Principal Component Analysis for STEM-EDS images: Optimal collection parameters
(2020) KOOL01 20201Centre for Analysis and Synthesis
- Abstract
- STEM-EDS is a common technique used for compositional mapping in materials. The vasts amounts of raw data produced by STEM EDS is suitable for advanced analysis using methods in unsupervised machine learning. One powerful method is Principal Component Analysis (PCA) which can automatically discover significant chemical correlations in a sample. Although more powerful than classical analysis, there are limits to how weak chemical correlations PCA can detect. Faced with a limited data collection time, it is of great importance to know how data collection parameters in the STEM affects PCA performance. In this work, the effect of image size and X-ray counts has been investigated using simulated STEM EDS datasets. It was found that decreasing... (More)
- STEM-EDS is a common technique used for compositional mapping in materials. The vasts amounts of raw data produced by STEM EDS is suitable for advanced analysis using methods in unsupervised machine learning. One powerful method is Principal Component Analysis (PCA) which can automatically discover significant chemical correlations in a sample. Although more powerful than classical analysis, there are limits to how weak chemical correlations PCA can detect. Faced with a limited data collection time, it is of great importance to know how data collection parameters in the STEM affects PCA performance. In this work, the effect of image size and X-ray counts has been investigated using simulated STEM EDS datasets. It was found that decreasing overall image size while increasing counts per pixel both increases the chances of discovering weak chemical correlations and improves the overall PCA accuracy. Furthermore, use of varimax factor rotations has been investigated as a way to improve interpretability of PCA results. Finally, a real sample originating from a tungsten-carbide cutting tool which had been turned in a titanium alloy was analysed using the presented methods. This analysis confirmed the existence of an expected pure tungsten phase but also lead to the discovery of an unexpected ridge-like structure in the adhered alloy material. (Less)
- Popular Abstract
- In the field of materials science, various methods are available for material analysis and characterisation. In the study of wear on hard tools used for metal machining, structural and chemical analysis of the used tool is essential for understanding how it wears down during machining.
One specific tool that can be used for analysis is the electron microscope which uses focused beams of electrons instead of light. In this type of microscope, electrons interact with the sample under study similarly to a classical microscope, but due to optical limitations of light, electron microscopes can reach a much higher resolution. This is needed when studying the very small structural and chemical changes that has occurred in the cutting tool. The... (More) - In the field of materials science, various methods are available for material analysis and characterisation. In the study of wear on hard tools used for metal machining, structural and chemical analysis of the used tool is essential for understanding how it wears down during machining.
One specific tool that can be used for analysis is the electron microscope which uses focused beams of electrons instead of light. In this type of microscope, electrons interact with the sample under study similarly to a classical microscope, but due to optical limitations of light, electron microscopes can reach a much higher resolution. This is needed when studying the very small structural and chemical changes that has occurred in the cutting tool. The type of electron microscope used in this work is the Scanning Transmission Electron Microscope or STEM for short.
Besides the actual image produced by the electron microscope, there are many interactions taking place between the electrons in the beam and the sample that can be recorded simultaneously for gaining additional information about the sample. In one of these interactions, the high energy electrons cause sample atoms to emit X-rays. The energy of these X-rays can be directly related to what kind of atoms that are present in the sample, and so a great deal of information on the chemical composition can be extracted. The study of these X-rays emitted from the sample is called Energy Dispersive X-ray Spectroscopy or EDS.
Because the range of X-ray energies that can be emitted by various elements is large, a lot of energy data needs to be recorded. In a STEM-EDS image, several thousand different X-ray energies can be recorded and stored for each pixel. This means that for a regularly sized image, the total data will consist of hundreds of millions of values. This huge amount of data is very difficult to analyse manually and so one preferably uses various computer tools and algorithms to identify different elements. In this work, the accuracy of one of these computer tools, called Principal Component Analysis (or PCA) has been investigated. The ability of it to automatically find elements was tested on simulated datasets were the real composition could be defined exactly beforehand. Some strategies for what microscope parameters to set for optimal PCA performance was discussed. Finally, PCA was used on a real dataset originating from a cutting tool made from tungsten carbide. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9030772
- author
- Lazar, Isac LU
- supervisor
-
- Martin Ek LU
- organization
- course
- KOOL01 20201
- year
- 2020
- type
- M2 - Bachelor Degree
- subject
- keywords
- Principal Component Analysis, PCA, STEM, data analysis, machine learning, metal machining, materials, materials chemistry, materialkemi
- language
- English
- id
- 9030772
- date added to LUP
- 2020-11-09 09:32:59
- date last changed
- 2020-11-09 09:32:59
@misc{9030772,
abstract = {{STEM-EDS is a common technique used for compositional mapping in materials. The vasts amounts of raw data produced by STEM EDS is suitable for advanced analysis using methods in unsupervised machine learning. One powerful method is Principal Component Analysis (PCA) which can automatically discover significant chemical correlations in a sample. Although more powerful than classical analysis, there are limits to how weak chemical correlations PCA can detect. Faced with a limited data collection time, it is of great importance to know how data collection parameters in the STEM affects PCA performance. In this work, the effect of image size and X-ray counts has been investigated using simulated STEM EDS datasets. It was found that decreasing overall image size while increasing counts per pixel both increases the chances of discovering weak chemical correlations and improves the overall PCA accuracy. Furthermore, use of varimax factor rotations has been investigated as a way to improve interpretability of PCA results. Finally, a real sample originating from a tungsten-carbide cutting tool which had been turned in a titanium alloy was analysed using the presented methods. This analysis confirmed the existence of an expected pure tungsten phase but also lead to the discovery of an unexpected ridge-like structure in the adhered alloy material.}},
author = {{Lazar, Isac}},
language = {{eng}},
note = {{Student Paper}},
title = {{Principal Component Analysis for STEM-EDS images: Optimal collection parameters}},
year = {{2020}},
}