Heat and Mass Transfer (2020) 55:1015–1036

https://doi.org/10.1007/s00231-019-02777-w

ORIGINAL

Performance prediction, numerical and experimental investigation

to characterize the flow field and thermal behavior of a cryogenic
turboexpander
Manoj Kumar1 · Debashis Panda1 · Ranjit K. Sahoo1 · Suraj K. Behera1

Received: 4 February 2019 / Accepted: 17 October 2019 / Published online: 19 November 2019
© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Radial inflow turbine and nozzle among the other components of the cryogenic turboexpander has a significant effect
on the efficiency of the system. This study proposes an effective one-dimensional design approach of a radial turbine by
introducing different loss correlations. The methodology also describes the effect of non-dimensional design variables on
the performance of the turbine. These variables (blade speed ratio, pressure ratio, hub and shroud to turbine inlet radius
ratio) undergo artificial intelligence-based model to predict their optimal range for better efficiency and power output of
the turbine. Based on these optimal ranges, two turbine and nozzle models are generated. The results of the optimized
configuration show that the turbine total-to-static efficiency and power output are higher by 4% and 18.9% respectively as
compared to the existing literature. Thereafter, the three-dimensional computational fluid dynamics (CFD) analysis is carried
out to visualize the fluid flow and thermal characteristics at different inlet temperatures in the flow passage using ANSYS
CFX® . The study also focuses to identify the flow separation zone, tip leakage flow, vortex formation, secondary losses and
its reasons at different spans of the turbine. An experimental platform is also established to validate the CFD results of a case
study. The experimental results show that the mass flow rate and rotational speed has major effect on temperature drop and
isentropic efficiency of the turboexpander. The study highlights the importance of the design methodology, the estimation
capability of artificial intelligence models, the experimental techniques and benchmarking model for numerical analysis at
different cryogenic temperature.

Keywords Radial turbine · Flow and thermal characteristics · CFD analysis · Experimental techniques ·
Artificial intelligence methods

Nomenclature D Turbine diameter (m)

Variables Dn Nozzle ring diameter (m)
b Blade height (m) Dh Mean passage hydraulic diameter (m)
bt Nozzle height (m) Dt Nozzle throat circle diameter (m)
C Absolute velocity (m/s) d Diameter (m)
C0 Spouting velocity (m/s) ds Specific diameter
Chn Chord length (m) h Enthalpy (kJ /kg)
Cmt Meridional velocity at throat of the nozzle (m/s) k Discharge coefficient
Cθ t Tangential velocity at throat of the nozzle (m/s) kxr Cross coupling coefficient
c Turbine blade chord (m) kp Constant
L Axial length (m)
Lt Total loss
Lh Mean passage hydraulic length (m)
 Manoj Kumar M Mach number
.
[email protected] m Mass flow rate (kg/s)
ns Specific speed
1 P Power (kW )
National Institute of Technology Rourkela, Odisha,
769008, India Pn Blade pitch length (m)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1016 Heat Mass Transfer (2020) 55:1015–1036

p Pressure (P a)   m Meridional
Q Volume flow rate m3 /s n Nozzle
R Turbine radius (m) o Total
Rh Hub radius (m) r Radial
Rs Shroud radius (m) rel Relative
r Radius (m) s Shroud, Isentropic
rp Pressure ratio T EL Trailing edge loss
s Meridional streamlength (m), t Throat
t Blade thickness(m) tip Tip (turbine exit)
U Blade speed (m/s) x Axial
vs Blade speed ratio 0 Stagnation state
W Relative velocity (m/s) 1 Nozzle inlet
Wt Throat width(m) 2 Turbine inlet
Z Number of blades 3 Turbine outlet
Zr Rotor axial length (m) θ Tangential

Acronyms
ANF I S Adaptive neuro-fuzzy inference system 1 Introduction
ANN Artificial neural network
CF D Computational fluid dynamics The branch of low-temperature physics, later renowned
MAE Mean absolute error as cryogenics started with liquefaction of oxygen by
MLP Multi-layer perceptron Raoul Pierre Pictet (Swiss physicist) and Louis Paul
ORC Organic Rankine cycle Cailletet (French mining engineer) but it is considered
P F H X Plate-fin heat exchanger that cryogenic science borns after liquefaction of helium.
RMSE Root mean squared error In 1896, Onnes proposed the concept of using an
TF Transfer function expansion turbine in liquefaction process of hydrogen [1,
2]. However, Claude implements turbine as an expander
Greek symbols in place of expansion device for the first time in his
α Absolute velocity angle (degree), experiment for air liquefaction system. After that, the
αt Nozzle throat angle (degree), era of cryogenic turboexpander begins for liquefaction
β Absolute flow angle (degree), cycles. The usage of cryogenic temperature is essential,
β2,opt Incidence angle (degree) for example, to produce an inert environment (shielding
εx Axial clearance (m) gas) in welding process, leak detectors, optical fibers,
εr Radial clearance (m) solar telescopes, superconductivity and superfluidity, space
γ Specific heat ratio appliances, biomedical and chemical instruments, etc. [3].
ηt−s Total-to-static
 efficiency
 Radial inflow turbine is extensively used in expanders,
ρ Density kg/m3 gas liquefaction units, cryogenic applications, chemical,
ηis Isentropic efficiency transportation, turbochargers, internal combustion engines,
ψz Zweifel number and petrochemical industries because of its higher perfor-
ψ Stage head coefficient mance, ruggedness, efficiency, and less cost as compared
λ Turbine outlet to inlet radius ratio to axial flow turbine [4]. It can also operate at a differ-
λs Stagger angle (degree) ent range of mass flow rate and rotational speeds. The
χ Ratio of absolute meridional velocity evolution of computational fluid dynamics explore many
φ Flow coefficient aspects like advancement in aerodynamic profile, flow field
ζ Hub ratio characteristics and various losses occurred inside different
Θ Rotor meridional velocity coefficient components of a turboexpander. Some internal and external
ω Rotational speed (rpm) losses exist inside the turboexpander. The external losses
(bearing loss) can not be determined through numerical sim-
Subscripts ulations. However, internal losses are considered. Several
BL Blade loading loss techniques (active and passive) are proposed to eliminate
Cl Clearance loss these losses for turbine and compressor [5, 6]. Whitfield and
h Hub Baines [7] propose the design methodology for computing
I Incidence loss the various losses inside the turbomachines. Harvey et al.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1017

