What fire modelling resolution is needed in AI training data for smart firefighting in tunnels?
Morales Mere, José Antonio LU ; Johansson, Nils LU
and Ronchi, Enrico
LU
(2025)
11th International Symposium on Tunnel Safety and Security
In International symposium on tunnel safety and security (ISTSS)
p.522-536
- Abstract
- The adoption of smart firefighting in tunnels – intended here as fire modelling coupled with sensor data - requires a broad discussion into the suitability of different tools and approaches for decision support in case of emergency. In particular, Artificial Intelligence (AI) tools have shown great potential to support real-time emergency management, given their ability to simulate and predict the evolution of fire scenarios with low computational cost. AI tools for fire modelling predictions generally rely on a given simulation database for their training. Such databases can be developed through a variety of fire modelling approaches with different resolutions, e.g. hand-calculations, (multi-)zone modelling and computational fluid... (More)
- The adoption of smart firefighting in tunnels – intended here as fire modelling coupled with sensor data - requires a broad discussion into the suitability of different tools and approaches for decision support in case of emergency. In particular, Artificial Intelligence (AI) tools have shown great potential to support real-time emergency management, given their ability to simulate and predict the evolution of fire scenarios with low computational cost. AI tools for fire modelling predictions generally rely on a given simulation database for their training. Such databases can be developed through a variety of fire modelling approaches with different resolutions, e.g. hand-calculations, (multi-)zone modelling and computational fluid dynamics (CFD). This study proposes an approach to investigate the suitability of different modelling resolutions for generating AI training datasets for tunnel fire scenarios, considering heat release rate as the key variable to predict. Regression modelling allows estimating this key variable. Testing has been performed through a set of fire scenarios within and outside the training dataset fire intensity intervals. An experimental dataset from real-world repeated experiments was used to understand the prediction performance based on different modelling resolutions. The prediction performance was assessed through the error evaluation for the testing and experimental datasets. The error of the testing data shows how strongly the regression model correlates to the assumptions taken in the training scenarios. Here, the regression model based on CFD shows the best fit when predicting fire intensity within the range used for training. The regression model based on multi-zone modelling shows a better fit for larger fires. Finally, the comparison to experimental data shows that handcalculations are not suitable for this type of applications while multi-zone and CFD modelling approaches can provide sufficient resolution to predict fire intensity when adopting them to train a regression model and using temperature sensor data. (Less)
- Abstract (Swedish)
- The adoption of smart firefighting in tunnels – intended here as fire modelling coupled with sensor data - requires a broad discussion into the suitability of different tools and approaches for decision support in case of emergency. In particular, Artificial Intelligence (AI) tools have shown great potential to support real-time emergency management, given their ability to simulate and predict the evolution of fire scenarios with low computational cost. AI tools for fire modelling predictions generally rely on a given simulation database for their training. Such databases can be developed through a variety of fire modelling approaches with different resolutions, e.g. hand-calculations, (multi-)zone modelling and computational fluid... (More)
- The adoption of smart firefighting in tunnels – intended here as fire modelling coupled with sensor data - requires a broad discussion into the suitability of different tools and approaches for decision support in case of emergency. In particular, Artificial Intelligence (AI) tools have shown great potential to support real-time emergency management, given their ability to simulate and predict the evolution of fire scenarios with low computational cost. AI tools for fire modelling predictions generally rely on a given simulation database for their training. Such databases can be developed through a variety of fire modelling approaches with different resolutions, e.g. hand-calculations, (multi-)zone modelling and computational fluid dynamics (CFD). This study proposes an approach to investigate the suitability of different modelling resolutions for generating AI training datasets for tunnel fire scenarios, considering heat release rate as the key variable to predict. Regression modelling allows estimating