Chinese Journal of Aeronautics, (2022), 35(11): 33–44

Chinese Society of Aeronautics and Astronautics

& Beihang University
Chinese Journal of Aeronautics
[email protected]
www.sciencedirect.com

Design and optimization of supersonic turbines for

detonation combustors
Jorge SOUSA a, Elena COLLADO-MORATA b, Guillermo PANIAGUA c,*

a
Flemish Institute for Technological Research, Mol 2400, Belgium
b
Commissariat à l´e´nergie atomique et aux énergies alternatives, Le Barp 33114, France
c
Mechanical Engineering, Purdue University, West-Lafayette IN 47906, USA

Received 29 September 2021; revised 19 October 2021; accepted 3 January 2022

Available online 18 April 2022

KEYWORDS Abstract Detonation-based engines offer a potential surge in efficiency for compact thermal power
Rotating detonation; systems. However, these cycles require ad-hoc components adapted to the high outlet velocity from
Shock waves; the detonation combustors. This paper presents the design methodology of turbine stages suitable
Supersonic flow; for supersonic inlet conditions and provides a detailed analysis of optimized turbine geometries. A
Turbines; reduced-order solver examines the supersonic blade rows’ functional design space, quantifies the
Unstarting turbine’s non-isentropic performance, and budgets the turbine loss for different optimized
leading-edge designs and chord to pitch ratios. The shock-wave interactions were identified as
the predominant contributor to turbine losses, and optimal pitch-chord ratios were determined
for various inlet Mach numbers. Finally, with this tool, the specific-power output for a wide range
of design configurations was computed; and the metal angle that ensures flow starting and maxi-
mizes power extraction was calculated. The detailed numerical study describes the flow interactions
in a supersonic turbine and offers new correlations to guide the design of future supersonic turbines.
Ó 2022 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd. This is
an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction steady growth in air transportation, increasing by 5% per

year,1 doubles the expected annual improvements in engine-
Gas turbines have played a significant part in advancing air- specific fuel consumption.2 Globalization and environmental
transport technology by improving comfort while reducing concerns drive the need for minimal pollutant emissions,3 com-
the time between distant human settlements. However, the pelling new efficient propulsion architectures.4 In detonation-
based engines5 combustors burn hydrogen through a detona-
* Corresponding author. tion process instead of the conventional deflagration reaction.
E-mail addresses: [email protected], [email protected] The fusion of this type of combustor with a short nozzle and
(G. PANIAGUA). a supersonic turbine6 is shown in Fig. 1(a). Numerous authors
Peer review under responsibility of Editorial Committee of CJA. have presented increased cycle efficiency, first using steady
assumptions7,8 and, more recently, reduced-order models.6,9,10
Several other authors have demonstrated the performance
and the aero-thermal challenges associated with turbine
Production and hosting by Elsevier stages operating with detonation combustors.11–13 The

https://doi.org/10.1016/j.cja.2022.04.003
1000-9361 Ó 2022 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
34 J. SOUSA et al.

Fig. 1 Rotating detonation engine architecture and performance with an axial supersonic turbine.

iso-temperature contours reveal the rotating detonation front, transported towards the outlet. After 2.7 ms, the traveling
resulting in a sudden increase in pressure consuming the shock wave arrives at the outlet, and a steady-state is achieved.
hydrogen-air mix at supersonic velocities along the tangential Fig. 3(a) sketches the two-dimensional supersonic turbine
clock-wise direction. An oblique shock wave and a slip line passage, defined by the inlet and outlet area (A) and the
are anchored at the triple point and propagate towards the inlet/outlet flow angle (a). The area can be expressed as a func-
combustor outlet into the nozzle and supersonic turbine. tion of outlet flow angle (a), the pitch (g), and channel height
Recent experimental and numerical research5,14 demonstrates (H): A ¼ gcosaH. The contraction ratio from the inlet to the
that the unsteady flow downstream of the Rotating Detonation throat must be limited to ensure that all the flow can pass
Combustor (RDC) lies within the low supersonic range.15–17 through the throat. To ensure self-staring, following Kantrow-
Hence, shock waves are generated at the airfoil leading-edges, itz and Donaldson,19 the throat must be sufficiently large to
creating numerous aerodynamic and heat transfer challenges. ensure that all the mass flow can be swallowed with the condi-
Fig. 1(b) is a temperature-entropy diagram that compares tions after the starting shock wave. Hence, the turning across
two small pressure ratio turbojets;6 a Rotating Detonation- the turbine passage must be limited to ensure that the contrac-
combustor Engine (RDE, black delta symbol) with a conven- tion ratio Aout/Ain, is above the Kantrowitz limit defined in Eq.
tional deflagration Brayton engine (red gradient symbol). Both (1), denoted as ‘‘design space” in Fig. 3(b), all the turbines
engines have an identical compression ratio, with the same design above the Kantrowitz limit will be self-starting for a
combustor inlet conditions p3 and T3 (p is the pressure and sudden change of conditions from subsonic into supersonic.
T is the temperature). However, the detonation combustor pre-
 c
sents a higher outlet pressure (p40 > p4), although a slightly Aout

