Applied Thermal Engineering 186 (2021) 116431

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Computational fluid-dynamics modelling of supersonic ejectors: Screening

of modelling approaches, comprehensive validation and assessment of
ejector component efficiencies
Giorgio Besagni a, *, Nicolò Cristiani a, b, Lorenzo Croci a, Gaël Raymond Guédon a, Fabio Inzoli a
a
Ricerca sul Sistema Energetico – RSE S.p.A., Power System Development Department, via Rubattino 54, 20134 Milan, Italy
b
Politecnico di Milano, Department of Energy, via Lambruschini 4a, 20156 Milano, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: The efficiency of ejector-based systems (the “system-scale”) relies on the behaviour of the ejector (the “component-
Ejector scale”), related to the flow phenomena within the component itself (the “local-scale”). As a consequence of this
Computational fluid-dynamics multi-scale connection, the precise prediction of the “local-scale” is of fundamental importance to sustain the
Validation
design of commercially viable ejector-based systems. Although it is widely accepted that computational fluid-
Ejector refrigeration
dynamics can achieve the prediction of the “local-scale” (CFD) modelling approaches, a broad agreement
regarding the performances of numerical methods is not reached: different authors applied different methods,
and a complete validation is missing so far. This paper contributes to the current discussion and closes the
knowledge gap by assessing the performances of a CFD approach for single-phase supersonic ejectors. To this
end, a comprehensive validation has been conducted, encompassing a wide range of ejector designs, boundary
conditions and working fluids; besides, a screening of modelling approaches is conducted, encompassing a wide
range of mesh criteria, geometrical modelling (2-Dimensional and 3-Dimensional approaches), solvers (density-
based and pressure-based) and turbulence models (k-ε RNG and k-ω SST). The extensive comparison with
experimental data allowed assessing and determining the influence of mesh criteria, geometrical modelling,
solvers and turbulence models. In particular, k-ω SST has shown the best agreement with the experimental
measurements concerning both global and local flow quantities, with an average entrainment ratio error of 14%
and a maximum of 20%, under on-design operating mode. Finally, the simulation outcomes have been further
post-processed to derive ejector component efficiencies, to contribute to the present discussion regarding clo­
sures in lumped parameter ejector modelling approaches. In conclusion, this paper thoroughly assesses the
performance of a CFD model for single-phase ejector simulations and poses precise guidelines to be applied in
future research activities and to support the design of ejector-based systems.

primary and secondary inlet pressures [2,3]. Thus, an ejector provides a

1. Introduction threefold effect (viz., entrainment, mixing and compression effects),
making it suitable in refrigeration systems either by replacing the
Ejector is a static device constituted by a primary nozzle, a suction lamination valve—to exploit the expansion work (e.g., in CO2 sys­
chamber, a mixing chamber and a diffuser (Fig. 1a). The primary high- tems)—or by providing pump-effect—to replace the compressor (e.g., in
pressure flow accelerates and expands through the primary nozzle, low-pressure refrigerant- systems). In particular, heat-driven ejector
creating a low-pressure region in the entrance of the mixing chamber. At refrigeration systems (ERSs) fall within the latter case and are a prom­
the same time, the secondary flow is entrained into the mixing chamber ising alternative to compressor-based technologies owing to the reli­
owing to the concurrent vacuum- and the shear-effects (interested ability, little maintenance, low initial and operational costs and no
readers may refer to Rao and Gopalan [1] for further details and visu­ refrigerant limitations [4]. Fig. 1b,d (forthcoming nomenclature refers
alizations of the entrainment process). Subsequently, the two streams to this figure) proposes the layout of an ERS cycle, whose operation is
mix within the mixing chamber and, finally, the mixed stream is com­ discussed in the following. The generator supplies low-grade heat to
pressed in the diffuser, achieving an intermediate value between produce the high-pressure stream sent towards the ejector primary inlet

* Corresponding author at: Ricerca sul Sistema Energetico – RSE S.p.A., Power System Development Department, via Rubattino 54, 20134 Milan, Italy.
E-mail address: [email protected] (G. Besagni).

https://doi.org/10.1016/j.applthermaleng.2020.116431
Received 21 August 2020; Received in revised form 19 November 2020; Accepted 2 December 2020
Available online 8 December 2020
1359-4311/© 2020 Elsevier Ltd. All rights reserved.
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Nomenclature E Primary inlet entry semi-diameter,[mm]

F Secondary inlet entry inner semi-diameter,[mm]
Abbreviations G Secondary inlet entry outer semi-diameter,[mm]
AR Aspect Ratio H Nozzle throat semi-diameter[mm]
[ ]
CFD Computational Fluid-Dynamics
h Specific enthalpy, kJ
COP Coefficient of Performance kg

CR Compression Ratio I Nozzle exit semi-diameter,[mm]

DB Density-Based J Connection thickness,[mm]
ERS Ejector Refrigeration System K Mixing chamber semi-diameter,[mm]
EWT Enhanced Wall Function L Diffuser outlet semi-diameter,[mm]
FMG Full Multi-Grid Ma Mach number,[ − ]
GCI Grid Convergence Index P Static pressure,[Pa]
GWP Global Warming Potential T Temperature,[K]
HFC Hydrofluorocarbons δ Entrainment ratio prediction relative error,[ − ]
HFO Hydrofluoroolefins η Ejector component efficiency,[ − ]
LPM Lumped Parameter Model θ Local pressure distribution prediction relative error,[ − ]
NXP Nozzle Exit Position ξ Critical temperature prediction relative error,[ − ]
ODP Ozone Depletion Potential ω Entrainment ratio,[ − ]
PB Pressure-Pased
RANS Reynolds Averaged Navier-Stokes Subscripts
SWF Standard Wall Function 1, … 6 Cycle points as in Fig. 1b
crit critical condition
Symbols dif diffuser
[ ]
ej ejector
ṁ Mass flow rate, kgs
exp experimental
A Nozzle converging section length,[mm] ext extrapolated
a Suction chamber semi-angle,[◦ deg] m mixing chamber
B Nozzle diverging section length,[mm] p primary nozzle
b Nozzle diverging semi-angle,[◦ deg] ref reference
C Mixing chamber length,[mm] s suction chamber
c Diffuser semi-angle,[◦ deg] sat saturation condition
D Diffuser length,[mm]

(point#1, Fig. 1b), which entrains the low-pressure steam from the
ṁ2
evaporator (point#2, Fig. 1b); finally, the mixed stream is conveyed to ω= (1)
ṁ1
the condenser (point#3, Fig. 1b). After the condensation process, the
resulting liquid is split into two parts: the first one is forced back in the In particular, three operational modes can be identified (local fluid
generator by a pump (point#6, Fig. 1b), whereas the remaining part is dynamics in the three operation mode are well described by Haghparast
expanded into a lamination valve (point#5, Fig. 1b) and sent towards et al. [8]):
the evaporator to provide cooling effect [5]. It is worth noting that the
outgoing flow coming from the generator and/or the evaporator could 1. critical mode (double-chocking, on-design): the primary flow is
be either saturated or superheated depending on the refrigerant used chocked, expands and reduces the passage area for the secondary
and the need to avoid condensation effects within the ejector [6]. The flow, which accelerates till supersonic conditions. ω has constant
performances of ERSs are lower compared with a vapour-compressor value owing to double chocking conditions; it should be noted that a
cycle, as the coefficient of performance (COP – determining the “sys­ two-plateau behaviour was observed by Poirier et al. [9] owing to a
tem-scale” performances) is in the range of 0.1–0.7. Besides, COP value “poor ejector design”;
(“system-scale”) is highly related to the ejector behaviour (“component- 2. subcritical mode (single-chocking, off-design): the primary flow is
scale”, i.e., the primary and secondary mass flow rates, the pressure chocked, whereas the secondary one is affected by the outlet pressure
recovery, the operation modes, …): owing to the absence of moving part, value;
the ejector is a fluid dynamic controlled device. In turn, the ejector 3. back-flow mode (malfunctioning): the secondary flow is no more
behaviour is the outcome of the complex fluid dynamic interaction entrained, and primary flow enters into the suction chamber.
characterizing the “local-scale” (i.e., boundary layers subject to adverse
pressure gradients, shock waves, under-expanded jets, flow reparation, To summarize, ejector performance is unrelated from the outlet
recirculation, turbulence mixing phenomena bounded by near-wall re­ boundary conditions until a critical point is reached (Tsat 3 = Tcrit); if
gions, …). It follows that a correct estimation of COP relies on the precise Tsat
3 > Tcrit ejector switches from on-design to off-design operating mode
knowledge of the “local-scale” and “component-scale” as well as on a and ω decreases. Relationship between ω and T3 determines the per­
multi-scale integration of these information [7]. The connection be­ formance of the cycle (Eq. (2)).
tween the “component-scale” and the “system-scale” is made clear looking
at ejector operating curve (Fig. 1c): the entrainment ratio (ω, defined in (2)
Eq. (1)) is a function of the outlet conditions (expressed as saturation
condenser temperature Tsat 3 ), at fixed inlet conditions. As mentioned above, the precise estimation of ejector behaviour, for
different boundary conditions, is of fundamental importance to deter­
mine system COP. To this end, two modelling approaches can be applied

2
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 1. Ejector refrigeration system.

3
 G. Besagni et al.
Table 1
Literature survey regarding CFD ejector modelling: model closures and mesh detail.
Ref. Year Fluid Code PV - coupling Solver Turbulence model Gas model Geometry Mesh size Mesh detail (*) Benchmark

[37] 2019 R-600a (a) – Coupled PB k-ω SST IdealReal 2D 67,923 (c) Mesh adaption [37]
R1234ze (b) gas (NIST) Axisymmetric 3,520,926
(c)3D (d) (d)
[10] 2019 Steam Fluent 17 Coupled DB k-ε (SWF) (a)k-ε (EWF) (b)k-ε Ideal 2D 46,352 Element quality > 0.9Maximum [10]
Realizable (SWF) (c)k-ε Realizable Axisymmetric skewness < 0.45Orthogonal quality > 0.7
(EWF) (d)k-ε RNG (SWF) (e)k-ε RNG
(EWF) (f)k-ω SST (g)inlet turbulence
intensity = 5%
[38] 2019 Steam Fluent – DB k-ε RNG (EWT) – 3D 350,000 30 < y+ < 60 [39]
[40] 2019 R141b Fluent 15.0 – DB k-ε RNG inlet turbulence intensity = 5% Real gas 2D 76,111 – [41]
(NIST) Axisymmetric
[36] 2018 R1234ze Fluent Coupled PB k-ω SST Ideal (a)Real 3D (1/4 domain) 1,700,000 Y+ < 5 [36]
gas (NIST)
(b)
[33] 2018 R141b Fluent 6.3 Coupled DB k-ε Realizable (SWF)inlet turbulence Ideal 2D 25,750 Orthogonal quality > 0.65Growth rate < 12% [33]
intensity = 5%inlet turbulent viscosity Axisymmetric Maximum skewness < 0.4510 < y+ < 65
ratio = 10
[8] 2018 R245fa Fluent 17.0 Coupled PB k-ω SST Real gas 2D 594,195 30 < y+ < 100 [8]
(NIST) Axisymmetric
[42] 2018 Steam Fluent 13 – DB k-ε Realizable (SWF) (a)k-ε RNG (SWF) Ideal 2D 45,000 Orthogonal quality > 0.92Aspect [42]
(b)k-ω SST (c) Axisymmetric ratio < 3.75y+ > 30
[43] 2018 R134a Fluent SIMPLE PB k-ε RNG (SWF) Real gas 2D 133,972 Mesh adaption [43]
(NIST) Axisymmetric
[44] 2018 Steam Fluent 6.3 – DB k-ε RNG Ideal 2D 67,894 Unstructured [44]
Axisymmetric
[31] 2017 Air Fluent 15.5 DB k-ε (a)k-ε Realizable (b)k-ω (c)k-ω SST Ideal 3D Rectangular 220,000 [31]
4

– –
(d)Reynolds Stress model (e)
[45] 2017 R134a Fluent SIMPLE PB k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Real gas 2D 82,197 StructuredMesh adaption [45]
RNG (SWF) (c) (NIST) Axisymmetric
[46] 2017 Air OpenFoam rhocentralfoam – k-ω SSTinlet turbulence intensity = 5% Ideal 2D 350,000 Mixed structured and unstructuredy+ = 1 [21]
2.3 inlet mixing length = 0.07*hydraulic Axisymmetric
diameter
[23] 2017 Steam Fluent 14.0 Coupled – k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 2D 49,000 Structured [24]
RNG (SWF) (c)k-ω (d)k-ω SST (e) Axisymmetric
Transition SST (SWF) (f)inlet turbulence
intensity primary (secondary) = 5%
(10%).Hydraulic diameter
[7] 2017 Steam Fluent 15.0 Coupled DB k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 2D 70,000 – [14]
RNG (SWF) (c)k-ω (d)k-ω SST (e) Axisymmetric
Reynolds Stress model (SWF) (f)Spalart-
Allmaras (g)primary (secondary) = 5%

Applied Thermal Engineering 186 (2021) 116431

(2%).Hydraulic diameter
[16] 2016 R134a Fluent 15 Coupled PB k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal, Real 2D 51,000 (a,b, Unstructured (a,b,c,d)Structured, 21 layer in [22]
RNG (SWF) (c)k-ω SST HR (d)k-ω SST gas (RKS), Axisymmetric c,d)645,000 BL, Growth rate = 1.15, y+ < 0.79 (e)
LR (e)inlet turbulence intensity = 5% Real gas (e)
hydraulic diameter (NIST)
[47] 2016 R134a (two Fluent 13.0 – – k-ε RNG (EWF) Real gas 2D 113,000 – [47]
phase) (Tillner- Axisymmetric
Roth and
Baehr)
[48] 2016 Steam Fluent 14.5 Coupled – k-ω SST Ideal 2D 42,900 Structured [48]
Axisymmetric
[30] 2015 Air Fluent 13 – – k-ε RNG (a)k-ω SST (b) Ideal 2D Rectangular 108,906 (a) Mesh adaptiony+ < 1 [30]
123,898 (b)
(continued on next page)
 G. Besagni et al.
Table 1 (continued )
Ref. Year Fluid Code PV - coupling Solver Turbulence model Gas model Geometry Mesh size Mesh detail (*) Benchmark

[15] 2015 R134a Fluent 14.5 Coupled DB k-ε (a)k-ε Realizable (b)k-ε RNG (c)k-ω Real gas 2D 90,000 Mesh adaption [15]
SST (d)inlet turbulence intensity = 5% (NIST) Axisymmetric
inlet turbulent viscosity ratio = 10
[21] 2015 Air Fluent 14.5 – DB k-ε (a)k-ε Realizable (b)k-ω SST (c) Ideal 2D Rectangular 75,000 (e) Structuredy+ < = 0.9 [21]
Reynolds Stress model (d)inlet (E)3D (1/4 750,000 (f)
turbulence intensity and viscosity ratio domain) (F)
as fully turbulent pipe flow
[49] 2015 R134a Phoenics SIMPLEST PB k-ε (SWF) Real gas 2D 10,350 – [49]
(NIST) Axisymmetric
[50] 2015 R245fa Fluent 14.5 Coupled DB k-ω SST (EWT) Real gas 2D 80,000 Structuredy+ < 1 at supersonic diffuser [50]
(Peng- Axisymmetric
Robinson)
[51] 2015 – Fluent 15.0 Coupled – k-ω SST Ideal 2D – y+ < 1.4 [51]
Axisymmetric
[52] 2014 Air Fluent SIMPLEC PB k-ε RNG (EWF) Ideal 2D 295,810 Structuredy+ = 1 [52]
Axisymmetric
[53] 2014 N2 Fluent 13.0 SIMPLEC PB k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 3D (1/2 domain) 620,000 – [53]
RNG (SWF) (c)k-ω (d)
[54] 2014 Steam Fluent Coupled – k-ε Realizable (SWF) Ideal 2D 3,858 – [54]
Axisymmetric
[29] 2014 Air Fluent 12.0 Coupled – k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 2D 24,705 (g) Mesh adaption [29]
RNG (SWF) (c)k-ω (d)k-ω SST (e) Axisymmetric 1,540,841
Reynolds Stress model (SWF) (f) (g)3D (h) (h)
[1] 2014 Air Fluent – – k-ω SST – 2D 500,000 – [1]
Axisymmetric
[27] 2014 Steam Fluent – – k-ε Realizable Ideal (a)Wet 2D 55,025 – [55](c)
steam (b) Axisymmetric [14](d)
5

[28] 2013 Steam Fluent 6.3 – DB k-ε Realizable (SWF) Ideal 2D 35,600 (a) Structured [28]
Axisymmetric 721,600 (b)
(a)3D (b)
[56] 2013 Air Fluent SIMPLE – k-ε (SWF) Ideal 2D 13,458 StructuredMesh adaption [56]
6.3.26 Axisymmetric
[57] 2013 R744 (two- Fluent – – k-ε RNG – 3D 1,500,000 Structured [57]
phase)
[19] 2013 Steam Fluent 6.3 – DB k-ε Realizable (SWF) (a)k-ω SST (b) Ideal 2D 40,000 – [19]
Axisymmetric
[58] 2013 R141b – – PB k-ε Realizable – 2D 35,282 StructuredMaximum element face [41]
Axisymmetric size = 0.25 mm
[59] 2013 R245fa Fluent 6.3 Coupled – k-ε Realizable (SWF) Ideal 2D 50,000 Triangular with quadrilateral boundary layer [59]
Axisymmetric
[60] 2013 R141b Fluent 12.0 – – k-ε RNG Real gas 3D 310,000 Majority of hexahedral and some tetrahedral [61]
(NIST) region

Applied Thermal Engineering 186 (2021) 116431

[62] 2012 R134a Fluent 6.3 SIMPLEC PB k-ε RNG Real gas 2D 34,650 Mesh adaption (pressure gradient > 10) [62]
(NIST) Axisymmetric
[63] 2012 N2 Fluent Coupled – k-ε Ideal 3D 986,287 Maximum skewness < 5Non- [63]
orthogonally < 70y + nozzle < 300Mesh
adaption
[64] 2012 CO2 (two- Fluent 12.0 SIMPLEC – k-ω SST Real gas – 80,000 – [64]
phase) (NIST)
[20] 2012 Steam Fluent – DB k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 3D 3,830,000 – [14]
RNG (SWF) (c)
[65] 2011 CH4 Phoenics – – k-ε RNG Real gas 2D 31,140 – [65]
3.4 Axisymmetric
[66] 2011 Steam Fluent 6.3 Segregated and – k-ε Realizable Ideal 2D 60,000 – [66]
Coupled Axisymmetric
(continued on next page)
 G. Besagni et al.
Table 1 (continued )
Ref. Year Fluid Code PV - coupling Solver Turbulence model Gas model Geometry Mesh size Mesh detail (*) Benchmark

[67] 2011 Air Fluent – – k-ω SST inlet turbulence intensity Ideal 2D 450,000 – [67]
primary (secondary) = 8% (2%) Axisymmetric
[68] 2010 Steam – Coupled – k-ε (SWF) (a)k-ε Realizable (SWF) (b) Ideal 2D 30,000 10 < y+ < 50 [68]
Axisymmetric
[69] 2010 Steam Fluent 6.2 Coupled – k-ε Realizable (EWF) Ideal 2D – – [14]
Axisymmetric
[12] 2009 Air Fluent 6.2 – – k-ε (a)k-ω SST (b) Ideal 2D 25,820 Y+ = 1 [12]
Axisymmetric
[70] 2009 R141b Fluent 6.2 SIMPLEC – k-ε RNG (SWF) Ideal 2D – Mesh adaption [70]
Axisymmetric
[71] 2008 R141b (a) Phoenics SIMPLEST PB k-ε Real gas 2D – – [41](a)
R245fa (b) 3.5.1 (NIST) Axisymmetric [72](b)
[26] 2007 Steam Fluent Coupled – k-ε Realizable (SWF) Ideal 2D 48,000 (a) – [26]
6.0.12 Axisymmetric 5,000,000
(c)3D (d) (b)
[14] 2007 steam Fluent 6.0 Coupled – k-ε Realizable (SWF) Ideal 2D 43,000 Structured [14]
6

