Thermal Enhancement of LED Cooling Systems Using Passive Tesla Valves for Turbulence-Induced Convection

The escalating demand for high-power lighting solutions has propelled Chip-on-Board (COB) Light Emitting Diodes (LEDs) to the forefront of modern illumination technologies. COB LEDs, characterized by multiple LED chips mounted directly onto a thermally conductive substrate, offer superior luminous efficacy and compactness. However, their high power density and compact form factor pose significant thermal management challenges. Efficient heat dissipation is paramount to maintain performance, reliability, and longevity of these devices¹.

In COB LEDs, a substantial portion of the electrical energy is converted into heat rather than light, necessitating effective thermal management strategies to prevent performance degradation and ensure device longevity². Traditional passive cooling methods, such as heat sinks and thermal interface materials, often fall short in dissipating the intense heat generated by high-power COB LEDs³. Consequently, there is a pressing need for innovative cooling solutions that can efficiently manage the thermal loads without compromising the compactness and efficiency of the lighting systems.

The LED lighting luminaire quality is composed of different parameters: color rendering index (CRI)⁴, luminous flux, temperature color, and lifespan. The mentioned parameters are sensitive to heat; some works reported that the luminous flux decreases its value by 0.108 lm per degree Celsius⁵. Abdelmlek et al.⁶ show a correlation that estimates the lifespan of COB LEDs as a function of temperature, which involves the temperature junction of the LED. Recent advances in microelectronics have led to continuous increases in device operation frequencies and progressive miniaturization of electronic components. These trends result in significantly higher power densities and heat fluxes that challenge conventional thermal management strategies^7,8. Such thermal loads, if not adequately managed, can severely degrade device reliability and performance.

To improve the heat dissipation on COB LEDs, some researchers have designed heat sinks with particular geometries and configurations that allow improving the heat transfer from the LED toward the ambient through free convection^{9,10,11,12,13,14}. These kinds of heat sinks are an effective option to reduce energy consumption during their operation, but the heat amounts are limited by the superficial area, increasing their size if the heat amount increases, resulting in larger devices if high heat fluxes are involved. High-power chip-on-board (COB) LEDs combine multiple dies on a compact substrate, enabling high luminous efficacy but also generating significant heat flux densities that threaten device reliability. Effective thermal management is therefore essential to maintain junction temperatures below critical limits. Traditionally, passive approaches such as finned heat sinks and natural convection have been employed; however, their performance is restricted by limited surface area and the relatively low convective coefficients achievable under buoyancy-driven flow. To overcome these limitations, active solutions including forced-air heat sinks, heat pipes, spray cooling, and liquid-cooled microchannels have been widely investigated, offering lower thermal resistances and more uniform temperature distributions. Building on this progression, our work explores an alternative strategy: integrating Tesla valve-enhanced microchannels as a compact cold plate. This design leverages the passive diodicity and flow mixing properties of Tesla geometries to intensify local convection while avoiding the need for moving parts or complex manifolds, thus bridging the gap between conventional microchannels and more complex active cooling systems.

To manage high heat fluxes in these applications, liquid cooling systems are a good option because they are able to increase the heat transfer coefficient and improve the heat transfer in more reducing spaces than passive cooling systems. There are some works reporting the use of liquid cooling fluid, such as cold plates with different channel configurations^15,16,17,18.

One promising approach involves the integration of microchannel heat sinks, which enhance heat transfer by increasing the surface area in contact with the cooling medium. Among various microchannel designs, the Tesla valve-a passive, no-moving-parts device-has garnered attention for its ability to control fluid flow directionally, thereby optimizing heat transfer and minimizing backflow. The unique geometry of Tesla valves facilitates unidirectional flow, which can be advantageous in thermal management applications by promoting efficient coolant circulation and enhancing overall heat dissipation. In order to increase the heat transfer coefficient, the researchers have included geometries that perturbed the path lines of the fluid. Tesla valve is a geometrical configuration of channels that allows the fluid to go in one direction; it works as a diodic valve (unidirectional valve). When a flow is introduced in the opposite direction, the pressure and the turbulence inside the system increase notably. Some works use the Tesla valve to increase the turbulence, being a consequence of the increase of the heat transfer coefficient. Some researchers have used parametric models to determine the fluid flow patterns in Tesla valves for use as a cooling system^19,20. The most common application of tesla valve implemented in the cooling system is through the cold plates^21,22,23. For instance, research has shown that incorporating Tesla valve structures into microchannel heat sinks can enhance thermal performance by promoting turbulent flow and increasing the convective heat transfer coefficient²⁴. These findings suggest that Tesla valve-based microchannel heat sinks could be a viable solution for managing the thermal challenges associated with high-power COB LEDs.

Several studies have addressed the thermal management of high-power COB LEDs through liquid-based cooling techniques. For instance, jet impingement and spray cooling approaches have demonstrated heat transfer coefficients on the order of 104-105 W·m^-2·K^-1, effectively maintaining junction temperatures below the critical 120 °C threshold with relatively low pumping power^25,26. Likewise, microchannel cold plates fabricated beneath COB arrays have achieved thermal resistances as low as 0.019 °C/W when water is circulated at moderate velocities, highlighting the strong potential of channel miniaturization for enhancing cooling efficiency^27,28. Other strategies include ferrofluid-enhanced heat sinks²⁹ and compact heat-pipe modules designed around COB packages³⁰, both of which demonstrate improved thermal spreading and reduced thermal resistance compared to conventional solid metal sinks.

