Smoke production in fire represents a threat because fire smoke reduces visibility and because fire smoke is toxic. One way to reduce the risk of persons being overcome by smoke during evacuation is by setting requirements to the building materials’ ability to contribute to the smoke production in a fire. By regulating the use of building products based on their contribution to the optical smoke production in a fire, the toxicity aspects of the smoke will be covered to a high degree as well.
In this work prediction models for optical smoke production in the Single Burning Item test (SBI) and in the Room Corner test have been developed. The models are of two kinds; classification models based on multivariate statistical analysis of Cone Calorimeter test results, and dynamic calculation models where empirically developed equations are combined with multivariate statistical classification models. The basic idea behind the dynamic smoke prediction models is that the smoke production rate is closely linked to the heat release rate. Prediction models for heat release rate in the two larger-scale methods were therefore a necessary starting point for the modelling of smoke production. Existing models simulating heat release in the SBI test and in the Room Corner test were modified to suit these needs; and were assessed to have high predictability after the modifications.
My work comprise the following prediction models:
•A modified version of the Wickström/Göransson model for prediction of heat release rate in the Room Corner test.
•A statistical model for predicting time to flashover in the Room Corner test using the concept of FO-categories.
•A model for predicting smoke production rate in the Room Corner test.
•A statistical model predicting the level of maximum and average smoke production rate in the Room Corner test.
•A modified version of the model by Messerschmidt et. al. for prediction of heat release rate in the Single Burning Item test.
•A model for predicting smoke production rate in the Single Burning Item test.
•A statistical model predicting the level of SMOGRA and the smoke classification in the Single Burning Item test.
All models, both for prediction of heat release and smoke production, use results from Cone Calorimeter tests at heat flux level 50 kW/m2 as input data. The empirical basis for the models is test data from a total of 65 different products. 32 of the products are tested both in the SBI test and in the Cone Calorimeter test; 56 are tested both in the Room Corner test and in the Cone Calorimeter test. Data from a total of 194 Cone Calorimeter tests have been analysed.
Both the statistical classification models and the dynamic calculation models can easily be implemented in a PC worksheet, and the prediction results are readily achieved.
The models’ predictability has been evaluated by comparing the predicted results to results from “real“ larger-scale tests, and by comparing predicted classification to the classification actually obtained. The actual and predicted classifications have been calculated according to the new European system for classification of building products based on reaction to fire test results, and according to the existing classification system based on the EUREFIC-programme.
The results show that both heat release and smoke production are possible to predict with these models. The predictions of the Single Burning Item test results are more precise than the Room Corner test results, this is probably because the ventilation conditions in the Cone Calorimeter test are more similar to the Single Burning Item test than to the conditions in the Room Corner test. The large-scale fire behaviour is found difficult to predict for some types of products where the fire behaviour depends on certain mechanical or chemical changes during the fire exposure. Such events are obviously not easily predicted from small-scale tests in the Cone Calorimeter, and will need more detailed modelling.
This thesis presents a generic method of designing prediction models where test results from small-scale methods are used to predict fire behaviour in larger scale. The main feature of this kind of models is the integration of multivariate statistical models in the calculations. Statistical information makes it possible to discriminate between different kinds of products and fire behaviour, and thereby to choose calculation algorithms specially designed for different product groups. Products with high flamespread ability, products with low heat release, products with high smoke production and wood-based products are examples of product types that require special treatment in the modelling of fire behaviour in the Room Corner test and in the SBI test. Modelling of other large-scale test methods may need the option of discriminating between other kinds of groups, based on e.g. product type, geometrical considerations etc.