SP2020_00068

LUMEN – DESIGN OF THE REGENERATIVE COOLING SYSTEM FOR AN EXPANDER

BLEED CYCLE ENGINE USING METHANE

Jan Haemisch(1) , Dmitry Suslov(2) , Günther Waxenegger-Wilfing(3) , Kai Dresia(4) , Michael Oschwald(5)
(1)-(5)
Institute of Space Propulsion, DLR Lampoldshausen, Langer Grund, D-74239 Hardthausen,
Jan.Haemisch@dlr.de, Dmitry.Suslov@dlr.de, Guenther.Waxenegger@dlr.de, Kai.Dresia@dlr.de,
Michael.Oschwald@dlr.de

KEYWORDS: LUMEN, HARCC, regenerative cool- 1 INTRODUCTION

ing, machine learning, design optimization
The regenerative cooling system for the LUMEN
engine faces two major features: 1. Methane
ABSTRACT as coolant. 2. Usage of the coolant in an ex-
pander bleed system architecture. Methane has a
much higher density, compared to hydrogen and
The regenerative cooling still is one of the most cru-
can therefore be stored in smaller, lighter tanks. Due
cial parts in designing a liquid rocket engine. The
to the higher density the turbopumps demand sig-
goal of an effective design is a trade-off between
nificantly less power. Additionally insulation can be
a sufficient cooling of the structure and a low pres-
saved due to a higher boiling point. Major drawback
sure drop in the cooling channels. For an expander
is a reduction of the specific impuls. Nevertheless
type cycle a third requirement arises: A sufficient
multiple LOX/CH4 or LOX/LNG engines are currently
enthalpy increase of the cooling fluid for the turbop-
under development [4, 16, 3].
umps to operate effectively. To prevent the combus-
Methane as coolant in regenerative cooling sys-
tion chamber from melting, the regenerative cooling
tems is a critical topic and a major concern for using
has to be designed with a sufficient margin to re-
methane as fuel for rocket engines. Methane has
spect all uncertainties. A neural network based sur-
a higher density and a lower specific heat at con-
rogate model is used to study the robustness of the
stant pressure cp , compared to hydrogen. The main
cooling channel design for other load points of the
different however is the vicinity to the critical point
engine.
that may lead to heat transfer deterioration (HTD)
In this paper, the design of the regenerative cool- [13, 15, 23].
ing system and the manufacturing of a subscale The main goal of a common cooling channel de-
combustion chamber will be discussed. sign is to maximize the cooling efficiency in other
words to reach the minimal hot gas side wall temper-
ature with a minimal pressure loss. The lower the
NOMENCLATURE hot gas side wall temperature, the better the cool-
ing performance since a temperature decrease of
dT = 40K doubles the engine life time [10]. For
α heat transfer coefficient [W/m/ K] the cooling channel design of an expander-type en-
cP specific heat capacity at constant pressure gine a third goal arises: A sufficient increase of the
[J/kg/K] coolant enthalpy to drive the turbopumps. This goal
h enthalpy [kJ/kg] is in opposite to the maximal cooling since a large
ks roughness [m] temperature increase of the coolant goes along a
L combustion chamber length [mm] high wall temperature. The enthalpy increase of the
ṁ mass flow [kg/s] coolant is the reason why expander-type engines
n number of cooling channels [-] typical display a long cylindrical combustion cham-
P pressure [bar] ber part [1].
Pcc combustion chamber pressure [bar]
Pin inlet pressure [bar] 1.1 LUMEN Engine
Pout outlet pressure [bar]
THE LUMEN (Liquide Upper Stage deMonstator
∆P pressure drop [bar]
ENgine) is a Nozzle-Expander-Bleed (NEB) engine
q̇w wall heat flux [W/m2 ]
that uses methane as fuel and therefore also as
Tin fluid inlet temperature [K]
coolant. The flow direction is counterflow for the
Tout fluid outlet temperature [K]
nozzle and combustion chamber and coflow for the
nozzle extension. Computational Fluid Dynamics

