Energy Conversion and Management 274 (2022) 116418

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Axial sCO2 high-performance turbines parametric design

Jan Stepanek a,b ,∗, Jan Syblik a , Slavomir Entler a,b
a
Department of Energy Engineering, Czech Technical University In Prague, Technicka 1902/4, 160 00 Prague, Czech Republic
b
Institute of Plasma Physics of the Czech Academy of Sciences, Za Slovankou 3, 182 00 Prague 8, Czech Republic

ARTICLE INFO ABSTRACT

Keywords: Supercritical CO2 thermal cycles provide low complexity and small size of equipment compared to the Rankine
Turbine design cycle. However, higher power units of hundreds of megawatts have not yet been implemented. This study
Supercritical carbon dioxide focuses on sCO2 turbines for high power cycles and presents an estimation of their dimensions, efficiency, and
Waste heat recovery
other parameters. For this purpose, a specialized calculation code named TACOS was built. The turbines are
Fusion power plant
designed for inlet pressures of 25 MPa and 30 MPa and inlet temperatures from 300 ◦ C to 600 ◦ C. In total, more
Optimization
Parametric analysis
than 460 high-performance axial turbines were designed in this study. The turbines have been optimized with
shaft power from 10 MW to 2,000 MW (one-flow configuration) for revolutions from 3,000 rpm to 12,000 rpm.
The results show that supercritical CO2 turbines are about one-fifth the size of their steam alternatives with
comparable efficiencies. Turbines with synchronous revolutions (50 Hz) achieve the highest efficiency for shaft
power in one-flow configuration up to 500 MW and this correspond to an efficiency close to 90%. High-speed
turbines are suitable for power levels up to 50 MW with efficiencies of around 87%. The vast majority of
turbines come out as three-stage or four-stage.

1. Introduction et al. published an article describing an optimization for 950 MW axial

turbine for a direct fired sCO2 cycle [7] with inlet temperatures up to
In power industry, two main heat cycles are used, the steam Rankine 1200 ◦ C. Centrifugal turbomachinery is suitable for power up to 300
and the gas Brayton cycle. While the steam cycle uses the heat of MW [8] and is prevalent in investigated turbine designs.
vaporization, and thus the water–steam phase change, the gas cycle Recently, studies have been carried out that propose a Brayton
works with the substance in a single gas phase. The advantage of this supercritical CO2 cycle as the power cycle of the DEMO fusion power
cycle over steam cycle is lower complexity and still relatively high
plant [9–12], and [13]. The thermal output of the main circuit of the
efficiency.
DEMO power plant will be approximately 2300 MWt. This application
The solution to many disadvantages of gas cycles is the use of a
requires a turbine with shaft power greater than 1000 MW or more
gas that has a critical point close to the ambient temperature, and it is
possible to significantly reduce the compression work and increase the turbines with power output in the order of hundreds of megawatts.
efficiency of the cycle. The critical point of carbon dioxide is located at These studies present this alternative cycle as more economical in
approximately 31 ◦ C and 7.37 MPa. At this point, it has a relatively high terms of investment costs and, above all, it has a simpler construction
density at the pressures selected in these cycles, especially in front of and incomparably smaller dimensions than the classical Rankine steam
the compressor (over 800 kg m−3 ). This leads to small-volume flows cycle.
and small dimensions of components such as compressors, turbines, The small size of the components due to the high density of super-
and heat exchangers. These thermal cycles have been studied since the critical CO2 is a common argument to promote this technology, but
1960s [1]. The specific applications and their thermal and pressure estimates of the sizes of these components for high performance are
levels are in Table 1. generally lacking.
Most studies dealing with the design of CO2 turbines focus on The aim and novelty of this study is to map the geometric and
relatively low output power at the level of hundreds of kW up to
thermodynamic parameters of axial high-power turbines. The results
tens of MW [2–4]. These applications often focus on the use of waste
of this study can serve as a performance map that will provide the
heat or renewable energy sources, such as solar energy. Turbines with
reader with sufficient accurate information on sCO2 turbines in a wide
outputs of hundreds of megawatts are rather an exception. An example
power range and the usual boundary conditions for their application.
is the work published by Rogalev [5] or Bidkar [6]. Recently, Selcuk

∗ Corresponding author at: Department of Energy Engineering, Czech Technical University In Prague, Technicka 1902/4, 160 00 Prague, Czech Republic.
E-mail address: [email protected] (J. Stepanek).

https://doi.org/10.1016/j.enconman.2022.116418
Received 19 August 2022; Received in revised form 14 October 2022; Accepted 26 October 2022
Available online 15 November 2022
0196-8904/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Table 1
Applications of sCO2 power cycles [15].
Application Power Maximum Maximum
[MWe] temperature [◦ C] pressure [MPa]
Nuclear 10–300 350–700 20–35
Fossil fuel 300–600 550–1500 15–35
Geothermal 10–50 100–300 15
Solar 10–100 500–1000 35
Waste heat 1–10 200–650 15–35

Fig. 2. Number of stages flow chart.

If we want to get a technically realistic design, it is necessary to limit

the calculation with several restrictions. The numerical calculation
itself does not take into account what is real, but can result in an
extremely large number of stages, excessive stress in the blades, and
so on. These restrictions are listed in Table 3. The maximum value
of the Mach number is chosen to ensure subsonic flow in the nozzle.
Hand in hand with this limitation go maximum values of absolute (𝛼1 )
and relative (𝛽2 ) blade outlet angles, which reduce the formation of too
sloping and thin blades. The restrictions also apply to the stress check
in the shaft and blades.
The initial values of optimized parameters were chosen so that they
were reasonably covered in the first Brute-force part of the calculation
code (described below). 𝛽2 and 𝛼1 angles were investigated in the range
from 12 to 20 degrees (with a step of 0.5), reaction 𝑅 from 0 to 0.6
(with a step of 0.025) and the values of the speed ratio 𝛱 from 0.5 to
Fig. 1. Turbine solver main loop flowchart. 0.96 with a step of 0.01. The estimation of the size of the flow annulus
of the inlet nozzle is based on the boundary conditions.