1 ðc þ 1ÞMain 2 c1
¼
lower average outlet temperature, due to the temperature fluc- Ain
Kantr: Main ðc  1ÞMain 2 þ 2
tuations. The higher total pressure is also associated with a  c1
1
! c1
lower entropy generation, while the enthalpy drop across the cþ1 1 þ c1 Main 2 2ðc1Þ
 2
ð1Þ
turbine is the same in both engines. Therefore, the pressure 2cMain 2  ðc  1Þ cþ1
2
downstream of the turbine stage remains higher in the detona-
tion engine, increasing the turbojet thrust. Hence, detonation where Main is the inlet Mach number, and c is the isentropic
engines offer a potential leap in terms of overall engine expansion factor or isentropic exponent for an ideal gas
performance. (c = cp/cv, cp specific heat capacity at constant pressure, and
This paper focuses on analyzing a new class of supersonic- cv the specific heat capacity at constant volume).Hence, Eq.
axial turbines instead of the traditional subsonic turbines (2) shows the allowable outlet conditions can be computed
designed for low inlet Mach numbers of about 0.2.18 A turbine to respect the Kantrowitz limit.


passage reduces the area along the axial direction. For a super- Aout

Hin
aout < arccos  cosa ð2Þ
Ain
Kantr: Hout
in
sonic inlet condition, the flow is gradually decelerated up to the
throat. During the starting phase of the engine, the flow tran-
sits from a subsonic to a supersonic operation. A normal shock A methodology to optimize the stator and rotor’s metal
wave must travel from the combustor to the turbine outlet fol- angles was developed conducive to maximum non-isentropic
lowing a sequence outlined in Fig. 2 below. At time 0.3 ms, one power output. We evaluated the losses and studied the impact
can observe a normal shock wave between the inlet and the of different chord-to-pitch ratios, leading-edge shape, rota-
leading-edges. The supersonic flow entering the turbine pas- tional speed, and end-walls design to achieve this goal. The
sage is defined by the conditions after this normal shock wave. performance predictions were validated against calibrated 3D
Hence, only if the post-shock flow is not chocked at the throat, Reynolds-Averaged Navier-Stokes simulations. Finally, we
the shock wave can progress further downstream. Between present a range of design parameters that minimize the super-
0.6 ms and 0.9 ms, the shock begins a complex interaction with sonic turbine’s losses. The stator/turbine unstarting constrains
the leading-edge and the airfoil walls, and a complex leading- the design space. After the design parameters are selected. The
edge oblique shock system is created. Downstream of the method of characteristics coupled with the optimization
trailing-edge, the oblique shocks converge into a normal shock algorithm20 is used to optimize the shape of the supersonic tur-
Design and optimization of supersonic turbines for detonation combustors 35

Fig. 2 Starting sequence using schlieren visualizations at different instants across a supersonic turbine.

Fig. 3 Definition of self-started passage.