1
 parameter unit value These goals are not independent since a higher
enthalpy increase would increase the turbopump
Tout [K] > 400 performance and enlarge the allowed pressure
Tw [K] < 900 drop. The other way around, a lower pressure drop
∆pchannel [bar] < 25 could potentially allow a lower enthalpy increase.
Since the turbopump is not part of the calculation,
the goals are set to be fixed. In future versions of
Table 1: Goals for the cooling system design. the calculation, the feedback of the turbomachinery
might be added.
parameter unit value Due to manufacturing and structural reasons the
cooling channel width [mm] ≥1 width of the cooling channel is at least 1 mm and
width between cooling channels [mm] ≥1 the distance between the cooling channels has to
distance to hot gas side [mm] 1 be at least 1 mm. The number of cooling channels
number of cooling channels [−] ≤ 86 is a consequence of these constraints in the noz-
zle throat. For the hot gas side wall temperature,
the optimum is a constant temperature slightly be-
Table 2: Constraints due to manufacturing and low the temperature limit.
structural reasons.
2.2 optimization parameters

(CFD) is a commonly used tool for the calculation of To reach the goals with the given boundary condi-
heat transfer in cooling channels [26, 22, 15]. How- tions the geometry of the cooling channel can be
ever most of the validation data is for hydrogen as varied. That includes the width and height along
cooling fluid. Data for methane as coolant is rare the cooling channel length, a varying length of the
[12, 14, 27, 20]. In addition to that, the physical cylindrical part of the combustion chamber and the
properties rapidly change in the vicinity of the crit- number of cooling channels. The nozzle geometry
ical point. These two points lead to significant de- remains fixed. Despite the geometry, the coolant
viations of the CFD simulations, especially close to mass flow can be used as another parameter. The
the critical point [15]. This has to be kept in mind LUMEN system architecture allows to vary the mass
when discussing sufficient margin for the analysis. flow for the cooling channels within a certain range.
The paper is divided into three parts: In the first The roughness is assumed to be constant.
part a method is described and used to find the opti- Due to simplification and manufacturing reasons,
mal cooling channel geometry for an expander type the cooling channel geometry can be changed at 5
engine. In the second part, a neural network model characteristic positions along the combustion cham-
is presented and used to validate the found solu- ber length. In between the values are interpolated
tion for different load points. The third part gives an linearly. Fig. 1 shows the positions depending on
overview of the manufacturing process of the sub- the assumed heat flux profile. The characteristic
scale combustion chamber. positions are the begin and end of the decreasing
heat flux in the cylindrical part, the nozzle and the
beginning and end of the combustion chamber. Ad-
2 PART I: OPTIMIZATION OF COOLING CHAN-
ditionally the length of the cylindrical part (between
NEL GEOMETRY
h2 and h3) can be varied.
Based on the boundary conditions and constrains,
that are defined by a system analysis for the whole 2.3 Boundary conditions
LUMEN engine, CFD simulations with varying ge-
ometry were performed to optimize the cooling The main boundary conditions for the cooling chan-
channel geometry. An optimization algorithm is nel design are the assumed heat flux profile, the
used to change the height, width and length of the maximal mass flow and the assumed roughness in
cooling channels to find the solution that addresses the channels.
all three goals: sufficient cooling, low pressure drop, The LUMEN engine is designed to work for load
large increase of coolant temperature. points between a combustion chamber pressure be-
tween Pcc = 35 bar and Pcc = 80 bar and the mixture
ratio (ROF) between 3 and 3.8.
2.1 goals for the design
The geometry optimization will be performed for
The optimal design features a wall temperature the nominal point, that is: Pcc = 60 bar and
below the allowed limit of 900K, an enthalpy in- ROF = 3.4. The boundary conditions for the CFD
crease of the coolant to get an outlet temperature of simulation for this load point can be found in Tab. 3.
Tout > 400 K, and the lowest pressure drop possible When a solution for this load point is found, the per-
(at least dP < 25 bar). The goals are summarized in formance for the other load points will be verified.
Tab. 1. The constraints due to manufacturing rea- The heat flux profile is the result of a ROCFLAM III
sons are summarized in Tab. 2. simulation [19, 21].

