1479 00

Patrick Van Hees
Tommy Hertzberg
Anne Steen Hansen, SINTEF

Development of a screening
method for the SBI and Room
Corner using the Cone
Calorimeter
Nordtest project 1479-00

SP Rapport 2002:11
Fire Technology
Bors 2002

2

Abstract

With the implementation of the classes of reaction to fire performance for construction
products, the so-called Euroclasses, a new test method has been introduced, namely the
SBI test method (EN 13823). This method is an intermediate scale test and it has been
developed to allow harmonization of the reaction to fire classification in Europe. As the
test is new and needs rather large samples there will be a need from the industry to have
an easy and cost effective tool for product development and quality control with respect
to this method. In the near future also the possibility of using an appeal procedure by
running the reference scenario test, ISO 9705, will be possible. Also for this test it would
be an advantage to have a screening method.
In this project such a tool has been developed by using cone calorimeter data according to
ISO 5660. The tool allows prediction of the major classification parameters for HRR and
SPR in the SBI and room corner. Within the project a software tool has also been
developed. The results of the project show that the predictions are satisfactory and that
the tool will be powerful for the product development by industry. This project can be
extended for other building products (e.g. cables, pipe insulation) once an European
classification system is in place for these products.

Key words: cone calorimeter, SBI, room tests, modeling, wall and ceiling linings

SP Sveriges Provnings- och SP Swedish National Testing and
Forskningsinstitut Research Institute
SP Rapport 2002:11 SP Report 2002:11
ISBN 91-7848-904-0
ISSN 0284-5172
Bors 2002
Postal address:
Box 857, SE-501 15 BORS
Sweden
Telephone: +46 33 16 50 00
Telefax: +46 33 13 55 02
E-mail: info@sp.se
Internet: www.sp.se
3

Contents
Abstract 2
Contents 3
Preface 5
Sammanfattning 5
1 Background 7
2 Scope 8
3 Research programme 9
3.1 WI 1 Development of screening protocol in ISO 5660 and fine
tuning of models 9
3.2 WI 2 Additional testing in ISO 5660 and SBI 9
3.3 WI 3 Validation of the models for prediction of FIGRA 9
3.4 WI 4 Development and validation of a prediction model for
SMOGRA in SBI and RCT 9
3.5 WI 5 Development of the software 10
3.6 WI 6 Conclusions and reporting 10
4 Overview of test methods 11
4.1 Cone Calorimeter test 11
4.2 SBI test method 11
4.3 Room corner test method 13
5 Euroclass system 14
6 Development of models for FIGRA and HRR 18
6.1 SBI test method 18
6.1.1 Description of model 18
6.1.2 Sensitivity study of model 22
6.1.2.1 Influence of HRR threshold and ignition time 22
6.1.2.2 Influence of backing board 23
6.1.2.3 Shiny materials 25
6.1.3 Guidance and description testing protocol 25
6.1.4 Comparison and discussion of simulation results 25
6.2 Room corner test method 27
6.2.1 Description of model(s) 27
6.2.2 Description of test protocol in ISO 5660 used for the simulations 34
6.2.3 Comparison and discussion of simulation results 34
7 Development of models for Smoke production 36
7.1.1 Statistical method 36
7.2 Test results from ISO 5660 used in the prediction models 37
7.3 Prediction of smoke production in the SBI test 38
7.3.1 General 38
7.3.2 Predicting the level of SMOGRA 40
7.3.3 Predicting the smoke classification s1, s2 or s3 41
7.4 Prediction of smoke production in the Room Corner test method 43
7.4.1 Flashover is determining for the smoke production 43
7.4.2 How predictable is the Room Corner SMOGRA value? 43
4

7.4.3 The EUREFIC classification system 44
7.4.4 Prediction of maximum smoke production rate in the Room Corner
test 45
7.4.5 Prediction of average smoke production rate in the Room Corner
test 48
7.5 Implementation of the prediction models 50
8 Software development 51
8.1 Principle 51
8.2 Different Button menus 52
8.2.1 Open file menu 52
8.2.2 Calculate menu 53
8.2.3 Select what to plot menu 54
8.2.4 Compare menu 55
8.2.5 Save SBI and SBI RCT menu 56
8.2.6 Print menu 57
8.2.7 Types menu 58
8.2.8 Help menu 59
8.2.9 Exit menu 60
8.3 Different Scroll menus 60
8.3.1 File menu 60
8.3.2 Simulation menu 60
8.3.3 Help menu 60
9 Conclusions 61
Annex A Simulation results for cone-RCT model for HRR 62
The Eurefic project 62
The SBI research project 67
Annex B Simulation results for Cone-SBI HRR model 81
Eurefic data 81
SBI RR materials 82
Additional data 96
Annex C Generic description of SBI RR materials 99
References 101
References 101

5

Preface

This work was sponsored by Nordtest as Nordtest project 1479-00. The authors would
like to thank all industrial sponsors for their support. Also the staff conducting the tests
and this work at SP, Interscience and SINTEF are thanked.

Sammanfattning

Infrandet av nya europaklasser fr byggnadsmaterial har medfrt implementering av en
ny testmetod, SBI (Single Burning Item) enligt Europanorm EN 13283. Metoden r att
karakterisera som en metod i mellanskala och den har utvecklats frmst fr att mjliggra
en harmonisering av brandklasser fr ytmaterial i Europa.
D metoden r ny och SBI medfr relativt omfattande experiment, finns behov av att ta
fram enkla, smskaliga och kostnadseffektiva verktyg fr industrin att anvnda i sin
produktutveckling och kvalitetsvervakning. Det finns ven ett behov av att ta fram ett
sdant verktyg fr det strre referensscenariotestet, ISO 9705, vilket kommer att kunna
anvndas fr att verklaga en produktklassificering som erhllits enligt SBI, EN 13283.

I detta projekt har ett simuleringsverktyg baserat p ett smskalig test utvecklats.
Metoden baseras p konkalorimeter test enligt internationell standard ISO 5660.
Verktyget mjliggr att uppskatta viktiga materialparametrar som styr
produktklassificeringen enligt EN 13283 eller ISO 9705.
Modelleringsverktyget visar en mycket god verensstmmelse mellan prediktion baserat
p smskaliga konkalorimeter frsk och experiment utfrda i stor (ISO 9705) eller
mellan (EN 13283) skala och kommer att kunna nyttjas fr ett stort antal industriella
produkter. Projektet och verktyget kan dessutom utvidgas till att omfatta andra
byggnadsprodukter ssom t.ex. kablar eller rrisoleringsmaterial, s fort en motsvarande
klassificering finns p plats fr dessa produkter.