bine passage. The impact of pulsating flow is then quantified tions, the first term on the left-hand side is zero, there are no
with unsteady Reynolds-averaged Navier-Stokes source terms.
simulations.21,22 2 3 2 3
e uðe þ pÞ
6 7 6 7
6 q 7 6 qu 7
2. Computational tool 6 7 6 2 7
Q¼6 7 6
6 qu 7; F1 ¼ 6 qu þ p 7;
7
6 7 6 7
The three-dimensional flow field across the supersonic pas- 4 qv 5 4 qvu 5
qw qwu
sages was retrieved by solving the Reynolds-Average Navier- 2 3 2 3
vðe þ pÞ wðe þ pÞ
Stokes (RANS) equations with the numerical solver CFD+ 6 7 6 7
6 qv 7 6 qw 7
+ 14.1.1 developed by Metacomp Technologies. The govern- 6 7 6 7
ing Navier-Stokes equations are expressed as: F2 ¼ 6 7 6
6 qvu 7; F3 ¼ 6 quw 7
7
6 2 7 6 7
4 qv þ p 5 4 qvw 5
@Q @ ðF1 þG1 Þ @ ðF2 þG2 Þ @ ðF3 þG3 Þ qwv qw2 þ p
þ þ þ ¼ 0 ð3Þ
@t @x @y @z
where e is the specific energy, q is density, u the velocity in the
where Q represents the dependent variables, F the inviscid flux axial direction, v the velocity in the tangential direction and w
vectors, G the viscous flux vectors. In our steady-state simula- the flow velocity in the radial direction.
36 J. SOUSA et al.
2 3
q_x  usxx  vsxy  wsxz The sink term in Eq. (5) involves the following terms:
6 7  1=6
6 0 7 1 þ c6w3  em
6 7
6 7 fw ¼ g 6 ; g ¼ r 1 þ cw2 r5  1 ; r ¼ ;
G1 ¼ 6 sxx 7; g þ cw3
6 e
S ðjdÞ2
6 7
6 sxy 7
4 5  
em v
Se ¼ x þ ; ft2 ¼ ct3 ect4 v
2
sxz 1
ðjdÞ 2
1 þ vfv1
2 3
q_y  usyz  vsyy  wsyz
6 7 d is the distance to the wall and m the kinematic molecular
6 0 7 viscosity. The model constants are the following:
6 7
6 7 cb1 ¼ 0:1355; r ¼ 2=3; cb2 ¼ 0:622; j ¼ 0:41;
G2 ¼ 6 sxy 7; ð4Þ
6 7
6 syy 7 cw1 ¼ cjb12 þ 1þcr b2 ; cw2 ¼ 0:3;
4 5
cw3 ¼ 2; cv1 ¼ 7:1, ct3 ¼ 1:1, ct4 ¼ 2, cr1 ¼ 1:0,
syz cr2 ¼ 12:0; cr3 ¼ 1:0.
2 3
q_z  uszx  vszy  wszz The thermodynamic properties are obtained utilizing
6 7 temperature-dependent coefficients in a real gas model.24 The
6 0 7
6 7 viscosity of air was computed based on the Sutherland Law
6 7
G3 ¼ 6 sxz 7 with the following coefficients (Tref = 273.11 K, rviscosity =
6 7
6 syz 7 1.716  105 kg/(m∙s), Cte viscosity = 111.0, kconductivity,ref =
4 5
2.41  103 kg∙m/s3 and Cte conductivity = 194). The finite volume
szz method is used for spatial discretization, with all control vol-
In Eq. (4) the flux vectors of Eq. (3) are represented. The umes being cell-centered. The inviscid flux function was a
first row represents energy balance (where e ¼ cv T and second-order upwind scheme using a Harten-Lax-Van Leer-
p ¼ qRT; R is the ideal gas constant). The continuity equa- Contact (HLLC) Riemann solver.25 The spatial variables’
tion is expressed in the second line. The last three rows are interpretation is performed through a second-order Total Vari-
accountable for the Navier-Stokes equations, where sij is the ation Diminishing (TVD) polynomial interpolation with a con-
shear stress tensor component and q_i denotes the heat flux tinuous limiter for stability. The solver with the same
  numerical setup was validated against experimental data
@T
(q_i ¼  k @x , k is the thermal conductivity). A density-based
i from a transonic turbine evaluated by Sousa and Paniagua.20
algorithm is used to accommodate for compressibility effects. The numerical domains were discretized with the software
The Spalart-Allmaras (SA) closure23 was added for the turbu- AutoGrid5. Only one passage was numerically computed
lence modelling, where the shear stress tensor component sij in by applying periodicity at the two sides of the computational
Eq. (4) is formulated as follows: domain in the tangential direction. Fig. 4(a) shows an exam-
  ple of the 3D structured grid with multi-block meshes. The
@ui @uj 2 @uk boundary layer is discretized and solved up to the wall. Static
sij ¼ mt Sij ; Sij ¼ þ   dij
@xj @xi 3 @xk inlet conditions are imposed at the inlet, and a supersonic
state is set at the outlet. A grid sensitivity study was con-
where dij is the Kronecker delta operator and the kinematic ducted to analyze the mesh dependency of the solution.
eddy viscosity is defined as: Fig. 4(b) plots the results of the ratio of the outlet total pres-
v3 ve sure to the inlet total pressure (p0,out/p0,in)probed at 0.5Cax
mt ¼ e
v fv1 ; fv1 ¼ ;v  (Cax is the airfoil axial chord) downstream of the trailing-
v3 þ c3v1 t
edge (x = xte + 0.5Cax, xte is the axial coordinate of the air-
where ve is the working variable of the turbulence model, cv1 is foil trailing edge)for three grid levels (coarse-Grid N3,
a constant (define below with all the constants), t is the molec- medium-Grid N2, and fine-Grid N1). Rh is the arc-lenght
ular kinematic viscosity. Hence, an additional equation needs along the tangential direction.
to be solved for ev . Since in our simulations, no rotation was To evaluate the grid convergence, we used the Richard-
included, and we assumed a fully turbulent boundary layer son extrapolation method26 to compute the Grid Conver-
the transport equation for e v is: gence Index (GCI) as recommended by Celik et al.27
@   2r h  i   Table 1 summarizes the three different grids’ results, with
uj v ¼ cb1 ð1 þ cr1 Þ 1  cr3 tan1 cr2 r  cr1 ð1  ft2 Þ S v a refinement ratio (r) close to the recommended value of
@xj 1 þ r
1 h 
 i  2 1.3. The quantity of interest was the total pressure ratio
þ r  m þ m r m þ cb2 r m
r (p0,out/ p0,in). These values were used to estimate the order
! of accuracy ( pe ¼ 1:7), close to the used 2nd order discretiza-
 cb1  v
 2

 cw1 fw  2 ft2 ð5Þ tion scheme. To quantify the relative uncertainty (erel.) the
j d
Richardson extrapolation is applied, plus extrapolated
where, uncertainty (eext.). Finally, we defined the GCI as
 ^   1:25erel:
^  2x S^ GCI ¼ 
r ¼ xS ; r ¼ ik^ 4 jk Dtij ; S^ij ¼ 12 @x
DS @ui @uj . Based on these results, we selected Grid N2
þ ;
j @x i rp  1
 D
 2 for its little uncertainty and lower computational times than
xij ¼ 12 @x@ui @u ^ ¼ 1 2S^ij S^ij þ 2xij xij ;
 @xji ; D
j 2 Grid N3. Advanced convergence acceleration techniques
used include unique pre-conditioning, relaxation, and
S^ ¼ 2S^ij S^ij ; x ¼ 2xij xij
2
multi-grid algorithms.
Design and optimization of supersonic turbines for detonation combustors 37

Fig. 4 Computational domain.

