In the context of global carbon reduction strategies, the energy storage industry has experienced remarkable growth, with the production scale of energy storage batteries reaching unprecedented levels. As a researcher deeply engaged in the thermal management of battery systems, I have focused my recent work on addressing the critical thermal challenges faced by large-scale battery energy storage systems. The battery energy storage system, particularly those utilizing lithium-ion technology, requires precise temperature control to ensure operational safety, performance stability, and extended lifespan. The normal operating temperature range for such systems is typically between 20 °C and 40 °C, with the temperature difference between individual cells needing to be maintained below 5 °C. When the temperature exceeds 70 °C, there is a significant risk of thermal runaway, which can lead to fire or explosion incidents. Therefore, developing an efficient, energy-saving, and safe thermal management system is paramount for enhancing the thermal safety and stability of battery energy storage systems.
In this study, I selected a multi-channel serpentine liquid cooling plate as the cooling solution for a 43 kWh battery energy storage system. This design choice was driven by the need for a balance between cooling performance, manufacturing complexity, and system integration feasibility. Unlike biomimetic or topology-optimized flow channels that offer marginal performance improvements at significantly higher manufacturing costs, the serpentine design provides a practical and effective solution for real-world applications. The liquid cooling plate operates on the principle of convective heat transfer, where a coolant fluid circulates through internal channels to absorb and dissipate heat generated by the battery cells. The pumping power required to drive the coolant through the system is a critical factor in overall system efficiency, making flow resistance optimization a key objective in my research.
My investigation began with a comprehensive analysis of six different flow channel configurations, with channel counts ranging from 3 to 8. I systematically evaluated the cooling performance and flow resistance characteristics of each configuration using computational fluid dynamics simulations. The results revealed several important insights that guided subsequent optimization efforts. Notably, I found that the number of flow channels had minimal impact on heat transfer performance across all configurations. All six designs successfully maintained the maximum battery pack temperature below 39 °C, with an average temperature of approximately 34.3 °C and a cell temperature difference of around 2.1 °C. These values are well within the safe operating range for lithium-ion batteries, demonstrating the inherent effectiveness of the serpentine liquid cooling approach. However, the flow resistance exhibited a clear downward trend as the number of channels increased, indicating potential for energy savings through channel count optimization.
Based on a comprehensive evaluation of both cooling performance and manufacturing complexity, I selected the five-channel liquid cooling plate as the optimal baseline design for further parameter optimization. The five-channel configuration offers a favorable balance between thermal performance, flow resistance, and practical manufacturability. Using a systematic optimization approach combining Latin hypercube sampling, response surface methodology, and the NSGA-II multi-objective genetic algorithm, I successfully reduced the flow resistance from 1,101 Pa to 640.6 Pa—a reduction of 41.8%—while maintaining the cooling performance within acceptable limits. This significant reduction in flow resistance translates directly to lower pumping power requirements and improved overall system efficiency for the battery energy storage system.
Mathematical Modeling of the Liquid Cooling System
To accurately predict the thermal and fluid dynamic behavior of the liquid cooling system, I developed a comprehensive mathematical model based on fundamental principles of fluid mechanics and heat transfer. The model accounts for both pressure losses and heat transfer mechanisms within the cooling channels, enabling precise performance prediction and optimization.
Flow Resistance Model
The flow resistance within the liquid cooling plate arises from two primary sources: frictional losses along the channel walls and local losses due to flow disturbances such as bends, expansions, and contractions. The total pressure drop is the sum of these two contributions. The frictional pressure loss, also known as the major loss, is calculated using the Darcy-Weisbach equation:
$$
\Delta P_{\lambda} = \lambda \frac{l}{d} \cdot \frac{\rho v^2}{2}
$$
where $\lambda$ is the friction factor, which depends on the flow regime characterized by the Reynolds number; $l$ is the channel length; $d$ is the hydraulic diameter; $\rho$ is the coolant density; and $v$ is the average flow velocity. The local pressure losses, caused by flow separation and vortex formation at geometric discontinuities, are calculated using:
$$
\Delta P_r = \zeta \frac{\rho v^2}{2}
$$
where $\zeta$ is the local loss coefficient, which is a function of the specific geometry of the flow disturbance. In the context of the serpentine liquid cooling plate, significant local losses occur at the inlet and outlet regions where the flow undergoes rapid expansion or contraction, as well as at the U-turn sections where the flow direction changes abruptly.