[8] propose the design methodology to increase the aero- different rotational speeds. Harinck et al. [24] perform
dynamic performance of turbine blade profile by increasing the three-dimensional numerical investigation for a set
efficiency at higher stage loading. of nozzle, radial turbine, and diffuser used in an ORC
Flow separation over turbine blade surfaces is a cycle. Yang et al. [25] investigate the coupling performance
significant concern among aerodynamic designers. It affects of a turboexpander compressor for air refrigerator based
the performance of rotor as well as turboexpander. on reversed Brayton cycle. Nitesh and Chatterjee [26]
Suhrmann et al. [9] propose different loss correlations develop a small-scale radial turbine with 2 kW power
for improving the aerodynamic profile of micro radial output for renewable energy conversion applications. Chang
turbine. The blade profile losses obtain due to boundary et al. [27] propose the reversed Brayton cycle-based two-
layer formation near the blade surface. It is minimized stage helium refrigerator in which two expanders are
by reducing the blade thickness at leading and trailing used in series operating at constant mass flow rate and
edge. Tako et al. [10] suggest the effect of blade thickness different pressure ratio. Streit and Razani [28] propose
on flow separation and secondary losses. This loss may thermodynamic optimization of the expander. Ke et al. [29]
exist because of reversed flow initiated by adverse pressure study the effect of vaneless space on performance of turbine
gradient. Also, the flow vortices are originated due to through experimental and numerical analysis.
mixing of these secondary flow to the mainstream. It can The literature review cited above unveils that very less
also obtain because of aerodynamic losses which are an literature exists which use the ANN and ANFIS models
important reason for decreasing the turbine performance in the design process of a turbine. Also, numerical and
[11, 12]. In this regard, some remarkable works are reported experimental analysis is important to envisage the fluid
considering various loss correlations in the preliminary flow behavior and thermal characteristics of a cryogenic
design process [13–16]. turboexpander. Present paper fulfills this literature gap.
The efficiency of small-scale radial turbine is lower In this paper, an effective and comprehensive 1 −
as compared to large turbines because of these losses. D design methodology for the development of a radial
Therefore, the efficiency can be improved by considering turbine is illustrated considering different loss correlations.
those losses in the one-dimensional (1 − D) design The non-dimensional design variable datasets are used to
procedure. Also, numerical studies of fluid flow and thermal develop an ANN and ANFIS networks to estimate their
characteristics of such kind of turbine are essential and optimal ranges for maximum total-to-static efficiency and
a matter of interest for extensive investigation. This can power output of a radial turbine. Based on the estimated
highlight the problem of flow separation, vortex generation, ranges, two turbine and nozzle models are developed.
primary and secondary losses, boundary layer separation, The obtained turbine dimensions are further validated with
and transitional turbulence. These type of losses can be previously published results. Thereafter, three-dimensional
reduced by optimizing the turbine blade profile. Therefore, Reynolds-averaged Navier-Stokes (RANS) computations
researchers are interested in increasing the efficiency of are carried out with k − ω shear stress transport (SST)
small turbine by minimizing these losses which may turbulence model to investigate the fluid flow and thermal
cause improvement in performance of the turboexpander characteristics of the turboexpander using ANSYS CFX® .
system [17]. Major performance parameters, like density, pressure
Furthermore, to increase the performance of reversed variation, Mach number, Reynolds number, static enthalpy,
Brayton cycle based turboexpander, the design of small static entropy, temperature variation, and internal flow
radial turbine needs improvement and non-dimensional and structures at different spans and meridional plane are
geometrical design variables must be optimized. Nowadays, reported. Additionally, to validate the numerical results, an
artificial neural network (ANN) and adaptive neuro-fuzzy experimental platform is developed.
inference system (ANFIS) are powerful optimization tool
and more demanding [18–20]. Some works are reported
on the application of ANN and ANFIS networks to 2 Design procedure for turbine and nozzle
characterize the performance of turbomachinery and radial
turbines [21, 22]. 2.1 Design of radial inflow turbine
Apart from this, some experimental and numerical
analysis of turboexpander is also reported in the literature. The preliminary design methodology of a radial inflow
Hongli et al. [23] perform an experimental analysis to turbine is introduced to save time because of excessive
investigate the performance of a turboexpander used in empirical correlations are employed in this process [30].
air refrigerator which is based on reversed Brayton cycle. The thermophysical properties of working fluid have a
Kumar et al. [22] conduct the experimental analysis to significant effect on aerodynamic design of radial inflow
study the performance of a cryogenic turboexpander at turbine [31]. Generally, cryogenic fluids are diatomic

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1018 Heat Mass Transfer (2020) 55:1015–1036

gas, and its thermophysical properties are complicated

and varied at high pressure and ultra-low temperature.
It is complicated to predict cryogenics fluid properties
during expansion process through ideal gas laws. Therefore,
accurate design of the turboexpander components needs a
real gas model which is implemented using REFPROP.
The preliminary design and calculation of turbine
geometry are performed in Matlab® 2017b by adopting an
analytical approach proposed by Hasselgruber and Balje
using primary data (thermodynamic properties) and some
non-dimensional parameters within the specified range.
Inlet pressure, mass flow rate, expansion ratio, rotational
speed, blade speed ratio, flow angles and number of blades,
etc. are the decision variables on which performance of
the turbine depends. Pressure ratio (rp ), blade speed ratio
(vs ), hub and shroud exit radius to turbine inlet radius ratio
are other parameters on which feasibility of turbine design
depends. The ranges of different parameters are mentioned
in Table 1.
In the design process, machining precisions are carefully Fig. 1 Preliminary design flowchart of a radial turbine
taken, for which stage loading (ψ) and flow coefficient (φ)
values are taken to be higher. The first step is to select the where Q3 is the volumetric flow rate and h1−3,s is
degree of reaction, stage loading, and flow coefficient. The isentropic enthalpy drop.
specific speed and specific diameter are selected using Balje The flow coefficient (φ) and static head coefficient (ψ)
chart to maintain the total-to-static efficiency of rotor in a are formulated as follows:
stratified zone and Mach number of the fluid at the turbine Cm3 Cm2
inlet are in subsonic regime, which ultimately indicates the φ= = (3)
U2 χU2
suitability of the design parameters. Figure 1 represents the
flowchart of 1 − D design methodology. h02 Cθ2 Cθ 3
ψ= 2
= −λ (4)
Specific speed: U2 U2 U2
√ h02 − h03
ω Q3 ηt−s = (5)
ns = (1) h02 − h03s
(h1−3s )3/4
Power output:
Specific diameter:
P = ṁ (h1 − h3 ) (6)
D2 (h1−3s ) 1/4
ds = √ (2) Blade velocity at turbine inlet:
Q3
ωD2
U2 = (7)
2
Table 1 Ranges of different parameters of radial turbine design for Spouting velocity:
cryogenic applications 
C0 = 2 (h1 − h3s ) (8)
Design variables Range
Blade speed ratio:
Inlet stagnation Pressure (bar) 8 − 10 U2
Inlet stagnation Temperature (K) 90 − 180 vs = (9)
. C0
m (kg/s) 0.01 − 0.09
The radial turbine has maximum efficiency for blade
Tip clearance (mm) 0.01
speed ratio in the range of 0.62 to 0.70. The eye tip to
ω (rpm) 80, 000 − 150, 000
inlet diameter ratio is 0.70 to restrain the shroud curvature.
L/D 0.35 − 0.46
Hub ratio (ζ ) is taken to be 0.60, corresponding to the
Blade speed ratio 0.60 − 0.82
maximum total-to-static efficiency within subsonic zone
Blade inlet angle (Degree) 72 − 82
and for obtaining the longer blade passages.
Blade exit angle (Degree) −2.8 − 3.6
Dtip
Z 10 − 17 ζ = (10)
D2

