Professional Documents
Culture Documents
Diabatix - P3-Battery Cold Plate Design
Diabatix - P3-Battery Cold Plate Design
Diabatix - P3-Battery Cold Plate Design
generative design
Designing battery
cold plates through
generative design
Diabatix publications, No. 2022.3
C
onventional cold plate designs are man- is crucial to achieving its longer lifespan. For this
ufactured widely to provide cooling for reason, the most common problems engineers
the batteries in electric cars. While com- face during the design of cold plates are reach-
panies seek more efficient alternatives to man- ing a low global temperature of the battery cells
age the battery cooling system, human partic- while keeping its uniformity and the pressure drop
ipation and trial-and-error scenarios are still inside the cold plate within an acceptable range.
dominant in the engineering design process. Furthermore, engineers should be attentive to the
The current technologies can make this process manufacturability of the product. This should al-
smoother using generative design with topology ways align with the minimization of the weight of
optimization. the product.
explanation of the steps for a topology optimiza- Generative design using topology opti-
tion analysis. A series of battery cold plate studies mization
provide the means for comparison of their perfor-
mance. This part explores the assumptions made In generative design, different techniques exist to
for the cold plate modeling and its analysis, fin- gather and process information from subsequent
ishing with the performance results. The paper design iterations. A popular generative design
concludes with a discussion comparing the cool- strategy uses topology optimization. For this tech-
ing systems. nique, optimizing the material distribution in the
design space results in a new design [3, 4, 5]. In
broad terms, a part of the parent material trans-
forms into a new material, which defines a design.
the temperature field throughout the solid do- tion describes the heat transfer in solid domains.
mains. The solutions for the different domains are further
2. Performance evaluation. The numerical so- conjugated.
lution obtained in (1) is used to evaluate the
performance of the design regarding the op-
timization targets. If the performance meets
Fluid region modeling
the optimization target, the optimization pro- The general equations solved in the fluid domain
cess ends, and the design is exported. Other- are the Navier-Stokes equations consisting of [7]:
wise, the optimization loop continues.
3. Sensitivity analysis. This step involves de- • Conservation of mass
termining how the design should change to
improve the performance. This is achieved ∇ · ρυ = 0 (1)
by assessing the sensitivity of the design per-
• Conservation of momentum
formance for the material distribution. The
adjoint approach calculates the sensitivities,
given a large degree of freedom in the design ∂ρυ
+ ∇ρυυ = −∇p + ∇ · µt ∇υ + f (2)
∂t
space. This mathematical framework allows
performing sensitivity analysis in a compu- • Conservation of energy
tationally efficient way. In this analysis, the
∂ρh k
adjoint is built following a continuous adjoint + ∇ρυh − ∇ · ∇h = Q (3)
∂t cp
approach [6].
4. Design update. The optimization algorithm where ρ is the density of the fluid, υ is its velocity,
proposes a new design based on the current p is its pressure, µt is its turbulent viscosity, f is a
design and the sensitivity analysis. This de- general momentum force term (e.g. gravity), h is
sign serves as the starting point for the next the enthalpy of the fluid, k its thermal conductivity,
optimization iteration. and cp is the specific heat capacity.
In the case study analysis, topology optimization Closure of the momentum equation is provided
is applied to the design of the cooling channels by applying turbulence models to compute the
of a cold plate. Topology optimization for fluid turbulent viscosity µt . Various turbulence models
cooling problems aims to find the optimal material exist, and the choice depends on the physics of
distribution. It seeks to minimize a cost objective each application.
while subject to operating and manufacturing con-
The two main turbulence model techniques are
straints. In other words, an optimal solid design is
RANS (Reynolds-Averaged Navier-Stokes) and
generated within a fluid parent region by increas-
LES (Large Eddy Simulation). RANS is obtained by
ing or reducing the solids at each cell.
applying a time-averaging and Reynolds decompo-
sition technique on the Navier-Stokes equations.
LES results from applying a spatial filter on the
MODELING THE COOLING SYS- Navier-Stokes equations. For wall-bounded flows,
TEM RANS is the most commonly used approach. Thus,
it is the turbulent model for the case study analy-
A battery pack typically consists of one or multiple sis.
solid parts and a liquid coolant. The simulation
of the type of system is known as a Conjugate
Solid region modeling
Heat Transfer (CHT) problem. In CHT problems,
Navier-Stokes and energy equations describe fluid For the solid region, only conservation of energy
behavior. On the other hand, the conduction equa- applies [1]:
optimal design.
