School of Mechanical Engineering Seminar
Monday, November 12, 2018 at 14:00
Wolfson Building of Mechanical Engineering, Room 206

Heat transfer During Phase Separation of Partially Miscible Liquid-Liquid Systems in Mini-Channels

by

Idan Shem Tov

This research was carried out in the School of Mechanical Engineering under the supervision of Prof. Amos Ullmann

and Prof. Neima Brauner

The increasing power density of miniaturized equipment requires advanced cooling technologies to maintain allowable temperatures and ensure safe operation. An experimental study has been conducted to explore the possibility of enhancing the convective heat transfer in mini-channels by temperature-induced phase separation of partially miscible liquid-liquid mixtures. The convective heat transfer of three different mixtures: triethylamine-water, propylene glycol propyl ether-water and lutidine-water was tested under conditions of constant wall heat fluxes. The phase separation of the mixtures showed up to 120% cooling enhancement compared to single-phase flow, up to 25% enhancement compared to water and up to 150% enhancement compared to the standard coolant of ethylene glycol-water mixture. The best results were achieved with triethylamine-water mixture under low Re number and high heating power. The creation and migration of droplets when phase-separation takes place and the resulting local fluid mixing in the heated wall vicinity, was found to be the main mechanism leading to the HTC enhancement. The local mixing can be represented as higher effective fluid thermal conductivity (k). The effective k values, circulated via CFD simulations, are up to 2.1 larger than those of the corresponding single-phase flow values. In parallel to this research, USA group, in collaboration with our research group, showed that the effect of HTC enchantment is preserved upon downscaling to micro-channel.

School of Mechanical Engineering Seminar
Wednesday, December 03, 2018 at 14:00
Wolfson Building of Mechanical Engineering, Room 206

Simulating photothermal actuation in resonant micro bridges

 Gilad Goldenberg

 MSc Student of Dr. Yoav Linzon

This work comes from the application need of optimizing the photothermal actuation in micro bridges that change their resonance frequencies with changes in mass due to gas absorption. The micro bridges are activated using 405[nm] (blue) laser close to resonance frequency, the actual bridge frequencies are sensed with red probe laser.

The research in this work is performed using Comsol Multiphysics, modeling the polysilicon micro bridges of various lengths (20 um, 30 um and 35 um) and their oscillation using thermal actuation in periodic and aperiodic actuation.

The research main results are:

1.    The expected displacements in 1.5um wide, 0.14um thick and 20um, 30um and 35 um long polysilicon bridges in the time domain.

2.    The expected maximum displacements in 1.5um wide, 0.14um thick and 20um, 30um and 35 um long polysilicon bridges in the frequency domain.

3.    The Q factors derived from the frequency domain analysis.

4.    An analysis of the bridge heating duty cycle effect on the maximum displacement.

5.    An analysis of unsynchronized activation.

Multidimensional diffusion MRI

Author: D. Topgaard
Affiliation: Department of Chemistry, Lund University, Sweden
E-mail: daniel.topgaard@fkem1.lu.se

The use Diffusion MRI is an excellent method for detecting subtle microscopic changes of the living human brain, but often fails to assign the experimental observations to specific structural properties such as cell density, size, shape, or orientation. When attempting to solve this problem, we have chosen to disregard essentially all previous work in the field of diffusion MRI, and instead translate data acquisition and processing schemes from multidimensional solid-state NMR spectroscopy [1, 2]. Key elements of our approach are free gradient waveforms, q-vector trajectories, b-tensors, and correlations between isotropic and directional diffusion encoding. By approximating the water displacement probability as a sum of anisotropic Gaussians, the voxel composition can be reported as a diffusion tensor distribution where each component of the distribution corresponds to a distinct tissue environment. Our new methods yield estimates of the complete diffusion tensor distribution or well-defined statistical properties thereof, such as the mean and variance of isotropic diffusivities, mean-square anisotropy, and orientational order parameter, which derive from analogous parameters in solid-state NMR and can be related to the structural properties of the tissue. This presentation will give an overview of the new methods, including basic physical principles, pulse sequences, and data processing, as well as examples of applications in healthy and diseased brain.

[1] Schmidt-Rohr K, Spiess HW. Multidimensional solid-state NMR and polymers. San Diego: Academic Press; 1994.
[2] Topgaard D. Multidimensional diffusion MRI. J Magn Reson 2017;275:98-113. https://dx.doi.org/10.1016/j.jmr.2016.12.007

(The talk will be given in English)

Speaker:     Dr. Kfir Levy
                   Institute for Machine Learning at ETH Zurich

SUNDAY, November 11^th, 2018
15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

Beyond SGD: Data Adaptive Methods for Machine Learning 

Abstract

The tremendous success of the Machine Learning paradigm heavily relies on the development of powerful optimization methods. The canonical algorithm for training learning models is SGD (Stochastic Gradient Descent), yet this method has its limitations. It is often unable to exploit useful statistical/geometric structure, it might degrade upon encountering prevalent non-convex phenomena, and it is hard to parallelize. In this talk, I will discuss an ongoing line of research where we develop alternative methods that resolve some of SGD’s limitations. The methods that I describe are as efficient as SGD, and implicitly adapt to the underlying structure of the problem in a data-dependent manner.
In the first part of the talk, I will discuss a method that is able to take advantage of hard/easy training samples. In the second part, I will discuss a method that enables an efficient parallelization of SGD. Finally, I will briefly describe a method that implicitly adapts to the smoothness and noise properties of the learning objective.

Bio
Kfir Levy is a post-doctoral fellow in the Institute for Machine Learning at ETH Zurich, advised by Prof. Andreas Krause. Kfir’s research is focused on Machine Learning and Stochastic Optimization, with a special interest in designing universal methods that apply to a wide class of learning scenarios. He is a recipient of the ETH Zurich Postdoctoral fellowship, as well as the Irwin and Joan Jacobs fellowship for excellence in research. Kfir received his degrees from the Technion—Israel Institute of Technology. He was advised by Prof. Elad Hazan during his Ph.D. and by Prof. Nahum Shimkin during his Master’s.‏