The output is an effort to bridge the gaps between few available specific 2.2. Main loop
designs and, in addition, to give new information beyond the limits of
these designs. The whole optimization process includes a large number of special-
ized subroutines that are called from the main calculation loop. The
2. Calculation flow chart of this loop is shown in Fig. 1. The variables and their
ranges are entered into the code. The following step is an estimate of
The turbine calculation adopted in this study is improved approach the number of stages. If the number of stages is lower than the limit,
based on turbine calculations used by Dostal [14]. The original code the calculation continues with per-stage calculations. The geometric
was written in Fortran programming language with lookup tables for
characteristics of the blades (length and profile) - are calculated for
sCO2 properties. The code was reviewed, optimized, and overall im-
each stage with relevant part of the code. This also produces values
proved, especially by the possibilities of modern computer technology.
for root diameters, velocities, efficiencies, etc. If all stages pass the
stress and other limits check, the last stage outlet pressure is compared
2.1. Boundary conditions, constants, variables and limits
with the input emission pressure. If the emission pressure differs, the
root diameter is corrected, and the task is repeated iteratively until a
The boundary conditions are the fluid admission and emission
solution is found.
parameters, the flow rate, and the shaft revolutions. The admission
parameters entering the turbine are temperature and pressure. The only
parameter for the turbine outlet is the emission pressure. The emission 2.3. Number of stages
temperature results from the final thermodynamic efficiency of the
turbine. In general, the choice of the number of turbine stages is based on
There are several constants that are used in the calculation. These economic requirements. A higher number of stages results in higher
constants include stator and rotor flow coefficients 𝜇1 and 𝜇2 , and a efficiency but also means higher complexity and investment cost of the
stator and rotor velocity coefficients 𝜙 and 𝜓. The flow coefficients 𝜇1 turbine. At the beginning, the first and the last stage is calculated and
and 𝜇2 represent flow reduction through stator and rotor nozzles, as can then the number of stages is determined based on these results and
be seen from Eqs. (7) and (11). The values of these flow coefficients are the total enthalpy drop. The flow chart of the subroutine flow chart is
equal to 0.96 [2,14]. The velocity coefficients 𝜙 and 𝜓 are portions of shown in Fig. 2.
the real and adiabatic velocities for the stator and rotor, respectively Basically, the number of stages is calculated as the total isentropic
(see Eqs. (8) and (12)). Their values were assumed to be 0.97 [14]. enthalpy drop divided by the average drop, which is calculated from
The optimized variables were degree of reaction 𝑅, velocity ratio 𝛱. the first and the last stage enthalpy drop. Isentropic enthalpy drop per
The degree of reaction 𝑅 represents the portion of the total adiabatic individual stage can be obtained through known diameter, revolutions
enthalpy drop processed within the rotor nozzle (see Eq. (4)). and velocity ratio 𝛱 using Eqs. (1), (2), and (3). The velocity ratio 𝛱

2
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 3. Angles and velocities in turbine stage.

Fig. 5. Example of blade lengths plan.

Fig. 4. Velocity triangles.

𝑤2 = 𝜓𝑤2𝑎𝑑 𝑢2 = 𝜋(𝐷𝑟𝑜𝑜𝑡 + 𝑙2 )𝑛 (12)
√
2 2
𝑐2 = 𝑤2 𝑢2 − 2𝑢2 𝑤2 𝑐𝑜𝑠(𝛽2 ) (13)
is usually chosen in the range of 0.7 to 1.0, but in this case it is one of
the optimized input parameters. Relative velocities 𝑤 and their angles 𝛽 from Eqs. (9), (12), and
(15) (𝛽2 is a parameter) are related to the rotor geometry. Analogously,
the absolute velocities 𝑐 and the corresponding angles 𝛼 from Eqs. (8),
𝑢 = 𝜋𝐷𝑛 (1) (13), and (17) (𝛼1 is a parameter) are related to the stator geometry. An
𝑢 example of a blade plan is shown in Fig. 5. It is evident from this figure
𝛱= (2)
𝑐𝑎𝑑 that the next blade is always longer than the previous one. This does
2
𝑐𝑎𝑑 ( )2 not automatically result from the calculation. In some cases, the next
1 𝑢
𝑑ℎ𝑎𝑑 = = (3) blade may be shorter than the previous one. This result must be fixed.
2 2 𝛱
Firstly, a new corrected blade length is calculated using a minimum
2.4. Blade length and velocities growth ratio and the new length is used for stage reaction recalculation.

The calculation of velocity triangles is the essential step in the geo-

metric design of the turbine stages. The velocity triangle is a graphical 𝑐1𝑎 = 𝑐1 𝑠𝑖𝑛(𝛼1 ) 𝑐1𝑢 = 𝑐1 𝑐𝑜𝑠(𝛼1 ) (14)
( )
representation of the absolute (𝑐), relative (𝑤), and peripheral velocity 𝑐1𝑢 − 𝑢1
𝛽1 = 90 − 𝑎𝑟𝑐𝑡𝑔 (15)
(𝑢) of the fluid at the examined point. An example of velocity triangles 𝑐1𝑎
is shown in Fig. 4. These points are (1) stator exit/rotor inlet and 𝑤2𝑎 = 𝑤2 𝑠𝑖𝑛(𝛽2 ) 𝑤2𝑢 = 𝑤2 𝑐𝑜𝑠(𝛽2 ) (16)
(2) rotor exit/next stator inlet (Fig. 3). The positions of individual ( )
𝑤2𝑢 − 𝑢2
velocities in a triangle give the corresponding angles. 𝛼2 = 90 − 𝑎𝑟𝑐𝑡𝑔 (17)
𝑤2𝑎
Another optimized parameter is the degree of reaction 𝑅, which is a
portion of the adiabatic enthalpy drop processed in the rotor (Eq. (4)).
The rest of the drop belongs to the stator (Eq. (5)). 2.5. Long blades

