Bionic Liquid-Cooling Plate for Vehicle Traction Battery Thermal Management

Global shipping has entered a transition era in which national “carbon-peak and carbon-neutrality” plans accelerate the adoption of clean energy carriers. In the marine sector, the electrification of propulsion systems depends strongly on the performance, cost and lifespan of high-capacity lithium-ion batteries. These batteries constitute a vehicle traction battery when installed in an electric vessel, where they supply energy to the propulsion motor through a DC power system. A typical arrangement uses several racks of prismatic lithium iron phosphate cells connected in series and parallel to reach the voltage and capacity requirements of the vessel. The resulting battery bank is one of the most expensive components of the ship, and its safe operation is directly coupled with the thermal state inside the container.

From the viewpoint of environmental benefit, replacing a diesel generator with a vehicle traction battery eliminates a large part of CO2, SO2 and NOx emissions in harbour areas. From an economic viewpoint, the electric propulsion chain has fewer moving parts, lower vibration and lower maintenance cost compared with a conventional engine. Nevertheless, this technology does not come without risk. Lithium-ion cells are sensitive to operating temperature because the kinetics of intercalation reactions depend exponentially on temperature. The recommended window for a lithium-based vehicle traction battery is normally between 25 °C and 40 °C, while a module temperature gradient beyond 5 °C can produce uneven state of charge between parallel cells. A hot spot can also trigger exothermic side reactions that propagate to adjacent cells. In recent years, fire accidents associated with battery-powered phones, electric vehicles and stationary storage systems have illustrated the importance of preventing thermal runaway. In the marine environment, the consequences of an uncontrolled battery fire are especially severe because evacuation routes are limited. Therefore, a robust battery thermal management system (BTMS) is not an accessory but a core safety subsystem for the marine vehicle traction battery.

Three main cooling strategies have been intensively investigated in the open literature: air cooling, passive phase-change-material cooling and liquid cooling. Air cooling is simple and light, but its convective heat-transfer coefficient is low and the duct volume becomes large when the heat flux is high. Phase change materials offer high latent-heat storage and excellent temperature smoothing, yet poor thermal conductivity and leakage risk limit their standalone application. Immersion cooling is extremely effective but introduces concerns about fluid compatibility and containment. The most compact and practical method for high-rate marine modules is indirect liquid cooling. In this method, a liquid cold plate is clamped between cells or under the module, and a coolant such as water or a water–glycol mixture flows through internal channels to remove heat. The global heat-transfer coefficient of liquid cooling is one order of magnitude higher than that of air cooling, which makes it the preferred solution for a 16.18 kWh vehicle traction battery module with a sustained heat generation rate of 250 W. The major drawback is that the pump consumes additional energy, and this energy consumption is determined by the pressure drop of the cold plate. Consequently, the cold-plate designer must simultaneously manage two conflicting indexes: maximum temperature and pressure drop. A low pressure drop is desirable because the pump work can be reduced, while a low maximum temperature is required to satisfy the recommended operating window of the vehicle traction battery.

In previous work, researchers have designed straight channels, serpentine channels, wavy channels and pin-fin arrays. A straight channel has small pressure drop, but its thermal boundary layer grows thick downstream, causing a continuous temperature rise along the flow direction. A serpentine channel enhances heat transfer by creating secondary flow, but the long path sharply increases the pressure drop. Several authors have borrowed natural structures, such as leaf veins, lung branches and honeycombs, to improve the homogeneity of heat transfer. For example, a bionic leaf-vein channel was reported to keep a battery below 33 °C at 3 C discharge rate. Another study adopted a butterfly-shaped channel to decrease the maximum temperature and reduce pressure drop simultaneously. Although these attempts have produced interesting geometry, most comparisons have changed the inlet width or inlet height together with the internal topology. In that situation, the volumetric flow rate is no longer identical, and the improvement cannot be attributed solely to the internal configuration. Few studies have paid sufficient attention to the relationship between energy loss and cooling performance under the same inlet area and same velocity.

The purpose of this thesis is therefore to propose a biomimetic liquid-cooling plate whose internal channels are derived from the hexagonal vein network of dicotyledon leaves and from the branching arrangement of tree roots. The cooling plate is studied by computational fluid dynamics for a marine vehicle traction battery with a heat generation rate of 250 W. A systematic orthogonal test is designed with four geometric parameters. The significance of each parameter on maximum temperature and pressure drop is ranked by range analysis. A multi-objective genetic algorithm NSGA-II is subsequently applied to find the best trade-off between temperature and pressure drop. The optimised plate is then compared with a conventional liquid-cooling plate under identical inlet height, inlet width, flow velocity and heat flux. Finally, the effects of coolant inlet temperature, coolant velocity and heat flux on the temperature response of the plate are discussed. The study demonstrates that a cleverly arranged hexagonal-root structure can reduce the maximum temperature by about 0.7 °C and the pressure drop by about 14.2 % with respect to a conventional finned plate. In other words, the proposed design uses less pumping energy to achieve a better thermal performance for a vehicle traction battery module.

