VACANCIES
PhD position: Bio-composites as coating for WAAM metal: deep-learning-enhanced multiscale modelling of moisture diffusion
Part of DurAMat (A Marie Skłodowska-Curie Actions – Doctoral Network (MSCA-DN) project). More information about DurAMat can be found here.
Applications should be sent through the university application portal and DurAMat website.
MASTER THESIS PROJECTS
Master thesis: Modelling deformation behavior of amorphous solids
Purpose: The aim of the project is to theoretically/numerically study the deformation behavior of amorphous solids, including macroscopic stress-strain response and microscopic shear band development. The effects of distribution of initial stability will be considered in the modelling process.
Image: Physical Review E, 98 (4), art. no. 040901 (2018).
Supervisors:
Mohsen Mirkhalaf (mohsen.mirkhalaf@physics.gu.se)
Johannes Hofmann, Department of Physics, University of Gothenburg, (johannes.hofmann@physics.gu.se)
Master/Bachelor thesis: Thermal conductivity of short fiber reinforced composites
Description: Computational homogenization, a modeling approach in composite materials, employs a Representative Volume Element (RVE) to capture micro-structural properties. It typically uses Finite Element Method for accurate results but faces high computational costs and challenges in generating accurate RVEs. An alternative method involves mean-field models, which consider average properties of micro-structures. In this project, we'll utilize both approaches.
Image: Polymers 2022, 14(16), 3360.
Supervisors:
Mohsen Mirkhalaf (mohsen.mirkhalaf@physics.gu.se)
Mats Granath, Department of Physics, University of Gothenburg, (mats.granath@physics.gu.se)
Master thesis: Developing deep-learning-enhanced structure-property relationship for hydrogels
Description: The project's goal is to create a deep-learning model for hydrogel analysis and potential design. Existing characterization data and experimental results will be integrated with artificial neural networks. Characterization data will be input, and material properties will be output. This efficient approach can enhance hydrogel characterization and aid in designing micro- and nano-structures.
Supervisors:
Mohsen Mirkhalaf (mohsen.mirkhalaf@physics.gu.se)
Jenny Malmström, Department of Chemical and Materials Engineering, University of Auckland, Auckland, New Zealand (j.malmstrom@auckland.ac.nz)
Master thesis: Multi-fidelity Bayesian deep-learning for short fiber reinforced composites
Description: In this project, we aim to develop an ANN model for heterogeneous SFRC microstructures, including imperfect fiber-matrix bonding with a cohesive zone. The challenge lies in generating sufficient data due to lengthy RVE simulations. We plan to address this using Transfer learning, leveraging a pre-trained RNN model for path-dependent elasto-plastic response of SFRCs. This approach mitigates the need for vast amounts of high-fidelity simulation data. Bayesian deep-learning will be employed to manage uncertainties arising from different RVE realizations.
Supervisors:
Mohsen Mirkhalaf (mohsen.mirkhalaf@physics.gu.se)
Hampus Linander, Department of Mathematical Sciences, Chalmers, (hampus.linander@chalmers.se)
Master thesis: Generative Adversarial Networks for generating multi-dimensional data for composites
Description: In this project, we investigate the use of Generative Adversarial Networks (GANs) in the context of micro-mechanical properties of elastic short fiber reinforced composites (SFRCs). GANs consist of two networks, a generator creating synthetic data and a discriminator distinguishing real from fake data. WE have already developed an ANN model for SFRC elastic properties, generating a comprehensive dataset using a Finite-Element-based method. In this project, GANs will be applied to a portion of this dataset to generate synthetic data resembling real patterns.
Supervisors:
Mohsen Mirkhalaf (mohsen.mirkhalaf@physics.gu.se)
Giovanni Volpe, Department of Physics, University of Gothenburg, (giovanni.volpe@physics.gu.se)