Since the peripheral velocity depends on radius, the velocity trian-

𝑑ℎ𝑎𝑑𝑅 = 𝑑ℎ𝑎𝑑 𝑅 (4) gles, and the corresponding angles change along the blade height. This
results in blade deformation. A relatively large number of analytical
𝑑ℎ𝑎𝑑𝑆 = 𝑑ℎ𝑎𝑑 (1 − 𝑅) (5)
√ approaches are used for the basic design of the blade geometry. The
𝑐1𝑎𝑑 = 2𝑑ℎ𝑎𝑑𝑆 + 𝑐02 (6) design depends on the type of stage, load, properties of the fluid, etc.
To avoid losses of swirling of the working fluid, it is necessary to design
The absolute adiabatic velocity value is the result of the energy
the stage so that the so-called radial equilibrium condition is met. The
balance within the stator in Eq. (6). The length of the blade can be long blade is defined in this work as a blade that is longer than the
calculated from conservation of mass if the velocities and angles are limit length given by Eq. (18).
known already; see Eqs. (7) and (11).
𝑙𝑙𝑖𝑚 = 𝜖𝑙 𝐷 (18)
[√ ]
2 + 4𝑚̇ The design of long blades is similar to the design of short blades but
𝑙1 = 0.5 𝐷𝑟𝑜𝑜𝑡 − 𝐷𝑟𝑜𝑜𝑡 (7)
𝜋𝜇1 𝑐𝑐1𝑎𝑑 𝑠𝑖𝑛(𝛼1 )𝜌1𝑎𝑑 for several different radii. A frequently used assumption is that the real
flow is very close to the potential flow. This is achieved if the velocity
𝑐1 = 𝜙𝑐1𝑎𝑑 𝑢1 = 𝜋(𝐷𝑟𝑜𝑜𝑡 + 𝑙1 )𝑛 (8) rotor 𝑐⃗ is close to 0. This is shown in Eq. (19).
√
2 2 ( )
𝑤1 = 𝑐1 𝑢1 − 2𝑢1 𝑐1 𝑐𝑜𝑠(𝛼1 ) (9) 𝜕𝑐 𝜕𝑐𝑟 𝜕𝑐𝑎 1 𝜕(𝑟𝑐𝑢 ) ⃗
√ 𝑐𝑢𝑟𝑙𝑐⃗ = − 𝑢 ⃗𝑖 + − 𝑗⃗ + 𝑘=0 (19)
𝜕𝑎 𝜕𝑎 𝜕𝑟 𝑟 𝜕𝑟
𝑤2𝑎𝑑 = 2𝑑ℎ𝑎𝑑𝑅 + 𝑤21 (10)
[√ ] Eq. (19) is fulfilled when the product 𝑟𝑐𝑢 is constant in the radial di-
2 + 4𝑚̇ rection. This constant is called the velocity circulation 𝛤 (see Eq. (20)).
𝑙2 = 0.5 𝐷𝑟𝑜𝑜𝑡 − 𝐷𝑟𝑜𝑜𝑡 (11)
𝜋𝜇2 𝑤2𝑎𝑑 𝑠𝑖𝑛(𝛽2 )𝜌2𝑎𝑑 New velocities and angles along the blade (Eqs. (23) and (24)) can be

3
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 6. Long blade profiles.

Fig. 7. Blade design geometry.

obtained with varying radius for the calculation of 𝑢(𝑟) and 𝑐𝑢(𝑟) (Eqs.
(21), (22)).

Fig. 8. Blade profile flow chart.

𝛤 = 𝑐𝑢 𝑟𝑚𝑒𝑎𝑛 = 𝑐𝑜𝑛𝑠𝑡. (20)
𝑢(𝑟) = 2𝜋𝑛𝑟 (21)
𝛤 these circles have their parametrically given coordinates and the spine
𝑐𝑢(𝑟) = (22) is shifted for smoother inlet in high-pressure side of the blade. The
𝑟 ( )
𝑡𝑔(𝛼)𝑟𝑚𝑒𝑎𝑛 width of the cascade throat 𝐴 is the point with the highest velocity
𝛼(𝑟) = 𝑎𝑟𝑐𝑡𝑔 (23) in the lattice. At this point, the maximum velocity is calculated and
𝑟
( ) checked for subsonic condition in the code. The number of blades and
𝑐𝑢(𝑟) − 𝑢(𝑟)
𝛽(𝑟) = 90 − 𝑎𝑟𝑐𝑡𝑔 (24) other geometric parameters are changed if the subsonic condition and
𝑐𝑎 stress check fail.
The resulting angles are sent to the blade profile subroutine. The The number of blades is also closely related to the blade profile. This
blade design subroutine calculates the envelope for each segment sep- number is based on the strength calculation and the maximum allowed
arately in the same way as described in Section 2.6. An example of speed of the fluid in the gap between the profiles. This is simplistically
overlaid profiles of long blades is shown in Fig. 6. indicated in Eq. (25). During the iteration process, these values are
checked and, if necessary, the number of teeth is increased or decreased
2.6. Blade 2D profile by 2. For example, the number of stator blades should be even and the
number of rotor blades odd. The reason is to prevent resonances in the
Quality blade profile design is a complex challenge that includes blade grid. The resulting number of blades from the stator is used as
an initial estimate of the number of rotor blades, which is adjusted by
fluid dynamics, material engineering, stress optimization, and others.
one.
A fast technique for approximate but sufficient blade profile creation
𝜋𝐷
has been developed for basic geometry and stress assessment. 𝑛𝑏 = = 𝑓 (𝑆𝑜𝑆, 𝛼1 ∕𝛽2 , 𝑆𝑡𝑟𝑒𝑠𝑠) [−] (25)
𝑡
Almost all basic characteristics, velocities, angles, etc., are based
only on thermodynamic calculations. Blade design is crucial for stress The stress itself is controlled by a set of concave hull processing
algorithms. It takes a set of points from the blade design subroutine
calculation and especially for geometry parameters such as chord
and calculates the polygon area, the polygon centroid (center of mass),
length and resulting blade width.
and the polygon inertia (second moment of the area). Using these
The blade profile design method adopted in the code is based on the
data and the known momentum change of the flow, the polygon bend
inscribed circles envelope with a basic geometry configuration inspired
module and the bend stress are subsequently calculated. Normal stress
by Gribin et al. [16]. An example blade profile is shown in Fig. 7.
is calculated from the known root area of the blade, rotation speed, and
Compared to this study, there is a shift of the central spine in the its mass (see Fig. 8).
profile, depending on its shape. This is a simplification over original
airfoil modifications to improve the suction side of the vane. 2.7. Stage efficiency
The input parameters for the blade design are velocity angles.
Temperatures, pressures, velocities, and blade length are important in One of the most important results of individual designs is the
the second step for stress evaluation and check. The main design aim thermodynamic efficiency of the turbine. It basically consists of partial
is to create a smooth transient from the inlet angle to the exit angle efficiencies of individual stages. The total thermodynamic efficiency
from the lattice with respect to structural strength. This smooth ‘spine’ of the turbine is defined as the proportion of the real and adiabatic
is conducted through four important circles: the leading edge circle (isentropic) enthalpy drop.
with radius 𝑅1 , the maximum radius circle 𝑅𝑚𝑎𝑥 , the cascade throat 𝑑ℎ𝑟𝑒𝑎𝑙 ℎ − ℎ𝑒
circle with radius 𝑅𝑢𝑡 , and the tailing edge circle with radius 𝑅2 . All 𝜂𝑡𝑑 = = 𝑎 [−] (26)
𝑑ℎ𝑎𝑑 ℎ𝑎 − ℎ𝑒𝑎𝑑