Compared to these approaches, the proposed Tesla-valve-based cooling plate introduces a novel passive mechanism that enhances mixing and turbulence generation without the need for moving parts or complex manifolds. While Tesla valves have been investigated in electronics and energy systems for their diodicity and flow control capabilities³¹, their application to COB LED cooling has not been extensively reported in the literature. This work, therefore, provides an original contribution by adapting Tesla-valve flow networks to the specific requirements of LED thermal management. Our results indicate that, under forced convection, the Tesla-valve plate achieves competitive surface temperature reductions relative to established liquid-cooling strategies, with the added benefit of simplified geometry and robustness against clogging or mechanical failure.

In this study, we investigate the efficacy of a CNC-machined aluminum microchannel heat sink incorporating Tesla valve structures for the thermal management of a 30 W COB LED. The design aims to leverage the directional flow control of Tesla valves to enhance coolant circulation and heat dissipation. By conducting experimental evaluations, we assess the thermal performance of the proposed cooling system, analyzing parameters such as temperature distribution, thermal resistance, and overall cooling efficiency. The outcomes of this research could inform the development of advanced passive cooling solutions for high-power LED applications, contributing to the optimization of thermal management strategies in compact electronic devices. The challenge of this project was the implementation of the Tesla Valve over an existing heat sink. With this arrangement, the power dissipated is major, allowing the use of devices with high power in a compact cooling system.

Figure 1 presents a schematic representation of the methodology followed in this study. The process begins with the geometry creation of the LED cooling device using CAD software. The geometry is then imported into the CFD solver (OpenFOAM/ANSYS Fluent), where mesh generation and boundary conditions are applied to perform the thermal-fluid simulations. In parallel, the experimental setup is implemented, including thermocouples and an infrared camera for temperature measurements. Finally, simulation results are validated against experimental data, ensuring the reliability of the proposed cooling approach.

Figure 1: Schematic of the experimental workflow: (1) geometry creation, (2) CFD simulations, (3) Manufacture for the prototype, (4) thermal measurements, and (5) validation. Please click here to view a larger version of this figure.

Geometry of the cooling device
The purpose of the project is to maintain a 50 W COB LED under a permissible temperature with the Tesla valve implementation. The COB used as a heat source is shown in Figure 2A, while Figure 2B shows 50 die chips turned on, indicating that each die has 1 W of power.

Figure 2: LED COB and the number of die chips. (A) LED COB used in the experiment (B) Demonstration of the number of die chips used. Please click here to view a larger version of this figure.

In the literature, some Tesla valves were reported to have a heat source mounted over the devices that contain the Tesla Valve, therefor, in this paper, the LED is part of the device because it is placed over the Tesla valve channels in direct contact with the cooling fluid in order to improve the heat transfer from the LED toward the fluid minimizing the thermal resistance. Figure 3A shows the geometry of the device. On the upper surface with transparency, the place where the LED is placed is shown. On the other hand, Figure 3B shows the proposed assembly for the cooling device.

Figure 3: Geometry and the proposed assembly of the cooling device. (A) Cooling device proposed showing the Tesla valve channels. (B) Cooling device proposed assembled. Please click here to view a larger version of this figure.

With the aim of placing the LED in direct contact with the fluid, the fluid inlet and outlet were designed over the above surface, as shown in Figure 4A. The fluid is conducted from the pump to the Tesla channels through the inlet terminal. Then, the fluid fills the inlet chamber, and the fluid is driven to the channels with the inlet channels shown in Figure 4B. In Figure 4B, the surface view above shows the fluid chambers of the device without any connections, including the upper cover.

Figure 4: Isometric behind view of the device. (A) Isometric view of the above surface cooling device. (B) Fluid chambers of the device. Please click here to view a larger version of this figure.

Materials and accessories
The materials and accessories used in the experiment are shown in Table 1. An aluminium flat bar was used to grind the Tesla channels with a CNC machine. The aluminium was selected due to its balance between thermal conductivity and cost. Furthermore, machining aluminium is easier and faster than other components, such as steel or harder materials. The upper cover shown in Figure 4A was designed with aluminium too. The inlet and outlet terminals are manufactured of brass and are connected through threaded holes on the upper cover. The COB metal-core was manufactured of aluminium alloy to improve the heat transfer.

Table 1: Accessories and materials used in the prototype manufactured.

Water was used as the working fluid due to its high specific heat capacity and widespread availability; furthermore, due to the low risk of generating flame and null toxicity. The flow through the channel was driven by a pump, which was characterized to determine the flow rate as a function of the voltage.

Boundary conditions
The pump's flow rate was characterized to determine the inlet boundary condition into the system. The inlet boundary condition is defined by an inlet velocity which was calculated with the flow rate and the transversal section of the inlet. The flow rate behavior of the pump is presented in Figure 5.

Figure 5: Flow rate of the pump and inlet velocity. Please click here to view a larger version of this figure.

Figure 5 shows the relationship between the applied voltage and two key hydraulic parameters: the average volumetric flow rate and the average fluid velocity . Each data point represents the mean value calculated from five independent measurements per voltage level, ensuring statistically robust interpretation.

Trend of volumetric flow rate
A clear increasing trend is observed in the average flow rate as the voltage increases. At the lowest voltage level (4 V), the average flow rate is approximately 1.63 × 10^-5 m³/s, while at 12 V it rises to nearly 3.49 × 10^-5 m³/s, which represents an 110% increase of over This behavior can be attributed to the influence of voltage on the electromechanical actuator driving the flow.

As voltage increases, the actuator speed rises, thereby increasing the volume displaced per unit time.