Axisymmetric
[72] 2006 Steam (a) Fluent Coupled DB k-ε (SWF) (e)k-ε Realizable (SWF) (f)k-ε Ideal 2D 19,795 (a) Mesh adaption [72]
R141b (b) RNG (SWF) (g)Spalart-Allmaras (h) Axisymmetric 6,891 (b,c,
R245fa (c)R d)
236fa (d)
[73] 2005 R141b Fluent 6.1 Segregated – k-ε Ideal 2D 120,000 Unstructured [41]
Axisymmetric
[18] 2005 Air Fluent Coupled – k-ε (SWF) (a)k-ε Realizable (SWF) (b)k-ε Ideal 2D – – [74]
RNG (SWF) (c)k-ω (d)k-ω SST (e) Axisymmetric
Reynolds Stress model (SWF) (f)inlet
turbulence intensity = 5%
[75] 2004 Air Fluent 6.1 Coupled – k-ε (a)k-ε RNG (b) Ideal 2D 22,400 – [75]
Axisymmetric

(*) 2D meshes are made of quadrilateral elements and 3D meshed of hexahedral element unless otherwise specified

Applied Thermal Engineering 186 (2021) 116431

 G. Besagni et al.
Table 2
Literature survey regarding CFD ejector modelling: boundary conditions and model performances.
Ref Year Benchmark Range of boundary conditions (Fig. 1) Error Remarks

Generator Evaporator Condenser δ(Eq. (3)) θ (Eq. (4)) ξ(Eq. (5))

On-design Off-design

P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat

3 [ C]
◦
mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%]

[37] 2019 [37] 974 (a)1170 68 (a)63.4 187 (a)203 7 (a)10.5 300 (a)420 42.6 (b) 4.1 (b, 4.1 (b, – – 6.7 (a, 16.2 – – –
(b) (b) (b) (b) (b) c) c) d) (a,d)
[10] 2019 [10] 310–390 – 2.33–3.17 – 3.5–6 – – – – – 9.7 (a) 28.6 – – Boundary layer
6.3 (b) (a) separation study
9.7 (c) 20.6 and how throat
4.1 (d) (b) dimension
11.1 28.5 influences it.
(e)6.8 (c)7.2
(f)6.5 (d)
(g) 30.2
(e)
25.1
(f)23.4
(g)
[38] 2019 [39] 200 120–130 – 5–10 12.35 – – – – – 5 5.6 – – –
[40] 2019 [41] 400.74–604.79 78–95 39.97–47.26 8–12 87.70–137.36 28–41 9.4 24.8 15 27.1 – – – – –
[36] 2018 [36] 1165–1205 61.0–68.6 203–320 5.8–11.8 420–612 – 34.3 38.8 212.8 441.9 15.8 117.7 – – –
(a)17.5 (a) (a) (a) (a)9.1 (a)
(b) 20.3 192.4 458.1 (b) 98.1
(b) (b) (b) (b)
7

[33] 2018 [33] – 90–94 – 4–16 85–115 27–37 5.4 9.1 – – – – 2.5 3.5 –
[8] 2018 [8] – 70–90 – 16–30 165–220 18–30 7.9 12.6 19.8 21.9 – – 7.2 – Exergy and
efficiency analysis
[42] 2018 [42] 160–350 120–139 1.2 – – 10–40 35.9 67.3 – – – – 15.2 24.8 (*)
(a)26.7 (a) (a) (a)
(b) 55.3 13.0 18.2
13.3 (b) (b)7.5 (b)
(c) 23.2 (c) 13.7
(c) (c)
[43] 2018 [43] 2100 70.23 414 10 750 – 6.1 6.5 – – – – – – (*)
[44] 2018 [44] 500 151.8 100 100 200 – 4.3 4.3 – – 12.5 22.7 – – –
[31] 2017 [31] 200–1000 – 60–100 – 100–200 – 26.4 66.4 – – – – – – Schlieren flow
(a)31.2 (a) visualization.
(b) 56.4 Supersonic core
19.4 (b) flow acquiresa

Applied Thermal Engineering 186 (2021) 116431

(c)17.2 45.1 circular cross-
(d) (c) section even
36.9 37.4 though the jet from
(e) (d) the nozzleemerges
99.1 from the
(e) rectangular cross-
section.
[45] 2017 [45] – 70–76 – 6–12 – 33 3.4 (a) 8.3 – – – – – – –
4.6 (b) (a)
7.4 (c) 12.6
(b)
16.2
(c)
(continued on next page)
 G. Besagni et al.
Table 2 (continued )
Ref Year Benchmark Range of boundary conditions (Fig. 1) Error Remarks

Generator Evaporator Condenser δ(Eq. (3)) θ (Eq. (4)) ξ(Eq. (5))

On-design Off-design

P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat

3 [ C]
◦
mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%]

[46] 2017 [21] 350 27 100 27 – – 0.2 0.2 – – – – 7.7 7.7 –

[23] 2017 [24] – 120–140 – 5–10 – – 8.2 (a) 14.0 – – – – 8.6 (a) 16.6 Calibration of
11.6 (a) 14.7 (a) turbulence model
(b)7.7 12.0 (b)7.6 23.0 constants of (a,c)
(c)17.9 (b) (c) (b) with which (c)
(d) 13.0 46.9 12.1 increases
12.3 (c) (d) (c) prediction error by
(e)4.5 26.0 25.7 52.7 2% and (a)
(f) (d) (e)2.8 (d) significant
18.0 (f) 31.6 improves accuracy
(e) (e)
14.0 11.4
(f) (f)
[7] 2017 [14] 198.7–270.3 120–130 0.8–1.2 5–15 3–4.5 24–40 10.2 24.5 – – 27.5 78.8 – – –
(a)10.2 (a) (a) (a)
(b) 18.4 29.5 70.2
12.9 (b) (b) (b)
(c)5.4 22.2 35.1 78.1
(d)6.5 (c) (c) (c)
(e)6.3 16.3 17.9 37.0
(f)13.3 (d) (d) (d)
8

(g) 10.3 17.5 35.8

(e) (e) (e)
11.7 29.8 64.0
(f) (f)19.8 (f)33.9
24.5 (g) (g)
(g)
[16] 2016 [22] 2598.0–3188.1 89.4–99.2 414.60 20 757.2–897.1 29.4–35.4 4.3 (a) 4.3 12.4 13.2 – – 0.2 (a) 0.2 Ideal gas δ = 19.0%
8.8 (b) (a)8.8 (a)16.9 (a)17.9 0.7 (d) (a) real gas NIST
6.0 (c) (b)6.0 (d)31.2 (d)49.1 0.2 (e) 0.7 δ = 3.8%real gas
6.8 (d) (c)6.8 (e) (e) (d) RKS δ = 3.8%
5.7 (e) (d)5.7 0.2
(e) (e)
[47] 2016 [47] 2986–2992 93–95.2 848 33 965 38 – – – – – – 9.9 9.9 Shadowgraph
imaging for
position and angle

Applied Thermal Engineering 186 (2021) 116431

of shocks
[48] 2016 [48] 143.3–270 – 1.2–1.6 – 4.2–6 – 8.9 10.4 – – 9.6 23.1 – – Extended mixing
layer study
[30] 2015 [30] 349–350 10.4–10.5 95–99 17.2–17.3 120–150 12.5–13.2 3.6 (a) 3.6 82.7 82.7 – – – – Shadowgraph
3.6 (b) (a)3.6 (a)77.6 (a)77.6 imaging
(b) (b) (b)
[15] 2015 [15] 2330–3190 84.2–100.0 350–489 15–25 650–807 – 19.9 26.3 140.7 140.7 4.7 (a) 20.8 – – 3 specific
(a)21.2 (a) (a) (a) 10.0 (a) validation cases on:
(b) 22.0 168.0 168.0 (b) 62.5 (i) Influence of the
21.2 (b) (b)94.0 (b)94.0 11.0 (d) turbulence model
(c)15.2 21.8 (c) (c) (c)5.5 on the mixing
(d) (c) 469.3 469.3 (d) between two
(d) (d) coaxial flows, (ii)
(continued on next page)
 G. Besagni et al.
Table 2 (continued )
Ref Year Benchmark Range of boundary conditions (Fig. 1) Error Remarks

Generator Evaporator Condenser δ(Eq. (3)) θ (Eq. (4)) ξ(Eq. (5))

On-design Off-design

P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat

3 [ C]
◦
mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%]

16.7 Shock wave e

(d) turbulent boundary
layer interaction,
(iii) Real gas
thermophysical
properties; K-ε (i)
and real gas (ii)
offers best results.
(EWT slightly
superior to SWF in
(ii))
[21] 2015 [21] 200–500 – 95–101 – 100–187 – 2.4 (a, 3.4 (a, 45.7 127.4 – – 7.7 7.7 Possible influence
e)3.1 e)5.4 (a,e) (a,e) of the wall that
(b,e) (b,e) 47.6 118.9 worsens 2D
2.8 (c, 5.9 (c, (b,e) (b,e) performances (2D
e)3.5 e)5.6 46.6 (c, 105.3 simulations miss
(d,e) (d,e) e)45.7 (c,e) the critical point).
2.3 (a, 3.4 (a, (d,e) 98.9 Wall roughness
f)1.6 f)3.1 17.9 (d,e) may affect the
(b,f) (b,f) (a,f) 40.9 transition to off-
4.1 (c, 10.0 16.4 (a,f) design condition.
9

f)2.2 (c,f) (b,f) 35.9 Water

(d,f) 3.1 (d, 14.1 (c, (b,f) condensation has
f) f)12.8 28.7 (c, negligible impact
(d,f) f)36.3 on ω.
(d,f)
[49] 2015 [49] 1934–2900 86–106 265–415 25–50 420–840 – 3.6 15.5 15.9 52.6 – – 6.3 9.5 Experimental two-
plateau behavior of
entrainment ratio
[50] 2015 [50] – 89–90 – 5–10 161–180 26–32 – – – – – – – – NIST and Peng-
Robinson give
differences on
ω < 1%, ideal gas
up to 6%. Different
wall roughness for
simulated cases are

Applied Thermal Engineering 186 (2021) 116431

tested. Higher wall
roughness
increases frictions
and cause the
critical state to
appear in advance
and the off-design
operation becomes
larger.
[51] 2015 [51] – – – – – – 5.3 9.9 – – – – – – –
[52] 2014 [52] 1000 – 500 – 400–700 – 6.5 15.0 26.2 36.0 13.2 31.8 5.0 5.0 –
[53] 2014 [53] 300–600 – 60 – 100 – – – – – – – – – Secondary mass
flow rate
(continued on next page)
 G. Besagni et al.
Table 2 (continued )
Ref Year Benchmark Range of boundary conditions (Fig. 1) Error Remarks

Generator Evaporator Condenser δ(Eq. (3)) θ (Eq. (4)) ξ(Eq. (5))

On-design Off-design

P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat

3 [ C]
◦
mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%]

error = 35.0% (a),

29.5%, (b) 11.6%
(c), 20.4% (d).
Schlieren optical
measurements best
agrees with (c,d)
partially with (b)
and (a) fails.
[54] 2014 [54] – 90–120 – 5–15 – – 4.2 5.6 – – – – – – (*)
[29] 2014 [29] 743 20.7 86.3 – 137 – – – – – – – – – Particle image
velocimetry (PIV)
supersonic area
best agreements
with (a).PIV vortex
scale best
agreements with
(a). (b) and (e)
underpredict it. (c,
d,f) don’t predict
vortex (failed)
[1] 2014 [1] 769 86 134 9.1 17.5 Schlieren flow
10

– – – – – – – – –
visualization and
non-mixing length
study
[27] 2014 [55](c) 87 (c)270 (d) 117 (c)130 1.228 (c) 10 (d) – – 20 (a, 20 (a, – – 24.6 42.4 6.4 (a, 6.4 –
[14](d) (d) d)6.7 d)6.7 (a,c) (a,c) d)2.1 (a,d)
(b,d) (b,d) 14.6 23.4 (b,d) 2.1
(b,c) (b,c) (b,d)
[28] 2013 [28] 1600 474 10 46 30 – 0.7 (a) 1.4 14.4 27.7 – – 2.1 (a) 2.1 –
1.6 (b) (a)2.4 (a)12.9 (a)24.3 1.7 (b) (a)
(b) (b) (b) 1.7
(b)
[56] 2013 [56] 570 32.85 101.3 26.85 140 – 6.4 6.4 – – – – – – –
[57] 2013 [57] 8006–8486 30.6–30.7 3548–3558 5.2–6.1 3594–3889 – 14.5 21.6 – – 4.0 15.1 – – –
[19] 2013 [19] 111.2–150 – 7.5 3.5–7.9 – 28.2 67.6 – – – – 13.8 21.0 (*)
(a)12.5 (a) (a)7.3 (a)

Applied Thermal Engineering 186 (2021) 116431

(b) 22.6 (b) 14.8
(b) (b)
[58] 2013 [41] – 90–95 – 8–12 – 36.7–38.7 4.4 4.4 – – – – – – –
[59] 2013 [59] – 90–110 – 12–20 – 35 4.6 13.9 – – – – 8.1 27.1 –
[60] 2013 [61] 476.1–930.2 80.3–115 97.2–101.8 31.4–32 127.9–185.1 – 5.9 13.3 – – – – – – –
[62] 2012 [62] 248–486 – 98–111 – 121.9 – – 25 – – – – – – Calibration of
turbulence model
constants (with
which maximum
δ = 8.9%, average
δ = 5.6%)
[63] 2012 [63] 206 20 – – – – – – – – 2.1 7.9 – – Circular and lobed
nozzles
(continued on next page)
 G. Besagni et al.
Table 2 (continued )
Ref Year Benchmark Range of boundary conditions (Fig. 1) Error Remarks

Generator Evaporator Condenser δ(Eq. (3)) θ (Eq. (4)) ξ(Eq. (5))

On-design Off-design

P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat

3 [ C]
◦
mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%]

[64] 2012 [64] 12,330 40 – − 5 3710 – 4.3 11.1 – – 1.4 5.1 – – –

[20] 2012 [14] – 130 – 10 3 – 5.0 (a) 5.0 – – 22.5 70.0 12.0 12.0 Geometrical nozzle
1.5 (b) (a)1.5 (a) (a) (a)6.0 (a) shape study
7.3 (c) (b)7.3 13.0 28.2 (b) 6.0
(c) (b) (b) 10.0 (b)
22.9 70.0 (c) 10.0
(c) (c) (c)
[65] 2011 [65] 10000–14000 – 1500–5000 – 5200 – 11.6 22.0 – – – – – – –
[66] 2011 [66] 42.5–117.9 – 0.85–1.19 – – – 10.0 27.3 88.8 145.9 – – – – Primary mass flow
rate error = 7.7%.
Secondary mass
flow rate
error = 18.8%
[67] 2011 [67] – – – – – – – – – – 7.6 10.9 – – –
[68] 2010 [68] 266–350 – 13–19 – 20–31 – 10.8 10.8 – – – – – – (*)
(a)9.6 (a)9.6
(b) (b)
[69] 2010 [14] 280 – 1.2–1.6 – 3–5.5 – – – – – 15.6 62.1 – – –
[12] 2009 [12] 300–600 – 100 – 100–200 – 5.1 (a) 8.6 10.6 47.1 – – 2.3 (a) 8.6 Validation based on
13.2 (a) (a)33.2 (a)47.4 3.0 (b) (a) ω only is not
(b) 19.1 (b) (b) 8.3 enough because of
11

(b) (b) the very different

flow feature of the
models but rather
similar ω
[70] 2009 [70] 450–600 – – 10–12 90–100 28.5–31.5 4.4 9.3 – – – – – – (*)
[71] 2008 [41](a) 400.6–604.6 78–95 (a) 40–47.3 (a) 8–12 (a) – 29.1–38.7 4.2 (a) 10.5 – – – – 2.3 (b) 8.3 –
[72] (b) (a)1261–1921 100–120 76.2–101.8 8–15 (b) (a) 8.3 (b) (a) (b)
(b) (b) (b) 10.8
(b)
[26] 2007 [26] – 120–130 – 5–15 – 2–7 6.5 17.1 – – – – 5.7 10.8 –
[14] 2007 [14] – 120–130 – 5–10 2.5–6 – 7.9 12.9 33 55.6 17.8 111.5 7 11.6 –
[72] 2006 [72] – 120 (a)90 – 5–10 (a)8 – 34–46 (c) 6.3 (a, 8.2 (a, – – – – 4.2 (c, 5.6 (*)
(b) (b)8–15 27.3–33 e)2.3 e)3.0 f)1.4 (c,f)
100–120 (c)4–12 (d) (a,f) (a,f) (d,f) 4.7
(c)82–88 (d) 7.4 (a, 9.5 (a, (d,f)

Applied Thermal Engineering 186 (2021) 116431

(d) g)24.8 g)
(a,h) 25.2
1.0 (b, (a,h)
e)4.8 1.0 (b,
(b,f) e)4.8
1.5 (b, (b,f)
g)9.6 1.5 (b,
(b,h) g)9.6
6.4 (c, (b,h)
f)25.7 18.4
(d,f) (c,f)
45.5
(d,f)
(continued on next page)
 G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 3

recovery. (e) shows

(C) and (e) are best
Mesh structure detail: radial parameter (y in Fig. 2).

strength and mean

error = 8.2 (a) 4.4

the shock phase,
suited to predict

performances in

Secondary mass
line of pressure

stream mixing
Parameter Number of throat cells

30

flow rate
α
Remarks

70

better
β
110

(b)
– γ

[2]: Lumped-Parameter- (LPM) and Computational Fluid Dynamic (CFD)

max
[%]

–
–

–
models. The former is based on one-dimensional steady-state equations
ξ(Eq. (5))

along with the ejector, and their prediction capabilities highly depend
mean

on the component efficiencies that account for the irreversibility and try
[%]

–
–

–
summarizing the local flow phenomena [10,11]. Conversely, CFD
models use partial differential equations along with domain discretiza­
max

tion techniques to solve local fluid dynamics. Thus, they can examine
[%]

–
–

–
how the “local-scale” affect the “component-scale”, and thus the “system-
θ (Eq. (4))

scale” [7]. Unfortunately, a broad agreement on the performance of CFD

mean
[%]

methods is not achieved yet: different authors apply different methods,

–
–

and a broad validation is missing so far. This paper contributes to the

current discussion and closes the knowledge gap by assessing the per­
formances of a single-phase CFD approach for supersonic ejectors. In
max
[%]

–
–

addition, the simulation outcomes have been further post-processed to

Off-design

derive ejector component efficiencies, to contribute to the present dis­

mean

cussion regarding closures in lumped parameter ejector modelling ap­

[%]

–
–

proaches. To better fit this paper within the correct body of knowledge, a
comprehensive literature survey has been performed and is presented in
max

the forthcoming part of the introduction.

[%]

10
–

–
On-design
δ(Eq. (3))

mean
Error

1.1. Literature survey

[%]

3.9
–

Table 1 and Table 2 collect 55 studies (coverage period: 2004-on­

7.7–38.7

ward) regarding CFD ejector modelling, validated with experimental

3 [ C]
◦

data: Table 1 focuses on the turbulence models, solver details and mesh
Tsat

criteria; conversely, Table 2 focuses on model performances. To derive a

consistent comparison between the different studies, the values in
(*) Errors do not distinguish between on-design error and off-design since no separation was possible

Table 2 have been computed by extracting published values and post-

Condenser

P3 [kPa]

processing with a consistent procedure, distinguishing between on-

design and off-design operation modes1. First, CFD accuracy is
100

100
–

assessed by computing the relative error (δ) between numerical (ωCFD)

and experimental (ωExp) entrainment ratios:
T2 [◦ C]

⃒ ⃒
8–12

⃒ωCFD − ωExp ⃒
δ= (3)
–

ωExp
In case more than one measurement is available, the average and
Evaporator

maximum δ are presented. Second, when a local validation was per­

P2 [kPa]

40–47.3
78–40

formed with N local-quantities (i.e., wall static pressure profiles), θ is

Range of boundary conditions (Fig. 1)

computed as follows:
N ⃒
⃒ ⃒
1 ∑ PCFD,i − PExp,i ⃒
θ= (4)
T1 [◦ C]

78–95
26.85

N i=1 PExp,i
–

The importance of a local validation was noted by Hemidi et al. [12],

who discussed how different local flow structures can obtain the same
400.6–604.6

entrainment ratio. Finally, the accuracy in predicting the critical point

Generator

400–600

100–600

(Fig. 1c) is assessed by computing ξ:

P1 [kPa]

⃒ ⃒
⃒Tcrit,CFD − Tcrit,Exp ⃒
ξ= (5)
Tcrit,Exp
Benchmark

The importance of predicting the position of the critical point is

[41]
[74]

[75]
Table 2 (continued )

1
it is commonly recognised that CFD models have better accuracy in critical
2005
2005

2004
Year

mode (Fig. 1c) owing to the absence of recirculating zones or fluid instabilities.
This choice is also explained by the fact that ejectors are supposed to operate in
[73]
[18]

[75]

double-choking mode, hence the necessity to know exactly the expected error
Ref

when predicting cycle performance.