4
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Table 2 Table 3
Boundary conditions. Calculation restrictions.
Boundary condition Value Restriction Value
𝑝𝑎 [MPa] 25/30 min. 𝛼1 , 𝛽2 [◦ ] 12
𝑇𝑎 [◦ C] 300–600 (step 100) max. Mach number [–] 0.9
𝑝𝑒 [MPa] 8 max. torsional stress [MPa] 80
rpm [min−1 ] 3000/6000/9000/12,000 max. bend stress [MPa] 20
𝑚̇ [kg s−1 ] 100–12,000 (variable step) max. normal stress [MPa] 300
max. no. of stages [–] 10

The adiabatic enthalpy drop of the individual stage calculated with

Eq. (3) is reduced by several enthalpy losses. These considered losses
in this study are stator loss, rotor loss, outlet loss, ventilation loss, and
trailing edge loss. The formulas for the thermodynamic efficiency of the
stage are gathered in Eqs. (27) to (33) and these equations are based
on the equations used in [14].

2
𝑐1𝑎𝑑
𝑧𝑠 = (1 − 𝜙2 ) (27)
2
𝑤2
𝑧𝑟 = 2𝑎𝑑 (1 − 𝜓 2 ) (28)
2
𝑐22
𝑧𝑜 = (29)
2
𝑐02
𝑑ℎ𝑎𝑑𝑠𝑡𝑎𝑔𝑛 = 𝑑ℎ𝑎𝑑 + (30)
2
⎛ 0.25 + 0.85 𝐷1 ⎞
1.5 ⎟ −3
𝜁𝑡 = ⎜ 103
10 (31)
⎜ 𝑙1 𝑠𝑖𝑛(𝛼1 ) ⎟
⎝ ⎠ Fig. 9. Example of a Nelder–Mead convergence progress.
( )
𝑙2 2
𝜁𝑣 = 0.77 (32)
𝐷2
𝑑ℎ𝑎𝑑𝑠𝑡𝑎𝑔𝑛 − 𝑧𝑜 − 𝑧𝑠 − 𝑧𝑟 these results and searches for a better optimized value of 𝜂𝑡𝑑 . It can
𝜂𝑠𝑡𝑎𝑔𝑒 = − 𝜁𝑡 − 𝜁𝑣 (33) simply be said that the brute-force parts find a global maximum of
𝑑ℎ𝑎𝑑𝑠𝑡𝑎𝑔𝑛
efficiency and the Nelder–Mead solver finds the local maximum near
With the obtained stage efficiency, the real enthalpy drop of the the global one.
stage is calculated, and subsequently the output values of enthalpy, The brute-force calculation is written with a focus on parallel com-
pressure, and temperature. These conditions are used as input for the puting, as the number of investigated combinations increases exponen-
next stage. tially with the number of variables. In practice, this means millions
of possible combinations. These combinations are divided between the
2.8. Optimization CPU cores, and the cores calculate the assigned turbine configurations.
If the combination of parameters is valid (the design has passed all the
The turbines optimization was performed in a custom built script checks), the result is stored in shared tables. Otherwise, the result is
written in Python programming language together with the CoolProp marked as invalid if the calculation fails.
fluid property tables [17]. The code is called TACOS (Turbine and Both the brute-force method and automatic optimization work on
compressor optimization software). The code is a subroutine for the the same principle of applying limits. If the input combination of
heat cycle optimization code used in previous studies [10,11]. The code parameters results in exceeding one of the limits, the corresponding
is independently executable and is composed of two parts. The first part function returns an efficiency value of 0. If the design is valid from the
uses the brute-force technique to find a set of possible solutions, and the point of view of the limits, the calculated efficiency value is returned. In
second part of the code uses the automatic gradient-based Nelder–Mead the case of brute-force, the result is an n-dimensional array of efficiency
values. The maximum non-zero value corresponds to the best combina-
algorithm for fine optimization of the found solution.
tion of input parameters. These values are sent to the automatic mode,
For the optimization of turbines and other components of thermo-
which searches for the minimum of the function 𝑎𝑚𝑖𝑛 = (1 − 𝜂) using the
dynamic cycles, genetic and PSO algorithms and recently also trained
Nelder–Mead minimizing algorithm. If the input parameters are invalid
neural networks in combination with CFD [18–20] designs are com-
and the efficiency value is 0, this value is equal to 1 and the algorithm
monly used. These approaches are validated (at least for steam and
chooses a new solution vector for the next iteration.
gas turbines) and give a detailed view of the design of a particular
Fig. 9 shows an example of the convergence of thermodynamic
turbine. In the case of parametric analysis of a large number of turbine
efficiency. Together with the efficiency, the courses of the parameters
designs, such design accuracy and depth are not always necessary. The entering the solution and the root diameter of the blades, which is the
chosen approach in this study enables the program to be effectively result of internal iterations of the geometry, can be seen here. From this
coupled with thermodynamic optimizations of entire thermodynamic particular progress, it is evident that in this case changes in inputs have
cycles as part of a large optimization package which is developed at practically no effect on increasing efficiency beyond the 80th iteration,
CTU in Prague. and the tolerance of the solution can thus be increased.
Fig. 10 shows the code flow chart. The purpose of the first brute
force part is to find the global maximum of the turbine efficiency (max. 2.9. Validation
𝜂𝑡𝑑 ) based on given variables and their ranges. The result is a set of
variable values which are sent to the automatic solver. The gradient- Code validation for the design of axial sCO2 turbines at this power
based Nelder–Mead solver [21], also called a downhill algorithm, takes level is somewhat difficult due to the lack of available designs. A