6

7

1 Background

Within the Euroclass system two methods are important with respect to classification of
wall and ceiling linings. On one hand there is the SBI (Single Burning Item) test, which is
the key test for Euroclasses A2 to D. On the other hand there is the room corner test (ISO
9705 or Nordtest NT FIRE 025), which is used as the reference scenario for the
Euroclasses. Both tests, however, cannot be considered as small-scale tests. Hence a need
for a small-scale test is necessary for screening but also as a tool for production control
and as a tool for product development. The cone calorimeter test according to ISO 5660 is
the most appropriate choice.
At the same time it should be noted that the room corner test results are used for
determining fire restricting materials in the HSC (High Speed Craft) Code within the
IMO regulations. Also in this case it is desirable with a small scale screening method for
production control and product development.
An important factor for the worldwide application of a combination of small and large-
scale tests is that the room corner test recently has been adopted in the Japanese building
code.
8

2 Scope

This project will develop a screening procedure for wall and ceiling lining testing in the
SBI and room corner tests. The outcome of this project will be a testing protocol and also
a multifunctional user-friendly software package allowing users to predict SBI and room
corner test results by means of cone calorimeter tests.
9

3 Research programme

The research programme sponsored by Nordtest had the following work items. In this
chapter an overview of these work items is given, and also a reference to the parts of this
report that are dealing with the specific work items.

3.1 WI 1 Development of screening protocol in ISO
5660 and fine tuning of models

This work package would develop a cone calorimeter test protocol for the screening test.
A number of materials from the RR in the SBI project would be chosen to optimise the
model. Following items will be investigated:
1. The optimum heat flux level in the cone calorimeter
2. The optimum sample preparation for testing
3. The optimum substrate for sample preparation
4. The optimum surface emissivity for testing shiny materials
5. The optimum ignition properties of the materials
In total a maximum of 35 cone calorimeter tests were planned
The work item will also include the fine-tuning of the models i.e. the cone-sbi and the
cone-room corner conetools model so that they can be used at the same heat flux level
and with the same sample preparation. One of the items in the cone-sbi model is the
improvement of the lateral flame spread routine. One of the items in the cone-rct model is
the adaptation of the model to use the same heat flux level as in the cone-sbi model.
The work with respect to this work item is summarised in chapter 6

3.2 WI 2 Additional testing in ISO 5660 and SBI

Five more materials would be checked and also the remaining RR materials of the SBI
project will be retested with the optimum test protocol defined in WI1, if necessary. The
five additional materials will also be tested in the SBI apparatus.
This part is included in the work described in chapter 6. More materials were provided
thanks to industrial support.

3.3 WI 3 Validation of the models for prediction of
FIGRA

This work item would validate the model for prediction of the FIGRA index for the SBI
test and for the room corner test. It will also validate the HRR curve obtained and fine
tune the model additionally if necessary. This work item is described in chapter 6.

3.4 WI 4 Development and validation of a prediction
model for SMOGRA in SBI and RCT

This work item would develop a first prediction model for the SMOGRA in the SBI and
the RCT. It will be based on a correlation model with different parameters from the cone
calorimeter test. As smoke prediction are very difficult to perform between different
scales of testing a modest approach should be taken.
This work item is described in chapter 7
10

3.5 WI 5 Development of the software

In this work item a multifunctional and user-friendly software package will be developed.
The request for Nordtest support was limited and more internal research resources were
used.
The software shall have the following properties:
1. user-friendly interface
2. multifunctional input routine i.e. different type of input files should be possible e.g.
Excel, FDMS and commercial software file formats
3. standard output report with links to Word/Excel

The development is described in chapter 8.

3.6 WI 6 Conclusions and reporting

This work package draws conclusions and gives a proposal for the screening test and a
manual for the software. These items are described in chapter 6, 7, 8 and 9.

11

4 Overview of test methods

4.1 Cone Calorimeter test

The cone calorimeter is described in ISO 5660
1
. The test method describes a test
specimen with an area of 100 mm x 100 mm, which is exposed to a constant radiant heat
flux. The heat flux can be adjusted from 10 kW/m
2
to 100 kW/m
2
. A spark plug
positioned over the test specimen ignites any flammable gasses produced by the test
specimen. The effluents from the test is collected in a hood and transported through a
duct. In the duct there is a thermocouple, a pressure sensor, smoke measurement system
and a sample probe. Furthermore the test specimen is positioned on a load cell, so the
mass loss of the test specimen can be recorded during the test. The test equipment is
shown in Figure 1.

The test results are heat release rate (calculated using oxygen depletion), time to ignition,
smoke production and weight loss.

Figure 1 Cone Calorimeter

4.2 SBI test method

The Single Burning Item (SBI) test
2
is developed by the Official Laboratories Group
(OLG) based on the guidelines set out by the EU Regulatory Group (RG). The SBI test is
one of the test methods to be used to determine the classification of building products in
the future European classification system.

The SBI test simulates a single burning item (e.g. a waste paper basket) burning in a
corner of a room. The test rig is shown in Figure 2. The dimensions of the test specimen
are 1.0 m x 1.5 m and 0.5 m x 1.5 m. The two wings of the test specimen overlap in the
12

corner behind the burner. Products are tested in their end use conditions as far a possible.
The burner is triangular with a side length of 250 mm, and is a diffusion burner supplied
with propane. The output of the burner is 30 kW for 21 minutes. The classification results
are only evaluated over a maximum of 20 minutes. There is a floor in the test
configuration but no ceiling. The effluents from the fire are collected in the hood and
transported through the duct. In the duct thermocouples, a pressure sensor, a smoke
measurement system and a sample probe are mounted. The test rig is placed in an
enclosure in order to avoid any draft around the test specimen and to protect the operator
from the produced smoke.

