Vipokezi katika minara ya kizazi kipya ya nishati ya jua vinaweza kufanya kazi na gesi zenye joto la juu na mtiririko wa turbulent kati ya nyuso ambazo joto lake hutofautiana kwa mamia ya kelvin. Katika hali hizi, joto, msongamano, kasi na turbulence huathiriana kwa nguvu. Ingawa inawezekana kutatua moja kwa moja mizani yote ya turbulence ya mtiririko, gharama ya kihesabu ya Direct Numerical Simulation ni kubwa sana kwa solar receivers zenye ukubwa halisi.

Kwa hiyo watafiti walitathmini Thermal Large Eddy Simulation, mbinu inayotatua moja kwa moja miundo mikubwa ya turbulence na kuwakilisha motions ndogo zaidi kwa mathematical models. Katika study, new sub-grid-scale models 12, zikiwemo five hybrid na seven functional, zililinganishwa na DNS reference data chini ya three different mesh resolutions na various numerical schemes.

Most balanced overall result ilipatikana kutoka two-layer hybrid model iliyoitwa M1 katika study. Model iliunganisha AMD na Gradient model katika momentum equation; na scalar AMD na Gradient model katika thermal closure. Kwa mean ya three meshes, normalized composite error score ya M1 ilikuwa asilimia 25,9. Same score ilikuwa asilimia 27,9 kwa previous best reference model MA na asilimia 27,4 kwa new M2 model iliyoshika nafasi ya pili.

Hata hivyo, hakuna model iliyotoa best result katika every quantity. M2 ilitoa lowest mean error score ya asilimia 14,8 katika second-order turbulence statistics, huku M1 ikibaki asilimia 17. Kwa Nusselt number, inayowakilisha wall heat transfer, solution isiyotumia any sub-grid model ilitoa best result kwa mean error ya asilimia 5,81. M1 iliimprove Nusselt number significantly relative na previous hybrid model; lakini old model ilibaki more accurate katika wall-friction coefficient.

Results zinaonyesha kwamba kuchagua turbulence model moja pekee haitoshi katika numerical design ya high-temperature solar receivers. Mesh resolution, numerical schemes zinazotumika katika momentum na mass transport, near-wall model coefficients na closures zinazowakilisha velocity-temperature interaction zinahitaji kutathminiwa together.

Kwa nini mtiririko katika solar receivers ni mgumu kumodeli?

Katika concentrated solar-power towers, solar radiation inayokusanywa kwa mirrors inaelekezwa kwenye central receiver. Katika next-generation receiver concept iliyochunguzwa katika study, pressurized air inazingatiwa kama heat-transfer fluid. Air inapopita kati ya high-temperature surface na opposite lower-temperature surface, significant changes katika density na fluid properties hutokea.

Turbulent flows zina broad range ya scales, kutoka large recirculating motions hadi very small eddies. Direct Numerical Simulation, au DNS, inajaribu kusolve all scales hizi kwenye computational grid. Approach hii inatoa detailed reference; lakini number ya cells na time steps zinazohitajika kwa all dimensions na operating duration ya real solar receiver zinaweza kuwa kubwa sana.

Large Eddy Simulation, au LES, inasolve directly large turbulence structures zinazobeba sehemu kubwa ya energy. Effects za eddies zilizo smaller kuliko computational grid kwenye momentum na heat transport zinaapproximateiwa kwa sub-grid-scale models. Thermal LES, au T-LES, inahitaji kutatua closure problem hii si kwa velocity field tu, bali pia kwa transport inayohusiana na temperature na density.

Main research question

Main objective ya study ni kuamua ni combination gani ya sub-grid-scale models inayotoa results zilizo closest na DNS data katika turbulent channel flows zenye high temperature difference. Watafiti walichunguza especially groups tatu za variables:

• Selection ya functional, structural na hybrid closure models,
• Resolution ya computational mesh,
• Second-, third- au fourth-order numerical schemes zinazotumika katika momentum na mass transport.

Study ina main aspects tatu zinazoiweka tofauti na previous research. Badala ya standard AMD closure katika energy equation, scalar AMD model inayofafanuliwa according to temperature gradients ilitestwa. Katika two-layer hybrid models, Gradient model ilitumiwa kama alternative kwa Bardina structural closure. Pia effects za second-order central, fourth-order central na third-order QUICK schemes kwenye model accuracy na stability zililinganishwa.

Representative channel geometry

Badala ya complex pipe na channel structures za real solar receivers, flat channel yenye periodicity katika directions mbili ilitumika ili kutenganisha fundamental physical interactions. Wakati flow inaenda katika x direction, z direction inaonyesha transverse direction ya channel na y direction inaonyesha height kati ya walls mbili.

