Electric vehicles (EVs) have become a global priority in the transition toward low-carbon transportation. The power battery pack is the core energy storage component of an EV, and its performance, safety, and lifespan are strongly dependent on operating temperature. An ideal EV battery pack should operate within a narrow temperature window, typically 20–45 °C, while maintaining a low maximum temperature difference to avoid uneven degradation and thermal runaway risks. Conventional air cooling often fails under high discharge rates and elevated ambient temperatures, while pure liquid cooling consumes substantial parasitic energy. In my research, I propose a novel hybrid thermal management architecture for an EV battery pack that couples phase change material (PCM) with an active liquid cooling plate. Through numerical simulation, I investigated the heat transfer characteristics and optimized the structural parameters of this PCM-liquid coupled system using multi-objective evolutionary algorithms.

1. Modeling Framework for the EV Battery Pack and Cooling System
1.1 Single-Cell Heat Generation Model
I selected a commercially available 18650 cylindrical lithium-ion cell with a nominal capacity of 2.2 Ah and an internal resistance of 40.85 mΩ. The cell was treated as a homogeneous heat source with anisotropic thermal conductivity. The volumetric heat generation rate \(q\) was calculated using the simplified Bernardi equation:
$$ q = \frac{I^2 R}{V_b} – \frac{I T}{V_b}\frac{dU_{ocv}}{dT} $$
where \(I\) is the discharge current, \(R\) is the equivalent internal resistance, \(V_b\) is the cell volume, \(T\) is the cell temperature, and \(U_{ocv}\) is the open-circuit voltage. Since I only simulated discharge processes at moderate temperatures and did not consider extreme overcharge or deep discharge, the internal resistance was assumed constant. The computed volumetric heat generation rates at 25 °C for various discharge rates are listed in Table 1.
| Discharge rate (C) | Current (A) | Volumetric heat generation (W/m³) |
|---|---|---|
| 1 | 2.2 | 27,815 |
| 1.5 | 3.3 | 50,688 |
| 2 | 4.4 | 79,538 |
| 2.5 | 5.5 | 114,364 |
| 5 | 11.0 | 354,046 |
The energy conservation equation for the battery domain is:
$$ \rho_b C_{p,b}\frac{\partial T}{\partial t} = \nabla \cdot (\lambda_b \nabla T) + q $$
where \(\rho_b\) is the equivalent density, \(C_{p,b}\) is the equivalent specific heat, and \(\lambda_b\) is the anisotropic thermal conductivity. By applying series/parallel thermal resistance models for the layered internal structure, I obtained the equivalent thermo-physical properties listed in Table 2.
| Property | Value |
|---|---|
| Density (kg/m³) | 2722 |
| Specific heat capacity (J/(kg·K)) | 970 |
| Radial thermal conductivity (W/(m·K)) | 2.6 |
| Axial thermal conductivity (W/(m·K)) | 28 |
I validated the single-cell model by simulating free convection cooling at 25 °C. The predicted maximum cell temperatures after discharge at 1 C, 1.5 C, 2 C, and 2.5 C were 35.83 °C, 43.32 °C, 51.23 °C, and 59.30 °C, respectively, which agree well with published experimental/numerical data.
1.2 PCM Model and Composite PCM Properties
I employed a paraffin wax/expanded graphite composite PCM (CPCM) to passively absorb heat from the EV battery pack. The enthalpy-porosity method was used to capture the melting process. The energy equation for the PCM domain is:
$$ \rho_{PCM}\frac{\partial H}{\partial t} = \nabla \cdot (\lambda_{PCM} \nabla T) $$
where the total enthalpy \(H = h + \Delta H\) consists of sensible enthalpy \(h\) and latent enthalpy \(\Delta H\). The liquid fraction \(\beta\) is defined as:
$$ \beta = \begin{cases} 0 & T < T_s \\ \frac{T – T_s}{T_l – T_s} & T_s \le T \le T_l \\ 1 & T > T_l \end{cases} $$
The latent heat content is expressed as \(\Delta H = \beta \gamma\), where \(\gamma\) is the latent heat of fusion. The CPCM was composed of 12 wt% expanded graphite (EG) and 88 wt% paraffin wax (PW). Its effective thermal conductivity was calculated using the model:
$$ \lambda_{CPCM} = \alpha_{EG} \lambda_{EG} \frac{\rho_{CPCM}}{\rho_{EG}} $$
with relevant properties given in Table 3.
