Crystal Structure Detection in Microgravity Experiments using Machine Learning Methods - Bachelor Thesis

Aug 3, 2026

Upload by Levi

This bachelor thesis analyzes machine learning methods for crystal structure detection in microgravity experiments. The objective is to identify crystal structures from particles using a 2D projection, while extending and modifying an existing algorithm originally designed for 3D data. Extensive testing validates accuracy and computational efficiency across multiple experimental conditions, including variations in experimental setup parameters. The work also investigates how these methods could be integrated to improve an overall experimental workflow. Finally, the thesis compares modified implementations and discusses alternative approaches and future research directions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages73

Reading Time3 h 4 min

AccessFree to download

Introduction

Structure

Crystalline Solids and Crystal Structures

Description of Crystal Structures

Microgravity Experiments

Physical Experiment for a 2D Projection

Methods

Feature Vector

Multilayer Perceptron

Sample Data Generation

Procedure to obtain 2D from 3D Data

Training and Evaluation

Parameters of the Experimental Setup

Results

Slice Thickness

Slice Rotation

Circle Range

Discussion

Experimental Setup Parameters

Comparison of Neighborhood Definitions

Challenges and Limitations

Future Research

Conclusion

Appendix

Mixed Crystal Signature (Refactored)

Projection

Notebooks

Faculty of Applied Computer Sciences and Biosciences BACHELOR THESIS Crystal Structure Detection in Microgravity Experiments using Machine Learning Methods Author: Ekaterina Filonenko Course of Study: Applied Mathematics Seminar Group: MA20w1-B First Examiner: Prof. Dr. rer. nat. Dr.-Ing. habil. Florian Zaussinger Second Examiner: Dr. Marika Kaden Submission: Mittweida, 22.09.2024 Defense/Evaluation: Mittweida, 2024 Bibliographic Description Filonenko, Ekaterina: Crystal Structure Detection in Microgravity Experiments using Machine Learning Methods.– 2024.– 24 S. Mittweida, Hochschule Mittweida – University of Applied Sciences, Faculty of Applied Computer Sciences and Biosciences, Bachelor Thesis, 2024 . Referat In this thesis, we analyze Machine Learning methods for crystal structure detection in microgravity experiments. Our objective is to identify crystal structures of the particles by a 2D projection. We modify an already existing algorithm for 3D structures [4] . Through extensive testing, we validate the accuracy and eﬃciency of our approach in various experimental conditions. Additionally, we explore the potential for integrating these methods to enhance the overall experimental workﬂow. Finally, we demonstrate the advantages of our modiﬁed implementations and discuss other possible approaches. Contents Contents I List of Figures III Source code IV Acknowledgment V 1 Introduction 1 1.1 Structure ........................................... 1 1.2 Crystalline Solids and Crystal Structures ......................... 2 1.3 Description of Crystal Structures ............................. 3 1.3.1 Body-Centered Cubic (BCC) Structure .......................... 3 1.3.2 Face-Centered Cubic (FCC) Structure ........................... 4 1.3.3 Hexagonal Close-Packed (HCP) Structure ........................ 4 1.4 Microgravity Experiments ................................. 5 1.5 Physical Experiment for a 2D Projection ......................... 6 2 Methods 7 2.1 Feature Vector ....................................... 7 2.2 Multilayer Perceptron ................................... 8 2.3 Sample Data Generation .................................. 8 2.4 Procedure to obtain 2D from 3D Data .......................... 9 2.5 Training and Evaluation .................................. 14 2.6 Parameters of the Experimental Setup .......................... 15 3 Results 16 3.1 Slice Thickness ....................................... 16 3.2 Slice Rotation ........................................ 17 3.3 Circle Range ......................................... 19 4 Discussion 22 4.1 Experimental Setup Parameters ............................. 22 4.2 Comparison of Neighborhood Deﬁnitions ........................ 22 4.3 Challenges and Limitations ................................ 23 4.4 Future Research ...................................... 23 4.5 Conclusion .......................................... 23 Appendix 25 A Mixed Crystal Signature (Refactored) 25 A.1 Signaure ........................................... 25 A.2 Data Generator ....................................... 28 A.3 Crystal Analyzer ....................................... 31 B Projection 35 B.1 Plane ............................................. 35 B.2 Slice ............................................. 36 C Notebooks 38 C.1 Slice ............................................. 39 C.2 Experimental Parameters ................................. 47 C.3 Experimental Parameters for Area Neighborhood ................... 52 Bibliography 62 Statutory Declaration in Lieu of an Oath 65 List of Figures 1.1 Schematic diagram highlighting the differences between the solid (left), liquid (centre) and gaseous (right) states of matter [25] ....................... 2 1.2 The bcc unit cell that consists of a net total of two atoms (left) and a schematic diagram of the bcc unit cell that has

FAQs

How does the thesis detect crystal structures in microgravity experiments?

It applies machine learning to infer crystal structures from particle observations represented as a 2D projection, rather than requiring full 3D information. The approach modifies an existing 3D algorithm to support this 2D setting.

What methods and model are used for the detection task?

The thesis builds a feature vector representation and uses a multilayer perceptron. It also includes procedures for generating sample data and transforming 3D data into 2D slices for training and evaluation.

Which experimental setup factors are studied in the results?

The results focus on slice thickness, slice rotation, and the circle range. These parameters are used to test how changes in experimental conditions affect detection accuracy and efficiency.

Aug 3, 2026

Upload by Levi