12
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 4 recent years, Croquer at al. [16] successfully tested a PB solver achieving
Mesh structure detail: axial parameter (x in Fig. 2). good agreement with experimental data (k-ω SST: δ = 6.78%; k-ε Stan­
Parameter AR nozzle AR diffuser dard: δ = 4.27%) and higher stability compared with a DB solver. N. Van
Vu and J. Kracik [17] compared a PB solver with a DB solver with a real
1 1 3
2 3 3 gas model (R32) and observed a discrepancy in predicting ω equal to 1%
3 3 9 and a difference in shockwaves at the entrance and at the end of the
4 9 9 mixing chamber. They also concluded that the PB solver requires less
computation time and is more stable compared with the DB solver.
Concerning turbulence modelling, the Reynolds Averaged Navier-
Table 5 Stokes (RANS) approach is generally used, but no broad agreement on
Mesh structure detail: wall-parameter. the most suitable turbulence model has been reached so far. Bartosie­
Parameter y + nozzle y + diffuser wicz et al. [18] (k-ε Standard, k-ε Realizable, k-ε RNG, , k-ω Standard, k-ω
SST, RSM - air) compared the performance of six RANS models and
– [30;200] [30;200]
n ≈1 [30;200]
concluded that k-ε RNG and k-ω SST are the most suitable ones to predict
d [30;200] ≈1 the shock phase, strength as well as the mean line of pressure recovery;
nd ≈1 ≈1 in particular, k-ω SST showed better performances in terms of stream
mixing prediction. Hemidi at al. [12] (k-ε Standard, k-ω SST - air) found
that k-ε Standard (δ = 5.1%) was more accurate compared with k-ω SST
made clear by looking at the change between on-design and off-design
(δ = 13.2%); they also observed that k-ε Standard and k-ω SST might
operation modes discussed in the previous paragraphs. If Tcrit,exp was
achieve close results in terms of global quantities, but with different
not stated by the authors themselves, its value is estimated as the
local flow. Ruangtrakoon et al. [19] (k-ε Realizable, k-ω SST - steam)
intersection between a linear interpolation of points in the on-design
found that k-ω SST exhibited better performances regarding both ω
and off-design operation modes. Based on the above indicators and
prediction (k-ω SST: δ = 12.5%; k-ε Realizable: δ = 28.2%) and critical
the information in Table 1 and Table 2, some detailed comments
condition estimation (k-ω SST: ξ = 7.3%; k-ε Realizable: ξ = 13.8%).
regarding solver, turbulence modelling, gas model, geometrical model­
They stated that the lower performances of k-ε Realizable might be
ling, mesh detail and working fluid can be made. Such considerations
caused by its inability to obtain an accurate prediction under strong
are also used to support the discussion of the results in the present paper.
adverse pressure gradient. Yang et al. [20] (k-ε Standard, k-ε Realizable,
Concerning the solver, the density-based (DB) approach was histor­
k-ε RNG - steam) found that k-ε Realizable achieved the best agreement
ically used for high-speed compressible flows, whereas the pressure-
with experimental data (δ = 1.5%; θ = 13.0%), followed by k-ε Standard
based (PB) one was mainly used for low-speed incompressible flows.
(δ = 5.0%; θ = 22.5) and k-ε RNG (δ = 7.3%; θ = 22.9%). Nevertheless,
In recent years, both methods have been extended to operate in a wide
none of these models accurately predicted the critical condition,
range of flow conditions [13]. For example, Sriveerakul et al. [14], in
underestimating it by ξ = 12.0% (k-ε Standard), 6.0% (k-ε Realizable)
their pioneering work, discussed the suitability of a DB solver for su­
and 10.0% (k-ε RNG). Del Valle et al. [15] (k-ε Standard, k-ε Realizable, k-
personic ejector modelling (δ = 7.9%); on the other hand, Del Valle et al.
ε RNG and k-ω SST – R134a) numerically investigated the performance
[15] pointed out that a DB solver requires attention in the Courant-
of a R134a ejector with a real-gas approach. k-ω SST was found to
Friedrichs-Lewy condition to prevent the solution from diverging. In

Fig. 2. Mesh structure detail examples for Geometry M (Table 6).

13
 G. Besagni et al.
Table 6
Benchmark geometrical detail. Ejector dimensions (in [mm]) refer to Fig. 3a and b.
Bench- Year Geom# Case NXP A B C D E F G H I J K L C1 C2 K1 a b c Fig. 3 Simulated by
mark

[10] 2019 #A 1 − 15(*) 8(*) 48.58 – 325(*) 6 8.25 (*) 35 1.25 5.5 2.75 14 35 (*) 184.5 196 35 – 5◦ 3.8◦ a [10]
(*) (*) (*) (*)
[79] 2018 #B 1,2 − 11.12 39.48 21.5 234.5 289 (**) 23.26 12.32 44 (**) 5.5 7.75 0.05 10.5 31.49 – – – 46.7 ◦
6.0 ◦
4.2 ◦
b [36]
(**) (**) (**) (**) (**) (**) (**) (**) (**) (**) (**) (**)
[9] 2018 #C 1,2,3,4 − 29 17.1 11.63 89 315 9 12.7 30 3.55 4.85 0.6 8.2 25.2 – – – 14.2◦ 6.4◦ 3.1◦ b [49]
(**)
[80] 2017 #D 1,2,3 0 10.85 20.69 – 85 6 8.54 18.5 1.6 3.41 0.7 4.5 12 63.46 45 7.75 – 5◦ 5◦ a [80]
(**) (**) (**) (**) (**)
[81] 2016 #E 1 − 8 30 7 20 70 3(*) 2.25 (*) 5.5 (*) 1.25 2.05 0.2 (*) 2.55 7.3 (*) – – – 20.2 ◦
6.5 ◦
3.9 ◦
b –
(*) (*)
[15] 2015 #F 1,2,3 − 4.37 7.81 (*) 20.46 – 52 (***) 2.4 (*) 2.88 (*) 16.01 1 1.5 0.1 (*) 2.4 5 40 3.2 2.85 30◦ 1.4◦ 2.9◦ a [15]
[82] 2015 #G 1,2,3 0 13.63 28.33 98.64 135.75 5 8 25.33 2.1 3.55 0.3 (*) 6.51 16 – – – 30◦ 3◦ 4◦ b –
(*)
#H 1,2,3 0 10.8 (*) 28.33 98.64 135.75 5 9 27.33 2.1 3.55 0.3 (*) 6.51 16 52.14 46.47 7.42 30◦ 3◦ 4◦ a –
[22] 2014 #I 1 − 9.72 7.81 (*) 20.46 – 52(***) 2.4 (*) 2.88 (*) 16.01 1 1.5 0.1 (*) 2.4 5 40 3.2 2.85 30◦ 1.4◦ 2.9◦ a [16]
14

#J 1 − 7.05 7.81 (*) 20.46 – 52(***) 2.4 (*) 2.88 (*) 16.01 1 1.5 0.1 (*) 2.4 5 40 3.2 2.85 30◦ 1.4◦ 2.9◦ a
#K 1 − 1.7 7.81 (*) 20.46 – 52(***) 2.4 (*) 2.88 (*) 16.01 1 1.5 0.1 (*) 2.4 5 40 3.2 2.85 30◦ 1.4◦ 2.9◦ a
[52] 2014 #L 1 − 2.8 34 5 (**) 47 212 10 6.63 10 3.2 3.7 0.5 4.7 10 – – – 14◦ 5.7◦ 1.43◦ b [52]
(**) (**)
[14] 2007 #M 1,2,3 0 2.88 (*) 34.3 – 180 3.88 7.5 (*) 27.5 (*) 1 4 0.5 (*) 9.5 20 (*) 94.91 95 12 – 5◦ 3.3◦ a [7,14,20,27,69]
(*) (*)
[41] 1999 #N 1 − 10.5 (*) 30.23 7.57 35.6 56.94 6.65 2.61 (*) 11.55 1.32 2.25 0.2 (*) 3.49 7.04 – – – 30◦ (*) 7◦ (*) 3.6◦ b [40,58,71]
(*) (*) (***) (***) (***) (***) (***) (*)
#O 1 − 10.5 (*) 30.23 7.57 35.6 56.94 6.65 2.61 (*) 11.55 1.32 2.25 0.2 (*) 3.67 7.04 – – – 30 (*)
◦
7 (*)
◦
3.6◦ b
(*) (*) (***) (***) (***) (***) (***) (*)

(*) information not available in the reference. Hypothesized based on the other data and figures.
(**) information not available in the reference. It was provided by the authors upon request.
(***) information not available in the reference. Dimension based on [71]
(***) the diffuser has a different section, see [15] for more detail about diffuser design.

Applied Thermal Engineering 186 (2021) 116431

G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 3. Ejector design.

predict the experimental data fairly well (δ = 15.2%; θ = 5.5%) followed Moreover, downstream of the non-equilibrium condensation, liquid
by k-ε Standard (δ = 19.9%; θ = 4.7%) and k-ε Realizable and k-ε RNG droplets might interact with the flow [6]. In such cases the inclusion of a
(δ > 20%; θ > 10%). The authors also tested the influence of the turbu­ wet-steam model might be beneficial as demonstrated by Wang et al.
lence model on the mixing between two coaxial flows concluding that, [27] (single phase ideal gas: δ = 20%, ξ = 6.4%; wet steam model:
despite all models slightly overestimated the spread of the jet, the ve­ δ = 6.7%, ξ = 2.1%). Dealing with synthetic refrigerant, instead, a real
locities in the mixing region was reasonably predicted by all the tur­ gas approach has been proved to be more accurate than using a perfect
bulence models. Mazzelli et al. [21] (k-ε Standard, k-ε Realizable, k-ω SST gas law by Croquer et al. [16] (ideal gas: δ = 19%; real gas: δ = 3.8%).
and RSM - air) used a 3-Dimensional approach and obtained promising Concerning the geometrical modelling, circular cross-section ejec­
results in on-design operation for all four-turbulence models (δ < 4.1%); tors are mainly axisymmetric except for the second inlet and the suction
conversely, in off-design operation mode δ > 10%. The worst perfor­ chamber. Pianthong et al. [26] compared a 2-Dimensional and 3-Dimen­
mances were observed by RSM (δ = 12.8%, along with numerical stiff­ sional simulations of a steam ejector; they concluded that the axisym­
ness and convergence issues) and k-ω SST (δ = 14.1%). Croquer et al. metric solver provides results close to the 3-Dimensional approach,
[16] (k-ε Standard, k-ε Realizable, k-ε RNG and k-ω SST – R134a) tested a owing to the low velocities at the suction chamber inlet. The same
PB solver against the experimental benchmark provided by Del Valle outcome was reached by Sharifi et al. [28], stating that an axisymmetric
et al. [22]: k-ε Standard was found to be the most accurate model in approach is sufficiently accurate (δ = 0.7%; ξ = 2.1%), compared a full
terms of ω prediction (δ = 4.3%), even if predicting a normal shock wave 3-Dimensional approach (δ = 1.6%; ξ = 1.7%). Gagan et al. [29]
in the mixing chamber instead of an oblique shock train, which was compared a 2-Dimensional and 3-Dimensional approaches and reported
instead captured by k-ω SST. Varga et al. [23] (k-ε Standard, k-ε Real­ a slight difference in the axial velocity and pressure nearby the diffuser,
izable, k-ε RNG, k-ω, k-ω SST and transition SST - steam) performed a six- causing a marginal influence on the whole ejector performance.
turbulence model comparison on the steam benchmark proposed by Conversely, significant discrepancies between 2-Dimensional and 3-
Eames at al. [24]. They concluded that the transition SST achieved the Dimensional domains might occur in rectangular cross-section ejectors
best performance (δ = 4.5%; ξ = 2.8%). On the other hand, k-ω showed a [21,30,31]. In these cases, a 3-Dimensional approach may be beneficial
very low agreement with experimental data resulting in reverse mode owing to the wall effects of the front and back walls: Mazzelli et al. [21]
while the other model correctly predicted on-design operations noted how this effect has a slight influence in the on-design operation
(ξ = 46.9%). Besagni and Inzoli [7] (k-ε Standard, k-ε Realizable, k-ε mode (2-Dimensional: δ = 3%; 3-Dimensional: δ = 2.6%) but has a large
RNG, k-ω, k-ω SST and Spalart-Allmaras - steam) performed a wide influence in the off-design operation mode (2-Dimensional: δ = 46.4%;
screening of turbulence model against the local wall pressure profile 3-Dimensional: δ = 15.3%).
derived by Svriveerakul et al. [14], suggesting k-ω SST for numerical Concerning the meshing criteria, a structured (hexahedral in 3-
investigation on supersonic ejectors (δ = 6.5%; θ = 17.5%). The closure Dimensional domains and quadrilateral in 2-Dimensional domains)
of the RANS system of equations also requires an equation of state mesh is the generally applied strategy. Indeed, the strong directionality
relating density, pressure, and temperature. To this end, the ideal gas of the flow (axial velocity component larger than the transversal
model has been widely used owing to its stability and simplicity but it component) allows the structured grid to be aligned with the momentum
should be adopted only under certain constrains (interested readers may fluxes, reducing numerical diffusion [32]. Wall treatment and y + re­
refer to Zucker and Biblarz [25] for a detailed discussion regarding quirements might vary widely among all studied models and mainly
suitability of the ideal gas assumption); Sriveerakul et al. [14] and depend on the turbulence model adopted. k-ε models and RSM require a
subsequently other authors [7,19,26] have successfully simulated steam mesh that displays y+ values between 30 and 300 if the standard wall
ejectors owing to the low pressure operation. However, when simulating function (SWF) is adopted. Otherwise, when the near-wall modelling
steam ejectors, non-equilibrium condensation may occur within the assumes an enhanced wall treatment (EWT) approach, the solution tends
nozzle. In such occurrence, nozzle behaviour is influenced by the to be y+-insensitive and the accuracy can be locally increased by solving
“condensation shock2” that increase the local pressure and temperature. the viscous sub-layer with a fine mesh nearby the wall (y+ < 1).
Conversely, Spalart-Allmaras, and k-ω models adopr a near-wall treat­
ment similar to the y+-insensitive enhanced wall treatment of the k-ε
2
abrupt collapse of the vapour state which condensate as a shock-like models. It should be noted that the choice of adopting a high (30 < y+ <
disturbance due to rapid drop of temperature and pressure. 300) or low (y+ < 1) Reynolds approach might largely affect the mesh

15
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 7
Benchmark boundary conditions and available data.
Benchmark Year Fluid Geom# Case Run P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat
3 [ C]
◦
Global data (ω) Local data

[10] 2019 Steam A 1 a 360.0 139.8 3.2 25.0 3.5 26.7 x x

b 360.0 139.8 3.2 25.0 4.0 29.0 x x
c 360.0 139.8 3.2 25.0 4.5 31.0 x x
d 360.0 139.8 3.2 25.0 5.0 32.9 x x
e 360.0 139.8 3.2 25.0 5.1 33.4 x
f 360.0 139.8 3.2 25.0 5.3 33.7 x
g 360.0 139.8 3.2 25.0 5.5 34.6 x x
h 360.0 139.8 3.2 25.0 5.8 35.4 x
i 360.0 139.8 3.2 25.0 5.8 35.5 x
j 360.0 139.8 3.2 25.0 6.0 36.2 x x
k 360.0 139.8 3.2 25.0 6.1 36.5 x
[79] 2018 R1234ze B 1 a 1179.0 63.7 218.0 8.4 420.0 42.6 x
b 1179.0 63.7 218.0 8.4 479.0 36.2 x x
c 1179.0 63.7 218.0 8.4 503.0 40.5 x
d 1179.0 63.7 218.0 8.4 509.0 43.6 x
e 1179.0 63.7 218.0 8.4 541.0 46.4 x
f 1179.0 63.7 218.0 8.4 552.0 47.4 x
g 1179.0 63.7 218.0 8.4 612.0 54.6 x
2 a 1160.0 64.7 272.0 11.0 441.0 30.6 x
b 1160.0 64.7 272.0 11.0 479.0 33.4 x x
c 1160.0 64.7 272.0 11.0 492.0 32.9 x
d 1160.0 64.7 272.0 11.0 497.0 44.1 x
e 1160.0 64.7 272.0 11.0 540.0 45.8 x
f 1160.0 64.7 272.0 11.0 556.0 43.5 x
g 1160.0 64.7 272.0 11.0 559.0 49.0 x
h 1160.0 64.7 272.0 11.0 570.0 49.9 x
o 1160.0 64.7 272.0 11.0 579.0 50.8 x
[9] 2018 R134a C 1 a 2633.0 100.0 350.0 15.0 610.0 22.1 x x
b 2633.0 100.0 350.0 15.0 630.0 23.2 x x
c 2633.0 100.0 350.0 15.0 650.0 24.2 x x
d 2633.0 100.0 350.0 15.0 670.0 25.2 x x
e 2633.0 100.0 350.0 15.0 690.0 26.2 x x
f 2633.0 100.0 350.0 15.0 700.0 26.7 x x
g 2633.0 100.0 350.0 15.0 710.0 27.2 x x
h 2633.0 100.0 350.0 15.0 720.0 27.7 x x
i 2633.0 100.0 350.0 15.0 730.0 28.1 x x
j 2633.0 100.0 350.0 15.0 740.0 28.6 x x
k 2633.0 100.0 350.0 15.0 750.0 29.1 x x
l 2633.0 100.0 350.0 15.0 760.0 29.5 x x
m 2633.0 100.0 350.0 15.0 770.0 30.0 x x
n 2633.0 100.0 350.0 15.0 780.0 30.4 x x
o 2633.0 100.0 350.0 15.0 790.0 30.9 x x
p 2633.0 100.0 350.0 15.0 800.0 31.3 x x
q 2633.0 100.0 350.0 15.0 810.0 31.8 x x
r 2633.0 100.0 350.0 15.0 820.0 32.2 x x
s 2633.0 100.0 350.0 15.0 830.0 32.6 x x
2 a 2633.0 100.0 382.0 17.6 720.0 27.7 x
b 2633.0 100.0 382.0 17.6 730.0 28.1 x
c 2633.0 100.0 382.0 17.6 740.0 28.6 x
d 2633.0 100.0 382.0 17.6 750.0 29.1 x
e 2633.0 100.0 382.0 17.6 760.0 29.5 x
f 2633.0 100.0 382.0 17.6 770.0 30.0 x
g 2633.0 100.0 382.0 17.6 780.0 30.4 x
h 2633.0 100.0 382.0 17.6 790.0 30.9 x
i 2633.0 100.0 382.0 17.6 800.0 31.3 x
j 2633.0 100.0 382.0 17.6 810.0 31.8 x
k 2633.0 100.0 382.0 17.6 820.0 32.2 x
l 2633.0 100.0 382.0 17.6 830.0 32.6 x
m 2633.0 100.0 382.0 17.6 840.0 33.1 x
n 2633.0 100.0 382.0 17.6 850.0 33.5 x
o 2633.0 100.0 382.0 17.6 860.0 33.9 x
3 a 2633.0 100.0 425.0 20.7 730.0 28.1 x x
b 2633.0 100.0 425.0 20.7 750.0 29.1 x x
c 2633.0 100.0 425.0 20.7 760.0 29.5 x x
d 2633.0 100.0 425.0 20.7 770.0 30.0 x x
e 2633.0 100.0 425.0 20.7 790.0 30.9 x x
f 2633.0 100.0 425.0 20.7 810.0 31.8 x x
g 2633.0 100.0 425.0 20.7 820.0 32.2 x x
h 2633.0 100.0 425.0 20.7 830.0 32.6 x x
i 2633.0 100.0 425.0 20.7 840.0 33.1 x x
j 2633.0 100.0 425.0 20.7 850.0 33.5 x x
k 2633.0 100.0 425.0 20.7 870.0 34.3 x x
l 2633.0 100.0 425.0 20.7 890.0 35.1 x x
4 a 2633.0 100.0 465.0 23.5 720.0 27.7 x
(continued on next page)

