Jul 22, 2023
Ultrahigh evaporative heat transfer measured locally in submicron water films
Scientific Reports volume 12,
Scientific Reports volume 12, Article number: 22353 (2022) Cite this article
795 Accesses
2 Altmetric
Metrics details
Thin film evaporation is a widely-used thermal management solution for micro/nano-devices with high energy densities. Local measurements of the evaporation rate at a liquid-vapor interface, however, are limited. We present a continuous profile of the evaporation heat transfer coefficient (\(h_{\text {evap}}\)) in the submicron thin film region of a water meniscus obtained through local measurements interpreted by a machine learned surrogate of the physical system. Frequency domain thermoreflectance (FDTR), a non-contact laser-based method with micrometer lateral resolution, is used to induce and measure the meniscus evaporation. A neural network is then trained using finite element simulations to extract the \(h_{\text {evap}}\) profile from the FDTR data. For a substrate superheat of 20 K, the maximum \(h_{\text {evap}}\) is \(1.0_{-0.3}^{+0.5}\) MW/\(\text {m}^2\)-K at a film thickness of \(15_{-3}^{+29}\) nm. This ultrahigh \(h_{\text {evap}}\) value is two orders of magnitude larger than the heat transfer coefficient for single-phase forced convection or evaporation from a bulk liquid. Under the assumption of constant wall temperature, our profiles of \(h_{\text {evap}}\) and meniscus thickness suggest that 62% of the heat transfer comes from the region lying 0.1–1 μm from the meniscus edge, whereas just 29% comes from the next 100 μm.
Spatial resolution of amplified evaporation rates in nanometer- and micrometer-thick liquid films, as are found in menisci, is a long-standing challenge1,2,3,4. Accurate measurements require submicron lateral precision and a modeling framework to interpret the results. Experimental measurements have probed evaporation in the macroscopic extended meniscus, where the evaporation heat transfer coefficient takes on its bulk value of 0.001–0.1 MW/\(\text {m}^2\)-K5,6,7. Theory intriguingly suggests an up to three orders-of-magnitude enhancement of the evaporation rate, and hence the heat transfer rate, in the thin film region of the meniscus, but these predictions have not yet been validated8,9,10,11,12,13.
The evaporation rate from a thin liquid film is controlled by a competition between the film thermal resistance and a suppressed liquid pressure. The latter results from the disjoining pressure \(P_d\), which measures the interaction strength between the solid substrate and the liquid film. A smaller film thickness: (i) Decreases the thermal resistance, leading to a higher superheat at the liquid-vapor interface, which enhances evaporation, and (ii) Increases \(P_d\), which suppresses evaporation8,9,10,14. These competing effects result in a non-monotonic profile for the evaporation heat transfer rate, as shown schematically in Fig. 1a. Quantifying this profile will reveal pathways to amplify heat transfer in micro/nanostructure thermal solutions used to manage high power density electronics, where single-phase air/liquid cooling cannot match the demand15,16,17,18,19,20,21. The efficiency of solar thermal generation22,23,24 and desalination processes25,26 will also be improved by engineering evaporation in thin liquid films to obtain high mass fluxes.
Experimental thin liquid film evaporation studies are often conducted by extracting the temperature profile along a meniscus on a heated surface. Infrared cameras11,12,13 and thermocouples8,9,10,27 with spatial resolution from 6 μm to 2 mm have been used to measure local temperatures. Reported heat flux and/or temperature profiles demonstrate enhanced heat transfer near the edge of the meniscus (i.e., the three-phase contact line). As an alternative, Höhmann et al.28 used thermochromic liquid crystals (TLCs) with a 1 μm spatial resolution. TLCs, however, suffer from limited life-spans and high measurement uncertainty12,29. Non-contact laser-based methods have also been used to study liquid-vapor phase change. Park et al. used ultrafast pump-probe spectroscopy to study the evaporation of a thin liquid film. They obtained the time-dependent film thickness response to a picosecond pump optical pulse, but did not report an evaporation rate profile30. Time-domain thermoreflectance was used by Mehrvand and Putnam to study microlayer evaporation in single bubbles during flow boiling of water.4 More recently, Che et al. combined time-domain thermoreflectance and numerical analysis to study the evaporation of an octane liquid film31. They report the variation of the overall heat transfer coefficient along the meniscus, obtaining a maximum value of 0.44 MW/\(\text {m}^2\)-K. This value includes the conductive thermal resistance of the liquid. Because Che et al. average over a 10 μm laser spot diameter, their overall heat transfer coefficient profile cannot resolve values that are less than 2 μm from the meniscus edge. Despite these advances, isolation of the evaporative heat transfer coefficient with microscale resolution across the entire meniscus has not been obtained experimentally.