5
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Table 4
Comparison of 450 MWe turbine design by Bidkar [6] and by the TACOS code.
Bidkar [6] TACOS r. diff. [%]
HPT
𝑝𝑎 [MPa] 25.06
𝑇𝑎 [◦ C] 700.0
𝑝𝑒 [MPa] 12.96
𝐷𝑟𝑜𝑜𝑡 [m] 0.762 0.819 +7
𝑙1 [mm] 71.1 80.0 +13
𝑙𝑛 [mm] n/a 119.2 n/a
𝑛𝑠𝑡𝑎𝑔𝑒𝑠 [–] 3 3 0
𝜂𝑡𝑑 [–] 90.6 89.2 −2
Power [MW] n/a 164.5 n/a
LPT
𝑝𝑎 [MPa] 12.96
𝑇𝑎 [◦ C] 680.0
𝑝𝑒 [MPa] 6.71
𝐷𝑟𝑜𝑜𝑡 [m] 0.762 0.827 +9
𝑙1 [mm] n/a 90.0 n/a
𝑙𝑛 [mm] 137.2 152.9 +11
𝑛𝑠𝑡𝑎𝑔𝑒𝑠 [–] 4 4 0
𝜂𝑡𝑑 [–] 91.6 88.8 −3
Power [MW] n/a 155.9 n/a

Fig. 10. Optimization code flow chart.

450 MWe turbine design by Bidkar [6] was chosen to compare the re-
sults of the TACOS code. In this work, the authors dealt with the design
of a powerful turbine for the use of heat from coal combustion. The
boundary conditions for the HPT and LPT part of the turbine included
in this paper were used for the calculation in order to reproduce the
results. The result of this comparison is in Table 4. The turbine was
designed as two-flow, so the power outputs in the table correspond to
half the shaft power of the entire turbine.
On the basis of these results, it can be concluded that they are in
good agreement. The turbine designed by TACOS has slightly larger di-
mensions (those available for comparison) and slightly lower efficiency. Fig. 11. Maximum power on revolutions.
This is probably due to the unavailability of parameters such as the load
coefficient and others of the original design. These parameters were
chosen in the same way as in this study. Since this is a comparison
The average slopes of the flow rate characteristics are in the range from
with a very detailed study, the result can be considered satisfactory.
6.3 kg s−1 MW−1 to 13 kg s−1 MW−1 . These values correspond to the
limit values of the inlet pressures and temperatures 600 ◦ C/30 MPa and
3. Results 300 ◦ C/25 MPa respectively.
One of the most important results is the thermodynamic efficiency
3.1. Thermodynamic parameters of a turbine. Fig. 13 shows the dependence of the thermodynamic
efficiency on the output power. Individual efficiency-performance char-
The found designs can be sorted according to many aspects, but acteristics show the expected influence of the inlet pressure and tem-
the main ones are revolutions, inlet pressures, and inlet temperatures, perature on thermodynamic efficiency. The efficiency decrease with
as Table 2 shows. The first conclusion that can be observed from lower inlet pressures and temperatures is obvious but only beyond the
valid calculated configurations is the maximum performance achiev- maximum efficiency point, which is situated at the top of the relatively
able for given revolutions. Higher revolutions go hand in hand with flat part of the characteristic. At least at low revolutions (Fig. 13,
higher stresses on blades, and this results in lower maximum power. 3000 rpm), the situation is opposite below this maximum point. In
The dependence of the maximum power on the revolutions is shown this region, the ventilation and trailing edge losses are relatively high
in Fig. 11. This figure also demonstrates the obvious influence of causing a shift of the expected characteristics for smaller turbines with
admission pressure on maximum power. lower volumetric flow rates.
Along with increasing power output, the required flow also in- The maximum resulting thermodynamic efficiencies for individual
creases, which is a function of admission pressure and thermodynamic turbine revolutions are listed in Table 5. These points are the inflection
efficiency for the given boundary conditions. These dependencies for points of Fig. 13. It is evident from this table and figure that the
individual revolutions are shown in Fig. 12. The maximum power in maximum efficiency decreases with rising revolutions. This behavior
Fig. 12 for 3000 rpm is limited to reasonable 1000 MW for reasons that is a consequence of higher velocities in the nozzles and smaller size,
will be discussed further. As the admission temperature and pressure which results in higher ventilation and tailing edge losses. This is in line
decrease, the thermodynamic efficiency and the adiabatic enthalpy with the assumption that small high-speed turbines are more suitable
drop decrease, leading to an increase in the required mass flow rate. radial ones.

6
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 12. Mass flow rates.

Fig. 13. Thermodynamic efficiency.

Table 5 Table 6
Maximum efficiencies [%] (𝑝𝑎 = 30∕25 MPa). Turbine BEP power [MW] (𝑝𝑎 = 30∕25 MPa), related to Table 5.
300 ◦ C 400 ◦ C 500 ◦ C 600 ◦ C 300 ◦ C 400 ◦ C 500 ◦ C 600 ◦ C
3,000 rpm 89.82/89.69 89.92/89.8 90.01/89.89 90.07/89.96 3,000 rpm 155/137 261/230 312/273 360/314
6,000 rpm 88.36/88.19 88.52/88.4 88.66/88.56 88.76/88.66 6,000 rpm 61/45 90/67 92/80 106/92
9,000 rpm 87.23/87.10 87.41/87.2 87.57/87.43 87.68/87.57 9,000 rpm 30/26 38/33 60/39 70/45
12,000 rpm 85.99/85.83 86.39/86.22 86.69/86.46 86.78/86.64 12,000 rpm 19/17 25/22 30/26 34/30

Resulting 𝑅 value of all optimized turbines is in the range from 0.4 3.2. Geometric parameters
to 0.6, as shown in Fig. 14. This means that all optimized turbines are
reaction turbines with a similar enthalpy drop within the stator and One of the most popular advantages of the sCO2 turbines is their
rotor. compactness. The dimensions of individual stages and the resulting

7
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 14. Degree of reaction.