16
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 7 (continued )
Benchmark Year Fluid Geom# Case Run P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat
3 [ C]
◦
Global data (ω) Local data

b 2633.0 100.0 465.0 23.5 740.0 28.6 x

c 2633.0 100.0 465.0 23.5 750.0 29.1 x
d 2633.0 100.0 465.0 23.5 770.0 30.0 x
e 2633.0 100.0 465.0 23.5 780.0 30.4 x
f 2633.0 100.0 465.0 23.5 790.0 30.9 x
g 2633.0 100.0 465.0 23.5 810.0 31.8 x
h 2633.0 100.0 465.0 23.5 830.0 32.6 x
i 2633.0 100.0 465.0 23.5 850.0 33.5 x
j 2633.0 100.0 465.0 23.5 870.0 34.3 x
k 2633.0 100.0 465.0 23.5 890.0 35.1 x
[80] 2017 R141b D 1 a 677.3 100.0 43.5 10.0 75.0 23.8 x
b 677.3 100.0 43.5 10.0 85.0 27.1 x
c 677.3 100.0 43.5 10.0 90.0 28.7 x
d 677.3 100.0 43.5 10.0 95.0 30.2 x
e 677.3 100.0 43.5 10.0 100.0 31.7 x
f 677.3 100.0 43.5 10.0 105.0 33.1 x
g 677.3 100.0 43.5 10.0 107.0 33.6 x
h 677.3 100.0 43.5 10.0 110.0 34.4 x
i 677.3 100.0 43.5 10.0 115.0 35.7 x
2 a 677.3 100.0 33.6 4.0 77.0 24.5 x
b 677.3 100.0 33.6 4.0 82.4 26.3 x
c 677.3 100.0 33.6 4.0 88.7 28.3 x
d 677.3 100.0 33.6 4.0 93.1 29.7 x
e 677.3 100.0 33.6 4.0 99.4 31.5 x
f 677.3 100.0 33.6 4.0 103.7 32.7 x
g 677.3 100.0 33.6 4.0 105.9 33.3 x
h 677.3 100.0 33.6 4.0 107.0 33.6 x
i 677.3 100.0 33.6 4.0 109.3 34.2 x
3 a 677.3 100.0 25.6 − 2.0 78.7 25.0 x
b 677.3 100.0 25.6 − 2.0 85.2 27.2 x
c 677.3 100.0 25.6 − 2.0 91.7 29.2 x
d 677.3 100.0 25.6 − 2.0 97.7 31.0 x
e 677.3 100.0 25.6 − 2.0 100.8 31.9 x
f 677.3 100.0 25.6 − 2.0 102.7 32.4 x
g 677.3 100.0 25.6 − 2.0 105.5 33.2 x
[81] 2016 R134a E 1 a 2300.0 73.7 410.0 9.7 730.0 28.1 x
b 2300.0 73.7 410.0 9.7 770.0 30.0 x
c 2300.0 73.7 410.0 9.7 810.0 31.8 x
d 2300.0 73.7 410.0 9.7 850.0 33.5 x
e 2300.0 73.7 410.0 9.7 890.0 35.1 x
[15] 2015 R134a F 1 a 2330.0 84.2 415.0 20.0 650.0 24.2 x x
2 a 3190.0 100.0 350.0 15.0 690.0 26.2 x x
3 a 3190.0 100.0 415.0 20.0 690.0 26.2 x x
[82] 2015 R245fa G 1 a 1129.8 95.0 75.8 8.0 184.5 31.0 x
b 1129.8 95.0 75.8 8.0 191.1 32.0 x
c 1129.8 95.0 75.8 8.0 197.9 33.0 x
d 1129.8 95.0 75.8 8.0 200.5 33.4 x
e 1129.8 95.0 75.8 8.0 204.8 34.0 x
f 1129.8 95.0 75.8 8.0 210.4 34.8 x
2 a 1129.8 95.0 89.5 12.0 197.9 33.0 x
b 1129.8 95.0 89.5 12.0 204.8 34.0 x
c 1129.8 95.0 89.5 12.0 208.3 34.5 x
d 1129.8 95.0 89.5 12.0 213.2 35.2 x
e 1129.8 95.0 89.5 12.0 214.8 35.4 x
f 1129.8 95.0 89.5 12.0 226.1 36.9 x
3 a 1129.8 95.0 105.2 16.0 204.8 34.0 x
b 1129.8 95.0 105.2 16.0 212.0 35.0 x
c 1129.8 95.0 105.2 16.0 220.2 36.1 x
d 1129.8 95.0 105.2 16.0 222.6 36.4 x
e 1129.8 95.0 105.2 16.0 225.2 36.8 x
f 1129.8 95.0 105.2 16.0 240.3 38.7 x
H 1 a 1129.8 95.0 75.8 8.0 178.1 30.0 x
b 1129.8 95.0 75.8 8.0 184.5 31.0 x
c 1129.8 95.0 75.8 8.0 190.1 31.9 x
d 1129.8 95.0 75.8 8.0 191.4 32.1 x
e 1129.8 95.0 75.8 8.0 193.0 32.3 x
f 1129.8 95.0 75.8 8.0 200.8 33.4 x
2 a 1129.8 95.0 89.5 12.0 184.5 31.0 x
b 1129.8 95.0 89.5 12.0 191.1 32.0 x
c 1129.8 95.0 89.5 12.0 200.8 33.4 x
d 1129.8 95.0 89.5 12.0 202.0 33.6 x
e 1129.8 95.0 89.5 12.0 203.2 33.8 x
f 1129.8 95.0 89.5 12.0 215.2 35.5 x
3 a 1129.8 95.0 105.2 16.0 191.1 32.0 x
b 1129.8 95.0 105.2 16.0 197.9 33.0 x
c 1129.8 95.0 105.2 16.0 211.6 35.0 x
(continued on next page)

17
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 7 (continued )
Benchmark Year Fluid Geom# Case Run P1 [kPa] T1 [◦ C] P2 [kPa] T2 [◦ C] P3 [kPa] Tsat
3 [ C]
◦
Global data (ω) Local data

d 1129.8 95.0 105.2 16.0 213.3 35.2 x

e 1129.8 95.0 105.2 16.0 215.8 35.5 x
f 1129.8 95.0 105.2 16.0 230.8 37.5 x
[22] 2014 R134a I 1 a 2889.0 94.4 414.6 20.0 695.0 26.5 x
b 2889.0 94.4 414.6 20.0 712.0 27.3 x
c 2889.0 94.4 414.6 20.0 732.0 28.3 x
d 2889.0 94.4 414.6 20.0 754.0 29.3 x
e 2889.0 94.4 414.6 20.0 770.0 30.0 x
f 2889.0 94.4 414.6 20.0 789.0 30.8 x
g 2889.0 94.4 414.6 20.0 808.0 31.7 x
J 1 a 2889.0 94.4 414.6 20.0 686.0 26.0 x
b 2889.0 94.4 414.6 20.0 709.0 27.0 x
c 2889.0 94.4 414.6 20.0 726.0 28.0 x
d 2889.0 94.4 414.6 20.0 750.0 29.1 x
e 2889.0 94.4 414.6 20.0 766.0 29.8 x
f 2889.0 94.4 414.6 20.0 790.0 30.9 x
g 2889.0 94.4 414.6 20.0 807.0 31.6 x
K 1 a 2889.0 94.4 414.6 20.0 694.0 26.4 x
b 2889.0 94.4 414.6 20.0 714.0 27.4 x
c 2889.0 94.4 414.6 20.0 723.0 27.8 x
d 2889.0 94.4 414.6 20.0 750.0 28.9 x
e 2889.0 94.4 414.6 20.0 766.0 29.8 x
f 2889.0 94.4 414.6 20.0 784.0 30.6 x
g 2889.0 94.4 414.6 20.0 805.0 31.6 x
h 2889.0 94.4 414.6 20.0 824.0 32.4 x
[52] 2014 air L 1 a 1000.0 25.0 500.0 25.0 400.0 25.0 x x
b 1000.0 25.0 500.0 25.0 450.0 25.0 x x
c 1000.0 25.0 500.0 25.0 450.0 25.0 x x
d 1000.0 25.0 500.0 25.0 520.0 25.0 x x
e 1000.0 25.0 500.0 25.0 550.0 25.0 x x
f 1000.0 25.0 500.0 25.0 600.0 25.0 x x
g 1000.0 25.0 500.0 25.0 640.0 25.0 x x
h 1000.0 25.0 500.0 25.0 670.0 25.0 x x
i 1000.0 25.0 500.0 25.0 700.0 25.0 x x
[14] 2007 Steam M 1 a 270.3 130.0 1.2 10.0 3.0 24.1 x x
b 270.3 130.0 1.2 10.0 3.5 26.7 x x
c 270.3 130.0 1.2 10.0 4.0 29.0 x x
d 270.3 130.0 1.2 10.0 4.5 31.0 x x
e 270.3 130.0 1.2 10.0 4.6 31.4 x
f 270.3 130.0 1.2 10.0 4.7 31.8 x
g 270.3 130.0 1.2 10.0 4.8 32.1 x
h 270.3 130.0 1.2 10.0 4.9 32.5 x
i 270.3 130.0 1.2 10.0 5.0 32.9 x
j 270.3 130.0 1.2 10.0 5.1 33.2 x
k 270.3 130.0 1.2 10.0 5.2 33.6 x
j 270.3 130.0 1.2 10.0 5.3 33.9 x
2 a 198.7 120.0 1.2 10.0 3.0 24.1 x x
b 198.7 120.0 1.2 10.0 3.3 25.7 x
c 198.7 120.0 1.2 10.0 3.5 26.7 x
d 198.7 120.0 1.2 10.0 3.7 27.6 x
e 198.7 120.0 1.2 10.0 3.8 28.1 x
f 198.7 120.0 1.2 10.0 3.9 28.5 x
g 198.7 120.0 1.2 10.0 4.0 29.0 x
h 198.7 120.0 1.2 10.0 4.1 29.4 x
i 198.7 120.0 1.2 10.0 4.2 29.8 x
3 a 270.3 130.0 0.9 5.0 3.0 24.1 x x
b 270.3 130.0 0.9 5.0 3.5 26.7 x
c 270.3 130.0 0.9 5.0 4.0 29.0 x
d 270.3 130.0 0.9 5.0 4.5 31.0 x
e 270.3 130.0 0.9 5.0 4.6 31.4 x
f 270.3 130.0 0.9 5.0 4.7 31.8 x
g 270.3 130.0 0.9 5.0 4.8 32.1 x
h 270.3 130.0 0.9 5.0 4.9 32.5 x
i 270.3 130.0 0.9 5.0 5.0 32.9 x
[41] 1999 R141b N 1 a 400.6 78.0 40.0 8.0 92.6 29.5 x
2 a 465.5 84.0 40.0 8.0 101.2 32.0 x
3 a 538.0 90.0 40.0 8.0 116.2 36.0 x
O 1 a 400.6 78.0 47.3 12.0 91.6 29.2 x
2 a 465.5 84.0 47.3 12.0 102.6 32.4 x
3 a 538.0 90.0 47.3 12.0 116.2 36.0 x
4 a 604.6 95.0 47.3 12.0 127.3 38.7 x
5 a 400.6 78.0 40.0 8.0 87.7 28.0 x
6 a 465.5 84.0 40.0 8.0 97.7 31.0 x
7 a 538.0 90.0 40.0 8.0 108.5 34.0 x
8 a 604.6 95.0 40.0 8.0 120.2 37.0 x

18
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 8 might reduce computational time without lacking in precision if general

Synthetic refrigerant properties. non-equilibrium conditions are negligible [7]. In many cases a detailed
Fluid R134a R141b R1234ze(E) R245fa wall resolution is adopted, hence the first cell close to a wall has y+ ≈ 1
along with a sufficient number of cells to solve the boundary layer, such
Formula C2F4H2 C2H3Cl2F C3F4H2 C3F5H3
Pcrit [kPa] 4059.3 4,212.00 3634.9 3651.0 as within the air-ejector studies of Mazzelli et al. [21] and Little et al.
Tcrit [◦ C] 101.1 477.5 109.4 153.9 [30]. In other cases, when dealing with an organic refrigerant and a real-
Molecularmass [g/mol] 102 116.96 114 134 gas model, a wall-function approach is used to reduce the computation
Boiling point[◦ C] − 26.15 32 − 18.97 15.3 effort. Thongtip and Aphornratana [33] (R141b ejector) used a SWT
Safety group (*) A1 A1 A2L B1
ODP1 0 0.12 0 0
with k-ε realizable, ensuring a y + in the range of 10 – 65, and obtained
GWP2100 1430 725 <1 1030 accurate predictions (δ = 5.4%; ξ = 2.5%). Haghparast et al. [8] (R245fa
(*) ASHRAE Standard 34 [83] ejector) used a similar approach when applying k-ω SST (δ = 7.9%;
1
Zero ozone depletion potential
ξ = 7.2%). To clarify the influence of a full boundary layer resolution
2
Global Warming Potential compared to a wall-function approach, Croquer et al. [16] performed
some mesh tests using k-ω SST. In the high-Reynolds formulation, grid
independency was obtained with 51,000 cells, whereas the low-
size and depends on the final goal of the study: if heat transfer or
Reynolds approach required 645,000 elements. The change in δ, in
complex near-wall phenomena are expected, a low Reynolds approach is
these two configurations was ≈ 1%. Besagni et al. [7] performed near-
likely to be the most appropriate. Conversely, a high Reynolds approach

Fig. 4. Tested working fluids: P-h representations and boundary conditions.

19
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 5. Pressure and velocity Mach number profiles: influence of the solver.

wall treatment sensitivity analysis comparing SWF, Non-Equilibrium- reached. Moreover, the validation cases of the presented studies are
Wall-Function and EWT that resulted in the choice of a SWF. This often limited in numbers and geometries, as well as the tested working
conclusion along with the adoption of high-Reynolds turbulence models fluids or boundary conditions with the consequence that a model which
was based on the absence of detachment or reattachment of the flow and has outstanding performance on one application could be not suitable in
general non-equilibrium conditions. Han et al. [10] proved that EWT other cases or, more simply, having unknown uncertainties when
improves the local wall pressure prediction with k-ε models: k-ε Stan­ applied out of its validation boundaries.
dard, k-ε Realizable, k-ε RNG with SWF resulted in θ equal to 9.7%, 9.7,
11.1%, whereas when adopting EWF it was equal to 6.3%, 4.1%, 6.8%, 1.2. Aim of this paper
respectively.
Finally, an ejector can be operated with different working fluids. Air This paper contributes to the current discussion and contributes to
and steam are mainly used in laboratory test rigs, owing to the simplicity the present discussion and tries filling the gap mentioned above by
and safety of the experimental facility. R134a, R245fa, R152a were assessing the performances of a single-phase CFD approach for super­
proved to be successfully simulated by CFD approaches but, given the sonic ejectors. In particular, the present paper proposes a “baseline”3
international regulations phasing-out hydrofluorocarbons (HFC) [34], ejector CFD model which could provide a basis for upcoming research
the choice of possible replacements (viz., hydrofluoroolefins, HFO – i.e., activities on ejector modelling. To this end, a comprehensive validation
R1132a, R1123, R1234yf, R1243zf, R1234ze, R1224yd, R1233zd, is conducted, encompassing a wide range of ejector designs, boundary
R1336mzz [35]) is attracting a growing discussion. To date, only a few conditions, and working fluids; in addition, a broad screening of
numerical studies are performed with these refrigerants, and the existing modelling approaches is conducted, encompassing a wide range of mesh
validations revealed low accuracy prediction of the off-design operation types, geometrical modelling (2-Dimensional and 3-Dimensional),
[36].
What emerges from the above literature survey is that several
methods have been applied when dealing with numerical simulation of 3
A “baseline model” is a CFD model setup which includes physics-based
supersonic ejectors. As it may be noted by Table 1, different strategies in models for relevant flow phenomena at least for a well-defined range of local
terms of solver, turbulence modelling, gas model, geometrical modelling flow conditions, as defined by Lucas et al. [76] D. Lucas, R. Rzehak, E. Krepper,
and mesh detail have been adopted with the result that some model T. Ziegenhein, Y. Liao, S. Kriebitzsch, P. Apanasevich, A strategy for the qual­
performed better than other (Table 2) but a general conclusion is not ification of multi-fluid approaches for nuclear reactor safety, Nuclear Engi­
neering and Design, 299 (2016) 2–11.

20
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 9
Mesh screening results with PB solver (it is worth noting that for M3a case, every mesh is considered as refined at the diffuser because of the obtained low y + value at
the diffuser even with the coarse meshes) - Pressure and Mach profile presented in Figures 6 – 8 refer to bold highlighted mesh.
Benchmark F1a (ωEXP = 0.592) Benchmark M3a (ωEXP = 0.309)

Mesh code Number of elements [-] ωCFD [-] Deviations Number of elements [-] ωCFD [-] deviations

ω Eq. (6) P Eq. (7) Ma Eq. (8) ω Eq. (6) P Eq. (7) Ma Eq. (8)

α1 21,526 0.664 − 4.7% 0.8% 7.1%

α1n 160,684 0.668 − 4.1% 0.8% 6.7%
α1d 73,366 0.662 − 4.9% 0.8% 7.2% 16,229 0.300 − 4.71% 3.0% 3.7%
α1nd 212,524 0.666 − 4.3% 0.8% 6.8% 51,788 0.312 − 0.73% 1.0% 2.6%
α2 16,533 0.654 ¡6.0% 1.0% 8.4%
α2n 116,361 0.674 − 3.2% 0.6% 6.1%
α2d 56,673 0.653 − 6.3% 1.0% 8.5% 14,103 0.300 ¡4.81% 3.7% 5.0%
α2nd 156,501 0.673 − 3.4% 0.6% 6.4% 36,810 0.321 2.18% 1.4% 3.0%
α3 8,121 0.659 − 5.3% 1.0% 10.4%
α3n 103,410 0.666 − 4.4% 0.9% 9.9%
α3d 25,422 0.657 − 5.7% 1.1% 10.8% 5,731 0.287 − 8.93% 4.8% 5.3%
α3nd 120,690 0.664 − 4.6% 1.0% 10.2% 19,555 0.308 − 2.13% 3.0% 4.7%
α4 6,172 0.611 − 12.3% 2.1% 14.9%
α4n 68,200 0.628 − 9.8% 1.4% 13.3%
α4d 19,852 0.610 − 12.4% 2.1% 15.2% 4,745 0.291 − 7.64% 5.1% 8.2%
α4nd 81,880 0.626 − 10.1% 1.4% 13.8% 14,222 0.307 − 2.53% 3.5% 5.5%
β1 94,743 0.676 ¡2.9% 0.5% 3.4%
β1n 256,740 0.679 − 2.6% 0.6% 3.3%
β1d 152,658 0.674 − 3.3% 0.5% 3.7% 72,898 0.305 ¡3.15% 1.1% 1.6%
β1nd 31,4475 0.676 − 2.9% 0.5% 3.6% 112,444 0.312 − 0.68% 0.8% 1.4%
β2 75,852 0.679 ¡2.5% 0.5% 3.4%
β2n 196,089 0.685 − 1.6% 0.4% 2.8%
β2d 122,157 0.677 − 2.8% 0.5% 3.8% 61,599 0.306 ¡2.74% 1.6% 2.2%
β2nd 242,394 0.683 − 1.9% 0.4% 3.3% 86,265 0.315 0.12% 0.9% 1.4%
β3 34,560 0.679 ¡2.6% 0.6% 7.1%
β3n 148,749 0.685 − 1.7% 0.6% 6.7%
β3d 60,207 0.676 − 2.9% 0.6% 7.3% 25,657 0.298 ¡5.28% 2.2% 3.4%
β3nd 174,396 0.682 − 2.1% 0.5% 6.8% 41,116 0.308 − 2.25% 1.7% 3.4%
β4 26,986 0.671 ¡3.6% 0.6% 8.3%
β4n 100,351 0.686 − 1.4% 0.6% 7.1%
β4d 42,691 0.670 − 3.8% 0.6% 8.5% 21,506 0.299 ¡4.91% 3.4% 4.4%
β4nd 116,455 0.688 − 1.3% 0.6% 6.9% 31,757 0.312 − 0.84% 2.3% 3.2%
γ1 228,748 0.682 − 2.1% 0.4% 2.4%
γ1n 35,2762 0.684 − 1.8% 0.4% 2.3%
γ1d 270,769 0.681 − 2.3% 0.4% 2.6% 173,052 0.308 − 2.09% 0.9% 1.3%
γ1nd 394,783 0.683 − 1.9% 0.4% 2.5% 234,453 0.314 − 0.34% 0.5% 0.8%
γ2 184,362 0.683 ¡1.9% 0.4% 2.2%
γ2n 276,624 0.687 − 1.4% 0.3% 1.8%
γ2d 218,340 0.682 − 2.1% 0.4% 2.5% 147,724 0.309 ¡1.74% 1.3% 1.9%
γ2nd 310,602 0.685 − 1.6% 0.3% 2.2% 188,107 0.315 0.09% 0.5% 0.8%
γ3 81,392 0.683 − 2.0% 0.5% 5.6%
γ3n 167,984 0.686 − 1.4% 0.4% 5.4%
γ3d 105,863 0.681 − 2.2% 0.4% 3.3% 60,152 0.303 − 3.58% 1.5% 2.4%
γ3nd 182,201 0.685 − 1.7% 0.4% 3.1% 84,260 0.310 − 1.43% 1.1% 2.3%
γ4 65,222 0.665 − 4.6% 0.8% 8.0%
γ4n 120,614 0.677 − 2.8% 0.5% 6.5%
γ4d 76,877 0.664 − 4.6% 0.8% 8.1% 50,163 0.306 − 2.75% 2.8% 3.6%
γ4nd 133,019 0.676 − 2.9% 0.5% 6.5% 65,343 0.315 0.07% 1.6% 2.6%
Reference case* 1,045,369 0.696 – – – 978,093 0.315 – – –