The objective of this work is to obtain a continuous profile for the evaporation heat transfer coefficient, \(h_\text {evap}\), for a thin water film using experimental measurements interpreted by a machine learned surrogate of the physical system. As shown in Fig. 1a, \(h_{\text {evap}}\) at a position \(\delta\) measured from the start of the meniscus is defined as the ratio of the evaporation heat flux \(q_{\text {evap}}\) to the temperature difference between the interface and the vapor (\(T_{lv} - T_v\))2,32. Water enclosed in a vertically-mounted cuvette forms a meniscus on its walls. At an ambient temperature of 295 K, the experiments are conducted using frequency domain thermoreflectance (FDTR), a non-contact laser-based method. A data-driven framework that uses finite element simulations to train a neural network is developed to extract the \(h_\text {evap}\) and meniscus thickness profiles from the experimental results. At a substrate temperature increase of 20 K, \(h_{\text {evap}}\) peaks at \(1.0_{-0.3}^{+0.5}\) MW/\(\text {m}^2\)-K at a location \(89_{-38}^{+88}\) nm from the edge of the meniscus, where the film thickness is \(15_{-3}^{+29}\) nm. The evaporation heat transfer coefficient decreases to the set bulk value of 0.01 MW/\(\text {m}^2\)-K at a location \(0.92_{-0.56}^{+0.56}\) μm from the edge of the meniscus, where the film thickness is \(1.6_{-1.4}^{+1.8}\) μm. The peak value is two orders of magnitude larger than the heat transfer coefficient for single-phase liquid forced convection7.
FDTR is an experimental approach primarily used to measure the thermal conductivity of solid thin films and the thermal conductances of the interfaces between solids33,34,35. The measurement is achieved by analyzing the phase lag between the periodic heat flux deposited by a periodically modulated pump laser and the induced periodic surface temperature oscillation measured by a coaxial probe laser. This phase lag is typically fit to an analytical solution to the heat diffusion equation36, from which the unknown properties are extracted. See "Materials and methods" for FDTR details.
We apply FDTR to an evaporating meniscus to extract \(h_{\text {evap}}\) and thickness profiles. The schematic in Fig. 1a illustrates the setup of a quartz slide with a 200 nm spin coated polymethyl methacrylate (PMMA) layer, onto which a 70 nm gold film is sputtered. Deionized water is the working fluid. To contain the liquid and form the meniscus, a cuvette is sealed onto the slide through a custom-built frame by applying mechanical pressure. The gold film serves as the transducer needed to absorb the pump laser and to provide a temperature dependent reflectance (i.e. thermoreflectance) for the FDTR measurements33. The two incoming lasers, which have an effective spot diameter of approximately 3.4 μm, transmit coaxially through the quartz lid from the back side and deposit heat into the gold film, which leads to evaporation of the water meniscus. The average power absorbed by the gold film is 600 μW and a superheat of around 20 K is achieved. The PMMA layer serves two purposes: (i) To channel heat to the meniscus by acting as a thermal insulation barrier on the quartz side, which enhances the sensitivity of the measured phase lag to \(h_{\text {evap}}\), and (ii) To improve the adhesion of the gold to the quartz and prevent its delamination under external effects, such as environmental humidity and thermo-mechanical stresses37,38. See "Materials and methods" for fabrication details.
(a) Rendering of the experimental setup. The pump laser is incident from the back of the quartz slide and transmits through the PMMA layer to periodically heat the gold layer. The laser power has a Gaussian profile in the radial direction. The dashed line in the right panel illustrates the expected shape of the \(h_{\text {evap}}\) profile32,39. (b) Static meniscus shape calculated from the augmented Young-Laplace equation without retardation effects. The red dashed curve is a parabolic fit.
The stage is moved vertically across the lasers in increments of \({1}\) μm, starting from the bulk liquid, across the meniscus, and into the bulk vapor. At a given position, four laser modulation frequencies (100 kHz, 178 kHz, 316 kHz and 562 kHz) are probed. The maximum frequency is set by the thermal penetration depth, which decreases as frequency increases [Eq. (S6)], making FDTR less sensitive to the liquid-vapor interface. The four subplots of Fig. 2b, each of which corresponds to one frequency, show the measured phase lag (plotted as black circles) as a function of position along the meniscus. While one scan is shown in Fig. 2b, six scans taken on different days are used in the analysis that follows. See "Materials and methods" for details on how the meniscus was located.
The standard analytical FDTR model assumes uniform layer thicknesses within its framework of the cylindrical heat diffusion equation40,41. As shown in Fig. 1b, a calculation based on the augmented Young–Laplace equation estimates a change in film thickness of up to 5 μm within the laser spot diameter of 3.4 μm. The uniform thickness FDTR model can therefore not be used to extract the \(h_{\text {evap}}\) and film thickness profiles from the phase lag data. See Sect. S3 for information on the theoretical meniscus shape calculation.
We developed a data-driven modeling framework to extract the \(h_{\text {evap}}\) and film thickness profiles from the FDTR measurements. The three-step workflow is shown in Fig. 2. In Step 1, finite element simulations representative of the experimental setup are performed. In Step 2, the known inputs (including parameters related to the laser, meniscus, and materials) and output (the phase lag) from the finite element simulations are used to train a neural network. In Step 3, FDTR results are fitted to the trained neural network to obtain the optimized neural network inputs, which include the \(h_{\text {evap}}\) and film thickness profiles.