Ignition
source
Fl ow
measurement
Gas analysis
(O
2
, CO, CO
2
)
Smoke
measurement
Exhaust
gases
Enclosure
Trolley with
specimen holder

Figure 2 Single Burning Item test

The test results are heat release rate (calculated using oxygen depletion), lateral flame
spread on the large wing of the test specimen, smoke production and burning
droplets/particles. From the results parameters such as FIGRA and SMOGRA are
calculated. An overview how the parameters are used inside the Euroclass system is given
in chapter 5.

The FIGRA(SBI) index is defined as the maximum value of 30 second averaged heat
release rate divided by time. The calculation of the FIGRA(SBI) index is described in
detail in EN 13823. The SMOGRA(SBI) is defined as the maximum value of 60 second
averaged smoke production rate divided by time.

13

4.3 Room corner test method

Optical density
(lamp/photocell)
Gas analysis (O
2
, CO, CO
2
)
Volume flow
Temperature and differential pressure
Exhaust hood
3,0x3,0x1,0
Exhaust gases
Doorway 0,8m x 2,0m
Gas
burner
2
,
4
0
m
3
,6
0
m
2
,
4
0
m

Figure 3 Room Corner Test

Room tests are performed according to ISO 9705, Room Corner Test
3
. The test room has
nominal internal dimensions of 3.6 m by 2.4 m by 2.4 m (length by width by height). The
test material is mounted so that the three inner walls and the ceiling in the room are
covered. Smoke gases are vented and air is let in through the door opening. The ignition
source is a gas burner, which is placed in one of the inner corners. The burner heat output
is 100 kW for the first ten minutes and then 300 kW for another ten minutes. The smoke
gases coming out through the doorway are collected by a hood and exhaust system from
where samples are taken for gas analysis. Heat release rate and smoke production rate are
measured continuously. A schematic drawing is given in Figure 3.

14

5 Euroclass system

The European Commission published the classes of reaction to fire performance of
construction products, the so called Euroclasses on February 8, 2000. Reaction to fire
testing will follow a new concept compared to existing procedures in Europe. Seven main
classes are introduced, the Euroclasses. These are A1, A2, B, C, D, E and F. A1 and A2
represent different degrees of limited combustibility. For linings, classes C to E represent
products that may go to flashover in a room and at certain times. F means that no
performance is determined. B means no flashover in a room corner test. Thus there are 7
classes for linings and 7 classes for floor coverings. Additional classes for smoke
production and any occurrence of burning droplets are also given, see Table 1 and Table
2.
15

Table 1: Classes of reaction to fire performance for construction products excluding floorings
(*)
Class Test method(s) Classification criteria

Additional classification
A1
EN ISO 1182 (
1
);
And
T 30C; and
m 50%; and
t
f

= 0 (i.e. no sustained flaming)
-
EN ISO 1716
PCS 2.0 MJ.kg
-1
(
1
); and
PCS 2.0 MJ.kg
-1
(
2
) (
2a
); and
PCS 1.4 MJ.m
-2
(
3
); and
PCS 2.0 MJ.kg
-1
(
4
)
-
A2
EN ISO 1182 (
1
);
Or
T 50C; and
m 50%; and t
f

20s
-
EN ISO 1716;

and
PCS 3.0 MJ.kg
-1
(
1
); and
PCS 4.0 MJ.m
-2
(
2
); and
PCS 4.0 MJ.m
-2
(
3
); and
PCS 3.0 MJ.kg
-1
(
4
)
-
EN 13823 (SBI) FIGRA 120 W.s
-1
; and
LFS < edge of specimen; and
THR
600s
7.5 MJ
Smoke production(
5
); and
Flaming droplets/ particles (
6
)
B EN 13823 (SBI);
And
FIGRA 120 W.s
-1
; and
THR
600s
7.5 MJ
Smoke production(
5
); and
6
)

EN ISO 11925-2(
8
):
Exposure = 30s
Fs 150mm within 60s
C EN 13823 (SBI);
And
FIGRA 250 W.s
-1
; and
THR
600s
15 MJ
Smoke production(
5
); and
6
)

EN ISO 11925-2(
8
):
Exposure = 30s
Fs 150mm within 60s
D EN 13823 (SBI);
And
FIGRA 750 W.s
-1

Smoke production(
5
); and
6
)

EN ISO 11925-2(
8
):
Exposure = 30s
Fs 150mm within 60s
E
EN ISO 11925-2(
8
):
Exposure = 15s
Fs 150mm within 20s
7
)
F No performance determined
(*) The treatment of some families of products, e.g. linear products (pipes, ducts, cables etc.), is still under review
and may necessitate an amendment to this decision.
(
1
) For homogeneous products and substantial components of non-homogeneous products.
(
2
) For any external non-substantial component of non-homogeneous products.
(
2a
) Alternatively, any external non-substantial component having a PCS 2.0 MJ.m
-2
, provided that the product
satisfies the following criteria of EN 13823(SBI) : FIGRA 20 W.s
-1
; and LFS < edge of specimen; and THR
600s

4.0 MJ; and s1; and d0.
(
3
) For any internal non-substantial component of non-homogeneous products.
(
4
) For the product as a whole.
(
5
) s1 = SMOGRA 30m
2
.s
-2
and TSP
600s
50m
2
; s2 = SMOGRA 180m
2
.s
-2
and TSP
600s
200m
2
; s3 = not s1 or s2.
(
6
) d0 = No flaming droplets/ particles in EN13823 (SBI) within 600s; d1 = No flaming droplets/ particles persisting
longer than 10s in EN13823 (SBI) within 600s; d2 = not d0 or d1; Ignition of the paper in EN ISO 11925-2 results in
a d2 classification.
(
7
) Pass = no ignition of the paper (no classification); Fail = ignition of the paper (d2 classification).
(
8
) Under conditions of surface flame attack and, if appropriate to enduse application of product, edge flame attack.