Katika Kielelezo 1, wall moja imefafanuliwa kuwa 900 K na nyingine 1.300 K. Temperatures hizi zinaendana na roughly 626,85 °C na 1.026,85 °C. Mean friction Reynolds number ni roughly 800. Kwa sababu flow na transverse directions ni periodic, fluid ikitoka boundary moja inaingia tena kupitia opposite boundary. Hivyo fully developed channel flow inaweza kumodeliwa bila kuunda long physical channel.

Cells katika wall-normal direction hazijawekwa kwa equal spacing. Small cells zimetumika near walls ambako turbulence na temperature gradients hubadilika fastest, na larger cells katikati ya channel. Cell positions zimesambazwa kwa hyperbolic tangent relation:

\[ y_k=L_y\left[ 1+\frac{1}{a} \tanh\left( \frac{k-1}{N_y-1}\tanh^{-1}(a) \right) \right] \]

Hapa y_k inawakilisha wall-normal position ya relevant grid point, L_y channel height, a mesh expansion coefficient na N_y number ya points katika wall-normal direction.

DNS na LES meshes

DNS mesh ina resolution mara kadhaa higher katika flow na transverse directions hata kuliko finest LES mesh. Objective ya study si LES kutoa same detail kama DNS, bali kuamua ni kwa kiwango gani inaweza kureproduce basic velocity, temperature, wall heat transfer na turbulence statistics katika much coarser meshes.

Low-Mach-number equations

Flow ilimodeliwa kwa low-Mach-number approach iliyotengenezwa na Paolucci. Approach hii iko kati ya incompressible equations na fully compressible flow equations. Inazingatia large density changes zinazosababishwa na temperature; lakini haisolve sound waves directly.

Mass conservation imeandikwa hivi:

\[ \frac{\partial \bar{\rho}}{\partial t} + \frac{\partial \bar{\rho}\widetilde{U}_j}{\partial x_j} =0 \]

Momentum equation, pamoja na transport ya resolved velocities, inajumuisha sub-grid-scale velocity-velocity correlation:

\[ \frac{\partial \bar{\rho}\widetilde{U}_i}{\partial t} = -\frac{\partial \left( \bar{\rho}\widetilde{U}_j\widetilde{U}_i + \bar{\rho}\widehat{U_jU_i} \right)} {\partial x_j} -\frac{\partial P}{\partial x_i} + \frac{\partial\Sigma_{ij}}{\partial x_j} \]

Ideal-gas relation inatoa relationship kati ya filtered temperature, density na thermodynamic pressure:

\[ \widetilde{T}=\frac{P_0}{\bar{\rho}r} \]

P₀ inawakilisha thermodynamic pressure inayochukuliwa constant in space, r specific gas constant ya air, ρ density na T temperature.

Dynamic viscosity na thermal conductivity ya air zilihesabiwa kwa temperature-dependent Sutherland relations. Expression iliyotumika kwa dynamic viscosity ni:

\[ \mu(T)=\mu_0 \left(\frac{T}{T_0}\right)^{3/2} \frac{T_0+S}{T+S} \]

Reference values zimetolewa kama μ₀ = 1,716 × 10⁻⁵ Pa·s, T₀ = 273,15 K na S = 110,4 K. Specific gas constant ya air ilichukuliwa kuwa 287 J/kg·K, na heat-capacity ratio 1,4.

Two thermal operating conditions

Ili kuchunguza behavior ya sub-grid closures chini ya different thermal loads, volumetric heat source katika energy equation ilibadilishwa:

• Base condition: Hakuna heat sink na hot-wall heat flux imetolewa kuwa 98 kW/m².
• S1 condition: Uniformly distributed volumetric heat sink ya 55 MW/m³ ilitumika. Study inaeleza kwamba hot-wall flux iliongezeka roughly 2,5 times hadi value ya 263.

Ingawa text inaeleza second value kama boundary heat flux, unit yake imeandikwa kW/m³. Kwa sababu boundary heat flux kwa kawaida inatarajiwa kuwa power per area, expression hii inapaswa kutathminiwa kama unit inconsistency katika preprint; value katika content haijabadilishwa wala kusahihishwa.

Functional models

Functional models zinafananishisha effect ya unresolved small eddies na additional turbulent viscosity. Approach hii inawakilisha energy transfer kutoka large scales kwenda small scales, yaani kwa dissipative manner. Ingawa kwa kawaida ni numerically stable, inaweza kutoa excessive damping kwa kuondoa energy nyingi kuliko inavyopaswa kutoka real flow.

