In modern electric vehicles, the traction battery pack is the primary energy storage unit and its performance is deeply coupled with the operating temperature. The EV battery pack must be maintained within a narrow temperature window to guarantee high power output, long cycle life, and operational safety. However, during high-rate discharge or fast charging, the internal heat generation rate of the cells can spike dramatically, leading to local hot spots and thermal non-uniformity. If the generated heat is not dissipated efficiently and uniformly, the EV battery pack may suffer from accelerated aging, capacity fade, or even thermal runaway. Therefore, an advanced battery thermal management system is of paramount importance for the reliable and efficient deployment of EV battery packs.
Liquid cooling is the most widely adopted solution for high-power EV battery packs due to its high heat capacity and favourable transport properties. Among liquid-cooled systems, the cold plate serves as the critical interface between the cells and the coolant. The design of the cold plate channels directly determines the thermal performance, temperature uniformity, and pumping power consumption. Traditional parallel straight channels are simple but often suffer from uneven flow distribution and limited heat transfer enhancement. In two-phase cooling, the situation becomes even more challenging because the downstream accumulation of vapor bubbles can lead to elongated slug flow, which further induces bubble backflow and flow oscillation, ultimately deteriorating the heat transfer performance of the cold plate. This issue is particularly severe in large-area cold plates used in EV battery packs, where parallel channels are prone to boiling non-uniformity and localized dry-out.
To overcome these limitations, this work proposes a novel topology-optimized structure array that reconstructs the parallel straight channels to suppress bubble backflow and enhance heat transfer. The research combines topology optimization, conjugate heat transfer simulation, multi-objective optimization, and experimental validation. The study demonstrates that the optimized cold plate can significantly improve the temperature uniformity and reduce the pressure drop penalty, making it a promising solution for high-efficiency thermal management of EV battery packs.
Topology Optimization of the Flow Structure
The first step of this study is to design a flow pattern that exhibits a strong flow diodicity effect, meaning that the pressure drop is much larger in one direction than in the opposite direction. This feature can prevent backflow during boiling in parallel channels. I adopted a density-based topology optimization method to redesign a two-dimensional cross-section of a straight channel. The design domain is a rectangular region with a width of 22.5 mm and a height of 10.5 mm, as shown in the earlier section. A velocity inlet is set at the left boundary, and a pressure outlet at the right boundary. The design variable is a continuous density field that indicates the presence of fluid or solid material.
In the topology optimization formulation, the domain is treated as a porous medium with variable permeability. The material interpolation is expressed as:
$$ \alpha(\gamma) = \alpha_f + (\alpha_s – \alpha_f) \frac{q(1-\gamma)}{q+\gamma} $$
where \(\gamma\) is the design density (0 for solid, 1 for fluid), \(q\) is a penalization factor, and \(\alpha_f\) and \(\alpha_s\) are the inverse permeabilities of the fluid and solid phases, respectively. The solid inverse permeability is defined as:
$$ \alpha_s = \frac{\mu}{Da \, L_c} $$
with \(\mu\) the dynamic viscosity, \(Da\) the Darcy number (taken as \(10^{-4}\)), and \(L_c\) the characteristic length.
