Finding Potential Detections of Dust in Protoplanetary Disk Winds. Using Machine Learning to Filter, Classify and Present ALMA data
Typ
Examensarbete för kandidatexamen
Bachelor Thesis
Bachelor Thesis
Program
Publicerad
2023
Författare
Fagrell, Peter
Kollberg, Jens
Rasmussen, Elias
Redmo Axelsson, Erik
Svensson, Oskar
Ybring, Alexander
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Star formation begins when clouds of dust and gas starts to collapse under the force
of their own gravity. Eventually, a protoplanetary disk is formed in the central
region. Because of magnetic fields, gravity and rotation, material is ejected through
jets and winds that sweep up gas from the surrounding envelope and form an outflow.
These outflows remove angular momentum which permits a star to form in the
center. In the disk, small grains eventually starts to clump together to become
planets. Whether lifting of dust via winds affect the planet formation process is
yet not understood, but it is a research topic (e.g. Pascucci et al., 2022) where this
project could contribute.
The project aims to automatically extract images from the Atacama Large Millime ter/submillimeter Array (ALMA) Science Archive that indicates dust lifting from
disks via the aforementioned winds, with the use of a Convolutional Neural Network
(CNN). Additionally, the network should be flexible enough to be applied to other,
similar problems. With the help of ALminer observations are fetched and then fed
to the CNN to filter out objects of interest. Due to a low number of known observa tions that could possibly depict dust in the wind, images are augmented to produce
sufficient training data. During training of the CNN, a portion of the training data
is reserved for determining the accuracy of the network. Finally, it is applied to a
part of the ALMA archive chosen through keywords and presents images it labels
as positive.
The methods mentioned led to an extraction of several observations. Prediction of
labels was performed by the CNN with high accuracy and it can, with modification,
act as a general model for other astronomical phenomena. In conclusion, a publicly
available and free tool was created that can support researchers in their collection of
data from the ALMA archive, minimizing manual labor and advancing the studies
of the universe.
Beskrivning
Ämne/nyckelord
Protoplanetary Disk, Outflows, Star, Winds, ALMA, CNN, ALminer