Governing Equations and Heat-Transfer Principle of the Vehicle Traction Battery

All simulations in the present study follow the principles of computational fluid dynamics. The flow is considered incompressible, steady in time and laminar for the final convergence because the Reynolds number is modest. The conservation equations include mass, momentum and energy. The mass conservation equation for an incompressible coolant can be written as:

$$\frac{\partial u_x}{\partial x} + \frac{\partial u_y}{\partial y} + \frac{\partial u_z}{\partial z} = 0$$

where $u_x$, $u_y$ and $u_z$ are the velocity components in the Cartesian directions. The momentum equation in vector form is:

$$\rho \left( \mathbf{u} \cdot \nabla \right) \mathbf{u} = -\nabla p + \mu \nabla^2 \mathbf{u}$$

where $\rho$ is the coolant density, $p$ is the static pressure and $\mu$ is the dynamic viscosity. In the solid aluminium plate, only heat conduction is considered, while in the fluid channel both advection and diffusion are present. The energy conservation equation can be expressed as:

$$\frac{\partial \left( \rho T \right)}{\partial t} + \nabla \cdot \left( \rho \mathbf{u} T \right) = \nabla \cdot \left( \frac{\lambda}{c_p} \nabla T \right) + S_T$$

Here, $T$ is temperature, $\lambda$ is the thermal conductivity, $c_p$ is the specific heat capacity and $S_T$ is the volumetric heat source. For the coolant domain the heat source is zero, whereas for the solid domain a uniform heat flux is imposed on the upper face of the plate. The above equations are solved with ANSYS Fluent, using a pressure-based coupled solver and second-order upwind discretisation for the convective terms.

The heat generation of a vehicle traction battery originates principally from four sources. The electrochemical reaction heat $Q_R$ is negative during charging and positive during discharging. The polarisation heat $Q_P$ is caused by the internal concentration and activation polarisation resistances. The Joule heat $Q_J$ is produced by the ohmic resistance of the electrodes, current collectors and electrolyte when a current flows through the cell. Side reactions $Q_S$ can generally be ignored under normal operation because they are relevant only when the cell enters the thermal-abuse regime. Therefore, the total heat generation rate can be summarised as:

$$Q_{\text{total}} = Q_R + Q_P + Q_J$$

where $Q_J = I^2 R_o$ and $Q_P = I^2 R_P$, with $I$ denoting the current, $R_o$ the ohmic resistance and $R_P$ the polarisation resistance. In the module considered here, the average heat flux is 785 W/m2. The heat is first transferred by conduction from the cell casing to the aluminium cold plate, then removed by convection from the channel walls to the coolant, and finally carried away by the bulk flow. Radiation is neglected because the temperature difference between the cold plate and the surrounding enclosure is small.

Geometry, Boundary Conditions and Mesh Verification

The geometry of the proposed bionic cold plate is illustrated by a rectangular aluminium plate with length 773 mm, width 634 mm and total thickness 6 mm. Two corrugated aluminium sheets are bonded to form the internal flow domain. The coolant enters through an inlet manifold with height 2 mm and width 32 mm, flows through an entrance region, then passes through a heat-transfer enhancement zone containing hexagonal obstacles, and finally leaves the plate through an outlet manifold of the same size. The entire structure follows the hexagonal pattern of a leaf vein and the hierarchical root-like branching of a tree.

The geometry of the internal heat-transfer enhancement zone is controlled by four structural parameters, namely $a$ which corresponds to the size of each hexagon, $b$ which corresponds to the transverse distance between two hexagons, $c$ which corresponds to the longitudinal distance between two hexagons, and $d$ which corresponds to the distance between the last hexagon row and the outlet. These four parameters were selected because they determine both the number of hexagons and the compactness of the array. The conventional liquid-cooling plate used for comparison has the same external dimensions and the same inlet area, but its internal structure consists of four rectangular partitions that divide the interior into five parallel straight channels. The honeycomb liquid-cooling plate used as a second baseline has the same hexagonal inserts but distributes them uniformly without the root-like gradient adopted in the bionic plate.

The thermophysical properties of the coolant and the solid material are listed in the following table. In this work, water is used as coolant and aluminium as the plate material. The properties are assumed to be constant because the temperature span is narrow.