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1019

Dh Thereafter, different loss correlations are incorporated

λ= (11)
Dtip in the design procedure. The incidence loss, friction loss,
passage loss, rotor clearance loss, blade loading loss, and
The number of turbine blades is computed as:
trailing edge loss are computed [9, 32]. Then, total-to-static
π efficiency is computed and used to update the initial guess
Z= (110 − α2 ) tan (α2 ) (12)
30 in the design process. The various loss correlations are
The thickness of blade is obtained from the correlations expressed as follows [33]:
proposed by Aungier [30] which are as follows:
2.1.1 Incidence loss
t2 = 0.04R2 (13)
The kinetic energy of the fluid is converted into internal
t3 = 0.02R2 (14) energy which increases the entropy. This loss is expressed as:
  
The outlet velocity triangle is expressed as: W22 cos β2 − β2,opt
hI = (16)
 2 2
b2
Θ =1+5 (15) where β2,opt is incidence angle which is computed as:
R2
Figure 2 represents the details of rotor geometry and −1.98cot (α2 )
cotβ2,opt = (17)
velocity triangles. Z(1 − 1.98
Z )

Fig. 2 a Rotor geometry details

b Inlet velocity triangle c Outlet
velocity triangle

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1020 Heat Mass Transfer (2020) 55:1015–1036

2.1.2 Passage loss The total-to-static efficiency is updated by adding the

computed total loss which is expressed as:
This loss is defined as a function of mean kinetic energy. It (h02 − h03 )
is expressed as: ηt−s =  (24)
(h02 − h03s ) + hLt
 
Lh Rt sinβ3
hpassage = kp + 0.68 1 − 2.2 Nozzle design
Dh R2 b3m /c
 
W2 + W3
2 2
× (18) The convergent type nozzle is designed in such a way
2 which provide subsonic flow regime at the nozzle exit.
The flow from nozzle should incident on turbine blade at
where Lh , Dh , and c are mean passage hydraulic length,
correct angle to minimize the incidence losses. The height
hydraulic diameter and chord length of turbine respectively.
(bt ), leading and trailing edge thickness and throat circle
kp is taken to be 0.2 (secondary loss constant).
diameter (Dt ) of the nozzle are designed based on the
correlations proposed by Kun and Sentz [36]. The nozzle
2.1.3 Rotor clearance loss
height is decided based on the concept of expansion in
annular space as well as to counter the axial misalignment
This loss is occurred in the tip clearance gap because of
of turbine. Generally, this height is less than tip width of the
the shear flow. In this region, the velocity variation is
turbine.
linear approaches from zero (casing) to one (blade surface
The meridional (C mt ) and throat (C θ t ) are two velocity
velocity). Also, axial (εx ) and radial (εr ) clearance plays a
components obtained at the exit of the throat. Cmt is
crucial role on this loss which are expressed as:
perpendicular to throat circle diameter, which determines
U22 Z  √  the mass flow rate whereas C θ t is tangential to throat.
hCl = kx εx cx + kr εr cr + kxr εx cx εr cr (19) The mass balance equation at the meridional plane is as
8π
follows:
where kx and kr are discharge coefficients and kxr is cross .
coupling coefficient. In this case, value of kx , kr , and kxr m
Cmt = (25)
are taken to be 0.40, 0.75, and −0.3 respectively [34]. π Dt ρ t b t

U2 D2
2.1.4 Trailing edge loss Cθ t = (26)
Dt
This loss can be computed by assuming the relative kinetic
energy at the turbine exit is equivalent to relative pressure bt = 0.8b2 (27)
drop which is expressed as:

ρ3 W32 Zt D t = 1.068D2 (28)

prel = p3rel − p2rel = (20)
2 π Rt cosβ3
2.3 Size of nozzle vane
Relative pressure drop is converted into trailing edge loss
coefficient and expressed as: The conservation of momentum and continuity equations
2 prel are used for calculating throat angle which is directly related
hT EL = 2
× (21) to trailing edge thickness of the nozzle. Aerodynamically,
γ M3rel p3rel
trailing edge thickness is as minimum as the mechanical
design limit can sustain. From the continuity equation,
2.1.5 Blade loading loss
throat width (W t ) and throat angle (α t ) are computed as
follows:
It is a type of secondary loss obtained because of boundary
mt
layer separation. It depends on the aerodynamics of the Wt = (29)
blade profile and expressed as [35]: Zn ρt bt Ct
 
(W2 − U2 )2 αt = tan−1
Cmt
hBL =2 (22) Cθ t
(30)
Zc/R2
The total losses (Lt ) are expressed as: The blade pitch length (pn ) is computed as:
 π Dt
hLt = hI +hpassage +hCl +hT EL +hBL (23) Pn = (31)
Zn

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1021

Fig. 3 Schematic of ANN

model network

The nozzle ring diameter (Dn ) is computed as: train the network, the validation datasets tests the accuracy
 of the network. The final model with high prediction
Dn = Dt2 + Wt2 − 2Wt Dt Cos (αt ) (32) accuracy rate has opted in which optimized input and output
The solidity value is obtained from aerodynamic load data sets are determined.
coefficient (ψ z ), which may be defined as ratio of actual
to ideal tangential force, renowned as Zweifel number. Its 3.2 ANFIS model development
value is taken to be 0.9. The chord length of nozzle is
obtained from equation of solidity and expressed as: ANFIS model is developed to estimate total-to-static
efficiency, total losses and power output of turbine. There
2s (cotαt − cotα0 ) sin2 αt
Chn = (33) are four input and three output variables which are divided
ψz sinλs into 80% and 20% datasets for training and validation
purpose.
3 Artificial intelligence model development 3.3 Structure of ANFIS
The ANN and ANFIS are soft computing models which are
Initially, the membership functions (MFs) are selected to
developed using design variable of a radial inflow turbine
determine an ANFIS network. MFs are arbitrary curves
to predict its optimal ranges in which it has the highest
which depend on type of application that can be utilized.
performance. The networks are developed to estimate the
In the present case, four MFs for every input has opted
total-to-static efficiency, total losses and power output of
[22]. There are various types of membership function
radial inflow turbine with four non-dimensional inputs
available, out of which, Gauss, Tri, Trap and Gbell MFs are
variables (Rs /R2 , Rh /R2 , vs , and rp ) in Matlab® toolbox.
implemented for input parameters to analyze which model
is suitable for this case. Additionally, linear function is used
3.1 ANN model development for output parameters. The basic structural information of
ANFIS model is shown in Fig. 4 and Table 3.
The number of layers, neurons, and transfer function type
are identified in the ANN model. The multi-layer perceptron
(MLP) neural network with a single hidden layer has opted
Table 2 The details of ANN model
for training process using a back-propagation algorithm.
Figure 3 illustrates a schematic of an ANN network with a Parameter Value
single hidden layer. The functional relation is expressed as:
⎛ ⎞ Input 1 (Rh /R2 ) 0.22 − 0.30
k  n
Input 2 (Rs /R2 ) 0.76 − 0.96
ym = g ⎝ 2
wmj ϕ wj1i xi ⎠ (34) Input 3 (vs ) 0.62 − 0.70
j =0 i=0  
Input 4 rp 2−5
Three transfer functions (TFs) are used for a comparative Output 1 (P in kW) 0.60 − 4.2
study of data prediction. From several trials, the best Output 2 (Lt ) 0.04 − 0.28
network is obtained by varying the hidden neurons in the Output 3 (ηt−s ) 0.60 − 0.84
range of 1 to 10. For every hidden neuron, training process Model structure MLP
is repeated. The other relevant details of ANN model are Error function MAE
presented in Table 2. Data filtering is necessary because it Transfer function Logsig, Purelin, Tansig
affects the network accuracy. Therefore, once the input and Hidden layers 1
output data are analyzed, the erroneous data are removed by Hidden neuron 1-10 with step size one
considering their ranges. The non-dimensional data sets are Training algorithm Back-propagation
divided into training (80%), testing (10%), and validation Number of epochs 1000
(10%) section to implement the model. The training data