Fig. 15. Root diameters.

dimensions of the turbine are the main result of this study. The low at some point (from the perspective of thermodynamic efficiency) it is
space requirements are the result of small volume flows due to the high more appropriate to reduce the root diameter and lengthen the blade.
density of sCO2 . The volume of the turbine (without the casing) is given Figs. 16 and 17 show the resulting blade lengths from the first and
by the root diameter, first and last blade length, number of stages, and last stages. These blade lengths are directly related to the flow rate
height of the blade profile for each stage. The blade profile height is and the corresponding root diameter, as described above. These plots
calculated from the 2D profile design described in Section 2.6. This is clearly show an increase in blade length with a decrease in admission
primarily a matter of stress caused by a change in the momentum of pressure. The relative increase in blade lengths in the first stage is
the flowing fluid. In contrast, root and tip diameters, and number of approximately 20–40%. In the last stage, this difference is not so great;
stages are more a matter of thermodynamics, i.e. velocity and density. it is up to 20%. This is due to the fact that the pressure in the last stage
The values of the root diameters are shown in Fig. 15. The minimum approaches the emission pressure, which is the same for both cases.
value of the root diameter was given by the maximum torsional stress In addition, Fig. 18 shows the total length of the turbine blade part.
on the shaft, but this limit was not a limiting factor in any case. The This length is calculated as the sum of the blade widths of the individual
value of the root diameter is given mainly by the revolutions and stages. These lengths are purely dependent on the stress calculations
the mass flow rate. This requires the corresponding flow area in the and profile shapes of the individual blades. We can state that the length
‘‘annulus’’ and this results in values of diameters and associated blade of the turbine increases exponentially with power (respectively, with
lengths. mass flow) as the force effects on the individual blades increase.
As can be seen in Fig. 15, the root diameters have a certain max- The number of turbine stages is directly related to the determination
imum for different revolutions. Behind this peak there is a slight of the total length above. This figure is shown in Fig. 19. As can be seen,
decrease as a result of the increase in blade lengths. This is due to the low-power turbines hit a limit of ten stages during optimization. Most
exponential growth of the flow area with increasing blade length, when designs then contained a smaller number of stages, usually 4 to 6. Due

8
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 16. The first blade length.

Fig. 17. The last blade length.

to the smaller pressure drop, turbines with a lower admission pressure negated, but the torques, of course, add up. Common thrust bearings
then of course have a lower number of stages, usually by one. High- are available for diameters from 20 to 1000 mm and for maximum
power turbines then required a higher number of stages for a given operating forces from 0.5 to 50 tons (4.9 to 4900 kN) [22]. In this
revolutions and inlet temperature due to higher power and greater work, it is assumed that the maximum thrust force for a one-flow
stress per stage. turbine arrangement is 50% of the maximum operating load for the
The axial forces that act on the turbine bearings are caused by appropriate bearing. This limit is 900 kN for 3000 rpm, 490 kN for
the change in the momentum of the flowing gas in the individual 6000 rpm, 235 kN for 9000 rpm and 140 kN for 12,000 rpm [23]. The
stages. The resulting value of this force is the sum of axial forces per axial force dependencies along with the thrust bearing limits are plotted
stage. There are several ways to reduce these axial (thrust) forces on in the graphs in Fig. 20. The second important limit that speaks to a
the bearings. A balance piston can be used for smaller and medium- two-flow arrangement for high-power turbines at given revolutions is
sized rotary machines; for larger machines, it is suitable to choose the the steep drop in efficiency, as shown in Fig. 13.
so-called two-flow rotor. The balance piston uses pressure differences Fig. 21 represents the methodology for dividing a turbine into
to suppress the thrust force, the two-flow arrangement divides the one-flow and two-flow designs. This method is based on doubling
flow into two opposite (mirrored) rotors, and the axial force is thus performance with two mirrored turbines while maintaining the same

9
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 18. Turbine length.

Fig. 19. Number of stages.

thermodynamic efficiency. Moreover, the two-flow design does not root diameter is given by the flow cross sections and is much larger
need a balance piston for thrust management. An important point than the stress limit. Fig. 22 brings an illustrative final comparison of
here is the breakpoint, from which the design of turbines as a two- steam and sCO2 turbine sizes with first-to-last blade lengths.
flow is more suitable. Together with the already presented efficiencies
with corresponding performances for one-flow design (Tables 5, and
4. Conclusion
6), the performance for two-flow design with maximum efficiency
(BEP–best efficiency point) is listed in Table 7. The breakpoints and
their respective performances are in Table 8. As mentioned above, The presented work aimed to create a parametric map of axial tur-
in the two-flow design, the torques add up, which could lead to the bines for sCO2 and thus create an approximate view of the dimensions
requirement for a shaft diameter larger than that calculated for the one- and properties of these devices. For these purposes, a computational
flow design. However, this did not happen in any case. The minimum code named TACOS was created that is able to optimize these turbines

10
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 20. Axial force.

