- Email: [email protected]
A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal
Applied Energy xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal Mrinal Jagirdar ⇑, Poh Seng Lee Department of Mechanical Engineering, National University of Singapore, Singapore 117575, Singapore
h i g h l i g h t s Novel flow regime detection technique for micro domains is presented. Method involves converting temporal data of temperature into frequency domain. It is corrected for damping due to thermal diffusivity of solid material. Magnitude of amplitudes were different for single phase, bubbly and slug flow. Applications involve smart feed-back loops for cooling and 3D IC packages.
a r t i c l e
i n f o
Article history: Received 15 August 2015 Received in revised form 25 December 2015 Accepted 26 December 2015 Available online xxxx Keywords: Flow boiling Microchannel Flow-regime detection Temperature transients
a b s t r a c t Flow boiling in microchannels has been receiving a lot of attention in recent years because of its high heat flux removal capabilities at low flow rates and low pumping power. An important aspect of flow-boiling experiments is prediction or detection of the prevalent flow-regime. Currently, most researchers use high-speed camera for flow visualization for regime detection. However, in some cases due to limitations of the experimental setup and test-piece, such as multi-layer cooling of 3D IC stack, this may not be feasible. In this paper, the influence of flow-regime on frequency domain of local temperature data of the wetted surface is studied. Experiments have been performed synchronously with high speed flow visualization on a single microchannel with width and length of 2.54 mm and 25.4 mm respectively. The microchannel heights tested were 0:14; 0:28 and 0:42 mm. De-gassed, de-ionized water was used as the working fluid. Mass fluxes tested ranged from 200 to 1000 kg=ðm2 sÞ. Depending on the prevalent flow regime, some of the highest of peak amplitudes in the frequency domain were quite distinct. Within the bounds of current experimental parameters, it is concluded that local transient temperature data can be a potential diagnostic tool for detection of flow-regimes. (A shorter version of this paper was presented at the 7th International Conference on Applied Energy (ICAE2015), March 28–31, 2015, Abu Dhabi, UAE (Original paper title: ‘‘Temperature transients for detection of flow-regimes in a mini/microchannel” and Paper No.: 430).) Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction With the current state-of-the-art electronics, the challenge is to remove heat fluxes of the order of 300 W=cm2 and this figure will keep rising with further miniaturization [1]. Especially for such applications, two-phase flow in micro-channels has been identified to be a very promising technology since the phase change process facilitates smaller thermal resistances [2] and the pumping power requirement too is quite small [3]. Other applications of microchannel two-phase flow include vapor compression heat ⇑ Corresponding author. Tel.: +65 65164657; fax: +65 67791459. E-mail address: [email protected] (M. Jagirdar).
pumps [4], vapor compression refrigerators [5], etc. An important aspect of microchannel two-phase flow is detection of flow regimes. In general, high-speed camera serves the purpose during experiments. Some researchers have also tried to come up with predictive flow regime maps or techniques to identify flow regimes for various experimental conditions. Revellin and Thome [6] developed a flow pattern map for circular microchannels that had the mass velocity and the vapor quality as co-ordinates. Transition curves were determined using Reynolds number, Weber number, Boiling number, fluid properties and microchannel dimensions. Regimes identified were isolated bubble regime, coalescing bubble regime, the annular zone and the post dry-out zone. Martin-Callizo et al.
http://dx.doi.org/10.1016/j.apenergy.2015.12.111 0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
2
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Nomenclature A c Co Dh f f G H I Il k L l qb qn Ql r S t
wetted surface area (m2 ) specific heat (J/(kg °C)) confinement number hydraulic diameter (m) frequency of temperature signal (Hz) non-dimensional frequency mass flux (kg=ðm2 sÞ) microchannel height (m) current (A) current during heat loss test (A) thermal conductivity (W=ðm KÞ) thickness of the solid silicon substrate (m) length of the microchannel (m) heat flux entering the solid substrate (W=m2 ) heat flux at time step ‘n’ at the wetted surface (W=m2 ) heat loss (W) number of future time-steps minimization function time (s)
[7] noted that larger inlet sub-cooling shifted all transition lines (except for Slug-annular/Churn to annular flow) to earlier vapor qualities. Moreover, an increase in the saturation temperature shifted all transition boundaries towards higher vapor qualities. A flow regime map was also proposed by Harirchian and Garimella [8]. Fluid properties, channel diameter, heat flux, mass flux as well as the heated length of the microchannels were considered and four regions namely slug, confined annular, bubbly, and alternating Churn/Annular/Wispy-annularflow were plotted. It may hence be argued that there are a number of variables affecting flowregimes and it seems to be a very difficult task to develop a regime prediction method that is generally applicable. Moreover, modifications made to the wetted surface characteristics or the geometry of microchannel fin structure can add to the difficulty in development of a generic prediction method for identification of flow regime. It was shown by Alam et al. [9] that transition from single phase to bubbly flow can be altered by changing the surface roughness. Even artificial nucleation sites (ANS) as shown by some researchers [10–13] can alter the flow-regime and heat transfer mechanism. Yang et al. [14] showed that by having superhydrophilic silicon nano-wires on the inner-walls of the microchannel, heat transfer can be very drastically enhanced by manipulation of the prevalent flow regime. Some researchers have also developed novel microchannel designs that can change heat transfer performance by changing prevalent regimes under certain experimental conditions [3,15,16]. Thus, development of a diagnostic tool for flow-regime detection seems to be relatively simple compared to the development of a generally applicable regime prediction method. Although, a diagnostic technique cannot fully replace flow-regime prediction techniques, it can still be of utility for important applications. Smart feed-back loops that control flow-rate to control the prevalent flow-regime for optimal heat transfer performance and minimum pressure drop can potentially use regime detection techniques. In fact, having noted that two-phase cooling can substantially reduce the cooling energy compared to air and water based cooling, Marcinichen et al. [17] showed the development of control strategies for multi-microchannel evaporators. Also, Marcinichen et al. [18] have pointed out that with the implementation of simple and cheap controllers, it is possible to ensure that on-chip cooling is optimal all the time even for transient loads.
T0 Ti ts Tw T sensor V Vl W x Yi
initial temperature of solid substrate ( C) temperature at time step i ( C) sensor depth (m) wetted surface temperature ( C) temperature measured by the sensor ( C) voltage (V) voltage during heat loss test (V) microchannel width (m) spatial co-ordinate along depth of solid substrate (m) measured temperature at time step i ( C)
Greek symbols a thermal diffusivity of substrate (m2 =s) DT temperature difference ( C) sensitivity coefficient at time step i at the sensor loca/i tion ( C m2 =W) q density (kg=m3 )
Another potential application for regime detection technique is for 3D IC packaging technology. Development in 3D chip packaging is being driven to overcome the limitations due to the physical distance between cores and memory as the computational performance of HPC systems continues to increase [19]. Thus, researchers have also shown interest towards assessment of cooling solutions for modular thermal management of 3D-ICs consisting of multiple inter-layer microchannels. Kim et al. [20] and Koo et al. [21] carried out numerical and theoretical investigations respectively, to explore 3D IC cooling. Since a high speed camera cannot be used for regime detection in internal microchannels, a diagnostic technique can be very useful. To the best of our knowledge, only Revellin et al. [22] developed a regime detection technique. Their method involved a novel optical technique that could characterize flow pattern transition in two-phase flows. Bubble frequency, percentage of surviving small bubbles, lengths of bubbles and flow pattern transitions could also be determined. In this paper, transients in local, wall temperature are investigated with synchronized flow-visualization to see if flow regimes have a unique temperature signature and whether it can be used for regime-detection without the need for optical access.
2. Experimental setup and procedure 2.1. Test section An exploded view of the test-section is as shown in Fig. 1 and a photograph of the assembled micro-channel test section is as shown in Fig. 2(a). The silicon test-chip (from Kokomo Semiconductors) is flip-chip packaged onto the PCB shown in Fig. 2(b). The Pyrex glass (shown in Fig. 1) insert is bonded to the Polycarbonate top-cover cavity and it provides good optical transparency required for flow-visualizations. A gap formed between the surface of the silicon shown in Fig. 2(b) and the Pyrex glass insert is a low aspect ratio micro-channel. Gasket sheet between the Polycarbonate top-cover and the PCB acts as a sealant. The inlet and outlet plenums in the polycarbonate top cover, as seen in Fig. 2(a), have fittings above them so that by slightly loosening them, the bubbles can be bled out. Two ports
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
3
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 1. Exploded view of the assembled test-section.
Fig. 2. (a) Assembled test-section. (b) Silicon test chip (wetted surface) on the PCB. (c) 1 10 array of thermal test dies behind the wetted surface of the silicon test chip, 630 lm below.
(see Fig. 2(a)) are provided through which RTD probes can be fitted to measure the fluid temperature at the inlet and outlet plenums. Similarly there are two more ports coming out of the other end of the plenums connected to the Pressure transducers. The Teflon bottom cover has an air–gap so that it is better insulated leading to smaller heat loss from the bottom. The microchannel width and length are 2540 lm and 25400 lm respectively. The currently used silicon test-chip consists of an array of 10 1 thermal test dies (see Fig. 2(c)) 630 lm below the top (wetted) surface. Each of the thermal test dies has an area of 2540 lm 2540 lm and each of them consists of a heater and a diode. Each heater covers an area of 2000 lm 2000 lm and each doped diode temperature sensor covers an area of 400 lm 400 lm at the center of the die. Each temperature sensor is a series connection of 5 diodes (p–n junction) with a total sensitivity of 10 m V= C and temperature is measured with an uncertainty of 0:4 C. The test-dies are the same as those used by Alam et al. [23], Harirchian and Garimella [24] and Lee and Garimella [25]. Fig. 3. Schematic of the flow loop.