Katika base AMD model, turbulent eddy viscosity ina structure ifuatayo:

\[ \nu_e^{AMD} = C^{AMD} \frac{ \max\left(0,-G_{ij}S_{ij}\right) }{ g_{kl}^{\,2} } \]

Hapa S_ij inaonyesha strain-rate tensor, g_ij velocity gradient na G_ij gradient-based tensor inayojumuisha directional effect ya mesh spacing.

Study ilichunguza three AMD variants:

• AMD: Standard model inayofafanuliwa according to energy distribution ya velocity field,
• AMDs: Model inayofafanuliwa according to gradients kwa scalar au temperature transport,
• AMDc: Compressibility adaptation inayozingatia pia trace ya strain tensor.

Structural models

Structural models zinajaribu kuiga mathematical structure ya actual sub-grid term. Kwa sababu hazitumii energy dissipation pekee, zinaweza kuruhusu reverse energy transfer kutoka small scales kwenda large scales. Feature hii inaweza kuhifadhi anisotropic structure ya flow vizuri zaidi; lakini inaweza kuongeza risk ya numerical instability.

Study ilizingatia Bardina, scale-similarity na Gradient models. Momentum closure ya Gradient model ni:

\[ \tau_{ij}^{Grad} = \frac{C^{Grad}}{12} \Delta_k^2 g_{ik}g_{jk} \]

na thermal au scalar closure ni:

\[ \pi_j^{Grad} = \frac{C^{Grad}}{12} \Delta_k^2 g_{jk}d_k \]

Δ_k inawakilisha filter au mesh width katika relevant direction, na d_k scalar gradient inayohusiana na temperature au density.

Kwa nini hybrid models zilitumika?

Hybrid models zinalenga kuunganisha stability ya functional model na uwezo wa structural model kuhifadhi turbulence structure:

\[ \tau_{ij} = \alpha_1\tau_{ij}^{func} + \beta_1\tau_{ij}^{struct} \]

\[ \pi_j = \alpha_2\pi_j^{func} + \beta_2\pi_j^{struct} \]

Katika two-layer models, coefficient ya functional closure si constant across channel. Near wall, viscous dissipation inatawala, hivyo contribution ya functional model inaongezwa; karibu na center ya channel, relative effect ya structural model inaongezwa.

Wall-dependent coefficient ilibadilishwa kwa function ifuatayo:

\[ C^{func,dyn}(y) = C^{func} + \left[ \frac{1}{2} + \frac{1}{2} \tanh\left( \frac{y-s_c}{s_f} \right) \right] \left( C^{center}-C^{func} \right) \]

Kielelezo 2 kinaonyesha coefficient ikipita smoothly kutoka high value near wall kwenda lower value katikati ya channel. Transition region inalingana na roughly asilimia 4 ya total channel height.

Structure ya M1 model

M1 model, iliyotoa lowest mean total error score katika study, ilitumia combinations zifuatazo kwenye wall:

\[ \tau_{ij} = 0{,}6\,\tau_{ij}^{AMD} + 0{,}5\,\tau_{ij}^{Grad} \]

\[ \pi_j = 0{,}6\,\pi_j^{AMDs} + 0{,}4\,\pi_j^{Grad} \]

Functional-model coefficient inashuka hadi 0,15 katika channel center. Fourth-order central scheme ilitumika kwa momentum transport na second-order central scheme kwa mass transport. Kwa short notation ya study, arrangement hii imeonyeshwa kama C4-C2.

Previous MA reference model ilitumia Bardina model badala ya Gradient model, standard AMD badala ya scalar AMD katika thermal closure, na C2-C2 numerical arrangement.

Numerical solution

Simulations zilifanywa katika TRUST/TrioCFD environment iliyotengenezwa na French Alternative Energies and Atomic Energy Commission. Main features za code ni:

• Finite-difference method kwenye staggered computational mesh,
• Third-order Runge–Kutta integration in time,
• Second- au fourth-order central scheme kwa momentum transport,
• Second-order central au third-order QUICK scheme kwa mass transport,
• Second-order central scheme kwa velocity divergence na temperature diffusion.

A posteriori evaluation ina maana gani?

Katika a priori tests, closure term ya model inaapplyiwa kwenye precomputed DNS fields na kulinganishwa na actual sub-grid terms. Katika a posteriori test, closure model inawekwa ndani ya LES equations na simulation inaendeshwa from beginning to end na model hiyo.

Kwa hiyo study haitathmini tu kama models zinatengeneza mathematically correct closure term, bali pia jinsi zinavyobadilisha velocity, temperature, friction, heat transfer na turbulence statistics over time. Uwezo wa kutambua models zinazokuwa unstable pia ni important advantage ya a posteriori approach.

Ni quantities zipi zililinganishwa?