Behavior of average velocity
The fluid velocity follows a similar increasing trend, which is consistent with the fundamental relationship between flow rate and velocity:

Q = A • v     (1)

where A is the cross-sectional area of the conduit and v is the average velocity. Since the area remains constant, changes in Q directly translate to proportional changes in v. In this experiment, the average velocity increased from approximately 6.5 ×10^-3 m·s^-1 (at 4 V) to 1.41 ×10^-2 m·s^-1 (at 12 V)

Comparison between curves
Both the flow rate and velocity curves exhibit a nearly linear relationship with voltage. Minor nonlinearities may be present and can be attributed to factors such as transient instabilities in the power supply, small variations in sensor response time, and internal friction or head losses at higher flow regimes. While these effects are not significant within the scope of the current study, they could be explored further using nonlinear regression or dynamic system modeling in future work.

Practical Implications
From an engineering perspective, the graph provides a useful correlation between an accessible control parameter (voltage) and critical flow variables. This is particularly relevant for applications in microfluidics, precision dosing, forced-convection cooling systems, or flow structures based on passive elements such as Tesla valves. The monotonic and predictable behavior of the flow rate also enables precise system calibration, ensuring operational stability and repeatability in experimental or automated setups.

Mathematical model for fluid dynamics in a Tesla valve
The internal fluid dynamics of the cooling device are not clearly understood from a physical standpoint. Therefore, a computational fluid dynamics (CFD) simulation was carried out to analyze the flow behavior, visualize turbulence that enhances the heat transfer coefficient, and identify stagnant fluid zones that can be optimized in the Tesla valve channels.

To simulate the fluid flow behavior within the Tesla valve, we employ the Reynolds-averaged Navier-Stokes equations in conjunction with the standard K-epsilon turbulence model. This approach provides a balance between computational cost and accuracy in capturing turbulence effects within internal flow passages with recirculation and anisotropic eddies, as typically observed in Tesla valves.

Governing equations
The fluid used is water, considered incompressible and Newtonian, and the flow is assumed to be steady-state. The governing equations are as follows:

     (2)

where  is the time-averaged velocity vector.

     (3)

where:

 ρ [kg/m³] is the fluid density,
 p [Pa] is the pressure,
μ [Pa • s] is the dynamic viscosity,
μ_t  [Pa • s] is the turbulent eddy viscosity, defined as:

     (4)

The turbulence kinetic energy k and its dissipation rate ε are governed by the following transport equations:

Turbulent kinetic energy (k):

     (5)

Turbulent dissipation rate (ε):

   (6)

Here:

 k [m²/s²] is the turbulence kinetic energy,
ε [m²/s³] is the turbulence dissipation rate,
P_k [kg/(m • s³)] is the production of turbulence kinetic energy, defined as:

   (7)

σ_k and σ_ε are the turbulent Prandtl numbers for k and ε, respectively,

 C_μ, C_1ε , and C_2ε are empirical constants.

The constants used in the standard k - ε model are:
C_μ  = 0.09, (calibrated for shear layers and free-stream turbulence)
σ_k  = 1.00, (ensures realistic diffusion of k)
σ_ε  = 1.30, (controls diffusion of ε)
C_1ε  = 1.44, (balances production and destruction of ε)
C_2ε  = 1.92. (ensures proper energy dissipation rate)

These values are derived from experimental validation in turbulent boundary layer flows and have been widely used in engineering applications with reliable accuracy^32,33,34,35.

The Tesla valve presents internal flow conditions characterized by boundary layer separation, recirculation zones, and directional resistance. These phenomena are inherently turbulent, and direct numerical simulation (DNS) or large eddy simulation (LES) approaches are computationally prohibitive for practical designs.

The k-ε model is for this study because it provides reliable predictions of the mean flow field and turbulence quantities in internal, high-Reynolds-number flows. It effectively captures the anisotropic turbulence effects induced by the turning channels and stagnation zones of the Tesla valve, ensuring an accurate representation of the complex flow behavior. Moreover, this model is robust and has been widely validated in engineering applications involving jet impingement, flow separation, and complex wall-bounded flows.

In this work, boundary conditions include a fully developed turbulent inlet velocity profile, no-slip walls, and a pressure outlet. The simulation domain corresponds to a multi-stage Tesla valve geometry, which is meshed using a hybrid structured-unstructured grid with boundary layer refinement to resolve near-wall effects using appropriate wall functions.

Boundary conditions and discussion
To ensure realistic flow simulation within the Tesla valve geometry, appropriate boundary conditions were defined for all relevant variables: velocity, pressure, turbulent kinetic energy (k), and turbulent dissipation rate (ε). The following conditions were imposed:

Inlet (velocity inlet):

A uniform velocity profile was prescribed at the inlet, assuming fully developed, incompressible, turbulent flow. The turbulent quantities k and ε were initialized based on the turbulence intensity I and the hydraulic diameter D_h, using the corresponding empirical relations.

     (8)

This approach ensures consistency with empirical turbulence generation models and preserves compatibility with the standard k-ε  formulation.

Outlet (pressure outlet):
A fixed static pressure (typically 0 Pa gauge) was set at the outlet to allow a fully developed outflow. Zero-gradient (Neumann) conditions were applied to all transported variables (velocity, k, and ε) to prevent artificial backflow effects and ensure numerical stability at the domain boundary.

Walls (no-slip condition):
The velocity at all solid boundaries was set to zero, applying the no-slip condition. For turbulence modeling, standard wall functions were used to represent near-wall behavior. The turbulent kinetic energy (k) was set to zero at the wall, while the dissipation rate (ε) was computed using standard wall-function approximations derived from the law of the wall. These functions eliminate the need to resolve the viscous sublayer, thereby reducing computational cost, under the assumption that the flow remains within the logarithmic region (y⁺ > 30).

