PhDs

 

 


D Kafka, 2020."Automated learning rates in machine learning for dynamic mini-batch sub-sampled losses"

Learning rate schedule parameters remain some of the most sensitive hyperparameters in machine

learning, as well as being challenging to resolve, in particular when mini-batch subsampling is considered. Mini-batch sub-sampling (MBSS) can be conducted in a number of ways, each with their own implications on the smoothness and continuity of the underlying loss function. In this study, dynamic MBSS, often applied in approximate optimization, is considered for neural network training. For dynamic MBSS, the mini-batch is updated for every function and gradient evaluation of the loss and gradient functions. The implication is that the sampling error between mini-batches changes abruptly, resulting in non-smooth and discontinuous loss functions. This study proposes an approach to automatically resolve learning rates for dynamic MBSS loss functions using gradient-only line searches (GOLS) over fifteen orders of magnitude. A systematic study is performed, which investigates the characteristics and the in ability of GOLS to resolve learning rates. GOLS are shown to compare favourably against the state-of-the-art probabilistic line search for dynamic MBSS loss functions. Matlab and PyTorch 1.0 implementations of GOLS are available for both practical training of neural networks as well as a research tool to investigate dynamic MBSS loss functions.

Supervisor: Prof DN Wilke


S M M Osman, 2020. "Experimental investigation into convection heat transfer in the transition flow regime by using nanofluids in a rectangular channel"

The growing demand for energy worldwide requires attention to the design and operating of heat exchangers and thermal devices to utilise and save thermal energy. There is a need to find new heat transport fluids with better heat transfer properties to increase convective heat transfer, and nanofluids are good alternatives to conventional heat transport fluids. Although extensive research has been done on the properties of nanofluids in recent decades, there is still a lack of research on convection heat transfer involving nanofluids, particularly in the transitional flow regime. This study focused on the application of nanofluids in heat exchangers as heat transport fluids by investigating forced convective heat transfer of alumina-water and titanium dioxide-water nanofluids prepared by using the one-step method. The particle size used was 46 nm and 42 nm for the aluminium oxide and the titanium dioxide respectively. Uniform heat flux boundary conditions were used by uniformly heating the rectangular channel electrically. Nanofluids with volume concentrations of 0.3, 0.5 and 1% were used for the alumina-water nanofluids, and volume concentrations of 0.3, 0.5, 0.7 and 1% were used for the titanium dioxide-water nanofluids. The viscosity of the nanofluids under investigation was determined experimentally, while the thermal conductivity and other properties were predicted by using suitable correlations from the literature. A Reynolds number range of 200 to 7 000 was covered, and the investigated flow rates included the laminar and turbulent flow regimes, as well as the transition regime from laminar to turbulent flow. Temperatures and pressure drops were measured to evaluate heat transfer coefficients, Nusselt numbers and pressure drop coefficients. Heat transfer and hydrodynamic characteristics in the transition flow regime were carefully studied and compared with those in the transition regime when flowing pure water in the same test section.

 

The study also investigated another approach of enhancing heat transfer in heat exchangers by increasing the heat transfer area of the heat exchanger itself, and this was done by filling the rectangular test section with porous media to increase the heat transfer surface area and thus enhance heat transfer. Hence in this study, the effect of using porous media was also studied by filling the rectangular test section with high-porosity nickel foam. The permeability of the used nickel foam was determined by conducting pressure drop measurements through the nickel foam in the test section, and heat transfer and pressure drop parameters were measured and compared with those in the empty test section.

 

The results showed that all the nanofluids used enhanced heat transfer, particularly in the transition flow regime. The 1.0% volume concentration alumina nanofluid showed maximum enhancement of the heat transfer coefficient, with values of 54% and 11% in the turbulent regime. The maximum enhancement of the heat transfer coefficient was 29.3% in the transition regime for the 1.0% volume concentration titanium dioxide-water nanofluid. The thermal performance factor in the transition flow regime was observed to be better than that in the turbulent and laminar flow regimes for all the nanofluids.

 

The results of the nickel foam test section showed that the values of the friction coefficient were 24.5 times higher than the values of the empty test section, and the Nusselt number was observed to be three times higher when using nickel foam than without foam in the test section. No transition regime was observed for the foam-filled test section on either the heat transfer results or the pressure drop results; however, transition from laminar to turbulent was found for the test section without foam. The results of the thermal factor of the foam-filled test section showed a thermal performance factor higher than unity through the entire Reynolds number range of 2 000 to 6 500, with better thermal performance factor at lower Reynolds number.

Supervisor: Prof Mohsen Sharifpur

Co-supervisor: Prof Josua Meyer


S. Giwa,2020."Investigation into thermal-fluid properties of hybrid ferrofluids as heat transfer "fluids.

Over two decades of extensive research on nanofluids have established them as a better cooling media than traditional fluids such as ethylene glycol (EG) and water. Recently, hybrid nanofluids have emerged as advanced thermal transport media with improved thermal and fluid properties relative to nanofluids. Experimentally, limited studies have been carried out on the thermo- and thermomagnetic convection heat transfer of nanofluids in cavities. However, there is a dearth of documentation on the thermo- and thermomagnetic convection of hybrid nanofluids in cavities in the public domain.

In this study, the thermo-convection heat transfer (Qav) performance of three magnetic hybrid nanofluids (MHNFs) contained in a rectangular cavity was experimentally investigated with and without magnetic stimuli. Aqueous MWCNT-ferrofluid (AMF) [MWCNT-Fe2O3/deionised water (DIW)], aqueous Al2O3-ferrofluid (AAF) [Al2O3-Fe2O3/DIW] and bi-aqueous Al2O3-ferrofluid (BAAF) [Al2O3-Fe2O3/EG-DIW] were formulated for volume concentrations (

- Author

Copyright © University of Pretoria 2024. All rights reserved.

COVID-19 Corona Virus South African Resource Portal

To contact the University during the COVID-19 lockdown, please send an email to [email protected]

FAQ's Email Us Virtual Campus Share Cookie Preferences