The flow is governed by the incompressible Navier–Stokes equations with a body force term representing the porous medium:
$$ \frac{\partial u_i}{\partial x_i} = 0 $$
$$ \rho \left( \frac{\partial u_i}{\partial t} + u_j \frac{\partial u_i}{\partial x_j} \right) = -\frac{\partial p}{\partial x_i} + \mu \frac{\partial^2 u_i}{\partial x_j \partial x_j} + f_i $$
Here, \(f_i = -\alpha(\gamma) u_i\) is the artificial body force. The objective is to maximize the diodicity \(Di\), defined as the ratio of pressure drop in the reverse flow to that in the forward flow:
$$ Di = \frac{\Delta P_{\text{reverse}}}{\Delta P_{\text{forward}}} $$
To ensure a manufacturable and clear solid–fluid interface, I applied a Helmholtz filter with radius \(R_{\min}\) followed by a hyperbolic tangent projection:
$$ \theta_f = -R_{\min}^2 \nabla^2 \theta_f + \theta_c $$
$$ \theta = \frac{\tanh(\beta(\theta_f – \theta_\beta)) + \tanh(\beta \theta_\beta)}{\tanh(\beta(1-\theta_\beta)) + \tanh(\beta \theta_\beta)} $$
where \(\beta=8\) and \(\theta_\beta=0.5\). The method of moving asymptotes is employed as the optimizer, and sensitivity information is computed via the adjoint method. The optimization problem is stated as:
$$ \begin{cases} \text{Maximize: } Di \\ \text{Subject to: } 0 < \gamma < 1 \\ \quad \quad \quad \frac{1}{V_\Omega} \int_\Omega \theta \, d\Omega \le V_f \\ \quad \quad \quad \Phi \le 0.6 \end{cases} $$
Three flow conditions were considered based on the inlet velocity: \(U_{in}=0.08\) m/s, \(0.12\) m/s, and \(0.16\) m/s, corresponding to Reynolds numbers of 40, 60, and 80 respectively. The optimized structures are denoted as case1, case2, and case3. The results show that as the Reynolds number increases, more fins and branched channels are generated, which increases the flow path tortuosity and enhances the diodicity. The optimized geometry for case1 is shown in the previous section, where the fins are distributed asymmetrically so that the reverse flow experiences a higher resistance.
Three-Dimensional Model Validation
To evaluate the diodicity in a realistic configuration, I transferred the two-dimensional topology into a three-dimensional cold plate model. The optimized pattern is arrayed on a straight channel with 4 units. The channel is 153 mm long, 3 mm wide, and 2 mm high, with inlet/outlet diameters of 3 mm. The working fluid is deionized water, and the solid is transparent acrylic for visualization. The boundary conditions are a mass flow inlet ranging from 3 to 5 g/s and a pressure outlet at 0 Pa. The fluid temperature is 298 K.
I performed a grid independence study for case1. The pressure drop converges when the mesh count reaches 6.56 million, with a deviation of less than 0.5% compared to a finer mesh. Therefore, all subsequent three-dimensional simulations use a mesh of approximately 6.56 million elements.
A dedicated experimental platform was built to validate the simulation accuracy. The test loop consists of a coolant reservoir, filter, gear pump, flowmeter, and the test cold plate. Pressure taps are installed at the inlet and outlet to measure the pressure difference in both forward and reverse flow. The cold plate models are manufactured either in aluminum or transparent acrylic. Figure 2.7 in the original text shows the machined models, but here I focus on the experimental validation.
The comparison between the experimental and simulated diodicity and reverse pressure drop is presented in Table 1. The maximum deviation of \(Di\) is 13.93% at 5 g/s, with an average deviation of 7.03%. The trends match well, confirming that the numerical model adequately predicts the flow characteristics.
| Mass flow (g/s) | \(Di\) (exp) | \(Di\) (sim) | Relative error (%) |
|---|---|---|---|
| 3.0 | 1.31 | 1.38 | 5.3 |
| 3.5 | 1.36 | 1.45 | 6.6 |
| 4.0 | 1.41 | 1.52 | 7.8 |
| 4.5 | 1.46 | 1.58 | 8.2 |
| 5.0 | 1.49 | 1.67 | 12.1 |
Figure 2.9 in the original paper shows the variation of \(Di\) for the three topologies and the straight channel. In the range of 3 to 5 g/s, the straight channel has a nearly constant \(Di\) close to 1, whereas all three optimized structures exhibit a higher \(Di\), with case1 achieving the highest value from 1.48 to 1.67. This improvement confirms that the topology-optimized fins create a preferential flow direction. The flow streamlines presented in the original text show that in the reverse direction, the fins create large recirculation zones and increase the flow resistance, while in the forward direction the main flow remains more aligned.
