The rapid growth and scaling of electronics are causing more severe thermal management challenges. For example, the high-performance computing processors are driving the data center power density to unprecedented levels, approaching the limit of conventional air cooling. In electric vehicles (EVs) and hybrid EVs, the power conversion electronics are integrated into a compact space, leading to ultra-high heat fluxes to dissipate. Among the available thermal management mechanisms, two-phase cooling that involves the phase-change process of the working fluid can maintain electronic devices at safe operating temperatures by taking advantage of the high latent heat of the fluid. Particularly, pool boiling plays a critical role in the two-phase immersion cooling of servers and other IT hardware, integrated cooling for three-dimensional electronic packaging, cooling of the core, and used fuel in nuclear reactors. Two-phase coolers are limited by instabilities such as the critical heat flux (CHF). At the critical heat flux, the temperature increases. It is important to be able to identify the CHF in order to prevent overheating. We aim to develop and compare boiling image classification models to distinguish between 2 boiling regimes. We will leverage principal component analysis (PCA) and K-means clustering to investigate the key differences between bubbles during nucleate boiling (pre-CHF) and transition boiling (post-CHF). We will also compare the results of the unsupervised learning model against popular supervised learning models that have been used for boiling regime classification in existing studies, such as convolutional neural networks, multiplayer perceptrons, and transformers. We successfully created 4 supervised and 1 unsupervised learning models to distinguish between the two types of boiling images.
- Award ID(s):
- 1738793
- NSF-PAR ID:
- 10338741
- Date Published:
- Journal Name:
- 2022 38th Semiconductor Thermal Measurement, Modeling & Management Symposium (SEMI-THERM)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Abstract -
more » « less
Abstract Jet impingement can be particularly effective for removing high heat fluxes from local hotspots. Two-phase jet impingement cooling combines the advantage of both the nucleate boiling heat transfer with the single-phase sensible cooling. This study investigates two-phase submerged jet impingement cooling of local hotspots generated by a diode laser in a 100 nm thick Hafnium (Hf) thin-film on glass. The jet/nozzle diameter is ∼1.2 mm and the normal distance between the nozzle outlet and the heated surface is ∼3.2 mm. Novec 7100 is used as the coolant and the Reynolds numbers at the jet nozzle outlet range from 250 to 5000. The hotspot area is ∼ 0.06 mm2 and the applied hotspot-to-jet heat flux ranges from 20 W/cm2 to 220 W/cm2. This heat flux range facilitates studies of both the single-phase and two-phase heat transport mechanisms for heat fluxes up to critical heat flux (CHF). The temporal evolution of the temperature distribution of the laser heated surface is measured using infrared (IR) thermometry. This study also investigates the nucleate boiling regime as a function of the distance between the hotspot center and the jet stagnation point. For example, when the hotspot center and the jet are co-aligned (x/D = 0), the CHF is found to be ∼ 177 W/cm2 at Re ∼ 5000 with a corresponding heat transfer coefficient of ∼58 kW/m2.K. While the CHF is ∼ 130 W/cm2 at Re ∼ 5000 with a jet-to-hotspot offset of x/D ≈ 4.2.
-
Jet impingement can be particularly effective for removing high heat fluxes from local hotspots. Two-phase jet impingement cooling combines the advantages of both the nucleate boiling heat transfer with the single-phase sensible cooling. This study investigates two-phase confined jet impingement cooling of local, laser-generated hotspots in a 100 nm thick Hafnium (Hf) thin film on glass. The jet/nozzle diameter is ∼1.2 mm and the normal distance between the nozzle outlet and the heated surface is ∼3.2 mm. The jet coolants studied are FC 72, Novec 7200, and Ethanol with jet nozzle outlet Reynolds numbers ranging from 250 to 5000. The hotspot area is ∼0.06 mm2 and the applied hotspot-to-jet heat fluxes range from 20 W/cm2 to 350 W/cm2. This heat flux range facilitates studies of both the single-phase and two-phase heat transport mechanisms for heat fluxes up to critical heat flux (CHF). The temporal evolution of the temperature distribution of the laser-heated surface is measured using infrared (IR) thermometry. This study focuses on the stagnation point heat transfer - i.e., the jet potential core is co-aligned with the hotspot center. For ethanol, the CHF is ∼315 W/cm2 at Re ∼ 1338 with a corresponding heat transfer coefficient of h ∼ 102 kW/m2·K. For FC 72, the CHF is ∼94 W/cm2 at Re ∼ 5000 with a corresponding h ∼ 56 kW/m2·K. And for Novec 7200, the CHF is ∼108 W/cm2 at Re ∼ 4600 with a corresponding h ∼ 50 kW/m2·K.more » « less
-
more » « less
Abstract Real-time thermal monitoring and regulation are critical to the mitigation of thermal runaways and device failures in two-phase cooling systems. Compared to conventional approaches that rely on the Joule effect, thermal gradient or transverse thermoelectric effect, acoustic emission (AE)-based remote sensing is more promising for robust and non-intrusive thermal monitoring. Nevertheless, due to the high stochasticity and noise of acoustic signals, existing implementations of AE in thermal systems have been limited to qualitative state monitoring. In this paper, we present a technology for real-time heat flux quantification during two-phase cooling by coupling acoustic sensing using hydrophones and condenser microphones and regression-based machine learning frameworks. These frameworks integrate a fast Fourier transform feature extraction algorithm with regressors, i.e., Gaussian process regressor and multilayer perceptron regressor for heat flux predictions. The acoustic signals and heat fluxes are collected from pool boiling tests under transient heat loads. It is shown that both hydrophone and condenser microphone signals are successful in predicting heat flux. Multiple models are trained and compared some using only one form of acoustic data while others combine both acoustic types (i.e., hydrophone and microphone) in fusion ML models (i.e., early, joint, late). The models using only hydrophone data are shown to perform better than the models using only microphone data. Also, some forms of fusion are shown to have better performance than either of the single input data type models. This AE-ML technology is demonstrated for accurate heatflux quantification. As such, this work will not only lead to a light, low-cost, and non-contact thermal measurement technology but also a new perspective for the physical explanation of bubble dynamics during boiling.
-
Increasing power densities in data centers due to the rise of Artificial Intelligence (AI), high-performance computing (HPC) and machine learning compel engineers to develop new cooling strategies and designs for high-density data centers. Two-phase cooling is one of the promising technologies which exploits the latent heat of the fluid. This technology is much more effective in removing high heat fluxes than when using the sensible heat of fluid and requires lower coolant flow rates. The latent heat also implies more uniformity in the temperature of a heated surface. Despite the benefits of two-phase cooling, the phase change adds complexities to a system when multiple evaporators (exposed to different heat fluxes potentially) are connected to one coolant distribution unit (CDU). In this paper, a commercial pumped two-phase cooling system is investigated in a rack level. Seventeen 2-rack unit (RU) servers from two distinct models are retrofitted and deployed in the rack. The flow rate and pressure distribution across the rack are studied in various filling ratios. Also, investigated is the transient behavior of the cooling system due to a step change in the information technology (IT) load.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