(*) as defined in Section 2.3

solvers (density-based and pressure based), and turbulence models (k-ε Section 1.1, two state-of-the-art turbulence models in ejector simulation
RNG and k-ω SST). The comprehensive comparison with experimental have been compared: k-ε RNG and k-ω SST. The former has been
data allowed assessing the influence of mesh types, geometrical employed along with a EWT, whereas the latter adopt a y+-insensitive
modelling, numerical solver and turbulence models. This paper is approach similar to EWT. For further information regarding their
structured as follows. In Section 2, the CFD approach and mesh sensi­ mathematical formulation, the reader should refer to Wilcox [77],
tivity analysis are outlined. In Section 3 benchmark is described and, in whereas their implementation in ANSYS-Fluent is described in ref. [13].
Section 4, the numerical results are presented and discussed. Finally, the To limit the numerical diffusion, second-order upwind numerical
main conclusions are outlined. schemes have been used for the spatial discretization, except for the
pressure equation, where the PRESTO! Scheme has been selected, as it
2. Physical model was designed for flows involving steep pressure gradients [16]. Gradi­
ents have been evaluated by a least-squares approach and second-order
2.1. Numerical model upwind schemes for the turbulence variables have been used. A two-step
approach is adopted for the initialization: a hybrid initialization fol­
The finite volume code ANSYS Fluent (Release 2020 – R1) has been lowed by a full multi-grid (FMG) initialization. The numerical solution
used to solve the steady-state RANS equations for the turbulent has been considered converged when the normalized difference of mass
compressible Newtonian flow. Based on the outcomes of Table 1 and flow rates at the inlets and at outlet were < 10-5 and the mass flow-rate

21
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 6. Pressure and velocity Mach number profiles: influence of the cell throat size.

variation of primary and secondary flow on the last 50 iterations [7].

was < 10-5. PB solver with Coupled algorithm has been adopted since it
is described in the literature as far more stable than density-based solver
2.3. Domain discretisation and mesh criteria
nevertheless sufficiently accurate and suitable for high-velocity
compressible flows. To clarify these concepts, a solver sensitivity anal­
After a preliminary study regarding the influence of 2-Dimensional
ysis has been performed and is presented in Section 4.2 and in Appendix
and 3-Dimensional approaches (presented in Appendix A) as well as
A. B. Pseudo-transient option was enabled which was found to speed up
the outcomes of refs [26,28,29], a 2-Dimensional axisymmetric
the steady-state solution [13]. Fluid properties have been evaluated with
approach has been performed and a comprehensive sensitivity analysis
the real-gas NIST database [78] except for steam (A1, M1, M2, M3
on the meshing criteria for ejector simulation has been carried out. The
benchmarks, see Sections 3) and air (L1, benchmarks, see Sections 3),
goal of sensitivity analysis is to derive general guidelines for grid inde­
where the ideal gas assumption is made.
pendency in ejector simulations. To this end, under the assumption of
structured meshes, three geometrical parameters have been identified to
2.2. Boundary conditions characterize the mesh fully: (i) the radial parameter (Table 3) is the
mixing chamber cell throat size, expressed as the number of cells in the
Ejector inlets have been modelled with pressure boundary conditions ejector’s throat; (ii) the axial parameter (Table 4) is the aspect ratio (AR,
(prescribing total pressures and temperatures). It is worth noting that the ratio between the axial and radial lengths of the element) of the cells
benchmark data generally consists in static pressure, but Mazzelli et al. at nozzle exit and diffuser entrance (with a smooth transition in be­
[21] proved that inlet variables are close to stagnant conditions; besides, tween); (iii) the wall parameter (Table 5) is y+. Finally, the application
their sensitivity analysis showed that a static/total change in the pres­ of a solution-adaptive mesh refinement based on a Mach gradient cri­
sure boundary conditions has a minor influence (<0.5%) in the pre­ terion has been tested as well. It should be noted that for the definition of
dicted mass flow rates. The outlet boundary condition is implemented as values in Tables 3–5, we referred to the numerical studies performed by
a pressure boundary condition. Finally, turbulence boundary conditions Del Valle et al. [15] on benchmark F1a (see Section 3 for a description of
are implemented as hydraulic diameter and turbulent intensity (5% for this code name), and by Besagni and Inzoli [7] on benchmark M3a (see
the primary flows and 2% for the secondary one), as suggested in ref. Section 3 for a description of this code name), such that a comparable

22
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 7. Pressure and velocity Mach number profiles: influence of the aspect ratio.

medium-mesh size is achieved (90,000 cells for F1a and 70,000 cells for ⃒ ⃒
N ⃒
M3a, respectively). 1 ∑ Mai − MaREF,i ⃒
Ma deviation = (8)
The combinations of above-reported 3 parameters have allowed N i=1 MaREF,i
obtaining 48 meshes, identified by a three-digits code name (built on the
combinations of the digits reported in Table 3, Table 4 and Table 5; an 3. Benchmark
example is given in Fig. 2.
The mesh sensitivity study has been conducted on a reduced A comprehensive set of experimental benchmarks have been
benchmark constituted by the 2 cases (F1a and M3a - see Section 3 for a collected, and is summarized in Table 6. The geometries dimensions
description of code names of such benchmarks). F1a has been simulated refer to Fig. 3a or Fig. 3b depending on the mixing chamber design:
with the real-gas approach, whereas M3a has been simulated with an either tapered section (Fig. 3a) or constant diameter (Fig. 3b). The
ideal-gas model. Finally, to evaluate grid independence, the deviation nozzle exit position (NXP) indicates the position of the primary nozzle
from a reference case (obtained as the refinement of the finest mesh) with respect to the entry plane of the converging entrance zone of the
have been computed. For the sake of clarity, ω deviation is computed mixing chamber: NXP < 0 if the nozzle exit is upstream and vice-versa.
using Eq. (6), whereas the static wall pressure and axial Ma number All experimental benchmarks in Table 6 are available with different
distributions have been computed in an equidistant finite number of boundary conditions, summarized in Table 7. From now the benchmark
point and compared with the reference case ones, obtaining the average data will be identified by a 3-digits code which refers to Table 7: the first
deviation of the quantities Eqs. (7) and (8): digit specifies the code name (benchmark reference, called geom#), the
ω − ωREF second digit specifies the case and the last one the run. Finally, the tested
ω deviation = (6) working fluid properties are presented in Table 8, whereas the P-h
ωREF
curves and boundary conditions are shown in Fig. 4.
N ⃒
⃒ ⃒
1 ∑ Pi − PREF,i ⃒
P deviation = (7) 4. Results
N i=1 PREF,i

Herein the mesh and solver screenings have been presented. Subse­
quently, the comparison between the turbulence models has been

23
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 8. Pressure and velocity Mach number profiles: influence of the solution-adaptive refinement.

solver comparison (k-ω SST as turbulence model). The results are pre­
Table 10 sented in Fig. 5, which displays the axial wall static and Mach profiles as
10 GCI study. well as in Appendix B, which presents the extended results for M3a case.
Case ωmedium-fineext GCImedium-fine numerical range It was found that PB and DB solvers achieved very close outcomes. In
particular, the normalized difference of ω is approximately 1.7% and
F1a 0.689 1.14% 0.681 0.697
M3a 0.316 2.61% 0.307 0.324 2.6% for F1a and M3a cases, respectively.
Looking at Fig. 5, it is noted that PB and DB axial profiles for case
M3a are close to each other: PB solver seems to compute more sweeping
presented and discussed. higher wall pressure oscillation and a more detailed shock train. It has
also been observed, as also mentioned by Croquer et al. [16], that PB
4.1. Solver comparison solver showed higher stability and faster convergence compared with
DB solver. In conclusion, both PB and DB solvers are suitable to simulate
Benchmarks F1 and M3a have been used to perform a PB and DB ejector fluid-dynamics, even if the PB one is suggested owing to the more

Fig. 9. Example of Mesh adaption - Case M3a, Mesh β2d, adapted.

24
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 11
Numerical results: local and global model performances.
Benchmark Year Fluid Geom# Case Run Figure ωEXP k-ω SST k-ε RNG

ωCFD δ [%] θ [%] ωCFD δ [%] θ [%]

[10] 2019 Steam A 1 a Fig. 10b 1.220 1.076 − 11.8 13.1 0.801 − 34.3 18.0
b 1.220 1.076 − 11.8 10.1 0.801 − 34.3 15.5
c 1.220 1.076 − 11.8 10.7 0.792 − 35.1 14.4
d 1.220 1.075 − 11.9 8.6 0.622 − 49.0 18.4
e 1.210 1.071 − 11.5 – 0.543 − 55.1 –
f 1.200 1.059 − 11.8 – 0.471 − 60.8 –
g 1.010 0.965 − 4.4 4.2 0.277 − 72.6 12.1
h 0.650 0.763 17.4 – 0.000 − 100.0 –
i 0.500 0.712 42.4 – − 0.060 − 112.0 –
j 0.180 0.492 173.6 2.1 − 0.299 − 266.2 5.5
k 0.010 0.353 3427.5 – − 0.445 − 4545.5 –
[79] 2018 R1234ze B 1 a Fig. 11b 0.242 0.287 18.6 – 0.281 16.1 –
b 0.238 0.287 20.6 8.5 0.281 18.1 11.3
c 0.239 0.095 − 60.1 – 0.087 − 63.7 –
d 0.24 0.045 − 81.4 – 0.036 − 84.8 –
e 0.166 − 0.112 − 167.7 – − 0.122 − 173.4 –
f 0.149 − 0.158 − 206.0 – − 0.166 − 211.6 –
g 0.062 (*) (*) – (*) (*) –
2 a Fig. 12b 0.408 0.394 − 3.3 – 0.391 − 4.2 –
b 0.407 0.394 − 3.1 19.5 0.391 − 4 28.4
c 0.408 0.394 − 3.3 – 0.391 − 4.2 –
d 0.339 0.394 16.3 – 0.391 15.2 –
e 0.291 − 0.029 − 110.1 – − 0.032 − 110.9 –
f 0.135 − 0.132 − 197.8 – − 0.129 − 195.8 –
g 0.172 − 0.15 − 187.4 – − 0.146 − 184.9 –
h 0.097 − 0.212 − 318.9 – − 0.205 − 311.2 –
o 0.049 − 0.262 − 635 – − 0.251 − 611.5 –
[9] 2018 R134a C 1 a Fig. 13b 0.338 0.391 15.7 6.3 0.371 9.7 4
b 0.335 0.391 16.8 16 0.371 10.8 14.2
c 0.337 0.391 15.9 4.8 0.371 10 15.2
d 0.333 0.391 17.5 3.8 0.371 11.5 16.6
e 0.336 0.391 16.2 11.5 0.371 10.3 26.6
f 0.339 0.391 15.3 12.8 0.371 9.4 29.1
g 0.334 0.391 17.1 13.9 0.371 11.1 31.3
h 0.329 0.391 18.9 14.4 0.371 12.8 22.1
i 0.313 0.391 24.8 17.3 0.371 18.4 22.9
j 0.3 0.391 30.4 19.8 0.371 23.7 22.2
k 0.292 0.391 33.8 20.9 0.371 26.9 21.5
l 0.279 0.391 40.3 8.4 0.371 33.1 8.9
m 0.264 0.391 48.3 7.9 0.371 40.7 22.9
n 0.253 0.391 54.3 6.7 0.371 46.4 23.6
o 0.249 0.391 56.8 7.3 0.371 48.8 23.8
p 0.252 0.323 28.1 4.6 0.371 47.1 7.4
q 0.207 0.246 18.7 4.5 0.307 48.1 8.5
r 0.164 0.173 5.6 3.4 0.216 31.7 1.7
s 0.123 0.107 − 12.7 4.2 0.141 14.8 4.2
2 a Fig. 14b 0.405 0.452 11.6 – 0.447 10.4 –
b 0.404 0.452 12.1 – 0.447 10.8 –
c 0.401 0.452 12.7 – 0.447 11.5 –
d 0.393 0.452 15.2 – 0.447 13.9 –
e 0.377 0.452 20.1 – 0.447 18.8 –
f 0.355 0.452 27.4 – 0.447 26 –
g 0.343 0.452 32 – 0.447 30.5 –
h 0.323 0.452 40.1 – 0.447 38.5 –
i 0.305 0.452 48.5 – 0.447 46.9 –
j 0.3 0.452 50.7 – 0.447 49 –
k 0.293 0.37 26.3 – 0.417 42.3 –
l 0.239 0.287 20 – 0.293 22.5 –
m 0.201 0.212 5.6 – 0.205 2.2 –
n 0.146 0.143 − 1.7 – 0.13 − 10.3 –
o 0.078 0.081 4.4 – 0.063 − 18.1 –
3 a Fig. 15b 0.497 0.545 9.6 4.7 0.54 8.6 6
b 0.5 0.545 9.1 5.7 0.54 8.1 1.2
c 0.495 0.545 10.2 10.8 0.54 9.1 4.2
d 0.489 0.545 11.6 12.1 0.54 10.5 5.7
e 0.448 0.545 21.7 21.9 0.54 20.5 16.1
f 0.404 0.545 34.8 25.9 0.54 33.6 20.4
g 0.396 0.545 37.6 26.3 0.54 36.3 20.9
h 0.39 0.545 39.8 26.6 0.54 38.5 21.2
i 0.326 0.507 55.6 17.2 0.54 65.7 26.2
j 0.302 0.4 32.7 2.5 0.421 39.4 5.5
k 0.219 0.234 6.8 7.1 0.228 4.2 6
l 0.099 0.097 − 1.4 5 0.081 − 17.8 5.8
(continued on next page)

25
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 11 (continued )
Benchmark Year Fluid Geom# Case Run Figure ωEXP k-ω SST k-ε RNG

ωCFD δ [%] θ [%] ωCFD δ [%] θ [%]

4 a Fig. 16b 0.565 0.618 9.5 – 0.613 8.6 –

b 0.562 0.618 10.1 – 0.613 9.1 –
c 0.564 0.618 9.6 – 0.613 8.7 –
d 0.562 0.618 10.1 – 0.613 9.1 –
e 0.554 0.618 11.6 – 0.613 10.7 –
f 0.535 0.618 15.5 – 0.613 14.6 –
g 0.49 0.618 26.2 – 0.613 25.2 –
h 0.467 0.618 32.5 – 0.613 31.3 –
i 0.438 0.618 41.1 – 0.613 39.9 –
j 0.363 0.488 34.5 – 0.53 46.1 –
k 0.251 0.314 25 – 0.316 25.9 –
[79] 2017 R141b D 1 a Fig. 17b 0.249 0.24 − 3.6 – 0.241 − 3.3 –
b 0.246 0.24 − 2.4 – 0.241 − 2.1 –
c 0.247 0.24 − 2.8 – 0.241 − 2.5 –
d 0.246 0.24 − 2.4 – 0.241 − 2.1 –
e 0.245 0.24 − 2 – 0.241 − 1.7 –
f 0.247 0.24 − 2.8 – 0.241 − 2.5 –
g 0.239 0.24 0.4 – 0.229 − 4.3 –
h 0.15 0.173 15.4 – − 0.002 − 101.5 –
i 0.011 − 0.1 − 1006.7 – − 0.282 − 2668 –
2 a Fig. 18b 0.123 0.165 34 – 0.165 34 –
b 0.121 0.165 36.2 – 0.165 36.3 –
c 0.125 0.165 31.8 – 0.165 31.9 –
d 0.125 0.165 31.8 – 0.165 31.9 –
e 0.119 0.165 38.5 – 0.165 38.6 –
f 0.123 0.165 34 – 0.165 34 –
g 0.116 0.165 41.9 – − 0.006 − 105.6 –
h 0.068 0.129 90.2 – − 0.072 − 206 –
i 0.009 0.003 − 72 – − 0.193 − 2247.7 –
3 a Fig. 19b 0.065 0.089 36.3 – 0.091 39.4 –
b 0.061 0.089 45.3 – 0.091 48.5 –
c 0.066 0.089 34.2 – 0.091 37.3 –
d 0.063 0.089 40.6 – 0.091 43.8 –
e 0.068 0.089 30.3 – 0.091 33.2 –
f 0.063 0.089 40.6 – − 0.043 − 167.7 –
g 0.008 0.001 − 93 – − 0.186 − 2425.8 –
[80] 2016 R134a E 1 a Fig. 20b 0.244 0.348 42.7 – 0.342 40.2 –
b 0.244 0.348 42.6 – 0.342 40.2 –
c 0.232 0.339 46.2 – 0.342 47.5 –
d 0.116 0.196 68.7 – 0.258 122.3 –
e 0.011 0.051 366.2 – 0.063 475.2 –
[15] 2015 R134a F 1 a Fig. 21 0.592 0.695 17.4 4.8 0.713 20.4 5.8
2 a 0.361 0.421 16.6 7.9 0.442 22.3 11.8
3 a 0.433 0.495 14.3 9.1 0.513 18.6 9.9
[81] 2015 R245fa G 1 a Fig. 22b 0.32 0.348 8.8 – 0.34 6.3 –
b 0.32 0.348 8.8 – 0.34 6.3 –
c 0.31 0.348 12.3 – 0.34 9.7 –
d 0.25 0.348 39.2 – 0.34 36 –
e 0.16 0.339 112.1 – 0.34 112.2 –
f 0.02 0.23 1049 – 0.263 1213.7 –
2 a Fig. 23b 0.42 0.445 5.9 – 0.436 3.8 –
b 0.42 0.445 5.9 – 0.436 3.8 –
c 0.4 0.445 11.2 – 0.436 9 –
d 0.29 0.444 53.3 – 0.436 50.4 –
e 0.23 0.416 80.9 – 0.436 89.7 –
f 0.02 0.177 783.2 – 0.175 773.1 –
3 a Fig. 24b 0.55 0.569 3.5 – 0.561 2 –
b 0.55 0.569 3.5 – 0.561 2 –
c 0.48 0.569 18.6 – 0.561 16.9 –
d 0.41 0.569 38.8 – 0.561 36.9 –
e 0.36 0.542 50.6 – 0.561 55.9 –
f 0.02 0.16 698.6 – 0.148 640.6 –
H 1 a Fig. 25b 0.41 0.478 16.7 – 0.469 14.4 –
b 0.41 0.478 16.7 – 0.469 14.4 –
c 0.34 0.478 40.7 – 0.469 38 –
d 0.3 0.447 49 – 0.469 56.4 –
e 0.25 0.418 67.3 – 0.469 87.6 –
f 0.01 0.233 2232.2 – 0.285 2750.7 –
2 a Fig. 26b 0.54 0.593 9.9 – 0.599 10.9 –
b 0.54 0.593 9.9 – 0.599 10.9 –
c 0.48 0.593 23.6 – 0.599 24.8 –
d 0.42 0.561 33.5 – 0.599 42.6 –
e 0.38 0.534 40.5 – 0.599 57.6 –
f 0.02 0.183 814.6 – 0.197 884.4 –
(continued on next page)

