Global decarbonization policies have accelerated the replacement of conventional internal combustion vehicles with electric vehicles. In my research, I focused on one of the most critical components of modern electric vehicles: the EV battery pack. The EV battery is expected to provide high energy density, long cycle life, and safe operation under highly variable loads. However, lithium-ion cells release heat during charging and discharging, and if this heat is not dissipated effectively, the temperature of the EV battery pack can rise beyond the recommended range. High temperatures accelerate degradation, reduce available power, and may create thermal runaway risks. Therefore, a robust thermal management system is essential for every EV battery.
I reviewed the main cooling technologies currently used for EV battery thermal management: air cooling, liquid cooling, phase-change material cooling, hybrid systems, and vapor-chamber cooling. Air cooling is simple and inexpensive but suffers from low heat-transfer coefficients and poor temperature uniformity. Liquid cooling provides higher heat-removal capability, but it increases system complexity, weight, cost, and leakage risk. Phase-change materials can absorb heat passively and suppress temperature spikes, but their low thermal conductivity and limited latent heat after melting require additional heat-recovery mechanisms. Hybrid systems combine two or more cooling strategies to overcome these limitations, yet they create integration challenges. In this context, vapor chambers provide an attractive solution because they use a two-phase evaporation–condensation cycle to spread heat with very small temperature differences. A vapor chamber can be inserted between the EV battery cells and an air-cooled heat sink, creating a compact, passive, and low-energy thermal management architecture.
My aim was to design, simulate, optimize, and experimentally verify a complete thermal management solution for a high-energy-density EV battery module. The solution is based on a newly developed vapor chamber with hollow fins and a Kelvin-lattice wick, combined with forced-air cooling through a topologically optimized four-channel duct. I also constructed a digital twin based on reduced-order models to enable real-time prediction and control of the EV battery thermal state. The work combines structural design, computational fluid dynamics, multi-objective optimization, laboratory experiments, and digital-twin implementation.
| Cooling method | Advantages | Disadvantages | Typical application |
|---|---|---|---|
| Air cooling | Simple, low cost, easy maintenance | Low heat transfer, large temperature gradient | Low-power EV battery packs |
| Liquid cooling | High heat transfer, accurate temperature control | Complex, leak risk, parasitic pump power | High-power EV battery packs |
| Phase-change material cooling | High latent heat, passive response | Low conductivity, heat accumulation after melting | Moderate-power EV battery packs |
| Hybrid cooling | Combines strengths of multiple methods | High integration complexity and cost | High-performance EV battery packs |
| Vapor chamber cooling | Very high effective conductivity, excellent temperature uniformity | Design of condenser and wick is challenging | High-power-density EV battery packs |
The vapor chamber used in my investigation relies on evaporation and condensation of a working fluid. The evaporator section is attached to the hot surface, usually the bottom or side of the EV battery cell. Liquid working fluid absorbs heat and evaporates. The resulting vapor travels through a central vapor core to the condenser section, where it releases latent heat and returns to liquid. Capillary forces in the porous wick, together with gravity in the vertical configuration, drive the liquid back to the evaporator. This continuous cycle makes the vapor chamber an extremely efficient heat spreader.
To predict the thermal behavior, I coupled several transport mechanisms: flow in the vapor core, flow in the porous wick, phase change at the liquid–vapor interface, and heat conduction in the metal casing. The conservation of mass for each phase can be expressed as
$$
\frac{\partial}{\partial t}\left(\alpha_n \rho_n\right)
+ \nabla\cdot\left(\alpha_n \rho_n \mathbf{v}_n\right)=S_M,
$$
where \(\alpha_n\) is the volume fraction, \(\rho_n\) is the density, \(\mathbf{v}_n\) is the velocity vector of phase \(n\), and \(S_M\) is the mass source caused by evaporation or condensation. The momentum balance follows the Navier–Stokes form
$$
\frac{\partial}{\partial t}\left(\rho \mathbf{v}\right)
+\nabla\cdot\left(\rho \mathbf{v}\mathbf{v}\right)
=-\nabla p + \nabla\cdot\left[\mu\left(\nabla\mathbf{v}+\nabla\mathbf{v}^T\right)\right]
+\rho\mathbf{g}+\mathbf{F}.
