Master’s Thesis: Integrating Convolutional Neural Networks with Finite Cell Method for Efficient Computational Modeling of Unit Cells

Institut

Professur für Computational Mechanics (TUM-ED)

Typ

Masterarbeit /

Inhalt

theoretisch /  

Beschreibung

For many solid mechanics problems, high-fidelity simulations (HFS)  can capture the most important physical phenomena. The finite element method (FEM) is a well-known time-tested method used to perform the HFS to predict the behavior of the relevant quantities of the design. An alternative formulation of FEM is the finite cell method (FCM). The FCM combines high-order elements with an unfitted domain, decoupling the physical domain from the discretization solution. Even for complex geometries, FCM generates a simple structured mesh and a solution with high convergence rates, avoiding the time-consuming meshing process without sacrificing accuracy. Additionally, because of the FCM discretization, the same underlying mesh can be used across multiple geometries. These characteristics make FCM attractive for the simulation of cellular structures' unit cells (UC). However, HFS are computationally expensive, creating a bottleneck in the design process. This is particularly true during the optimization phase, where multiple design variables are varied to find the best performance. To evaluate the performance of each design, HFS are carried out. A parametric reduced order model (pROM) can be generated for each new simulation to accelerate the process.

 

Scope of the work:

 

• Understand the basic principles of the finite cell method.
• Research on different types of unit cells for cellular structures.
• Literature review on the state-of-the-art convolutional neural networks and similar architectures for solid mechanics.
• Implementation of different deep learning methods, whose inputs come from FCM simulations.
• Comparative analysis of the accuracy and the computational cost of the obtained results.
• Documentation and presentation of results.

 

If you are interested in this project or have further questions, please email juan.vargas<script>document.write('@');</script>

<noscript>(at)</noscript>tum.de with only your CV and transcript of records.

Voraussetzungen

We are looking for candidates who:

 

• Completed Bachelor’s degree in Mechanical Engineering, Civil Engineering, or a related field
• Strong programming skills, particularly in Python
• Ability to work independently and proactively
• Very good command of English (written and spoken)
• Strong interest in finite element methods (FEM) and machine learning approaches

 

Nice to have experience/knowledge in:

• Git
• OpenRadioss
• Model Order Reduction
• PyTorch

 

Möglicher Beginn

sofort

Kontakt

Juan Angelo Vargas Fajardo
juan.vargastum.de

Ausschreibung