| Parameter | Value |
|---|---|
| Material composition | 12% EG + 88% PW |
| Thermal conductivity (W/(m·K)) | 5.74 |
| Density (kg/m³) | 897 |
| Specific heat capacity (J/(kg·K)) | 1852 |
| Latent heat (J/kg) | 224,400 |
| Phase change temperature range (°C) | 41–44 |
1.3 Liquid Cooling Model
For active liquid cooling, I used a 50% aqueous ethylene glycol solution flowing through serpentine channels in an aluminum cold plate. The flow was assumed laminar, incompressible, and steady. The Reynolds number was calculated as:
$$ \mathrm{Re} = \frac{\rho_f v D_h}{\mu} $$
where \(D_h = 4A_c/P\) is the hydraulic diameter. For all mass flow rates used (0.005–0.030 kg/s), the maximum Reynolds number remained below 2300, thus a laminar model was applied. The governing equations are the continuity, momentum, and energy equations:
$$ \nabla \cdot \mathbf{v} = 0 $$
$$ \rho_f (\mathbf{v}\cdot\nabla)\mathbf{v} = -\nabla p + \mu \nabla^2 \mathbf{v} $$
$$ \rho_f C_{p,f}(\mathbf{v}\cdot\nabla T) = \lambda_f \nabla^2 T $$
Table 4 summarizes the properties of the aluminum plate and the coolant.
| Material | Density (kg/m³) | Specific heat (J/(kg·K)) | Thermal conductivity (W/(m·K)) |
|---|---|---|---|
| Aluminum cold plate | 2707.7 | 903.2 | 238.4 |
| 50% ethylene glycol/water | 1063.66 | 3358 | 0.394 |
2. PCM-Only Cooling of the EV Battery Pack
I first investigated the stand-alone PCM cooling performance of the EV battery pack under an extreme condition: ambient temperature of 40 °C and 5C discharge (354,046 W/m³). The pack consisted of 25 cells arranged in a 5×5 grid. I studied three packing layouts – in-line, triangular staggered (equilateral), and rectangular staggered (right-triangle) – while varying the surface-to-surface spacing \(d\) from 2 mm to 10 mm. The computational domain was discretized with tetrahedral elements, and a mesh independence study confirmed that a mesh with approximately 850,650 cells was sufficient to keep temperature variations below 0.02 °C.
Figure 1 compares the maximum temperature and maximum temperature difference versus battery spacing for the three layouts. The simulation results showed that increasing the cell spacing generally decreased the maximum temperature because more PCM filled the interstitial gaps, absorbing more heat. However, beyond a certain spacing, the benefit diminished sharply due to the finite thermal conductivity of the PCM. Moreover, the maximum temperature difference did not always decrease with spacing; it first decreased and then reached a plateau or even increased due to thermal conduction limitations through the PCM layer.
For the in-line arrangement, when \(d\) increased from 2 mm to 6 mm, the maximum temperature dropped from above 75 °C to under 60 °C. Further increasing \(d\) to 10 mm reduced the maximum temperature by only 1–2 °C. At \(d = 8\) mm, the in-line layout achieved a maximum temperature of approximately 70.2 °C and a maximum temperature difference below 5 °C. In contrast, the equilateral triangular layout required at least 10 mm spacing to bring the maximum temperature difference below 5 °C, while the rectangular staggered layout behaved similarly to the in-line layout but required more PCM volume and yielded a heavier module.