26
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 11 (continued )
Benchmark Year Fluid Geom# Case Run Figure ωEXP k-ω SST k-ε RNG

ωCFD δ [%] θ [%] ωCFD δ [%] θ [%]

3 a Fig. 27b 0.68 0.735 8.1 – 0.729 7.3 –

b 0.68 0.735 8.1 – 0.729 7.3 –
c 0.59 0.735 24.5 – 0.729 23.6 –
d 0.54 0.735 36.1 – 0.729 35.1 –
e 0.47 0.65 38.3 – 0.729 55.2 –
f 0.02 0.125 526.4 – 0.118 487.7 –
[22] 2014 R134a I 1 a Fig. 28b 0.465 0.471 1.2 – 0.465 0 –
b 0.459 0.475 3.6 – 0.465 1.3 –
c 0.459 0.475 3.6 – 0.464 1.1 –
d 0.416 0.366 − 11.9 – 0.328 − 21.2 –
e 0.292 0.257 − 12 – 0.23 − 21.2 –
f 0.094 0.146 55.5 – 0.12 27.4 –
g 0.046 0.037 − 19.5 – 0.008 − 82.9 –
J 1 a Fig. 29b 0.466 0.512 10 – 0.496 6.4 –
b 0.479 0.512 7 – 0.496 3.5 –
c 0.481 0.512 6.5 – 0.496 3.1 –
d 0.476 0.512 7.6 – 0.462 − 3 –
e 0.359 0.365 1.6 – 0.295 − 17.8 –
f 0.192 0.152 − 21.1 – 0.112 − 41.4 –
g 0.024 0.046 91.9 – 0.013 − 46 –
K 1 a Fig. 30b 0.491 0.597 21.7 – 0.571 16.2 –
b 0.5 0.598 19.5 – 0.571 14.1 –
c 0.504 0.597 18.6 – 0.571 13.2 –
d 0.5 0.598 19.5 – 0.571 14.1 –
e 0.504 0.598 18.6 – 0.571 13.2 –
f 0.416 0.597 43.6 – 0.424 1.9 –
g 0.24 0.121 − 49.7 – 0.079 − 67.2 –
h 0.068 − 0.1 − 247.64 – − 0.112 − 264.71 –
[52] 2014 Air L 1 a Fig. 31b 0.459 0.454 − 1.2 19.1 0.454 − 1.1 17.2
b 0.46 0.454 − 1.2 18.6 0.456 − 0.8 19.5
c 0.46 0.454 − 1.2 16.9 0.456 − 0.8 19.8
d 0.462 0.454 − 1.8 16 0.456 − 1.4 19.7
e 0.462 0.453 − 2 18.5 0.456 − 1.4 19.6
f 0.452 0.454 0.5 17.4 0.455 0.8 18.5
g 0.36 0.436 21.3 13.6 0.449 24.9 19.6
h 0.27 0.36 33.3 8.6 0.347 28.7 8
i 0.176 0.233 32.6 4 0.222 26.2 4.4
[14] 2007 Steam M 1 a Fig. 32b 0.397 0.431 8.7 9.8 0.47 18.4 25.7
b 0.401 0.431 7.6 12 0.47 17.2 30.8
c 0.399 0.431 8.1 14.3 0.47 17.8 8.7
d 0.403 0.431 7.1 13.3 0.304 − 24.6 42.5
e 0.4 0.432 7.9 – 0.246 − 38.4 –
f 0.398 0.43 8.2 – 0.187 − 53.1 –
g 0.4 0.41 2.5 – 0.124 − 68.9 –
h 0.4 0.368 − 8 – 0.062 − 84.5 –
i 0.399 0.312 − 21.7 – − 0.001 − 100.2 –
j 0.36 0.243 –32.4 – − 0.065 − 118.2 –
k 0.293 0.173 − 41.1 – − 0.126 − 142.9 –
j 0.198 0.101 − 49 – − 0.184 − 193.1 –
2 a Fig. 33b 0.528 0.592 12.1 6.3 0.622 17.9 35.7
b 0.53 0.592 11.7 – 0.494 − 6.8 –
c 0.533 0.592 11.1 – 0.352 –33.9 –
d 0.531 0.59 11.2 – 0.187 − 64.7 –
e 0.513 0.559 9.1 – 0.103 − 79.9 –
f 0.471 0.484 2.7 – 0.016 − 96.6 –
g 0.345 0.375 8.8 – − 0.075 − 121.6 –
h 0.292 0.251 − 14 – − 0.158 − 154.1 –
i 0.167 0.132 − 21.1 – − 0.237 − 242 –
3 a Fig. 34b 0.31 0.313 1 9.6 0.3 − 3.1 22.3
b 0.31 0.313 1 – 0.3 − 3.1 –
c 0.31 0.313 1 – 0.3 − 3.1 –
d 0.31 0.283 − 8.7 – 0.05 − 83.9 –
e 0.309 0.252 − 18.6 – − 0.01 − 103.3 –
f 0.31 0.204 − 34.1 – − 0.068 − 122.1 –
g 0.31 0.149 − 52 – − 0.126 − 140.7 –
h 0.23 0.091 − 60.3 – − 0.192 − 183.3 –
i 0.1 0.037 − 63.3 – − 0.252 − 352.4 –
[41] 1999 R141b N 1 a Fig. 35 0.392 0.417 6.3 – 0.415 5.9 –
2 a 0.312 0.351 12.6 – 0.348 11.8 –
3 a 0.309 0.264 13.6 – 0.304 11.9 –
O 1 a 0.613 0.62 1 – 0.617 0.6 –
2 a 0.479 0.503 5.1 – 0.501 4.6 –
3 a 0.403 0.421 4.4 – 0.42 4.1 –
4 a 0.35 0.375 7 – 0.371 6 –
(continued on next page)

27
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 11 (continued )
Benchmark Year Fluid Geom# Case Run Figure ωEXP k-ω SST k-ε RNG

ωCFD δ [%] θ [%] ωCFD δ [%] θ [%]

5 a 0.439 0.487 10.8 – 0.48 9.3 –

6 a 0.388 0.407 4.8 – 0.398 2.6 –
7 a 0.304 0.356 17 – 0.342 12.4 –
8 a 0.255 0.322 26.2 – 0.307 20.3 –

(*) convergence not reached

stable behaviour and lower computational time. Accordingly, the up­ a reference cell size that obtained as a fraction of the mixing throat size
coming mesh screening has been performed with the PB solver; for the (i.e., 70 cells) and build the entire mesh satisfying a reasonable
sake of completeness, the corresponding DB mesh study has been pre­ maximum aspect ratio of 3. Such mesh is further refined during simu­
sented in Appendix B. lation by using a Mach gradient criterion to capture the thin shock layers
and the region around primary jet-core where first mixing with sec­
4.2. Mesh study ondary flow occurs (Fig. 9). Based on the obtained results, nozzle and
diffuser wall refinement influence is relatively negligible. The same
The outcomes of the mesh screening (k-ω SST as turbulence model) approach was suggested by Besagni et Inzoli [7] who confirmed the non-
are presented in Table 9 and in Figs. 6–8: for the different meshes, it dependencies of k-ω SST with low Reynolds meshes. In the forthcoming
summarizes the number of elements, ωCFD as well as Eqs. (6)–(8) out­ section, even though mesh β2 gave a reasonable converged solution,
comes; conversely Figs. 6–8 presents the local axial profiles. Concerning especially employing solution-based adaption, γ2 mesh criteria (Mach
the effect of the number of throat elements (Fig. 6), expressed by the first gradient-based refinement) is adopted; indeed, this mesh sensitivity
mesh code index (Table 3), it is noted that β mesh (viz., 70 throat cells) study has been performed on two cases, and fair confidence is needed.
with a constant aspect ratio equal to 3 seems to offer a sufficient grid
resolution to capture the axial Mach distribution as well as the wall 4.3. Turbulence model comparison and extended validation
pressure profiles, having < 5.0% deviations from the reference case (as
defined in Section 2.3). This outcome is supported by the local profiles This section presents a comprehensive validation of the proposed
presented in Fig. 6: despite the coarser mesh (α2) is able to predicts the CFD approach, based on simulating all cases listed in Table 7. For all
location of the shockwaves, it lacks in precision when predicting the cases, both k-ω SST and k-ε RNG turbulence models have been applied.
second shock train, especially for case M3a. Besides, when looking at the The outcome of the extended validation is presented in Table 11 and
pressure profile predictions, passing from mesh α2 towards mesh β2 Table 12. The former compares numerical and experimental ω, δ value
induces a considerable improvement, which is lower when passing from (Eq. (3)) and, when available, local pressure comparison is quantified by
mesh β2 towards mesh γ2. Concerning the effect of the aspect ratio using θ (Eq. (4)). The latter compares numerical and experimental Tcrit
(Fig. 7), it is noted that mesh β1 and mesh β2 provide close results (or Pcrit)4 as well as ξ (Eq. (5)). The ejector operating curves and the
(difference in ωCFD < 0.5%), suggesting that as long as the cell size is Mach contours, for the different benchmarks are presented in
small enough, there is no need to excessively refine the nozzle exit re­ Figs. 10–35 (which have been recalled in Table 11); in case other CFD
gion. Conversely, increasing the aspect ratio values above 3, negatively studies (Table 1 and Table 2) have simulated such benchmarks, their
affects the shock waves predictions as the low axial resolution does not outcomes are included as well in the corresponding figures. Also, for the
allow to capture the shock wave phenomena. This outcome is in different cases, the filled contours of Mach number for the on-design
agreement with AR = 3.75 suggested by Ramesh et al. [42]. Concerning operation condition is presented for both turbulence models. It is
wall discretization, the refinement at the diffuser section for case F1a worth noting that we have chosen to confront the computed fluid-
does not largely improve the accuracy of the CFD simulation, obtaining dynamics by the two models in the double-choking operating mode
very similar results to the standard mesh; conversely, the refinement at and not at the critical condition thereby comparing the two Mach con­
the nozzle section slightly affects the entrainment ratio prediction but tours for the same outlet conditions (which is not always the same when
largely increases the number of elements. considering the critical point). Finally, the comparison between nu­
The mesh independence can also be evaluated by the grid conver­ merical and experimental local pressure profiles (determining the θ
gence index (GCI), which has been computed on meshes α2, β2, γ2, values) are provided in Appendix C.
considered as the “coarse”, “medium” and “fine”, respectively. The pro­ Concerning benchmark in ref. [10] (Geom#A, Table 6 and Table 7 -
cedure, based on Richardson extrapolation described by Roache [84], Fig. 10), it is noted that experimental data have been correctly predicted
has been performed on both cases for primary and secondary mass flow by k-ω SST, which gave satisfactory results in terms of on-design
rates, evaluating the apparent order of convergence p and the error entrainment ratio (δ = 11.8%) and critical temperature (ξ = 0.7%) pre­
band. Coupling the outcomes, entrainment ratio ωmedium-fineext is diction. Conversely, k-ε RNG has shown poor performances, exhibiting
computed along with the numerical range of uncertainty (Table 10): this errors in on-design operation > 30% as well as largely under predicting
result supports above-mentioned discussion, based on Table 9. the critical condition (ξ = 6.9%). This outcome is supported by the local
Finally, the effects of adaptive mesh refinement are shown for both pressure profiles (Appendix C), were k-ω SST has been able to correctly
benchmarks using mesh β2 as reference case (Fig. 8). A double-cycle capture the local phenomena: the proposed model is characterized
refinement based on Mach gradient allows to locally improve the θ = 8.2% (k-ω SST) and 13.6% (k-ε RNG), which are close to the values
detail of the predicted shockwave and it has been found to improve the obtained by to Han et al. [10] (Table 1 and Table 2), where k-ω SST
entrainment ratio prediction; the deviation of ω from the reference case showed θ = 6.5%, whereas k-ε RNG showed θ = 11.1%.
is reduced (from 2.4% to 0.3% for F1a; from 2.9% to 1.8% for M3a). The benchmark using R1234ze [79] (Geom#B, Table 6, Table 7,
Accordingly, a closer agreement with the extrapolated ω is reached Fig. 11 and Fig. 12) have been reasonably well predicted in the on-
(Table 10). Thus, the adaptive mesh refinement, which application is design mode by both turbulence models (δ < 20%); conversely, in the
widely employed (Table 1), represents a valuable strategy to locally
increase the spatial discretization where needed, optimizing the solution
accuracy and the number of elements. To sum up, the suggested 4
these values are computed as intersection of the on-design and off-design
approach to obtain a mesh for ejector simulation is as follows: start from ejector operating curves.

28
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 12
Numerical results: critical condition evaluation.
Benchmark Year Fluid Geom# Case Model Tcrit [◦ C] ξ [%]

[10] 2019 Steam A 1 exp 34.2 /

k-ω SST 34 − 0.7
k-ε RNG 31.8 − 6.9
[79] 2018 R1234ze B 1 exp 25.7 /
k-ω SST 23.7 − 7.8
k-ε RNG 23.7 − 7.8
2 exp 24.5 /
k-ω SST 24.9 1.6
k-ε RNG 24.9 1.6
[9] 2018 R134a C 1 exp 31.3 (*) /
k-ω SST 29.9 − 4.4
k-ε RNG 31.4 0.3
2 exp 32.2 (*) /
k-ω SST 31.7 − 1.6
k-ε RNG 32 − 0.8
3 exp 32.5 (*) /
k-ω SST 32.8 0.8
k-ε RNG 32.9 1
4 exp 33.7 (*) /
k-ω SST 33.7 − 0.2
k-ε RNG 34 0.7
[79] 2017 R141b D 1 exp 33.5 /
k-ω SST 34.1 1.7
k-ε RNG 33.5 0
2 exp 33.2 /
k-ω SST 33.4 0.7
k-ε RNG 32.5 − 2.2
3 exp 32.4 /
k-ω SST 32.4 − 0.1
k-ε RNG 31.9 − 1.6
[80] 2016 R134a E 1 exp 31.6 /
k-ω SST 31.7 0.3
k-ε RNG 32.8 3.8
[81] 2015 R245fa G 1 exp 33 /
k-ω SST 33.9 3
k-ε RNG 34.4 4.3
2 exp 34.3 /
k-ω SST 35.2 2.6
k-ε RNG 35.4 3.2
3 exp 35.7 /
k-ω SST 36.7 2.7
k-ε RNG 36.8 3.1
H 1 exp 31.5 /
k-ω SST 31.9 1.1
k-ε RNG 32.6 3.5
2 exp 33.1 /
k-ω SST 33.5 1.1
k-ε RNG 34.1 3
3 exp 34.6 /
k-ω SST 35.2 1.8
k-ε RNG 35 2.8
[22] 2014 R134a I 1 exp 28.9 /
k-ω SST 28.4 − 1.7
k-ε RNG 28.2 − 2.4
J 1 exp 29.2 /
k-ω SST 29 − 0.9
k-ε RNG 28.5 − 2.5
K 1 exp 30.2 /
k-ω SST 29.8 − 1.5
k-ε RNG 30.1 − 0.4
[52] 2014 Air L 1 exp 600 (**) /
k-ω SST 637.2 (**) 6.2
k-ε RNG 639.6 (**) 6.6
[14] 2007 Steam M 1 exp 33.1 /
k-ω SST 32.1 − 2.9
k-ε RNG 30.1 − 9.1
2 exp 28.1 /
k-ω SST 28 − 0.3
k-ε RNG 25.1 − 10.7
3 exp 32.3 /
k-ω SST 30.9 − 4.4
k-ε RNG 29.5 − 8.7

(*) Experimental second plateau is considered as the end of the on-design operation.
(**) outlet conditions expressed as pressure [kPa] instead of temperature because air is the working fluid (absence of saturation conditions).

29
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 10. Comparison between experimental and numerical ejector operating curves, Case A1.

off-design mode a decrease in the performances have been detected. In Coming to the R141b experimental benchmark of Thongtip et al.
the latter case, a sharp decrease in the entrained secondary mass flow [80] (Geom#D, Table 6 and Table 7), it is noted that case D1 is char­
rate has led to significant differences compared with the experimental acterized by δ < 5% for both turbulence models. Conversely, case D2 and
results. Similar behaviour was observed by Kracik et al. [36] (3- case D3 are characterized by δ > 30%. This issue could be explained by
Dimensional approach and employing k-ω SST – see Table 1 and Table 2) the low secondary mass flow rate of case D2 and case D3 that magnifies
who concluded that the on-design operation is fairly predicted by the the relative error of the entrainment ratio. Comparing the two turbu­
numerical model (δ = 20%), whereas the off-design approach had some lence models, their on-design performances are very close, whereas in
issues. Based on the results of the present simulations, it can be stated the off-design operation mode a better accuracy of k-ω SST is exhibited. In
that both k-ω SST and k-ε RNG computed close global performance pa­ such case, k-ω SST still predicts the single chocked operating mode
rameters (ω and Tcrit), even if with a different local fluid dynamics, in whereas k-ε RNG fails and predicts the backflow regime. Looking at the
agreement with the previous outcome of Hemidi et al. [12]. In partic­ Mach contours comparison (Fig. 17b, Fig. 18b and Fig. 19b) no signifi­
ular, k-ω SST predicts a more detailed second shock train upstream than cant difference is observed, as both models accurately predict the
the normal shock predicted by k-ε RNG (Fig. 11b and Fig. 12b). For this change in primary nozzle flow regime: as the primary conditions are
reason, k-ω SST detects a wall pressure profile consistent with the fixed and the secondary pressure reduces, the nozzle in case D1 operates
experimental one: k-ω SST is characterized by θ = 14.0% against in over expanded regime, exhibiting a succession of oblique shocks at
θ = 19.9% of k-ε RNG. the nozzle exit owing to the lower pressure compared to the secondary
The ω values of the R134a experimental benchmark proposed by flow. On the other hand, case D2 operating conditions cause the nozzle
Poirier at al. [9] (Geom#C, Table 6 and Table 7) have been systemati­ to be close to adapted conditions, causing the jet core to flow almost
cally overestimated by the CFD model with a slightly better performance straight and presenting weak shock waves. Thus, a further decrease of
of k-ε RNG (δ = 9.3%) compared with k-ω SST (δ = 12.1%). It is worth the secondary pressure leads the nozzle in case D3 to operate in the
mentioning that the ejector operating curve of this benchmark was under-expanded regime, continuing to expand in the very first part of
characterized by a two-plateau behaviour: according to the authors, this the mixing chamber [85].
behaviour might be related to an improper NXP. Considering the The comparison of numerical performances with the benchmark
experimental second plateau as the end of the on-design operation, the provided by Yan et al. [81] (Geom#E Table 6 and Table 7) showed poor
CFD model correctly predicts the critical condition (k-ω SST: ξ = 1.8%; k- agreement with experimental data in the on-design operation mode,
ε RNG: ξ = 0.7%). Conversely, k-ω SST obtained a better agreement with registering high deviations on entrainment (δ > 40%). Unfortunately, no
the local profiles (θ = 11.4%) compared with k-ε RNG (θ = 15.0%). As other CFD studies have been performed on this benchmark; hence,
noted for Geom#B, k-ω SST predicts an oblique shock wave followed by comparisons with other numerical results are unavailable. One of the
a more detailed second shock train near the diffuser entrance, instead of possible reasons of the reported poor performance could be related to
a single normal shock wave predicted by k-ε RNG (see Fig. 13b, Fig. 14b, the wrong suction chamber geometry and/or the diffuser exit di­
Fig. 15b and Fig. 16b). mensions which have been assumed owing to the missing details in the

30
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 11. Comparison between experimental and numerical ejector operating curves, Case B1.

