Real-time Bayesian Inversion in Resin Transfer Moulding using Neural Surrogates

Thumbnail
View/Open
Publication date
2024-07-19
Creators
Causon, Michael
Metadata
Show full item record
Description
The aim of this study is to rapidly estimate the properties of fibrous reinforcements during the injection phase of Resin Transfer Moulding. There are five data sets associated with this study. The first data set was generated by simulating the resin injection process for 50,000 samples of reinforcement properties. This data was used to train a surrogate model to emulate the injection simulator. The remaining four data sets correspond to the lab experiments included within the paper. These show time series plots for resin pressure at each sensor within the tool, recorded by the data acquisition system.
External URI
DOI
Subjects
  • Resin transfer molding
  • Plastics -- Molding
  • Fiber-reinforced plastics
  • Resin Transfer Moulding, Moving boundary problems, Neural networks, Surrogate models, Machine Learning
  • Engineering::Chemical, process & energy engineering
  • T Technology::TP Chemical technology
Divisions
  • University of Nottingham, UK Campus::Faculty of Engineering
Research institutes and centres
  • University of Nottingham, UK Campus::Advanced Manufacturing, Institute for
Deposit date
2024-07-19
Data type
Surrogate training data are text files consisting of the inputs and outputs of the injection simulator. Experimental data recorded via the data acquisition system are in MATLAB data files.
Contributors
  • Iglesias, Marco
  • Matveev, Mikhail
  • Endruweit, Andreas
  • Tretyakov, Michael
Funders
  • Engineering & Physical Sciences Research Council
Grant number
  • EP/P006701/1
Collection dates
  • Surrogate training data were generated on 14/01/24. Experimental data were collected between 11/12/23 - 04/01/24.
Data collection method
The surrogate training data were generated using the "Control volume FEM solver for 2D moving boundary problems" in Matlab (doi:10.5281/zenodo.10914584). The experimental data were collected via a data acquisition system, recording fluid pressure at each sensor within the tool at a rate of 10 per second.
Resource languages
  • en
Publisher
The University of Nottingham
CC-BY
Except where otherwise noted, this item's license is described as Creative Commons by Attribution