16

Table 2. Classes of reaction to fire performance for floorings
Class Test method(s) Classification criteria

Additional classification
A1
FL

EN ISO 1182 (
1
);
And
T 30C; and
m 50%; and
t
f

= 0 (i.e. no sustained flaming)
-
EN ISO 1716
PCS 2.0 MJ.kg
-1
(
1
); and
PCS 2.0 MJ.kg
-1
(
2
); and
PCS 1.4 MJ.m
-2
(
3
); and
PCS 2.0 MJ.kg
-1
(
4
)
-
A2
FL

EN ISO 1182 (
1
);
Or
T 50C; and
m 50%; and
t
f

20s
-
EN ISO 1716;

and
PCS 3.0 MJ.kg
-1
(
1
); and
PCS 4.0 MJ.m
-2
(
2
); and
PCS 4.0 MJ.m
-2
(
3
); and
PCS 3.0 MJ.kg
-1
(
4
)
-

EN ISO 9239-1 (
5
) Critical flux (
6
) 8.0 kW.m
-2
Smoke production (
7
)
B
FL

EN ISO 9239-1 (
5
)
and
Critical flux (
6
) 8.0 kW.m
-2
Smoke production (
7
)

EN ISO 11925-2(
8
):
Exposure = 15s
Fs 150mm within 20s

C
FL

EN ISO 9239-1 (
5
)
And
Critical flux (
6
) 4.5 kW.m
-2
Smoke production (
7
)

EN ISO 11925-2(
8
):
Exposure = 15s
Fs 150mm within 20s

D
FL

EN ISO 9239-1 (
5
)
And
Critical flux (
6
) 3.0 kW.m
-2
Smoke production (
7
)

EN ISO 11925-2(
8
):
Exposure = 15s
Fs 150mm within 20s

E
FL

EN ISO 11925-2(
8
):
Exposure = 15s
Fs 150mm within 20s
-
F
FL
No performance determined
(
1
) For homogeneous products and substantial components of non-homogeneous products.
(
2
) For any external non-substantial component of non-homogeneous products.
(
3
) For any internal non-substantial component of non-homogeneous products.
(
4
) For the product as a whole.
(
5
) Test duration = 30 minutes.
(
6
) Critical flux is defined as the radiant flux at which the flame extinguishes or the radiant flux after a test period of 30
minutes, whichever is the lower (i.e. the flux corresponding with the furthest extent of spread of flame).
(
7
) s1 = Smoke 750%.min; s2 = not s1.
(
8
) Under conditions of surface flame attack and, if appropriate to the enduse application of the product, edge flame
attack.
Symbols. The characteristics are defined with respect to the appropriate test method.
T temperature rise
m mass loss
t
f
duration of flaming
PCS gross calorific potential
FIGRA fire growth rate
THR
600s
total heat release
LFS lateral flame spread
SMOGRA smoke growth rate
TSP
600s
total smoke production
Fs flame spread
17

Definitions
Material : A single basic substance or uniformly dispersed mixture of substances, e.g.
metal, stone, timber, concrete, mineral wool with uniformly dispersed binder, polymers.
Homogeneous product : A product consisting of a single material, of uniform density
and composition throughout the product.
Non-homogeneous product : A product that does not satisfy the requirements of a
homogeneous product. It is a product composed of one or more components, substantial
and/or non-substantial.
Substantial component : A material that constitutes a significant part of a non-
homogeneous product. A layer with a mass per unit area 1.0 kg/m
2
or a thickness 1.0
mm is considered to be a substantial component.
Non-substantial component : A material that does not constitute a significant part of a
non-homogeneous product. A layer with a mass per unit area < 1.0 kg/m
2
and a thickness
< 1.0 mm is considered to be a non-substantial component.
Two or more non-substantial layers that are adjacent to each other (i.e. with no substantial
component(s) in-between the layers) are regarded as one non-substantial component and,
therefore, must altogether comply with the requirements for a layer being a non-
substantial component.
For non-substantial components, distinction is made between internal non-substantial
components and external non-substantial components, as follows :
Internal non-substantial component : A non-substantial component that is covered on
both sides by at least one substantial component.
External non-substantial component : A non-substantial component that is not covered
on one side by a substantial component.

An Euroclass is intended to be declared as for example Bd1s2. B represents the main
class, d1 means droplets/particles class no 1 and s2 means smoke class no 2. This gives
theoretically a total of about 40 classes of linings and 11 classes of floor coverings to
choose from. However, each country is expected only to use a very small fraction of the
possible combinations.

18

6 Development of models for FIGRA and
HRR

6.1 SBI test method

6.1.1 Description of model

The model
4,5
presented here uses ignition time as well as the complete heat release rate
curve from the Cone Calorimeter. In principle results from a single small scale test is
used to predict the first part of the heat release rate curve in the SBI and hence the
FIGRA(SBI) index. Other models
6,7
exist but for this project the above-mentioned model
was used.

The calculation model, based on the conetools model
8
for the room corner, is described in
detail below in sections containing principles, area growth, criterion for flame spread and
heat release rate, respectively.

Principles of prediction model

Three major assumptions have been made in the prediction model of heat release rate in
the SBI test:

1) The burning area growth rate and the heat release rate are decoupled.

2) The burning area growth rate is proportional to the ease of ignition, i.e. the inverse
of the time to ignition in small scale.

3) The history of the heat release rate per unit area at each location in the SBI test is
the same as in small scale.

Burning area growth rate

The fire spread can follow two different routes as shown in Figure 4. All products start to
spread along route I. A product is assumed to continue to spread along route III if the
calculated sustained flame height is at least 1.5 m, which is equal to the height of the test
sample. Otherwise the product is assumed to spread along route II. The calculation of
flame height will be outlined below.

19

0 100 200 300 400 500 600
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time [sec.]
A
r
e
a

[
m
2
]
I
II
III

Figure 4 Burning area curve modelled for the SBI

Within the different flame spread regimes the burning area growth rate of a product
depends on ignitability, i.e. time to ignition in the Cone Calorimeter. Once the flame
spread rate is determined the heat release rate is calculated assuming that products always
give the same heat release rate per unit area as a function of time in small scale as in the
SBI test. In other words all parts of the tested product are assumed to burn in the same
way in the SBI as in small scale. This is of course a simplification. The heat release rate
depends more or less on the actual heat flux level received by the product as a function of
time. However, the experience so far of the model shows, that the errors average out and
can be included in the empirical constants.