2
 h5 4
m CH parameter unit value
h1 h2 h3 L [mm] 247.2
h4
60 h1 [mm] 8.2
h2 [mm] 4.0
50
heat flux [MW/m²]

h3 [mm] 4.0
40 h4 [mm] 1.8
30 h5 [mm] 4.1
nchannel [−] 86
20
ṁCH4 [−] 2.35
10

0
-350 -300 -250 -200 -150 -100 -50 0 50 100 150 Table 4: Geometry parameter for the final design.
89 mm 60 mm 119 mm
L
length [mm]
2.4 The numerical setup
Figure 1: Positions at which the cooling channel ge- RANS simulations were performed with the com-
ometry is optimized. mercial software ANSYS CFX [2]. The SST turbu-
lence model was used. The used mesh is adapted
parameter unit value to each iteration step so that the cell width and
2 heights remain constant to get comparable results.
q̇w [M W/m ] Fig. 2
The the y+-value is <1. More details for the numeri-
Tein [K] 120
cal setup can also be found in [15].
Pout [bar] 68.4
ks [m] 5e-6
ṁCH4 [kg/s] < 2.8 2.4.1 optimization procedure
As optimization scheme the Multi-Objective Generic
Algorithm (MOGA) algorithm is used [9]. The width
Table 3: Boundary conditions for the CFD simula- and heights are changed after each step and a CFD
tions. calculation was performed.

The profile that is shown in Fig. 2 was calculated 3 RESULTS

for a fixed combustion chamber contour, however
it was adapted to use it for different lenghts of the In this section, the results of the optimization will
cylindrical part of the combustion chamber. The first be discussed. Two different approaches were ana-
peak after the injector head and the slope of the de- lyzed: At first, a constant distance between the cool-
creasing heat flux is held constant. This leads to a ing channels was assumed. This approach leads to
lower heat flux at the beginning of the nozzle part smaller channels in the nozzle part, where the heat
for a longer cylindrical part. The overall heat flux load is higher and wider channels in the nozzle ex-
is of course larger for a longer combustion cham- tension where heat load is lower. This should lead
ber. The heat flux in the nozzle and nozzle extension to a better cooling where it is necessary and a low
is assumed to be independent from the combustion pressure drop where the cooling demand is less rel-
chamber length. evant. However the results were insufficient and no
solution was found that fulfills all three goals. The
approach was changed to a constant cooling chan-
60 nel width of 1 mm. This approach improved the re-
L small
50 L large sults significantly and led to the final cooling channel
geometry.
heat flux [MW/m²]

40

30 3.1 constant cooling channel width

20 A constant cooling channel width is comparatively
10 much easier to manufacture, and the results show
that also the cooling performance is enhanced com-
0
-400 -300 -200 -100 0 100 200 pared to the constant distance between cooling
length [mm] channels approach. This is due to a higher aspect
ratio (height-to-width-ratio) compared to the case
with constant distance between the cooling chan-
Figure 2: Assumed heat flux profile q̇w for differ- nels.This enhances the cooling performance and
ent lengths of the cylindrical part of the combustion also lowers the risk of the occurrence of heat trans-
chamber L. fer deterioration [13, 17].

3
 parameter unit value 1000
q design
q +10%
Tout [K] 408.6 q - 10%
800

Temperature [K]
Twh2 [K] 874.1
Twh4 [K] 879.1
600
∆pchannel [bar] 23.7
400

Table 5: Results for the CFD simulation of the final

200
design. -300 -200 -100 0 100
length [mm]
wall temperature h5 4
m CH
pressure
(a) heat flux
h1 h2 h3
h4
1000
m design
900 95 m +10%
m - 10%
800 90 800

Temperature [K]
Temperatur [K]

700
85
Druck [bar]
600 600
80
500
75 400
400

300 70

200
200
65
-350 -300 -250 -200 -150 -100 -50 0 50 100 150 -300 -200 -100 0 100
247.2 mm 60 mm 119 mm length [mm]
length [mm] (b) massflow

Figure 3: Temperature and pressure distribution for 1000

ks design
ks +20%
the optimized designpoint. ks - 20%
800
Temperature [K]

