Triangulator. In Ming C. Lin and Dinesh Manocha, editors, Applied Computational Geometry: Towards Geometric Engineering, volume 1148 of Lecture Notes in Computer Science, pages 203–222. Springer-Verlag, May 1996. From the First ACM Workshop on Applied Computational Geometry. References [69] E. Shima and K. Kitamura. Multidimensional numerical noise from captured shock wave and its cure. AIAA Journal, 51(4):992–998, February 2013. [70] H. Si. Tetgen a quality tetrahedral mesh generator and a 3d delaunay triangulator, 2011. http://tetgen.berlios.de/. [71] E. van der Weide, H. Deconinck, E. Issman, and G. Degrez. A parallel, implicit, multi-dimensional upwind, residual distribution method for the Navier-Stokes equations on unstructured grids. Computational Mechanics, 23:199–208, 1999. [72] E. van der Weide, H. Deconinck, E. Issman, and G. Degrez. A parallel, implicit, multi-dimensional upwind, residual distribution method for the navier-stokes equations on unstructured grids. Computational Mechanics, 23:199–208, 1999. 10.1007/s004660050401. [73] W. G. Vincenti and C. H. Krüger. Introduction to physical gas dynamics, volume 26. Wiley, 1965. [74] M. Vinokur and J. L. Montagné. Generalized flux-vector splitting and roe average for an equilibrium real gas. Journal of Computational Physics, 89(2):276 – 300, 1990. [75] J. Vlcek. A collisional-radiative model applicable to argon discharges over a wide range of conditions. i. formulation and basic data. Journal of Physics D: Applied Physics, 22(5):623, 1989. [76] X. Wang and X. Zhong. A High-Order Shock-Fitting NonEquilibrium Flow Solver for DNS of Strong Shock and Turbulence Interactions. In 7th International Conference on Computational Fluid Dynamics, 1996. ICCFD7-2305. [77] Y. Yamamoto and K. Karashima. Floating shock fitting for threedimensional inviscid supersonic flows. AIAA Journal, 20(1):9–17, January 1982. [78] Y. B. Zel’dovich and Y. P. Raizer. Physics of shock waves and hightemperature hydrodynamic phenomena. New York: Academic Press, 1966/1967, edited by Hayes, W.D.; Probstein, Ronald F., 1967. 99