This boundary configuration effectively captures the main flow features within the Tesla valve, including flow separation, recirculation zones, and turbulence intensification along the curved passages.

Thermographic image acquisition and measurement considerations
To evaluate the thermal performance of the cooling system and the COB-type LED module, thermographic measurements were conducted using an uncooled infrared camera operating in the 8-14 µm spectral range. The thermographic inspection aimed to characterize the surface temperature distribution under steady-state operating conditions. Special attention was paid to minimizing uncertainties commonly associated with infrared thermography.

One of the most critical aspects was the adjustment of the surface emissivity values for accurate temperature retrieval. The cooling system was composed of CNC-machined aluminum plates with a polished finish, whose low and angle-dependent emissivity poses a significant challenge for infrared measurements. To address this, the aluminum surfaces were coated with a high-emissivity matte black paint (ε ≈ 0.95), previously calibrated using contact thermocouples and infrared reference materials. This treatment ensured uniform emissivity and eliminated specular reflections that would otherwise compromise the accuracy of the measurements.

For the LED COB module, whose encapsulating material is typically a phosphor-coated silicone with higher emissivity (ε ≈  0.92), no additional coating was necessary. However, care was taken to avoid thermal gradients during camera setup by allowing the LED to reach thermal equilibrium before capturing images. A waiting period of at least 60 min was imposed after powering the LED to ensure steady-state conditions were established.

Ambient conditions were controlled during the experiments. The measurements were carried out in a closed laboratory with stable ambient temperature (25 ± 1 °C) and negligible air currents.

Reflected apparent temperature was estimated using a crumpled aluminum foil method and manually set in the Infrared (IR) camera. Relative humidity was not taken into account during the measurements. The camera was mounted on a vibration-free tripod at a perpendicular angle to the surface of interest, ensuring consistent viewing geometry. The distance between the camera and the target was kept constant (approximately 0.5 m), and the field of view was adjusted to maximize spatial resolution without introducing parallax errors. Calibration was performed before each session using a blackbody reference source to validate the radiometric response of the camera. Figure 6 illustrates the experimental setup used during the thermographic acquisition. Table 2 summarizes the key parameters and procedures implemented to ensure accurate and repeatable temperature measurements.

Figure 6: System instrumentation and specimen coating. (A) System instrumentation for thermocouple measurements. (B) Specimen coated in black paint for thermography measurements. Please click here to view a larger version of this figure.

Table 2: Thermography parameters and procedures used for error minimization.

The parameters used in the thermography measurements are listed in Table 2.

Infrared thermography procedure and uncertainty analysis
Infrared (IR) thermography was employed to non-invasively map the surface temperature field of a 30 W COB LED assembly cooled by a Tesla-valve microchannel cold plate. The camera operated in the long-wave band (LWIR), mounted on a rigid tripod, normal to the surface of interest to minimize view-angle effects. All acquisitions were performed after the system reached steady conditions. A fixed imaging distance was selected so that the instantaneous field-of-view (IFOV) spot size was at least 3× smaller than the smallest region of interest (ROI), ensuring adequate spatial sampling across the steep temperature gradients induced by the Tesla-valve features.

To minimize common thermographic artifacts, several precautions were implemented throughout the experimental procedure. The optics were carefully cleaned, and the lens was manually focused on the region of interest (ROI) using the maximum-gradient criterion, while a non-uniformity correction (NUC) was executed prior to each acquisition. To control emissivity, all bare metallic surfaces, such as aluminum and copper-which are inherently low-emissivity and reflective in the long-wave infrared (LWIR) range-were coated with matte black high-emissivity paint (nominal ε ≈ 0.95) or covered with calibrated emissivity tape. Thermographic ROIs were defined exclusively on these treated areas. The apparent reflected temperature (T₍ref₎) was measured using the crumpled-aluminum-foil method and incorporated into the camera's radiometric model; shiny or angled surfaces were shielded from external radiation sources to prevent false hot spots. Measurements were performed inside a draft-minimized enclosure, with the operator positioned behind a screen and external heat sources such as power supplies or lamps kept outside the field of view to reduce parasitic reflections. The optical axis was maintained close to normal incidence to limit directional emissivity effects and parallax errors. The same camera distance and ROI definitions were preserved throughout all tests to ensure repeatability.

Surface emissivities were assigned as follows for radiometric temperature retrieval, and only treated (painted or taped) regions were used for quantitative reporting: black-painted aluminum heat sink/Tesla valve, ε = 0.95 ± 0.02; solder mask on MCPCB (white), ε ≈ 0.90 ± 0.03; and silicone encapsulant, ε ≈ 0.94 ± 0.03. However, encapsulant regions were not used for quantitative ROIs to avoid semi-transparency effects in the LWIR range.

The camera was configured with the emissivity of the specific ROI (typically ε = 0.95) and with the measured T_ref and ambient air temperature T_∞. Each thermogram consisted of N ≥ 10 frames averaged to reduce random noise. For every ROI (e.g., near-junction area on the COB substrate and representative points on the cold plate), the mean temperature and standard deviation s_T were recorded. The resultant thermal map was used to compute the temperature rise for subsequent thermal-resistance and heat-transfer analyses.

Let the reported LED surface temperature be T_IR. The combined standard uncertainty u_c(TIR) was estimated using the law of propagation of uncertainty, assuming uncorrelated inputs:

 =     (9)

where: (i) u_acc accounts for the camera's radiometric accuracy (manufacturer specification, converted to k=1), (ii) u_T,ε captures the sensitivity to emissivity, (iii) u_focus represents defocus/IFOV mismatch, (iv) u_NUC covers detector non-uniformity residuals, (v) u_ref  models uncertainty in reflected apparent temperature, and (vi) u_rep is repeatability (frame-to-frame and run-to-run).