The drawback is a larger distance between the 600

cooling channels. However due to a wall material
with a high thermal conductivity and low heat flux in 400
the nozzle extension the cooling performance can
still be enhanced. 200
-300 -200 -100 0 100
After almost 500 CFD Simulations, an optimized length [mm]
solution was found. Fig. 3 shows the Temperature
(c) roughness
and pressure distribution for the final cooling chan-
nel geometry. The final geometric parameter are
summarized in Tab. 4 and the results for the final Figure 5: Influence of uncertainties to the hot gas
CFD simulation in Tab. 5 and Fig. 4. The design side wall temperature.
fulfills all requirements that were defined in Tab. 1.
The hot gas side wall temperature distribution
Although the heat flux peak at h2 is significant lower
shows two peaks at positions h2 and h4 that cor-
both peaks show a comparable high temperature of
relate with the peaks of heat flux profile (Fig. 2).
about Tw ≈ 880K. This distribution is in accordance
with the optimal temperature distribution, defined in
section 2.1 (a constant temperature slightly below
the temperature limit) and leads to a minimal pres-
sure drop.
The length of the cylindrical part is enlarged by
almost 50 mm compared to the initial design. Due to
the additional heat flux, the outlet temperature Tout
was maximized.

3.2 impact of uncertainties

To check how the results change due to uncertain-
ties of the boundary conditions, simulations were
performed with ±10% to the assumed value. It was
Figure 4: Temperature distribution for the optimized analyzed: Heat flux distribution, q̇w , massflow, ṁ,
designpoint. and roughness, ks . Fig. 5 shows the results for

4
these simulations. It can be seen that the highest 4.1 Comparison NN with CFD
deviations occur close to the nozzle throat. For the
After training, the NN can predict the maximum wall
design point, two peaks in the hot gas side wall tem-
temperature for previously unseen test cases, in-
perature occur. For deviations in heat flux (q̇w +10%)
cluding different channel geometries and operation
or mass flow (ṁ − 10%) the peak close to the nozzle
conditions. For the subsequent results, the model
throat is significant higher. On the other hand the
was extended for different channel curvatures and
peak at −250 mm is higher for the deviations with
rib thicknesses. Figure 6 shows an exemplary archi-
changing signs (q̇w − 10 % ; ṁ + 10 %). The opti-
tecture with two hidden layers, four neurons per hid-
mum is only valid for the design point. Deviations in
den layer, and all input parameters. The NN is com-
roughness (ks ± 20 %) only play a minor role.
bined with further reduced-order models that cal-
culate the stream-wise development of the coolant
pressure and enthalpy. Thus, predictions with a pre-
cision similar to full CFD calculations are possible.
4 PART II: NEURAL NETWORK BASED SUR-
The prediction of an entire channel segment takes
ROGATE MODEL FOR THE MAXIMUM WALL
only 0.6 s, which is at least 1000 times faster than
TEMPERATURE
comparable three-dimensional CFD simulations.

The main disadvantage of CFD simulations is that

they are not suitable for design space exploration 1000
and extensive sensitivity analysis due to their large
calculation effort. By constructing surrogate models
using samples of the computationally expensive cal-
wall temperature [K] 800
culation, one can alleviate this burden. However, it is
crucial that the surrogate model mimics the behavior
600
of the simulation model as closely as possible and
generalizes well to unsampled locations, while being 400 NN
computationally cheap to evaluate. Neural networks CFD
(NNs) are known to be universal function approxi- 200
mators [11] and have been successfully applied as 300 200 100 0 100
surrogate models in many domains, including avion- length [mm]
ics [24, 8] and space [7, 18].
To overcome the low accuracy of empirical heat-
Figure 7: Comparison between CFD and NN-based
transfer correlations for methane, Waxenegger-
calculations for a combustion chamber pressure of
Wilfing et al. [29] developed an NN-based surro-
60 bar.
gate model to predict the maximum wall temperature
along the cooling channel of the combustion cham-
ber. Trained with data from full three-dimensional
CFD simulations, the model achieves high accuracy parameter unit CFD NN
and can predict the maximum wall temperature even Tout [K] 408.6 409.1
close to the critical point. The NN uses a fully con- ∆pchannel [bar] 23.7 20.4
nected, feedforward architecture with 4 hidden lay- Twh2 [K] 874.1 920.0
ers and 408 neurons per layer. Twh4 [K] 879.1 877.2

Table 6: Comparison between CFD and NN-based

Hidden Hidden Output
Input Layer calculations for a combustion chamber pressure of
Layer 1 Layer 2 Layer
Mass Flux 60 bar.
Heat Flux
Pressure
Figure 7 and Table 6 compare the predictions of
Enthalpy
the NN with CFD results for a combustion chamber
Hot-Gas pressure of 60 bar. Figure 7 shows that the NN cap-
Channel Area
Wall tures the wall temperature curve very well, espe-
Aspect Ratio Temp. cially in the cylindrical segment. Deviations down-
Wall Thickness stream of the nozzle throat (length < 0 mm) are
Wall Roughness caused by inertia effects, which are not taken into
Curvature account. Studies are currently underway to deter-
Fin Thickness mine how these effects can also be included in the
NN. As Table 6 indicates, the full surrogate model
Figure 6: Exemplary NN architecture. matches the outlet temperature very well (conser-
vation of energy), but the pressure drop is slightly
underestimated.