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1022 Heat Mass Transfer (2020) 55:1015–1036

Fig. 4 The basic structure of

ANFIS

The ANFIS structure is expressed by if-then rules derived 3.4 ANN and ANFIS results
from Takagi-Sugeno (T − S) model as follows:
Figure 5 illustrates the regression curves for comparison of
Rule-1: If x is Ai and y is Bj then calculated and predicted results of total-to-static efficiency,
fi = pi x1 + qi yn + ri (35) total losses and power output from ANN network.
The ANFIS networks are trained based on non-
The performance of the model is verified using three dimensional variables used in the design process. There are
error functions including R-squared, Mean Absolute error three networks: ANFIS-1 estimate total-to-static efficiency,
(MAE) and Root Mean Squared Error (RMSE) which are ANFIS-2 estimate total losses, and ANFIS-3 estimate power
as follows: output. The samples of non-dimensional input and output
parameters and its predicted values from ANN and ANFIS
N
t=1 (ft − Ot )2 are summarised in Table 4. It shows that the Gauss MF
R2 = 1 −   (36)
− 2 possesses less eroor as compared to other MF.
N
t=1 ft − ft
3.5 Preliminary design results and its validation
1
MAE = × t=1
N
| ft − Ot | (37)
N The aforementioned 1 − D design methodology and pre-
dicted results from artificial intelligence techniques, two tur-
 bine and nozzle models are developed. The obtained 1 − D
1
RMSE = × t=1
N
(ft − Ot )2 (38) results are compared with previously published data which
N
is shown in Table 5. It shows that the 1−D design method is
where Ot is model output, ft is overall output, N is total reliable. Other relevant specifications of turbine and nozzle
number of samples and t is number of samples. are shown in Table 6.

Table 3 ANFIS model specification

4 Numerical issues
Parameters Description

Type of membership function (input) Gauss, Tri, Trap, Gbell 4.1 Computational model and mesh generation
Type of membership function (output) Linear
Number of membership function 4 For high computational accuracy of numerical results, fine
Optimization type Hybrid
hexahedral grids are generated for turbine and nozzle model
Number of training data 54
using ANSYS Turbo-Grid. The H -type topology is used for
Number of epochs 200
inlet and outlet blocks whereas J /O-type is selected for
passage block to solve near-wall region which is required

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1023

Fig. 5 Regression curve of

ANN a ηt−s b Lt c P

for boundary layer flow visualization. The tip clearance gap domain slide with respect to each other. This methodol-
is discretized with O-type grid. Also, O-grid (20) includes ogy is realistic and more demanding [38]. The steady-state
in the periodic boundary and 100 points are located from simulations are used as an initial solution for a transient
hub to shroud. The node distribution of grid and its scheme rotor-stator simulation for faster convergence of numerical
are mentioned in Table 7. solution.
Apart from this, y+ value which shows non-dimensional SST k − ω turbulence model is used to capture the turbu-
distance from the wall are mentioned in Table 8 [37]. lence closure and flow separation effects on the eddy viscos-
ity. This model can accurately estimate the boundary layer
4.2 Numerical methodology separations in the turbomachinery [39–41]. It combines the
advantages of two commonly used turbulence models: k −ω
The numerical simulations are conducted using commer- model and k −  model. k − ω model can predict the flow
cially available CFD platform ANSYS CFX® . It is based near-wall region whereas k −  model in wakes and free-
on three-dimensional, time-dependent, viscous and com- shear regions in the outer boundary layer (bulk flow). A
pressible RANS equations using finite volume method. blending function establishes a smooth transition between
Firstly, a moving reference frame is defined using frozen two models. In this aspect, k − ω SST model can predict
rotor model to compute rotation of turbine in a steady- the boundary layer through the passage (excellent turbulent
state condition. In this method, fluid around the turbine boundary layer modeling behavior), in which ω near the
blade is set as a moving reference frame whereas blade wall is more stable than that of .
and hub are assumed to be stationary with respect to Additionally, high-resolution scheme opts for discretiza-
the inner fluid. In this way, there is no need for grid tion of advection term and turbulent viscosity term and
movement during the steady-state simulation. Secondly, a second-order backward Euler scheme is used for dis-
sliding mesh method opts during transient blade row simu- cretization of transient terms. Turbulence numerics are
lation. Therefore, transient rotor-stator boundary condition set to be high-resolution. The dynamic model control of
is imposed at the interface of the two domain. The rotor solver is achieved using automatic pressure level infor-
domain rotates at each time step which is defined by the mation, temperature damping, velocity pressure coupling
pitch ratio. Thus the boundary nodes of stator and rotor and Rhie-Chow fourth-order model [38]. Body forces are

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1024 Heat Mass Transfer (2020) 55:1015–1036

Table 4 Comparison of input (non-dimensional design variables) and output (performance parameters) with predicted data from ANN and ANFIS
methods

(a) Non-dimensional input variables and output performance parameters

Non-dimensional input parameters Output parameters
Rs /R2 Rh /R2 rp vs η Lt P (kW )
0.92 0.25 3.10 0.62 0.802 0.22 0.608
0.89 0.27 2.62 0.62 0.821 0.188 1.331
0.86 0.25 2.89 0.68 0.823 0.178 3.191
0.85 0.28 3.14 0.72 0.836 0.188 0.465
0.86 0.22 4.20 0.73 0.778 0.241 1.345
0.9 0.25 2.68 0.74 0.769 0.256 1.289
0.90 0.25 3.42 0.70 0.802 0.181 1.780
0.85 0.26 3.44 0.69 0.803 0.165 1.724
0.8 0.26 3.68 0.66 0.784 0.187 2.169
0.91 0.26 4.14 0.65 0.816 0.154 2.815

(b) ANN results (predicted) from different TFs

LOGSIG PURELIN TANSIG
η Lt P η Lt P η Lt P
0.802 0.218 0.607 0.802 0.201 0.698 0.805 0.221 0.707
0.820 0.186 1.281 0.817 0.203 1.212 0.817 0.188 1.241
0.823 0.179 3.201 0.825 0.188 3.159 0.825 0.168 3.061
0.820 0.188 0.482 0.817 0.173 0.492 0.808 0.187 0.485
0.784 0.241 1.575 0.798 0.187 1.735 0.799 0.241 1.499
0.794 0.221 1.308 0.79 0.197 1.399 0.792 0.204 1.478
0.804 0.180 1.794 0.813 0.170 1.712 0.801 0.176 1.751
0.806 0.169 1.778 0.805 0.177 1.640 0.808 0.172 1.648
0.788 0.185 1.986 0.784 0.209 1.920 0.787 0.194 1.871
0.811 0.158 2.838 0.814 0.163 2.737 0.809 0.158 2.682