I evaluated the PCM volume fraction and overall volume for each layout at identical spacing. Although the rectangular staggered layout provided acceptable thermal performance at \(d = 8\) mm, it consumed the largest PCM volume. Therefore, considering both thermal performance and lightweight design, the optimal configuration was the in-line layout with \(d = 8\) mm, which provided a good compromise between heat dissipation and module compactness. Even with this optimum, the maximum cell temperature remained above 45 °C under 5C discharge at 40 °C ambient, indicating that pure PCM cooling cannot maintain the EV battery pack within the safe threshold under high-rate and high-temperature conditions. The liquid fraction contour showed that the PCM adjacent to the cells was almost fully melted, while the outer regions remained solid, confirming the need for active cooling integration.
3. Hybrid PCM-Liquid Cooling System Design
3.1 System Architecture and Flow Direction Optimization
To improve the cooling performance of the EV battery pack, I added two serpentine liquid cooling plates above and below the PCM-filled module. A schematic of this hybrid system is shown in Figure 2. The cooling plates were made of aluminum, each containing an S-shaped channel with rounded corners to avoid stagnation zones. The coolant enters at 40 °C, equal to the ambient temperature, eliminating the need for an energy-intensive refrigeration cycle.
I first evaluated six possible configurations of the top and bottom plates with respect to their relative orientation and coolant flow direction. The configurations were: Plan 1 (same orientation, unidirectional flow), Plan 2 (same orientation, counter-flow), Plan 3 (perpendicular orientation, unidirectional), Plan 4 (perpendicular orientation, counter-flow), Plan 5 (reverse perpendicular orientation, unidirectional), and Plan 6 (reverse perpendicular orientation, counter-flow). All parameters except flow arrangement were kept identical. The simulated maximum cell temperature and temperature difference after a complete 5C discharge are shown in Table 5.
| Configuration | T_max (°C) | ΔT_max (°C) |
|---|---|---|
| Plan 1 | 45.30 | 3.41 |
| Plan 2 | 45.22 | 3.24 |
| Plan 3 | 45.34 | 3.46 |
| Plan 4 | 45.28 | 3.35 |
| Plan 5 | 45.31 | 3.38 |
| Plan 6 | 45.27 | 3.32 |
Among these, Plan 2 (same plate orientation with counter-flow) delivered the best thermal performance: the lowest maximum temperature (45.22 °C) and the lowest maximum temperature difference (3.24 °C). Therefore, I adopted this flow arrangement for all subsequent analyses.
3.2 Effect of Turbulence Promoter Structures
To further enhance the heat transfer of the liquid cooling plates, I introduced disturbance structures (also called pin fins) inside the channels. Each pin is a small columnar protrusion aligned coaxially with each battery cell. I investigated four different distributions of pins within the serpentine channel: 12121, 23232, 32323, and 55555, where the digits indicate the number of pins in each straight segment. A baseline channel without any pin was also simulated. The pin height was initially set to 1 mm.
Simulations were performed for coolant mass flow rates from 0.005 to 0.030 kg/s, and I monitored the maximum cell temperature \(T_{max}\), the maximum temperature difference \(\Delta T_{max}\), the coolant pressure drop \(\Delta P_{max}\), and the maximum PCM liquid fraction \(\beta_{max}\). The results for various pin configurations at a mass flow rate of 0.025 kg/s are summarized in Table 6.
| Configuration | T_max (°C) | ΔT_max (°C) | ΔP_max (Pa) | β_max |
|---|---|---|---|---|
| No pin | 44.98 | 3.31 | 535.26 | 0.95 |
| 12121 | 44.92 | 3.34 | 568.20 | 0.94 |
| 23232 | 44.88 | 3.36 | 622.40 | 0.93 |
| 32323 | 44.87 | 3.38 | 689.70 | 0.93 |
| 55555 | 44.85 | 3.40 | 780.50 | 0.92 |
As expected, adding more pins improved the heat transfer coefficient by promoting flow mixing and thinning the thermal boundary layer. Consequently, \(T_{max}\) decreased monotonically with increasing pin number. However, the pressure drop increased more sharply, because the pins obstruct the flow and create additional form drag. The maximum temperature difference was still below 3.5 °C for all cases, well within the 5 °C safety threshold, owing to the superior temperature uniformity provided by the PCM layer. The maximum liquid fraction of PCM decreased as the coolant removed more heat, which is advantageous because it preserves the latent heat capacity of the PCM for transient thermal shocks.