First-order statistics:

• Mean streamwise velocity,
• Mean wall-normal velocity,
• Mean temperature,
• Nusselt number,
• Wall-friction coefficient.

Second-order statistics:

• Velocity variances katika three directions,
• Reynolds shear stress,
• Streamwise turbulent temperature transport,
• Wall-normal turbulent temperature transport.

Temperature variance pia ilionyeshwa kwa diagnostic purpose; lakini kwa sababu hakuna separate sub-grid closure term kwa quantity hii, haikujumuishwa katika total error score.

Nusselt number na friction coefficient

Wall heat transfer ilitathminiwa kwa Nusselt number:

\[ Nu = \frac{ D_h \left\langle \lambda\frac{\partial T}{\partial y} \right\rangle_w }{ \lambda_w(T_w-T_b) } \]

D_h inaonyesha hydraulic diameter, λ thermal conductivity, T_w wall temperature na T_b bulk-mean temperature.

Wall-friction coefficient ilihesabiwa kwa:

\[ C_f = \frac{ 2\mu_w \left\langle \frac{\partial U_x}{\partial y} \right\rangle_w }{ \rho_bU_b^2 } \]

Quantities hizi mbili zinatoa separate information kuhusu kiasi cha heat ambacho fluid inaweza kupata katika solar receiver na kiasi cha hydrodynamic resistance kinachotokea wakati huo.

Error scores zilitengenezwaje?

Wall-normal profiles kwanza zilimatchiwa kati ya DNS mesh na LES mesh. Logarithmic wall-distance weighting ilitumika katika profile error ili near-wall regions zisipotee katika evaluation.

First- na second-order error sums zilinormalizeiwa relative na largest error value observed among all models. Kisha composite error ilihesabiwa kwa kuzingatia number ya variables katika category:

\[ \varepsilon_{tot}^{LES,i} = \frac{ n\varepsilon_{mean}^{LES,i} + m\varepsilon_{rms}^{LES,i} }{ n+m } \]

Hapa n ni number ya first-order statistics na m number ya second-order statistics. Kwa hiyo total values kama asilimia 25,9 zinazotolewa katika study si raw relative error ya single physical quantity. Ni normalized comparison scores zilizoundwa kwa kutumia different profiles na wall quantities.

Mpangilio wa jumla wa mifano

Grafu yenye sehemu tatu kwenye ukurasa wa 13 inaonyesha kwa pamoja alama za makosa ya daraja la kwanza, daraja la pili na makosa ya jumla ya mifano kwa meshes A, B na C. Mifano mipya ya hybrid iko upande wa kushoto wa grafu, functional models katikati, na reference models kutoka utafiti uliotangulia upande wa kulia.

Model ya M1 ilitoa alama ya kosa la jumla ya asilimia 18,8 kwenye mesh A, asilimia 21,6 kwenye mesh B na asilimia 37,2 kwenye mesh C. Mgawanyo huu unaonyesha kuwa model ina utendaji mzuri kwenye meshes fine na medium, lakini accuracy yake inapungua kwa kiasi kikubwa kwenye coarse mesh.

M2 iliongoza katika second-order statistics

Ingawa M1 ilishika nafasi ya kwanza katika mpangilio wa jumla, alama ya chini zaidi ya wastani katika second-order statistics, zinazojumuisha fluctuations za turbulence na heat-flux correlations, ilipatikana kwa model ya M2:

Matokeo haya yanaonyesha kwamba scalar AMD closure inaweza kuboresha utendaji wa jumla, lakini si lazima itoe kosa la chini zaidi kwa turbulence correlations zote.

Mifano ilitofautiana katika first-order quantities

Grafu ya nguzo kwenye ukurasa wa 14 inaonyesha tofauti za makosa katika mean velocity, wall-normal velocity, temperature, Nusselt number na friction coefficient. Inaonekana wazi kwamba model ileile inaweza kuwa bora katika quantity moja na dhaifu katika nyingine.

Model ya M1 ilileta uboreshaji wa wazi katika utabiri wa wall heat flux ikilinganishwa na MA. Kwa upande mwingine, model ya MA inayotegemea Bardina ilikuwa bora zaidi kwa friction coefficient. Kutoweza kuzalisha heat transfer na friction kwa accuracy sawa kwa closure ileile kunaonyesha kwamba engineering quantity inayolengwa lazima ielezwe wazi wakati wa kuchagua model.

Kwa nini kutotumia sub-grid model kulionekana vizuri katika baadhi ya matokeo?

Solution ya NA, ambayo haina closure model, ilitoa kosa dogo kuliko models zote katika Nusselt number. Katika Reynolds shear stress pia, solutions za NA na F1 zilifanya vizuri sana kwenye baadhi ya meshes.