2.2. Flow loop The flow loop schematic is as shown in Fig. 3. The test-section is mounted on a X–Y stage so that it is easy to adjust the field of view of the camera. The temperature at the inlet and outlet plenums of the test-section are measured using RTDs (Omega 1/10 DIN class) with an uncertainty of 0:12 C. The reservoir is filled with de-ionized water. Immersion heaters (2 600 W) fitted into the reservoir are used to boil the water for degassing. The gear pump (Micropump driven by Cole-parmer gear
pump drive) drives the flow through the loop and the McMillan liquid flow sensor (Model 104) measures the flow rate. The water bath and liquid to liquid heat exchanger control the temperature of the fluid supplied to the test-section. Hot water that leaves the test section is cooled using a condenser Thermatron liquid-to-air heat exchanger (Model 735), before it flows back into the reservoir. All these components in the flow-loop are connected using Swagelok tubing and fittings.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
4
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
The data from all the sensors are collected using a National Instruments high speed Data Acquisition System. This system consists of a chassis (NI PXIe-1082), controller (NI PXIe-8135), a voltage module (NI PXIe-6363) and a RTD module (NI PXIe-4357). High speed camera (Photron FASTCAM SA5) is used for flow visualization. A metal halide light source (Daitron MLDS250) provides the required lighting.
Table 1 Microchannel dimensions and experimental test conditions.
2.3. Experiment procedure Before carrying out the main flow boiling experiments, temperature sensors were calibrated, heat loss tests were performed and water was boiled for degassing purpose. Procedures for these are the same as those used by Alam et al. [23]. For the heat loss test, test-section is evacuated of the coolant. Power is then supplied to the heaters. At a steady-state, all the supplied heat is lost and it is a function of the average wall temperature. During actual experiments, flow rate as well as inlet fluid temperature (86 C) were maintained throughout a session. Supplied heat flux was incremented and a quasi-steady state was considered to be achieved after the temperature fluctuations reduced to within 0:5 C for 1 min. A TTL signal was then sent by the camera to the DAQ thus triggering it to capture data synchronously with flow visualization. This data was captured at 1 kHz while the video was captured at a frame rate of 5 kHz, for a short duration of about 1 s. Under the same experimental conditions, data (not synchronous with video) was also captured at a frequency of 10 Hz for 5 min to get time-averaged values. Same procedure was repeated for several heat fluxes. Table 1 shows the experimental conditions tested. Confinement number [26] was always large. Hence, under all tested conditions, the term ‘‘microchannel” is considered most appropriate. However, it may be noted that the aspect ratio of all microchannels is small. Yet, the aspect ratios are much larger compared to the so called micro-gaps used by some researchers [9,27]. 3. Data reduction procedure Data reduction for transient temperature is applied to data captured by diode 8 (diode 1 is the most upstream while diode 10 is the most downstream on the thermal test-chip). The reason for the choice of this diode is that it is nearly in center of the field of view of the high speed camera used to capture video frames when it is adjusted at the most downstream end of the microchannel (the 8th sensor’s location is shown by a blue dot in video frames in Fig. 6). 3.1. Steady state The base heat flux is calculated by subtracting the heat loss from the power supplied to the heaters during main experiments and then dividing it by the wetted surface area. For measurement of the voltage drop across the heaters, kelvin connection (4-wire sensing) is used to eliminate the effect of voltage drops in the wires carrying high current. Current is calculated from the voltage measurement across a shunt resistor connected in series with the load (heaters).
qb ðtÞ ¼
VI Q l A
ð1Þ
Case
l (mm)
W (mm)
H (mm)
Dh ðmmÞ
G actual (nominal) (kg/(m2s))
Co
1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3 3.4
25.4 25.4 25.4 25.4 25.4 25.4 25.4 25.4 25.4
2.54 2.54 2.54 2.54 2.54 2.54 2.54 2.54 2.54
0.14 0.14 0.28 0.28 0.28 0.42 0.42 0.42 0.42
0.265 0.265 0.504 0.504 0.504 0.721 0.721 0.721 0.721
557 (600) 994 (1000) 393 (400) 600 (600) 998 (1000) 182 (200) 380 (400) 580 (600) 1028 (1000)
9.51 9.51 5.01 5.01 5.01 3.5 3.5 3.5 3.5
3.2. Fast Fourier Transform (FFT) of the wetted surface temperature Fourier analysis is performed using Discrete Fourier Transform (DFT) on the transient temperature measured by the diode sensor. DFT breaks down the signal into constituent sinusoidal signals of different frequencies and amplitudes. The governing equation for temperature distribution within the solid is the 1D transient heat conduction Eq. (5). Because of thin substrate, temperature difference within the solid domain (along the thickness) are small enough so that changes in substrate properties can be neglected. This implies that this equation is linear. The principle of superposition is hence valid. Thus, the original signal can be broken down into a number of constituent signals. The algorithm used for computing the DFT is Fast Fourier Transform (FFT). FFT transforms the time-dependent temperature signal into frequency domain data. During actual experiments, temperature measurement is done 630 lm below the wetted surface. The requirement however, is to get FFT of temperature on the wetted surface. The difficulty here is that due to finite thermal diffusivity of the silicon substrate, the temperature signal on the wetted surface gets damped as it penetrates the substrate. To correct for this damping and to get the actual frequency domain that represents temperature variations on the wetting surface, amplitude ratio (damping) is derived. This amplitude ratio is a function of the sine wave signal frequency (or non-dimensional frequency given by Eq. (4) for general applicability since nondimensionalization makes the damping independent of depth of the sensor and thermal diffusivity of the solid substrate).
f ¼
ts 2
a
f
ð4Þ
To derive each of these amplitude ratios, Finite-Volume Method (FVM) is used for discretization of the 1D transient heat conduction equation. Sinusoidal temperature signals are simulated on the top (wetting surface) and the corresponding temperature signal at the bottom (sensor location) of the 630 lm solid domain is derived and amplitude of this damped sine wave is calculated. The ratio of this damped amplitude and the originally simulated sine wave amplitude (on the wetting surface) is the amplitude ratio. This amplitude ratios are derived for several frequencies of the sine wave and are then plotted in Fig. 4 and curve fitted. The amplitudes in the frequency domain (by FFT) of the sensor (630 lm below the wetted surface) temperature are then divided by these amplitude ratios (corresponding to the individual frequencies) and the frequency domain of the temperature signal at the wetting surface is thus derived.
Heat loss is calculated as
Q l ¼ V l Il
ð2Þ
Wetted area is given as
A ¼ Wl
ð3Þ
3.3. Transient heat transfer Calculation of transient wetted surface temperature involves solving the 1D transient heat conduction equation for a solid domain schematically represented by Fig. 5
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
5
Fig. 6. Flow regimes observed during experiments were (a) Single phase flow, (b) Bubbly flow, (c) Slug flow, (d) Churn flow, (e) Wispy annular flow.
Fig. 4. Amplitude ratio vs. frequency.
larger frequencies. Hence, for small noise level at the sensor location, large changes are artificially introduced in the calculated wetted surface temperature. Standard text-books on this subject (e.g. Beck et al. [28]) can be referred to for details. Such methods have also been verified experimentally to give very good results [29]. A brief description regarding the solution method used in this work is described in Appendix A. This method is used instead of the direct numerical method to eliminate the effect of noise for calculation of T w . 4. Results and discussion
Fig. 5. Schematic of the solid domain.
qc
@T @ k@T ¼ @t @x @x
ð5Þ
The top (wetted) surface temperature T w ð0; tÞ and top surface heat flux qw ðtÞ are unknown.
k
@T ¼ qw ðtÞ @x x¼0
ð6Þ
The initial temperature T 0 is known.
