Student Posters Repository

Molecular dynamics (MD) simulations of proteins produce large data sets - long trajectories of atomic coordinates - and provide a representation of the sampling of a given molecule’s structural ensemble. A deep quantitative analysis using advanced machine learning techniques is a means to interpret MD trajectories. To visualize the conformational space of the molecule and properly identify conformational states, we suggest combining clustering methods and dimensionality reduction algorithms. We investigate different choices of features to represent individual structures, clustering algorithms, similarity metric, and methods to assign the number of clusters. 

Through the large amounts of mass and energy exchanged with the atmosphere, tropical ecosystems play an important role in the global climate and carbon cycle. However, seasonal variation in tropical vegetation activity is not well represented in land surface models. These models are the terrestrial component of Earth System Models, which are currently our most important tool for future climate prediction. Given the large ecosystem-atmosphere feedbacks, the incorrect simulation of seasonality in the tropics may affect climate and carbon exchange predictions from Earth System Models. In my doctoral research, I propose three projects to help improve the simulation of tropical ecosystems’ seasonality. In the first study, I looked at model simulations of intra-annual variation in gross primary productivity (GPP) and its relationship with climate at the regional level. In the second project, I am looking at the effect of changes in land cover on the seasonality of carbon and water exchange with the atmosphere. In the last project, I am testing the incorporation of different phenology formulations in a carbon model and how they affect GPP simulations for tropical forests.

Mesh creation and refinement is one of the most time-consuming steps in any CFD simulation; even automated mesh generation requires high levels of expertise and fine-tuning. This project attempts to circumvent some of this complexity by leveraging deep convolutional neural networks to predict mesh densities for arbitrary geometries.

An automated pipeline was created to generate random geometries and run CFD simulations, iteratively performing targetted mesh refinement utilizing adjoint sensitivies. A comprehensive 6TB dataset consisting of 65,000 geometry-mesh pairs was assembled via an extensive post-processing and evaluation setup.

Current literature indicated that the UNet architecture extended by Thuerey et al. was suitable to predict flow-related quantities, but had never been used for mesh prediction. In this work, we present a deep, fully convolutional network that estimates mesh densities based off geometry data. The most recent model, tuned with network depth, channel size and kernel size, had an accuracy of 98% on our testing dataset.

The current pipeline provides a proof-of-concept that convolutional neural networks can, for specific use-cases, generate accurate mesh densities without the need manual fine-tuning. Such a product, if further tuned and extended, can provide significant time savings in future CFD workflows, completely independent of personnel expertise.

Spatially-developing turbulent boundary layers (SDTBL) are very challenging to numerically model by virtue of appropriate and realistic turbulent inflow information needed. In fact, the problem becomes harder if the idea is to account for compressibility effects, i.e. high Mach-number flows. Direct Numerical Simulation (DNS) is a promising numerical tool that resolve all scales according to the analyzed problem. In this study, DNS of turbulent spatially-developing boundary layers in the supersonic regime is performed using the standard Smagorinsky model over an isothermal plate at a Mach number of 2.5. The code used (PHASTA) is based on a Finite-Element Approach and has been parallelized on MPI C++/Fortran90-95.

The first detection of gravitational waves (GWs) from a binary neutron star (BNS) merger, GW170817, in August of 2017 signaled the beginning of the era of GW-multimessenger astronomy.
Our work emphasizes to simulate this kind of signals in full General Relativity using HPC and analysing the data afterwards to acquire knowledge regarding the underlying physics. Neutron stars
are among the most extreme entities in the universe. Their central density exceeds the nucleus density of an atom. The physics of the inner cores of Neutron Stars is unknown to Physics yet.
Gravitational Waves from such events give us the opportunity to extract information about those extreme events in the universe. After analysing full Numerical Relativity simulations we produced
empirical relations for Neutron star characteristics. Finally, we used Machine Learning algorithms and proved the theoretical classification that was proposed by Bauswein & Stergioulas 2015 PRD.

The nucleotide excision repair (NER) DNA repair pathway is responsible for drug resistance against gold standard DNA damaging chemotherapies such as platinum-based drugs. I will present the computer-aided drug design pipeline implemented at the Alberta DNA Repair Consortium (ALDRECON) to identify new drugs inhibitors for key NER proteins, to improve the efficacy of current chemotherapies. The pipeline is based on molecular dynamics simulations and large-scale virtual screening experiments, followed by in-house biological validations of the in-silico hits, and has been successfully applied to the discovery of a promising pre-clinical drug candidate for colorectal cancer treatment.

Objectives:

To study the Electrohydrodynamics (EHD) and Thermo-capillary flows impact on rising and break-up mechanisms of droplets (migration and deformation) using a multiphase Incompressible Smoothed Particle Hydrodynamics (ISPH) method, with many industrial applications such as oil and gas industry. 

Methodology:

SPH Incompressible multiphase flow solver 2D MPI parallel code.

Large eddy simulations (LES) have proven to be a powerful tool in modeling turbulent premixed flames because they resolve some of the energy carrying eddies unlike RANS. However, the LES sub-filter-scale (SFS) modelling uncertainties are of particular importance in premixed turbulent combustion because the entire reaction occurs within the sub-grid scales. For this reason, the LES SFS model and and combustion model should be extensively validated and verified. A laboratory scale turbulent premixed methane-air Bunsen flame is simulated and compared to experimental results. A LES using a WALE SFS model coupled with the FPI-PCM combustion model is used. The FPI approach is a flamelet type model based on premixed laminar flamelets with chemistry tabulation in terms of a transported progress-of-reaction variable with a presumed PDF to model chemistry-turbulence interactions. Recent experimental measurements have shown a drastic increase in the turbulent intensity in the region near the burner walls. Previous numerical studies did not represent this increased turbulent intensity and it is believed that this resulted in the persistent over prediction of the flame height. To accurately represent the experimentally measured burner inflow turbulence while also controlling computational costs, an equilibrium stress near-wall model will be used to recreate the experimentally observed turbulence. The numerical flame height will be compared to the experimental results to determine if near wall effect can account for previous discrepancies between experimental and numerical predictions.

We study the common-envelope evolution of a 1.2 Msol AGB star. By including recombination energy, we succeed in ejecting and gravitationally unbinding a substantial part of the envelope. Hydrodynamical simulations of giant stars are difficult due to the wide range in spatial scales and in dynamical timescales. Especially the weakly bound envelope of the AGB star has caused problems previously. After mapping a one-dimensional stellar profile from mesa onto the grid and relaxing the giant’s atmosphere to obtain a stable model, we perform a three-dimensional binary simulation of the hydrodynamics with an adaptive mesh code, where dynamical instabilities can be observed. Considering different companion star masses, we discuss the ejection of the envelope. The dynamics are governed by dynamical friction and shocks. Taking into account recombination energy, the unbound mass fraction can be increased significantly, ejecting the envelope completely. The common-envelope phase thus can be considered as a mechanism leading to close binary systems, driven by recombination.