A multilayer feedforward neural network, as an universal function approximator, provides a high degree of flexibility to investigate a rich input parameter space42. To construct the neural network, eight input features [Fig. 2c], were selected to describe the FDTR experiment. Four have small uncertainties: (i) the laser frequency f, (ii) the laser spot radius r, (iii) the laser position relative to the starting point of the meniscus \(\Delta _\text {laser}\), and (iv) the thermal conductivity of the PMMA layer \(k_\text {PMMA}\). The \(h_{\text {evap}}\) profile is approximated as three straight lines with three features: (v) The peak value \(h_{\text {evap,peak}}\), achieved at (vi) a film thickness of \(t_{\text {peak}}\), and (vii) the film thickness \(t_{\text {end}}\) where \(h_{\text {evap}}\) reduces to the set bulk value of 0.01 MW/\(\text {m}^2\)-K43,44,45. The meniscus thickness profile is described as \(t(\delta )=c\delta ^2\), where (viii) c is the interface shape coefficient. When the predicted static meniscus profile is fitted to this functional form, a correlation coefficient of 0.9997 is achieved with \(c=0.42\) μ\(\mathrm{m}^{-1}\), as shown in Fig. 1b. The value of c is expected to increase when the meniscus is heated32.
The three-step modeling framework includes finite element simulations, neural network training, and neural network fitting. (a) Finite element simulation of the temperature distribution in an FDTR experiment. The temperature increase is caused by a periodic heating and the phase lag is extracted from the spatially weighted gold temperature. (b) FDTR phase lag along the scanned region of the meniscus at four frequencies. The black circles are the measurements, obtained by scanning from the bulk liquid to the bulk vapor. The beige curves show the predictions from the trained neural network that minimize the mean square error. (c) The eight input features [(i)–(viii)] for the neural network.
Finite element simulations that mimic the FDTR experiments, based in the transient mechanical module in ANSYS Workbench46, are used to train the neural network. The pump laser is modeled as a periodic heat flux at the gold/PMMA interface at a specified frequency with a Gaussian radial profile. The temperatures of the nodes at the gold/PMMA interface are extracted and weighted spatially by the probe laser profile to calculate the average temperature at any instant in time. The phase lag is calculated as the phase difference of the applied heat flux and this average temperature. This approach is validated by comparing the calculated phase lag against the analytical prediction in bulk liquid and bulk vapor systems. See "Materials and methods" and Fig. S5 for details on the finite element simulations and their validation. Evaporation from the meniscus is modelled as a convection boundary condition at the liquid-vapor interface specified by the \(h_{\text {evap}}\) profile. Fluid dynamics within the meniscus are not considered. Prior studies have shown that the effects of inertial forces47, thermocapillary convection48, and thermal resistance at the interface49,50 on the fluid flow and heat transfer in the evaporating thin film are small. As we will show, our model's predictions are in good agreement with the experimental results, indicating that fluid dynamics can be neglected in this analysis.
By varying the inputs (i)-(viii) over a rich parameter space, 2653 distinct evaporating meniscus systems are simulated by the finite element simulations (Step 1 in Fig. 2). The parameter space construction is discussed in Sect. S4. The finite element simulation results are used to train the neural network (Step 2 in Fig. 2). A three-layer neural network with hidden layer sizes of 12 and 15 is selected through the random search technique. See "Materials and methods" for details about the neural network structure. The correlation coefficient for the neural network prediction of the validation data is \(0.9990\pm 0.0001\).
Once trained, the neural network becomes a surrogate for the finite element simulation and is used to fit the FDTR experimental data (Step 3 in Fig. 2). For each of the six data sets, the phase lags at the four frequencies are fit simultaneously. The Powell optimization technique is used to extract the feature values that minimize the mean square error (MSE) between the neural network-predicted and FDTR-measured phase lags. During the optimization process, frequency is fixed as a known parameter taken directly from the experiment. \(k_{\text {PMMA}}\) is set to the literature value of \(0.240\pm 0.005\) W/m-K51 with its uncertainty incorporated in the analysis by a uniform grid search over the reported range. Six features (r, \(\Delta _\text {laser}\), \(h_\text {evap,peak}\), \(t_\text {peak}\), \(t_\text {end}\) and c) are thus extracted from the Powell optimization. More information on the Powell optimization is provided in Sect. S6.
An example of the neural network phase lag prediction from Step 3 for one of the data sets is plotted in Fig. 2b as solid beige lines. The neural network predictions capture the FDTR data well at all frequencies and scan positions.
Neural networks of the same structure but implemented with different algorithmic factors yield different predictions52,53. Here, the effect of uncertainty in algorithmic factors is evaluated by changing the random seeds for: (i) Splitting the dataset into training and testing subsets,52 and (ii) Initializing the weights53. One thousand neural networks with different seeds are created. Each of the one thousand neural networks is then used to fit each of the six FDTR datasets through the Powell optimization. The set of six features resulting in the lowest MSE is extracted for each data set/neural network pair (i.e., we obtain six thousand values for the six features).