To characterize the flow regime within the cooling channels, I calculated the Reynolds number at the operating conditions. With a coolant inlet flow rate of 0.05 kg/s and channel dimensions of 20 mm width and 6 mm height, the Reynolds number at the inlet was found to be 5,273.5, indicating turbulent flow conditions. The turbulent flow regime enhances convective heat transfer but also increases frictional losses. The friction factor for turbulent flow was determined using the Konakov correlation:
$$
f = (1.8 \log Re – 1.5)^{-2}
$$
This correlation is valid for the Reynolds number range encountered in my study and provides accurate predictions for smooth channels.
Heat Transfer Model
The heat transfer process in the liquid cooling system involves two distinct mechanisms: conduction through the solid components and convection between the coolant and the channel walls. The heat generated by the battery cells is first conducted through the cell materials and the thermal interface material to the liquid cooling plate, and then transferred to the coolant through forced convection.
The conductive heat transfer follows Fourier’s law:
$$
\Phi = -\lambda_s A \frac{dT}{dx}
$$
where $\lambda_s$ is the thermal conductivity of the solid material, $A$ is the cross-sectional area perpendicular to heat flow, $T$ is the temperature, and $x$ is the distance along the heat conduction path. In the battery cell, the thermal conductivity is anisotropic due to the layered structure of electrode and separator materials. The in-plane thermal conductivity (y and z directions) is significantly higher than the through-plane conductivity (x direction), which affects the temperature distribution within the cell.
The convective heat transfer between the coolant and the channel walls is described by:
$$
\Phi = h A (T_w – T_f)
$$
where $T_w$ is the wall temperature, $T_f$ is the bulk fluid temperature, and $h$ is the convective heat transfer coefficient. The heat transfer coefficient is determined using the Nusselt number correlation:
$$
h = \frac{\lambda_l Nu}{d}
$$
where $\lambda_l$ is the thermal conductivity of the coolant. For turbulent flow in circular and non-circular channels, the Gnielinski correlation provides the most accurate prediction of the Nusselt number across a wide range of operating conditions:
$$
Nu = \frac{(f/8)(Re – 1000)Pr}{1 + 12.7(f/8)^{1/2}(Pr^{2/3} – 1)} \left[ 1 + \left( \frac{d}{l} \right)^{2/3} \right]
$$
where $Pr$ is the Prandtl number of the coolant, which represents the ratio of momentum diffusivity to thermal diffusivity. The Gnielinski correlation is valid for Reynolds numbers ranging from 2,300 to 10^6 and Prandtl numbers from 0.6 to 10^5, which covers the operating range of my liquid cooling system.
Geometric Modeling and Simulation Setup
Battery Pack Configuration
The battery energy storage system under investigation is a 43 kWh system consisting of 48 lithium iron phosphate cells arranged in two modules. Each cell has a nominal capacity of 280 Ah, an internal resistance of 0.22 mΩ, and dimensions of 72 mm × 208 mm × 173 mm in the x, y, and z directions, respectively. The cells are bonded together using thermally conductive adhesive and tightly secured with metal straps, with negligible spacing between adjacent cells. The key parameters of the cell are summarized in the following table:
| Parameter | Value |
|---|---|
| Nominal capacity (Ah) | 280 |
| Internal resistance (mΩ) | 0.22 |
| Dimensions (x, y, z) (mm) | 72, 208, 173 |
| Density (kg/m³) | 2,219 |
| Specific heat capacity (J/kg·K) | 1,000 |
| Thermal conductivity (x, y, z) (W/m·K) | 2.5, 12, 12 |
| Maximum charge/discharge rate | 1C |
The anisotropic thermal conductivity of the cell is a critical factor in the thermal management design. The through-plane thermal conductivity (x direction) is 2.5 W/m·K, while the in-plane thermal conductivity (y and z directions) is 12 W/m·K. This anisotropy results from the layered structure of the cell, where heat conduction along the electrode and separator materials is more efficient than conduction across the interface between layers. To simplify the simulation while maintaining accuracy, I modeled only the essential components—the cells and the liquid cooling plate—and omitted the mechanical supports, control modules, and other auxiliary components that have minimal impact on the thermal behavior.