Table 1 Thermophysical properties used in the CFD model
Item Material Density / kg m-3 Dynamic viscosity / kg m-1s-1 Thermal conductivity / W m-1K-1 Specific heat / J kg-1K-1
Coolant Water 998.2 0.001003 0.6 4182
Solid Aluminium 2719 — 155.5 3485

At the inlet, a uniform velocity boundary condition is prescribed. The values chosen for the baseline are 30 °C for coolant temperature and 0.2 m/s for velocity. At the outlet, the static pressure is set to zero gauge pressure. The top surface is treated as a heated wall with a constant heat flux of 785 W/m2. All other external walls are adiabatic. The Reynolds number is defined as:

$$Re = \frac{\rho U D_h}{\mu}$$

where $U$ is the inlet velocity and $D_h$ is the hydraulic diameter of the inlet rectangle. For the reference condition, the Reynolds number is approximately 680, which indicates that the laminar model is suitable. The choice of the numerical scheme is equally important. The pressure-based coupled algorithm was selected because it provides robust convergence for laminar flows involving conjugate heat transfer. The spatial discretisation uses second-order upwind for momentum and energy.

Since the hexagonal arrangement contains inclined surfaces, the computational domain cannot be easily filled with a single structured block. To reduce the memory usage and computational time, I generated a multi-block structured mesh with ICEM CFD. First, the two-dimensional cross-section of the plate was divided into blocks that follow the hexagonal walls. Then the two-dimensional mesh was extruded in the thickness direction to create a high-quality hexahedral mesh. More than 88 % of the resulting cells have an orthogonal quality above 0.9. Five mesh resolutions were tested to check whether the results were grid-independent. The mesh sizes were 987,282; 1,976,241; 3,482,563; 4,740,770 and 9,548,515 cells. A larger mesh number normally improves accuracy but increases the computing time and memory cost. The maximum temperature and pressure drop obtained with the last two meshes differ by less than 0.2 °C and less than 1 Pa, respectively. Therefore, the mesh with 4,740,770 cells was adopted for the subsequent simulations. The following table summarises the influence of mesh number on the two monitored quantities.

Table 2 Grid-independence study: effect of cell count on maximum temperature and pressure drop
Case Number of cells $T_{\max}$ / °C $\Delta P$ / Pa
M1 987,282 35.92 155.1
M2 1,976,241 35.70 161.3
M3 3,482,563 35.59 162.8
M4 4,740,770 35.51 163.8
M5 9,548,515 35.42 164.0

A validation study was also carried out. Because a direct experimental rig for the proposed bionic plate was not available, I reproduced a published liquid-cooling plate experiment of similar scale. The reference plate was 154 mm long, 79 mm wide and 6 mm thick, and the channel thickness was 2 mm. The test was performed at a 5 C discharge rate, corresponding to a simulated heat flux of 3412 W/m2 on the heating surface, with an inlet coolant temperature of 25 °C. Four mass-flow rates were investigated. The numerical results were compared with the data extracted from that experiment. At flow rates of 0.5, 1.0, 1.5 and 2.0 g/s, the predicted maximum temperatures all lie within 5 % of the experimental values. Therefore, the present numerical model is believed to be sufficiently accurate for the comparative and parametric studies that follow.

Influence of External Parameters on the Thermal-Hydraulic Behaviour

Before optimising the internal structure, it is necessary to choose two external operating parameters that remain fixed in the comparison. The first is the thickness of the fluid domain $h_f$, which is not exactly identical to the inlet height because the plate consists of a cover plate and a base plate. In a maritime container, the space allocated to cooling is restricted, so the outer thickness of the plate is always maintained at 6 mm while only the internal channel height is changed. Four fluid-domain thicknesses, 1, 2, 3 and 4 mm, were investigated. The inlet velocity is kept equal to 0.2 m/s, so an increase in thickness directly produces a proportional increase in volumetric flow rate.

The following table presents the maximum temperature and pressure drop as functions of the fluid-domain thickness.

Table 3 Effect of fluid-domain thickness at fixed inlet velocity 0.2 m/s
Fluid-domain thickness / mm $T_{\max}$ / °C $\Delta P$ / Pa
1 36.1 301.5
2 35.5 163.8
3 35.8 127.2
4 36.3 103.4

Several competing mechanisms explain these results. When the channel becomes thicker, the mass-flow rate through the channel rises and thereby increases the heat carried away by the coolant. At the same time, the conductive path in the aluminium between the heated surface and the bottom channel wall also changes. A very thin wall above the channel improves conduction because the aluminium path is short; a very thick channel reduces the remaining wall thickness and degrades conduction. The optimum is therefore not monotonic. The pressure drop always decreases with thickness because a larger hydraulic diameter reduces the velocity gradient at the wall and the frictional resistance. In this study, 2 mm was chosen because it gives the lowest temperature. Although the pressure drop at 3 mm is lower by about 36 Pa, the temperature is 0.3 °C higher, and the thinner wall also leaves more aluminium for mechanical stiffness. Thus, the fluid-domain thickness was fixed to 2 mm for the remaining simulations.