Tðx; 0Þ ¼ T 0
ð7Þ
The heat flux at the bottom surface qb is also known.
k
@T ¼ qb @x x¼L
ð8Þ
The temperature measured by temperature sensor at the bottom surface is given as
TðL; tÞ ¼ T sensor ðtÞ ¼ YðtÞ
ð9Þ
Eq. (5) along with initial condition i.e. Eq. (7) and two boundary conditions at the bottom surface i.e. Eqs. (8) and (9) can be directly solved numerically using Finite Volume Method. However, the problem is ill-posed since both boundary conditions at the top (wetted) surface are unknown. Such a problem is called an Inverse Heat Conduction Problem (IHCP). These problems amplify the effect of noise (within the measured variables) in the calculated quantities. This may be understood well from Fig. 4. Especially for higher frequencies, small deviations in temperature signal measured at the sensor location indicate a large change in temperature at the wetted surface since the amplitude ratio becomes small for
Sample video frames showing various flow regimes encountered for various experimental conditions mentioned in Table 1 are shown in Fig. 6. Location of diode 8 is shown by a blue1 dot. Fig. 7 shows the typical sensor temperature verses time data that was taken at a frequency of 1 kHz, for some of the selected conditions under which various flow regimes were observed. A cursory look gives an idea about the unique temperature signature of each of the flow regimes. While single phase flow has the least fluctuations, the same for bubbly flow are relatively larger. For slug flow as well as for Churn/Wispy annular flow, the fluctuations are even larger. It can be observed that the amplitudes of fluctuations are often small, especially for single phase and bubbly flow compared to the uncertainty in temperature measurement. However, it should be noted that uncertainty in temperature measurement only affects the measured average value of temperature. The transient fluctuations (relative to the time-averaged value) are affected only by noise in temperature signal measurement. For the present case, the standard deviation in the noise is 0:1 C. There were three probable data-reduction procedures considered that could be used to identify flow regimes. 1. Standard deviation of sensor temperature data: Although very simple to apply, the difficulty with standard deviation is that temperature data is typically measured below the wetted surface. This leads to damping (due to thermal capacitance of the solid substrate) of temporal fluctuations in temperature due to phenomena occurring on the wetted surface. This damping
1 For interpretation of color in Fig. 6, the reader is referred to the web version of this article.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
6
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 7. Typical temperature (measured by sensor) verses time curve for different flow regimes (a) Single phase flow for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ, (b) Bubbly flow for H ¼ 0:42 mm; G ¼ 1000 kg=ðm2 sÞ, (c) Slug flow for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ, (d) Churn/Annular flow for H ¼ 0:14 mm; G ¼ 600 kg=ðm2 sÞ.
Fig. 8. Comparison of FFT obtained directly using transient wall (wetted surface) temperature and corrected FFT of temperature sensor data for (a) Single phase flow, (b) Bubbly flow, (c) Slug flow, (d) Slug/Churn flow, (e) Churn/Wispy annular flow.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
7
Fig. 9. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:14 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
is not only a function of substrate properties and depth of temperature sensor from the wetted surface but also the frequency components present in the temperature–time data as shown in Fig. 4. Even if regimes could be differentiated for the present experimental conditions from standard deviation values, it would lack general applicability. This method is thus avoided. 2. Direct application of FFT on temperature sensor data followed by the use of methodology described in Section (3.2) to correct for damping of amplitudes of different frequencies. 3. The use of IHCP solution method to first calculate the wetted surface temperature followed by application of FFT on the wetted surface data.
Fig. 8 shows a comparison between the frequency domain curve obtained by the second (denoted in the legend as ‘T sensor based’) and third (denoted in the legend as ‘T w based’) data reduction procedure considered for various experimental conditions under which different flow regimes were observed. Especially from Fig. 8(a), it may be observed that the amplitudes are much smaller than the uncertainty in temperature measurement as well as the standard deviation in the noise. It may again be noted that uncertainty in temperature measurement only affects the time-averaged value. The amplitudes are affected by the noise in the temperature sensor. However, the noise being random, does not have any distinct peaks and consists of many small amplitudes that are distributed over various frequencies.
Fig. 10. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
8
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 11. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 400 kg=ðm2 sÞ for various heat fluxes.
It can be observed from Fig. 8(c), (d) and (e) that there is a very good agreement in the frequency domain, between the two methods, especially for lower frequencies (below 80 Hz). For higher frequencies, the data reduction procedure based on the third method (involving the use of T w ) leads to smaller amplitudes compared to that obtained from the second method (involving the direct use of T sensor ). For Fig. 8(a) and (b) the difference in amplitude seems much more prominent, since the y-scale is relatively smaller. The discrepancy in the frequency domain for the two methods increases as the frequency increases. This is due to the presence
of noise in the T sensor data and due to the correction for higher frequency being much larger (refer to Fig. 4 and note the decreasing value of amplitude ratio with increasing frequency) leading to much greater amplification of noise at higher frequencies. Contrastingly, for T w , since IHCP solution dampens the noise very well, this effect is not observed for T w based curve. Hence, there is a difference in the frequency-domain of the two methodologies. The third method (‘T w based’) seems to be the best due to clear distinction in the some of the highest peak amplitudes for different flow regimes. However, the algorithm for post-processing is much
Fig. 12. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
9
Fig. 13. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
more complex than that for the second method (‘T sensor based’). This could make things difficult for real-time applications. Fortunately, within certain limits the third method still seems applicable. For frequency components less than 80 Hz, the second method is nearly as good as the third for distinction of the flow regimes. Moreover, the relative simplicity of the method makes it more practical for application purpose. Hence, unless much higher frequencies are dominant and unless the sensor noise (having standard deviation of 0:1 C in the present case) is too large, the second method is good enough. If sensor noise is much smaller
and/or if the sensor depth is smaller and/or the thermal diffusivity of the solid substrate material is large (non-dimensional frequency would be less in that case and amplitude ratio would be large implying a smaller amplification of noise), it is expected that frequencies much greater than 80 Hz can also be well captured. For each of the conditions presented in Table 1, a number of heat fluxes were tested. However, for brevity, results of only some of the heat fluxes have been shown in Figs. 9–17 to demonstrate the differences in some of the maximum peak amplitudes in the frequency for different flow regimes.
Fig. 14. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 200 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
10
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 15. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 400 kg=ðm2 sÞ for various heat fluxes.
Fig. 18, shows the regime detection chart summarizing the observations. For single phase flow, amplitude of all peaks were below 0:03 C for frequencies less than 80 Hz. For bubbly flow several of the highest of amplitude peaks ranged from 0.04 to 0.25 °C. While Slug flow, Churn flow and Wispy annular flow could not be distinguished among each other, some of the highest of peaks were much greater (> 0:35 C) compared to single phase and bubble flow. Single phase flow (for steady-state condition) in Figs. 9–17 should ideally not have had any frequency components and the
curve should have simply coincided with the x-axis. However, due to non-ideal experimental conditions, primarily the inherent noise in temperature measurement leads to some temperature fluctuations leading to non-zero amplitudes in frequency domain. During bubbly flow, the temperature fluctuations are relatively higher than that of single-phase flow since single-phase is steady (ideally) while boiling phenomena is unsteady. Immediately following bubble nucleation, there is a sudden increase in heat transfer coefficient due to release of accumulated superheat (utilized in evaporation, owing to the latent heat) in the surrounding liquid
Fig. 16. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
11
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 17. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
[30]. This increased heat transfer coefficient in turn leads to decreased wall temperature. However, during the waiting time (time period between bubble detachment and ebullition of another bubble), since there is single phase flow, the heat transfer coefficient again reduces and consequently, the wall temperature increases. Hence, amplitudes are larger for bubbly flow compared to single phase flow. Slug flow regime generally involves three distinct features as shown in Fig. 19; (i) elongated bubble flow, during which the heat transfer coefficient is very high owing to thin film evaporation, (ii) local dry-out, during which the heat transfer coefficient is very low, (iii) liquid slug flow, during the passage of which heat transfer coefficient is intermediate [31]. Due to multiplicity of heat transfer mechanisms and large difference in heat transfer coefficient associated with each of them, the temperature fluctuations during slug flow are very large. During Churn flow and Wispy annular flow, the hydrodynamics and heat transfer are complex. These flow regimes may include thin film evaporation of liquid film whose thickness may vary with time, nucleation of smaller bubbles and periods of partial-dry-out, again leading to temperature fluctuations which are as high as those observed during slug flow regime. It was anticipated that with increase in heat flux, due to increase in number of bubbles and slugs, the dominant frequency could shift to higher frequencies. But contrarily to expectations, this was not observed. The lack of increase in dominant frequencies with increasing heat flux is likely due to stochastic nature of
the bubble/slug ebullition and passage leading to distributed peaks rather than a single large peak. Moreover, temperature variation during a complete slug cycle is not sinusoidal. These are the reasons, that for conditions under which slug flow regime was
Fig. 18. Regime detection chart.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
12
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 19. Schematic of slug flow.
observed, although a relatively large peak was observed in the frequency-domain at a frequency corresponding to the frequency of passage of slug (as seen from high speed flow visualization), there were other peaks too, which could often have larger amplitudes. Hence, neither was a pattern such as increase in dominant frequencies was observed with increase in heat flux nor was it possible to detect the slug frequency from peaks in frequency domain.