$$
For the energy balance, I used
$$
\frac{\partial}{\partial t}\left(\rho E\right)
+\nabla\cdot\left(\mathbf{v}\left(\rho E + p\right)\right)
=\nabla\cdot\left(k_{eff}\nabla T\right) + S_E,
$$
where \(E\) is the total internal energy, \(k_{eff}\) is the effective thermal conductivity, and \(S_E\) includes latent heat exchange.
Because the vapor chamber requires a capillary wick to return the condensed liquid, I modeled the wick as a porous medium. The velocity in the saturated wick is related to the pressure gradient by Darcy’s law:
$$
Q = K A J,
$$
where \(K\) is the hydraulic permeability, \(A\) is the cross-sectional area, and \(J\) is the hydraulic gradient. In more general form, the steady momentum equation in the porous wick is
$$
\nabla p = -\frac{\mu}{K}\mathbf{v} + \rho \mathbf{g}.
$$
The overall thermal resistance of the vapor chamber is a key performance indicator. I defined it as
$$
R_{VC} = \frac{T_e-T_c}{Q_{in}},
$$
where \(T_e\) is the average temperature of the evaporator surface, \(T_c\) is the average condenser-surface temperature, and \(Q_{in}\) is the input heat. A lower thermal resistance indicates better heat-spreading performance. From the theoretical analysis, the total resistance includes the evaporator-wall resistance, the vapor-flow resistance, and the condenser-side convection resistance. Increasing the condenser surface area is particularly effective. This insight led me to design hollow fins on the condenser section of the vapor chamber.
In the first stage of the design, I modeled a conventional vapor chamber using COMSOL. The computational domain consisted of a metal casing with wall thickness \(0.5\,\text{mm}\), a vapor core, and a \(0.5\,\text{mm}\)-thick porous wick. I set the porosity to \(0.5\), the evaporator heat flux to \(1500\,\text{W/m}^2\), and the natural-convection heat-transfer coefficient on the condenser to \(5\,\text{W/(m}^\circ\text{C)}\). The resulting steady-state temperature ranged from \(37.4\,^\circ\text{C}\) to \(38.1\,^\circ\text{C}\). This baseline model confirmed that the conventional vapor chamber had acceptable uniformity but insufficient heat rejection to maintain a low maximum temperature.
To improve the condenser performance, I introduced hollow fins into the upper part of the vapor chamber. The hollow fins created additional pathways for vapor and increased the heat-transfer area without adding excessive mass. I identified four important geometrical factors: the number of fins on one side \(A\), fin width \(B\), fin length \(C\), and fin inclination angle \(D\). The table below lists the three levels used in the orthogonal experiments.
| Level | \(A\) / number of fins | \(B\) / mm | \(C\) / mm | \(D\) / degree |
|---|---|---|---|---|
| 1 | 2 | 1.5 | 10 | 15 |
| 2 | 4 | 2.0 | 15 | 30 |
| 3 | 6 | 2.5 | 20 | 45 |
I evaluated nine combinations according to an \(L_9\) orthogonal array. The calculated evaporator average temperature \(T_e\) was used as the response. The results are summarized as follows:
| Run | \(A\) | \(B\) | \(C\) | \(D\) | \(T_e\) / \(^\circ\text{C}\) | \(T_c\) / \(^\circ\text{C}\) |
|---|---|---|---|---|---|---|
| 1 | 1 | 1 | 1 | 1 | 32.4 | 31.3 |
| 2 | 1 | 2 | 3 | 2 | 33.2 | 32.3 |
| 3 | 1 | 3 | 2 | 3 | 34.2 | 33.5 |
| 4 | 2 | 1 | 3 | 3 | 31.2 | 29.8 |
| 5 | 2 | 2 | 2 | 1 | 32.0 | 30.9 |
| 6 | 2 | 3 | 1 | 2 | 33.4 | 32.5 |
| 7 | 3 | 1 | 2 | 2 | 30.7 | 29.3 |
| 8 | 3 | 2 | 1 | 3 | 32.4 | 31.3 |
| 9 | 3 | 3 | 3 | 1 | 29.9 | 28.0 |
I performed a range analysis to determine the most significant factors. For each factor \(i\) and each level \(j\), the mean response is
$$
\overline{T}_{i,j}=\frac{1}{n}\sum_{k=1}^{n}T_{e,k},
$$
and the range is
$$
R_i=\max_j\left(\overline{T}_{i,j}\right)-\min_j\left(\overline{T}_{i,j}\right).
