Fire resistance evaluation of gypsum plasterboard walls using machine learning method

Abstract

Gypsum plasterboard walls are commonly used in constructions and they provide the passive fire protection as separating elements. The fire resistance of the wall is largely dependent on the thermo-mechanical properties of the gypsum plasterboard. Numerical modelling of the fire behaviour of gypsum plasterboard cracking, fall-off and gypsum induced water transport is a great challenge. The fire resistance of the wall is assessed by performing expensive fire tests or conservative empirical calculations. This paper presents the evaluation of fire resistance of steel stud gypsum plasterboard lined walls by means of machine learning. A series of large, medium and small scale fire tests are performed or gathered on the gypsum plasterboard walls in order to create a data base. Various tracks are considered when using machine learning in order to predict the failure times and the temperature evolution on the unexposed side of these walls. The machine learning results are compared with test results, with the separating function method and finite element method results.