this key variable. Testing has been performed through a set of fire scenarios within and outside the training dataset fire intensity intervals. An experimental dataset from real-world repeated experiments was used to understand the prediction performance based on different modelling resolutions. The prediction performance was assessed through the error evaluation for the testing and experimental datasets. The error of the testing data shows how strongly the regression model correlates to the assumptions taken in the training scenarios. Here, the regression model based on CFD shows the best fit when predicting fire intensity within the range used for training. The regression model based on multi-zone modelling shows a better fit for larger fires. Finally, the comparison to experimental data shows that hand-calculations are not suitable for this type of applications while multi-zone and CFD modelling approaches can provide sufficient resolution to predict fire intensity when adopting them to train a regression model and using temperature sensor data. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/e99247dc-0bf7-4cda-be76-63aa3c931202
- author
- Morales Mere, José Antonio
LU
; Johansson, Nils
LU
and Ronchi, Enrico
LU
- organization
- publishing date
- 2025
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- Proceedings from the 11th International Symposium on Tunnel Safety and Security
- series title
- International symposium on tunnel safety and security (ISTSS)
- editor
- Li, Ying Zhen ; Lönnermark, Anders ; Gehandler, Jonatan and Ingason, Haukur
- pages
- 522 - 536
- publisher
- Rise, Research Institute of Sweden
- conference name
- 11th International Symposium on Tunnel Safety and Security
- conference location
- Reykjavik, Iceland
- conference dates
- 2025-04-09 - 2025-04-11
- ISBN
- 978-91-90036-19-8
- project
- Utilization of Innovative Digital Tools for Efficient and Smart Fire Fighting
- language
- English
- LU publication?
- yes
- id
- e99247dc-0bf7-4cda-be76-63aa3c931202
- date added to LUP
- 2026-08-07 09:11:45
- date last changed
- 2026-08-11 14:40:17
@inproceedings{e99247dc-0bf7-4cda-be76-63aa3c931202,
abstract = {{The adoption of smart firefighting in tunnels – intended here as fire modelling coupled with sensor data - requires a broad discussion into the suitability of different tools and approaches for decision support in case of emergency. In particular, Artificial Intelligence (AI) tools have shown great potential to support real-time emergency management, given their ability to simulate and predict the evolution of fire scenarios with low computational cost. AI tools for fire modelling predictions generally rely on a given simulation database for their training. Such databases can be developed through a variety of fire modelling approaches with different resolutions, e.g. hand-calculations, (multi-)zone modelling and computational fluid dynamics (CFD). This study proposes an approach to investigate the suitability of different modelling resolutions for generating AI training datasets for tunnel fire scenarios, considering heat release rate as the key variable to predict. Regression modelling allows estimating this key variable. Testing has been performed through a set of fire scenarios within and outside the training dataset fire intensity intervals. An experimental dataset from real-world repeated experiments was used to understand the prediction performance based on different modelling resolutions. The prediction performance was assessed through the error evaluation for the testing and experimental datasets. The error of the testing data shows how strongly the regression model correlates to the assumptions taken in the training scenarios. Here, the regression model based on CFD shows the best fit when predicting fire intensity within the range used for training. The regression model based on multi-zone modelling shows a better fit for larger fires. Finally, the comparison to experimental data shows that handcalculations are not suitable for this type of applications while multi-zone and CFD modelling approaches can provide sufficient resolution to predict fire intensity when adopting them to train a regression model and using temperature sensor data.}},
author = {{Morales Mere, José Antonio and Johansson, Nils and Ronchi, Enrico}},
booktitle = {{Proceedings from the 11th International Symposium on Tunnel Safety and Security}},
editor = {{Li, Ying Zhen and Lönnermark, Anders and Gehandler, Jonatan and Ingason, Haukur}},
isbn = {{978-91-90036-19-8}},
language = {{eng}},
pages = {{522--536}},
publisher = {{Rise, Research Institute of Sweden}},
series = {{International symposium on tunnel safety and security (ISTSS)}},
title = {{What fire modelling resolution is needed in AI training data for smart firefighting in tunnels?}},
url = {{https://lup.lub.lu.se/search/files/257396689/28_review_Morales_-_Johansson_-_Ronchi_final.pdf}},
year = {{2025}},
}