S¼
r X ðY Mþi1 T Mþi1 Þ2
where
T Mþi1 ¼ Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 þ /i qM
Acknowledgement Financial support given by MOE, Singapore (Ref: WBS No. R265000-423-112) is gratefully acknowledged. Appendix A. Solution methodology for Inverse Heat Conduction Problem (IHCP) The solution method used for solving IHCP is Beck’s Functional Specification Method-constant heat flux functional form [28]. The objective for the functional specification method is to minimize the function S (Eq. (A.1)) with respect to top (wetted surface) surface heat flux qM where M refers to the index of the current timestep. Function S is the summation for r future-time steps of the square of the difference between the sensor temperature Y and the numerical model imposed temperature T at the sensor location. Temporarily the heat flux qM at time step t M is assumed constant over ‘r’ future times. Future temperature information is used to reduce the sensitivity of the estimated heat flux to measurement noise and stabilize the computation as the time step size reduces. The heat flux qM is determined such that the least square error between the model and experimentally measured temperatures is minimized. This function is expressed as
ðA:2Þ
The resulting equation after minimization of S w.r.t qM for the Functional Specification Method-constant heat flux functional form is
Pr
5. Summary and conclusions Experiments were performed for sub-cooled flow boiling phenomena in a single low aspect ratio microchannel with width and length of 2.54 mm and 25.4 mm respectively and heights 0:14; 0:28 and 0:42 mm for mass fluxes ranging from 200 to 1000 kg/(m2 s). Several heat fluxes were tested for each condition. A simple methodology based on transient wall temperature is developed to diagnose flow regimes. Single phase flow, Bubbly flow and Slug/Churn/Wispy-annular flow could be clearly differentiated by looking at some of the highest of peaks in amplitude in the corrected frequency domain of wall temperature for frequencies below 80 Hz. For single phase flow, all peaks were below 0:03 C. For bubbly flow, several of the highest of peaks ranged from 0.04 to 0.25 °C. While for Slug flow, Churn flow and Wispy annular flow, the highest of peaks were greater than 0:35 C. Suggestions have also been given to increase the limit of maximum frequency for which results can be relied upon.
ðA:1Þ
i¼1
qM ¼
i¼1 ðY Mþi1
Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 Þ/i ; M ¼ 1; 2; . . . ; M max Pr 2 i¼1 /i ðA:3Þ
where
Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 ¼ T 0 þ
M 1 X
qn D/Mnþi1
ðA:4Þ
n¼1
and the sensitivity coefficient for temperature at the sensor location and heat flux at the top surface is written as
@T @qti
ðA:5Þ
D/i ¼ /iþ1 /i
ðA:6Þ
/i ¼
T w , that is the temperature at the top (wetted) surface can then be obtained by numerically solving the transient heat conduction equation using two boundary conditions qM (at top surface) obtained from Eq. (A.3) and T obtained from Eq. (A.2) (which is the model temperature, instead of the sensor temperature Y) along with the initial condition. References [1] Agostini Bruno, Fabbri Matteo, Park Jung E, Wojtan Leszek, Thome John R, Michel Bruno. State of the art of high heat flux cooling technologies. Heat Transfer Eng 2007;28(4):258–81. [2] Zhang Lian, Koo Jae-mo, Jiang Linan, Asheghi Mehdi, Goodson Kenneth E, Santiago Juan G, et al. Measurements and modeling of two-phase flow in microchannels with nearly constant heat flux boundary conditions. J Electromech Syst 2002;11(1):12–9. [3] David Milnes P, Miler Josef, Steinbrenner Julie E, Yang Yizhang, Touzelbaev Maxat, Goodson Kenneth E. Hydraulic and thermal characteristics of a vapor venting two-phase microchannel heat exchanger. Int J Heat Mass Transfer 2011;54(25–26):5504–16. [4] Shao Liang-Liang, Yang Liang, Zhang Chun-Lu. Comparison of heat pump performance using fin-and-tube and microchannel heat exchangers under frost conditions. Appl Energy 2010;87(4):1187–97. [5] Yuan Weixing, Yang Bo, Yang Yufei, Ren Kexian, Xu Jian, Liao Yibing. Development and experimental study of the characteristics of a prototype miniature vapor compression refrigerator. Appl Energy 2015;143:47–57. [6] Revellin R, Thome JR. A new type of diabatic flow pattern map for boiling heat transfer in microchannels. J Micromech Microeng 2007;17(4):788–96. [7] Martin-Callizo Claudi, Palm Bjorn, Owhaib Wahib, Ali Rashid. Flow boiling visualization of R-134a in a vertical channel of small diameter. J Heat Transfer 2010;132(3):031001.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx [8] Harirchian Tannaz, Garimella Suresh V. Flow regime-based modeling of heat transfer and pressure drop in microchannel flow boiling. Int J Heat Mass Transfer 2012;55(4):1246–60. [9] Alam Tamanna, Lee Poh Seng, Yap Christopher R. Effects of surface roughness on flow boiling in silicon microgap heat sinks. Int J Heat Mass Transfer 2013;64:28–41. [10] Zhang Lian, Wang Evelyn N, Goodson Kenneth E, Kenny Thomas W. Phase change phenomena in silicon microchannels. Int J Heat Mass Transfer 2005;48 (8):1572–82. [11] Sitar Anze, Sedmak Ivan, Golobic Iztok. Boiling of water and FC-72 in microchannels enhanced with novel features. Int J Heat Mass Transfer 2012;55(23–24):6446–57. [12] Lu Chun Ting, Pan Chin. Convective boiling in a parallel microchannel heat sink with a diverging cross section and artificial nucleation sites. Exp Therm Fluid Sci 2011;35(5):810–5. [13] Kuo CJ, Peles Y. Local measurement of flow boiling in structured surface microchannels. Int J Heat Mass Transfer 2007;50(23–24):4513–26. [14] Yang Fanghao, Dai Xianming, Peles Yoav, Cheng Ping, Khan Jamil, Li Chen. Flow boiling phenomena in a single annular flow regime in microchannels (I): Characterization of flow boiling heat transfer. Int J Heat Mass Transfer 2014;68:703–15. [15] Kandlikar Satish G, Widger Theodore, Kalani Ankit, Mejia Valentina. Enhanced flow boiling over open microchannels with uniform and tapered gap manifolds. J Heat Transfer 2013;135(6):061401. [16] Law Matthew, Lee Poh-Seng, Balasubramanian Karthik. Experimental investigation of flow boiling heat transfer in novel oblique-finned microchannels. Int J Heat Mass Transfer 2014;76:419–31. [17] Marcinichen Jackson Braz, Olivier Jonathan A, De Oliveira Vinicius, Thome John R. A review of on-chip micro-evaporation: experimental evaluation of liquid pumping and vapor compression driven cooling systems and control. Appl Energy 2012;92:147–61. [18] Marcinichen Jackson Braz, Wu Duan, Paredes Stephan, Thome John R, Michel Bruno. Dynamic flow control and performance comparison of different concepts of two-phase on-chip cooling cycles. Appl Energy 2014;114:179–91. [19] Garimella Suresh V, Persoons Tim, Weibel Justin, Yeh Lian-Tuu. Technological drivers in data centers and telecom systems: multiscale thermal, electrical, and energy management. Appl Energy 2013;107:66–80.
13
[20] Kim Yoon Jo, Joshi Yogendra K, Fedorov Andrei G, Lee Young-Joon, Lim Sungkyu. Thermal characterization of interlayer microfluidic cooling of threedimensional integrated circuits with nonuniform heat. J Heat Transfer 2014;132(April 2010):1–9. [21] Koo Jae-Mo, Im Sungjun, Jiang Linan, Goodson Kenneth E. Integrated microchannel cooling for three-dimensional electronic circuit architectures. J Heat Transfer 2005;127(January 2005):49. [22] Revellin Rémi, Dupont Vincent, Ursenbacher Thierry, Thome John R, Zun Iztok. Characterization of diabatic two-phase flows in microchannels: flow parameter results for R-134a in a 0.5 mm channel. Int J Multiphase Flow 2006;32(7):755–74. [23] Alam Tamanna, Lee Poh Seng, Yap Christopher R, Jin Liwen, Balasubramanian K. Experimental investigation and flow visualization to determine the optimum dimension range of microgap heat sinks. Int J Heat Mass Transfer 2012;55(25–26):7623–34. [24] Harirchian Tannaz, Garimella Suresh V. A comprehensive flow regime map for microchannel flow boiling with quantitative transition criteria. Int J Heat Mass Transfer 2010;53(13–14):2694–702. [25] Lee PS, Garimella S. Saturated flow boiling heat transfer and pressure drop in silicon microchannel arrays. Int J Heat Mass Transfer 2008;51(3–4):789–806. [26] Kew Peter A, Cornwell Keith. Correlations for the prediction of boiling heat transfer in small-diameter channels. Appl Therm Eng 1997;17(8–10):705–15. [27] Dae Whan Kim, Emil Rahim, Avram Bar-Cohen, Bongtae Han. Thermofluid characteristics of two-phase flow in micro-gap channels. In: 2008 11th IEEE intersociety conference on thermal and thermomechanical phenomena in electronic systems, I-THERM; 2008. p. 979–92. [28] Beck James V, Blackwell Ben, Clair Jr Charles RSt. Inverse heat conduction: illposed problems. Wiley Interscience Publication; 1985. [29] Beck JV, Blackwell B, Haji-Sheikh A. Comparison of some inverse heat conduction methods using experimental data. Int J Heat Mass Transfer 1996;39(17):3649–57. [30] Kandlikar Satish G. Heat transfer mechanisms during flow boiling in microchannels. J Heat Transfer 2004;126(1):8. [31] Thome J, Dupont V, Jacobi Anthony M. Heat transfer model for evaporation in microchannels. Part I: Presentation of the model. Int J Heat Mass Transfer 2004;47(14–16):3375–85.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal Mrinal Jagirdar ⇑, Poh Seng Lee Department of Mechanical Engineering, National University of Singapore, Singapore 117575, Singapore
h i g h l i g h t s Novel flow regime detection technique for micro domains is presented. Method involves converting temporal data of temperature into frequency domain. It is corrected for damping due to thermal diffusivity of solid material. Magnitude of amplitudes were different for single phase, bubbly and slug flow. Applications involve smart feed-back loops for cooling and 3D IC packages.