∂ρh k
− ∇ · ∇h = Q (4)
∂t cp
5b) displays an increase in the battery tempera- onal temperature stream is displayed with a lower
ture along the way toward the outlet. This uneven peak temperature (Fig. 5c). The model converges
temperature distribution causes cold regions and a to a uniform flow velocity along the plate channels.
thermal hotspot, which is harmful to the battery in
the long term. A low flow velocity is acknowledged
Generative design
for this model.
The generative design sets targets and constraints
to obtain an optimized model. The pressure
drop and temperature constraints are adjusted
for 5.5 kPa and ∆T = 5 ◦ C, respectively, while
the manufacturing constraint is the sheet metal
foaming. The peak temperature optimization is
the thermal design target. Fig. 6 shows the final
generative design.
DISCUSSION
(c) Fine dimpled design
The temperature distribution of the battery is an
Figure 5: Conventional cold plate designs results: the important variable in deciding on a cold plate de-
top picture is displaying the temperature on
the battery and the bottom one refers to the sign. The chart presented in Fig. 8 shows the dif-
velocity of the coolant inside the plate chan- ference in temperature between the battery and
nels.
the inlet, ∆T = Tbattery − Tinlet .
Table 1: Summary of the results (peak temperature, temperature uniformity, pressure drop, and manufacturability)
related to each one of the cold plate designs.
Table 1 summarizes the results for each one of The performance is compared to two traditional
the designs. It contains the peak and uniformity designs, one S-shaped and two dimpled designs.
of the temperature, the pressure drop inside the In conclusion, the conventional designs can only
channel, and whether or not the plate is manufac- perform similarly to the generative design at the
turable. expense of a higher pressure drop or a higher tem-
perature distribution.
The conventional designs can operate as de-
sired, either for peak temperate, uniformity of tem-
perature, or pressure drop. However, they do not
fulfill all the scenarios simultaneously, as with the
custom design. With the overview of the results,
it is clear the cold plate generative design has an
ultimate performance.
Extra examples
(a)
All analyses and design were created using Dia-
batix proprietary platform ColdStream [8]. In Fig.
11, two more examples of battery cold plates gen-
erated by ColdStream are shown.
(b)
CONCLUSION
Figure 11: Other examples of battery cold plate designs
The added value of the generative thermal design generated with ColdStream.
process is illustrated through the design of a bat-
tery cold plate. The methodology autonomously
suggests cooling system geometries in natural
freeform shapes reaching an optimized design.
AUTHORS
Sarah da Silva Andrade
Marketing Engineer, Diabatix
Joao Miranda
R&D Engineer, Diabatix
Lieven Vervecken
CEO, Diabatix
References
[1] K. Shea, R. Aish, and M. Gourtovaia. “Towards
integrated performance-driven generative
design tools”. In: Automation in Construction
14.2 (2005), pp. 253–264.
[2] S. Oh et al. “Deep generative design: Integra-
tion of topology optimization and generative
models”. In: Journal of Mechanical Design
141.11 (2019).
[3] M. P. Bendsoe and O. Sigmund. Topology op-
timization: theory, methods, and applications.
Springer Science & Business Media, 2003.
[4] O. Sigmund and K. Maute. “Topology op-
timization approaches”. In: Structural and
Multidisciplinary Optimization 48.6 (2013),
pp. 1031–1055.
[5] J. Alexandersen and C. S. Andreasen. “A re-
view of topology optimisation for fluid-based
problems”. In: Fluids 5.1 (2020), p. 29.
[6] S. Nadarajah and A. Jameson. “A comparison
of the continuous and discrete adjoint ap-
proach to automatic aerodynamic optimiza-
tion”. In: 38th Aerospace Sciences Meeting
and Exhibit. 2000, p. 667.
[7] C. Hirsch. Numerical computation of inter-
nal and external flows: The fundamentals of
computational fluid dynamics. Elsevier, 2007.
[8] Diabatix ColdStream ©. Version Coldstream
2. url: https://coldstream.diabatix.com.