5
4.2 Results for 30 and 85 bar
1000
The NN-based surrogate model is now used to study

wall temperature [K]

the cooling channel performance for other loads 800
points of the LUMEN engine, i.e. a combustion
chamber pressure of 35 bar and 80 bar. Table 7
600
shows the boundary conditions, where α is the scal-
ing factor to adjust the heat flux profile to different
3.4 kg/s
3.6 kg/s
combustion chamber pressures. α is calculated un- 400 3.8 kg/s
der the assumption that the heat flux scales with an
exponent of 0.8 with respect to the chamber pres-
4.0 kg/s
200
sure [25]. The mass flow rate ṁ and the cooling
300 200 100 0 100
channel outlet pressure pout are varied to study their
implications on the wall temperature. Typical values
length [mm]
were chosen based on the last iteration of the LU-
MEN cycle analysis. Figure 9: Wall temperature for a combustion cham-
ber pressure of 80 bar and a cooling channel outlet
parameter unit 35 bar 80 bar pressure of 105 bar.
Tin [K] 120 120
α [−] 0.65 1.26
pout [bar] 50-95 90-135 It is also interesting to study the pressure depen-
ṁ [kg s−1 ] 1.4-2.0 3.4-4.0 dency of the results. Figure 10 and Figure 11 show
the maximum wall temperature for given coolant
mass flow and cooling channel outlet pressure. As
Table 7: Boundary conditions for a combustion
already mentioned, the maximum wall temperature
chamber pressure of 35 bar and 80 bar, respectively.
for small coolant mass flows and 35 bar combustion
chamber pressure is close to the nozzle throat. The
temperature maximum decreases with higher pres-
1000 sure because transcritical effects and the influence
of HTD become weaker. The drop is followed by
wall temperature [K]

800 a horizontal evolution. This is due to the fact that

the maximum wall temperature then appears close
to the injector, where the coolant is supercritical and
600 no significant pressure dependence is present. The
1.4 kg/s same applies to the 80 bar case.
1.6 kg/s
400 1.8 kg/s
2.0 kg/s 1200
maximum wall temperature [K]

200 1.4 kg/s

300 200 100 0 100 1100 1.6 kg/s
length [mm] 1.8 kg/s
1000 2.0 kg/s
Figure 8: Wall temperature for a combustion cham- 900
ber pressure of 35 bar and a cooling channel outlet
pressure of 80 bar. 800
700
Figure 8 and Figure 9 show the result for a cham-
ber pressure of 35 bar and 80 bar and a fixed cooling 600
channel outlet pressure. For the given cooling chan- 50 60 70 80 90
nel geometry, the wall temperature maximum is ei- outlet pressure [bar]
ther close to the nozzle throat or in the cylindrical
part of the chamber. For higher coolant mass flows,
the maximum is in the cylindrical part, whereas the Figure 10: Pressure dependency of the maximum
wall temperature for lower mass flows rises sharply wall temperature for 35 bar.
close to the nozzle throat. In the 80 bar case the tem-
perature maximum is always in the cylindrical part In the LUMEN project, a maximum pressure drop
whereas in the 35 bar case both situations can occur in the cooling channel was specified. In the future, it
and transcritical effects play an essential role. Over- might be interesting to include other elements in the
all, the cooling channel design can achieve maxi- design process, such as the turbopumps, additive
mum wall temperatures below 850 K for both operat- manufacturing, fatigue life expectations of the com-
ing conditions and realistic coolant mass flows. bustion chamber [28], or control aspects [6]. Here,

6
 1200
maximum wall temperature [K]
3.4 kg/s
1100 3.6 kg/s
3.8 kg/s
1000 4.0 kg/s
900
800
700
600
90 100 110 120 130
outlet pressure [bar]

Figure 11: Pressure dependency of the maximum

wall temperature for 80 bar.

an NN-based wall temperature prediction is an ap-

pealing alternative. Finally, it might also be possi-
ble to design an entire engine or launch vehicle in a
global multidisciplinary design process to lower the
expected costs [5].