a r t i c l e
i n f o
Article history: Received 15 August 2015 Received in revised form 25 December 2015 Accepted 26 December 2015 Available online xxxx Keywords: Flow boiling Microchannel Flow-regime detection Temperature transients
a b s t r a c t Flow boiling in microchannels has been receiving a lot of attention in recent years because of its high heat flux removal capabilities at low flow rates and low pumping power. An important aspect of flow-boiling experiments is prediction or detection of the prevalent flow-regime. Currently, most researchers use high-speed camera for flow visualization for regime detection. However, in some cases due to limitations of the experimental setup and test-piece, such as multi-layer cooling of 3D IC stack, this may not be feasible. In this paper, the influence of flow-regime on frequency domain of local temperature data of the wetted surface is studied. Experiments have been performed synchronously with high speed flow visualization on a single microchannel with width and length of 2.54 mm and 25.4 mm respectively. The microchannel heights tested were 0:14; 0:28 and 0:42 mm. De-gassed, de-ionized water was used as the working fluid. Mass fluxes tested ranged from 200 to 1000 kg=ðm2 sÞ. Depending on the prevalent flow regime, some of the highest of peak amplitudes in the frequency domain were quite distinct. Within the bounds of current experimental parameters, it is concluded that local transient temperature data can be a potential diagnostic tool for detection of flow-regimes. (A shorter version of this paper was presented at the 7th International Conference on Applied Energy (ICAE2015), March 28–31, 2015, Abu Dhabi, UAE (Original paper title: ‘‘Temperature transients for detection of flow-regimes in a mini/microchannel” and Paper No.: 430).) Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction With the current state-of-the-art electronics, the challenge is to remove heat fluxes of the order of 300 W=cm2 and this figure will keep rising with further miniaturization [1]. Especially for such applications, two-phase flow in micro-channels has been identified to be a very promising technology since the phase change process facilitates smaller thermal resistances [2] and the pumping power requirement too is quite small [3]. Other applications of microchannel two-phase flow include vapor compression heat ⇑ Corresponding author. Tel.: +65 65164657; fax: +65 67791459. E-mail address: [email protected] (M. Jagirdar).
pumps [4], vapor compression refrigerators [5], etc. An important aspect of microchannel two-phase flow is detection of flow regimes. In general, high-speed camera serves the purpose during experiments. Some researchers have also tried to come up with predictive flow regime maps or techniques to identify flow regimes for various experimental conditions. Revellin and Thome [6] developed a flow pattern map for circular microchannels that had the mass velocity and the vapor quality as co-ordinates. Transition curves were determined using Reynolds number, Weber number, Boiling number, fluid properties and microchannel dimensions. Regimes identified were isolated bubble regime, coalescing bubble regime, the annular zone and the post dry-out zone. Martin-Callizo et al.
http://dx.doi.org/10.1016/j.apenergy.2015.12.111 0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
2
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Nomenclature A c Co Dh f f G H I Il k L l qb qn Ql r S t
wetted surface area (m2 ) specific heat (J/(kg °C)) confinement number hydraulic diameter (m) frequency of temperature signal (Hz) non-dimensional frequency mass flux (kg=ðm2 sÞ) microchannel height (m) current (A) current during heat loss test (A) thermal conductivity (W=ðm KÞ) thickness of the solid silicon substrate (m) length of the microchannel (m) heat flux entering the solid substrate (W=m2 ) heat flux at time step ‘n’ at the wetted surface (W=m2 ) heat loss (W) number of future time-steps minimization function time (s)
[7] noted that larger inlet sub-cooling shifted all transition lines (except for Slug-annular/Churn to annular flow) to earlier vapor qualities. Moreover, an increase in the saturation temperature shifted all transition boundaries towards higher vapor qualities. A flow regime map was also proposed by Harirchian and Garimella [8]. Fluid properties, channel diameter, heat flux, mass flux as well as the heated length of the microchannels were considered and four regions namely slug, confined annular, bubbly, and alternating Churn/Annular/Wispy-annularflow were plotted. It may hence be argued that there are a number of variables affecting flowregimes and it seems to be a very difficult task to develop a regime prediction method that is generally applicable. Moreover, modifications made to the wetted surface characteristics or the geometry of microchannel fin structure can add to the difficulty in development of a generic prediction method for identification of flow regime. It was shown by Alam et al. [9] that transition from single phase to bubbly flow can be altered by changing the surface roughness. Even artificial nucleation sites (ANS) as shown by some researchers [10–13] can alter the flow-regime and heat transfer mechanism. Yang et al. [14] showed that by having superhydrophilic silicon nano-wires on the inner-walls of the microchannel, heat transfer can be very drastically enhanced by manipulation of the prevalent flow regime. Some researchers have also developed novel microchannel designs that can change heat transfer performance by changing prevalent regimes under certain experimental conditions [3,15,16]. Thus, development of a diagnostic tool for flow-regime detection seems to be relatively simple compared to the development of a generally applicable regime prediction method. Although, a diagnostic technique cannot fully replace flow-regime prediction techniques, it can still be of utility for important applications. Smart feed-back loops that control flow-rate to control the prevalent flow-regime for optimal heat transfer performance and minimum pressure drop can potentially use regime detection techniques. In fact, having noted that two-phase cooling can substantially reduce the cooling energy compared to air and water based cooling, Marcinichen et al. [17] showed the development of control strategies for multi-microchannel evaporators. Also, Marcinichen et al. [18] have pointed out that with the implementation of simple and cheap controllers, it is possible to ensure that on-chip cooling is optimal all the time even for transient loads.
T0 Ti ts Tw T sensor V Vl W x Yi
initial temperature of solid substrate ( C) temperature at time step i ( C) sensor depth (m) wetted surface temperature ( C) temperature measured by the sensor ( C) voltage (V) voltage during heat loss test (V) microchannel width (m) spatial co-ordinate along depth of solid substrate (m) measured temperature at time step i ( C)
Greek symbols a thermal diffusivity of substrate (m2 =s) DT temperature difference ( C) sensitivity coefficient at time step i at the sensor loca/i tion ( C m2 =W) q density (kg=m3 )
Another potential application for regime detection technique is for 3D IC packaging technology. Development in 3D chip packaging is being driven to overcome the limitations due to the physical distance between cores and memory as the computational performance of HPC systems continues to increase [19]. Thus, researchers have also shown interest towards assessment of cooling solutions for modular thermal management of 3D-ICs consisting of multiple inter-layer microchannels. Kim et al. [20] and Koo et al. [21] carried out numerical and theoretical investigations respectively, to explore 3D IC cooling. Since a high speed camera cannot be used for regime detection in internal microchannels, a diagnostic technique can be very useful. To the best of our knowledge, only Revellin et al. [22] developed a regime detection technique. Their method involved a novel optical technique that could characterize flow pattern transition in two-phase flows. Bubble frequency, percentage of surviving small bubbles, lengths of bubbles and flow pattern transitions could also be determined. In this paper, transients in local, wall temperature are investigated with synchronized flow-visualization to see if flow regimes have a unique temperature signature and whether it can be used for regime-detection without the need for optical access.
2. Experimental setup and procedure 2.1. Test section An exploded view of the test-section is as shown in Fig. 1 and a photograph of the assembled micro-channel test section is as shown in Fig. 2(a). The silicon test-chip (from Kokomo Semiconductors) is flip-chip packaged onto the PCB shown in Fig. 2(b). The Pyrex glass (shown in Fig. 1) insert is bonded to the Polycarbonate top-cover cavity and it provides good optical transparency required for flow-visualizations. A gap formed between the surface of the silicon shown in Fig. 2(b) and the Pyrex glass insert is a low aspect ratio micro-channel. Gasket sheet between the Polycarbonate top-cover and the PCB acts as a sealant. The inlet and outlet plenums in the polycarbonate top cover, as seen in Fig. 2(a), have fittings above them so that by slightly loosening them, the bubbles can be bled out. Two ports
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
3
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 1. Exploded view of the assembled test-section.
Fig. 2. (a) Assembled test-section. (b) Silicon test chip (wetted surface) on the PCB. (c) 1 10 array of thermal test dies behind the wetted surface of the silicon test chip, 630 lm below.
(see Fig. 2(a)) are provided through which RTD probes can be fitted to measure the fluid temperature at the inlet and outlet plenums. Similarly there are two more ports coming out of the other end of the plenums connected to the Pressure transducers. The Teflon bottom cover has an air–gap so that it is better insulated leading to smaller heat loss from the bottom. The microchannel width and length are 2540 lm and 25400 lm respectively. The currently used silicon test-chip consists of an array of 10 1 thermal test dies (see Fig. 2(c)) 630 lm below the top (wetted) surface. Each of the thermal test dies has an area of 2540 lm 2540 lm and each of them consists of a heater and a diode. Each heater covers an area of 2000 lm 2000 lm and each doped diode temperature sensor covers an area of 400 lm 400 lm at the center of the die. Each temperature sensor is a series connection of 5 diodes (p–n junction) with a total sensitivity of 10 m V= C and temperature is measured with an uncertainty of 0:4 C. The test-dies are the same as those used by Alam et al. [23], Harirchian and Garimella [24] and Lee and Garimella [25]. Fig. 3. Schematic of the flow loop.