(c) ANFIS results (predicted ) from different MFs

Gauss TRI TRAP GBELL
η Lt P η Lt P η Lt P η Lt P
0.802 0.219 0.612 0.802 0.22 0.638 0.802 0.22 0.619 0.802 0.205 0.641
0.821 0.193 1.249 0.821 0.188 1.391 0.821 0.186 1.286 0.821 0.191 1.216
0.823 0.174 3.221 0.823 0.178 3.248 0.823 0.18 3.221 0.823 0.182 3.092
0.836 0.162 0.453 0.836 0.163 0.478 0.836 0.164 0.476 0.823 0.165 0.491
0.778 0.171 1.351 0.778 0.173 1.371 0.778 0.24 1.362 0.778 0.182 1.357
0.78 0.18 1.308 0.78 0.183 1.157 0.78 0.187 1.303 0.78 0.191 1.340
0.802 0.162 1.761 0.802 0.163 1.724 0.802 0.164 1.744 0.802 0.165 1.688
0.804 0.171 1.696 0.803 0.173 1.701 0.803 0.183 1.739 0.803 0.182 1.813
0.784 0.195 2.163 0.784 0.195 2.362 0.784 0.20 2.348 0.784 0.205 2.181
0.816 0.162 2.792 0.816 0.163 2.743 0.816 0.155 2.943 0.816 0.165 2.841
 −3

Training errors ×10 0.193 0.012 0.431 0.230 0.019 0.80 0.196 0.020 0.619 0.198 0.022 0.793
Testing error 0.736 0.721 0.694 0.954 0.793 0.968 0.872 0.737 0.741 0.894 0.743 0.866

volume-weighted average type. The maximum coefficient 4.3 Boundary conditions

loops are 10 for better convergence control. The simula-
tions are assumed to be converged when normalized RMS The flow field is characterized using inlet boundary as total
residuals and conservation target are less than 10−6 and pressure (8 bar) and total temperature (120, 130, 140, and
10−5 respectively. 150 K). The static pressure has opted at the outlet boundary

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1025

Table 5 A comparison with

previous published literature (a) Range of input parameters
Input parameters Range
Fluid Nitrogen
T02 (K) 90 − 180
P02 (bar) 8.0 − 10.0
.
Mass flow rate m (kg/s) 0.01 − 0.09
α2 (Degree) 72 − 82
Rotational speed (rpm) 80, 000 − 150, 000

(b) Output parameters

Present results Ghosh et al. [33] Relative error
(Model 1 / Model 2)
D2 (mm) 24.90 / 25.64 25.19 0.012 / 0.018
D3 (mm) 15.60 / 15.71 15.75 0.01 / 0.003
b2 (mm) 2.11 / 2.14 2.00 0.06 / 0.07
b3 (mm) 1.005 / 1.01 0.996 0.009 / 0.014
U2 (m/sec) 143.38 / 144.62 130.28 0.09 / 0.09
U3 (m/sec) 124.42 / 122.73 121.88 0.02 / 0.01
ψ 0.862 / 0.87 0.88 0.02 / 0.01
φ 0.237 / 0.24 0.241 0.017 / 0.004
ηts 79.20 / 79.41 78.20 0.01 / 0.02
ηtt 80.20 / 79.89 79.51 0.01 / 0.005
β2 (Degree) −54.98 / −55.12 −69.78 0.21 / 0.21
β3 (Degree) −85.04 / −84.76 −85.74 0.01 / 0.01
M2 0.78 / 0.79 0.85 0.08 / 0.07
M3 0.25 / 0.28 0.31 0.194 / 0.097
ns 0.54 / 0.55 0.47 0.13 / 0.15
Total losses (Lt ) 0.226 / 0.208 0.242 0.06 / 0.14
Power (kW ) 3.52 / 3.49 2.83 0.196 / 0.189

Table 6 Other relevant

parameters of turbine and Parameters Model 1 (case 1 and 2) Model 2 (case 3 and 4)
nozzle
Turbine blades (Z) 13 11
Nozzle pitch circle diameter 40.14 40.25
Nozzle outlet diameter 25.58 25.73
Nozzle blades (Zn ) 20 17

Table 7 Node distribution and

its scheme in different blocks Inlet block Outlet block Passage block
of the turbine and nozzle
(Model 1/2) Streamwise Blade-to-blade Streamwise Blade-to-blade Streamwise Blade-to-blade

Turbine 20/20 50/50 25/26 50/54 80/85 40/42

Nozzle 16 / 16 35 / 36 14 / 14 34 / 32 54 / 54 28 / 28

Table 8 Non-dimensional wall

distance of the turbine and Sections y+ value Sections y+ value
nozzle (Model 1/2)
Turbine inlet 8/8 Nozzle inlet 10/10
Near turbine blade 3/3 Near Nozzle blade 5/5
Center of the flow passage 14/14 Center of the flow passage 18/18
Turbine outlet 20/20 Nozzle outlet 15/15

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1026 Heat Mass Transfer (2020) 55:1015–1036

Fig. 6 Computational domain of

nozzle and turbine

Table 9 Grid independence

analysis (Model 1/2) Nozzle (Nodes in Million) Turbine (Nodes in Million) Isentropic efficiency CPU time (hrs)

0.48/0.49 0.81/0.80 0.68/0.73 142/140

0.64/0.68 0.96/0.99 0.72/0.76 146/149
0.72/0.75 1.11/1.12 0.76/0.78 151/153
0.85/0.89 1.24/1.26 0.76/0.79 156/159

Fig. 7 Selected grid size for

turbine and nozzle (a) Model 1
(b) Model 2

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1027

Fig. 8 Schematic of the experimental test-rig

which is in rotating coordinate frame. Periodic boundary Figure 6 illustrates the details of computational domain.
conditions are used for side walls along the pitchwise direc- The implementation of the equation of state significantly
tion which shows physical condition (rotating or stationary) affects the accuracy of numerical solution. In this regard,
of each blade row. The flow direction is normal to inlet sur- Peng-Robinson model is selected for each simulation [40].
face of nozzle and the flow regime is opted to be subsonic.
The turbulence intensity is set to be 5%. The hub, shroud, 4.4 Grid independence study
walls and blade surfaces are assumed to be hydraulically
smooth, adiabatic and no-slip. Additionally, the gen- To validate the accuracy of numerical results, a grid
eral grid interface (GGI) is used for mesh connection. independence test is carried out. In this work, four grid

Fig. 9 Experimental set-up

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1028 Heat Mass Transfer (2020) 55:1015–1036

Fig. 10 a Components of a
turboexpander b Solidworks®
drawing c Instrumentations

Table 10 Specification of
measuring components Parameters Instrument Accuracy Company

Pressure Pressure transducer ±0.2% Endevco

Temperature RT D scanner, ADAM 80 − 600 K Santronix
Mass flow rate Rotameter ±2.5%. Alflow
Rotational speed Speed sensor ±0.005% Micro-Epsilon