Hali hii haimaanishi kwamba sub-grid model si muhimu. LES meshes zilizotumika zinaweza kuwa fine kiasi cha kusolve sehemu kubwa ya baadhi ya quantities tayari. Aidha, numerical diffusion ya central schemes au mesh resolution inaweza kwa bahati ikafidia athari ya closure iliyokosekana. Study inatafsiri hali hii kama error compensation.

Model kuwa na kosa dogo katika quantity moja pekee haimaanishi kwamba inawakilisha velocity, temperature na turbulence zote kwa usahihi. Solution ya NA ilitoa matokeo dhaifu zaidi katika temperature profile na temperature variance.

Mean velocity na temperature profiles

Ukurasa wa 15 unaonyesha mean streamwise velocity, wall-normal velocity na temperature profiles kwa hot-wall na cold-wall sides. Mistari nyekundu inawakilisha hot-wall DNS data, na mistari ya bluu cold-wall DNS data. Mistari yenye rangi nyepesi zaidi ya model ileile inaonyesha meshes zilizo coarse zaidi.

Kwa mean streamwise velocity, models nyingi zinakaribia DNS nje ya viscous sublayer. Solution ya NA bila sub-grid model ina mwelekeo wa kukadiria velocity chini, wakati F1 inayotumia functional closure pekee ina mwelekeo wa kuikadiria juu. Hybrid models M1 na MA zinazalisha umbo la jumla la profile kwa balance nzuri zaidi.

Mpangilio unaofanana unaonekana katika mean temperature. NA inakadiria temperature profile chini, na F1 juu. Profiles za M1 na MA zinabaki karibu na DNS. M1 inatoa agreement ya karibu zaidi kuliko MA upande wa cold wall, huku MA ikiwa na faida ndogo katika baadhi ya maeneo upande wa hot wall.

Reynolds stresses

Kurasa za 17 na 18 zinachunguza variances za velocity components tatu. DNS profiles zinatengeneza peak dhahiri karibu na y⁺ ≈ 15 katika wall units. Peak upande wa cold wall ni kubwa zaidi kwa sababu ya higher local friction Reynolds number.

Solutions za NA na F1 zilikadiria zaidi hasa streamwise velocity variance upande wa cold wall. Wakati DNS peak value ni takriban 5, baadhi ya solutions zilifikia karibu 8. Model ya MA ina mwelekeo wa kukadiria peaks chini. M1, hasa kwenye fine mesh, ilikamata peak amplitudes kwa balance nzuri zaidi; hata hivyo, ilisogeza peak position kidogo kuelekea outer region ya channel.

Grafu kwenye kurasa za 29-31 katika appendices zinatenganisha kwa undani closure contributions za models hizohizo kwenye fine mesh A na coarse mesh C. Kwenye coarse mesh, M1, NA na F1 zinaonekana kukadiria velocity variances juu kwa kiasi kikubwa, jambo linaloonyesha kuwa mesh resolution ina athari kubwa kwenye statistics hizi.

Athari ya Gradient model kwenye Reynolds shear stress

Grafu kwenye ukurasa wa 19 inaonyesha total Reynolds shear stress pamoja na contributions za functional na structural closures. Bardina closure katika MA ilitoa contribution kubwa kuliko ilivyopaswa kuondoa kutoka actual shear stress katika baadhi ya maeneo, hivyo kusogeza total profile mbali na DNS.

Direct contribution ya Gradient closure katika M1 kwenye shear stress ilibaki karibu na sifuri. Hivyo resolved LES field iliweza kuzalisha DNS profile kwa distortion ndogo zaidi. Mean error values za quantity hii zilikuwa:

• F1: asilimia 4,8,
• NA: asilimia 5,8,
• M1: asilimia 7,0,
• MA: asilimia 15,1

Ingawa F1 ilitoa lowest numerical error hapa, advantage ya M1 ni kutoa matokeo yenye balance nzuri zaidi inapozingatiwa pamoja na velocity na temperature statistics nyingine.

Turbulent heat transport

Kurasa za 20, 32 na 33 zinaonyesha streamwise na wall-normal turbulent temperature transport. Correlations hizi zinawakilisha heat inayosafirishwa kwa pamoja na temperature fluctuations pamoja na velocity fluctuations.

Katika streamwise U′T′ correlation, coarse mesh ilikadiria heat transport juu, hasa upande wa cold wall. M1 ilibaki karibu na DNS kwenye fine mesh, lakini deviation yake iliongezeka kwenye coarse mesh. MA ilikadiria peak values chini katika pande zote mbili.