Table 1 Summary of discretization error analysis.

Item Size r erel. (%) eext. (%) GCI (%)
Grid N1 3197952
Grid N1/Grid N2 1.33 0.41 0.64 0.81
Grid N2 1333248
Grid N2/Grid N3 1.47 0.035 0.36 0.045
Grid N3 419968

3. Aerodynamic performance from Point A to Point B are first submitted to the leading-
edge bow shock, followed by the combined effect of the
3.1. Development of a shock and viscous loss prediction tool boundary layer growth and reflected shock as the particles tra-
vel from Point B to Point C. However, this analysis does not
consider other losses created by the trailing- edge shock system
The supersonic flow pattern in turbine airfoils is shown in
interacting with the wakes and secondary flows. Still, they rep-
Fig. 5(a). Strong leading-edge bow shock waves propagate
resent a significant contribution in low aspect ratio supersonic
throughout the passage due to reflections creating total pres-
profiles.13
sure loss (p0(x)). Additionally, the boundary layer develops
To characterize the detached leading-edge bow shock, we
along the airfoil; due to the turning, a supersonic diffuser is
utilized the Moeckel empirical method.28 Fig. 6(a) sketches
created, magnifying the viscous layer loss, usually called profile
the leading-edge bow shock profile. SB is the sonic point along
losses. The boundary layer then detaches at the trailing-edge
the leading-edge, while hd represent the detachment angle for
and begins a mixing process. The present reduced-order model
the given inlet Mach (Main). Upstream of the leading-edge,
considers both the inviscid contribution arising from the
the shock wave geometry follows an asymptotic line, with b
leading-edge shock waves as well as the viscid loss created by
function of the incoming Mach number, defined by Moeckel
the profile and mixing losses. In Fig. 5(b), particles flowing
with Eq. (6).

Fig. 5 Flow topology and losses in a supersonic blade passage.

38 J. SOUSA et al.

Fig. 7 Viscous, shock wave and total losses (compared with

Euler and Navier-Stokes simulations).

sent the momentum thickness at the trailing-edge where, hss

and hps are the momentum thickness at the suction and pres-
sure side respectively. The empirical correlation outlined by
Stratford and Beavers30 was used to compute those
thicknesses.

3.2. Assessment of loss models

Fig. 6 Leading edge shock reduced-order model.
Fig. 7 depicts the total loss prediction, 1  p02/p01 (solid blue
1=2 line); the corresponding viscous contribution (black dashed-
x2shock  x20
yshock ¼ ð6Þ dotted line) from the mixing and profile; the leading-edge
b bow shock losses (red dashed line). To assess the methodology,
where xshock and yshock are the coordinates of the shock wave Fig. 7 also includes the results from the higher-order two-
while, x0 is the axial coordinate of the shock inception. dimensional numerical simulations. The red circular symbols
A study was conducted to analyze the shape of the reduced- represent the non-viscous (Euler) simulations, while the blue
order model results of the leading-edge bow shock geometry squared symbols represent the Navier-Stokes simulation. The
and benchmark against the CFD results as a function of inlet agreement between the Computational Fluid Dynamics simu-
Mach numbers. Fig. 6(b) compares the estimated shock wave lations (CFD) and the shock and total loss prediction is within
profile, red line, with a numerical schlieren visualization of 2% and 5%, respectively. The most significant contribution to
the CFD results, depicted in grayscale. To study where the loss is caused by shock waves, which increase dramatically as
entropy is generated, the loss gradient in the pitch-wise direc- the inlet Mach number is increased. For the selected aspect
tion across one passage (g = pitch) is plotted in Fig. 6(b) right ratio (height to chord). The contribution to entropy generation
for three Mach numbers. The maximum of entropy is achieved of the trailing-edge shock system or secondary flows are mar-
at the edge of the passage (y = 0, g). In this region shock wave ginal for the selected aspect ratio (height to chord); and conse-
is perpendicular to the flow. quently are neglected in the frame of this analysis. For high
Both profile and mixing losses are correlated and are eval- chord to height ratios, the endwall effects dominate, as demon-
uated by applying a mass balance29 between Plane 20 and Plane strated for the non-optimal chord to pitch ratios.13
2, which particularized for supersonic passages yield Eq. (7):
3.3. Effect of leading-edge design
cosa20 ð1  d  dte ÞðqVÞ20 ¼ cosa2 ðqVÞ2 ð7Þ

where V is the magnitude of the local flow velocity. The bal- The main contribution of this methodology is to quickly per-
ance of axial and tangential momentum results in Eqs. (8) form evaluations over a large span design space with a hetero-
and (9): geneous turbine profile topology. We ran several viscous
simulations to evaluate the current methodology’s ability to
sina20 cosa20 ð1  d  dte  h Þ qV2 2
0 ¼ sina2 cosa2 qV2 2
ð8Þ predict the losses’ values and derivatives for comparison pur-
poses. This reduced-order tool was coupled with a genetic
:
gp20 þ cos2 a20 ð1  d  dte  h Þ qV2 0
2 algorithm to optimize the turbine passages using a low number
¼ gp2 þ cos a2 qV
2 2
ð9Þ of design parameters.20 The camber line was built using a
2
quadratic Bezier curve. Among the free design parameters,
where dte is the ratio of trailing-edge thickness to throat area we have the leading-edge thickness T1,le in Fig. 8(a), and the
(t/(gcos a20 )); d ¼ dss þ dps =ðgcosa20 ) is the boundary layer wedge leading-edge angle ale . Fig. 8(b) presents the thickness
displacement, where dss ; dps the displacement at the suction to chord distribution along the axial direction for the baseline
and pressure side respectively; h ¼ hss þ hps = gcosa20 repre- and optimized geometry. Fig. 8(c) shows that large leading-
Design and optimization of supersonic turbines for detonation combustors 39