Fig. 11 The comparison of (a)

pressure ratio (b) isentropic
efficiency with mass flow rate
for experiment and numerical
result

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1029

5 Experimental test-rig development

The experimental set-up comprises of different compo-

nents such as a screw compressor, an air filter, an air
purifier, precooler, plate-fin heat exchanger (PFHX), surge
tank, heat exchanger, and turboexpander assembly. The
schematic diagram, detailed experimental test set-up, and
other assembly components are shown in Figs. 8, 9, and 10
respectively.
The impurities like dust particles, water, and oil
molecules, etc. present in compressed air, diminishes the
performance of aerodynamic bearings, turbine profile, and
other turboexpander components. Thus, a purification unit
is necessary to overcome these problems. In this regard, a
Fig. 12 Effect of rotational speed on temperature drop screw compressor, air filter (dust and carbon particles
absorber), oil and water particle absorber and purifier sys-
tem together produce clean and compressed air. Addition-
ally, to produce dry and dehumidified air, a Freon precooler
resolutions for both the models are used to discretize is situated after the compressor unit. Thereafter, the process
the computational domain. The isentropic efficiency and gas flows through liquid nitrogen cooled PFHX [42] which
computational time are compared which are mentioned in decreases its temperature up to the desired limit and sub-
Table 9. It is observed that among the four investigated sequently it will be stored in a surge tank. Finally, a high-
grid resolutions, the difference in isentropic efficiency of pressure and low temperature (150 K) air is supplied to the
the third and fourth row is least. After that, computational turboexpander unit.
time is increased with less effect on isentropic efficiency. The experimental test-rig also contains different measur-
Therefore, as a compromise between the accuracy of ing instruments like rotameter, ADAM, temperature sensor,
numerical results and computational time, the third row of pressure gauge, and pressure transducer to measure flow
Table 9 is opted for the remainder of the analysis as shown rate, temperature, and pressure at various positions. Also,
in Fig. 7. speed sensors are used to determine the rotational speed of

Fig. 13 Pressure distribution at

0.5 span for different inlet
temperature

(a) 120 K (b) 130 K

(c) 140 K (d) 150 K

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1030 Heat Mass Transfer (2020) 55:1015–1036

shaft, aerodynamic journal and thrust bearings are used. The

brake compressor is mounted on the opposite end of the
same shaft to dissipate the power produced by the turbine.
This dissipated power shows the refrigerating capacity of
the turboexpander. The turboexpander assembly is placed
inside the bearing housing which is made up of stainless
steel (external diameter 160 mm and height 157.50 mm).
Also, O-rings are used to avoid leakage from the turboex-
pander assembly. Initially, the compressed air is supplied
to the aerodynamic journal and thrust bearings to float the
bearing unit. The bearing pressure is maintained to be 5 bar.
After that, the working fluid is supplied to the turbine at
different mass flow rates.

6 Results and discussions

6.1 Experimental validation

Fig. 14 Area-averaged density variation along the streamwise location
A series of experiments have been performed to analyze
the turbine. Table 10 shows the specifications of various the performance of turboexpander. The numerical results
measuring components. are validated with experimental results by operating the
Chromium-vanadium steel and stainless steel materials turboexpander unit at off-design condition. Figure 11 rep-
are chosen to manufacture shaft and brake compressor, and resents a comparison of experimental and numerical results
other stationary components (nozzle and bearings) respec- of isentropic efficiency and pressure ratio variation with
tively. Aluminum alloy (Al 6082) is used to manufacture mass flow rate. It is noticed that pressure ratio decreases
the turbine blades because of its high fatigue strength which by increasing the mass flow rate whereas isentropic effi-
can sustain the higher rotational speed (60, 000 − 120, 000 ciency increases up to the mass flow rate of 0.04 kg/s after
rpm). To overcome the axial and radial forces act on the that it decreases with increase in mass flow rate at constant

(a) 120 K (b) 130 K

(c) 140 K (d) 150 K

Fig. 15 Mach number (rotating frame) contours at meridional plane

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1031

There are two turbines and nozzle model is designed for

numerical investigation. Model 1 is used for fluid flow visu-
alization at 120 and 130 K (case 1 and 2) inlet temperature
whereas model 2 is used for 140 and 150 K (case 3 and 4)
inlet temperature at same inlet pressure (8 bar) boundary
condition. The results are represented for pressure contours,
Mach number, Reynolds number, Prandtl number, velocity
vectors, etc. Besides, the attention is also taken place on
temperature drop, static enthalpy, static entropy, and density.

6.2 Effect of pressure variation on fluid flow

properties

The fluid flow characteristics inside nozzle and turbine

are three-dimensional and completely different because of
variation in density and viscosity of fluid at cryogenic
temperature. Therefore, it is essential to compare the flow
properties of fluid under these conditions for significant use
Fig. 16 Variation of Reynolds number along the streamwise location in a turboexpander. Figure 13 represents pressure contours,
obtained from all the cases at 50% blade height (span). The
pressure drop inside the nozzle and turbine is approximately
rotational speed. The decrease in isentropic efficiency may 1.2 and 3.4 bar respectively. It is noticed that fluid inlet
happen due to the possibility of choking at a higher mass temperature affects the pressure variation inside the nozzle
flow rate. The maximum isentropic efficiency is obtained and turbine. It happens because of severe changes in density
at a pressure ratio in the range of 4.38 − 4.58. It means and molecular viscosity of cryogenic fluids at high-pressure
that isentropic efficiency increases with decrease in pres- and ultra-low temperature (Fig. 14).
sure ratio up to the mass flow rate of 0.04 kg/s. After that, it Figure 15 illustrates Mach number distribution of the
decreases with increase in mass flow rate. Figure 12 repre- turbine at meridional plane. It indicates that Mach number at
sents the effect of rotational speed of turbine on temperature inlet of the turbine is in subsonic regime which is important
drop. It is noted that temperature drop increases with an for flow stability. The results also show that Mach number
increase in rotational speed. at the inlet of the turbine is highest for case 1 and 2
Fig. 17 Static entropy variation
for different inlet temperature of
the fluids at 0.5 span

(a) 120 K (b) 130 K

(c) 140 K (d) 150 K

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1032 Heat Mass Transfer (2020) 55:1015–1036

(model 1). It happens due to sonic velocity is proportional to incurred. These losses can be quantified in terms of entropy
square root of absolute temperature of the fluid. Figure 16 generation for an adiabatic machine. The entropy generation
illustrates the changes in Reynolds number along with the may occur due to either of these processes: (i) viscous
streamwise location. It is noticed that the Reynolds number friction in boundary layer, (ii) heat transfer and (iii) sudden
is maximum for case 1 near the turbine inlet. For the expansion process (non-equilibrium process). Figure 17
remaining cases, the variation is approximately similar. represents the static entropy contours at 0.50 span. The
entropy remains almost constant throughout the nozzle for
6.3 Loss analysis all cases except near the trailing edges where it increases
slightly but increases rapidly in the turbine near leading and
The main aim of the turboexpander designers is to increase trailing edges where the flow separation exist. The entropy
its efficiency which is directly entitled to the losses generation takes place inside the turbine due to boundary