These results show the suitability of individual turbine config-

urations for certain power levels. In general, it can be said that
synchronous turbines are suitable for large outputs approaching
1000 MW, turbines with higher revolutions are more suitable
for outputs in the order of several tens of megawatts to several
hundred megawatts.
2. The resulting power range, which was given by the sCO2 mass
flow, covers the range from 10 MW to 1000 MW in one-flow
configuration. These results are supplemented by an overview of
the geometric values of the designs. Based on this, the size and
basic parameters of the turbine for a specific application can be
estimated.
Fig. 21. Turbine one/two-flow design according to the best thermodynamic efficiency. 3. The breaking point for the two-flow configuration is outlined.
High-power turbines are more suitable to choose in a two-flow
Table 7
Two-flow turbine BEP power [MW] (𝑝𝑎 = 30∕25 MPa).
configuration with flow split. The breaking point shows the
◦
300 C ◦
400 C 500 C◦ ◦
600 C performance values from which such a configuration is more
3,000 rpm 311/275 523/460 624/546 720/628 advantageous from the point of view of efficiency.
6,000 rpm 122/90 180/135 184/161 212/185 4. It is shown, that sCO2 turbines are approximately one-fifth in
9,000 rpm 60/53 76/67 121/79 140/91
size and have at least comparable power to their steam alter-
12,000 rpm 39/35 50/44 60/52 69/60
natives. This statement is rather indicative, as it depends on the
specific boundary conditions and the specific design of the given
steam turbine.
Table 8
One/Two-flow turbine breakpoint power [MW] (𝑝𝑎 = 30∕25 MPa).
300 ◦ C 400 ◦ C 500 ◦ C 600 ◦ C It can be stated that high-power sCO2 cycles are a promising al-
3,000 rpm 227/207 321/278 397/349 484/392 ternative to steam cycles for nuclear and now for fusion power plants.
6,000 rpm 89/74 104/99 146/119 175/144 They promise small size, low complexity and investment costs. The only
9,000 rpm 45/39 63/57 81/64 97/59
12,000 rpm 32/17 29/36 50/43 45/30 obstacle is the few realized practical applications of this technology and
the virtually complete absence of development of cycles with power
outputs in the order of hundreds of megawatts.
for specified boundary conditions, including the design of a 2D blade In the near future, a similar study is planned regarding axial com-
profile. pressors of large capacities. The optimization package, which includes
The main conclusions of this work can be summarized as follows:
the TACOS code, also includes the design of microchannel recuperators,
1. In this parametric study, more than 470 turbines with different which should lead to more comprehensive studies of energy conversion
revolutions, inlet pressures, and temperatures were designed. systems based on sCO2 .

11
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Fig. 22. Illustrative comparison of steam and sCO2 turbine sizes with first-to-last blade lengths.

Abbreviations 𝑝𝑒 [Pa] - Emission pressure

𝑇𝑎 [◦ C] - Admission temperature
𝑧𝑠 [J kg−1 ] - Stator loss
CTU - Czech technical university in Prague
𝑧𝑟 [J kg−1 ] - Rotor loss
BEP - Best efficiency point
DEMO - Demonstration fusion power plant 𝑧𝑜 [J kg−1 ] - Outlet loss
rpm - Revolutions per minute 𝜁𝑡 [–]([%]) - Tailing edge loss
sCO2 - Supercritical carbon dioxide 𝜁𝑣 [–]([%]) - Ventilation loss
SoS - Speed of sound 𝑚̇ [kg s−1 ] - Mass flow rate
𝜇1 [–] - Stator flow coefficient
𝜇2 [–] - Rotor flow coefficient
𝜙 [–] - Stator velocity coefficient
Nomenclature 𝜓 [–] - Rotor velocity coefficient
𝑛 [s−1 ] - Revolutions per second
𝑏 [m] - Blade width
R [–] - Degree of reaction
𝑡 [m] - Blade pitch
𝛼1 [◦ ] - Stator tailing angle
𝛼2 [◦ ] - Stator leading angle
𝛽1 [◦ ] - Rotor tailing angle CRediT authorship contribution statement
𝛽2 [◦ ] - Rotor leading angle
𝑐 [m s−1 ] - Absolute velocity Jan Stepanek: Conceptualization, Methodology, Software, Writ-
𝑤 [m s−1 ] - Relative velocity ing – original draft. Jan Syblik: Conceptualization, Writing – review
𝑢 [m s−1 ] - Peripheral velocity & editing. Slavomir Entler: Conceptualization, Writing – review &
𝛤 [m2 s−1 ] - Velocity circulation editing.
𝛱 [–] - Velocity ratio
𝜂𝑡𝑑 [–]([%]) - Thermodynamic efficiency Declaration of competing interest
𝑟 [m] - Radius
𝐷𝑟𝑜𝑜𝑡 [m] - Root diameter The authors declare that they have no known competing finan-
𝐷1 [m] - Stator mean diameter cial interests or personal relationships that could have appeared to
𝐷2 [m] - Rotor mean diameter
influence the work reported in this paper.
𝑙1 [m] - Stator blade length
𝑙2 [m] - Rotor blade length
ℎ𝑎 [J kg−1 ] - Admission enthalpy Data availability
ℎ𝑒 [J kg−1 ] - Emission enthalpy
𝑝𝑎 [Pa] - Admission pressure No data was used for the research described in the article.

12
J. Stepanek et al. Energy Conversion and Management 274 (2022) 116418

Acknowledgment [11] Stepanek J, Entler S, Syblik J, Vesely L, Dostal V, Zacha P. Comprehensive