Fig. 8 Leading edge and thickness geometry definition and impact on total pressure loss.

edge thickness induces a surge in the losses for thick airfoils. where shock waves are generated. The optimized results
The leading-edge design determines the bow shock’s location obtained by Senoo31 are shown in Fig. 9(a); similarly to our
ahead of the leading-edge, the prime contribution to the loss conclusion, the cusped airfoil presents a leading-edge shock
generation. Fig. 8(d) shows that diminishing the wedge angle wave with a narrower angle that helps to minimize the shock
cause a decrease in the shock losses. Therefore, the mechanical strength and their consequent interaction with the boundary
constraints are to be balanced with the aerodynamic penalty. layer of the neighborhood blade. Šafařı́k’s team designed mul-
In the last stage of steam turbines, due to the high- tiple supersonic turbine airfoils. Based on the interferometric
peripheral speeds, we also observe supersonic flows near the results,32 we note the appearance of a terminal shock in Profile
tip in the relative frame of reference but with a subsonic axial A and Profile C due to the over acceleration around the
component. The design and optimization of such profiles are leading-edge. Design B with a cusped geometry allowing to
essentially focused on optimizing the leading-edge region eliminate the terminal shock.

Fig. 9 Leading-edge geometries of supersonic steam turbine rotor tips.

40 J. SOUSA et al.

Fig. 10 Supersonic airfoil loading.

3.4. Effect of chord to pitch ratio Several Navier-Stokes simulations are also included in
Fig. 11 to reassess the validity of the reduced-order model to
Fig. 10(a) displays the isentropic Mach number (Mais) along assess the passage loss. For a given pitch, we note a monotonic
the airfoil’s pressure and suction side for the three pitch to increase of the loss with the chord. At Main = 2.0, a 10%
chord ratios. The flow accelerates rapidly to sonic conditions change on the chord to pitch ratio results in approximately
near the stagnation region. A Prandtl-Mayer expansion 5% variation in total pressure loss. Fig. 10(a) evidence that a
around the leading-edge further accelerates the flow to super- higher chord to pitch ratio is conducive to two shock waves
sonic conditions. The acceleration continues along the suction impacting the suction side.
side due to the convex curvature, while a smoother decelera- To evaluate the optimal c/g ratio, the outlet flow angle
tion occurs along the pressure side due to its concave curva- (aout) was mass flow averaged at a plane located 0.5Cax down-
ture. Around the airfoil leading-edge, all the Mach number stream of the trailing-edge. Fig. 12(a) represents outlet flow
lines overlap, as in this region, the flow is outside of the zone angle as a function of the c/g ratio for several inlet Mach num-
of influence of the neighbor airfoil, as sketched in Fig. 10(b). bers. One can observe the existence of an optimal c/g at which
The first traceable impact of the neighborhood blade is the first the outlet flow angle is maximized, in fact exceeding the metal
shock impingement. For the chord-to-pitch c/g = 1.5, no Suc- angle by approximately 5°. This occurs precisely for a config-
tion Side (SS) shock impingement can be identified on the uration where the reflected shock from the suction side
Mach number distribution. Therefore, this part of the airfoil impinges on the trailing edge, as seen in Fig. 12(b) for c/
operates without any flow influence from the neighborhood g = 2.1, Fig. 12(a) reveals that the optimal c/g rises with the
airfoil. The shock impingement appears on both sides of the Mach number due to the narrower cones of influence seen in
airfoil and moves incrementally upstream, increasing the chord Fig. 12(b). Hence, the c/g pitch ratio needs to be adapted to
to pitch ratio. At this location, a reflected shock wave is gener- obtain the same shock features. Accordingly, the optimal c/g
ated and propagated downstream. For c/g = 2.1, the reflected correlates with the Mach number for an outlet metal angle
shock from the pressure side impacts close to the trailing-edge, of 20°. as, c=g ¼ Main  0:5.
as exemplified in Fig. 10(b). While for c/g = 2.5 this shock Fig. 13 evaluates the outlet Mach number (Maout) and flow
wave interacts again with the suction side, as seen by the Mach angle (aout) for three c/g values. At low c/g (Fig. 13(a)), the
number abatement at around x/Cax = 0.75, the loading is lack of flow guidance across the passage induces an important
inverted. velocity step at the trailing-edge that promotes the appearance
Fig. 11 shows the predicted total pressure loss in function of of a strong shock wave and an expansion fan. Consequently,
the chord to pitch ratio (c/g) at three inlet Mach numbers. the downstream flow presents high pitch-wise velocity gradi-

Fig. 11 Total pressure loss prediction at three Mach numbers.

