Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

A Digital Twin for Smart Firefighting - A comparison of FDS and CFAST as input for a machine learning model

Björkqvist, Ivar LU and Jonsson, Oscar LU (2025) In LUTVDG/TVBB VBRM10 20251
Division of Fire Safety Engineering
Abstract
This thesis explores the development of a digital twin for a real-life tunnel facility in Revinge, Skåne, using fire simulations and machine learning. The study aimed to identify preferred machine learning algorithms and compare fire simulation tools in terms of efficiency and performance. A literature review highlighted Long Short Term Memory (LSTM) networks as a suitable choice. Fire simulations were conducted in CFAST and FDS, modeling a 13-compartment corridor scenario. Data from 96 simulations were used to train MATLAB® models predicting, using a premade deep learning toolbox, heat release rate (HRR) and fire location.
The models were tested against experimental tunnel fire data, with HRR predictions achieving around 50% accuracy per... (More)
This thesis explores the development of a digital twin for a real-life tunnel facility in Revinge, Skåne, using fire simulations and machine learning. The study aimed to identify preferred machine learning algorithms and compare fire simulation tools in terms of efficiency and performance. A literature review highlighted Long Short Term Memory (LSTM) networks as a suitable choice. Fire simulations were conducted in CFAST and FDS, modeling a 13-compartment corridor scenario. Data from 96 simulations were used to train MATLAB® models predicting, using a premade deep learning toolbox, heat release rate (HRR) and fire location.
The models were tested against experimental tunnel fire data, with HRR predictions achieving around 50% accuracy per timestep. The models performance could have been improved significantly if the conditions in the training data had a better resemblance to the experimental data. The CFAST-based model outperformed FDS in predicting fire location. Results emphasized the need for a broader training dataset and showed that simulation tool preference depends on the specific application, though CFAST offered advantages in speed and usability for this case. (Less)
Please use this url to cite or link to this publication:
author
Björkqvist, Ivar LU and Jonsson, Oscar LU
supervisor
organization
alternative title
En digital tvilling för smart brandbekämpning – En jämförelse mellan FDS och CFAST som input data till en maskinlärningsmodell
course
VBRM10 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Smart Firefighting, Digital Twin, FDS, CFAST, RNN, LSTM, Machine Learning, Hotel room
publication/series
LUTVDG/TVBB
report number
5749
other publication id
LUTVDG/TVBB--5749--SE
language
English
id
9209861
date added to LUP
2025-08-28 07:59:40
date last changed
2025-08-28 07:59:40
@misc{9209861,
  abstract     = {{This thesis explores the development of a digital twin for a real-life tunnel facility in Revinge, Skåne, using fire simulations and machine learning. The study aimed to identify preferred machine learning algorithms and compare fire simulation tools in terms of efficiency and performance. A literature review highlighted Long Short Term Memory (LSTM) networks as a suitable choice. Fire simulations were conducted in CFAST and FDS, modeling a 13-compartment corridor scenario. Data from 96 simulations were used to train MATLAB® models predicting, using a premade deep learning toolbox, heat release rate (HRR) and fire location.
The models were tested against experimental tunnel fire data, with HRR predictions achieving around 50% accuracy per timestep. The models performance could have been improved significantly if the conditions in the training data had a better resemblance to the experimental data. The CFAST-based model outperformed FDS in predicting fire location. Results emphasized the need for a broader training dataset and showed that simulation tool preference depends on the specific application, though CFAST offered advantages in speed and usability for this case.}},
  author       = {{Björkqvist, Ivar and Jonsson, Oscar}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{LUTVDG/TVBB}},
  title        = {{A Digital Twin for Smart Firefighting - A comparison of FDS and CFAST as input for a machine learning model}},
  year         = {{2025}},
}