3.3 Influence of Pin Height
Fixing the pin distribution to 55555, I varied the pin height \(e\) from 1 mm to 4 mm. The coolant mass flow rate was again varied from 0.005 to 0.030 kg/s. Increasing the pin height caused a larger portion of the channel cross-section to be occupied by solid obstacles, intensifying turbulent mixing and improving convective heat transfer. Table 7 presents the results at 0.025 kg/s.
| Pin height e (mm) | T_max (°C) | ΔT_max (°C) | ΔP_max (Pa) | β_max |
|---|---|---|---|---|
| 1 | 44.85 | 3.40 | 780.50 | 0.92 |
| 2 | 44.81 | 3.42 | 1099.53 | 0.91 |
| 3 | 44.78 | 3.44 | 1436.35 | 0.91 |
| 4 | 44.70 | 3.46 | 1890.20 | 0.90 |
While the maximum temperature decreased by 0.15 °C when the pin height was increased from 1 mm to 4 mm, the pressure drop more than doubled. Thus, there is a trade-off between enhanced cooling and pumping power. The maximum temperature difference increased slightly with pin height because the local heat extraction became more non-uniform, but it never exceeded 3.5 °C.
3.4 Influence of Channel Height and Width
Maintaining the 55555 pin distribution with a pin height of 2 mm, I varied the channel height \(h\) from 5 to 8 mm and the channel width \(l\) from 13 to 16 mm. The effects on thermal performance and pressure drop were evaluated at a constant mass flow rate of 0.025 kg/s. Tables 8 and 9 summarize the simulation results.
| Channel height h (mm) | T_max (°C) | ΔT_max (°C) | ΔP_max (Pa) |
|---|---|---|---|
| 5 | 44.80 | 3.41 | 1345.9 |
| 6 | 44.81 | 3.42 | 1099.5 |
| 7 | 44.82 | 3.43 | 891.2 |
| 8 | 44.82 | 3.44 | 689.7 |
Decreasing the channel height increased the flow velocity for a fixed mass flow rate, slightly improving the heat transfer. However, the pressure drop increased disproportionately. The impact on \(T_{max}\) was only 0.02 °C, whereas \(\Delta P_{max}\) increased by nearly a factor of two when shifting from 8 mm to 5 mm channel height. Hence, an excessively small channel height is not beneficial from an energy-saving perspective.
| Channel width l (mm) | T_max (°C) | ΔT_max (°C) | ΔP_max (Pa) |
|---|---|---|---|
| 13 | 44.82 | 3.40 | 1002.4 |
| 14 | 44.81 | 3.42 | 891.2 |
| 15 | 44.82 | 3.43 | 795.5 |
| 16 | 44.82 | 3.44 | 710.8 |
Wider channels reduced the flow resistance because the cross-sectional area increased and the velocity decreased. The maximum temperature remained almost unchanged, showing that the increase in heat transfer surface area compensated for the reduced velocity. Thus, a wider channel could lower the pressure drop without sacrificing thermal performance.
3.5 Sensitivity Analysis
To quantify the relative importance of the four design parameters (pin distribution \(R\), pin height \(e\), channel height \(h\), and channel width \(l\)) on the three performance metrics (\(T_{max}\), \(\Delta T_{max}\), \(\Delta P_{max}\)), I performed a sensitivity analysis using the normalized index:
$$ SA_i = \frac{f(x_i^{max}) – f(x_i^{min})}{\sum_{j=1}^{n} \left[ f(x_j^{max}) – f(x_j^{min}) \right]} $$
The computed sensitivity coefficients at a coolant mass flow rate of 0.025 kg/s are given in Table 10.
| Parameter | Sensitivity for T_max | Sensitivity for ΔT_max | Sensitivity for ΔP_max |
|---|---|---|---|
| Pin number/layout R | 52% | 25% | 10% |
| Pin height e | 23% | 43% | 40% |
| Channel height h | 15% | 7% | 42% |
| Channel width l | 10% | 25% | 8% |
The pin layout had the greatest effect on the maximum cell temperature, while the pin height dominated the temperature uniformity and the channel height dominated the pressure drop. These findings guided my subsequent multi-objective optimization process.