Based on these results, I selected case1 as the most promising topology for application in an EV battery pack cooling plate.
Numerical Model of the EV Battery Pack Cold Plate System
After establishing the enhanced diodicity of case1, I applied this topology to a full-size battery cold plate for an EV battery pack. The battery module studied in this work consists of 24 prismatic NCM lithium-ion cells, each with a nominal capacity of 105 Ah and a nominal voltage of 3.6 V. The cell dimensions are 50 mm × 114 mm × 173 mm. The cold plate is made of aluminum and has an overall size of 650 mm × 530 mm × 9.5 mm. The internal flow path contains a central main channel and multiple branched channels, where the topology-optimized structures are arrayed to enhance heat transfer and suppress bubble backflow.
Heat Generation Model of the Cell
To accurately simulate the temperature field of the EV battery pack, I used a simplified Bernardi heat generation model:
$$ q_{\text{gen}} = \frac{I^2 R_T}{V_c} $$
where \(I\) is the discharge current, \(R_T\) is the internal resistance depending on the state of charge (SOC) and temperature, and \(V_c\) is the cell volume. At 298 K ambient temperature, the internal resistance was fitted as a function of the instantaneous SOC:
$$ R_{T=\text{298K}} = 1.99 – 3.75 \, SOC_t + 7.47 \, SOC_t^2 – 5.21 \, SOC_t^3 $$
where \(SOC_t = SOC_0 – \frac{I t}{3600 C}\).
For a 1.5 C discharge rate, the current is \(I = 157.5\) A. The discharge depth is 80%, meaning \(SOC_0=1\) and the final SOC is 0.2. The total discharge time is 1920 s. This heat generation model was implemented as a user-defined function (UDF) in the Fluent simulation.
Conjugate Heat Transfer Model
The cooling plate is equipped with a 50% ethylene glycol-water mixture as the coolant. The flow and heat transfer in the channels are governed by the continuity, momentum, and energy equations:
$$ \frac{\partial \rho_f}{\partial t} + \nabla \cdot (\rho_f \mathbf{v}) = 0 $$
$$ \frac{\partial}{\partial t} (\rho_f \mathbf{v}) + \nabla \cdot (\rho_f \mathbf{v}\mathbf{v}) = -\nabla P + \nabla \cdot \mathbf{\tau} $$
$$ \frac{\partial}{\partial t} (\rho_f c_f T_f) + \nabla \cdot (\rho_f c_f \mathbf{v} T_f) = \nabla \cdot (k_f \nabla T_f) $$
Heat conduction in the solid regions (battery, thermal pad, aluminum plate) is described by:
$$ \frac{\partial}{\partial t} (\rho_s c_s T_s) = \nabla \cdot (k_s \nabla T_s) $$
The computational model includes the battery cells, thermal interface pads (1.5 mm thick), and the aluminum cold plate. The material properties used in the simulation are listed in Table 2.
| Material | Density (kg/m³) | Specific heat (J/kg·K) | Thermal conductivity (W/m·K) |
|---|---|---|---|
| Aluminum | 2719 | 871 | 202.4 |
| 50% Ethylene glycol | 1065 | 2574.7+3.06T | 0.419 |
| Thermal pad | 2094.96 | 2684 | 4 |
| Battery (NCM) | 2535 | 1014 | 1.696 (x), 29.94 (y,z) |
Boundary Conditions and Grid Independence
The inlet of the flow channel is set as a mass flow rate of 0.015 kg/s, with a coolant temperature of 298.15 K. The outlet is a pressure outlet at 0 Pa. All external surfaces exposed to the environment have a convective heat transfer coefficient of 8 W/m²·K. The battery internal heat source is applied through the UDF. A transient solver is used with a time step of 20 s and 96 total steps.
A grid independence study was performed for the full battery-module model. Four mesh sizes were compared: 9.6 million, 12.9 million, 17.7 million, and 22.1 million elements. The maximum temperature and pressure drop changed negligibly beyond 17.7 million elements, so the third mesh strategy was adopted for all subsequent simulations.
