Javascript must be enabled for the correct page display

Automatic Timestep Selection for In Situ Visualization of Large Data

Feltham, Alpheaus (2023) Automatic Timestep Selection for In Situ Visualization of Large Data. Master's Thesis / Essay, Computing Science.


Download (5MB) | Preview
[img] Text
Restricted to Registered users only

Download (131kB)


This thesis examines in-situ simulation timestep selection and reconstruction, through the use of a predictive autoencoder system. This autoencoder is used to detect various anomalies within a simulated 2D Kármán Vortex Street ensemble in order to select timesteps of interest, as well as to reconstruct timestep data from a given sub-sampling based on the selected timesteps. This autoencoder setup is trained to predict subsequent timesteps in a series. The detection capacity of this autoencoder is tested against a non-predictive autoencoder system. This is done using a number of datasets which are copied from the original ensemble and subsequently modified to contain artificial anomalies of various types. Then the detection and reconstruction ability of both systems are compared against each other. The thesis finds that both autoencoder setups are similar in their detection capacity, but the predictive model performs slightly better in data reconstruction. The detection system is found to be only somewhat effective otherwise, and various possible reasons therefore as well as some potential solutions and alternatives are discussed.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Kosinka, J. and Frey, S.D.
Degree programme: Computing Science
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 11 Oct 2023 12:15
Last Modified: 11 Oct 2023 12:15

Actions (login required)

View Item View Item