The flame spread of the product is described by an S-shaped curve, which is a function of
time. In the beginning of the test, the product ignites at one point on the test sample. This
ignition time is assumed to be a time equal to half of the ignition time found in the Cone
Calorimeter at 40 kW/m
2
. Immediately after ignition, the area growth rate of the product
is slow. The area growth rate will then accelerate, depending on the time to ignition in the
Cone Calorimeter, until the involved area gets close to its maximum value. Then the area
growth rate slows down again.

The area growth rate is described by the following function:

(
(
(
(
|
|
|
|
.
|
\
|
|
|
|
|
.
|
\
|
+ =
ign
ign
ign
ign
max
t
2
t
t
exp
t
2
t
t
1 1 A A(t) [1]

where A
max
is the maximum area involved and t
ign
is time to ignition in the Cone
Calorimeter.
20

In the beginning of the test, all products are assumed to follow the same area growth
function. However, if the sustained flame height reaches the top of the test specimen,
which is 1.5 m, then the maximum area in the area growth function changes. This is the
only parameter that is changed when changing from one flame spread regime to another.
The sustained flame height is a function of the calculated total heat release in the test as
explained in the section about criteria for flame spread.

The area growth function and the different values for the maximum area are empirically
chosen. However, they agree very well with those observed during the SBI round robin
test series at both SP and Danish Institute of Fire Technology (DIFT).

The maximum area is assumed to be 0.35 m
2
for the products, which do not have a
sustained flame height of 1.5 m. This area is roughly equal to the area behind the burner
flames. For products where the sustained flame height exceeds 1.5 m the maximum area
is 0.60 m
2
. This maximum area is chosen based on the configuration of the SBI test. The
burner has a side length of 250 mm and is positioned at a distance of 40 mm from the test
specimen. If the flames were spreading to the top of the test specimen in the entire width
of the burner the maximum area should be 0.87 m
2
. However, since the burner is
triangular, the thickness of the flame varies. During the tests, it was observed that the
flame leaned into the corner and in no way has a width identical to the width of the burner
in its entire height. Using these areas as maximum areas for flame spread the model gives
good agreement with observations during tests.

Some products will spread the flames more than 0.6 m
2
before reaching their first peak in
heat release rate. These are products with an extremely short ignition time in the Cone
Calorimeter, or thermoplastic products, which create pool fires before reaching their
maximum heat release. The model does therefore not give correct results for these types
of products when it comes to peak heat release. But as will be shown later, the
FIGRA(SBI) index for these products is predicted quite good by the model which is due
to a good estimation of the initial inclination of the heat release curve.

Criteria for flame spread

As shown in Figure 4 the flames will either spread over a small area or over a larger part
of the test sample. The criterion used in this model to decide what the maximum area of
the flame spread will be, is the sustained flame height in the corner.

In the original Cone Tools model the criteria used to determine if flames were spreading
over a larger area than what was initially involved, is an assumed surface temperature.
This assumed surface temperature depended on the temperature of the combustion gasses
passing over the surface and on the thermal response of the product surface. This
approach agrees well with the fact that there is a hot gas layer under the ceiling in the
Room Corner. However in the SBI test there is no ceiling and hence no hot gas layer to
heat the product. The temperature of the surface of the product in the SBI test depends
primarily on the radiation and convective heat transfer from the flame.

Using the flame height to determine the maximum area over which the flames will spread
is based on observations from SBI tests, and on the assumption that the part of the
product behind the sustained flame will receive a heat flux from the flame sufficient to
ignite that part of the product. In the SBI test the product will first ignite in the corner
behind the burner flames. The involved area will then spread mostly upwards behind the
flame from the burner. Depending on the burning behaviour of the product, the heat
release from the burner and the product can be high enough to create a flame in the
21

corner, which will have a sustained height equal to, or higher, than the height of the test
sample. If this is the case the upward flame spread will continue to the top of the test
specimen.

The flame height in a wall corner geometry is given as
9
:

H
D
3 Q
*
2/ 3
=
&
[2]

where

& &
Q Q (D 1110)
*
total
5/3
= [3]

D is the diameter of the burner, which was assumed 150 mm considering that the burner
is triangular. Using these expressions gave the criteria that the total heat release
&
Q
total

shall be greater than 59 kW if the flames shall spread to the top of the specimen. This
criterion also agrees well with what was observed in the SBI tests.

Calculation of heat release rate

The total heat release from the SBI test is obtained by summing up the contributions from
each part of the total burning area and the burner.

& & &
Q Q Q
total product burner
= + [4]

&
Q
burner
is constant at 30 kW while
&
Q
product
varies with time as the fire spreads, the
involved area A(t) increases as described above and the burning intensity at each position
is time dependent.
&
Q
product
is obtained by adding the contributions from burning parts
which have started to burn at various times. The heat release rate of the specimen at each
location is then assumed to go through the same history as was measured in bench-scale,
i.e. the Cone Calorimeter.

&
Q
product
is calculated using the Duhamels integral:

=
t
bs product
d t q A Q
0
) ( ) (
&
& &
[5]

where
&
Ais the time derivation of the burning area, t is time, & q
bs
is the heat release per
unit area as recorded in the Cone Calorimeter and is a dummy variable.

It is not possible to obtain a closed-form solution of this integral unless & q
bs
is given a
suitable analytical form. The following very simple numerical solution to the Duhamel's
integral is the approach used in this model:

i N
bs i product
q A Q
=

&
&
[6]
22

Where A
i
is the incremental burning area growth at the time increment i, and
i N
bs
q

&
is
the heat release rate per unit area after (N-i) time increments as recorded in the Cone
Calorimeter.

Correction for cone calorimeter data obtained at other heat flux levels

The model has been developed to use cone calorimeter at a heat flux level of 50 kW/m2.
In order to be able to use the model also with cone resultants different from the preset
value a correction was introduced for both the ignition time and the HRR level. The
correction is based mainly on fine-tuning the results:

t
ignCorr
= t
ignCone
.* (ConeFlux / SBIFlux)
HRR
Corr
= HRR
Cone
(SBIFlux / ConeFlux)^
0.5

Where:
t
ignCorr:
Corrected ignition time used in the model
t
ignCone
Ignition time in the cone calorimeter test
HRR
Corr
Corrected heat release rate
HRR
Cone
Heat release rate in the cone calorimeter test
ConeFlux Flux level in the cone calorimeter test
SBIFlux Corresponding reference flux for the cone-SBI model being 40 kW/m
2

It is understood that for ignition the correction is based on thermally thin theory but this
has shown to give the best simulation results. The exponent for the HRR correction was
determined in a similar way.