The second external parameter is the inlet coolant velocity. Five values from 0.1 to 0.3 m/s were considered for the bionic liquid-cooling plate, the conventional liquid-cooling plate and the honeycomb liquid-cooling plate. The results are summarised in the following table. Increasing the velocity generally lowers the maximum temperature, but the amount of improvement decreases as the velocity grows because the thermal resistance in the boundary layer becomes less dominant. The energy required by the pump, however, increases approximately with the square of velocity. Thus, a speed that is too high leads to diminishing returns and excessive energy consumption. A velocity of 0.2 m/s offers a good compromise because the temperature reduction from 0.2 to 0.3 m/s is smaller than the reduction from 0.1 to 0.2 m/s for all three plates. This velocity also keeps the Reynolds number below the laminar-to-turbulent transition, which simplifies the numerical modelling and improves the stability of the solution.

Table 4 Effect of inlet velocity on maximum temperature for three cooling plates
Velocity / m s-1 $T_{\max}$: conventional / °C $T_{\max}$: honeycomb / °C $T_{\max}$: bionic / °C
0.10 39.9 39.3 39.0
0.15 37.5 36.9 36.6
0.20 36.2 35.6 35.5
0.25 35.3 34.8 34.7
0.30 34.7 34.3 34.1

Orthogonal Design and Structural Optimisation

After the external parameters are fixed, the attention is focused on the internal arrangement. The four structural parameters $a$, $b$, $c$ and $d$ form a four-factor-four-level test matrix. A full factorial design would require $4^4=256$ configurations, which would be prohibitively expensive. The orthogonal test method is a fractional factorial design that significantly reduces the number of cases while preserving the statistical independence of the factors. An L16(44) orthogonal array was selected, which means that only sixteen configurations are needed. The levels of each factor are listed in the table below.

Table 5 Level assignment for the four geometric parameters
Factor Description Level 1 Level 2 Level 3 Level 4
$a$ size of hexagon / mm 20 26 32 38
$b$ transverse spacing / mm 8 12 16 20
$c$ longitudinal spacing / mm 30 40 50 60
$d$ distance from last hexagon to outlet / mm 80 90 100 110

Each configuration was built, meshed and run under the same boundary condition: coolant temperature 30 °C, velocity 0.2 m/s and heat flux 785 W/m2. Since the inlet and outlet dimensions are identical in all cases, the volumetric flow rate remains constant throughout the orthogonal matrix. Therefore, the pressure drop can be interpreted as a direct measure of the pumping energy required for the vehicle traction battery cooling loop. The maximum temperature and pressure drop for all sixteen cases are listed in the following table. The maximum temperature varies by almost 1 °C, while the pressure drop varies by more than 25 Pa.

Table 6 L16 orthogonal test results for the bionic liquid-cooling plate
Case $a$ / mm $b$ / mm $c$ / mm $d$ / mm $T_{\max}$ / °C $\Delta P$ / Pa
1 20 8 30 80 35.43 173.47
2 20 12 40 90 36.08 158.21
3 20 16 50 100 36.14 157.52
4 20 20 60 110 36.33 153.46
5 26 8 50 90 35.46 171.68
6 26 12 60 80 35.88 158.92
7 26 16 30 110 36.06 157.39
8 26 20 40 100 36.35 157.34
9 32 8 60 100 35.45 167.94
10 32 12 50 110 35.84 165.11
11 32 16 40 80 36.16 170.11
12 32 20 30 90 36.13 154.75
13 38 8 40 110 35.47 180.66
14 38 12 30 100 35.81 158.73
15 38 16 60 90 35.91 158.69
16 38 20 50 80 35.96 158.38

The spatial temperature distributions of all sixteen configurations share a common feature: the hottest region always appears at the two corners near the outlet. This happens because the coolant gradually absorbs heat while it travels through the plate, so the area close to the outlet has the highest fluid temperature. The lowest temperature is found near the inlet where the coolant is fresh. The temperature cloud maps also demonstrate that a proper combination of the parameters can make the cloud map more uniform. The conventional plate shows a ‘convex’ shape in the isotherms of the rear part, while the bionic plate generally shows a more ‘concave’ pattern. The concave isotherms indicate that the heat-transfer wall is cooled more effectively in the streamwise direction.

A range analysis was performed to quantify which geometric factor has the strongest influence. For each factor, the average value of the maximum temperature at each level was calculated. Let $K_i$ denote the average of the results for level $i$. The range $R$ is defined as:

$$R = \max\left(K_1, K_2, K_3, K_4\right) – \min\left(K_1, K_2, K_3, K_4\right)$$

A larger value of $R$ indicates that the corresponding factor is more influential. The following table gives the complete range-analysis results, where the column represented by $T$ refers to maximum temperature and the column represented by $P$ refers to pressure drop.