2.2. Flow loop The flow loop schematic is as shown in Fig. 3. The test-section is mounted on a X–Y stage so that it is easy to adjust the field of view of the camera. The temperature at the inlet and outlet plenums of the test-section are measured using RTDs (Omega 1/10 DIN class) with an uncertainty of 0:12 C. The reservoir is filled with de-ionized water. Immersion heaters (2 600 W) fitted into the reservoir are used to boil the water for degassing. The gear pump (Micropump driven by Cole-parmer gear
pump drive) drives the flow through the loop and the McMillan liquid flow sensor (Model 104) measures the flow rate. The water bath and liquid to liquid heat exchanger control the temperature of the fluid supplied to the test-section. Hot water that leaves the test section is cooled using a condenser Thermatron liquid-to-air heat exchanger (Model 735), before it flows back into the reservoir. All these components in the flow-loop are connected using Swagelok tubing and fittings.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
4
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
The data from all the sensors are collected using a National Instruments high speed Data Acquisition System. This system consists of a chassis (NI PXIe-1082), controller (NI PXIe-8135), a voltage module (NI PXIe-6363) and a RTD module (NI PXIe-4357). High speed camera (Photron FASTCAM SA5) is used for flow visualization. A metal halide light source (Daitron MLDS250) provides the required lighting.
Table 1 Microchannel dimensions and experimental test conditions.
2.3. Experiment procedure Before carrying out the main flow boiling experiments, temperature sensors were calibrated, heat loss tests were performed and water was boiled for degassing purpose. Procedures for these are the same as those used by Alam et al. [23]. For the heat loss test, test-section is evacuated of the coolant. Power is then supplied to the heaters. At a steady-state, all the supplied heat is lost and it is a function of the average wall temperature. During actual experiments, flow rate as well as inlet fluid temperature (86 C) were maintained throughout a session. Supplied heat flux was incremented and a quasi-steady state was considered to be achieved after the temperature fluctuations reduced to within 0:5 C for 1 min. A TTL signal was then sent by the camera to the DAQ thus triggering it to capture data synchronously with flow visualization. This data was captured at 1 kHz while the video was captured at a frame rate of 5 kHz, for a short duration of about 1 s. Under the same experimental conditions, data (not synchronous with video) was also captured at a frequency of 10 Hz for 5 min to get time-averaged values. Same procedure was repeated for several heat fluxes. Table 1 shows the experimental conditions tested. Confinement number [26] was always large. Hence, under all tested conditions, the term ‘‘microchannel” is considered most appropriate. However, it may be noted that the aspect ratio of all microchannels is small. Yet, the aspect ratios are much larger compared to the so called micro-gaps used by some researchers [9,27]. 3. Data reduction procedure Data reduction for transient temperature is applied to data captured by diode 8 (diode 1 is the most upstream while diode 10 is the most downstream on the thermal test-chip). The reason for the choice of this diode is that it is nearly in center of the field of view of the high speed camera used to capture video frames when it is adjusted at the most downstream end of the microchannel (the 8th sensor’s location is shown by a blue dot in video frames in Fig. 6). 3.1. Steady state The base heat flux is calculated by subtracting the heat loss from the power supplied to the heaters during main experiments and then dividing it by the wetted surface area. For measurement of the voltage drop across the heaters, kelvin connection (4-wire sensing) is used to eliminate the effect of voltage drops in the wires carrying high current. Current is calculated from the voltage measurement across a shunt resistor connected in series with the load (heaters).
qb ðtÞ ¼
VI Q l A
ð1Þ
Case
l (mm)
W (mm)
H (mm)
Dh ðmmÞ
G actual (nominal) (kg/(m2s))
Co
1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3 3.4
25.4 25.4 25.4 25.4 25.4 25.4 25.4 25.4 25.4
2.54 2.54 2.54 2.54 2.54 2.54 2.54 2.54 2.54
0.14 0.14 0.28 0.28 0.28 0.42 0.42 0.42 0.42
0.265 0.265 0.504 0.504 0.504 0.721 0.721 0.721 0.721
557 (600) 994 (1000) 393 (400) 600 (600) 998 (1000) 182 (200) 380 (400) 580 (600) 1028 (1000)
9.51 9.51 5.01 5.01 5.01 3.5 3.5 3.5 3.5
3.2. Fast Fourier Transform (FFT) of the wetted surface temperature Fourier analysis is performed using Discrete Fourier Transform (DFT) on the transient temperature measured by the diode sensor. DFT breaks down the signal into constituent sinusoidal signals of different frequencies and amplitudes. The governing equation for temperature distribution within the solid is the 1D transient heat conduction Eq. (5). Because of thin substrate, temperature difference within the solid domain (along the thickness) are small enough so that changes in substrate properties can be neglected. This implies that this equation is linear. The principle of superposition is hence valid. Thus, the original signal can be broken down into a number of constituent signals. The algorithm used for computing the DFT is Fast Fourier Transform (FFT). FFT transforms the time-dependent temperature signal into frequency domain data. During actual experiments, temperature measurement is done 630 lm below the wetted surface. The requirement however, is to get FFT of temperature on the wetted surface. The difficulty here is that due to finite thermal diffusivity of the silicon substrate, the temperature signal on the wetted surface gets damped as it penetrates the substrate. To correct for this damping and to get the actual frequency domain that represents temperature variations on the wetting surface, amplitude ratio (damping) is derived. This amplitude ratio is a function of the sine wave signal frequency (or non-dimensional frequency given by Eq. (4) for general applicability since nondimensionalization makes the damping independent of depth of the sensor and thermal diffusivity of the solid substrate).
f ¼
ts 2
a
f
ð4Þ
To derive each of these amplitude ratios, Finite-Volume Method (FVM) is used for discretization of the 1D transient heat conduction equation. Sinusoidal temperature signals are simulated on the top (wetting surface) and the corresponding temperature signal at the bottom (sensor location) of the 630 lm solid domain is derived and amplitude of this damped sine wave is calculated. The ratio of this damped amplitude and the originally simulated sine wave amplitude (on the wetting surface) is the amplitude ratio. This amplitude ratios are derived for several frequencies of the sine wave and are then plotted in Fig. 4 and curve fitted. The amplitudes in the frequency domain (by FFT) of the sensor (630 lm below the wetted surface) temperature are then divided by these amplitude ratios (corresponding to the individual frequencies) and the frequency domain of the temperature signal at the wetting surface is thus derived.
Heat loss is calculated as
Q l ¼ V l Il
ð2Þ
Wetted area is given as
A ¼ Wl
ð3Þ
3.3. Transient heat transfer Calculation of transient wetted surface temperature involves solving the 1D transient heat conduction equation for a solid domain schematically represented by Fig. 5
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
5
Fig. 6. Flow regimes observed during experiments were (a) Single phase flow, (b) Bubbly flow, (c) Slug flow, (d) Churn flow, (e) Wispy annular flow.
Fig. 4. Amplitude ratio vs. frequency.
larger frequencies. Hence, for small noise level at the sensor location, large changes are artificially introduced in the calculated wetted surface temperature. Standard text-books on this subject (e.g. Beck et al. [28]) can be referred to for details. Such methods have also been verified experimentally to give very good results [29]. A brief description regarding the solution method used in this work is described in Appendix A. This method is used instead of the direct numerical method to eliminate the effect of noise for calculation of T w . 4. Results and discussion
Fig. 5. Schematic of the solid domain.
qc
@T @ k@T ¼ @t @x @x
ð5Þ
The top (wetted) surface temperature T w ð0; tÞ and top surface heat flux qw ðtÞ are unknown.
k
@T ¼ qw ðtÞ @x x¼0
ð6Þ
The initial temperature T 0 is known.
Tðx; 0Þ ¼ T 0
ð7Þ
The heat flux at the bottom surface qb is also known.
k
@T ¼ qb @x x¼L
ð8Þ
The temperature measured by temperature sensor at the bottom surface is given as
TðL; tÞ ¼ T sensor ðtÞ ¼ YðtÞ
ð9Þ
Eq. (5) along with initial condition i.e. Eq. (7) and two boundary conditions at the bottom surface i.e. Eqs. (8) and (9) can be directly solved numerically using Finite Volume Method. However, the problem is ill-posed since both boundary conditions at the top (wetted) surface are unknown. Such a problem is called an Inverse Heat Conduction Problem (IHCP). These problems amplify the effect of noise (within the measured variables) in the calculated quantities. This may be understood well from Fig. 4. Especially for higher frequencies, small deviations in temperature signal measured at the sensor location indicate a large change in temperature at the wetted surface since the amplitude ratio becomes small for
Sample video frames showing various flow regimes encountered for various experimental conditions mentioned in Table 1 are shown in Fig. 6. Location of diode 8 is shown by a blue1 dot. Fig. 7 shows the typical sensor temperature verses time data that was taken at a frequency of 1 kHz, for some of the selected conditions under which various flow regimes were observed. A cursory look gives an idea about the unique temperature signature of each of the flow regimes. While single phase flow has the least fluctuations, the same for bubbly flow are relatively larger. For slug flow as well as for Churn/Wispy annular flow, the fluctuations are even larger. It can be observed that the amplitudes of fluctuations are often small, especially for single phase and bubbly flow compared to the uncertainty in temperature measurement. However, it should be noted that uncertainty in temperature measurement only affects the measured average value of temperature. The transient fluctuations (relative to the time-averaged value) are affected only by noise in temperature signal measurement. For the present case, the standard deviation in the noise is 0:1 C. There were three probable data-reduction procedures considered that could be used to identify flow regimes. 1. Standard deviation of sensor temperature data: Although very simple to apply, the difficulty with standard deviation is that temperature data is typically measured below the wetted surface. This leads to damping (due to thermal capacitance of the solid substrate) of temporal fluctuations in temperature due to phenomena occurring on the wetted surface. This damping
1 For interpretation of color in Fig. 6, the reader is referred to the web version of this article.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
6
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 7. Typical temperature (measured by sensor) verses time curve for different flow regimes (a) Single phase flow for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ, (b) Bubbly flow for H ¼ 0:42 mm; G ¼ 1000 kg=ðm2 sÞ, (c) Slug flow for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ, (d) Churn/Annular flow for H ¼ 0:14 mm; G ¼ 600 kg=ðm2 sÞ.