$$
The range values are shown below.
| Metric | \(A\) | \(B\) | \(C\) | \(D\) |
|---|---|---|---|---|
| Level-1 mean | 33.3 | 31.4 | 32.7 | 31.4 |
| Level-2 mean | 32.2 | 32.5 | 32.3 | 32.4 |
| Level-3 mean | 31.0 | 32.7 | 31.4 | 32.6 |
| Range | 2.3 | 1.3 | 1.3 | 1.2 |
According to the range analysis, the fin count had the strongest effect on the evaporator temperature, followed by fin length, fin width, and inclination angle. The best combination was \(A_3B_1C_3D_1\), which represented six fins per side, a fin width of \(1.5\,\text{mm}\), a fin length of \(20\,\text{mm}\), and an inclination angle of \(15^\circ\). With this optimum configuration, the maximum temperature of the vapor chamber decreased to \(29.4\,^\circ\text{C}\), while the minimum temperature was \(27.3\,^\circ\text{C}\). The condenser length was reduced to one third of the entire vapor-chamber height, and the hollow fins helped maintain a uniform temperature over the evaporator surface.
After optimizing the hollow fins, I turned my attention to the wick structure. Conventional vapor-chamber wicks are made from sintered metal powder. Although this type of wick offers strong capillary pressure, it is difficult to print with additive manufacturing while preserving a controlled pore size and uniform pore distribution. I therefore designed a Kelvin-lattice wick, which can be formed by existing metal 3D printing technology and still provides the necessary porosity and flow channels. The Kelvin cell is an ideal space-filling structure that creates an open network of interconnected pores. I replaced the metal-powder wick in my simulation with a Kelvin lattice wick and kept the porosity at \(0.5\). The entire vapor chamber, including the hollow fins and the lattice wick, could be integrated into a single printed component.
For the phase-change simulation of the Kelvin-lattice vapor chamber, I switched from COMSOL to Fluent because of its strong computational fluid dynamics capabilities for two-phase flow. I used the mixture model with the evaporation–condensation model proposed by Lee. The mass-transfer rate is governed by the deviation between the fluid temperature and the saturation temperature. For evaporation, when \(T>T_{sat}\), the source term can be written as
$$
\dot{m}_{lv}=r_e\alpha_l\rho_l\frac{T-T_{sat}}{T_{sat}},
$$
and for condensation,
$$
\dot{m}_{vl}=r_c\alpha_v\rho_v\frac{T_{sat}-T}{T_{sat}}.
$$
The simulation domain was discretized with tetrahedral elements, and the time step was initially \(10^{-6}\,\text{s}\), increasing to \(10^{-5}\,\text{s}\) after the residual curves became stable. The metallic shell was defined as stainless steel, while the wick was defined as copper. The working fluid was a \(50\%\) ethanol aqueous solution, selected because of its low saturation temperature and high latent heat.
Before performing the lattice simulations, I validated the Fluent setup against COMSOL using the same sintered-powder wick geometry and the same boundary conditions. At steady state, the peak temperatures from Fluent and COMSOL were \(29.6\,^\circ\text{C}\) and \(29.4\,^\circ\text{C}\), respectively, giving an error of only about \(0.7\%\). This agreement gave me confidence to use Fluent for the remaining simulations. With the Kelvin-lattice wick, the steady-state maximum temperature was \(29.1\,^\circ\text{C}\), which was \(0.5\,^\circ\text{C}\) lower than that obtained with the powder wick. The Kelvin lattice therefore improved both manufacturability and heat-transfer performance.