For emissivity sensitivity, linearization of the Stefan-Boltzmann inversion about (T,ε) yields the widely used approximation:

    (10)

with T in kelvin. The expanded uncertainty at ~ 95% confidence was reported as U(T_IR) ≈ 2uc(T_IR).

The following representative values correspond to measurements taken on black-painted regions near 65 °C (T ≈ 338 K). The camera has an accuracy specification of ±2 °C or ±2% of the reading, treated as k = 2. The emissivity was set to ε = 0.95 ± 0.02, with well-focused optics and controlled reflections. The individual uncertainty contributions were as follows: the camera accuracy (k = 1) yields u_acc = 1.0 K; the emissivity contribution, estimated from Equation (10), is u₍T,ε₎≈ (338 × 0.02)/(4 × 0.95) ≈ 1.78 K; the focus or IFOV effect gives u_focus = 0.3 K; the non-uniformity correction residual u_NUC = 0.2 K; the reflected apparent temperature u_ref = 0.4 K; and the repeatability between frames or runs u_rep = 0.2 K.

Thus,

 =  

The uncertainty results are shown in Table 3.

Table 3: Uncertainty budget for IR temperature on black-painted ROI (illustrative).

The overall thermal behavior of the COB LED assembly cooled by a Tesla-valve microchannel plate can be quantified through the concept of thermal resistance. Analogous to Ohm's law in electrical circuits, the thermal resistance R_θ relates the temperature rise of the LED to the input power dissipated as heat:

     (11)

where T_LED is the surface temperature of the LED package, T_∞ is the ambient air temperature, and P is the electrical power supplied to the LED (assuming nearly all of it is converted into heat). A low value of R_θ indicates efficient heat removal, which is critical for maintaining luminous efficacy and ensuring the long-term reliability of high-power LEDs.

The heat generated at the LED junction follows a conduction-convection path. Initially, heat flows by conduction through the LED substrate, the solder interface, and the Tesla valve material. Once transferred to the microchannels, the heat is removed by convection to the coolant flowing inside. Each interface contributes to the overall R_θ:

     (12)

with the convective resistance R_{θ,convective}, typically dominating in compact, high-flux LED systems. The resistance diagram is shown in Figure 7.

Figure 7: Thermal resistance network of the COB LED coupled to the Tesla-valve microchannel. Please click here to view a larger version of this figure.

The heat flows from the LED junction through conduction paths and into the coolant via convection, with each stage represented as a thermal resistance (R_js,R_cond,R_conv).

The Tesla-valve microchannels act as a passive turbulence promoter, increasing the local convective heat transfer coefficient h without the need for moving parts. Unlike straight microchannels, the Tesla design introduces curvature and recirculation zones that intensify mixing, thus enhancing convective cooling. The effective convective resistance can be expressed as:

     (13)

where A is the effective heat transfer area. Higher h values achieved through the Tesla-valve effect directly reduce R_θ, leading to lower LED operating temperatures. However, this enhancement comes at the expense of additional pressure drop, which must be balanced against pumping power requirements.

For the tested 30 W COB LED, typical measurements yielded a surface temperature rise of ΔT ≈ 40 K above ambient, with an effective dissipated power of P ≈ 29.9 W. Substituting into Eq. (11), the overall thermal resistance was:

This value is notably lower than that of conventional passive sinks of similar footprint, highlighting the beneficial effect of the Tesla-valve microchannel plate. Furthermore, the spatial temperature distribution obtained via infrared thermography confirmed that the regions directly coupled to Tesla channels exhibited more uniform temperature fields, evidencing improved convective performance.

Implications
A reduced thermal resistance translates into a lower LED junction temperature, directly mitigating lumen depreciation and color shift, both of which are temperature-dependent.

The Tesla-valve microchannel approach therefore provides a practical passive enhancement for compact LED cooling systems, merging manufacturability with thermal-hydraulic performance. Nevertheless, the increased hydraulic resistance introduced by the Tesla geometry must be considered in pump selection and system-level energy efficiency analyses.

Prototype manufacturing

The prototype was fabricated using CNC milling to create both the Tesla channels and the chambers. The machined Tesla channels are shown in Figure 8A, while the fabricated chambers are presented in Figure 8A. Subsequently, a drilling process was carried out to form the inlet and outlet ports on the chambers.

Figure 8: Manufactured Tesla valve. (A) Top view. (B) Bottom view. Please click here to view a larger version of this figure.

To prevent fluid leakage, a static sealing groove was incorporated to accommodate a rubber gasket around the Tesla channels, as illustrated in Figure 8A. The complete assembly of the cooling device is shown in Figure 9. The fluid inlet and outlet are positioned above the Tesla valve design to achieve a compact LED cooling configuration, as shown in Figure 9B.

Figure 9: Assembly of the cooling system. (A) Front view. (B) Above view. (C) Lateral view. Please click here to view a larger version of this figure.

The Tesla valve was implemented to enhance the convective heat transfer coefficient by inducing turbulence through its internal geometry. Therefore, a numerical model, as described in previous sections, was employed to estimate the internal fluid dynamics. The simulation was preprocessed in FreeCAD using the CFDof module. The boundary and operating conditions were configured to obtain the velocity field shown in Figure 10.

Figure 10: Velocity profiles within the Tesla channels. Please click here to view a larger version of this figure.