Of the extracted features describing the shape of the meniscus and the continuous \(h_\text {evap}\) profile [features (v)–(viii) in Fig. 2c], the neural network is most sensitive to \(h_{\text {evap,peak}}\) and the coefficient c, as shown in Fig. S8b. To analyze the variation in the obtained values, we plotted histograms of \(h_{\text {evap,peak}}\) and c for each of the six data sets (Fig. S9) and investigated their similarities by the two-sample Kolmogorov-Smirnov null hypothesis test (Sect. S8). The results do not prove that the distributions are drawn from the same population. Potential sources of the differences are variations in the lab environment and scan positions because the six FDTR data sets were taken on six different days. That being said, the median value of each histogram is within the 10th and 90th percentiles of the other histograms for both features. To illustrate the data spread, the six histograms are combined to obtain the distributions for c and \(h_{\text {evap,peak}}\) plotted in Fig. 3a and b. The distributions are non-Gaussian. In subsequent discussion, the 10th and 90th percentile values are used to quantify the uncertainty. Using these statistics, the \(h_{\text {evap}}\) and thickness profiles are plotted in Fig. 3c. For the \(h_{\text {evap}}\) profiles, the median values of \(t_{\text {peak}}\), \(t_{\text {end}}\), and c are used.
Distributions of (a) interface shape coefficient c and (b) \(h_{\text {evap,peak}}\) from the neural network run with randomized seeds. The dashed vertical lines show the 10th percentile, median value, and 90th percentile. (c) \(h_{\text {evap}}\) and meniscus shape profiles. The solid lines indicates the profiles using the median values. The dashed lines and dashed dotted lines indicate the profiles using the 10th and 90th percentile values.
Wang et al. theoretically studied changes in an octane meniscus shape for different substrate temperatures32. For a 20 K increase in the substrate temperature, they found the interface shape coefficient c to increase by a factor of eight compared to the static meniscus (i.e., no superheat). Here, c for the static meniscus is 0.42 μ\(\text {m}^{-1}\) [Fig. 1b]. For the calculated superheat of 20 K, the c extracted from our modeling framework is \(1.9_{-0.9}^{+1.6}\) μ\(\text {m}^{-1}\), which is approximately four times larger than the static value.
As shown in Fig. 3c, at single nanometer film thicknesses, high \(P_d\) suppresses the evaporation. As the thickness increases, \(P_d\) decreases and the evaporation rate increases. \(h_{\text {evap}}\) reaches its largest value of \(1.0_{-0.3}^{+0.5}\) MW/\(\text {m}^2\)-K at \(t_{\text {peak}}=15_{-3}^{+29}\) nm, which is \(89_{-38}^{+88}\) nm from the meniscus edge. Further increasing the liquid film thickness increases the liquid conduction resistance. \(h_{\text {evap}}\) thus decreases and reaches the set bulk value of 0.01 MW/\(\text {m}^2\)-K at \(t_{\text {end}}=1.6_{-1.4}^{+1.8}\) μm, which is \(0.92_{-0.56}^{+0.56}\) μm from the meniscus edge. Reported values from theoretical calculations of \(h_{\text {evap,peak}}\) are between 0.8 and 8 MW/\(\text {m}^2\)-K for superheats between 3 and 50 K39,54,55,56. We note that the heating in an FDTR experiment is applied locally with a laser, resulting in a temperature distribution at the point of incidence, in contrast to the constant wall temperature in most past analyses.
We assess our results by performing two calculations of the mass flux (\(\dot{m}\)) profile. First, from the Hertz-Kundsen-Schrage (HKS) relation1,14,
where \(T_{lv}\) is the spatially-dependent liquid-vapor temperature, \(\alpha\) is the mass accommodation coefficient at the liquid-vapor interface, M is the molar mass of the liquid, R is the universal gas constant, \(P_v\) is the pressure in the vapor, and \(P_{eq}\) is the so-called equilibrium pressure, which is the modified saturation vapor pressure under the influence of disjoining pressure and capillary pressure. Second, from a mass-energy balance as
where \(h_{fg}\) is the latent heat of vaporization. See Sect. S9 for details about the calculations from Eqs. (1) and (2).
We applied a temperature-dependent \(\alpha\) profile, with \(\alpha = 0.995\) at our peak mass flux, from the molecular dynamics simulations of Chandra and Keblinski57. The resulting peak value of \(\dot{m}''_{\text {evap,HKS}}\) is 40% higher than the peak value of \(\dot{m}''_{\text {evap,direct}}\), as shown in Fig. S10. Consistent with this result, Chandra and Keblinski found \(\dot{m}''_{\text {evap,HKS}}\) to be 6-19% larger than that extracted directly from their self-consistent simulations57. They showed that their difference is due to a non-zero averaged vapor molecule velocity tangential to the interface, which violates a key assumption of the HKS relation. Given the spatial profile and lateral temperature gradients of the liquid-vapor interface, this assumption is likely violated in our experiments. An additional source of the discrepancy between our \(\dot{m}''_{\text {evap,HKS}}\) and \(\dot{m}''_{\text {evap,direct}}\) values is challenges in applying the HKS relation, especially the proper choice of \(\alpha\)58. For water, reported values span over four orders of magnitude, from \(10^{-4}\) to 11,59. A temperature-independent \(\alpha\) of 0.8 makes the peak values of \(\dot{m}''_{\text {evap,HKS}}\) and \(\dot{m}''_{\text {evap,direct}}\) match.