6.1.2 Sensitivity study of model

6.1.2.1 Influence of HRR threshold and ignition time

The main input parameters to the model are the ignition time and the heat release curve.
The HRR curve is automatically registered by a computer, but the ignition time is
obtained from visually observing the experiment. In this project also a HRR threshold
value was used to investigate whether it is possible to run this as an alternative. This is
especially interesting for material with heavy flashing behaviour (FR materials) and for
materials with low HRR levels where maybe even no ignition occurs. From the results in
Table 3 it can be seen that a threshold of 10 kW/m2 can be used as alternative for a visual
ignition time. Using 50 kW/m2 as a threshold gives mainly lower Figra values i.e. results
with a better classification. However, the use of a HRR threshold value should also be
done after studying the actual HRR curve. This can be done in the conetools software
package before the calculations are performed.
From our experience it is also advisable to run a small sensitivity study on the ignition
time in order to investigate whether it has a great influence on the result. If so it can be
advisable to run at another flux level. This is mainly the case for materials with short
ignition times.
23

Table 3 Results of 5 materials with visual ignition time and HRR threshold
values as input for the ignition time
Material Cone test Ignition criterion FIGRA* FIGRA0.2 FIGRA0.4 THR
CP1 Test 1 Visual = 7s 383 274 77 0.61
HRR= 50 kW/m
2
254 176 58 0.61
HRR= 10 kW/m
2
383 274 77 0.61
CP1 Test 2 Visual = 10s 255 197 79 0.68
HRR= 50 kW/m
2
185 135 63 0.68
HRR= 10 kW/m
2
255 197 79 0.68
CP2 Test 1 Visual = 11s 79 67 32 0.58
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
79 67 32 0.58
CP2 Test 2 Visual = 14s 47 43 27 0.61
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
47 43 27 0.61
CP3 Test 1 Visual = 15s 62 44 0 0.45
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
59 41 0 0.45
CP3 Test 2 Visual = 17s 46 36 14 0.53
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
34 27 11 0.53
CP4 Test 1 Visual = None NA2 NA2 NA2 NA2
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
27 25 17 1.1
CP5 Test 1 Visual = None NA2 NA2 NA2 NA2
HRR= 50 kW/m
2
NA1 NA1 NA1 NA1
HRR= 10 kW/m
2
19 0 0 0.72

* without threshold level of THR, only HRR > 3kW
NA1: not applicable since HRR is lower than the threshold (Values would be zero)
NA2: not applicable since no visual ignition occurred (Values would be zero)

6.1.2.2 Influence of backing board

Figure 5 and Figure 6 give the difference between a sample preparation with and without
the standard backing board used in the SBI. It can be seen that this improves the quality
of the simulation, especially in the second part of the SBI curve. It is hence advisable to
use as often as possible a backing board or substrate identical to the one that will be used
in the SBI test.

24

M22 with backingboard
0
20
40
60
80
100
120
140
0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 510 540 570 600
hrrSBIsim(kW)
HRRSBI

Figure 5 Simulation of particle board (M22) with backing board
M22
0
20
40
60
80
100
120
140
0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 510 540 570 60
hrrSBIsim(kW)
HRRSBI

Figure 6 Simulation of particle board (M22) without backing board
25

6.1.2.3 Shiny materials

The total heat flux towards the specimen in case of the cone calorimeter consists mainly
of radiation (more than 90%). This means that materials with a shiny surface such as M4
and Insulation material 2 in Table 4 will reflect a large part of the incident heat flux from
the cone heater. In the SBI test, however, the radiation will be lower than in the cone
calorimeter as a larger part of the incident heat flux is based on convection. Moreover, the
materials will be sooted very fast and hence receive more radiation energy due to an
increase of the surface emissivity. This could be observed for the two above-mentioned
materials. Without a sooted or painted surface the materials did not ignite at a heat flux
level of 50 kW/m
2
(Insulation material 2) or showed a very high ignition time (M4)
resulting respectively in a >B and C classification.

6.1.3 Guidance and description testing protocol

The following guidance can be given when preparing test specimens in the cone
calorimeter:
1. Materials should by preference be tested at 50 kW/m2 unless very short ignition
times are observed. In this case a lower heat flux level can be chosen
2. The preparation of the sample should closely follow the mounting as in the SBI
test. So it is advisable to run the materials in the cone calorimeter with the
backing boards described in the SBI standard.
3. Shiny materials, e.g materials with Aluminium foil facing, should also be tested
with the surface sooted or blackened by paint (with limited combustibility e.g.
heat flux meter paints). The results may in this case be more conservative, but
will allow a better overall prediction.
4. If very short ignition times (less than 5 s) are obtained at the cone heat flux level,
it can be advisable to reduce the heat flux level in the cone calorimeter.

6.1.4 Comparison and discussion of simulation results

In annex B an overview is given of the graphs comparing the SBI test results with the
prediction of the conetools model. The data given here are all SBI RR materials (except
cables and pipes), one Eurefic material (used as market place material in the SBI project),
5 ceiling panels and 3 insulation materials. The generic description of the RR materials is
given in annex C.
From the results in Table 4 it can be seen that a satisfactory prediction tool has been
developed. The marked materials are those where a wrong classification is obtained. In
two cases the materials are melting products (M2 and M7). Here some more research is
needed to try to improve the model if possible,.
In the two other cases the results are so-called borderline results (M5 and M26).