(a) Span 0.2

(a) Span 0.2

(b) Span 0.5

(b) Span 0.5

(c) Span 0.8 (c) Span 0.8

Fig. 18 Velocity vectors for Model 1 (Pressure side (P ), Suction side Fig. 19 Velocity vectors for Model 2 (Pressure side (P ), Suction side
(S)) (S))

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1033

layer separation and vortices formation. It is highest at the downstream causing the formation of passage vortices in
turbine exit due to non-ideal conditions such as viscous the turbine blade channel flow passages. These vortices
friction, flow reversal, and other dissipative effects. Apart occur at 0.20 and 0.50 span near the leading edge which
from this, boundary layer formation and free stream layer is disappeared when they have strong interaction with
are also responsible for entropy generation which increases rotating blades in the mainstream fluids. It also inflicts
irreversibility in the flow passages. It is also observed that the negative flow incidence which is responsible for large
mixing of fluid streams from pressure and suction side of passage vortices and disturbing the flow towards the
the blade increases shear strain which leads to increase the second turbine blade. Due to this, the viscous friction and
entropy of the fluid for case 3 and 4 (model 2). irreversibility in the flow increases which ultimately leads
Figures 18 and 19 represent the velocity vectors at 0.20, to entropy generation in these regions. Also, scraping flow
0.50 and 0.80 span for model 1 and 2 respectively. The vortices are occurred near the leading edge at 0.50 span in
tip leakage flow appears due to pressure difference which model 2 which generates higher entropy in these regions
further extends to the downstream where it is added to (Fig. 21b). At higher span, these vortices are completely
secondary flow from the suction side. It introduces the disappeared.
non-uniformity and turbulence in fluid flow and a major
cause of losses in rotating machines. It is noticed that 6.4 Effect of pressure variation on the thermal
tip leakage flow occurs near the leading edge in model properties
1. For model 2, it occurs in the middle portion and
starts changing its direction. The secondary flow occurs The law of conservation of energy states that pressure
in vicinity of the blade wall boundary layer because of energy is converted into kinetic energy of the fluid during
relatively higher radial component of Coriolis acceleration the expansion process. In the energy conversion process,
as compared to radial pressure gradient. This phenomenon velocity of fluid increases and enthalpy decreases, as a
is also noticed by Zangeneh et al. [43] for the centrifugal result, decrease in temperature takes place. Figure 20
compressor. The secondary flow also appears due to illustrates the temperature contours of the fluid at four
induced pressure gradient by curvature of blade profile. different inlet temperatures. It shows that temperature drop
The adverse pressure gradient along the downstream is approximately 20, 24, 27, and 30 K for 120, 130, 140,
direction is responsible for flow separation inside the and 150 K inlet temperature fluids. Generally, for normal
turbine which finally results in vortex generation. In the expansion process, the temperature and enthalpy drop
first turbine blade, the velocity vectors deviate towards the depends on pressure change, but at cryogenic temperature,

Fig. 20 Variation of temperature

distribution at different inlet
temperature of the fluids at 0.5
span

(a) 120 K (b) 130 K

(c) 140 K (d) 150 K

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1034 Heat Mass Transfer (2020) 55:1015–1036

Fig. 21 a Static enthalpy b Static entropy variation along the streamwise location

it severely depends on other properties of the fluid. The The isentropic efficiency of turboexpander is calculated
high-temperature nitrogen (case 4) contains higher enthalpy as:
as compared to other cases at the same inlet pressure
(Fig. 21a). During the expansion stage, it decreases at a h1 − h3
ηis = (39)
relatively higher rate due to which higher temperature drop h1 − h3s
takes place. For other cases, temperature drop decreases
due to fluid approaches nearer to its boiling point. In The isentropic efficiency and power output obtained from
those cases, thermodynamic properties of the cryogenic the 1 − D design and numerical results are mentioned in
fluids are randomly changed which severely affect the Table 11.
fluid flow and thermal behavior during expansion process.
Figure 21a represents the static enthalpy variation along
with the streamwise location. It shows that enthalpy drop 7 Conclusion
inside the turbine is higher as compared to nozzle. The
relatively higher enthalpy drop leads to higher velocity In this work, a comprehensive 1-D design procedure
inside the turbine. The main reason behind this is selection of a radial inflow turbine including loss models are
of degree of reaction. In the design process, degree of presented. The work is further incorporated with artificial
reaction is considered to be 0.62 and 0.78 for model 1 and 2 intelligence techniques to estimate the optimum range of
respectively. non-dimensional design variables for higher efficiency of
Isentropic efficiency (ηis ) is an important parameter the radial turbine. Results show that total-to-static efficiency
which can be used to define the irreversibility during expan- and power output of the turbine is increased by 4% and
sion process. The ideal properties of the fluid are obtained 18.9% as compared to the existing one. An experimental
from REFPROP software at corresponding pressure and set-up is developed to determine the performance of the
temperature. turboexpander at different rotational speed using air as a
working fluid. It is noticed that the isentropic efficiency
of the turboexpander is maximum (0.78 − 0.81) when the
Table 11 Efficiency of the turbine and nozzle for different cases (1-D mass flow rate and pressure ratio vary in the range of
/ Numerical) 0.04 kg/s and 4.38 − 4.58 respectively. After that, a CFD
analysis is carried out to visualize the fluid flow and thermal
Isentropic efficiency(ηis ) Power (kW )
characteristics in the flow passage. The analysis deduced
Case 1 (Model 1) 0.771/0.722 3.15/2.87 that the temperature drop (approximately 30 K) is maximum
Case 2 (Model 1) 0.771/0.765 3.15/3.04 for an inlet temperature of 150 K due to the higher enthalpy
Case 3 (Model 2) 0.813/0.783 3.39/3.24 drop. Additionally, a comparison between experimental and
Case 4 (Model 2) 0.813/0.801 3.39/3.36 CFD results (case 4) indicates that the present numerical
studies are reliable. The author believes the proposed work

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Heat Mass Transfer (2020) 55:1015–1036 1035