Using the obtained \(h_{\text {evap}}\) and thickness profiles, heat transfer in the meniscus under the assumption of a constant wall temperature can be studied. The spatially-dependent thermal conductance \(G(\delta )\) between the wall and the vapor is calculated as
where \(k_{\text {water}}\) is the water thermal conductivity (0.6 W/m-K)60. The thermal conductance profile across the meniscus is shown as the black line in Fig. 4. The normalized cumulative heat transfer across the meniscus is calculated as the cumulative area under the thermal conductance profile, shown as the blue line in Fig. 4. 62% of the heat transfer comes from the region 0.1–1 μm from the meniscus edge, while only 29% comes from the next 100 μm.
Thermal conductance profile across the meniscus (black line). The blue line shows the normalized cumulative heat transfer. The two vertical dashed lines mark the positions where the distances to the meniscus edge are 0.1 μm and 1 μm. The two horizontal dashed lines show the corresponding normalized cumulative heat transfer percentage, which are 9% and 71%.
High spatial resolution thermoreflectance measurements and a finite element simulation/neural network interpretation of the results enabled us to obtain a continuous \(h_{\text {evap}}\) profile in the thin film region of a water meniscus. The peak \(h_{\text {evap}}\) is within the range of reported theoretical and experimental results and definitively quantifies competing effects from liquid conduction resistance and disjoining pressure. Under the assumption of constant wall temperature, our profiles of \(h_{\text {evap}}\) and meniscus thickness define a target length scale of 0.1–1 μm for the design of surface features and pore sizes to enhance evaporation heat transfer rates. The framework we developed for analyzing FDTR data through training a neural network with representative computational results could be applied to interpret other thermal transport measurements that do not lend themselves to analytical solutions.
In FDTR, the transducer layer (here, gold) is heated by a 488 nm continuous-wave pump laser (Coherent) that is intensity-modulated over a frequency range of 100-562 kHz. The periodic heating generates a periodic change in the surface temperature of the gold layer, with a phase lag relative to the heat flux that depends on the properties of the sample. Thermoreflectance of the gold layer causes modulation of a reflected 532 nm probe laser (Coherent). Its phase lag relative to the pump is monitored by a lock-in amplifier (Zurich Instruments Model HF2LI). A double-modulation scheme is implemented where a mechanical chopper is added in the probe path to remove coherent noise from the signal. The average power absorbed by the transducer layer is 600 μW, which was measured with a digital handheld optical power console (Thorlabs). From this value, the steady-state temperature rise of the gold is estimated to be 20 K with a periodic temperature amplitude that ranges from 6-13 K depending on the modulation frequency.
The water is confined by a demountable cuvette with a 5 mm light path purchased from Starna Cells, Inc. (Type 49). To ensure a clean environment for the evaporation experiment, the cuvette was cleaned with a nanostrip after purchase and filled with water in a class 100 cleanroom. The quartz substrate is from VWR Corp. (Grade GE124).
The PMMA layer is prepared by spin coating 495 PMMA onto the quartz slide, which is taped onto a 6 inch silicon wafer. The sample is first spun at 500 RPM for 6 s then accelerated to 4000 RPM for 60 s to achieve a thickness of \(200\pm 4\) nm measured by profilometry. The coated slide is baked in vacuum at a temperature of 180 \(^\circ\)C for 120 s to cure. The transmittance of the PMMA-coated quartz slide is calculated at the FDTR wavelengths (pump: 488 nm, probe: 532 nm) using the Fresnel equation61. The reflection loss at the interface between air and the quartz slide is 4%, and that between the quartz and PMMA is below 1%, with negligible absorption in quartz and PMMA at the FDTR wavelengths62,63.
The gold layer is sputtered onto the fully-cured PMMA using the Perkin Elmer 6J Sputtering System. The thickness of the gold layer is measured to be \(70\pm 1\) nm using profilometry. The typical RMS roughness of the sputtered gold from this system is 2 nm as measured by X-ray reflectivity. The electrical conductivity of the gold layer is measured using a four-point probe and used to calculate its thermal conductivity by the Wiedemann-Franz law.
Before assembling the sample, the quartz slide and cuvette are both cleaned with isopropyl alcohol and rinsed by deionized water in a class 100 cleanroom. To prevent any change in the water content, proper sealing between the cuvette and the quartz substrate is essential. To improve the seal, a high-temperature silicone gasket cut into the shape of the cuvette edge is placed between the cuvette wall and the quartz slide. The gasket is placed in compression by a 3D printed nylon 12 frame, as shown in Fig. S1. The setup is placed at room temperature for a 48-hour period before measurement to degas the water and ensure that there is no water level change. One measurement scan takes three to five hours.
During the FDTR experiment, the meniscus is located by moving the sample vertically downwards so that the fixed lasers scan from the bulk liquid to the bulk vapor, as shown in Fig. S2a. The total scan distance is 20 μm and the distance between each data point is 1 μm. In Fig. S2b, the abrupt change in signal locates the transition region.