Validation of the Numerical Model
To validate the numerical model, an experimental test was conducted using a heated cold plate with a total heating power of 35 W. The heating film was attached to the top surface of the cold plate to simulate the battery heat source. The coolant flow rate was varied from 0.002 to 0.005 kg/s. The measured pressure drop and temperature were compared with the simulation results, as shown in Table 3.
| Mass flow (kg/s) | ΔP_exp (kPa) | ΔP_sim (kPa) | Error (%) | T_diff_exp (K) | T_diff_sim (K) |
|---|---|---|---|---|---|
| 0.002 | 1.05 | 1.15 | 9.5 | 4.2 | 4.5 |
| 0.003 | 1.48 | 1.55 | 4.7 | 2.9 | 3.1 |
| 0.004 | 1.92 | 1.98 | 3.1 | 2.2 | 2.4 |
| 0.005 | 2.31 | 2.37 | 2.6 | 1.8 | 1.9 |
The maximum relative error of pressure drop is about 11%, while the average temperature error is 0.79 K, demonstrating that the simulation model can accurately predict the thermal and flow behavior of the EV battery pack cold plate.
Comparison between Topology-Optimized and Straight-Channel Cold Plates
Using the validated model, I compared the performance of the topology-optimized cold plate (case1 array) with that of a conventional parallel straight-channel cold plate under identical operating conditions (1.5 C discharge, 0.015 kg/s coolant, 298.15 K inlet). The results are summarized in Table 4.
| Configuration | Tmax (K) | ΔP (Pa) | ΔT (K) |
|---|---|---|---|
| Topology-optimized array | 312.7 | 7890 | 4.4 |
| Parallel straight channels | 313.1 | 6240 | 4.7 |
The topology-optimized cold plate reduces the maximum battery temperature by 0.4 K and the temperature difference by 0.3 K, but at the expense of a 26.4% increase in pressure drop. This pressure drop penalty is mainly due to the reduced flow cross-section and the increased flow tortuosity introduced by the fins. Nevertheless, the thermal performance improvement is significant, motivating a further multi-objective optimization to balance heat transfer and pumping power.
Multi-Objective Optimization of the EV Battery Pack Cold Plate
To improve the overall performance of the topology-optimized cold plate, I applied a multi-objective optimization framework using the NSGA-II algorithm combined with Kriging surrogate models. The objective is to minimize the maximum battery temperature \(T_{\max}\), the temperature difference \(\Delta T\), and the pressure drop \(\Delta P\). The design variables are the topology structure spacing \(A\), the channel depth \(B\), and the inlet mass flow rate \(C\). Their ranges are listed in Table 5.
| Design variable | Description | Minimum | Maximum |
|---|---|---|---|
| A | Topology structure spacing (mm) | 30 | 60 |
| B | Channel depth (mm) | 1 | 2 |
| C | Inlet mass flow rate (kg/s) | 0.01 | 0.03 |
The base model has \(A=30\) mm, \(B=1\) mm, and \(C=0.015\) kg/s, yielding \(T_{\max}=312.7\) K, \(\Delta P=7890\) Pa, and \(\Delta T=4.4\) K.
I employed the Optimal Latin Hypercube Design (OLHD) to generate 31 sample points within the design space. Each sample point was evaluated via CFD simulation. The resulting responses are listed in Table 6 (only a subset is shown here for brevity).
