Implementation of LSTM network for generating design temperature–time curves of mass timber compartment fire

2026-06-08

Tongchen Han

Solomon Tesfamariam *

DOI：10.1016/j.firesaf.2026.104888

Original

Save

View PDF

Abstract

En 中文

Mass timber products have been extensively used to support the construction of tall timber buildings, while the fire safety of such structures remains a critical public concern. This paper aims to use machine-learning techniques to facilitate the mass timber compartment fire design. First, a timber compartment is designed. Fire scenarios are defined by parameters including movable fire load density, fire growth rate, opening factor, and the extent of exposed timber surfaces. The contribution of the exposed combustible surfaces within the compartment is quantified through the heat release rate (HRR). A comparative study is carried out for different HRR models. Subsequently, a long short-term memory (LSTM) network is developed to predict the temperature–time evolution of the compartment based on the calculated HRR. Sensitivity analysis is carried out in terms of input parameters. To extend the applicability of the model across compartments of varying sizes, generalization is achieved through the concept of equivalent fire load density. The derivation of HRR and the developed LSTM network are integrated into a user-oriented program, Compartment fire predictor, which allows for customized compartment-specific input parameters. Finally, the developed predictor is validated against large-scale compartment fire tests, demonstrating its effectiveness in predicting the temperature evolution of compartments prior to CLT delamination. The limitations of the predictor are also illustrated.

Keywords:

Long short-term memory

Mass timber

Compartment fire

Fire design

Structural fire safety

Implementation of LSTM network for generating design temperature–time curves of mass timber compartment fire

Abstract

AI Summary

Journal

Organization

Cited Papers

Citing Papers