Table 7 Range analysis for maximum temperature and pressure drop
Statistical variable Factor $a$ Factor $b$ Factor $c$ Factor $d$
$K_1$ for $T_{\max}$ / °C 36.00 35.45 35.86 35.86
$K_2$ for $T_{\max}$ / °C 35.94 35.90 36.02 35.90
$K_3$ for $T_{\max}$ / °C 35.90 36.07 35.85 35.94
$K_4$ for $T_{\max}$ / °C 35.79 36.19 35.89 35.93
$R$ for $T_{\max}$ 0.21 0.74 0.17 0.08
$K_1$ for $\Delta P$ / Pa 160.67 173.44 161.09 165.22
$K_2$ for $\Delta P$ / Pa 161.33 160.24 166.58 160.83
$K_3$ for $\Delta P$ / Pa 164.48 160.93 163.17 160.38
$K_4$ for $\Delta P$ / Pa 164.12 155.98 159.75 164.16
$R$ for $\Delta P$ 3.81 17.46 6.83 4.84

For the maximum temperature, the significance order is $b > a > c > d$. This means that the transverse spacing between hexagons is the most important parameter that determines the thermal performance. A smaller transverse spacing creates more hexagons in each row and therefore increases the heat-transfer area, but it also restricts the flow area and raises the pressure drop. The distance $d$ between the last row and the outlet has the weakest effect on temperature because the flow in that zone has already been fully developed by the preceding hexagonal rows. For the pressure drop, the significance order is $b > c > a > d$. The transverse spacing again dominates the pressure drop, followed by the longitudinal spacing. This is consistent with the fact that a narrow gap between two cylinders or hexagons produces a large local velocity and thus a high wall shear stress.

To obtain a trade-off between the conflicting objectives, I used a regression approach to transform the discrete orthogonal-table results into continuous approximate functions. Let $x_1=a$, $x_2=b$, $x_3=c$ and $x_4=d$. The fitted linear equations are expressed as:

$$y_p = 0.224917 x_1 – 1.292 x_2 – 0.07405 x_3 – 0.03645 x_4 + 181.0079$$

$$y_t = -0.01108 x_1 + 0.059625 x_2 – 0.0006 x_3 + 0.00245 x_4 + 35.18467$$

where $y_p$ denotes the pressure drop in Pascal and $y_t$ denotes the maximum temperature in degrees Celsius. The coefficient of determination is defined as:

$$r^2 = 1 – \sum_{i=1}^{n}\left( \frac{Z_i – Z_{i,\mathrm{CFD}}}{Z_{i,\mathrm{CFD}}} \right)^2$$

where $Z_i$ is the fitted value and $Z_{i,\mathrm{CFD}}$ is the simulated value for the $i$-th orthogonal case. The calculated coefficient is 0.81 for temperature and 0.63 for pressure drop. Although the pressure-drop regression is not perfect, the accuracy is sufficient for the purpose of initial screening during multi-objective optimisation. A further verification simulation at the selected optimum is then carried out to ensure that the predicted value is realistic.

The optimisation problem can be stated as a two-objective minimisation problem:

$$\min F(x_1,x_2,x_3,x_4) = \begin{cases} \min y_t(x_1,x_2,x_3,x_4) \\[4pt] \min y_p(x_1,x_2,x_3,x_4) \end{cases}$$

$$\text{subject to} \quad \begin{cases} 20 \leq x_1 \leq 38 \\[3pt] 8 \leq x_2 \leq 20 \\[3pt] 30 \leq x_3 \leq 60 \\[3pt] 80 \leq x_4 \leq 110 \end{cases}$$

The NSGA-II algorithm was selected because of its excellent performance in handling non-convex multi-objective optimisation problems. NSGA-II sorts the population according to non-dominated fronts and uses a crowding-distance metric to preserve diversity. In this study, the population size was set to 50, the maximum number of generations was set to 150 and the fitness-function tolerance was set to 0.0001. At the end of the evolution, the Pareto front was obtained. The front clearly shows that the maximum temperature and the pressure drop cannot be improved simultaneously. When the maximum temperature is reduced, the pressure drop tends to increase, and vice versa. Because both the temperature and the pressure drop affect the performance of the marine vehicle traction battery cooling system, it is not possible to choose an all-encompassing optimised solution from the front solely by mathematical reasoning. In engineering practice, the balance between them depends on the size of the pump and the available power.