5 PART III: MANUFACTURING OF THE COM-

BUSTION CHAMBER

The following steps were performed to build the

combustion chamber:

1. turning of the first outer contour

2. milling of the cooling channels and drilling of

holes for measurement technique
(a) Picture of the manufacturing pro-
3. filling the cooling channels with wax and clos- cess. Combustion chamber after step 2.
ing with electro deposit copper (3 layers). After
each layer: drilling of holes for measurement
technique and openings for the inlet and outlet
manifolds.

4. turning of the final outer contour

5. turning of the final inner contour

5.1 Timeframe
From the beginning of the manufacturing (freezing
CAD model and drawings) until the delivery of the
final combustion chamber it took nearly 18 months.
(b) Picture of the manufacturing process. Detail of the nozzle
throat with measurement holes ( = 0.25 mm) after step 2.
It took some effort to mill the cooling channels to
the high accuracy that is necessary to gain reliable
results in the test campaigns. However the by far
longest time took the electro deposit of the copper
jacket (>12 months). Due to the technical specifics It
is not possible to accelerate this manufacturing step.
Alternatives that have to be analyzed to solve
this problem are for example the manufacturing with
SLM techniques.

7
 6 CONCLUSION

In this paper the regenerative cooling system for an

expander bleed-type engine was optimized. CFD
simulations were used to calculate the cooling chan-
nel performance and to optimize the cooling channel
geometry. An optimized solution was found that en-
sures sufficient cooling, sufficient enthalpy increase
and a minimal pressure drop.
The performance of the cooling system was vali-
dated for all load points with a neural network model
that significantly reduces the computational effort
compared with CFD simulations.
The manufacturing of the combustion chamber is
nearly completed and the tests for the finalized com-
bustion chamber are scheduled for this year.
To reduce uncertainties and to prepare future sim-
ulations, the cooling channels are extensively mea-
sured during the manufacturing process. That in-
(c) Picture of the manufacturing process. Detail of the nozzle with
openings for the manifold after the firs electro deposit layers. cludes roughness measurements at the side and
bottom part of the channels. This knowledge in addi-
tion to hot fire tests will help to design an even more
efficient cooling channel design in the future.

7 ACKNOWLEDGEMENTS

The authors would like to thank Sebastian Schulze

who prepared the optimization scheme and per-
formed the simulations, Andreas Märklen who is re-
sponsible for the combustion chamber manufactur-
ing and ArianeGroup in Ottobrunn who performed
the hot gas simulations for the heat flux boundary
condition.

REFERENCES

[1] Alliot, P. : The VINCI upper stage engine: to-

ward the demonstration of maturity. In: AIAA
2013-4054, 49 th AIAA/ASME/SAE/ASEE Joint
Propulsion Conference and Exhibit July 2013,
San Jose, California, 2013

[2] ANSYS: Theory Guide, 18.0

[3] Asakawa, H. ; Nanri, H. ; Aoki, K. ; Kubota, I. ;

Mori, H. ; Ishikawa, Y. ; Kimoto, K. ; Ishihara, S.
; Ishizaki, S. : The Status of the Research and
Development of LNG Rocket Engines in Japan.
In: Springer Aerospace Technology, Chemical
Rocket Propulsion: A Comprehesive Survey of
(d) Picture of the manufacturing process.
Combustion chamber after step 4 with final
Energetic Materials. – doi: 10.1007/978–3–
outer contour (without manifolds). 319–27748–6–19, 2017

Figure 12: Pictures of the manufacturing process [4] Börner, M. ; Manfletti, C. ; Hardi, J. ; Suslov,
D. ; Kroupa, G. ; Oschwald, M. : Laser igni-
tion of a multi-injector LOX/methane combus-
tor. In: CEAS Space Journal, 10:273-286. –
doi: 10.1007/s12567–018–0196–6, 2018