Fig. 8. Comparison of FFT obtained directly using transient wall (wetted surface) temperature and corrected FFT of temperature sensor data for (a) Single phase flow, (b) Bubbly flow, (c) Slug flow, (d) Slug/Churn flow, (e) Churn/Wispy annular flow.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
7
Fig. 9. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:14 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
is not only a function of substrate properties and depth of temperature sensor from the wetted surface but also the frequency components present in the temperature–time data as shown in Fig. 4. Even if regimes could be differentiated for the present experimental conditions from standard deviation values, it would lack general applicability. This method is thus avoided. 2. Direct application of FFT on temperature sensor data followed by the use of methodology described in Section (3.2) to correct for damping of amplitudes of different frequencies. 3. The use of IHCP solution method to first calculate the wetted surface temperature followed by application of FFT on the wetted surface data.
Fig. 8 shows a comparison between the frequency domain curve obtained by the second (denoted in the legend as ‘T sensor based’) and third (denoted in the legend as ‘T w based’) data reduction procedure considered for various experimental conditions under which different flow regimes were observed. Especially from Fig. 8(a), it may be observed that the amplitudes are much smaller than the uncertainty in temperature measurement as well as the standard deviation in the noise. It may again be noted that uncertainty in temperature measurement only affects the time-averaged value. The amplitudes are affected by the noise in the temperature sensor. However, the noise being random, does not have any distinct peaks and consists of many small amplitudes that are distributed over various frequencies.
Fig. 10. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:14 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
8
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 11. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 400 kg=ðm2 sÞ for various heat fluxes.
It can be observed from Fig. 8(c), (d) and (e) that there is a very good agreement in the frequency domain, between the two methods, especially for lower frequencies (below 80 Hz). For higher frequencies, the data reduction procedure based on the third method (involving the use of T w ) leads to smaller amplitudes compared to that obtained from the second method (involving the direct use of T sensor ). For Fig. 8(a) and (b) the difference in amplitude seems much more prominent, since the y-scale is relatively smaller. The discrepancy in the frequency domain for the two methods increases as the frequency increases. This is due to the presence
of noise in the T sensor data and due to the correction for higher frequency being much larger (refer to Fig. 4 and note the decreasing value of amplitude ratio with increasing frequency) leading to much greater amplification of noise at higher frequencies. Contrastingly, for T w , since IHCP solution dampens the noise very well, this effect is not observed for T w based curve. Hence, there is a difference in the frequency-domain of the two methodologies. The third method (‘T w based’) seems to be the best due to clear distinction in the some of the highest peak amplitudes for different flow regimes. However, the algorithm for post-processing is much
Fig. 12. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
9
Fig. 13. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:28 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
more complex than that for the second method (‘T sensor based’). This could make things difficult for real-time applications. Fortunately, within certain limits the third method still seems applicable. For frequency components less than 80 Hz, the second method is nearly as good as the third for distinction of the flow regimes. Moreover, the relative simplicity of the method makes it more practical for application purpose. Hence, unless much higher frequencies are dominant and unless the sensor noise (having standard deviation of 0:1 C in the present case) is too large, the second method is good enough. If sensor noise is much smaller
and/or if the sensor depth is smaller and/or the thermal diffusivity of the solid substrate material is large (non-dimensional frequency would be less in that case and amplitude ratio would be large implying a smaller amplification of noise), it is expected that frequencies much greater than 80 Hz can also be well captured. For each of the conditions presented in Table 1, a number of heat fluxes were tested. However, for brevity, results of only some of the heat fluxes have been shown in Figs. 9–17 to demonstrate the differences in some of the maximum peak amplitudes in the frequency for different flow regimes.
Fig. 14. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 200 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
10
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 15. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 400 kg=ðm2 sÞ for various heat fluxes.
Fig. 18, shows the regime detection chart summarizing the observations. For single phase flow, amplitude of all peaks were below 0:03 C for frequencies less than 80 Hz. For bubbly flow several of the highest of amplitude peaks ranged from 0.04 to 0.25 °C. While Slug flow, Churn flow and Wispy annular flow could not be distinguished among each other, some of the highest of peaks were much greater (> 0:35 C) compared to single phase and bubble flow. Single phase flow (for steady-state condition) in Figs. 9–17 should ideally not have had any frequency components and the
curve should have simply coincided with the x-axis. However, due to non-ideal experimental conditions, primarily the inherent noise in temperature measurement leads to some temperature fluctuations leading to non-zero amplitudes in frequency domain. During bubbly flow, the temperature fluctuations are relatively higher than that of single-phase flow since single-phase is steady (ideally) while boiling phenomena is unsteady. Immediately following bubble nucleation, there is a sudden increase in heat transfer coefficient due to release of accumulated superheat (utilized in evaporation, owing to the latent heat) in the surrounding liquid
Fig. 16. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 600 kg=ðm2 sÞ for various heat fluxes.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
11
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 17. Corrected FFT of temperature (measured by sensor) verses time data for H ¼ 0:42 mm; G ¼ 1000 kg=ðm2 sÞ for various heat fluxes.
[30]. This increased heat transfer coefficient in turn leads to decreased wall temperature. However, during the waiting time (time period between bubble detachment and ebullition of another bubble), since there is single phase flow, the heat transfer coefficient again reduces and consequently, the wall temperature increases. Hence, amplitudes are larger for bubbly flow compared to single phase flow. Slug flow regime generally involves three distinct features as shown in Fig. 19; (i) elongated bubble flow, during which the heat transfer coefficient is very high owing to thin film evaporation, (ii) local dry-out, during which the heat transfer coefficient is very low, (iii) liquid slug flow, during the passage of which heat transfer coefficient is intermediate [31]. Due to multiplicity of heat transfer mechanisms and large difference in heat transfer coefficient associated with each of them, the temperature fluctuations during slug flow are very large. During Churn flow and Wispy annular flow, the hydrodynamics and heat transfer are complex. These flow regimes may include thin film evaporation of liquid film whose thickness may vary with time, nucleation of smaller bubbles and periods of partial-dry-out, again leading to temperature fluctuations which are as high as those observed during slug flow regime. It was anticipated that with increase in heat flux, due to increase in number of bubbles and slugs, the dominant frequency could shift to higher frequencies. But contrarily to expectations, this was not observed. The lack of increase in dominant frequencies with increasing heat flux is likely due to stochastic nature of
the bubble/slug ebullition and passage leading to distributed peaks rather than a single large peak. Moreover, temperature variation during a complete slug cycle is not sinusoidal. These are the reasons, that for conditions under which slug flow regime was
Fig. 18. Regime detection chart.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
12
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx
Fig. 19. Schematic of slug flow.
observed, although a relatively large peak was observed in the frequency-domain at a frequency corresponding to the frequency of passage of slug (as seen from high speed flow visualization), there were other peaks too, which could often have larger amplitudes. Hence, neither was a pattern such as increase in dominant frequencies was observed with increase in heat flux nor was it possible to detect the slug frequency from peaks in frequency domain.
S¼
r X ðY Mþi1 T Mþi1 Þ2
where
T Mþi1 ¼ Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 þ /i qM
Acknowledgement Financial support given by MOE, Singapore (Ref: WBS No. R265000-423-112) is gratefully acknowledged. Appendix A. Solution methodology for Inverse Heat Conduction Problem (IHCP) The solution method used for solving IHCP is Beck’s Functional Specification Method-constant heat flux functional form [28]. The objective for the functional specification method is to minimize the function S (Eq. (A.1)) with respect to top (wetted surface) surface heat flux qM where M refers to the index of the current timestep. Function S is the summation for r future-time steps of the square of the difference between the sensor temperature Y and the numerical model imposed temperature T at the sensor location. Temporarily the heat flux qM at time step t M is assumed constant over ‘r’ future times. Future temperature information is used to reduce the sensitivity of the estimated heat flux to measurement noise and stabilize the computation as the time step size reduces. The heat flux qM is determined such that the least square error between the model and experimentally measured temperatures is minimized. This function is expressed as
ðA:2Þ
The resulting equation after minimization of S w.r.t qM for the Functional Specification Method-constant heat flux functional form is
Pr
5. Summary and conclusions Experiments were performed for sub-cooled flow boiling phenomena in a single low aspect ratio microchannel with width and length of 2.54 mm and 25.4 mm respectively and heights 0:14; 0:28 and 0:42 mm for mass fluxes ranging from 200 to 1000 kg/(m2 s). Several heat fluxes were tested for each condition. A simple methodology based on transient wall temperature is developed to diagnose flow regimes. Single phase flow, Bubbly flow and Slug/Churn/Wispy-annular flow could be clearly differentiated by looking at some of the highest of peaks in amplitude in the corrected frequency domain of wall temperature for frequencies below 80 Hz. For single phase flow, all peaks were below 0:03 C. For bubbly flow, several of the highest of peaks ranged from 0.04 to 0.25 °C. While for Slug flow, Churn flow and Wispy annular flow, the highest of peaks were greater than 0:35 C. Suggestions have also been given to increase the limit of maximum frequency for which results can be relied upon.