In the present plate, the pressure drop of all sixteen configurations is below 181 Pa, which is already lower than the conventional plate’s 190.91 Pa mentioned earlier. Therefore, the designer has some freedom to select a solution that favours maximum temperature. From the Pareto front, I selected an intermediate point with a maximum temperature of 35.49 °C and a pressure drop of 170.08 Pa. The corresponding geometric parameters are $a=30$ mm, $b=8$ mm, $c=50$ mm and $d=90$ mm. A detailed three-dimensional simulation was then run using the selected parameters. The simulation gives a maximum temperature of 35.51 °C and a pressure drop of 163.79 Pa. The comparison between the algorithm prediction and the CFD result is summarised in the following table. The maximum-temperature error is 0.00056 and the pressure-drop error is 0.038, both of which are considered acceptable. Thus, the selected optimum configuration is reliable and is adopted in the next section for further analysis.

Table 8 Comparison between NSGA-II prediction and independent CFD simulation at the selected optimum
Quantity Algorithm-predicted value CFD-simulated value Relative error
Maximum temperature / °C 35.49 35.51 0.00056
Pressure drop / Pa 170.08 163.79 0.038

Comparative Analysis of the Bionic Liquid-Cooling Plate and a Conventional Plate

The optimal bionic plate is now compared with a conventional straight-channel plate and a uniform honeycomb plate under identical conditions: inlet velocity 0.2 m/s, coolant temperature 30 °C, heated-surface heat flux 785 W/m2 and fluid-domain thickness 2 mm. The conventional plate is denoted as CLCP, the plate with uniform hexagons as HLCP, and the proposed bionic liquid-cooling plate as BLCP. The pressure fields of the three plates are presented in Figure-like contour maps. Although direct cloud images are not reproduced in this text, the numerical values can be clearly compared.

The pressure drop of the CLCP is 190.91 Pa, that of the HLCP is 165.66 Pa and that of the optimised BLCP is 163.79 Pa. The bionic plate produces a pressure drop that is 14.2 % lower than the conventional plate and about 1.1 % lower than the uniform honeycomb plate. Because the flow rate is identical in all three cases, the reduction in pressure drop directly corresponds to a reduction in pumping power. The reason for this advantage is visible in the difference between the internal topology. In the conventional plate, the inlet jet enters a straight manifold and then splits into five separate channels separated by long rectangular partitions. The coolant travels the entire length of the plate in one straight line. Although the five channels are wide, the total wetted area is much smaller than that of the bionic plate. In the bionic plate, the flow repeatedly encounters hexagonal obstacles whose sharp corners face the oncoming flow. The combination of a streamlined front edge and the splitting of the flow into branches reduces the formation of a large recirculation zone. The pressure-loss reduction is thus caused not by lowering the velocity but by arranging the obstacles in an optimal staggered pattern that permits a smoother directional change of the coolant.

For the temperature field, the CLCP reaches a maximum temperature of 36.20 °C, the HLCP reaches 36.00 °C and the BLCP reaches 35.51 °C. The bionic plate therefore cools the heated surface 0.69 °C more effectively than the conventional plate and about 0.5 °C more effectively than the honeycomb plate. It should be remembered that this temperature improvement is achieved with a lower pressure drop, which is unusual because most heat-transfer enhancement techniques increase pressure loss. The good performance of the bionic plate is associated with the division of the plate into two thermal zones. In the entrance zone, where the conventional straight plate already has a fully developed laminar boundary layer, the bionic plate has no obstacle and thus behaves similarly to a plain duct. Once the coolant reaches the enhancement region, the row of hexagonal elements breaks the thermal boundary layer repeatedly. Each time the coolant flows around a hexagonal obstacle, a new boundary layer starts to develop, and the local heat-transfer coefficient near the leading edge of the obstacle is very high. Between the staggered rows, the coolant wakes interact with the downstream obstacles, increasing turbulence-like mixing even in a laminar regime. This effect is stronger in the bionic plate than in the uniform honeycomb plate because the longitudinal distance and transverse distance were selected to maximise the interaction between the wake of one row and the next row.

In addition to the maximum temperature, the temperature uniformity is an important index for a vehicle traction battery. A large temperature difference inside the module directly produces a state-of-charge mismatch between cells and accelerates the ageing of the hottest cell. The isotherm patterns in the CFD results show that the conventional plate has a ‘convex’ bulge in the rear part of the plate, meaning that the coolant near the centre of the domain is heated much faster than near the lateral walls. The bionic plate, in contrast, produces a temperature distribution with a more regular gradient. The highest temperature still appears at the two rear corners, but the area of this hot zone is smaller than that of the conventional plate. The corresponding maximum temperature difference across the heated surface is below 5 °C. Such uniformity satisfies the recommended thermal specification of a vehicle traction battery pack.