26

Table 4 Summary of simulation results for the cone-SBI model
Material FIGRA_02
(MW/S
2
)
FIGRA_04
(MW/S
2
)
THR (MJ) Euroclass
according
to
simulation
Euroclass
according
to test
result
M01 28 6 0.5 B B
M02 262 262 0.6 D >B
M03 1554 1554 19.0 E E
M04 2109 2109 0.6 E E
M05 1212 1212 26.0 E D
M06 0 0 0.4 B B
M07 428 428 0.6 D >B
M08 37 24 0.35 B B
M09 147 114 1.1 C C
M10 675 658 6.0 D D
M11 60 35 0.35 B B
M12 592 592 23.9 D D
M13 47 28 0.35 B B
M14 96 88 4.9 B B
M15 0 0 0.1 B B
M16 335 335 0.6 D D
M19 10 7 0.35 B B
M20 361 361 0.6 D D
M21 6 3 0.35 B B
M22 473 473 33.6 D D
M23 430 430 0.6 D D
M24 450 450 31.0 D D
M25 421 421 36.7 D D
M26 734 734 35.0 D E
M27 42 36 0.35 B B
M28 21 3 0.35 B B
M29 153 127 1.45 C C
M30 2236 2227 0.6 E E
Eurefic3 476 476 3.2 D D
Ceiling P1 236 78 0.65 C C
Ceiling P2 55 30 0.60 B B
Insulation 1 1404 1052 7.6 E E
Insulation 2 767 747 4.5 D D
Insulation 3 448 448 16.7 D D

27

6.2 Room corner test method

6.2.1 Description of model(s)

In the Wickstrm/Gransson model
8
ignition time as well as the complete heat release
rate curve from the Cone Calorimeter are used. In principle results from a single small-
scale test could be used to predict the full-scale behavior of a product.

Certainly there are lots of products for which it is not possible to predict their behavior in
large scale based on small-scale tests. Examples are products with a protective surface or
with joints, which after some heating suddenly cracks and exposes highly flammable
materials. They need to be tested in large scale to get results that can be used for
evaluating their potential fire hazard.

The calculation model is described in detail below in four sections containing principles,
area growth, heat release rate and criterion for flame spread, respectively. Finally
calculated full-scale heat release rates are shown and compared with measured values.

Principles of prediction model

Three major assumptions have been made in the prediction model of heat release rate in
full scale:

1) The burning area growth rate and the heat release rate are decoupled.

2) The burning area growth rate is proportional to the ease of ignition, i.e. the inverse of
the time to ignition in small scale.

3) The history of the heat release rate per unit area at each location in full scale is to be
the same as in small scale.

As shown Figure 7 the fire spread may follow three different routes. A product is
assumed to spread along routes II and V if a calculated fictitious surface temperature is
higher than a critical value. The calculation is based on data from the Cone Calorimeter
as will be outlined below.

Within the different flame spread regimes the burning area growth rate depends on
ignitability, i.e. time to ignition in the Cone Calorimeter. Once the flame spread rate is
determined the heat release rate is calculated assuming that products always give the
same heat release rate per unit area as a function of time in small and large scale. In other
words all parts of the tested product is assumed to burn in the same way in full scale as in
small scale. This is of course a vast simplification. The heat release rate depends more or
less on the actual heat flux level received by the product as a function of time. The
experience so far of the model shows, however, that the errors average out and can be
included in empirical constants.

28

0
5
10
15
0 5 10 15 20
B
u
r
n
i
n
g

a
r
e
a

(
m
2
)
Time (min)
II
I
V
IV
III
VI
A
B

Figure 7 Area growth as a function of time for the model

Burning area growth rate depends on the time to ignition

The flame spread at the beginning of the test is divided into two subsequent parts. First
the area in the corner behind the burner is ignited. The size of the area is assumed to be
the same for all products. In the second part, the burning area is assumed to grow
according to a given function of time. It will, however, start to grow only if a fictitious
surface temperature is reached. This assumed temperature depends on ignitability as well
as on heat release properties of the product. These parameters are obtained from the Cone
Calorimeter and together they give sufficient information for the model to predict the
flame spread development.

Figure 3 shows how the area behind the burner first ignites first (I) and burns at a certain
heat release rate. As mentioned above products then behave in one of two ways; either
there is a progressive flame spread that eventually will involve the entire room (II), or
there is no further flame spread outside the burner flame region (III). Which category the
product belongs to be determined by the flame spread propagation criterion, which is
described below.

When the burner heat output is raised to 300 kW the flames will become bigger and a
larger surface area will almost immediately be involved (IV). The situation is then similar
to the one at the beginning of the test. Either there is a progressive flame spread (V), or
not (VI).

The two continuous flame spread developments, (II) and (V), are assumed to follow given
functions of time as shown below in this section.

29

A more detailed description of how the flame-spread rate is obtained in the model follows
now. The formulae are empirical and must be seen as proposals that at least with the
current experience have proven to give acceptable results.

First the area behind the burner ignites, see Figure 7. The size of this area (2 m2) is
assumed to be the same for all products while the growth rate of the burning area is
assumed to vary with time normalized with the ignition time, tign, of the product
considered:

A(t) =4(
t
tign
) -1
(1)

Along routes (II) and (V) in Figure 7 the involved area as a function of time is written as,

A(t) = A0 (1 + a
(t-tx)2
tign
)
(2)

A0 is the area behind the burner and a is an empirical constant found to be 0.025 s-1 for
route (II) and tx is chosen so that the burning area growth rate is the same as in eq (1)
when equation (2) starts to be used, i.e. when the surface temperature criterion is
exceeded.

After 10 minutes when the burner output is raised, the burning area is initially expressed
as:

A(t) = 2 +
24
tign
(t-t10)
(3)

until A(t) is 5 m2; t10 = 600 s (10 minutes). If the surface temperature criterion is
reached, the burning area growth will continue as in equation (2) with the parameters A0
= 5 m2 and a = 0.1 s-1 and tx determined as above.

Calculation of the heat release rate

The total heat release from the room is obtained by summing the contributions from each
part of the total burning area and the burner.

Q
total
= Q
product
+ Q
burner (4)

Q
burner
is constant while Q
product
varies with time as the fire spreads; the involved area
A(t) increases as described above and the burning intensity at each position is time
dependent. Q
product
is obtained by adding the contributions from burning parts, which
have started to burn at various times. The heat release rate of the specimen at each
location is then assumed to go through the same history as was measured in bench-scale,
i.e. the Cone Calorimeter.

A time incremental approach is used to calculate Q
product
. At each time step t, an area
increment A is calculated and the Q
product
is obtained by adding the contributions from
each area increment. Then we get at time t = N t

30

N
Q
product
= A
i
q
bs
N-i

i=1
(5)

In an alternative form this equation may be written as an integral (Duhamels integral):

t
Q
product
= eq
0
A'() q
bs
(t-) d
(6)
where A' is the time derivative of the burning area and is a dummy variable. No closed
form solution may, however, be obtained unless q
bs
is given a suitable analytical form.