is quite beneficial for the development of an experimental 17. Khalil KM, Mahmoud S, Al-Dadah R, Ennil AB (2017)
prototype of a cryogenic turboexpander. Investigate a hybrid open-Rankine cycle small-scale axial nitrogen
expander by a camber line control point parameterization
optimization technique. Appl Therm Eng 127:823–836
Compliance with Ethical Standards 18. Ghorbanian K, Gholamrezaei M (2009) An artificial neural
network approach to compressor performance prediction. Appl
Conflict of interests The authors declared that there is no conflict of Energy 86(7-8):1210–1221
interest to any person or organisation. 19. Jovanovic RZ, Bozic IO (2018) Feedforward neural network and
ANFIS-based approaches to forecasting the off-cam energy char-
acteristics of Kaplan turbine. Neural Comput Applic 30(8):2569–
2579
References 20. Asil AG, Pour AN, Mirzaei S (2018) Reliable prediction of
heat transfer coefficient in three-phase bubble column reactor
1. Gavroglu K, Goudaroulis Y (1988) Heike kamerlingh onnes’ via adaptive neuro-fuzzy inference system and regularization
researches at leiden and their methodological implications. Studs network. Heat Mass Transf 54(10):2975–2986
History Philos Sci Part A 19(2):243–274 21. Bartolini C, Caresana F, Comodi G, Pelagalli L, Renzi M, Vagni
2. Linde H (1995) Aus der geschichte der tieftemperaturverfahren- S (2011) Application of artificial neural networks to micro gas
stechnik. KI Luft-und Kaltetechnik 31(9):419–424 turbines. Energy Convers Manag 52(1):781–788
3. Reif-Acherman S (2009) Several motivations, improved proce- 22. Kumar M, Panda D, Behera SK, Sahoo R (2019) Experimental
dures, and different contexts: the first liquefactions of helium investigation and performance prediction of a cryogenic turboex-
around the world. Int J Refrig 32(5):738–762 pander using artificial intelligence techniques. Appl Thermal Eng,
4. Boyce MP (2012) Gas turbine engineering handbook, ch. 8 - 114273
radial-inflow turbines, pp 357–383. Gulf Professional 23. Hongli Z, Yu H, Liang C (2009) Experimental study on a small
5. Carter C, Guillot S, Ng W, Copenhaver W (2001) Aerodynamic Brayton air refrigerator under- 120 ◦ C. Appl Therm Eng 29(8–
performance of a high-turning compressor stator with flow 9):1702–1706
control. In: 37th Joint propulsion conference and exhibit, p 3973 24. Harinck J, Pasquale D, Pecnik R, van Buijtenen J, Colonna P
6. McAuliffe BR, Sjolander SA (2004) Active flow control using (2013) Performance improvement of a radial organic Rankine
steady blowing for a low pressure turbine cascade. In: ASME cycle turbine by means of automated computational fluid dynamic
turbo expo 2004: power for land, sea, and air. American Society design. Proc Institut Mech Eng Part A: J Power Energy
of Mechanical Engineers, pp 1223–1235 227(6):637–645
7. Whitfield A, Baines N (1976) A general computer solution for 25. Yang S, Chen S, Chen X, Zhang X, Hou Y (2016) Study on the
radial and mixed flow turbomachine performance prediction. Int J coupling performance of a turboexpander compressor applied in
Mech Sci 18(4):179–184 cryogenic reverse Brayton air refrigerator. Energy Convers Manag
8. Harvey NW, Rose MG, Taylor MD, Shahpar S, Hartland J, 122:386–399
Gregory-Smith DG (2000) Non-axisymmetric turbine end wall 26. Nithesh K, Chatterjee D (2016) Numerical prediction of the
design: part i-three-dimensional linear design system. In: Journal performance of radial inflow turbine designed for ocean thermal
of turbomachinery, vol 122. American Society of Mechanical energy conversion system. Appl Energy 167:1–16
Engineers, pp 278–285 27. Chang H-M, Ryu KN, Baik JH (2018) Thermodynamic design
9. Suhrmann JF, Peitsch D, Gugau M, Heuer T, Tomm U (2010) of hydrogen liquefaction systems with helium or neon Brayton
Validation and development of loss models for small size radial refrigerator. Cryogenics 91:68–76
turbines. In: ASME turbo expo 2010: power for land, sea, 28. Streit J, Razani A (2013) Thermodynamic optimization of
and air. American Society of Mechanical Engineers, pp 1937– reverse Brayton cycles of different configurations for cryogenic
1949 applications. Int J Refrig 36(5):1529–1544
10. Takao M, Thakker A, Abdulhadi R, Setoguchi T (2006) Effect of 29. Ke C, Xiong L, Peng N, Dong B, Li K, Li J, Liu L (2017)
blade profile on the performance of a large-scale wells turbine for Numerical and experimental study of the performance effects
wave-energy conversion. Int J Sustain Energy 25(01):53–61 of varying vaneless space in high-speed micro turbine stators.
11. Acharya S, Mahmood G (2006) Turbine blade aerodynamics. Gas Cryogenics 88:10–16
Turb Handbook 1:364–380 30. Aungier RH (2006) Aerodynamic design and performance
12. Denton JD (1993) Loss mechanisms in turbomachines. In: analysis of exhaust diffusers. ASME Press
ASME 1993 international gas turbine and aeroengine congress 31. Li Y, Ren X-d (2016) Investigation of the organic Rankine cycle
and exposition. American Society of Mechanical Engineers, (ORC) system and the radial-inflow turbine design. Appl Therm
pp V002T14A001–V002T14A001 Eng 96:547–554
13. Benson RS (1970) A review of methods for assessing loss 32. Moustapha H, Zelesky MF, Baines NC, Japikse D (2003) Axial
coefficients in radial gas turbines. Int J Mech Sci 10:905–932 and radial turbines. Concepts NREC White river junction, VT, vol
14. Costall A, Hernandez AG, Newton P, Martinez-Botas R (2015) 2
Design methodology for radial turbo expanders in mobile organic 33. Ghosh SK (2008) Experimental and computational studies
Rankine cycle applications. Appl Energy 157:729–743 on cryogenic turboexpander. National Institute of Technology
15. Ventura CA, Jacobs PA, Rowlands AS, Petrie-Repar P, Sauret Rourkela, PhD thesis
E (2012) Preliminary design and performance estimation of 34. Baines N (1998) A meanline prediction method for radial turbine
radial inflow turbines: an automated approach. J Fluids Eng efficiency. In: IMECHE conference transactions
134(3):031102 35. Rodgers C (1987) Mainline performance prediction for radial
16. Joseph J, Subramanian S, Vigney K, Prasad B, Biswas D (2018) inflow turbine in small high pressure ratio turbine, tech rep. Von
Thermodynamic wetness loss calculation in nozzle and turbine Karman Institute Lecture Series
cascade: nucleating steam flow. Heat Mass Transf 54(8):2521– 36. Kun L, Sentz R (1985) High efficiency expansion turbines in air
2531 separation and liquefaction plants. In: International conference

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1036 Heat Mass Transfer (2020) 55:1015–1036

on production and purification of coal gas & separation of air, of non-axisymmetric convergent nozzle. Engineering Science and
Beijing, pp 1–21 Technology, an International Journal
37. Ariff M, Salim SM, Cheah SC (2009) Wall y+ approach for 41. Taremi F (2013) Endwall flows in transonic turbine cascades.
dealing with turbulent flow over a surface mounted cube: part Carleton University, PhD thesis
1-low Reynolds number. In: Seventh international conference on 42. Alur S (2012) Experimental studies on plate fin heat exchangers.
CFD in the minerals and process industries, pp 1–6 PhD thesis, National Institute of Technology, Rourkela
38. Ansys C (2013) Ansys CFX solver modeling guide, release 15.0 43. Zangeneh M, Goto A, Harada H (1998) On the design criteria
39. Lei Q, Zhengping Z, Huoxing L, Wei L (2010) Upstream wake– for suppression of secondary flows in centrifugal and mixed flow
secondary flow interactions in the endwall region of high-loaded impellers. J Turbomach 120(4):723–735
turbines. Comput Fluids 39(9):1575–1584
40. Kumar M, Sahoo R, Behera S (2018) Design and numerical Publisher’s note Springer Nature remains neutral with regard to
investigation to visualize the fluid flow and thermal characteristics jurisdictional claims in published maps and institutional affiliations.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]