The transient response was also computed. Starting from a uniform temperature of 30 °C, the maximum temperature of the BLCP and CLCP rises rapidly at the beginning because the heat flux is larger than the instantaneous heat-removal rate. After roughly 50 s, the slope changes and the temperature gradually approaches a plateau. The rate at which the temperature approaches the plateau is higher for the BLCP, which means that the bionic plate reaches the quasi-steady state sooner. At steady state, the final maximum temperatures differ by approximately 0.7 °C. In the first 50 s, the conventional plate appears slightly better, but this is simply because the heat has not yet propagated to the entrance of the enhancement zone in the bionic plate. From the perspective of long-duration marine operation, only the steady-state behaviour is relevant.

Figure-like velocity contour plots reveal a significant difference in the local velocity distribution between the CLCP and the BLCP. In the conventional plate, the flow velocity inside the five straight channels is almost uniform and has a value close to 0.07 m/s. The maximum velocity appears near the outlet manifold. In the bionic plate, the velocity field outside the enhancement zone remains moderate, while inside the enhancement zone the flow is strongly decelerated as it spreads around the hexagons. The local velocity in the narrow gap between adjacent hexagons can be higher than the inlet velocity, whereas the velocity in the wake region behind each hexagon is very small. This combination of high-velocity regions and low-velocity regions increases the residence time of the coolant in contact with the heated wall and thus improves the heat exchange. However, a greater residence time is not always beneficial because the coolant becomes warmer; in this geometry, the tree-root arrangement directs the coolant progressively from the central inlet to the lateral zones, ensuring that the whole surface is washed by relatively fresh coolant.

A useful approach for interpreting the heat-transfer enhancement is the field-synergy principle. The synergy angle between the velocity vector and the temperature-gradient vector can be expressed as:

$$\cos \beta = \frac{\mathbf{U} \cdot \nabla T}{\left| \mathbf{U} \right| \left| \nabla T \right|}$$

When the absolute value of the synergy angle is small, i.e. when the velocity vector is nearly parallel to the temperature gradient vector, the convective heat transfer is most effective for the same pump power. In a fully developed straight channel, the velocity is parallel to the wall, while the temperature gradient is largely perpendicular to the wall, so the synergy angle approaches 90 ° and the convective term is weak. By inserting obstacles into the flow, the velocity direction is locally deflected and the synergy angle decreases. The CFD-POST software was used to post-process the synergy-angle distribution. In the CLCP, only the leading and trailing edges of the rectangular partitions generate a zone with a synergy angle below 90 °. The area of these zones is limited. In the BLCP, every hexagonal element produces a wake and a stagnation region that alternately change the direction of the velocity. Consequently, the area of the domain with a synergy angle below 90 ° is larger. The field-synergy angle map proves that the equivalent heat-transfer coefficient of the bionic plate is genuinely higher than that of the conventional plate, and not merely caused by an increased contact area.

Table 9 Comparison of steady-state performance indexes for the three cold plates
Configuration Maximum temperature / °C Pressure drop / Pa Pressure drop reduction relative to CLCP / % Temperature reduction relative to CLCP / °C
CLCP 36.20 190.91 — —
HLCP 36.00 165.66 13.2 0.20
BLCP 35.51 163.79 14.2 0.69

Influence of Operating Conditions on the Bionic Plate

The external climate on a ship changes with the navigation area, and the discharge power of the vehicle traction battery changes with the sailing condition. Therefore, the response of the optimised bionic plate to coolant temperature, coolant velocity and heat flux was investigated separately.

Keeping the heat flux at 785 W/m2 and the velocity at 0.2 m/s, the coolant inlet temperature was changed from 25 to 30 to 35 °C. The maximum plate temperature at the outlet was 30.46, 35.51 and 40.41 °C, respectively. Although the absolute temperature increases with the inlet temperature, the difference between the outlet and the inlet remains almost constant at about 4.17 °C. This result is consistent with the energy balance: when the inlet boundary temperature is increased while the flow rate and heat flux are unchanged, the absorbed heat is the same, so the coolant temperature rise is nearly identical. The lower coolant temperature has the advantage of bringing the plate closer to the recommended operating window of the vehicle traction battery, and it also shortens the time required for the maximum temperature to stabilise. The stabilisation time is around 150 s at 25 °C and increases to about 250 s at 35 °C. In a module where a periodic charge-discharge cycle is shorter than 250 s, the transient thermal behaviour must be considered because the thermal mass of the aluminium plate has an important buffering effect.