In Figure 10, the red circle on the right represents the fluid inlet, while the one on the left indicates the outlet. The inlet velocity corresponds to the pump operating at 12 V, as referenced in Figure 5. Although the figure indicates a maximum velocity of approximately 3 m/s, the actual inlet velocity is around 8.9 m/s. However, the velocity scale was intentionally limited in the plot to better visualize vortex structures and flow behavior within the Tesla channels.

The fluid was introduced in reverse through the diodic Tesla valve to promote the formation of turbulent vortices. These vortices are evidenced by localized increases in velocity at the intersections of the curved channels, characteristic of reverse flow behavior. High-velocity zones are also observed at the beginning and end of the channel paths. These zones are not directly related to the Tesla valve effect but result from the geometric features of the cooling system, specifically the inlet positioning on the upper surface. To accommodate this, geometric steps were included, as shown in Figure 11.

Figure 11: Geometrical step implemented to introduce fluid from the top of the system. Please click here to view a larger version of this figure.

The maximum pressure obtained under the prescribed boundary conditions was approximately 1.5 KPa at the inlet, which was subsequently distributed within the inlet chamber. Figure 12 shows the pressure field. A sharp pressure drop is observed at the entrance of each Tesla channel, caused by the sudden reduction in cross-sectional area. Along the channel, the pressure continues to decrease and then rises near the outlet, attributed to the corresponding increase in area.

Figure 12: Pressure distribution within the Tesla valve. Please click here to view a larger version of this figure.

To verify that the numerical estimations are not biased by spatial discretization, we conducted a mesh independence study centered on the fluid velocity field. Four successively refined meshes (coarsevery fine) were evaluated under identical boundary conditions and solver tolerances. As the primary metric, we used the area-averaged outlet velocity,  , defined as the mass-flux-weighted average over the outlet patch; this quantity is directly linked to the global momentum balance and is sensitive to near-wall resolution. As a secondary check, we monitored the maximum velocity magnitude in the domain,umax , to ensure that peak values in shear layers and recirculation cores were captured stably with refinement.

For each mesh, simulations were advanced until residuals dropped below the prescribed criterion and integral monitors reached steady values. The relative variation between successive meshes was quantified as

where φ ∈ { out,umax} and ε is a small positive number to avoid division by zero. Mesh convergence was deemed satisfactory when εk < 0.5% for out (and < 4% for umax). As summarized in Figure 13, the variation of uout between the third and fourth meshes falls below the adopted tolerance, indicating mesh-independent predictions. Consequently, the third mesh was selected for the rest of the simulations as it offers an optimal compromise between accuracy and computational cost.

Figure 13: Mesh independence based on maximum velocity. Grid independence using velocity-based metrics: area-averaged outlet velocity uout (primary) and peak velocity umax (secondary) for four mesh densities. The last refinement satisfies the adopted tolerance for uout. Please click here to view a larger version of this figure.

In the experiment, two tests were conducted to evaluate the thermal performance of the system under free convection, as originally designed for these LEDs. The tests were performed with two different input powers: 30 W and 50 W. The heat sink used was specifically designed to support a thermal load of 30 W. To avoid overheating, the LED temperature must remain below 100 °C. Figure 14 shows the resulting temperature distributions.

Figure 14: Thermographic images of the LEDs under free convection. (A) Temperature distribution with 30 W input power. (B) Temperature distribution with 50 W input power. Please click here to view a larger version of this figure.

As shown in Figure 14A, the maximum temperature reached by the 30 W LED was 92.2 °C, maintaining the LED within an acceptable operating range. This behavior is expected since the heat sink was designed for 30 W dissipation. However, in Figure 14B, the temperature exceeds the permissible threshold, leading to LED overheating. Notably, the LED presented in Figure 14B experienced partial failure, as some die chips ceased to function after the test.

Conversely, the cooling system proposed in this study was evaluated using the 50 W LED under the same operating conditions as those used in the free convection tests. The resulting temperature distributions are presented in Figure 15.

Figure 15: Thermographic images of the LEDs under forced convection. (A) Temperature distribution at a flow rate of 3.5×10-5 m3/s (12 V). (B) Temperature distribution at a flow rate of 3.14×10-5 m3/s (9 V). (C) Temperature distribution at a flow rate of 1.06×10-5 m3/s (5 V). Please click here to view a larger version of this figure.

According to Figure 5, which presents the flow rate as a function of pump voltage, the cooling system was configured to operate at 12 V, 9 V, and 5 V. As expected, the flow rate increases with voltage. In Figure 15A, a maximum temperature of 75.8 °C was recorded at 12 V. When the voltage was reduced to 9 V, the temperature slightly decreased to 75.3 °C (Figure 15B), and at5 V, the temperature increased to 76.4 °C (Figure 15C). The observed temperature difference of approximately 0.5 °C between the 12 V and 9 V cases can be attributed to two main factors: the small difference in flow rates between the two cases and the measurement uncertainty inherent to the laboratory environment. These results demonstrate that the system can operate efficiently at lower flow rates, which is advantageous for reducing energy consumption and optimizing spatial integration during implementation.

Thermal measurement reliability

The thermal camera was employed to monitor surface temperatures in a non-contact manner, and the reliability of its readings depends on several factors. The manufacturer specifies a sensor accuracy of ±2 °C or ±2 % of the measured value, whichever is greater. Because emissivity has a strong influence on temperature estimation, incorrect emissivity settings-particularly for shiny or metallic materials-can lead to significant deviations. To mitigate this effect, all aluminum components were coated with matte black paint, providing a stable and well-defined emissivity close to 0.95. Reflected radiation from nearby sources, including the operator and surrounding heat emitters, was minimized through careful camera alignment and shielding. The optical resolution, governed by the distance-to-spot ratio, was maintained by keeping a fixed and optimized measurement distance, ensuring pixel-level accuracy within the defined regions of interest. To further reduce measurement uncertainties, calibration checks were performed against a blackbody reference, and the camera parameters were adjusted prior to each experimental run.