Liquid Cooling Plate Design
I designed six different liquid cooling plate configurations with channel counts ranging from 3 to 8. Each configuration features a serpentine flow pattern with parallel channels connected at both ends by manifold sections that distribute and collect the coolant. The inlet and outlet are located on the same side of the plate, with a diameter of 12 mm. Each channel has a width of 20 mm and a height of 6 mm. The manifold sections are designed to ensure uniform flow distribution among the parallel channels while minimizing the overall footprint of the cooling plate.
The serpentine flow pattern provides extended coolant residence time and enhanced heat transfer compared to straight-through designs. However, it also introduces additional flow resistance due to the multiple U-turns. The balance between heat transfer enhancement and pressure drop is a key consideration in the design optimization. The following table summarizes the geometric parameters of the six liquid cooling plate configurations:
| Configuration | Number of Channels | Channel Width (mm) | Channel Height (mm) | Inlet/Outlet Diameter (mm) |
|---|---|---|---|---|
| 1 | 3 | 20 | 6 | 12 |
| 2 | 4 | 20 | 6 | 12 |
| 3 | 5 | 20 | 6 | 12 |
| 4 | 6 | 20 | 6 | 12 |
| 5 | 7 | 20 | 6 | 12 |
| 6 | 8 | 20 | 6 | 12 |
Simulation Methodology and Grid Independence Study
I performed all simulations using ANSYS Fluent 2022 R1, a computational fluid dynamics software package widely used for thermal and fluid flow analysis. The boundary conditions were carefully selected to represent realistic operating conditions of the battery energy storage system. The cell heat generation rate was set to 17 W per cell, corresponding to 1C discharge rate. The initial temperature of the entire system was set to 25 °C, and the coolant inlet temperature was maintained at 25 °C throughout the simulation. The outlet pressure was set to 0 Pa (gauge pressure), and a natural convection heat transfer coefficient of 5 W/m²·K was applied to the external surfaces of the battery pack. The turbulent flow was modeled using the k-ε turbulence model, which provides a good balance between accuracy and computational efficiency for internal flow problems.
For the numerical solution, I used the COUPLED algorithm to solve the mass, momentum, and energy equations simultaneously. All spatial discretizations were performed using second-order upwind schemes to minimize numerical diffusion and improve solution accuracy. The convergence criterion was set based on the residuals of the continuity, momentum, and energy equations, with a maximum of 600 iterations to ensure temperature convergence.
To ensure the accuracy of the simulation results, I conducted a grid independence study using five different mesh sizes. The Poly-Hexcore meshing method was employed, which automatically adjusts the mesh density based on the local geometry, allowing for efficient mesh generation with appropriate resolution in critical regions. The mesh sizes tested were 4,259,451; 13,677,238; 16,302,827; 17,970,457; and 19,557,812 elements. The results showed that the maximum battery pack temperature became stable when the mesh size exceeded 17,970,457 elements. Based on this analysis, I selected the mesh with 17,970,457 elements for all subsequent simulations, with a minimum element size of 1 mm and a maximum element size of 16 mm.
| Mesh Size (elements) | Maximum Temperature (°C) | Temperature Change (°C) |
|---|---|---|
| 4,259,451 | 40.2 | — |
| 13,677,238 | 39.8 | -0.4 |
| 16,302,827 | 39.6 | -0.2 |
| 17,970,457 | 38.5 | -1.1 |
| 19,557,812 | 38.4 | -0.1 |
Results and Discussion
Effect of Channel Number on Cooling Performance
I simulated the thermal behavior of the battery energy storage system with all six liquid cooling plate configurations under identical operating conditions. The results revealed that the number of flow channels has a minimal effect on the overall cooling performance, with all configurations maintaining the battery pack temperature well within the safe operating range. The maximum battery pack temperature ranged from 38.3 °C to 38.6 °C across the six configurations, while the average temperature was consistently around 34.2 °C. The temperature difference between individual cells was approximately 2.1 °C for all configurations, indicating uniform cooling performance irrespective of the channel count.