Keeping the coolant temperature at 30 °C and the heat flux at 785 W/m2, the inlet velocity was changed from 0.1 to 0.2 to 0.3 m/s. The maximum plate temperature decreases from 39.25 to 35.51 to 34.12 °C, respectively. The difference between the outlet and inlet temperatures also decreases from about 8 K to 4.17 K to 2.84 K. At the same heat flux, the higher velocity brings a larger mass-flow rate, so a given amount of heat produces a smaller temperature rise of the coolant. The transient stabilisation time is about 250 s at 0.1 m/s, 150 s at 0.2 m/s and only 100 s at 0.3 m/s. This trend demonstrates that the inertia of the thermal system is reduced by increasing the coolant flow rate.

Keeping the coolant temperature at 30 °C and the velocity at 0.2 m/s, the heat flux was varied from 400, 600, 785, 1000 to 1200 W/m2. The corresponding maximum temperatures are 32.77, 34.10, 35.51, 37.10 and 38.30 °C. The outlet-to-inlet temperature differences are 2.12, 3.18, 4.17, 5.30 and 6.37 K, respectively. In this set of calculations, the measured thermal time to steady state is roughly 150 s for the low heat-flux cases and 200 s for the highest heat-flux cases. Thus, the heat flux has a small effect on the thermal time constant compared with the coolant velocity or inlet temperature. The results show that the proposed bionic plate is capable of keeping the maximum temperature below 40 °C for all investigated heat fluxes up to 1200 W/m2 when the coolant is 30 °C. This provides a comfortable safety margin for the marine vehicle traction battery application.

Conclusions and Outlook

This thesis has presented a systematic numerical investigation of a biomimetic liquid-cooling plate that imitates a hexagonal leaf-vein pattern and a tree-root-like branching structure. The study was carried out for a marine vehicle traction battery module with a reference heat generation of 250 W. Sixteen orthogonal configurations were proposed to explore the influence of the hexagon size, the transverse distance, the longitudinal distance and the distance from the last row to the outlet. A multi-objective optimisation based on NSGA-II was subsequently employed to balance the maximum temperature and the pressure drop. The optimised plate was then compared in detail with a conventional straight-fin plate. The conclusions can be summarised as follows.

First, the transverse spacing between hexagons is the dominant factor for both the maximum temperature and the pressure drop. The influence order for maximum temperature is $b > a > c > d$, while the order for pressure drop is $b > c > a > d$. A smaller transverse spacing generally improves the heat-transfer area but increases the pressure loss. The distance from the last row of hexagons to the outlet is of secondary importance in both objectives.

Second, the structural optimisation produces a meaningful improvement at a fixed flow rate. The selected optimum geometry with $a=30$ mm, $b=8$ mm, $c=50$ mm and $d=90$ mm yields a maximum temperature of 35.51 °C and a pressure drop of 163.79 Pa under the reference boundary condition. The NSGA-II prediction agrees well with the CFD verification simulation, thereby proving that the fitted model and the optimisation procedure are sufficiently reliable for the considered design space.

Third, the bionic plate differs from a conventional plate mainly in the distribution of local speed and the repeated interruption of the thermal and hydrodynamic boundary layers. In the enhancement region, the flow repeatedly splits and recombines around the hexagonal obstacles. Each obstacle produces a recirculation zone and a new boundary layer, which increases the heat-transfer coefficient in a laminar flow domain. The synergy-angle analysis convincingly demonstrates that a larger fraction of the bionic flow domain has an angle below 90 ° between the velocity vector and the temperature-gradient vector. As a result, the bionic plate lowers the maximum temperature by 0.7 °C while also lowering the pressure drop by 14.2 % compared with a conventional plate.

Fourth, the external operating parameters strongly determine the plate behaviour. The coolant temperature has a negligible effect on the coolant temperature rise across the plate, but it directly shifts the absolute temperature and affects the thermal stabilisation time. An increase in coolant velocity reduces both the maximum temperature and the outlet-to-inlet temperature difference, but above 0.3 m/s the additional gain becomes smaller. Finally, an increase in heat flux linearly raises the outlet temperature difference, and the bionic plate is capable of maintaining the maximum temperature below 40 °C for heat fluxes up to 1200 W/m2 with 30 °C coolant at 0.2 m/s.

Some limitations of the present work should be mentioned. The CFD model assumes constant thermophysical properties and ignores the temperature dependence of the cell heat source. The mechanical deformation of the cold plate and the contact resistance between the battery and the plate are not included. In future work, a coupled electrochemical-thermal model of the vehicle traction battery should be incorporated so that local current and heat-generation distributions can be predicted more accurately. Experimental validation of the optimised bionic plate should also be performed on a full-size module test bench. In addition, more advanced optimisation algorithms such as grey-wolf optimisation or whale optimisation can be compared with the NSGA-II results to verify the global optimum. The proposed hexagonal-tree-root structure opens a promising avenue for low-energy and high-uniformity thermal management of marine batteries.

Scroll to Top