The phase-fraction contours revealed the internal working process. At the initial instant, the working fluid in the vapor chamber was mostly liquid. As heat continued to be applied, vapor bubbles appeared near the evaporator and rose toward the condenser. The lattice structure increased the contact area between the fluid and the heated wall, which promoted more localized boiling. In the later stage, the vapor distribution became more uniform across the vapor chamber, indicating that the lattice wick provided stable and uniform evaporation sites and reduced the thermal resistance of the evaporator.
I fabricated a prototype vapor chamber using additive manufacturing to validate the simulations and to study the real performance under practical conditions. The experimental platform included a silicone heater with an effective area of \(100 \times 80\,\text{mm}\), a programmable direct-current power supply, an adjustable tilting stand, a fan with a variable voltage controller, T-type thermocouples, and a data-acquisition system. Six thermocouples were placed on the evaporator and condenser surfaces to measure temperature distributions. The average evaporator surface temperature \(T_e\) was calculated from three sensors on the evaporator, and the average condenser temperature \(T_c\) was obtained from three sensors on the condenser. All experiments were performed at steady state.
| Instrument | Specification | Measurement range | Accuracy |
|---|---|---|---|
| Thermocouple | T-type | -200 to 350 \(^\circ\text{C}\) | \(\pm0.4\%\) |
| DC power supply | eTm-3060PC | 0–30 A, 0–60 V | \(\pm0.2\%\) current, \(\pm0.02\%\) voltage |
| Data acquisition system | Keysight 32972A | 0–5 A channel | \(\pm0.05\%\) |
Comparison between the measured temperature history and the simulation showed close agreement for a heat flux of \(1500\,\text{W/m}^2\). The transient temperature increased rapidly at the beginning and gradually approached steady state after about \(600\) seconds. The maximum difference between the simulated and experimental results was less than \(5\%\), confirming that the numerical model could be used for subsequent parametric design.
I then studied the effect of heat flux on the evaporator-surface temperature and thermal resistance. The heat flux was increased stepwise from \(500\) to \(2500\,\text{W/m}^2\) in intervals of \(500\,\text{W/m}^2\). At heat fluxes below \(1000\,\text{W/m}^2\), the evaporator temperature increased slowly. Above \(1500\,\text{W/m}^2\), the temperature rise accelerated, indicating that the vapor chamber was approaching its heat-transfer limit. When the heat flux changed suddenly in the range of \(1500\) to \(2000\,\text{W/m}^2\), a small transient jump occurred because the heat was initially stored on the evaporator surface before being conducted into the working fluid. Once the evaporation mechanism became fully active, the temperature stabilized without overshoot.
I also tested the orientation sensitivity of the vapor chamber by rotating the prototype from \(-90^\circ\) to \(90^\circ\) in \(45^\circ\) increments. The orientation angle was defined as positive when the evaporator was below the condenser, which is the gravity-assisted orientation. The measured thermal resistance curves show that at \(-90^\circ\), the heat source was located above the condenser, and the liquid was unable to return through the lattice wick because the wick did not connect the two evaporating surfaces. As a result, the thermal resistance reached approximately \(0.44\,^\circ\text{C/W}\) at a low heat flux of \(500\,\text{W/m}^2\). In the range of \(-45^\circ\) to \(45^\circ\), however, the thermal resistance remained relatively stable, indicating that the Kelvin-lattice wick could maintain sufficient liquid supply against moderate adverse gravity. At the highest heat flux and the most favorable orientation, the thermal resistance decreased to \(0.12\,^\circ\text{C/W}\).