Fig. 12. Comparison between experimental and numerical ejector operating curves, Case B2.

31
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 13. Comparison between experimental and numerical ejector operating curves, Case C1.

Fig. 14. Comparison between experimental and numerical ejector operating curves, Case C2.

32
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 15. Comparison between experimental and numerical ejector operating curves, Case C3.

Fig. 16. Comparison between experimental and numerical ejector operating curves, Case C4.

33
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 17. Comparison between experimental and numerical ejector operating curves, Case D1.

Fig. 18. Comparison between experimental and numerical ejector operating curves, Case D2.

34
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 19. Comparison between experimental and numerical ejector operating curves, Case D3.

Fig. 20. Comparison between experimental and numerical ejector operating curves, Case E1.

35
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 21. Mach contours, case F1, F2, F3.

36
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 22. Comparison between experimental and numerical ejector operating curves, Case G1.

Fig. 23. Comparison between experimental and numerical ejector operating curves, Case G2.

37
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 24. Comparison between experimental and numerical ejector operating curves, Case G3.

Fig. 25. Comparison between experimental and numerical ejector operating curves, Case H1.

38
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 26. Comparison between experimental and numerical ejector operating curves, Case H2.

Fig. 27. Comparison between experimental and numerical ejector operating curves, Case H3.

39
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 28. Comparison between experimental and numerical ejector operating curves, Case I1.

Fig. 29. Comparison between experimental and numerical ejector operating curves, Case J1.

40
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 30. Comparison between experimental and numerical ejector operating curves, Case K1.

Fig. 31. Comparison between experimental and numerical ejector operating curves, Case L1.

41
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 32. Comparison between experimental and numerical ejector operating curves, Case M1.

Fig. 33. Comparison between experimental and numerical ejector operating curves, Case M2.

42
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 34. Comparison between experimental and numerical ejector operating curves, Case M3.

original paper. Conversely, regarding the critical point, a satisfactory Geom#H (Geom#G has a constant area mixing chamber after the suc­
agreement is achieved, with slight better performances of k-ω SST tion chamber, whereas Geom#H has tapered section before the constant
(ξ = 0.3%) compared to the k-ε RNG (ξ = 3.8%) area mixing chamber, Fig. 3) for fixed inlet conditions, increases the
The numerical simulations conducted on the benchmark provided by entrained secondary mass flow rate, hence the entrainment ratio, but
Del Valle et al. [15] (Geom#F, Table 6 and Table 7) assessed the better decreases the critical pressure.
performance of k-ω SST over k-ε RNG in predicting the double chocking The same ejector design of benchmark [15] has been used for the
operation mode. Besides, the present outcomes are very close to the experimental study in ref. [22] (Geom#I, #J and #k, Table 6 and
numerical results provided by Del Valle et al. [15]. The present nu­ Table 7) by the same authors, also including an additional NXP study
merical model obtained δ = 16.1% (k-ω SST) and δ = 20.2% (k-ε RNG), and obtaining the whole ejector operating curve. The CFD model has
whereas Del Valle et al. obtained δ = 15.2% − 21.2%, respectively. It is proved to successfully predict the on-design mode with an average error
worth noting that the current CFD approach employed a PB solver, δ = 11.7% (k-ω SST) and δ = 7.5% (k-ε RNG). The critical point is also
whereas Del Valle et al. [15] used a DP; the comparable outcomes, correctly computed by both turbulence model (k-ω SST: ξ = 1.4%; k-ε
hence, suggest once again the fair independence of the global quantities RNG: ξ = 1.8%).
from the employed solver, as stated in Section 4.1. Similarly, the local The benchmark provided by Chong et al. [52] (Geom#L, Table 6 and
pressure profile computed by k-ω SST (θ = 7.3%) is more consistent with Table 7) has been very well predicted by the CFD model which showed
experimental pressure measurements compared with k-ε RNG very good agreement with the on-design operating mode of the tested
(θ = 9.2%). ejector (δ < 2% for both turbulence models). Conversely, both k-ω SST
The experimental data provided by Shestopalov et al. [82] (Geom#G and k-ε RNG failed in predicting the critical pressure , with ξ > 6%.
and H, Table 6 and Table 7), describing the ejector operating curve for Looking at the predicted flow phenomena within the ejector (Fig. 31b),
two R245fa ejector designs, have been well predicted by CFD simula­ the two models computed rather different flow structures. Comparing
tions with average on-design entrainment ratio deviation δ = 8.8% (k-ω the wall pressure with experimental data, it can be concluded that k-ω
SST) and δ = 7.5% (k-ε RNG). The two turbulence models resulted in SST is more consistent with local experimental data (θ = 14.7%)
similar Mach contours (Fig. 22b, Fig. 23b, Fig. 24b, Fig. 25b, Fig. 26b compared with k-ε RNG (θ = 16.2%). Similar outcomes have been ob­
and Fig. 27b) as both k-ε RNG and k-ω SST seem to predict the second tained by Chong et al. [52], who employed a k-ε RNG with EWT (δ = 6.5,
shock train with a series of oblique shockwaves. This outcome is θ = 13.2%, ξ = 5.0%).
different from the previously observed results, where k-ε RNG predicted The steam ejector described by Sriveerakul et al. [14] (Geom#M,
a normal shockwave (Geom#B and Geom#C - Table 6 and Table 7). Table 6 and Table 7) has been already widely used to conduct CFD model
Concerning the critical point prediction, k-ω SST (ξ = 2.1%) is charac­ validations. In the present study, k-ω SST obtained good agreement with
terized by better accuracy compared with k-ε RNG (ξ = 3.3%). It is worth experimental data with an average prediction error of the on-design
noting that changing the ejector design from geometry Geom#G to mode equal to δ = 7.3% and a maximum δ = 12.1%. Conversely, k-ε

43
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 35. Mach contours, case N1, N2, N3, O1, O2, O3, O4, O5, O6, O7, O8.

44
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 35. (continued).

45
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 35. (continued).

46
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 35. (continued).

RNG predicted the double-chocking mode with an average error prediction capability and the critical point prediction capability are
δ = 13.1% (and maximum δ = 18.4% in case M2). Based on the previous presented. Every benchmark is fairly predicted by the CFD approach
CFD studies which have simulated this benchmark, some uncertainties with an on-design δ < 20%, except for Benchmark [81] and [80] which
in the critical point prediction are observed: Sriveerakul et al. [14] ob­ suffered from the above-discussed issues. Regarding the influence of the
tained ξ = 7% (under predicting the experimental critical condition), turbulence model, k-ω SST showed better agreement compared with k-ε
Yang et al. [20] under predicted the critical point (ξ = 9%), while Wang RNG in terms of entrainment ratio prediction in both on-design and off-
et al. [27] under predicted the critical point (ξ = 6.4%); it should be design modes. The average deviation δ for k-ω SST is 14.2% in double-
noted that all these CFD approaches employed the two-equation k-ε chocking mode and 62.8% in off-design operation; conversely, the
model or one of its modifications (k-ε Realizable and k-ε RNG). The average deviation δ for is 16.1% in double-chocking mode and 102.9%
proposed CFD approach suffered the same problem using k-ε RNG as in off-design operation. The great difference between the on-design and
turbulence model; in this case, the average registered ξ > 9%. the off-design performance can be attributed to two main reasons: (i)
Conversely, k-ω SST achieved ξ = 2.5%. The better performances of k-ω while in the double-choking mode the secondary flow entrainment is not
SST are supported by the extended local validation where k-ω SST affected by the outlet pressure because of the second shock train
showed fair agreement (θ = 10.9%) with experimental data, while k-ε occurring at the end of the mixing chamber which prevent the down­
RNG poorly performed with θ = 27.6% (Appendix C). stream information to travel upwards, in the subcritical operation
The main problem with the well-known dataset provided by Huang instead, due to higher outlet pressure, the second shock is no longer
et al. [41] (Geom#N, #O, Table 6 and Table 7) is the lack of geometrical occurring. In such a condition, while the primary flow is not affected, the
information regarding the ejector; as detailed in Table 6, some missing entrainment process and mixing are highly influenced by the outlet flow
data are obtained from Scott et al. [71]. Although the original experi­ reversing in the mixing chamber, with the result that the secondary mass
mental benchmark is obtained under the double-chocking mode, some flow rate (which affect the entrainment ratio) is predicted with less
of the tested cases resulted in single-chocking operation adopting the accuracy in average. (ii) : in the off-design operating mode the value of
author’s specified discharge pressure. To assess the CFD error in the on- the entrainment ratio decreases (compared with the on-design value)
design operating mode, hence, the back-pressure of the problematic which leads to a greater relative error (even when the absolute differ­
cases has been reduced until the double-chocking regime was obtained, ence between the predicted value and the experimental one is the same).
allowing a comparison between numerical and experimental results. It is The critical point is also better predicted by k-ω SST, having average
noted that the numerical approach can predict the ejector performance ξ = 2.3%, whereas k-ε RNG is characterized by ξ = 4.3%; this outcome is
with an average deviation of 9.9% adopting the k-ω SST, and 8.1% with important since it supports that the present CFD model can effectively
k-ε RNG. This result is slightly worst compared with previous CFD predict the transition point of its two operating regimes. Finally, the
studies had. Rusly et al. [73] obtained a δ = 3.9%, while Scott et al. [71] extended local validation (Appendix C) aimed to understand which
obtained δ = 4.2%. A more recent study of Mohamed et al. [40] is in model could better predict the local flow phenomena suggest k-ω SST as
better agreement with the present outcomes having δ = 9.4%. It should the most suitable model. In conclusion, the “baseline” CFD approach
be noted that the outdated experimental data are affected by a ± 5% which is simulated by k-ω SST model can predict a wide range of
error on mass flow rates calculation [86], which makes such differences boundary conditions on different supersonic ejector design operating
negligible. The Mach contours highlight how, for this benchmark, except with different working fluids. The average deviation with experimental
for the starting position of the second shock train, both models well data δ is equal to 14.2% (which decreases to 11.4% if not considering
described the oblique waves. benchmark [81], for which the important errors suggest a problem in the
To summarize, validation outcomes are listed in Table 13, which has geometrical or boundary input rather than CFD model failure). The
the same structure as Table 2, giving the possibility to compare the stability of the simulation and the rapidity reaching a converged solu­
present CFD model to other approaches. For all tested benchmarks data tion (granted by the pressure-based approach and a systematic mesh
regarding entrainment ratio, for both on and off-design mode, local generation criteria) is furthermore a very significant result which could

47
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Table 13
Model validation outcomes.
Benchmark Fluid Code k-ω SST k-ε RNG
name
δ θ ξ δ θ ξ

On-design Off-design (*) On-design Off-design (*)

mean max mean max mean max mean max mean max mean max mean max mean max
[%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%]

[10] Steam A 11.8 11.8 59.4 173.6 8.2 13.1 0.7 0.7 34.3 34.3 137.7 266.2 13.6 18.4 6.9 6.9
[79] R1234ze B 11 18.6 229.9 635.0 14 19.5 4.7 7.9 10.2 16.1 226.8 611.5 19.9 28.4 4.7 7.9
[9] R134a C 11.6 15.6 12.1 32.7 11.4 26.3 1.8 4.4 9.3 10.4 17.4 39.4 15 31.3 0.7 1
[80] R141b D 24.6 36.3 48.8 90.2 / / 0.8 1.7 25.6 39.4 158.4 206.0 / / 1.3 2.2
[81] R134a E 42.7 42.7 57.5 68.7 / / 0.3 0.3 40.2 40.2 84.9 122.3 / / 3.8 3.8
[15] R134a F 16.1 17.4 / / 7.3 9.1 / / 20.4 22.3 / / 9.2 11.8 / /
[82] R245fa G, H 8.8 16.7 42.8 112.1 / / 2.1 3.0 7.5 14.4 46.5 112.2 / / 3.3 4.3
[22] R134a I, J, K 11.7 21.7 55.4 247.6 / / 1.4 1.7 7.5 16.2 59.2 264.7 / / 1.8 2.5
[52] Air L 1.2 1.2 29.1 33.3 14.7 19.1 6.2 6.2 1.1 1.1 26.6 28.7 16.3 19.8 6.6 6.6
[14] Steam M 7.3 12.1 30.2 63.3 10.9 14.3 2.5 4.4 13.1 18.4 168.4 352.4 27.6 42.5 9.5 10.7
[41] R141b N, O 9.9 26.2 / / / / / / 8.1 20.3 / / / / / /
Average / / 14.2 20.0 62.8 161.8 11.1 16.9 2.3 3.4 16.1 21.2 102.9 222.6 16.9 25.4 4.3 5.1

(*) relative errors when experimental ω was <0.02 were not considered due to very high σ that distorted the overall off-design performance prediction

be useful for future numerical study on supersonic ejector component. of Ruangtrakoon et al. [19]. The over-expansion would cause a much
significant contraction of the primary jet core increasing the available
area for the secondary flow, which enhances the entrainment ratio. On
4.4. Comparison between the fluid dynamic phenomena in the different the other hand, an under-expanded jet would continue the expansion in
ejectors the mixing chamber reducing the entrained secondary mass flow rate
and hence, the ejector performance.
In this section, some comments regarding the computed fluid dy­
namic concerning the different geometries and boundary conditions will 5. Numerical assessment of ejector component efficiencies
be presented. First, it is noticed an essential difference in the supersonic
flow structure when considering a tapered mixing chamber geometry In this section, a broad assessment of the ejector component effi­
(Fig. 3a) or a constant diameter mixing chamber (Fig. 3b). It is in fact ciencies is provided (for a complete discussion about this topic the
observed that in long-tapered mixing chambers, the secondary flow is reader may refer to Varga et al. [87]). In particular, the efficiencies of
not reaching supersonic conditions till the last section of the mixing the primary nozzle, suction chamber, mixing chamber diffuser have
chamber, which is usually a constant area section. This is quite evident, been calculated from the numerical outcomes of every simulated case
for instance, in Fig. 17, Fig. 18, Fig. 19, Fig. 21, Fig. 28, Fig. 29, Fig. 30 and have been presented in Fig. 37, Fig. 38, Fig. 39, Fig. 40, respectively.
and less evident in Fig. 25, Fig. 26, Fig. 27. This leads to the formation of Besides, the overall ejector efficiency defined by Elbel [88] has also been
a normal shock-wave rather than a shock train entering the diffuser (this made available in Fig. 41. Each efficiency has been presented as a
feature is also often called second-shock or second-shock train to function of the primary, secondary and outlet pressure, as well as a
distinguish it from the primary diamond-shaped shock structure). In dimensionless parameter which is named compression ratio (CR) and
case of a constant diameter mixing chamber, the observed pattern is the defined in Eq. (9) as the ratio of the pressure lift produced by the ejector
presence of a detailed shock train as in Fig. 13, Fig. 14, Fig. 15, Fig. 16, and the available motive pressure difference. It should be noted that
Fig. 17 which however is not necessary present at the diffuser entrance owing to the better validation results, efficiencies have been evaluated
(Fig. 11, Fig. 12, Fig. 22, Fig. 23, Fig. 24, Fig. 35). Directly relating the by using the k-ω SST turbulence model. The herein efficiencies will
component geometry to the ejector performance is quite tricky but provide a useful guideline for anyone dealing with a lumped-parameter
generally, a tapered mixing chamber allows a greater cross-section area model of supersonic ejectors since it has been already highlighted how
for secondary flow entrainment which improves the entrainment pro­ their performance are greatly influenced by the assumed values. Primary
cess at the expanse of a lower critical temperature, as seen for Geom#G and secondary performances (Eq. (10) and Eq. (11), respectively) are
and Geom#H. As regards the boundary conditions, it has been observed defined as isentropic efficiency of the primary and secondary flow, as
how, regardless of the geometry, the ejector performance is greatly well as the diffuser efficiency (Eq. (12)). Mixing chamber efficiency (Eq.
influenced by the primary flow expansion. According to the ratio be­ (13)) is instead defined as momentum transfer efficiency Finally, the
tween the primary flow pressure at the exit of the nozzle and the suction overall ejector efficiency (Eq. (14)) is the ratio between the isentropic
pressure, the primary jet could be over-expanded, adapted or under- recovered compression energy in the secondary flow and the available
expanded. It is found that, generally, over-expanded jets registered theoretical energy in the isentropic expansion of the motive stream.
higher entrainment effect, which could be explained by the conclusion

Fig. 36. Definition of ejector sections.

48
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 37. Primary nozzle component efficiencies. Fig. 38. Suction chamber component efficiencies.

49
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 39. Mixing chamber component efficiencies.

Fig. 40. Diffuser efficiencies.

50
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

ṁ1 u4p + ṁ2 u4s

ηmix = (12)
(ṁ1 + ṁ2 )u5

his3 − h5
ηdif = (13)
h3 − h5

h(P3 , s2 ) − h2
ηej = ω (14)
h1 − h(P3 , s1 )
* above subscripts refer to ejector sections defined in Fig. 36

6. Conclusions

This paper contributes to the current discussion and closes the

knowledge gap by assessing the performances of a CFD approach for
single-phase supersonic ejector modelling. To this end, a comprehensive
validation is conducted, encompassing a wide range of ejector designs,
boundary conditions, and working fluids. In this procedure, a broad
screening of modelling approaches is conducted. First, a detailed mesh
screening study has been performed, and guidelines for the definition of
a grid-independent mesh are achieved, based on a reference cell size, a
maximum aspect ratio, wall requirements and mesh refinement. Also, a
comparison between 2-Dimensional and 3-Dimensional approaches is
conducted, showing no significant differences between the two. Second,
a comparison between PB and DB solvers is conducted, showing no
significant differences and a faster convergence and stability of the PB
solver. Third, a broad validation has been conducted, and two turbu­
lence models (k-ε RNG and k-ω SST) have been extensively compared:
the k-ω SST has shown better agreement with the experimental mea­
surements concerning both global and local flow quantities. The
entrainment ratio has been well predicted with an average on-design
error of 14% and a maximum of 20%. The off-design operation mode
is instead affected by a higher average error of 63% which is however
aligned with other studies. The higher deviations registered are caused
by the inherent complex fluid dynamics originating from the single-
chocking mode that characterize the subcritical operation. On the
other hand, good performances on the critical pressure prediction are
obtained with an average relative error equal to about 3%. Finally, the
simulation outcomes have been further post-processed to derive ejector
component efficiencies, to contribute to the present discussion regarding
closures in lumped parameter ejector modelling approaches. In
conclusion, the present paper proposes a “baseline” ejector CFD model
which could provide a basis for upcoming research activities on ejector
modelling.

CRediT authorship contribution statement

Giorgio Besagni: Conceptualization, Data curation, Investigation,

Methodology, Supervision, Writing - original draft. Nicolò Cristiani:
Conceptualization, Data curation, Investigation, Writing - original draft.
Lorenzo Croci: Investigation, Supervision, Project administration,
Funding acquisition. Gaël Raymond Guédon: Methodology. Fabio
Inzoli: Supervision.

Declaration of Competing Interest

Fig. 41. Ejector efficiency.
The authors declare the following financial interests/personal re­
P3 − P2 lationships which may be considered as potential competing interests:
CR = (9)
P1 − P2 [Giorgio Besagni is Associate Editor for Applied Thermal Engineering].

h1 − h4p
ηp = (10) Acknowledgments
h1 − his4p
This work has been financed by the Research Fund for the Italian
h2 − h4s
ηs = (11) Electrical System in compliance with the Decree of Minister of Economic
h2 − his4s
Development April 16, 2018.

51
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Appendix A

see Fig. 42.

Fig. 42. comparison of predicted inside fluid-dynamics for case M3a with 2D-3D meshes and DB-PB solvers - Mach contours comparisons: 2D-3D and DB-PB solvers.
Case M3a: T1 = 130.0 ◦ C, P1 = 270,300 Pa T2 = 5.0 ◦ C, P2 = 870 Pa, Tsat
3 = 24.1 C, P3 = 3,000 Pa, ωEXP = 0.309.
◦

Appendix B

see Table 14.