This insensitivity of cooling performance to channel number can be attributed to the complex interplay between two competing factors. On one hand, increasing the number of channels increases the total heat transfer surface area, which enhances convective heat transfer. On the other hand, the coolant velocity in each channel decreases as the total cross-sectional area increases, leading to lower Reynolds numbers and reduced convective heat transfer coefficients. These two effects partially offset each other, resulting in similar overall heat transfer performance across the range of channel counts studied. The following table presents the detailed simulation results for all six configurations:
| Number of Channels | Maximum Temperature (°C) | Average Temperature (°C) | Temperature Difference (°C) | Flow Resistance (Pa) |
|---|---|---|---|---|
| 3 | 38.6 | 34.3 | 2.1 | 1,454 |
| 4 | 38.5 | 34.3 | 2.1 | 1,310 |
| 5 | 38.4 | 34.2 | 2.1 | 1,101 |
| 6 | 38.3 | 34.1 | 2.1 | 1,045 |
| 7 | 38.3 | 34.1 | 2.1 | 1,012 |
| 8 | 38.3 | 34.1 | 2.1 | 1,000 |
Effect of Channel Number on Flow Resistance
In contrast to the cooling performance, the flow resistance showed a significant dependence on the number of channels. As the channel count increased from 3 to 8, the flow resistance decreased from 1,454 Pa to 1,000 Pa, representing a reduction of 31.2%. This trend can be explained by the fundamental relationship between flow velocity and pressure drop. According to the Darcy-Weisbach equation, the pressure drop is proportional to the square of the flow velocity. When the number of channels increases, the total cross-sectional area for flow increases, leading to a decrease in the average velocity in each channel. This velocity reduction directly translates to lower frictional pressure losses.
However, it is important to note that the relationship between channel count and flow resistance is not linear. The marginal benefit of adding additional channels diminishes as the channel count increases. The difference in flow resistance between the 3-channel and 4-channel configurations is 144 Pa, while the difference between the 7-channel and 8-channel configurations is only 12 Pa. This diminishing return is due to the increasing influence of local losses at the manifold sections and U-turns, which become more significant as the channel count increases and the flow distribution becomes more complex.
Furthermore, I observed that increasing the channel count beyond a certain point leads to practical manufacturing challenges. The channels become narrower, making the fabrication of the cooling plate more difficult and expensive. The manifold sections also become more complex, requiring precise flow distribution design to avoid maldistribution. Based on a comprehensive evaluation of cooling performance, flow resistance, and manufacturing feasibility, I selected the five-channel configuration as the optimal baseline design for further optimization.
Coolant Flow Rate Effect on Cooling Performance
To establish the optimal operating conditions for the five-channel liquid cooling plate, I investigated the effect of coolant flow rate on the cooling performance. The flow rate was varied from 0.03 kg/s to 0.10 kg/s, and the resulting maximum and average temperatures were recorded. The results showed a clear trend of decreasing temperature with increasing flow rate, confirming the expected relationship between coolant flow and heat transfer performance. At a flow rate of 0.03 kg/s, the maximum temperature approached 40 °C, which is at the upper limit of the safe operating range. Increasing the flow rate to 0.05 kg/s reduced the maximum temperature to 38.4 °C, providing a comfortable safety margin. Further increases in flow rate resulted in diminishing temperature reductions, while significantly increasing the pumping power requirements. Based on this analysis, I selected 0.05 kg/s as the optimal flow rate for the battery energy storage system.
| Flow Rate (kg/s) | Maximum Temperature (°C) | Average Temperature (°C) |
|---|---|---|
| 0.03 | 39.8 | 34.9 |
| 0.05 | 38.4 | 34.2 |
| 0.07 | 37.8 | 33.5 |
| 0.10 | 37.2 | 32.8 |
Multi-Objective Optimization Using NSGA-II Algorithm
Having established the five-channel configuration as the optimal baseline design, I proceeded to optimize the manifold geometry to further reduce flow resistance while maintaining cooling performance. The optimization problem was formulated as a multi-objective problem with the coolant outlet temperature as a constraint and the flow resistance as the objective function to be minimized.
Design Variables and Response Surface Methodology
Based on the flow velocity distribution analysis of the five-channel configuration, I identified the manifold gaps at both ends of the channels as the key geometric parameters influencing flow resistance. These gaps, where the coolant undergoes expansion or contraction, are the primary sources of local pressure losses due to vortex formation and flow separation. I defined six design variables (X₁ through X₆) representing the four inter-channel gaps and two serpentine return gaps. The dimensions of these gaps were varied within a range of 20 mm to 60 mm, based on the observed vortex size from the initial simulation results.