Katika wall-normal V′T′ correlation, NA ndiyo solution iliyozalisha transition region vizuri zaidi. Mean error values zilikuwa asilimia 10,9 kwa NA, asilimia 17,6 kwa F1, asilimia 21,9 kwa M1 na asilimia 22,6 kwa MA. Matokeo haya yanaonyesha kwamba ingawa M1 ni ya kwanza katika overall ranking, si model yenye accuracy ya juu zaidi kwa kila turbulent heat-flux component.

Temperature variance ilitoa diagnostic muhimu

Hakuna direct sub-grid closure iliyotumika kwa temperature variance. Kwa hiyo T′² profiles kwenye kurasa za 20 na 34 ni diagnostic ya kuonyesha jinsi velocity na temperature closures zinavyofanya kazi kwa dynamic compatibility.

Solutions za NA na F1 zilikadiria temperature fluctuations juu katika near-wall na logarithmic regions, na tatizo hili liliongezeka kwenye coarse mesh. M1 ilikamata near-wall peak vizuri zaidi lakini ikabaki juu katika outer region. Ingawa MA ilikuwa more dissipative, ilitoa lowest value kati ya models nne zilizochaguliwa katika temperature variance kwa mean error ya asilimia 14,9. Mean error ya M1 kwa quantity hii ilikuwa asilimia 32.

Mesh resolution haikuwa na athari ya mwelekeo mmoja kila wakati

General trend ni kwamba fine mesh A huzalisha velocity na temperature profiles kwa usahihi zaidi kuliko coarse meshes. Hata hivyo, baadhi ya statistics zilionyesha kosa dogo zaidi kwenye coarse mesh. Watafiti wanaeleza hili kwa compensation kati ya model diffusion na numerical diffusion.

Kwa mfano, closure iliyo excessively dissipative inaweza kudhoofisha real turbulence kupita kiasi kwenye fine mesh, wakati numerical error ya coarse mesh inaweza kwa bahati kuleta profile karibu zaidi na DNS. Kwa hiyo comparison iliyofanywa kwenye mesh moja pekee inaweza kutathmini vibaya model quality.

QUICK scheme haikutoa faida iliyotarajiwa

Functional models zilizotumia QUICK kwa mass transport kwa ujumla zilitoa higher errors. QUICK scheme ilifanya kazi kama additional implicit filter katika problem hii na kuungana na dissipative effect ya sub-grid model.

Vivyo hivyo, kuongeza compressible AMD formulation kwenye momentum closure hakukuongeza accuracy katika functional models zilizochunguzwa; models nyingi za F2-F7 zilifanya vibaya zaidi kuliko F1 iliyotumia standard AMD.

Numerical instability

Baadhi ya hybrid models zikawa unstable wakati wa simulation. Hasa momentum closures zilizochanganya compressible AMD na Bardina hazikufanya kazi kwa stability kwenye finest A mesh. Kuongeza Bardina coefficient kutoka 0,4 hadi 0,6 kulisababisha additional instability kwenye A na B meshes.

Watafiti wanaona kwamba kwa sababu fine mesh huzalisha less numerical diffusion, reverse energy transfer ya structural model haikuweza kudamped vya kutosha. Matokeo yanaonyesha kwamba higher order au finer resolution si lazima moja kwa moja kumaanishe T-LES iliyo stable na accurate zaidi.

Maana kwa engineering

Model ya M1 ilitoa lowest overall error score among models zilizochunguzwa katika turbulent channel flow yenye high temperature difference inayowakilisha solar-receiver conditions. Hasa ilitoa kwa pamoja sifa zifuatazo:

• Balanced accuracy katika mean velocity na temperature profiles,
• Better Nusselt-number prediction kuliko previous MA model,
• More balanced reproduction ya velocity-variance peaks,
• Less distortion ya Reynolds shear stress kuliko Bardina model,
• Generally good performance kwenye fine na medium meshes.

Hata hivyo, M1 ni dhaifu kuliko MA katika friction coefficient, kuliko M2 katika combined second-order error score, na kuliko solution isiyo na model katika wall-normal turbulent heat flux. Kwa hiyo study inaonyesha M1 si kama universal closure, bali kama model inayotoa most balanced compromise kati ya statistics zilizotathminiwa.

Uwezekano wa kuhamisha mbinu kwa Uturuki

Uturuki ina mazingira ya kijiografia na engineering ya kuchunguza concentrated solar-energy systems, solar towers na high-temperature thermal-storage technologies katika maeneo yenye high direct solar irradiation. Method ya study inatoa model-selection principles zinazoweza kutumika kutathmini numerically heat transfer na pressure loss zinazoweza kutokea katika receiver channels za facilities kama hizi.