Because forced-air cooling is the main heat-removal mechanism in my proposed system, I tested the effect of fan input power on the thermal resistance of the vapor chamber. Three fan power levels were considered: \(0\), \(2.5\), and \(5\,\text{W}\). The results showed that at low heat fluxes, increasing fan power from \(0\) to \(2.5\,\text{W}\) produced only a small improvement. At a heat flux of \(2500\,\text{W/m}^2\), however, the evaporator surface temperature was controlled at \(36.2\,^\circ\text{C}\) when the fan was operated at \(5\,\text{W}\), which was \(3.4\,^\circ\text{C}\) lower than the temperature without the fan. The thermal resistance dropped from \(0.35\,^\circ\text{C/W}\) at no fan to \(0.12\,^\circ\text{C/W}\) at maximum fan power. This result demonstrates that the hollow-fin condenser needs forced convection to realize its full potential.
The start-up performance was evaluated at heat fluxes of \(500\), \(1500\), and \(2500\,\text{W/m}^2\). The times required to reach steady state were approximately \(180\), \(540\), and \(750\) seconds, respectively. Although a higher heat flux required a longer time to reach equilibrium, the initial temperature response was faster because the larger heat input produced a steep temperature gradient. The evaporator temperature did not show a visible overshoot, which indicates good dynamic capture ability of the designed vapor chamber.
The next stage of my project addressed the integration of the vapor chamber into an actual EV battery module. The battery was a lithium iron phosphate cell with a nominal voltage of \(3.2\,\text{V}\), a nominal capacity of \(110\,\text{Ah}\), and dimensions of \(149\,\text{mm} \times 52\,\text{mm} \times 115\,\text{mm}\). The module was arranged as \(24\) cells in series and \(8\) cells in the length direction. The table below lists the thermal and electrical properties of the cell.
| Property | Value |
|---|---|
| Nominal voltage / V | 3.2 |
| Nominal capacity / Ah | 110 |
| Voltage range / V | 2.5–3.5 |
| Dimensions \(x \times y \times z\) / mm | 149 × 52 × 115 |
| Density / kg/m³ | 2024 |
| Specific heat / J/(kg·K) | 1054 |
| Thermal conductivity \(k_x\), \(k_y\), \(k_z\) / W/(m·K) | 24, 0.8, 24 |
| Volumetric heat generation at 1C / W/m³ | 18022 |
| Single-cell heat generation at 1C / W | 16.08 |
I modeled the EV battery using an anisotropic heat-conduction equation:
$$
\rho C_p\frac{\partial T}{\partial t}
=\frac{\partial}{\partial x}\left(k_x\frac{\partial T}{\partial x}\right)
+\frac{\partial}{\partial y}\left(k_y\frac{\partial T}{\partial y}\right)
+\frac{\partial}{\partial z}\left(k_z\frac{\partial T}{\partial z}\right)
+\dot q_{gen}.
$$
The model was validated by comparing the simulated maximum temperature of the EV battery module under natural convection at discharge rates of \(0.33C\), \(0.5C\), and \(1C\) with experimental measurements. The simulated maximum temperatures were \(40.8\), \(48.0\), and \(56.2\,^\circ\text{C}\), while the measured values were \(41.1\), \(46.5\), and \(56.7\,^\circ\text{C}\), respectively. The maximum error was below \(5\%\). The largest temperature differences under those three discharge rates were \(1.0\), \(2.0\), and \(3.0\,^\circ\text{C}\). Since the \(1C\) case exceeded the acceptable \(45\,^\circ\text{C}\) limit, subsequent thermal management studies focused on the \(1C\) discharge condition; if the temperature could be reduced below \(45\,^\circ\text{C}\) at \(1C\), lower discharge rates would also satisfy the requirement.
My base thermal management structure consisted of the EV battery module, the optimized vapor chamber, and an epoxy baseplate. The vapor chamber was placed vertically between the battery module and the air-cooling side. In the simulation, the vapor chamber was not modeled as a full two-phase device. Instead, I used a user-defined function in Fluent to update its effective thermal conductivity according to the local evaporator and condenser temperatures. The effective conductivity was defined as
$$
k_{VC}=\frac{Q_{VC}L}{A_{VC}\Delta T_{VC}},
$$
where \(Q_{VC}\) is the heat transported by the vapor chamber, \(L\) is the distance between the evaporator and condenser centers, \(A_{VC}\) is the average cross-sectional area, and \(\Delta T_{VC}\) is the temperature difference between the evaporator and condenser. By using this dynamic effective conductivity, I captured the nonlinear heat-spreading behavior of the vapor chamber without performing a full phase-change simulation over the whole module.