Design and optimization of supersonic turbines for detonation combustors 41

Fig. 12 Effect of chord to pitch ratio on outlet flow angle and flow structures.

Fig. 13 Downstream conditions for several chord to pitch ratios.

ents. In this case, the Mach number fluctuates around 0.45, the iso-contour in Fig. 13(c), it appears that the better flow
which denotes 30% of the mean outlet value. Regarding the guidance, in this configuration, provides a more homogenous
flow angle, its mean value (aout) is below the metal angle flow field; nonetheless, the amplitudes of the mid-span Mach
(ametal) and the fluctuations are within 3°. For the test case number are similar to the previous case. Due to the airfoils’
with c/g = 1.5, Fig. 13(b), a decrease in the Mach number additional flow guidance, the mean flow angle gets closer to
amplitudes (Ma = ±0.25) is shown, while the mean value of the metal angle, and the extra turning provided by the shock
the flow angle is considerably higher (aout = ametal + 4.4°). wave is smeared out. From this analysis, it appears that c/
The flow angle fluctuations present a lower spatial frequency g = 1.5 for the Main = 2, represents an optimal value that
content; however, the amplitude is increased to ±4°. From could allow a higher work extraction in the downstream row.
42 J. SOUSA et al.

utilizing Cantera,33 chemical kinetics and thermodynamic tool-

box, and the p0,in/p0,out evaluated with the previously discussed
reduced-order loss solver. An iterative procedure was imple-
mented in MATLAB to calculate the outlet Mach number,
both in the vane and in the rotor row, using the relative frame
of reference. The overall stage work (DWmax) extraction is
evaluated by first computing the conditions isentropically
across the stator: p01 = p02; and then repeating the approach
in the rotor the relative rotor inlet total pressure (Plane 2)
remains the same at the rotor outlet (Plane 3):
prelative,02 = prelative,03.

4.2. Parametric study on design parameters

Fig. 14 Turbine stage layout. Fig. 15(a) maps the specific enthalpy, in the function of the
vane outlet flow angle (a2), the rotor exit flow angle in the rel-
ative frame of reference (b3) for a constant turbine passage
4. Stage design and evaluation height (H). The stator’s design space is bounded by the maxi-
mum area contraction or outlet stator angle, ensuring a stable
4.1. 1D methodology for stage design supersonic operation. The same is noted for b3, the ultimate
turning in the rotor passage is dictated by the inlet relative
In axial supersonic turbines, the downstream conditions are rotor angle (b2) and the inlet velocity in the relative frame of
not defined by the outlet static pressure but by the geometry reference (MaR2) as dictated by the stator design choice.
of the passage, inlet conditions, and irreversibility. Let us con- Enlarging a2 causes more turning, bounding further the maxi-
sider isentropic flows, in a stationary blade row the total tem- mum inlet Mach number. To respect the Kantrowitz limit, the
perature at the inlet is identical to the outlet (T0,in = T0,out), maximum absolute outlet rotor angle ought to be diminished.
the outlet flow properties can be estimated using Eq. (10): Therefore, the maximum power output compromises a balance
 0:5ðcþ1 between a2 and b3. With identical inlet velocity, Fig. 15(b) dis-
c1Þ
c1 plays the specific enthalpy (Dh) results with a vane channel
Maout 1 þ Maout
2 height increase of 20% from inlet to outlet, H2 = 1.2H1,
 0:5ðcþ1
c1Þ
and 20% increase along the rotor passage H3 = 1.2H2. The
Hin cosain p0;in c1 design space is widened for both vane and the entire turbine
¼  Main 1 þ Main ð10Þ
Hin cosaout p0;out c stage. The available flow turning on each passage was
In the previous equation, Hin/Hout is the height inlet to out- enhanced with the increase of outlet area. The turning across
let ratio. The Kantrowitz self-starting boundary (Eq. (2)) the vane was increased more than 10°, which in turn also
defines the maximum turning, and thus the exit flow angle enhanced the rotor range of operation. The increase of 20%
(aout). Fig. 14(a) sketches the blade-to-blade geometry and in the rotor channel height resulted in nearly doubling the
defines the angle conventions. Fig. 14(b) displays an axial cut stage specific enthalpy.
of the turbine passage to illustrate the change of radius along Fig. 16(a) depicts the specific power, at three inlet Mach
the vane and rotor endwalls. The specific heat ratio and numbers, as a function of the rotational speed (U). Each point
enthalpy were computed at each station along the turbine along the line is the result of evaluating the optimal

Fig. 15 Specific enthalpy for different stator outlet angles (a2) and rotor outlet angle (b3).
Design and optimization of supersonic turbines for detonation combustors 43

Fig. 16 Specific enthalpy in function of rotational speed (U).