Hasa kwa research centers na engineering teams zinazotengeneza solar receivers, study inatoa lessons zifuatazo:

• Heat transfer na friction huenda zisihesabiwe kwa accuracy sawa na model ileile.
• Standard isothermal LES closures zinaweza kutotosha katika high temperature differences.
• Thermal scalar closure inapaswa kuchaguliwa independently of momentum closure.
• Model na numerical scheme zinapaswa kuvalidateiwa pamoja.
• Matokeo mazuri kwenye mesh resolution moja hayapaswi kuchukuliwa kama mesh-independent model success.

Hata hivyo, channel geometry katika research haijumuishi pipe bends, inlet region, solar-flux distribution, material conduction au full-system geometry ya real solar receiver. Kwa hiyo results haziwezi kutumika directly kama design parameters kwa facility ya Uturuki.

Matokeo ambayo study inaonyesha na isiyoonyesha

Study inaonyesha kwamba M1 hybrid model ilitoa lowest overall error score chini ya channel, temperature, Reynolds number, meshes na error metrics zilizochunguzwa. Haionyeshi kwamba model ndiyo most accurate closure katika solar receivers zote au turbulent heat-transfer problems zote.

Research pia hairipoti kwa numerical terms ni processor time au energy kiasi gani M1 inaokoa relative na DNS. Ingawa ni wazi LES meshes zina resolution ya chini sana kuliko DNS, specific cost advantage haiwezi kuhesabiwa kwa sababu computation time, core-hours au speedup factor hazijatolewa.

Hakuna physical solar-receiver experiment, wall-temperature measurement au laboratory velocity data iliyotumika. Validation ilifanywa kwa DNS data iliyozalishwa kwa same physical problem. Huu ni strong numerical comparison, lakini hauwezi kuchukua nafasi ya experimental validation ya model katika real-device conditions.

Mbinu na Matokeo ya Utafiti

Muundo wa utafiti

Study ni comparative computational-fluid-dynamics research ambapo 12 new sub-grid-scale T-LES models zilitestwa a posteriori katika turbulent channel flows zenye high temperature difference.

Model iliyofanikiwa zaidi kwa ujumla

M1: Katika momentum closure, 0,6 AMD + 0,5 Gradient; katika scalar closure, 0,6 AMDs + 0,4 Gradient; 0,15 functional coefficient at channel center; C4-C2 numerical scheme.

• Mesh A total error score: asilimia 18,8.
• Mesh B total error score: asilimia 21,6.
• Mesh C total error score: asilimia 37,2.
• Mean ya three meshes: asilimia 25,9.

Main comparative findings

• M1 ilitoa lowest mean composite error score kati ya models zote.
• M2 ilitoa lowest mean error katika second-order statistics kwa asilimia 14,8.
• Previous best hybrid model MA ilikuwa na total score ya asilimia 27,9.
• Nusselt-number error ya M1 ilikuwa asilimia 11,1, wakati ya MA ilikuwa asilimia 24,9.
• Kwa wall-friction coefficient, MA ilitoa asilimia 11,7 error na M1 asilimia 19,3.
• NA solution bila sub-grid model ilitoa lowest Nusselt-number error kwa asilimia 5,8.
• F1 model ilitoa lowest mean temperature error kwa asilimia 21,1.
• M1 ilizalisha velocity-variance peaks kwa balance nzuri zaidi kuliko MA.
• Gradient model ilipunguza distortion iliyosababishwa na Bardina closure katika Reynolds shear stress.
• Functional models zenye QUICK scheme kwa ujumla zilifanya vibaya zaidi kuliko F1 yenye central scheme.
• Coarse meshes ziliongeza errors hasa katika velocity variance, temperature transport na temperature variance.
• Baadhi ya combinations za compressible AMD na Bardina zikawa numerically unstable kwenye fine meshes.

Tathmini ya grafu na majedwali

• Ukurasa 4, Kielelezo 1: Unaonyesha channel geometry yenye periodicity katika directions mbili kati ya walls za 900 K na 1.300 K.
• Ukurasa 8, Kielelezo 2: Unaonyesha transition ya functional-model coefficient kutoka high value near wall hadi low value at channel center.
• Ukurasa 13, Kielelezo 3: Unalinganisha normalized first-order, second-order na total error scores za 12 new na 6 reference models.
• Ukurasa 14, Kielelezo 4: Unatenganisha mean velocity, temperature, Nusselt-number na friction-coefficient errors za M1, MA, NA na F1.
• Ukurasa 15, Kielelezo 5: Unalinganisha mean velocity na temperature profiles upande wa hot na cold wall na DNS.
• Ukurasa 17, Kielelezo 6: Unaonyesha second-order statistical errors za selected models kama separate bars.
• Ukurasa 18, Kielelezo 7: Unaonyesha wall-normal profiles za Reynolds normal stresses katika directions tatu.
• Ukurasa 19, Kielelezo 8: Unatenganisha resolved LES, functional-closure na structural-closure contributions katika Reynolds shear stress.
• Ukurasa 20, Kielelezo 9: Unalinganisha turbulent temperature transport na temperature variance.
• Majedwali kwenye kurasa 25-28: Yanatoa total, first-order, second-order na variable-based error values kwa kila model.
• Appendix plots kwenye kurasa 29-34: Zinaeleza kwa undani contributions za closure terms kwenye velocity na temperature statistics katika fine mesh A na coarse mesh C.