4. Multi-Objective Optimization of the Liquid Cold Plate Structure
4.1 Problem Formulation and Surrogate Modeling
Because the design objectives often conflict – e.g., lowering \(T_{max}\) typically requires more pins or higher pins, which increases \(\Delta P_{max}\) – I formulated a multi-objective optimization problem. The objective functions were to minimize the maximum temperature \(T_{max}\), the maximum temperature difference \(\Delta T_{max}\), and the maximum pressure drop \(\Delta P_{max}\) of the EV battery pack cooling channel. The design variables were the pin number \(R\), pin height \(e\), channel height \(h\), and channel width \(l\). I selected a coolant mass flow rate of 0.025 kg/s to provide sufficient cooling while limiting pumping power.
To reduce the computational cost of running full CFD simulations for every candidate design, I built a surrogate model based on an orthogonal experimental design. I used the \(L_{16}(4^4)\) orthogonal array, which only required 16 CFD simulations. The design ranges are shown in Table 11.
| Variable | Level 1 | Level 2 | Level 3 | Level 4 |
|---|---|---|---|---|
| R | 7 | 12 | 13 | 25 |
| e (mm) | 1 | 2 | 3 | 4 |
| h (mm) | 5 | 6 | 7 | 8 |
| l (mm) | 13 | 14 | 15 | 16 |
I applied the method of least squares to fit quadratic polynomial response surfaces. The resulting surrogate models are:
$$ Y_1 = 44.919 – 0.00904864 R – 0.0305 e + 0.01 h + 0.0055 l $$
$$ Y_2 = 3.02238 + 0.00283262 R + 0.0135 e + 0.0035 h + 0.0235 l $$
$$ Y_3 = 2631.32 – 16.6704 R + 136.222 e – 254.027 h – 62.247 l $$
where \(Y_1\) corresponds to \(T_{max}\) in °C, \(Y_2\) to \(\Delta T_{max}\) in °C, and \(Y_3\) to \(\Delta P_{max}\) in Pa. The coefficient of determination \(r^2\) for each surrogate was computed as:
$$ r^2 = 1 – \frac{\sum_{i=1}^{n}(Y_{i,P} – Y_{i,T})^2}{\sum_{i=1}^{n} Y_{i,T}^2} $$
The \(r^2\) values for the training set were 0.999, 0.999, and 0.966 for \(T_{max}\), \(\Delta T_{max}\), and \(\Delta P_{max}\), respectively; for the test set they were 0.999, 0.999, and 0.989. These values indicate excellent predictive accuracy. I then formulated the multi-objective optimization problem as:
$$ \min \; F(R,e,h,l) = \left[ f_1(R,e,h,l), f_2(R,e,h,l), f_3(R,e,h,l) \right]^T $$
subject to:
$$ 7 \le R \le 25, \quad R \in \mathbb{Z} $$
$$ 1 \le e \le 4, \quad 5 \le h \le 8, \quad 13 \le l \le 16 $$
$$ f_1 \le 45\,^\circ C, \quad f_2 \le 5\,^\circ C $$
4.2 NSGA-II Optimization Results
I employed the improved NSGA-II algorithm in Python to solve the multi-objective problem. The algorithm parameters are listed in Table 12.