comparison of various working media and corresponding power cycle layouts
for the helium-cooled DEMO reactor. Fusion Eng Des 2021;166:112287. http:
The work was supported by European Regional Development Fund
//dx.doi.org/10.1016/j.fusengdes.2021.112287, URL https://www.sciencedirect.
- Project ‘‘Center for Advanced Applied Science’’ (No. CZ 02.1.01/0.0/ com/science/article/pii/S0920379621000636.
0.0/16-019/0000778) and by the Strategy AV21 of the Czech Academy [12] Entler S, Syblik J, Dostal V, Stepanek J, Zacha P. Optimization of the
of Sciences within the research program ‘Systems for Nuclear Energy’. supercritical CO2 power conversion system based on the net efficiency under
conditions of the pulse-operated fusion power reactor DEMO. Appl Therm Eng
2021;194:116884. http://dx.doi.org/10.1016/j.applthermaleng.2021.116884,
References URL https://www.sciencedirect.com/science/article/pii/S135943112100332X.
[13] Syblík J, Entler S, Veselý L, Štěpánek J, Dostál V. Fusion DEMO sCO2 layout
[1] Feher E. The supercritical thermodynamic power cycle. Energy Convers design with battery farm. Energy 2022;249:123730. http://dx.doi.org/10.1016/
1968;8(2):85–90. http://dx.doi.org/10.1016/0013-7480(68)90105-8. j.energy.2022.123730, URL https://www.sciencedirect.com/science/article/pii/
[2] Shi D, Zhang L, Xie Y, Zhang D. Aerodynamic design and off-design performance S0360544222006338.
analysis of a multi-stage S-CO2 axial turbine based on solar power generation [14] Dostal V. A supercritical carbon dioxide cycle for next generation nuclear
system. Appl Sci 2019;9(4). http://dx.doi.org/10.3390/app9040714, URL https: reactors (Ph.D. thesis), Massachusetts Institute of Technology. Dept. of Nuclear
//www.mdpi.com/2076-3417/9/4/714. Engineering.; 2004.
[3] Kumaran RS, Alone DB, Nassar A, Kumar P. Preliminary aerodynamic [15] Brun K, Friedman P, Dennis R. Fundamentals and applications of supercriti-
design of a S-CO2 axial turbine. In: ASME 2021 gas turbine india cal carbon dioxide (SCO2 ) based power cycles. Woodhead Publishing; 2017,
conference. 2021, http://dx.doi.org/10.1115/GTINDIA2021-76454, arXiv:https: p. 433. http://dx.doi.org/10.1016/B978-0-08-100804-1.00001-3, URL https://
//asmedigitalcollection.asme.org/GTINDIA/proceedings-pdf/GTINDIA2021/ www.sciencedirect.com/science/article/pii/B9780081008041000013,
85536/V001T12A009/6812511/v001t12a009-gtindia2021-76454.pdf, [16] Alexeev R, Tishchenko V, Gribin V, Gavrilov I. Turbine blade profile design
V001T12A009. method based on bezier curves. J Phys Conf Ser 2017;891:012254. http://dx.
[4] Zhenya L, Wenjie B, Li J, Chuanliang L, Shi Jinyuan HN. Supercritical doi.org/10.1088/1742-6596/891/1/012254.
carbon dioxide turbine design and arrangement optimization. Front Energy [17] Bell IH, Wronski J, Quoilin S, Lemort V. Pure and pseudo-pure fluid thermophys-
Res 2022;10. http://dx.doi.org/10.3389/fenrg.2022.922542, URL https://www. ical property evaluation and the open-source thermophysical property library
frontiersin.org/articles/10.3389/fenrg.2022.922542. CoolProp. Ind Eng Chem Res 2014;53(6):2498–508. http://dx.doi.org/10.1021/
[5] Rogalev A, Grigoriev E, Osipov S, Rogalev N. The design approach for super- ie4033999.
critical CO2 gas turbine. In: AIP conference proceedings 2189. 2019, http:// [18] Maghsoudi P, Sadeghi S, Xiong Q, Aminossadati SM. A multi-factor methodology
dx.doi.org/10.1063/1.5138630, URL https://aip.scitation.org/doi/pdf/10.1063/ for evaluation and optimization of plate-fin recuperators for micro gas turbine
1.5138630. 020018. applications considering payback period as universal objective function. Internat
[6] Bidkar R, Mann A, Singh R, Sevincer E, Cich S, Day M, Kulhanek CD, Thatte A, J Numer Methods Heat Fluid Flow 2020;30(5):2411–38. http://dx.doi.org/10.
Peter A, Hofer D, Moore J. Conceptual designs of 50MWe and 450MWe 1108/HFF-04-2019-0333.
supercritical CO2 turbomachinery trains for power generation from coal. Part [19] Pedram H, Maghsoudi P. Exergy, economy and pressure drop analyses for optimal
1: Cycle and turbine. 2016. design of recuperator used in microturbine. Energy Equip Syst 2017;5(2):95–
[7] Can Uysal S, Weiland N. Turbomachinery design of an axial turbine for a direct 113. http://dx.doi.org/10.22059/ees.2017.25717, arXiv:http://www.
fired sCO2 cycle. Energy Convers Manage 2022;267:115913. http://dx.doi.org/ energyequipsys.com/article_25717_308e673ba6057953610ddd74ac856e09.pdf,
10.1016/j.enconman.2022.115913, URL https://www.sciencedirect.com/science/ URL http://www.energyequipsys.com/article_25717.html.
article/pii/S0196890422007099. [20] Saeed M, Berrouk A, Burhani B, Alatyar A, Wahedi Y. Turbine Design and
[8] Sienicki J, Moisseytsev A, Fuller R, Wright S, Pickard P. Scale dependencies Optimization for a Supercritical CO2 Cycle Using a Multifaceted Approach Based
of supercritical carbon dioxide brayton cycle technologies and the optimal size on Deep Neural Network. Energies 2021;14(7807):1–27. http://dx.doi.org/10.
for a next-step supercritical CO2 cycle demonstration. In: SCO2 power cycle 3390/en14227807, URL https://www.mdpi.com/1996-1073/14/22/7807/pdf.
symposium may 24-25. 2011. [21] Gao F, Han L. Implementing the nelder-mead simplex algorithm with adaptive
[9] Syblik J, Vesely L, Entler S, Stepanek J, Dostal V. Analysis of super- parameters. Comput Optim Appl 2012;51:1:259–77.
critical CO2 Brayton power cycles in nuclear and fusion energy. Fusion [22] Bloch HP. Tilting pad thrust bearings. Mach Lubr 2006;7. URL https://www.
Eng Des 2019;146:1520–3. http://dx.doi.org/10.1016/j.fusengdes.2019.02.119, machinerylubrication.com/Read/906/tilting-pad-thrust-bearings.
SI:SOFT-30. [23] Miba Industrial Bearings Germany OGaCK. Tilting pad thrust bearings.
[10] Stepanek J, Entler S, Syblik J, Vesely L, Dostal V, Zacha P. Parametric study of 2022, URL https://www.miba.com/fileadmin/user_upload/Divisionen/Bearing/
S-CO2 cycles for the DEMO fusion reactor. Fusion Eng Des 2020;160:111992. MIB/MIB_Products/Downloads_MIB/MIBA_Tilting_Pad_Thrust_Bearing.pdf.
http://dx.doi.org/10.1016/j.fusengdes.2020.111992.

13