Without any forced convection, the vapor chamber reduced the maximum temperature of the EV battery at \(1C\) discharge from \(56.2\,^\circ\text{C}\) to \(53.4\,^\circ\text{C}\), a drop of \(2.8\,^\circ\text{C}\). The temperature distribution was also more uniform because the vapor chamber transferred heat from the hot center of the module toward the edges. However, \(53.4\,^\circ\text{C}\) was still too high, so I added a forced-air duct to improve the condenser heat rejection.
I initially designed a single-channel air duct above the vapor chamber. The inlet air temperature was \(25\,^\circ\text{C}\), and the mass-flow rate was varied from \(0.05\) to \(0.5\,\text{kg/s}\). At a low flow rate of \(0.05\,\text{kg/s}\), the maximum EV battery temperature was \(50.5\,^\circ\text{C}\), which was above the \(45\,^\circ\text{C}\) target. At a higher flow rate of \(0.35\,\text{kg/s}\), the temperature decreased to \(32.6\,^\circ\text{C}\), but the maximum temperature difference exceeded \(4.5\,^\circ\text{C}\). At \(0.5\,\text{kg/s}\), the maximum temperature was \(31.5\,^\circ\text{C}\), yet the temperature difference rose to \(5.4\,^\circ\text{C}\). Thus, the single-channel duct could not simultaneously satisfy the temperature and uniformity requirements with a reasonable flow rate.
| Duct type | Inlet flow / kg/s | Maximum temperature / \(^\circ\text{C}\) | Minimum temperature / \(^\circ\text{C}\) | Temperature difference / \(^\circ\text{C}\) |
|---|---|---|---|---|
| Single channel | 0.05 | 50.5 | 47.8 | 2.7 |
| Single channel | 0.20 | 37.4 | 33.8 | 3.6 |
| Single channel | 0.35 | 32.6 | 28.1 | 4.5 |
| Single channel | 0.50 | 31.5 | 26.1 | 5.4 |
| Double channel | 0.05 | 47.8 | 44.8 | 3.0 |
| Double channel | 0.50 | 30.5 | 25.4 | 5.1 |
| Four channel original | 0.05 | 45.3 | 42.0 | 3.3 |
| Four channel original | 0.50 | 29.9 | 25.1 | 4.8 |
The velocity distributions showed that a single-channel duct produced a single dominant flow path with an average velocity of only \(2.8\,\text{m/s}\). A double-channel duct increased the maximum velocity to \(4.1\,\text{m/s}\), but the flow distribution remained uneven. A four-channel duct increased the maximum velocity to \(5.1\,\text{m/s}\) and produced a much more uniform flow field over the condenser fins. Nevertheless, the initial four-channel geometry still had some small regions with poor ventilation. I therefore used topology optimization to further improve the design.
Topology optimization was performed in COMSOL with the solid isotropic material with penalization, or SIMP, algorithm. The design domain was the air duct around the inlet and outlet plenums. The material density distribution was used as the design variable. The objective was to minimize the maximum EV battery temperature while maintaining the air-flow resistance at an acceptable level. I applied a Helmholtz filter and hyperbolic-tangent projection to obtain a clean and manufacturable topology. Four candidate designs were obtained. Three of them produced fragmented material distributions, while one candidate, labeled optimum \(2\), removed material from the center of the duct inlet without creating unnecessary small features. This solution was selected for final validation.