| # | A (kg/s) | B (mm) | C (mm) | Tmax (K) | ΔP (Pa) | ΔT (K) |
|---|---|---|---|---|---|---|
| 1 | 0.01667 | 45.53 | 1.967 | 312.9 | 2647 | 4.36 |
| 2 | 0.012 | 38.3 | 1.033 | 313.4 | 5422 | 4.84 |
| 3 | 0.026 | 29 | 1.533 | 311.3 | 7056 | 3.49 |
| 4 | 0.01 | 44.5 | 1.8 | 314.0 | 1509 | 5.11 |
| 5 | 0.01333 | 50.7 | 1.133 | 313.1 | 5077 | 4.79 |
| … | … | … | … | … | … | … |
| 31 | 0.02467 | 58.97 | 1.4 | 311.5 | 7682 | 3.84 |
Before constructing the surrogate model, I performed a sensitivity analysis to evaluate the impact of each design variable. The sensitivity of a variable is calculated as:
$$ S_{A_i} = \frac{f_{\max}(x_i) – f_{\min}(x_i)}{f_{\max}(x) – f_{\min}(x)} \times 100\% $$
The results, shown in the original figure, indicate that the mass flow rate \(C\) is the most influential parameter for \(T_{\max}\) with a sensitivity of 97.7%, and for \(\Delta P\) it shares influence with channel depth \(B\) (52.7% and 47.2%). The topology spacing \(A\) has negligible influence.
Correlation analysis using the Pearson coefficient revealed a strong positive correlation between \(T_{\max}\) and \(\Delta T\) (\(r=0.99\)), and negative correlations between \(T_{\max}\) and \(\Delta P\) (\(r=-0.76\)), and between \(\Delta P\) and \(\Delta T\) (\(r=-0.68\)). Therefore, in the multi-objective optimization, I focused on \(T_{\max}\) and \(\Delta P\) as the two conflicting objectives, while monitoring \(\Delta T\) as a constraint.
I then built a Kriging surrogate model mapping the design variables to the responses. The accuracy of the surrogate was checked via cross-validation coefficients of determination \(R^2\). The \(R^2\) values were 0.961 for \(T_{\max}\) and 0.970 for \(\Delta P\), both exceeding the acceptable threshold of 0.9.
The NSGA-II algorithm was applied to the surrogate model to find the Pareto front. Two representative Pareto-optimal solutions were selected and verified by CFD simulations. These are denoted as Opt-1 and Opt-2. Their design parameters and performances are listed in Table 7.
| A (mm) | B (mm) | C (kg/s) | Tmax (K) | Error (%) | ΔP (kPa) | Error (%) | |
|---|---|---|---|---|---|---|---|
| Straight | – | – | 0.015 | 313.1 | – | 6.24 | – |
| Base | 30 | 1 | 0.015 | 312.7 | – | 7.89 | – |
| Opt-1 | 46.8 | 1.99 | 0.0219 | 312.0 (CFD 312.1) | 0.03 | 3.705 (CFD 3.79) | 2.3 |
| Opt-2 | 38.8 | 1.84 | 0.0275 | 311.3 (CFD 311.4) | 0.03 | 5.826 (CFD 5.75) | 1.3 |
Both optimized designs significantly reduce the pressure drop while maintaining a lower maximum temperature than the straight-channel configuration. In particular, Opt-1 reduces the pressure drop by 51.96% compared to the Base model and by 40% compared to the straight channel, while also lowering the maximum temperature by 0.7 K relative to the straight channel and 0.7 K relative to the Base model.
Thermo-fluid Performance Indicators
To evaluate the comprehensive performance of the optimized cold plates, I calculated the following dimensionless and energy-based indicators.
The effective heat transfer enhancement factor \(p_f\) is defined as:
$$ p_f = \frac{Nu/Nu_0}{(\Delta P/\Delta P_0)^{1/3}} $$
where \(Nu\) and \(\Delta P\) are the Nusselt number and pressure drop of the tested cold plate, and the subscript 0 denotes the reference (straight or Base) values. The Nusselt number is:
$$ Nu = \frac{h_{ave} D_h}{k_{liq}} $$
with \(h_{ave} = \frac{q_{plate}}{T_{ave,plate}-T_{ave,liq}}\).
