Determination of Intrinsic Material Flammability Properties from Material Tests assisted by Numerical Modelling

Computational Fluid Dynamics (CFD) codes are being increasingly used in the field of fire safety engineering. They provide, amongst other things, velocity, species and heat flux distributions throughout the computational domain. The various sub-models associated with these have been developed sufficiently to reduce the errors below 10%-15%, and work continues on reducing these errors yet further. However, the uncertainties introduced by using material properties as an input for these models are considerably larger than those from the other sub-models, yet little work is being done to improve these. Most of the data for these material properties comes from traditional (standard) tests. It is known that these properties are not intrinsic, but are test-specific. Thus, it can be expected that the errors incurred when using these in computations can be significant. Research has been held back by a lack of understanding of the basic factors that determine material flammability. The term “flammability” is currently used to encompass a number of definitions and “properties” that are linked to standardised test methodologies. In almost all cases, the quantitative manifestations of “flammability” are a combination of material properties and environmental conditions associated with the particular test method from which they were derived but are not always representative of parameters linked intrinsically with the tested material. The result is that even the best-defined parameters associated with flammability cannot be successfully introduced into fire models to predict ignition or fire growth. The aim of this work is to develop a new approach to the interpretation of standard flammability tests in order to derive the (intrinsic) material properties; specifically, those properties controlling ignition. This approach combines solid phase and gas modelling together with standard tests using computational fluid dynamics (CFD), mass fraction of flammable gases and lean flammability limits (LFL). The back boundary condition is also better defined by introducing a heat sink with a high thermal conductivity and a temperature dependant convective heat transfer coefficient. The intrinsic material properties can then be used to rank materials based on their susceptibility to ignition and, furthermore, can be used as input data for fire models. Experiments in a standard test apparatus (FPA) were performed and the resulting data fitted to a complex pyrolysis model to estimate the (intrinsic) material properties. With these properties, it should be possible to model the heating process, pyrolysis, ignition and related material behaviour for any adequately defined heating scenario. This was achieved, within bounds, during validation of the approach in the Cone Calorimeter and under ramped heating conditions in the Fire Propagation Apparatus (FPA). This work demonstrates that standard flammability and material tests have been proven inadequate for the purpose of obtaining the “intrinsic” material properties required for pyrolysis models. A significant step has been made towards the development of a technique to obtain these material properties using test apparatuses, and to predict ignition of the tested materials under any heating scenario. This work has successfully demonstrated the ability to predict the driving force (in-depth temperature distribution) in the ignition process. The results obtained are very promising and serve to demonstrate the feasibility of the methodology. The essential outcomes are the “lessons learnt”, which themselves are of great importance to the understanding and further development of this technique. One of these lessons is that complex modelling in conjunction with current standard flammability test cannot currently provide all required parameters. The uncertainty of the results is significantly reduced when using independently determined parameters in the model. The intrinsic values of the material properties depend significantly on the accuracy of the model and precision of the data