ðA:1Þ
i¼1
qM ¼
i¼1 ðY Mþi1
Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 Þ/i ; M ¼ 1; 2; . . . ; M max Pr 2 i¼1 /i ðA:3Þ
where
Tb Mþi1 jqM ¼qMþ1 ¼...¼qMþi1 ¼0 ¼ T 0 þ
M 1 X
qn D/Mnþi1
ðA:4Þ
n¼1
and the sensitivity coefficient for temperature at the sensor location and heat flux at the top surface is written as
@T @qti
ðA:5Þ
D/i ¼ /iþ1 /i
ðA:6Þ
/i ¼
T w , that is the temperature at the top (wetted) surface can then be obtained by numerically solving the transient heat conduction equation using two boundary conditions qM (at top surface) obtained from Eq. (A.3) and T obtained from Eq. (A.2) (which is the model temperature, instead of the sensor temperature Y) along with the initial condition. References [1] Agostini Bruno, Fabbri Matteo, Park Jung E, Wojtan Leszek, Thome John R, Michel Bruno. State of the art of high heat flux cooling technologies. Heat Transfer Eng 2007;28(4):258–81. [2] Zhang Lian, Koo Jae-mo, Jiang Linan, Asheghi Mehdi, Goodson Kenneth E, Santiago Juan G, et al. Measurements and modeling of two-phase flow in microchannels with nearly constant heat flux boundary conditions. J Electromech Syst 2002;11(1):12–9. [3] David Milnes P, Miler Josef, Steinbrenner Julie E, Yang Yizhang, Touzelbaev Maxat, Goodson Kenneth E. Hydraulic and thermal characteristics of a vapor venting two-phase microchannel heat exchanger. Int J Heat Mass Transfer 2011;54(25–26):5504–16. [4] Shao Liang-Liang, Yang Liang, Zhang Chun-Lu. Comparison of heat pump performance using fin-and-tube and microchannel heat exchangers under frost conditions. Appl Energy 2010;87(4):1187–97. [5] Yuan Weixing, Yang Bo, Yang Yufei, Ren Kexian, Xu Jian, Liao Yibing. Development and experimental study of the characteristics of a prototype miniature vapor compression refrigerator. Appl Energy 2015;143:47–57. [6] Revellin R, Thome JR. A new type of diabatic flow pattern map for boiling heat transfer in microchannels. J Micromech Microeng 2007;17(4):788–96. [7] Martin-Callizo Claudi, Palm Bjorn, Owhaib Wahib, Ali Rashid. Flow boiling visualization of R-134a in a vertical channel of small diameter. J Heat Transfer 2010;132(3):031001.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111
M. Jagirdar, P.S. Lee / Applied Energy xxx (2016) xxx–xxx [8] Harirchian Tannaz, Garimella Suresh V. Flow regime-based modeling of heat transfer and pressure drop in microchannel flow boiling. Int J Heat Mass Transfer 2012;55(4):1246–60. [9] Alam Tamanna, Lee Poh Seng, Yap Christopher R. Effects of surface roughness on flow boiling in silicon microgap heat sinks. Int J Heat Mass Transfer 2013;64:28–41. [10] Zhang Lian, Wang Evelyn N, Goodson Kenneth E, Kenny Thomas W. Phase change phenomena in silicon microchannels. Int J Heat Mass Transfer 2005;48 (8):1572–82. [11] Sitar Anze, Sedmak Ivan, Golobic Iztok. Boiling of water and FC-72 in microchannels enhanced with novel features. Int J Heat Mass Transfer 2012;55(23–24):6446–57. [12] Lu Chun Ting, Pan Chin. Convective boiling in a parallel microchannel heat sink with a diverging cross section and artificial nucleation sites. Exp Therm Fluid Sci 2011;35(5):810–5. [13] Kuo CJ, Peles Y. Local measurement of flow boiling in structured surface microchannels. Int J Heat Mass Transfer 2007;50(23–24):4513–26. [14] Yang Fanghao, Dai Xianming, Peles Yoav, Cheng Ping, Khan Jamil, Li Chen. Flow boiling phenomena in a single annular flow regime in microchannels (I): Characterization of flow boiling heat transfer. Int J Heat Mass Transfer 2014;68:703–15. [15] Kandlikar Satish G, Widger Theodore, Kalani Ankit, Mejia Valentina. Enhanced flow boiling over open microchannels with uniform and tapered gap manifolds. J Heat Transfer 2013;135(6):061401. [16] Law Matthew, Lee Poh-Seng, Balasubramanian Karthik. Experimental investigation of flow boiling heat transfer in novel oblique-finned microchannels. Int J Heat Mass Transfer 2014;76:419–31. [17] Marcinichen Jackson Braz, Olivier Jonathan A, De Oliveira Vinicius, Thome John R. A review of on-chip micro-evaporation: experimental evaluation of liquid pumping and vapor compression driven cooling systems and control. Appl Energy 2012;92:147–61. [18] Marcinichen Jackson Braz, Wu Duan, Paredes Stephan, Thome John R, Michel Bruno. Dynamic flow control and performance comparison of different concepts of two-phase on-chip cooling cycles. Appl Energy 2014;114:179–91. [19] Garimella Suresh V, Persoons Tim, Weibel Justin, Yeh Lian-Tuu. Technological drivers in data centers and telecom systems: multiscale thermal, electrical, and energy management. Appl Energy 2013;107:66–80.
13
[20] Kim Yoon Jo, Joshi Yogendra K, Fedorov Andrei G, Lee Young-Joon, Lim Sungkyu. Thermal characterization of interlayer microfluidic cooling of threedimensional integrated circuits with nonuniform heat. J Heat Transfer 2014;132(April 2010):1–9. [21] Koo Jae-Mo, Im Sungjun, Jiang Linan, Goodson Kenneth E. Integrated microchannel cooling for three-dimensional electronic circuit architectures. J Heat Transfer 2005;127(January 2005):49. [22] Revellin Rémi, Dupont Vincent, Ursenbacher Thierry, Thome John R, Zun Iztok. Characterization of diabatic two-phase flows in microchannels: flow parameter results for R-134a in a 0.5 mm channel. Int J Multiphase Flow 2006;32(7):755–74. [23] Alam Tamanna, Lee Poh Seng, Yap Christopher R, Jin Liwen, Balasubramanian K. Experimental investigation and flow visualization to determine the optimum dimension range of microgap heat sinks. Int J Heat Mass Transfer 2012;55(25–26):7623–34. [24] Harirchian Tannaz, Garimella Suresh V. A comprehensive flow regime map for microchannel flow boiling with quantitative transition criteria. Int J Heat Mass Transfer 2010;53(13–14):2694–702. [25] Lee PS, Garimella S. Saturated flow boiling heat transfer and pressure drop in silicon microchannel arrays. Int J Heat Mass Transfer 2008;51(3–4):789–806. [26] Kew Peter A, Cornwell Keith. Correlations for the prediction of boiling heat transfer in small-diameter channels. Appl Therm Eng 1997;17(8–10):705–15. [27] Dae Whan Kim, Emil Rahim, Avram Bar-Cohen, Bongtae Han. Thermofluid characteristics of two-phase flow in micro-gap channels. In: 2008 11th IEEE intersociety conference on thermal and thermomechanical phenomena in electronic systems, I-THERM; 2008. p. 979–92. [28] Beck James V, Blackwell Ben, Clair Jr Charles RSt. Inverse heat conduction: illposed problems. Wiley Interscience Publication; 1985. [29] Beck JV, Blackwell B, Haji-Sheikh A. Comparison of some inverse heat conduction methods using experimental data. Int J Heat Mass Transfer 1996;39(17):3649–57. [30] Kandlikar Satish G. Heat transfer mechanisms during flow boiling in microchannels. J Heat Transfer 2004;126(1):8. [31] Thome J, Dupont V, Jacobi Anthony M. Heat transfer model for evaporation in microchannels. Part I: Presentation of the model. Int J Heat Mass Transfer 2004;47(14–16):3375–85.
Please cite this article in press as: Jagirdar M, Lee PS. A diagnostic tool for detection of flow-regimes in a microchannel using transient wall temperature signal. Appl Energy (2016), http://dx.doi.org/10.1016/j.apenergy.2015.12.111