After modifying the four-channel duct according to the topology-optimization result, the velocity distribution was much more uniform. The high-velocity regions above \(4.5\,\text{m/s}\) were properly aligned with the hollow fins of the vapor chamber. The maximum velocity difference inside the duct was reduced compared with the original four-channel geometry, which decreased the local pressure loss and improved the convective heat transfer. I simulated the optimized system under an inlet temperature of \(25\,^\circ\text{C}\) and a low mass-flow rate of \(0.05\,\text{kg/s}\). The maximum temperature of the EV battery was \(44.7\,^\circ\text{C}\), satisfying the target of less than \(45\,^\circ\text{C}\). The maximum temperature difference was \(3.5\,^\circ\text{C}\), satisfying the target of less than \(5\,^\circ\text{C}\). This result confirmed that the combination of a Kelvin-lattice vapor chamber and a topologically optimized four-channel air duct can provide effective thermal management for a high-energy-density EV battery with minimal fan power.
The final component of my work was a digital twin for the EV battery thermal management system. A digital twin is beneficial because real-time operation requires rapid evaluation of different inlet temperatures and flow rates. Full three-dimensional computational fluid dynamics models are too slow for real-time control. I therefore created a reduced-order model using the Twin Builder platform from Ansys. The workflow began with a parametric Fluent model of the optimized EV battery thermal management system. I chose the inlet air temperature and inlet mass-flow rate as the input parameters. To cover the design space efficiently, I used a Latin-hypercube experimental design with nine points. The inlet temperature ranged from \(15\) to \(30\,^\circ\text{C}\), and the mass-flow rate ranged from \(0\) to \(0.5\,\text{kg/s}\).
| Run | Inlet temperature / \(^\circ\text{C}\) | Inlet flow / kg/s | Simulated \(T_{max}\) / \(^\circ\text{C}\) |
|---|---|---|---|
| 1 | 20.83 | 0.08 | 40.9 |
| 2 | 25.83 | 0.19 | 40.1 |
| 3 | 15.83 | 0.03 | 40.0 |
| 4 | 24.17 | 0.25 | 36.7 |
| 5 | 17.50 | 0.42 | 23.8 |
| 6 | 29.17 | 0.14 | 44.2 |
| 7 | 22.50 | 0.47 | 27.0 |
| 8 | 27.50 | 0.36 | 35.8 |
| 9 | 19.17 | 0.31 | 30.2 |
From these simulation results, I exported the temperature field and generated a low-order representation of the thermal response. The reduced-order model was then imported into Twin Builder and connected to a real-time visualization environment. To validate the reduced-order model, I compared its temperature field with the reference Fluent simulation for the design operating condition of \(25\,^\circ\text{C}\) inlet temperature and \(0.05\,\text{kg/s}\) flow rate. Both models produced very similar temperature distributions. The isotherms followed the same pattern, and the overall temperature range was preserved. The maximum local difference was about \(0.2\,^\circ\text{C}\), while the average temperature difference was only \(0.095\,^\circ\text{C}\). This small error proves that the reduced-order model can reproduce the high-fidelity simulation with excellent accuracy while enabling nearly instantaneous response.
| Run | Simulated \(T_{max}\) / \(^\circ\text{C}\) | Predicted \(T_{max}\) / \(^\circ\text{C}\) | Error / \(^\circ\text{C}\) |
|---|---|---|---|
| 1 | 40.9 | 40.9 | 0.0 |
| 2 | 40.1 | 40.3 | -0.2 |
| 3 | 40.0 | 40.1 | -0.1 |
| 4 | 36.7 | 36.8 | -0.1 |
| 5 | 23.8 | 23.8 | 0.0 |
| 6 | 44.2 | 44.3 | -0.1 |
| 7 | 27.0 | 27.0 | 0.0 |
| 8 | 35.8 | 36.0 | -0.2 |
| 9 | 30.2 | 30.1 | 0.1 |
The reduced-order model also allowed me to investigate the coupled influence of inlet temperature and flow rate on the maximum EV battery temperature. For a fixed inlet temperature, increasing the flow rate always reduced the maximum temperature because more air was available to carry heat away from the condenser. For a fixed flow rate, increasing the inlet temperature increased the maximum EV battery temperature because the driving temperature difference between the air and the condenser surface became smaller. The two parameters interacted strongly: an inlet temperature of \(30\,^\circ\text{C}\) with a low flow rate of \(0.05\,\text{kg/s}\) produced a predicted maximum temperature of \(47.6\,^\circ\text{C}\), while an inlet temperature of \(15\,^\circ\text{C}\) with a higher flow rate of \(0.45\,\text{kg/s}\) produced a maximum temperature close to \(21\,^\circ\text{C}\). This response surface provides a quantitative calibration for designing an intelligent controller that can adjust the fan speed based on the measured ambient temperature and the target EV battery temperature.