A temporally-periodic heat flux with a Gaussian profile in the radial direction is set to each node on the gold surface (the interface with PMMA) using the ANSYS Parametric Design Language (APDL) to simulate the FDTR pump laser. Because FDTR considers the phase lag between the heat flux and the surface temperature (gold surface), only the AC component of the laser heat flux is applied. After the simulation, the time series of temperatures for each node on the gold layer are output through APDL commands.
Evaporation is set as the convection boundary on the meniscus surface using the \(h_{\text {evap}}\) profile. Even with the bulk value of 0.01 MW/m\(^2\)-K, the heat transfer by the phase change process is three orders of magnitude larger than that by natural convection7. Neglecting mass transfer and fluid dynamics simplifies the model, which enables exploration of fine increments in the parametric space. More information on the finite element simulation meniscus setup and validation is provided in Sect. S4.
A three-layer neural network with two hidden layers (12, 15) is selected through the random search technique using the Scikit-Learn package64. The logistic activation function has the best performance in terms of accuracy and training time. The limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm is used as the optimization solver. The max iteration is set to 10,000 and the tolerance is set to 0.001. More information on the random search is provided in Sect. S5.
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.
Persad, A. H. & Ward, C. A. Expressions for the evaporation and condensation coefficients in the hertz-knudsen relation. Chem. Rev. 116, 7727–7767 (2016).
Article CAS Google Scholar
Wayner, P. Jr., Kao, Y. & LaCroix, L. The interline heat-transfer coefficient of an evaporating wetting film. Int. J. Heat Mass Transf. 19, 487–492 (1976).
Article Google Scholar
Lu, Z. et al. An ultrathin nanoporous membrane evaporator. Nano Lett. 17, 6217–6220 (2017).
Article ADS CAS Google Scholar
Mehrvand, M. & Putnam, S. A. Transient and local two-phase heat transport at macro-scales to nano-scales. Commun. Phy. 1, 1–13 (2018).
Google Scholar
Hallinan, K., Chebaro, H., Kim, S. & Chang, W. Evaporation from an extended meniscus for nonisothermal interfacial conditions. J. Thermophys. Heat Transfer 8, 709–716 (1994).
Article ADS CAS Google Scholar
Park, K., Noh, K.-J. & Lee, K.-S. Transport phenomena in the thin-film region of a micro-channel. Int. J. Heat Mass Transf. 46, 2381–2388 (2003).
Article MATH Google Scholar
Incropera, F. P. et al. Fundamentals of Heat and Mass Transfer Vol. 6 (Wiley, 1996).
Google Scholar
Wayner, P. Jr. & Coccio, C. Heat and mass transfer in the vicinity of the triple interline of a meniscus. AIChE J. 17, 569–574 (1971).
Article CAS Google Scholar
Panchamgam, S. S., Chatterjee, A., Plawsky, J. L. & Wayner, P. C. Jr. Comprehensive experimental and theoretical study of fluid flow and heat transfer in a microscopic evaporating meniscus in a miniature heat exchanger. Int. J. Heat Mass Transf. 51, 5368–5379 (2008).
Article MATH Google Scholar
Kazemi, M. A., Nobes, D. S. & Elliott, J. A. Experimental and numerical study of the evaporation of water at low pressures. Langmuir 33, 4578–4591 (2017).
Article CAS Google Scholar
Ibrahem, K., Abd Rabbo, M., Gambaryan-Roisman, T. & Stephan, P. Experimental investigation of evaporative heat transfer characteristics at the 3-phase contact line. Exp. Thermal Fluid Sci. 34, 1036–1041 (2010).
Article CAS Google Scholar
Dhavaleswarapu, H. K., Garimella, S. V. & Murthy, J. Y. Microscale temperature measurements near the triple line of an evaporating thin liquid film. J. Heat Transf. 131 (2009).
Buffone, C., Minetti, C., Boussemaere, L., Roudgar, M. & De Coninck, J. Marangoni convection in evaporating meniscus with changing contact angle. Exp. Fluids 55, 1–8 (2014).
Article CAS Google Scholar
Wang, X., Li, Y., Malen, J. A. & McGaughey, A. J. Assessing the impact of disjoining pressure on thin-film evaporation with atomistic simulation and kinetic theory. Appl. Phys. Lett. 116, 213701 (2020).
Article ADS CAS Google Scholar
Pop, E. Energy dissipation and transport in nanoscale devices. Nano Res. 3, 147–169 (2010).
Article CAS Google Scholar
Duan, C., Karnik, R., Lu, M.-C. & Majumdar, A. Evaporation-induced cavitation in nanofluidic channels. Proc. Natl. Acad. Sci. 109, 3688–3693 (2012).
Article ADS CAS Google Scholar
La Mer, V. K. & Healy, T. W. Effects of mechanically produced capillary waves on the evaporation of water through monolayers. Nature 201, 488 (1964).
Article ADS Google Scholar
Xu, N. et al. A water lily-inspired hierarchical design for stable and efficient solar evaporation of high-salinity brine. Sci. Adv. 5, eaaw7013 (2019).
Article ADS CAS Google Scholar
Li, Y., Alibakhshi, M. A., Zhao, Y. & Duan, C. Exploring ultimate water capillary evaporation in nanoscale conduits. Nano Lett. 17, 4813–4819 (2017).