Nguvu za utafiti

• Models zilitestwa si theoretically tu bali a posteriori ndani ya full T-LES simulations.
• Velocity, temperature, wall heat transfer na turbulence correlations zilitathminiwa pamoja kwa kutumia DNS reference.
• Mesh dependence ya model performance ilionyeshwa kwa kutumia three mesh resolutions.
• Interaction kati ya model formulation na numerical scheme ilichunguzwa wazi.
• Unstable models ziliripotiwa, hivyo kuepuka kuchagua successful results pekee.
• Variable-based results zilitolewa pamoja na total error score.
• Previous best model iliingizwa directly katika comparison.

Critical limitations

• Study ni preprint ambayo haijapitia peer review.
• Hakuna direct validation iliyofanywa kwa physical solar receiver au channel experiment.
• Geometry ni simplified flat channel isiyojumuisha details za real solar receiver.
• Only specific operating conditions around friction Reynolds number 800 zilichunguzwa.
• Wall temperatures na thermal loads ziliwakilishwa kwa limited number ya fixed scenarios.
• Model coefficients ni fixed; dynamic-coefficient models hazikutathminiwa.
• Total error score ni normalized composite metric na si direct error percentage ya single physical variable.
• Hakuna separate sub-grid closure kwa temperature variance.
• M1 haikutoa best result katika all first- na second-order quantities.
• Kwenye coarse mesh, total error score ya M1 ilipanda hadi asilimia 37,2.
• Processor-time au core-hour savings relative na DNS hazijaripotiwa.
• Inlet development, pipe curvature, surface roughness, solar-flux distribution na heat conduction ndani ya solid material katika real receivers hazikumodeliwa.
• Unit ya hot-wall heat flux katika S1 condition imeandikwa inconsistently na boundary-flux definition katika text.
• Data hazikuwekwa kwenye open repository; imeelezwa only kwamba zinaweza kupatikana kutoka corresponding author upon reasonable request.

Maelezo ya Chanzo na Mbinu

Original study: “A posteriori study of Thermal-Large Eddy Simulation in solar receiver operating conditions.”

Authors: Yanis Zatout, Françoise Bataille na Adrien Toutant.

Corresponding author: SSRN record inaonyesha Adrien Toutant kama corresponding author. Hakuna equal-contribution statement.

Institutions: PROMES-CNRS, UPR 8521, Université de Perpignan Via Domitia, Perpignan, Ufaransa; LISN-CNRS, UMR 9015, Université Paris-Saclay, Orsay, Ufaransa. Yanis Zatout ana affiliation na institutions zote mbili, wakati Françoise Bataille na Adrien Toutant wana affiliation na PROMES-CNRS.

Source type: Original computational-fluid-dynamics research na preprint.

Preprint date: 10 Juni 2026.

SSRN upload date: 16 Juni 2026.

Peer-review status: Study haijapitia peer review. Findings hazipaswi kutathminiwa kama final na peer-reviewed journal results.

Journal na publisher status: Text inaeleza kwamba preprint iliwasilishwa kwa Elsevier, lakini specific journal name haijatolewa. Accepted au published peer-reviewed journal version haijathibitishwa.

Platform: SSRN. Study pia ina arXiv record.

DOI:10.2139/ssrn.6948778

SSRN link:SSRN study record

arXiv link:arXiv:2606.17596

Funding na computing resources: Study ilifadhiliwa na project ANR-21-CE50-0031. High-performance computing resources zilitolewa kupitia GENCI allocations A0132A05099, A0152A14652 na SS012A15404. CEA STMF team ilishukuriwa kwa development ya TrioCFD software.

Data access: Imeelezwa kwamba data zinazounga mkono findings zinaweza kupatikana kutoka corresponding author upon reasonable request. Hakuna public data repository au source-code link iliyotolewa.

Mbinu, equations, mesh resolutions, model coefficients, error values, tafsiri za grafu na majedwali, findings na limitations katika content hii zinategemea study iliyochunguzwa. External sources zilitumika only kwa bibliographic verification ya title, author order, corresponding author, DOI, institutional records na official study links; hakuna scientific result ambayo haipo katika study iliyoongezwa.