[5] Dresia, K. ; Jentzsch, S. ; Waxenegger-Wilfing,

G. ; Hahn, R. ; Deeken, J. ; Oschwald, M.

8
 ; Mota, F. : Multidisciplinary Design Opti- Fluid Dynamics for Propulsion and Power. –
mization of Reusable Launch Vehicles for Dif- doi: 10.1007/978–3–030–49626–5, 2020
ferent Propellants and Objectives. In: AIAA
Journal of Spacecraft and Rockets. – doi: [15] Haemisch, J. ; Suslov, D. ; Oschwald, M.
10.2514/1.A34944, 2021 : Experimental and Numerical Investigation
of Heat Transfer Processes in Rocket Engine
[6] Dresia, K. ; Waxenegger-Wilfing, G. ; Deeken, Cooling Channels Operated with Cryogenic
J. ; Oschwald, M. : Nonlinear Control of Hydrogen and Methane at Supercritical Con-
an Expander-Bleed Rocket Engine Using Re- ditions. In: Transactions of the Japan So-
inforcement Learning. In: Proceedings of the ciety for Aeronautical and Space Sciences,
Space Propulsion 2020+1 Conference. Virtual Aerospace Technology Japan 19(1):96-105. –
Event, 2021 doi: 10.2322/tastj.19.96, 2021
[7] Dresia, K. ; Waxenegger-Wilfing, G. ; Riccius,
J. ; Deeken, J. ; Oschwald, M. : Numeri- [16] Kato, T. ; Terakado, D. ; Nanri, H. ; Morito,
cally Efficient Fatigue Life Prediction of Rocket T. ; Masuda, I. ; Asakawa, H. ; Sakaguchi,
Combustion Chambers Using Artificial Neural H. ; Ishikawa, Y. ; Inoue, T. ; Ishihara, S. ;
Networks. In: Proceedings of the 8th Euro- Sasaki, M. : Subscale Firing Test for Regen-
pean Conference for Aeronautics and Space erative Cooling LOX/Methane Rocket Engine.
Sciences (EUCASS). Madrid, Spain, 2019 In: 7th European Conference for Aeronautics
and Space Science (EUCASS), 2017
[8] EASA: Artificial Intelligence Roadmap - A
Human-Centric Approach to AI in Aviation. [17] Leonardi, M. ; Pizzarelli, M. ; Nasuti, F. :
2020 Analysis of thermal stratification impact on the
design of cooling channels for liquid rocket
[9] Fonseca, C. M.: Genetic Algorithms for Mul- engines. In: International Journal of Heat
tiobjective Optimization: Formulation, Discus- and Mass Transfer 135 (2019) 811-821. –
sion and Generalization. In: Genetic Algo- doi: 10.1016/j.ijheatmasstransfer.2019.02.028,
rithms: Proceedings of the Fifth International 2019
Conference (S. Forrest, ed.), San Mateo, CA:
Morgan Kaufmann, July 1993, 1993 [18] Ma, H. ; Zhang, Y.-x. ; Haidn, O. J. ; Thuerey,
[10] Fröhlich, A. ; Popp, M. ; Schmidt, G. ; Thele- N. ; Hu, X.-y. : Supervised Learning Mix-
mann, D. : Heat Transfer Characteristics ing Characteristics of Film Cooling in a Rocket
of H2/O2 - Combustion Chambers. In: Combustor Using Convolutional Neural Net-
AIAA/SAE/ASME/ASEE 29th Joint Propulsion works. In: Acta Astronautica. – doi:
Conference June 28-30, 1993 / Monterey, CA. 10.1016/j.actaastro.2020.05.021, Okt. 2020. –
– doi: 10.2514/6.1993–1826, 1993 ISSN 0094–5765

[11] Goodfellow, I. ; Bengio, Y. ; Courville, A. : [19] M.Frey ; B.Kniesner ; O.Knab: Considera-