Article ADS CAS Google Scholar
Farokhnia, N., Irajizad, P., Sajadi, S. M. & Ghasemi, H. Rational micro/nanostructuring for thin-film evaporation. J. Phys. Chem. C 120, 8742–8750 (2016).
Article CAS Google Scholar
Nazari, M. et al. Ultrahigh evaporative heat fluxes in nanoconfined geometries. Langmuir 35, 78–85 (2018).
Article Google Scholar
Tao, P. et al. Solar-driven interfacial evaporation. Nat. Energy 3, 1031–1041 (2018).
Article ADS Google Scholar
Ghasemi, H. et al. Solar steam generation by heat localization. Nat. Commun. 5, 1–7 (2014).
Article Google Scholar
Luo, X., Huang, C., Liu, S. & Zhong, J. High performance of carbon-particle/bulk-wood bi-layer system for solar steam generation. Int. J. Energy Res. 42, 4830–4839 (2018).
Article Google Scholar
Humplik, T. et al. Nanostructured materials for water desalination. Nanotechnology 22, 292001 (2011).
Article CAS Google Scholar
Culp, T. E. et al. Electron tomography reveals details of the internal microstructure of desalination membranes. Proc. Natl. Acad. Sci. 115, 8694–8699 (2018).
Article ADS CAS Google Scholar
Renk, F. & Wayner, P. An evaporating ethanol meniscus-part i: Experimental studies. J. Heat Transfer 101, 55–58 (1979).
Article CAS Google Scholar
Höhmann, C. & Stephan, P. Microscale temperature measurement at an evaporating liquid meniscus. Exp. Thermal Fluid Sci. 26, 157–162 (2002).
Article Google Scholar
Migliaccio, C. P., Dhavaleswarapu, H. K. & Garimella, S. V. Temperature measurements near the contact line of an evaporating meniscus v-groove. Int. J. Heat Mass Transf. 54, 1520–1526 (2011).
Article CAS Google Scholar
Park, J., Xie, X., Li, D. & Cahill, D. G. Plasmonic sensing of ultrafast evaporation and condensation. Nanoscale Microscale Thermophys. Eng. 21, 70–80 (2017).
Article ADS CAS Google Scholar
Che, Z., Wang, T., Sun, F. & Jiang, Y. Research on heat transfer capability of liquid film in three-phase contact line area. Int. J. Heat Mass Transf. 195, 123158 (2022).
Article CAS Google Scholar
Wang, H., Garimella, S. V. & Murthy, J. Y. Characteristics of an evaporating thin film in a microchannel. Int. J. Heat Mass Transf. 50, 3933–3942 (2007).
Article MATH Google Scholar
Ong, W.-L., Rupich, S. M., Talapin, D. V., McGaughey, A. J. & Malen, J. A. Surface chemistry mediates thermal transport in three-dimensional nanocrystal arrays. Nat. Mater. 12, 410–415 (2013).
Article ADS CAS Google Scholar
Regner, K. T. et al. Broadband phonon mean free path contributions to thermal conductivity measured using frequency domain thermoreflectance. Nat. Commun. 4, 1–7 (2013).
Article Google Scholar
Schmidt, A. J., Cheaito, R. & Chiesa, M. A frequency-domain thermoreflectance method for the characterization of thermal properties. Rev. Sci. Instrum. 80, 094901 (2009).
Article ADS Google Scholar
Cahill, D. G. Analysis of heat flow in layered structures for time-domain thermoreflectance. Rev. Sci. Instrum. 75, 5119–5122 (2004).
Article ADS CAS Google Scholar
Kim, D. W. & Kong, B.-S. The effect of hygro-mechanical and thermo-mechanical stress on delamination of gold bump. Microelectron. Reliab. 46, 1087–1094 (2006).
Article CAS Google Scholar
Majeed, B., Paul, I., Razeeb, K. M., Barton, J. & O’Mathuna, S. C. Effect of gold stud bump topology on reliability of flip chip on flex interconnects. IEEE Trans. Adv. Packag. 30, 605–615 (2007).
Article Google Scholar
Zhao, J.-J., Duan, Y.-Y., Wang, X.-D. & Wang, B.-X. Effect of nanofluids on thin film evaporation in microchannels. J. Nanopart. Res. 13, 5033–5047 (2011).
Article ADS CAS Google Scholar
Schmidt, A. J. Pump-probe thermoreflectance. Annu. Rev. Heat Transf. 16 (2013).
Wang, X., Jeong, M., McGaughey, A. J. & Malen, J. A. Reducing the uncertainty caused by the laser spot radius in frequency-domain thermoreflectance measurements of thermal properties. Rev. Sci. Instrum. 93, 023001 (2022).
Article ADS CAS Google Scholar
Hornik, K., Stinchcombe, M. & White, H. Multilayer feedforward networks are universal approximators. Neural Netw. 2, 359–366 (1989).
Article MATH Google Scholar
Lemmon, E., Huber, M. & McLinden, M. Nist standard reference database 23: Reference fluid thermodynamic and transport properties-refprop, version 8.0 (2007).
Nishikawara, M. & Nagano, H. Optimization of wick shape in a loop heat pipe for high heat transfer. Int. J. Heat Mass Transf. 104, 1083–1089 (2017).