To efficiently explore the design space, I employed Latin hypercube sampling, which provides excellent space-filling properties and is well-suited for building accurate surrogate models. A total of 43 sample points were generated, and each configuration was evaluated using CFD simulations. The results were used to construct a response surface model, which provides a mathematical relationship between the design variables and the performance metrics. The response surface model was built using a second-order polynomial regression, with the accuracy assessed using the coefficient of determination R²:
$$
R^2 = 1 – \frac{\sum_{i=1}^{n}(y_i – \hat{y}_i)^2}{\sum_{i=1}^{n}(y_i – \bar{y})^2}
$$
The response surface model for the coolant outlet temperature (Y₁) achieved an R² value of 0.91064, while the model for the flow resistance (Y₂) achieved an R² value of 0.98779. Both values exceed the threshold of 0.9, indicating that the surrogate models provide accurate predictions and can be used confidently for optimization. The high R² value for flow resistance indicates that the response surface model captures the relationship between the geometric variables and the pressure drop with excellent fidelity.
NSGA-II Optimization Algorithm
The NSGA-II (Non-dominated Sorting Genetic Algorithm II) is a widely used multi-objective optimization algorithm known for its efficiency and ability to find diverse Pareto-optimal solutions. The algorithm operates by maintaining a population of candidate solutions and iteratively applying genetic operators such as selection, crossover, and mutation to generate improved solutions. The key features of NSGA-II include non-dominated sorting, which classifies solutions based on their dominance relationships, and crowding distance calculation, which maintains diversity among solutions on the Pareto front.
The crowding distance is calculated using the following formula:
$$
d_i^{\text{distance}} = \sum_{k=1}^{m} \frac{z_k^{(k+1)} – z_k^{(k-1)}}{z_k^{\max} – z_k^{\min}} \quad (2 \le i \le n-1)
$$
where $m$ is the number of objectives, $z_k$ is the $k$-th objective value, and $z_k^{\max}$ and $z_k^{\min}$ are the maximum and minimum values of the $k$-th objective in the population. The crowding distance measures the density of solutions surrounding a particular solution, with higher values indicating more isolated solutions that contribute to diversity.
In my optimization problem, I set the coolant outlet temperature (Y₁) as a constraint with a maximum allowable value of 29.5 °C, and the flow resistance (Y₂) as the objective to be minimized. The NSGA-II algorithm was configured with a population size of 100 and a maximum of 200 generations, providing a thorough exploration of the design space.
Optimization Results
The NSGA-II optimization yielded a set of Pareto-optimal solutions representing the trade-off between cooling performance and flow resistance. From these solutions, I selected the configuration that minimized flow resistance while satisfying the temperature constraint. The optimal design variables were found to be: X₁ = 54 mm, X₂ = 44 mm, X₃ = 52 mm, X₄ = 52 mm, X₅ = 56 mm, and X₆ = 60 mm. This configuration corresponds to a predicted flow resistance of 590.9 Pa.
To validate the optimization results, I performed a CFD simulation of the optimized configuration and compared the predicted performance with the response surface model predictions. The CFD simulation yielded a flow resistance of 640.6 Pa, which is within 7.8% of the predicted value, confirming the accuracy of the surrogate model. The slight discrepancy can be attributed to the inherent approximation error of the response surface model and the complexity of the turbulent flow phenomena in the manifold regions.
| Parameter | Baseline (5-channel) | Optimized | Change (%) |
|---|---|---|---|
| X₁ (mm) | — | 54 | — |
| X₂ (mm) | — | 44 | — |
| X₃ (mm) | — | 52 | — |
| X₄ (mm) | — | 52 | — |
| X₅ (mm) | — | 56 | — |
| X₆ (mm) | — | 60 | — |
| Flow resistance (Pa) | 1,101 | 640.6 | -41.8 |
| Outlet temperature (°C) | 29.5 | 29.5 | 0 |
| Maximum battery temperature (°C) | 38.4 | 38.4 | 0 |
| Cell temperature difference (°C) | 2.1 | 2.1 | 0 |
Performance Validation of the Optimized Design
To fully validate the optimized liquid cooling plate design, I integrated it into the complete battery energy storage system model and performed a comprehensive thermal simulation. The results confirmed that the optimized design maintains the same high level of cooling performance as the baseline configuration, with the maximum battery pack temperature remaining at 38.4 °C and the cell temperature difference at 2.1 °C. The temperature distribution within the battery pack follows a similar pattern to the baseline design, with higher temperatures observed in the upper region of the cells and the coolant outlet side.