| Parameter | Value |
|---|---|
| Population size | 125 |
| Maximum generations | 500 |
| Pareto fraction | 0.32 |
| Adaptation tolerance | 1×10⁻⁴ |
The optimization produced 40 Pareto-optimal solutions. Analysis of the Pareto front revealed that the pin layout \(R = 25\) (i.e., the 55555 distribution) appeared in 75% of the optimal solutions, while the 23232 distribution appeared in only 5%. Regarding pin height, \(e = 1\) mm was preferred in 37.5% of the solutions, and \(e = 3\) mm appeared only 7.5%. For channel height, \(h = 8\) mm appeared in 57.5% of the solutions, and \(h = 7\) mm only 2.5%. For channel width, \(l = 13\) mm appeared in 67.5% of the solutions, while \(l = 14\) mm never appeared. Therefore, I selected the design with \(R = 25\), \(e = 1\) mm, \(h = 8\) mm, \(l = 13\) mm as the final optimal configuration. The surrogate models predicted \(T_{max} = 44.81\) °C, \(\Delta T_{max} = 3.44\) °C, and \(\Delta P_{max} = 342.87\) Pa. A verification CFD simulation for this design produced \(T_{max} = 44.80\) °C, \(\Delta T_{max} = 3.40\) °C, and \(\Delta P_{max} = 344.03\) Pa, corresponding to relative errors of 0.02%, 1.18%, and 0.34%, respectively. This excellent agreement confirmed the reliability of the surrogate-based optimization.
5. Performance Comparison Between Optimized and Conventional Cold Plate
I compared the optimized cold plate to the original serpentine channel without pins (the conventional baseline). Both were evaluated under identical conditions: 40 °C ambient, 5C discharge, coolant inlet at 40 °C, and a mass flow rate of 0.025 kg/s. Table 13 summarizes the results.
| Metric | Conventional cold plate | Optimized cold plate | Improvement |
|---|---|---|---|
| T_max (°C) | 44.98 | 44.80 | -0.40% |
| ΔT_max (°C) | 3.31 | 3.40 | +2.72% (still safe) |
| ΔP_max (Pa) | 535.26 | 344.03 | -35.72% |
The optimized cold plate reduced the maximum cell temperature by 0.4% while reducing the pressure drop by an impressive 35.72%. Although the maximum temperature difference increased slightly from 3.31 °C to 3.40 °C, it remained far below the 5 °C safety threshold. This improvement stems from the fact that the selected design uses a relatively large number of low-height pins and a tall, narrow channel, which enhances heat transfer without excessively increasing flow resistance. The presence of PCM in the EV battery pack also maintains excellent temperature uniformity, so the slight increase in \(\Delta T_{max}\) is acceptable in exchange for substantial energy savings in the liquid cooling system.
6. Conclusions
In this study, I developed a hybrid phase change material-liquid coupled thermal management system for an EV battery pack composed of 18650 cells. The main findings are:
- Pure PCM cooling cannot keep the EV battery pack below 45 °C under a 5C discharge at 40 °C ambient temperature, even with an optimal in-line layout and a cell spacing of 8 mm. However, the PCM maintains the maximum temperature difference well below 5 °C.
- Adding dual serpentine liquid cooling plates with counter-flow arrangement on the top and bottom of the PCM module successfully lowers the maximum cell temperature to the safe range while preserving the thermal uniformity advantage of PCM.
- The introduction of pin-type disturbance structures significantly enhances the heat transfer of the liquid cooling plate, but also increases the pressure drop. The pin layout has the largest influence on \(T_{max}\), the pin height dominates \(\Delta T_{max}\), and the channel height has the strongest effect on \(\Delta P_{max}\).
- A surrogate-based multi-objective optimization using NSGA-II produced a Pareto-optimal design with a 55555 pin layout, 1 mm pin height, 8 mm channel height, and 13 mm channel width. This design reduces the maximum temperature by 0.4% and lowers the pressure drop by 35.72% compared to the conventional channel, making the thermal management system more efficient and energy-saving for EV battery packs.
The proposed methodology combining CFD, orthogonal experiments, response surface modeling, and evolutionary optimization provides a robust framework for designing high-performance, low-energy thermal management solutions for EV battery packs under extreme operating conditions.