Critical steps influencing the method's success

The reproducibility and success of the proposed methodology are highly dependent on several critical steps. First, the definition of the computational domain and the quality of the mesh play a decisive role. A mesh that is too coarse can smooth temperature gradients and underestimate heat transfer, whereas excessive refinement leads to high computational costs without significant gains in accuracy. Second, the choice of boundary conditions, particularly the imposed heat fluxes on the LED surface and the thermal properties of the working fluid, must accurately represent the physical system. Small deviations in these inputs can produce large discrepancies between simulations and experimental measurements. Third, solver stability and convergence settings are essential for reliability. In our case, the use of the solver with a residual criterion of 10-6 was necessary to ensure consistent and physically meaningful results. Finally, the quality of experimental validation depends strongly on the placement and calibration of thermocouples and infrared imaging. Even slight misalignments in sensor location or improper calibration can introduce errors that mask the true thermal performance of the system. Since water was used as the working fluid, maintaining leak-free operation and steady flow conditions was also critical to avoid artifacts in the temperature distribution.

Overall, careful control of mesh generation, boundary conditions, solver settings, and measurement practices constitutes the foundation for the robustness of the methodology and strengthens the agreement between CFD predictions and experimental results.

We quantify the cooling performance of the Tesla-valve microchannel plate using standard dimensionless metrics: the Reynolds number (Re), Prandtl number (Pr), Nusselt number (Nu), the Colburn heat-transfer factor (j), and the performance evaluation criterion (PEC).

         (14)

From the pump characterization (Table 4), we take the measured volumetric flow rates at the three voltages used in the experiments.

Table 4: Operating points from pump characterization (experimental).

The thermophysical properties of water at 25 °C used in the numerical case are shown in Table 5. Since the microchannel dimensions for LED applications have not yet been reported, we present in Table 6 the main features of the Tesla valve used for the computation of the dimensionless numbers.

Table 5: Water properties at 25 °C (Elsevier tables).

Table 6: Microchannel geometry (illustrative, replace with measured values).

Experimentally, the overall thermal resistance was about Rθ  1.34 K/W. As a conservative bound, we take Rθ,conv Rθ, leading to:

     (15)

With Aw  = 4.0 × 10-4m2 and Rθ = 1.34 K • W-1, Eq. (15) yields h  1/(1.34 × 4.0 × 10-4) ≈ 1865 W • m-2 • K-1  . This is a lower bound on h because Rθ includes non-convective contributions.

Velocity, Reynolds, Nusselt, j and PEC (Case A, 12 V)

Using QA = 3.49 × 10-5 m3/s and the illustrative At = 5.0 × 10-6 m2, ,

With h 1865 Wm-2K-1 and Dh = 6.67 × 10-4 m,

Hence, the Colburn factor (with Pr 6.2):

For benchmarking against a straight duct at the same Re, we use (i) fully developed laminar Nu0 = 4.36 when appropriate36 or (ii) a turbulent reference via the Gnielinski formulation37. Since the present illustrative Re ~ 5.2×103 is transitional, we report the laminar baseline to keep a conservative stance:

Taking Nu0 = 4.36 and the laminar friction reference f0 = 64/Re for a smooth straight duct, one needs the Tesla-plate friction factor f (from measured Δp and L) to complete PEC. If we adopt the measured maximum pressure rise near the inlet Δp ~ 1.5 kPa over an effective path L, then

Table 7: Illustrative dimensionless results at 12 V (replace geometry/length to update).

Table 7 presents illustrative dimensionless results at 12 V (replace geometry/length to update). Because Rθ contains conduction and interfacial contributions, Eq. (15) yields a conservative (low) h, hence a lower bound on Nu and j. A tighter (larger) Nu can be obtained from a calorimetric balance combined with ΔTwall in the channels. Once the hydraulic length L and section-wise Δp are inserted, f and PEC follow directly. A value PEC > 1 indicates a net advantage of the Tesla geometry over the straight-duct benchmark for a fixed pumping criterion38.

Benchmark consistency checks

For fully developed laminar internal flow with constant wall heat flux, the canonical straight-tube value is Nu0=4.36. For higher-Re regimes, the Gnielinski correlation (widely validated in Elsevier literature) provides Nu0 (Re,Pr) and couples naturally with friction-factor data to produce j and PEC 37. The Colburn formulation j = Nu/(RePr1⁄3) is standard for comparing heat-transfer augmentation at equal flow conditions36.

Applicability of the system

To illustrate the applicability of the proposed cooling system, a 90 W commercial LED street lamp was selected as a reference (Figure 16). This lamp employs Surface-Mount Device (SMD) LEDs, whose main drawback is limited heat dissipation due to the additional thermal resistance of the PCB attached to the heat sink.

Figure 16: Commercial 90 W LED street lamp employing SMD technology. Please click here to view a larger version of this figure.

Two modifications are suggested to adapt the lamp for the proposed system. First, SMD LEDs are replaced by Chip-on-Board (COB) LEDs to improve heat transfer and increase luminous flux per unit of power. Second, a pocket is machined into the heat sink to accommodate the cooling device (Figure 17). The pocket depth is limited to 6 mm to preserve the upper fins, and the system is attached with screws. With this arrangement, the cooling device dissipates most of the thermal load, while the heat sink handles only residual heat. The fluid reservoir and micropump are housed within the lamp casing.