Article Google Scholar
Hodes, M., Steigerwalt Lam, L., Cowley, A., Enright, R. & MacLachlan, S. Effect of evaporation and condensation at menisci on apparent thermal slip. J. Heat Transf. 137, 071502 (2015).
Article Google Scholar
Ansys® academic research mechanical, release 18.1 (2017).
Ma, H., Cheng, P., Borgmeyer, B. & Wang, Y. Fluid flow and heat transfer in the evaporating thin film region. Microfluid. Nanofluid. 4, 237–243 (2008).
Article Google Scholar
Dhavaleswarapu, H. K., Murthy, J. Y. & Garimella, S. V. Numerical investigation of an evaporating meniscus in a channel. Int. J. Heat Mass Transf. 55, 915–924 (2012).
Article MATH Google Scholar
Stephan, P. & Hammer, J. A new model for nucleate boiling heat transfer. Heat Mass Transf. 30, 119–125 (1994).
CAS Google Scholar
Stephan, P. & Busse, C. Analysis of the heat transfer coefficient of grooved heat pipe evaporator walls. Int. J. Heat Mass Transf. 35, 383–391 (1992).
Article CAS Google Scholar
Katz, J. S. et al. Nanostencil fabrication with double exposure optical lithography for scalable resist-free patterning of metal on polymers. In 2018 Transducer Research Foundation Hilton Head Workshop (2018).
Goodman, S. N., Fanelli, D. & Ioannidis, J. P. What does research reproducibility mean?. Sci. Transl. Med. 8, 071502 (2016).
Article Google Scholar
He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770–778 (2016).
Ranjan, R., Murthy, J. Y. & Garimella, S. V. A microscale model for thin-film evaporation in capillary wick structures. Int. J. Heat Mass Transf. 54, 169–179 (2011).
Article CAS MATH Google Scholar
Zhao, J.-J., Duan, Y.-Y., Wang, X.-D. & Wang, B.-X. Effects of superheat and temperature-dependent thermophysical properties on evaporating thin liquid films in microchannels. Int. J. Heat Mass Transf. 54, 1259–1267 (2011).
Article MATH Google Scholar
Lay, J. H. & Dhir, V. K. Shape of a Vapor Stem During Nucleate Boiling of Saturated Liquids. J. Heat Transf. 117, 394–401 (1995).
Article CAS Google Scholar
Chandra, A. & Keblinski, P. Investigating the validity of schrage relationships for water using molecular dynamics simulations. J. Chem. Phys. 153, 124505 (2020).
Article ADS CAS Google Scholar
Ward, C. & Stanga, D. Interfacial conditions during evaporation or condensation of water. Phys. Rev. E 64, 051509 (2001).
Article ADS CAS Google Scholar
Marek, R. & Straub, J. Analysis of the evaporation coefficient and the condensation coefficient of water. Int. J. Heat Mass Transf. 44, 39–53 (2001).
Article CAS MATH Google Scholar
Cho, W., Kwon, S. & Choi, J. The thermal conductivity for granite with various water contents. Eng. Geol. 107, 167–171 (2009).
Article Google Scholar
Joram, C. Transmission curves of plexiglass (pmma) and optical grease (Tech. Rep, 2009).
Malitson, I. H. Interspecimen comparison of the refractive index of fused silica. Josa 55, 1205–1209 (1965).
Article ADS CAS Google Scholar
Sultanova, N., Kasarova, S. & Nikolov, I. Dispersion proper ties of optical polymers. Acta Phys. Polon. Ser. A Gen. Phys. 116, 585 (2009).
Article ADS CAS Google Scholar
Pedregosa, F. et al. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011).
MathSciNet MATH Google Scholar
Download references
We gratefully acknowledge Xiaoyue Zhao, Matthew Bartnof, and Wee-Liat Ong, who helped to develop the experimental setup. This project was supported by the NSF ENG Directorate, CBET Division, Award Number 1804752 and the Natural Sciences and Engineering Research Council of Canada.
These authors contributed equally: Xiaoman Wang and S. Arman Ghaffarizadeh.
Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh, PA, 15213, USA
Xiaoman Wang, S. Arman Ghaffarizadeh, Xiao He, Alan J. H. McGaughey & Jonathan A. Malen
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
X.W. made the sample and ran the experiments; S.A.G. and X.W. designed and assembled the setup; X.W. and S.A.G. designed the finite element method; X.W. performed the finite element simulations; X.H. and X.W. and S.A.G. constructed the machine learning algorithms; X.W., S.A.G. A.J.H.M. and J.A.M. analyzed the data; all authors wrote and edited the manuscript.
Correspondence to Alan J. H. McGaughey or Jonathan A. Malen.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Wang, X., Ghaffarizadeh, S.A., He, X. et al. Ultrahigh evaporative heat transfer measured locally in submicron water films. Sci Rep 12, 22353 (2022). https://doi.org/10.1038/s41598-022-26182-2
Download citation
Received: 18 August 2022
Accepted: 12 December 2022
Published: 26 December 2022
DOI: https://doi.org/10.1038/s41598-022-26182-2
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.