Deep Learning. Cambridge, Massachusetts : tion of Real Gas Effects and Condensation in
The MIT Press, 2016 (Adaptive Computation a Spray Combustion Rocket Thrust Chamber
and Machine Learning). – ISBN 978–0–262– Design Tool. In: 3rd European Conference
03561–3 for Aerospace Sciences, Versailles, Frabce, 6-
9 July 2009, EUCASS2009-85, 2009
[12] Haemisch, J. ; Suslov, D. ; Oschwald, M. :
Experimental Analysis of Heat Transfer Pro- [20] Negishi, H. ; Daimon, Y. ; Kawashima, H. ;
cesses in Cooling Channels of a Subscale Yamanashi, N. : Flowfield and Heat Trans-
Combustion Chamber at Real Thermal Con- fer Characteristics of Cooling Channel Flows in
ditions for Cryogenic Hydrogen and Methane. a Methane-Cooled Thrust Chamber. In: 48th
In: 6th Space Propulsion Conference, 14-18th AIAA/ASME/SAE/ASEE Joint Propulsion Con-
May 2018, Sevilla, 2018 ference & Exhibit 30 July - 1 August 2012, At-
[13] Haemisch, J. ; Suslov, D. ; Oschwald, M. : Ex- lanta, Georgia. – doi: 10.2514/6.2012–4122,
perimental study of Methane Heat Transfer De- 2012
terioration in a Subscale Combustion Chamber.
In: Journal of Propulsion and Power. – doi: [21] O.Knab ; M.Frey ; J.Görgen ; C.Maeding
10.2514/1.B37394, 2019 ; K.Quering ; D.Wiedmann: Progress in
Combustion and Heat Transfer Modelling in
[14] Haemisch, J. ; Suslov, D. ; Oschwald, M. : Rocket Thrust Chamber Applied Engineering.
Experimental Investigations of Heat Transfer In: 45th AIAA/ASME/SAE/ASEE Joint Propul-
Processes in Cooling Channels for Cryogenic sion Conference & Exhibit 2 - 5 August 2009,
Hydrogen and Methane at Supercritical Pres- Denver, Colorado, AIAA 2009-5477. – doi:
sure. In: In book: Non-Ideal Compressible 10.2514/6.2009–5477, 2009

9
[22] Pizzarelli, M. : A CFD-derived Correla-
tion for Methane Heat Transfer Deteriora-
tion. In: Numerical Heat Transfer, Part
A 2016, Vol.69, No. 3, 242-264. – doi:
10.1080/10407782.2015.1080575, 2016
[23] Pizzarelli, M. : The status of the research
on the heat transfer deterioration in super-
critical fluids: A review. In: International
Communications in Heat and Mass Transfer,
Volume 95, July 2018, Pages 132-138. – doi:
10.1016/j.icheatmasstransfer.2018.04.006,
2018
[24] Sun, G. ; Wang, S. : A Review of the
Artificial Neural Network Surrogate Modeling
in Aerodynamic Design. In: Proceedings
of the Institution of Mechanical Engineers,
Part G: Journal of Aerospace Engineering. –
doi: 10.1177/0954410019864485, Dez. 2019.
– ISSN 0954–4100

[25] Sutton, G. P. ; Biblarz, O. : Rocket Propulsion

Elements. 7th ed. New York : John Wiley &
Sons, 2001. – ISBN 978–0–471–32642–7
[26] Urbano, A. ; Nasuti, F. : Onset of Heat Trans-
fer Deterioration in Supercritical Methane Flow
Channels. In: Journal of Thermophysics and
Heat Transfer Vol. 27, No. 2, April-June. – doi:
10.2514/1.T4001, 2013
[27] Votta, R. ; Battista, F. ; Salvatore, V. ; Piz-
zarelli, M. ; Leccese, G. ; Nasuti, F. ; Meyer,
S. : Experimental investigation of transcritical
methane flow in rocket engine cooling chan-
nel. In: Applied Thermal Engineering, Vol-
ume 101, 25 May 2016, Pages 61-70. – doi:
10.1016/j.applthermaleng.2015.12.019, 2016

[28] Waxenegger, G. ; Riccius, J. ; Zametaev, E. ;

Deeken, J. ; Sand, J. : Implications of Cycle
Variants, Propellant Combinations and Operat-
ing Regimes on Fatigue Life Expectancies of
Liquid Rocket Engines. 2017

[29] Waxenegger-Wilfing, G. ; Dresia, K. ; Deeken,

J. C. ; Oschwald, M. : Heat Transfer Pre-
diction for Methane in Regenerative Cooling
Channels with Neural Networks. In: Journal
of Thermophysics and Heat Transfer. – doi:
10.2514/1.T5865, 2020. – ISSN 1533–6808

10