Figure 17: Modified LED lamp with COB technology and integrated cooling system. Please click here to view a larger version of this figure.

When working with high-power LEDs, proper safety measures must be considered due to the significant amount of heat generated during operation. Direct contact with heated surfaces, including the LED module, heat sink, and fluid channels, should be avoided to prevent burns. Appropriate thermal insulation and protective gloves are recommended when handling the device under operation. In addition, eye protection should be used to avoid potential damage from the intense light emitted by the LEDs. In this study, water was employed as the working fluid for the cooling device because water is generally safe and non-toxic. Potential leaks must be carefully monitored to avoid short circuits in the measurement electronics and to prevent unintended contact with heated components. All tubing connections should be properly sealed prior to operation, and the system must be drained after experiments to prevent corrosion or microbial growth. Spilled water should be promptly cleaned to reduce slipping hazards in the workspace.

In practical terms, the integration of the cooling device into a commercial street luminaire requires only minor modifications to the existing heat sink geometry. The cooling pocket and internal tubing can be manufactured using standard machining processes or die-casting for mass production. The micropump and fluid reservoir can be integrated within the luminaire enclosure, maintaining an IP65 or higher ingress protection rating by employing sealed connectors and corrosion-resistant materials. These characteristics make the system compatible with existing manufacturing lines and regulatory standards for outdoor lighting equipment.

Outside the laboratory, the cooling system can be adapted for different environmental conditions. In temperate climates, water can be safely used as the working fluid; however, in colder regions or in applications exposed to freezing temperatures, a mixture of water and glycol or dielectric fluids can be employed to prevent freezing and ensure electrical insulation. For applications in high-temperature environments, such as industrial halls or foundries, the system can operate in a closed loop with a small heat exchanger or radiator attached to the lamp casing to maintain a stable fluid temperature.

The implementation of the cooling device can significantly extend the lifespan of LED modules in real installations by reducing the junction temperature and mitigating thermal fatigue in solder joints and encapsulant materials. This contributes to lower maintenance intervals and improved luminous performance over time. Furthermore, the compact nature of the system makes it suitable for other confined or vibration-prone environments such as automotive headlights, tunnel illumination, and architectural lighting, where space constraints and reliability are critical.

From an economic standpoint, the device presents a low-cost solution that bridges the gap between passive and fully active cooling technologies. Unlike fan-based systems that require periodic replacement, the Tesla valve-based approach ensures reliability without noise generation or mechanical wear. Its scalability allows it to be adapted for various power ranges simply by adjusting the number of channels or the flow rate, enabling manufacturers to integrate it across multiple product lines.

In future work, optimization of heat transfer performance and pump selection will be carried out to enhance industrial applicability and achieve a compact system capable of dissipating significant heat loads in confined spaces. Additionally, we will focus on the experimental integration of the proposed system into the commercial lamp.

DATA AVAILABILITY:

All raw and processed data supporting the findings of this study are publicly available in the Zenodo repository under the title "Experimental and Numerical Data for a Tesla-Valve-Based LED Cooling Device." The dataset includes experimental temperature measurements, infrared thermography images, numerical simulation files, and post-processing scripts used for analysis. The complete dataset can be accessed at https://doi.org/10.5281/zenodo.17428394. Researchers are encouraged to reuse these data under the terms of the Creative Commons Attribution 4.0 International License.

Unlike conventional heat sinks^{9,10,11,12,13,14}, the system integrates a forced convection mechanism within a compact configuration that significantly enhances heat dissipation. Importantly, the integration of inlet and outlet ports in a radial configuration allows the system to occupy minimal space, making it ideal for confined installations.

This is the first time a thermally optimized compact device of this nature has been applied to LED lighting systems. Its application opens a pathway for implementing advanced cooling strategies in small-scale luminaires, enabling both thermal protection and design flexibility.

The experimental and numerical results demonstrate that the Tesla-valve-based cooling system provides a significant improvement in heat dissipation compared to conventional cooling. The formation of vortices and recirculating zones, clearly visible in the velocity field, enhances mixing and local convective heat transfer, contributing to the reduction in junction temperature. The pressure drop distribution corroborates the diodic nature of the Tesla channels, which promote turbulence generation while maintaining acceptable flow resistance for small-scale pumping systems.

The mesh independence verification supports the robustness of the CFD predictions, ensuring that the numerical trends-particularly in velocity and pressure-are physically meaningful. The strong agreement between the computational and experimental temperature fields reinforces the reliability of the methodology.

Thermal performance tests confirmed that the device effectively maintained the LED temperature well below critical limits even at 50 W, far outperforming the free convection configuration. The narrow temperature variation observed between 12 V and 9 V operation suggests that moderate reductions in flow rate can be achieved without compromising thermal performance, an important consideration for energy-efficient operation.

The computed Nusselt number and Colburn factor confirm that the Tesla geometry promotes convective enhancement in the transitional Reynolds regime. Although the PEC value depends on the pressure drop characteristics, preliminary results indicate a favorable trade-off between heat transfer and flow resistance.

From a design standpoint, the compact radial configuration enables integration into confined LED assemblies, providing a scalable thermal management strategy for high-power or high-density COB arrays. The proposed design could also be adapted for other electronics cooling applications where spatial constraints and energy efficiency are critical.

Future work will focus on optimizing channel geometry, pump selection, and the integration process in commercial luminaires. Enhancing the predictive coupling between experimental thermography and CFD modeling will further refine the understanding of localized heat transfer mechanisms in Tesla-based microchannel systems.