combination of vane outlet (a2) and rotor outlet flow angle 5. Conclusions
(b3). We observe a monotonic increase in the specific power
with the turbine inlet velocity. Running at Main = 2.5 repre- This paper presents both a new approach to optimize axial
sents an increase of 120% in specific power relative to the base- supersonic turbines suitable for detonation-based engines as
line Main = 2.0 case, due to the higher flow velocities and well as guidelines for turbine and engine designers. The paper
lower supersonic starting limitations. Based on the Euler equa- presents first the design space that guarantees the startability
tion, the turbine enthalpy is directly propositional to rotational of the supersonic passages during the engine starting phase.
speed. However, a counter effect relates the rotational speed Secondly, the primary sources of entropy generation were
with the amount of possible flow turning across the rotor pas- diagnosed. A strategy was explained to evaluate the shock
sage (b3  b2) that ensures a started supersonic flow. For and viscous losses, comprising profile, associated with the
example, when Main increased to 2.5 the optimal rotational boundary layer spread, and mixing loss caused by the shear
speed rose by 60%. Thus, there is an ideal revolutions per min- between the wakes and the core flow. The main cause of irre-
ute for each operational Mach number. In Fig. 16(b), we com- versibility is the leading-edge bow shock. Shock loss is pre-
pare different growth rates of the turbine passage along the dominant in high inlet Mach numbers and is responsible for
axial direction, 10%, and 20% increase in the height, relative a steep performance decrease with a higher chord to pitch ratio
to the constant height. Enlarging the height of the turbine pas- and leading-edge thickness. The results of the reduced-order
sage along the turbine yields higher specific enthalpy and model were benchmarked against 3D Navier-Stokes simula-
increases the rotor’s optimal rotational speeds. Hence, this tions at different inlet supersonic velocities and geometrical
analysis/tool provides the necessary correlations to perform a properties.
complete engine matching when other turbomachinery compo- Additionally, we used various Navier-Stokes simulations to
nents are involved and attached to the same shaft. characterize the downstream flow field and identify the optimal
The parametric analysis defines the relevant design space chord to pitch ratio for different operating conditions. The
conducive to maximum power extraction. Afterward, a com- proposed loss models were then combined with an algorithm
plete airfoil optimization can be launched based on direct to determine the vane and rotor outlet flow angles conducive
evaluations of the losses and the isentropic Mach number to maximum non-isentropic enthalpy from the single super-
distribution, using the method of characteristics coupled sonic stage. This approach’s low computational demand
with an optimization scheme based on genetic algorithms. allowed us to study the impact of different geometrical and
This meshless design tool was assessed by comparing the operational conditions. We observed that an increase of the
results with Reynolds Averaged Navier-Stokes.20 A reduc- meridional passage allows more turning in the stator and
tion in the turbine rotational speed causes positive incidence rotor. For example, a 20% blade height variation offered a
into the supersonic blading. At the leading-edge, we observe 100% gain in specific power output. Finally, we showed that
a steeper acceleration along the suction side and a stronger the optimal peripheral speed increases with the inlet Mach
right-running leading-edge shock, which originate a recircu- number and the blade height variation. This methodology pro-
lation bubble, abating the turbine performance.13 Interest- vides the fundamental steps to guide supersonic turbines’ ini-
ingly, the off-design performance of supersonic internal tial design suitable for high-speed inlet conditions. The
passages is highly dependent on the excitation frequency. reduced-order approach is used to estimate the non-
At low reduced frequencies, a hysteresis-like behavior is isentropic power output from a single stage.
observed between the inlet and the outlet distinct from the
steady prediction.21 At higher reduced frequencies, the inlet
Declaration of Competing Interest
flow fluctuations are faster than the traveling time of a char-
acteristic wave between the inlet and outlet planes. Hence,
The authors declare that they have no known competing
the slowest characteristic waves are overtaken by waves of
financial interests or personal relationships that could have
the opposite family; the amplitudes are significantly damped
appeared to influence the work reported in this paper.
across the turbine passage.
44 J. SOUSA et al.

Acknowledgments 17. Schwer D, Kailasanath K. Fluid dynamics of rotating detonation