Table 14
Mesh sensitivity analysis results for case M3a with DB solver.
Benchmark M3a (ωEXP = 0.309)

Mesh code n cell ω Deviation

ω Eq. (6) P Eq. (7) Ma Eq. (8)

α1d 16,229 0.2926 4.63% 3.01% 4.15%

α1nd 51,788 0.3066 0.07% 1.15% 3.24%
α2d 14,103 0.2892 5.74% 3.62% 5.03%
α2nd 36,810 0.3048 0.65% 1.87% 3.89%
α3d 5,731 0.2781 9.37% 4.12% 5.62%
α3nd 19,555 0.2971 3.16% 2.38% 4.97%
α4d 4,745 0.2792 9.02% 4.78% 7.52%
α4nd 14,222 0.2961 3.50% 3.04% 6.20%
β1d 72,898 0.2979 2.91% 1.37% 2.12%
β1nd 112,444 0.3062 0.22% 0.52% 1.59%
β2d 61,599 0.2977 2.99% 1.87% 1.96%
β2nd 86,265 0.3039 0.94% 0.73% 3.56%
β3d 25,657 0.2915 5.01% 2.36% 3.11%
β3nd 41,116 0.2996 2.36% 1.26% 4.60%
β4d 21,506 0.2888 5.86% 2.88% 3.11%
β4nd 31,757 0.3009 1.94% 2.13% 4.18%
γ1d 173,052 0.3007 1.99% 0.81% 1.28%
γ1nd 234,453 0.3059 0.29% 0.52% 0.95%
γ2d 147,724 0.3001 2.18% 1.03% 1.56%
γ2nd 188,107 0.3046 0.72% 0.49% 1.19%
γ3d 60,152 0.2961 3.51% 1.38% 2.25%
γ3nd 84,260 0.3022 1.50% 0.78% 2.07%
γ4d 50,163 0.2981 2.83% 2.44% 3.65%
γ4nd 65,343 0.3046 0.73% 1.82% 3.22%
reference 1,163,362 0.3068 – – –

52
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Appendix C

see Fig. 43.

Fig. 43. Local pressure profiles - Every case is named accordingly to Table 7.

53
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 43. (continued).

54
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 43. (continued).

55
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 43. (continued).

56
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

Fig. 43. (continued).

References [21] F. Mazzelli, A.B. Little, S. Garimella, Y. Bartosiewicz, Computational and

experimental analysis of supersonic air ejector: Turbulence modeling and
assessment of 3D effects, Int. J. Heat Fluid Flow 56 (2015) 305–316.
[1] S. Rao, J. Gopalan, Observations on the non-mixed length and unsteady shock
[22] J. García Del Valle, J.M. Saíz Jabardo, F. Castro Ruiz, J.F. San José Alonso, An
motion in a two dimensional supersonic ejector, Phys. Fluids 26 (2014), 036103.
experimental investigation of a R-134a ejector refrigeration system, International
[2] Z. Aidoun, K. Ameur, M. Falsafioon, M. Badache, Current advances in ejector
Journal of Refrigeration, 46 (2014) 105-113.
modeling, experimentation and applications for refrigeration and heat pumps. Part
[23] S. Varga, J. Soares, R. Lima, A.C. Oliveira, On the selection of a turbulence model
1: Single-phase ejectors, Multidisciplinary Digital Publishing Institute 4 (2019) 15.
for the simulation of steam ejectors using CFD, Int. J. Low-Carbon Technolog. 12
[3] B.M. Tashtoush, M.d.A. Al-Nimr, M.A. Khasawneh, A comprehensive review of
(2017) 233–243.
ejector design, performance, and applications, Applied Energy, 240 (2019) 138-
[24] I.W. Eames, S. Aphornratana, H. Haider, A theoretical and experimental study of a
172.
small-scale steam jet refrigerator, Int. J. Refrig 18 (1995) 378–386.
[4] G. Besagni, R. Mereu, F. Inzoli, Ejector refrigeration: A comprehensive review,
[25] O.B. Robert, D. Zucker, Fundamentals of Gas Dynamics, John Wiley & Sons Inc,
Renew. Sustain. Energy Rev. 53 (2016) 373–407.
2002.
[5] Y. Fang, S. Croquer, S. Poncet, Z. Aidoun, Y. Bartosiewicz, Drop-in replacement in a
[26] K. Pianthong, W. Seehanam, M. Behnia, T. Sriveerakul, S. Aphornratana,
R134 ejector refrigeration cycle by HFO refrigerants, Int. J. Refrig 77 (2017)
Investigation and improvement of ejector refrigeration system using computational
87–98.
fluid dynamics technique, Energy Convers. Manage. 48 (2007) 2556–2564.
[6] F. Mazzelli, F. Giacomelli, A. Milazzo, CFD modeling of condensing steam ejectors:
[27] X. Wang, J. Dong, A. Li, H. Lei, J. Tu, Numerical study of primary steam
Comparison with an experimental test-case, Int. J. Therm. Sci. 127 (2018) 7–18.
superheating effects on steam ejector flow and its pumping performance, Energy
[7] G. Besagni, F. Inzoli, Computational fluid-dynamics modeling of supersonic
78 (2014) 205–211.
ejectors: Screening of turbulence modeling approaches, Appl. Therm. Eng. 117
[28] N. Sharifi, M. Boroomand, An investigation of thermo-compressor design by
(2017) 122–144.
analysis and experiment: Part 1. Validation of the numerical method, Energy
[8] P. Haghparast, M.V. Sorin, H. Nesreddine, The impact of internal ejector working
Conversion Manage. 69 (2013) 217–227.
characteristics and geometry on the performance of a refrigeration cycle, Energy
[29] J. Gagan, K. Smierciew, D. Butrymowicz, J. Karwacki, Comparative study of
162 (2018) 728–743.
turbulence models in application to gas ejectors, Int. J. Therm. Sci. 78 (2014) 9–15.
[9] M. Poirier, D. Giguère, H. Sapoundjiev, Experimental parametric investigation of
[30] A. Little, Y. Bartosiewicz, S. Garimella, Visualization and Validation of Ejector Flow
vapor ejector for refrigeration applications, Energy 162 (2018) 1287–1300.
Field With Computational and First-Principles Analysis, J. Fluids Eng. 137 (2015)
[10] Y. Han, X. Wang, H. Sun, G. Zhang, L. Guo, J. Tu, CFD simulation on the boundary
51107.
layer separation in the steam ejector and its influence on the pumping
[31] K.M. Arun, S. Tiwari, A. Mani, Three-dimensional numerical investigations on
performance, Energy 167 (2019) 469–483.
rectangular cross-section ejector, Int. J. Therm. Sci. 122 (2017) 257–265.
[11] G. Besagni, R. Mereu, E. Colombo, CFD study of ejector efficiencies, in: Engineering
[32] G. Grazzini, A. Milazzo, F. Mazzelli, Ejectors for Efficient Refrigeration: Design.
Systems Design and Analysis, Vol. 45844, American Society of Mechanical
Applications and Computational Fluid Dynamics, Springer, 2018.
Engineers, 2014, pp. V002T011A004.
[33] T. Thongtip, S. Aphornratana, Development and performance of a heat driven
[12] A. Hemidi, F. Henry, S. Leclaire, J.-M. Seynhaeve, Y. Bartosiewicz, CFD analysis of
R141b ejector air conditioner: Application in hot climate country, Energy 160
a supersonic air ejector. Part I: Experimental validation of single-phase and two-
(2018) 556–572.
phase operation, Appl. Therm. Eng. 29 (2009) 1523–1531.
[34] Directive 2006/40/EC of the European Parliament and of the Council of 17 May
[13] A. Fluent, Ansys Fluent Theory Guide, ANSYS Inc., USA, 2020.
2006 relating to emissions from air conditioning systems in motor vehicles and
[14] T. Sriveerakul, S. Aphornratana, K. Chunnanond, Performance prediction of steam
amending Council Directive 70/156/EC., in, 2006.
ejector using computational fluid dynamics: Part 1. Validation of the CFD results,
[35] A. Cavallini, The state of the art on Refrigerants, in: J.o.P.C. Series (ed.) 37th UIT
Int. J. Therm. Sci. 46 (2007) 812–822.
Heat Transfer Conference, Vol. 1599, Padova, Italy, 2019.
[15] J. García Del Valle, J. Sierra-Pallares, P. Garcia Carrascal, F. Castro Ruiz, An
[36] J. Kracik, V. Dvorak, V. Nguyen Van, K. Smierciew, Experimental and Numerical
experimental and computational study of the flow pattern in a refrigerant ejector.
Study on Supersonic Ejectors Working with R-1234ze(E), EPJ Web Conf., 180
Validation of turbulence models and real-gas effects, Applied Thermal Engineering,
(2018).
89 (2015) 795-811.
[37] K. Śmierciew, J. Gagan, D. Butrymowicz, Application of numerical modelling for
[16] S. Croquer, S. Poncet, Z. Aidoun, Turbulence modeling of a single-phase R134a
design and improvement of performance of gas ejector, Appl. Therm. Eng. 149
supersonic ejector. Part 1: Numerical benchmark, Int. J. Refrig 61 (2016) 140–152.
(2019) 85–93.
[17] V. Nguyen, J. Kracik, CFD simulation of ejector: is it worth to use real gas models?,
[38] M. Atmaca, C. Ezgi, Three-dimensional CFD modeling of a steam ejector, Energy
EPJ Web of Conferences, 180 (2018) 2075-2075.
Sources, Part A: Recovery, Utilizat. Environ. Effects (2019) 1–12.
[18] Y. Bartosiewicz, Z. Aidoun, P. Desevaux, Y. Mercadier, Numerical and
[39] C. Ezgi, I. Girgin, Design and thermodynamic analysis of a steam ejector
experimental investigations on supersonic ejectors, Int. J. Heat Fluid Flow 26
refrigeration/heat pump system for naval surface ship applications, Entropy 17
(2005) 56–70.
(2015) 8152–8173.
[19] N. Ruangtrakoon, T. Thongtip, S. Aphornratana, T. Sriveerakul, CFD simulation on
[40] S. Mohamed, Y. Shatilla, T. Zhang, CFD-based design and simulation of
the effect of primary nozzle geometries for a steam ejector in refrigeration cycle,
hydrocarbon ejector for cooling, Energy 167 (2019) 346–358.
Int. J. Therm. Sci. 63 (2013) 133–145.
[41] B.J. Huang, J.M. Chang, C.P. Wang, V.A. Petrenko, 1-D analysis of ejector
[20] X. Yang, X. Long, X. Yao, Numerical investigation on the mixing process in a steam
performance, Int. J. Refrig 22 (1999) 354–364.
ejector with different nozzle structures, Int. J. Therm. Sci. 56 (2012) 95–106.

57
G. Besagni et al. Applied Thermal Engineering 186 (2021) 116431

[42] A.S. Ramesh, S.J. Sekhar, Experimental and numerical investigations on the effect [65] W. Chen, D. Chong, J. Yan, J. Liu, Numerical optimization on the geometrical
of suction chamber angle and nozzle exit position of a steam-jet ejector, Energy 164 factors of natural gas ejectors, Int. J. Therm. Sci. 50 (2011) 1554–1561.
(2018) 1097–1113. [66] S. Varga, A.C. Oliveira, X. Ma, S.A. Omer, W. Zhang, S.B. Riffat, Comparison of CFD
[43] H. Zhang, L. Wang, L. Jia, X. Wang, Assessment and prediction of component and experimental performance results of a variable area ratio steam ejector, Int. J.
efficiencies in supersonic ejector with friction losses, Appl. Therm. Eng. 129 (2018) Low-Carbon Technolog. 6 (2010) 119–124.
618–627. [67] J. Kolář, V. Dvořák, Verification of K-ω SST turbulence model for supersonic
[44] Y. Wu, H. Zhao, C. Zhang, L. Wang, J. Han, Optimization analysis of structure internal flows, World Academy of Science, Eng. Technol. (2011).
parameters of steam ejector based on CFD and orthogonal test, Energy 151 (2018) [68] M. Ji, T. Utomo, J. Woo, Y. Lee, H. Jeong, H. Chung, CFD investigation on the flow
79–93. structure inside thermo vapor compressor, Energy 35 (2010) 2694–2702.
[45] L. Wang, J. Yan, C. Wang, X. Li, Numerical study on optimization of ejector [69] X.-D. Wang, J.-L. Dong, Numerical study on the performances of steam-jet vacuum
primary nozzle geometries, Int. J. Refrig 76 (2017) 219–229. pump at different operating conditions, Vacuum 84 (2010) 1341–1346.
[46] O. Lamberts, P. Chatelain, Y. Bartosiewicz, New methods for analyzing transport [70] Y. Zhu, W. Cai, C. Wen, Y. Li, Numerical investigation of geometry parameters for
phenomena in supersonic ejectors, Int. J. Heat Fluid Flow 64 (2017) 23–40. design of high performance ejectors, Appl. Therm. Eng. 29 (2009) 898–905.
[47] A.B. Little, S. Garimella, Shadowgraph visualization of condensing R134a flow [71] D. Scott, Z. Aidoun, O. Bellache, M. Ouzzane, CFD Simulations of a Supersonic
through ejectors, Int. J. Refrig 68 (2016) 118–129. Ejector for Use in Refrigeration Applications. International Refrigeration and Air
[48] K. Ariafar, D. Buttsworth, G. Al-Doori, N. Sharifi, Mixing layer effects on the Conditioning Conference, Purdue, 2008.
entrainment ratio in steam ejectors through ideal gas computational simulations, [72] A.E. Ablwaifa, A theoretical and experimental investigation of jet-pump
Energy 95 (2016) 380–392. refrigeration system, in (2006).
[49] A. Hakkaki-Fard, M. Ouzzane, Z. Aidoun, D. Giguère, M. Poirier, An Experimental [73] E. Rusly, L. Aye, W.W.S. Charters, A. Ooi, CFD analysis of ejector in a combined
Study of Ejectors Supported by CFD. in: 24th IIR International Congress of ejector cooling system, Int. J. Refrig 28 (2005) 1092–1101.
Refrigeration ICR2015, 2015. [74] P. Desevaux, O. Aeschbacher, Numerical and experimental flow visualizations of
[50] F. Mazzelli, A. Milazzo, Performance analysis of a supersonic ejector cycle working the mixing process inside an induced air ejector, Int. J. Turbo Jet Engines 4 (2002)
with R245fa, Int. J. Refrig 49 (2015) 79–92. 311.
[51] F. Kong, H.D. Kim, Analytical and computational studies on the performance of a [75] H.A.M. Al-Ansary, S.M. Jeter, Numerical and Experimental Analysis of Single-
two-stage ejector–diffuser system, Int. J. Heat Mass Transf. 85 (2015) 71–87. Phase and Two-Phase Flow in Ejectors, HVAC&R Research 10 (2004) 521–538.
[52] D. Chong, M. Hu, W. Chen, J. Wang, J. Liu, J. Yan, Experimental and numerical [76] D. Lucas, R. Rzehak, E. Krepper, T. Ziegenhein, Y. Liao, S. Kriebitzsch,
analysis of supersonic air ejector, Appl. Energy 130 (2014) 679–684. P. Apanasevich, A strategy for the qualification of multi-fluid approaches for
[53] Y. Zhu, P. Jiang, Experimental and numerical investigation of the effect of shock nuclear reactor safety, Nucl. Eng. Des. 299 (2016) 2–11.
wave characteristics on the ejector performance, Int. J. Refrig 40 (2014) 31–42. [77] D.C. Wilcox, Turbulence Modeling for CFD, DCW Industries Inc, La Canada,
[54] V.V. Chandra, M.R. Ahmed, Experimental and computational studies on a steam jet California, 2006.
refrigeration system with constant area and variable area ejectors, Energy Convers. [78] E.W. Lemmon, M.L. Huber, M.O. McLinden, NIST standard reference database 23,
Manage. 79 (2014) 377–386. NIST reference fluid thermodynamic and transport properties—REFPROP, Version
[55] F. Bakhtar, M.T. Mohammadi Tochai, An investigation of two-dimensional flows of (2010).
nucleating and wet steam by the time-marching method, Int. J. Heat Fluid Flow 2 [79] J. Gagan, K. Śmierciew, D. Butrymowicz, Performance of ejection refrigeration
(1980) 5–18. system operating with R-1234ze(E) driven by ultra-low grade heat source, Int. J.
[56] V. Kumar, G. Singhal, P.M.V. Subbarao, Study of supersonic flow in a constant rate Refrig 88 (2018) 458–471.
of momentum change (CRMC) ejector with frictional effects, Appl. Therm. Eng. 60 [80] T. Thongtip, S. Aphornratana, An experimental analysis of the impact of primary
(2013) 61–71. nozzle geometries on the ejector performance used in R141b ejector refrigerator,
[57] K. Banasiak, M. Palacz, A. Hafner, Z. Buliński, J. Smołka, A.J. Nowak, A. Fic, Appl. Therm. Eng. 110 (2017) 89–101.
A CFD-based investigation of the energy performance of two-phase R744 ejectors [81] J. Yan, C. Lin, W. Cai, H. Chen, H. Wang, Experimental study on key geometric
to recover the expansion work in refrigeration systems: An irreversibility analysis, parameters of an R134A ejector cooling system, Int. J. Refrig 67 (2016) 102–108.
Int. J. Refrig 40 (2014) 328–337. [82] K.O. Shestopalov, B.J. Huang, V.O. Petrenko, O.S. Volovyk, Investigation of an
[58] A. Gutiérrez, N. León, Conceptual Development and CFD Evaluation of a High experimental ejector refrigeration machine operating with refrigerant R245fa at
Efficiency – Variable Geometry Ejector for Use in Refrigeration Applications, design and off-design working conditions. Part 2. Theoretical and experimental
Energy Procedia 57 (2014) 2544–2553. results, Int. J. Refrig 55 (2015) 212–223.
[59] R.H. Yen, B.J. Huang, C.Y. Chen, T.Y. Shiu, C.W. Cheng, S.S. Chen, K. Shestopalov, [83] A. Ashrae, Standard 34–2001, Designation and safety classification of refrigerants.
Performance optimization for a variable throat ejector in a solar refrigeration American society of heating, refrigerating, and air-conditioning engineers, Atlanta,
system, Int. J. Refrig 36 (2013) 1512–1520. GA, 2001.
[60] J. Smolka, Z. Bulinski, A. Fic, A.J. Nowak, K. Banasiak, A. Hafner, A computational [84] P.J. Roache, Quantification of uncertainty in computational fluid dynamics, Annu.
model of a transcritical R744 ejector based on a homogeneous real fluid approach, Rev. Fluid Mech. 29 (1997) 123–160.
Appl. Math. Model. 37 (2013) 1208–1224. [85] A. Bouhanguel, Etude numérique et expérimentale de l’interaction entre deux
[61] E. Rusly, Ejector cooling with reference to combined ejector-vapour compression écoulements compressibles dans un éjecteur supersonique, in (2013).
system, in (2004). [86] B.J. Huang, J.M. Chang, Empirical correlation for ejector design, Int. J. Refrig 22
[62] C. Lin, W. Cai, Y. Li, J. Yan, Y. Hu, Pressure recovery ratio in a variable cooling (1999) 379–388.
loads ejector-based multi-evaporator refrigeration system, Energy 44 (2012) [87] S. Varga, A.C. Oliveira, B. Diaconu, Numerical assessment of steam ejector
649–656. efficiencies using CFD, Int. J. Refrig 32 (2009) 1203–1211.
[63] M.J. Opgenorth, D. Sederstrom, W. McDermott, C.S. Lengsfeld, Maximizing [88] S. Elbel, P. Hrnjak, Experimental validation of a prototype ejector designed to
pressure recovery using lobed nozzles in a supersonic ejector, Appl. Therm. Eng. 37 reduce throttling losses encountered in transcritical R744 system operation, Int. J.
(2012) 396–402. Refrig 31 (2008) 411–422.
[64] M. Yazdani, A.A. Alahyari, T.D. Radcliff, Numerical modeling of two-phase
supersonic ejectors for work-recovery applications, Int. J. Heat Mass Transf. 55
(2012) 5744–5753.

58