The velocity distribution analysis of the optimized design revealed a significant reduction in vortex formation at the manifold sections. The manifold gaps in the optimized design are larger than those in the baseline configuration, allowing for smoother flow transition between the inlet pipe and the parallel channels. This reduction in flow disturbance is the primary mechanism responsible for the 41.8% reduction in flow resistance. Additionally, the optimized design achieves more uniform flow distribution among the parallel channels, further contributing to the reduction in local pressure losses.
The energy savings resulting from the reduced flow resistance are substantial. For a typical battery energy storage system operating continuously, the pumping power is directly proportional to the product of the flow rate and the pressure drop. A 41.8% reduction in pressure drop translates to an equivalent reduction in pumping power, assuming the flow rate remains constant. Over the lifetime of the system, this energy saving can significantly reduce the operating cost and improve the overall efficiency of the energy storage system.
Conclusions
In this comprehensive study, I have systematically investigated the thermal performance and flow resistance characteristics of a multi-channel serpentine liquid cooling plate for a 43 kWh battery energy storage system. Through a combination of CFD simulations, response surface modeling, and multi-objective optimization, I have developed an optimized cooling plate design that achieves a 41.8% reduction in flow resistance while maintaining the cooling performance within the required specifications.
The key findings of my research can be summarized as follows:
First, the number of flow channels in the serpentine liquid cooling plate has a minimal effect on the overall cooling performance of the battery energy storage system. All configurations with 3 to 8 channels maintained the maximum battery pack temperature below 39 °C and the average temperature around 34.3 °C, with a cell temperature difference of approximately 2.1 °C. This insensitivity to channel count arises from the competing effects of increased heat transfer area and decreased convective heat transfer coefficient.
Second, the flow resistance decreases significantly as the number of channels increases, with a 31.2% reduction observed when increasing from 3 to 8 channels. However, the marginal benefit diminishes at higher channel counts, and practical manufacturing considerations limit the maximum feasible channel count. Based on a balanced evaluation of performance and manufacturability, the five-channel configuration was selected as the optimal baseline design.
Third, the manifold gap geometry has a substantial impact on the flow resistance of the liquid cooling plate. The optimized design, with larger manifold gaps arranged in a crescent-shaped pattern, reduces the vortex formation and flow separation that are the primary sources of local pressure losses. The NSGA-II algorithm successfully identified the optimal combination of design variables, achieving a flow resistance reduction from 1,101 Pa to 640.6 Pa.
Fourth, the optimized liquid cooling plate maintains the same cooling performance as the baseline design, with the maximum battery pack temperature at 38.4 °C and the cell temperature difference at 2.1 °C. This confirms that the optimization successfully reduced the pumping power requirement without compromising thermal management capability.
The methodology developed in this study provides a systematic approach for designing and optimizing liquid cooling plates for battery energy storage systems. The combination of CFD simulation, response surface modeling, and multi-objective optimization offers a powerful framework for exploring the design space and identifying optimal solutions that balance competing performance objectives. The specific results obtained for the 43 kWh system demonstrate the potential for significant energy savings through careful geometric optimization, contributing to the development of more efficient and sustainable energy storage technologies.
Future work should focus on experimental validation of the optimized design, as well as investigation of transient operating conditions and aging effects on the thermal management performance. The integration of advanced cooling strategies, such as variable flow rate control and phase change materials, could further enhance the efficiency and safety of battery energy storage systems. Additionally, the application of machine learning techniques for real-time optimization and predictive maintenance of the thermal management system represents a promising direction for future research.
In conclusion, this study has successfully demonstrated that significant improvements in the efficiency of battery energy storage system thermal management can be achieved through systematic geometric optimization of the liquid cooling plate. The 41.8% reduction in flow resistance not only reduces the energy consumption of the cooling system but also contributes to the overall reliability and sustainability of the energy storage system. As the global demand for energy storage continues to grow, such efficiency improvements will play an increasingly important role in making energy storage systems more economical and environmentally friendly.