The cooling efficiency factor is defined as:
$$ \eta = \frac{Q_{liq}/t}{\Delta P \cdot q_v} $$
where \(Q_{liq}\) is the heat absorbed by the coolant, \(t\) is the discharge time, and \(q_v\) is the volumetric flow rate.
Table 8 summarizes the computed performance metrics for the straight, Base, Opt-1, and Opt-2 configurations.
| Configuration | Nu | p_f | η (relative) | R_total (K/W) |
|---|---|---|---|---|
| Straight | Baseline | 1.0 | 1.0 | 0.0214 |
| Base | 1.18 | 1.12 | 1.05 | 0.0201 |
| Opt-1 | 2.15 | 1.61 | 1.74 | 0.0178 |
| Opt-2 | 2.10 | 1.47 | 1.55 | 0.0174 |
Opt-1 has the highest enhancement factor (161% higher than Base) and the highest relative cooling efficiency (74.8% higher than Base). Opt-2 offers the lowest total thermal resistance, but its higher flow rate leads to a lower cooling efficiency factor. Based on the balanced performance, I selected Opt-1 for experimental fabrication and testing.
Experimental Study of Single-Phase and Two-Phase Heat Transfer
To verify the numerical predictions and to examine the real thermal-hydraulic behavior of the topology-optimized cold plate under both single-phase and two-phase conditions, I fabricated the cold plate using CNC machining. The cold plate consists of an upper cover (650 mm × 530 mm × 3.5 mm) and a lower base (650 mm × 530 mm × 6 mm). The channel depth is 1.99 mm, and the topology structure spacing is 46.87 mm. An experimental loop was constructed to control the inlet temperature, flow rate, and heating power.
Single-Phase Experiments
The single-phase working fluid was a 50% ethylene glycol-water mixture. The inlet temperature was maintained at 298 K. Two sets of tests were performed: constant heat flux with variable flow rate, and constant flow rate with variable heat flux. The pressure drop across the cold plate was measured, and temperatures were recorded at multiple points on the heating film.
Figure 5.4 (in the original text) revealed that the pressure drop increases nonlinearly with the volumetric flow rate, with a steeper slope at higher flow rates. This behavior is attributed to the complex internal geometry of the topology-optimized channels, which causes additional frictional losses and local accelerations when the flow becomes turbulent.
The temperature distribution at a given flow rate is shown in the original figure, where the lowest temperature occurs near the inlet (measurement point 1) and the highest near the outlet edge (point 24). At a heat flux of 2429 W/m², increasing the flow rate from 1.0 to 1.55 L/min reduced the maximum temperature by about 1.3 °C, as summarized in Table 9.
| Flow rate (L/min) | Heat flux (W/m²) | Tmax (°C) | Tmean (°C) |
|---|---|---|---|
| 1.0 | 2429 | 48.2 | 45.1 |
| 1.55 | 2429 | 46.9 | 44.3 |
These results demonstrate that increasing the flow rate enhances convective heat transfer within the EV battery pack cold plate, effectively lowering the temperature.
Two-Phase Flow Boiling Experiments
For the two-phase experiments, I used Noah 2100A fluorinated liquid as the coolant. The inlet temperature was set to either 311 K or 313 K, corresponding to subcooling degrees of 9 K or 7 K relative to the saturation temperature at the operating pressure. The flow rate varied between 1.7 and 2.0 L/min. The heating power was gradually increased to trigger nucleate boiling. The pressure drop and wall temperatures were recorded.
The onset of nucleate boiling (ONB) was identified by a sudden change in the slope of the wall superheat versus heat flux curve. Figure 5.8 in the original text shows the boiling curves. Table 10 presents the ONB heat flux and the corresponding wall superheat for different conditions.
| Flow rate (L/min) | Subcooling ΔT_sub (K) | q_ONB (W/m²) | ΔT_wall (K) |
|---|---|---|---|
| 1.7 | 7 | 1240 | 0.3 |
| 1.7 | 9 | 1450 | 0.45 |
| 2.0 | 7 | 1360 | 0.5 |
| 2.0 | 9 | 1600 | 0.55 |
The results show that a higher subcooling degree delays ONB, requiring a higher heat flux to initiate boiling. Likewise, increasing the flow rate also delays ONB due to enhanced forced convection removing heat before the saturation point is reached.