The digital twin was displayed through a three-dimensional interactive visualization interface. The upper panel contained two input sliders: one for the inlet air temperature and one for the mass-flow rate. The lower panel showed the temperature field of the EV battery module. By moving the sliders, I could immediately observe how the thermal distribution changed. Such a visualization tool is useful for both thermal engineers and control-system developers because it clearly communicates the effect of operating parameters on the spatial thermal pattern of the EV battery. It also supports anomaly detection; if a measured point deviates from the predicted field, the operator can detect a possible duct blockage, fan degradation, or sensor failure.
From a broader perspective, my work demonstrates a complete design loop for advanced EV battery thermal management. The first contribution is the design of a vapor chamber with hollow fins and a Kelvin-lattice wick. The hollow fins increased the condenser area and reduced the required condenser length to one third of the total height. The Kelvin-lattice wick made additive manufacturing feasible while decreasing the maximum equilibrium temperature by \(0.5\,^\circ\text{C}\) compared with a conventional powder wick. The experimental thermal resistance reached \(0.12\,^\circ\text{C/W}\) under a heat flux of \(2500\,\text{W/m}^2\) and a \(5\,\text{W}\) fan, demonstrating the excellent heat-spreading capability.
The second contribution is the coupled architecture of EV battery, vapor chamber, and forced-air duct. By representing the vapor chamber with a dynamic equivalent thermal conductivity in Fluent, I was able to simulate the full module efficiently while preserving the key effect of the phase-change heat spreader. The systematic comparison of single-, double-, and four-channel air ducts showed that increasing the number of channels improves both the maximum temperature and the temperature uniformity. Topology optimization further improved the flow distribution, enabling the EV battery to satisfy the recommended thermal limits at an inlet temperature of \(25\,^\circ\text{C}\) and a mass-flow rate of only \(0.05\,\text{kg/s}\).
The third contribution is the construction of a digital twin for the integrated system. The reduced-order model built with Latin-hypercube experiments and Twin Builder accurately reproduced the temperature field of the full computational-fluid-dynamics model with only a small error. The ability to predict the EV battery maximum temperature in real time makes it possible to optimize the fan power during transient driving cycles, thereby saving energy while maintaining safe operating temperatures. This breaks through the limitations of offline thermal management design and enables condition-aware thermal control.
In the future, additional work is needed to verify the coupled system under more extreme operating conditions, such as fast-charging cycles and high ambient temperatures. Thermal cycling and vibration tests should be carried out to confirm the reliability of the Kelvin-lattice vapor chamber over the full lifetime of an EV battery. The digital twin could also be extended to include cell-aging models, enabling predictive maintenance based on thermal impedance changes. Multi-scale coupling between the phase-change phenomena inside the vapor chamber and the module-level airflow remains an interesting topic; high-fidelity two-phase simulations can be used to generate more accurate thermal-response maps for the reduced-order model. Finally, embedded sensors and real-time telemetry could be combined with the digital twin to create a closed-loop thermal management system that adjusts the fan speed, duct opening, and refrigerant flow whenever the EV battery approaches a thermal limit.
In conclusion, this research has shown that the combination of an additively manufactured vapor-chamber heat spreader with a low-power forced-air duct can meet the demanding temperature requirements of modern EV battery packs. The hollow-fin condenser and Kelvin-lattice wick are effective, manufacturable innovations. The topology-optimized four-channel air duct achieves excellent temperature uniformity even with a very small airflow rate. The digital twin enables real-time visualization and control, which brings EV battery thermal management closer to the intelligent and adaptive operation expected in next-generation electric vehicles. I believe that this work provides a practical pathway for developing compact, efficient, and safe thermal management systems for high-energy-density EV battery packs.