engines with hydrogen and hydrocarbon fuels. Proc Combust Inst
2013;34(2):1991–8.
We would like to acknowledge the French Agency ANRT for
18. Liu Z, Braun J, Paniagua G. Thermal power plant upgrade via a
supporting the first author through a CIFRE Ph.D. grant. The rotating detonation combustor and retrofitted turbine with opti-
authors also thank the US Department of Energy for the mized endwalls. Int J Mech Sci 2020;188:105918.
appointment of Prof. Paniagua to the Faculty Research Partic- 19. Kantrowitz A, Donaldson CD. Preliminary investigation of
ipation Program at the NETL. The same authors presented supersonic diffusers. Washington, D.C.: NACA; 1945. Report
some figures included in this paper at the ASME Power and No.: NACA ACR L5D20.
Energy Conference, 2017, Paper POWER-ICOPE2017- 20. Sousa J, Paniagua G. Entropy minimization design approach of
3624,34 permission for publication was granted by ASME. supersonic internal passages. Entropy 2015;17(8):5593–610.
21. Sousa J, Paniagua G, Saavedra J. Aerodynamic response of
internal passages to pulsating inlet supersonic conditions. Comput
References Fluids 2017;149:31–40.
22. Liu Z, Braun J, Paniagua G. Characterization of a supersonic
1. Grote M, Williams I, Preston J. Direct carbon dioxide emissions turbine downstream of a rotating detonation combustor. J Eng
from civil aircraft. Atmos Environ 2014;95:214–24. Gas Turbines Power 2019;141(3) :031501.
2. Masiol M, Harrison RM. Aircraft engine exhaust emissions and 23. Spalart P, Allmaras S. A one-equation turbulence model for
other airport-related contributions to ambient air pollution: A aerodynamic flows. Reston: AIAA; 1992. Report No.: AIAA-
review. Atmos Environ 2014;95:409–55. 1992-0439.
3. Gaspar RMP, Sousa JMM. Impact of alternative fuels on the 24. McBride BJ, Gordon S, Reno MA. Thermodynamic data for fifty
operational and environmental performance of a small turbofan reference elements. Washington, D.C.: NASA; 2001. Report No.:
engine. Energy Convers Manag 2016;130:81–90. NASA/TP-3287.
4. Lee JJ. Can we accelerate the improvement of energy efficiency in 25. Toro EF, Spruce M, Speares W. Restoration of the contact surface
aircraft systems? Energy Convers Manag 2010;51(1):189–96. in the HLL-Riemann solver. Shock Waves 1994;4(1):25–34.
5. Ma JZ, Luan MY, Xia ZJ, et al. Recent progress, development 26. Richardson LF. The approximate arithmetical solution by finite
trends, and consideration of continuous detonation engines. AIAA differences of physical problems involving differential equations,
J 2020;58(12):4976–5035. with an application to the stresses in a masonry dam. Phil Trans R
6. Sousa J, Paniagua G, Collado ME. Thermodynamic analysis of a Soc Lond A 1911;210(459–470):307–57.
gas turbine engine with a rotating detonation combustor. Appl 27. Celik IB, Ghia U, Roache PJ. Procedure for estimation and
Energy 2017;195:247–56. reporting of uncertainty due to discretization in CFD applications.
7. Heiser WH, Pratt DT. Thermodynamic cycle analysis of pulse J Fluids Eng 2008;130(7) :078001.
detonation engines. J Propuls Power 2002;18(1):68–76. 28. Moeckel W. Approximate method for predicting form and
8. Alhussan K, Assad M, Penazkov O. Analysis of the actual location of detached shock waves ahead of plane or axially
thermodynamic cycle of the detonation engine. Appl Therm Eng symmetric bodies. Washington, D.C.: NACA; 1949. Report No.:
2016;107:339–44. NACA TN D-1921.
9. Sousa J, Braun J, Paniagua G. Development of a fast evaluation 29. Stewart WL. Analysis of two-dimensional compressible-flow loss
tool for rotating detonation combustors. Appl Math Model characteristics downstream of turbomachine blade rows in terms
2017;52:42–52. of basic boundary-layer characteristics. Washington, D.C.:
10. Mizener AR, Lu FK. Low-order parametric analysis of a rotating NACA; 1955. Report No.:NACA-TN-3515.
detonation engine in rocket mode. J Propuls Power 2017;33 30. Stratford B, Beavers GS. The calculation of the compressible
(6):1543–54. turbulent boundary layer in an arbitrary pressure gradient: A
11. Fernelius MH, Gorrell SE. Predicting efficiency of a turbine driven correlation of certain previous methods. London: Her Majesty’s
by pulsing flow. New York: ASME; 2017. Report No.: GT2017- Stationery Office; 1961. Report No.: R. & M. 3207.
63490. 31. Senoo S. Development of design method for supersonic turbine
12. Zhou SB, Ma H, Li S, et al. Effects of a turbine guide vane on aerofoils near the tip of long blades in steam turbines: Part 1—
hydrogen-air rotating detonation wave propagation characteris- Overall configuration. Proceedings of ASME turbo expo 2012:
tics. Int J Hydrog Energy 2017;42(31):20297–305. Turbine technical conference and exposition; New York: ASME;
13. Paniagua G, Iorio MC, Vinha N, et al. Design and analysis of 2012. p. 355–65.
pioneering high supersonic axial turbines. Int J Mech Sci 32. Dvořák R, Šafařı́k P, Luxa M, et al. Optimizing the tip section
2014;89:65–77. profiles of a steam turbine blading. Proceedings of ASME 2013
14. Bedick C, Sisler A, Ferguson DH, et al. Development of a lab- turbine blade tip symposium; New York: ASME; 2014.
scale experimental testing platform for rotating detonation engine 33. Goodwin D, Moffat H, Speth R. Cantera: An object-oriented
inlets. Reston: AIAA; 2017. Report No.: AIAA-2017-0785. software toolkit for chemical kinetics, thermodynamics, and
15. Knowlen C, Wheeler E, Mendez D, et al. Thrusting pressure and transport processes [Internet]. 2018 May 17[cited 2022 Jan 10].
supersonic exhaust velocity in a rotating detonation engine. Available from: http://www.cantera.org.
Reston: AIAA; 2018. Report No.: AIAA-2018-0884. 34. Sousa J, Paniagua G, Collado-Morata E. Analysis of the
16. Braun J, Saracoglu BH, Paniagua G. Unsteady performance of aerodynamic losses in a supersonic turbine. New York: ASME;
rotating detonation engines with different exhaust nozzles. J 2017. Report No.: POWER-ICOPE2017-3624.
Propuls Power 2016;33(1):121–30.