The pressure drop behavior during boiling is illustrated in Figure 5.9. Before ONB, the pressure drop increases slowly with heat flux. Once boiling starts, the pressure drop rises sharply because vapor bubbles obstruct the flow. Table 11 lists the measured pressure drop at several heat fluxes for a fixed condition.
| q_eff (W/m²) | ΔP (kPa) |
|---|---|
| 880 | 4.12 |
| 1360 | 4.35 |
| 1600 | 4.98 |
| 2080 | 5.72 |
In addition, the time-resolved pressure signals showed increased fluctuation amplitudes after ONB, indicating flow instability caused by bubble generation and departure. The root-mean-square fluctuation increased from 0.1 kPa in single-phase to 0.48 kPa at high heat flux.
The wall temperature curves in Figure 5.11 show that after ONB, the temperature rise slows down significantly due to the latent heat absorption during boiling. The average heat transfer coefficient was calculated using:
$$ h_{ave,two-phase} = \frac{q_{eff}}{T_{heat,ave} – T_{sat}} $$
Figure 5.12 in the original text illustrates that the heat transfer coefficient decreases with increasing heat flux in the fully developed boiling region, likely due to partial dry-out or vapor blanketing on the heated surface. The effect of subcooling and flow rate on the average heat transfer coefficient is summarized in Table 12.
| Flow rate (L/min) | Subcooling ΔT_sub (K) | h_ave (W/m²·K) |
|---|---|---|
| 1.7 | 7 | 4133 |
| 1.7 | 9 | 2000 |
| 2.0 | 7 | 3125 |
| 2.0 | 9 | 1563 |
The experiments demonstrate that lower subcooling and lower flow rate promote earlier ONB and yield a higher heat transfer coefficient, although the pressure drop fluctuation may increase. The topology-optimized structure effectively suppresses severe bubble backflow, allowing stable boiling operation in the EV battery pack cold plate.
Conclusion
In this work, I developed a topology-optimized structure array for direct cooling plates used in EV battery pack thermal management. The following conclusions can be drawn:
- Topology optimization produced an asymmetric fin structure that increases the flow diodicity of a straight channel by a factor of 0.2 to 0.67, with the best configuration (case1) achieving a diodicity of 1.48–1.67 in the mass flow range of 3–5 g/s. This design suppresses bubble backflow during boiling.
- When applied to a full-size EV battery pack cold plate, the topology-optimized structure reduced the maximum battery temperature by 0.4 K and the temperature difference by 0.3 K, albeit with a 26.4% increase in pressure drop.
- Multi-objective optimization via NSGA-II and Kriging identified an optimal design (Opt-1) with a channel depth of 1.99 mm, spacing of 46.8 mm, and inlet mass flow rate of 0.0219 kg/s. This design reduced the pressure drop by 51.96% and the maximum temperature by 0.7 K compared to the baseline, while increasing the cooling efficiency factor by 74.8%.
- Single-phase experiments confirmed that increasing the flow rate enhances heat transfer, lowering the maximum temperature by 1.3 °C at a heat flux of 2429 W/m².
- Two-phase experiments revealed a clear onset of nucleate boiling, after which the pressure drop fluctuations increased and the heat transfer coefficient decreased with increasing heat flux. Lower subcooling promoted earlier ONB and higher heat transfer. The topology-optimized cold plate operated stably under the tested conditions, validating its potential for high-performance EV battery pack cooling.
In summary, the proposed topology-optimized direct cooling plate offers a high-efficiency, low-energy-consumption solution for EV battery pack thermal management, particularly suitable for two-phase cooling applications. Future work will extend the study to dynamic load profiles and extreme ambient conditions to further validate the robustness of this design.