To understand the summation in the equation (5) we will go through a simple numerical
example. Assume that the heat release rate as measured in the Cone Calorimeter and the
involved area are as shown in Figure 8 and Figure 9, respectively. The incremental
contribution can then be calculated and added as shown in Table 5. The total heat release
rate at the third time increment may for instance be calculated as:

(0.2)(250) + (0.45)(125) + (0.65)(50) = 50 + 56 + 32 = 138.

This incremental technique is computationally very fast and may be carried out with such
a small time increment that the details of the small-scale heat release curve are adequately
considered.

0
100
200
300
400
500
600
1 2 3 4 5 6
5
0
1
2
5
2
5
0
4
7
5
2
2
5
5
0
H
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
k
W
/
m
2
)
Time increments (delta t)
t
ign

Figure 8 Schematic curve from heat release rate as measured in the cone
calorimeter
31

0
1
2
3
4
5
6
7
8
1 2 3 4 5 6
0
.
2
0
.
6
5
1
.
3
2
.
1
5
3
.
5
5
.
9
5
Time increment (delta t)
B
u
r
n
i
n
g

a
r
e
a

(
m
2
)

Figure 9 Burning area growth rate described by parabolic relation

Table 5. Example of heat release rate calculation.

Time A A q
bs
Incremental contributions kW Total
s
m
2
m
2
/s kW/m
2

1 2 3 4 5 6 kW
1t 0.2 0.2 50 10 10
2t 0.65 0.45 125 25 22 47
3t 1.3 0.65 250 50 56 32 138
4t 2.15 0.85 475 95 112 81 42 330
5t 3.5 1.35 225 45 214 162 106 68 595
6t 5.95 2.45 50 10 101 309 212 169 122 914

Legend to Table 5: q
bs
column is the cone calorimeter result, the A column is the
estimated burning area growth. By using the superposition technique with respect to when
the different areas are ignited the total heat release rate is obtained.

Time to ignition plays an important role in this analysis as it governs the growth rate of
the burning area. An alternative calculation form can be obtained by differentiating
equation (2):

A = (2A
0
a/t
ign
) t t (7)

Thus the involved area for a given time is proportional to the inverse of the ignition time.
The summation (equation 5) may then be written in a more convenient form as [7]

32

Q
product
= (2A
0
a / t
ign
) (t
i
q
bs
N-i
t) (8)

This formula reveals how sensitive the calculated heat release rate is to the ignition time
measured in small scale. Therefore several tests ought to be carried out to achieve reliable
ignition time results. For this, some kind of averaging technique need to be developed so
that results from various irradiance levels can be considered. Note that all results reported
here are based on single tests at an irradiance of 25 kW/m
2
in the Cone Calorimeter.

Criterion for flame propagation is a fictive surface temperature

As indicated in Figure 7 the fire may or may not spread away from the vicinity of the
burner at the beginning or after 10 minutes when the burner output is raised to 300 kW,
respectively. The criterion assumed in this model for flame spread to occur is that the
surface temperature,
s
, at an imaginary point just beyond the flame front at some time
reaches a certain critical value. The calculation procedure as outlined below is very
schematic but it is found for the tested products that when the calculated surface
temperature is above a particular value, flame spread occurs.

The surface temperature,
s
, depends on the temperature of the combustion gases passing
by,
gas
, and on the thermal response of the product surface.
gas
in turn is in this model
assumed to depend on the sum of the heat release rate from the burner and the specimen
product in the vicinity of the burner, i.e. Q
total
.

The heat release from the product is calculated with involved areas as outlined above. A
fictitive gas temperature is then in these calculations obtained as

gas
= Q
total
2/5
(9)

where is a proportionality factor here empirically chosen to be 50 and 35 K/W
2/5
at the
burner rates of 100 and 300 kW, respectively. The expression could be compared with the
plume temperatures beyond the flame (the intermittent regime) as calculated by for
instance McCaffrey. The power 2/5 is derived for free flames but as the flames in this
case reaches the ceiling a somewhat higher value could possibly be argued for.

The thermal response of the surface is assumed to be determined by the ignition time in
the Cone Calorimeter only. That gives combined information about the thermal response
of the surface and the ignition temperature. It is assumed in the calculation that the
product is semi-infinite i.e. thermally thick. The thermal response may then be expressed
by the thermal inertia (kc). This parameter varies with orders of magnitude while the
ignition temperature lies within a relatively narrow range. The ignition time gives a
measure mainly of the thermal response characteristics.

Below the thermal response is calculated assuming gas passing by with a temperature
varying with time as shown above. The thermal response of the surface is then calculated
by superposition. The response function is chosen assuming that the body is semi-
infinite and that kc is proportional to the ignition time measured in the Cone
Calorimeter. Other response functions would of course be more adequate in many cases
but for the final result it seem from our experience not necessary to employ other
functions.

33

The thermal inertia, kc, which in theory is relevant only for semi-infinite homogeneous
products, is here replaced by what we call IRV = Ignition Response Value. IRV is
assumed proportional to tign and by matching it with known thermal inertia of wood we
get IRV = 1250 tign, where tign is measured in the Cone Calorimeter at an irradiance
level of 25 kW/m
2
. It would of course also be possible to obtain the IRV value at other
irradiance levels or even by other methods.

The response function of the surface, , expression for semi-infinite bodies exposed to a
constant gas temperature g, as taken from textbooks, is now employed:

=

s

g
= [1 - exp(
t

erfc((
t

)]
(10)

where

= IRV/h2 (11)

The convection heat transfer coefficient, h, is for the numerical calculation assumed to be
50 W/m
2
K. This rather high value is deemed reasonable for a point near the burner
plume.

For gas temperature varying with time, surface temperature can now be obtained by
superposition expressed as (the same numerical technique as outlined above for
calculating heat release rate):

N
s
(t) =
g
i

N-i
(12)
i=1

where
g
i
is the gas temperature at the i:th time increment and
N-i
the response function
at time N-i.

Correction for cone calorimeter data obtained at other heat flux levels

The original model has been developed to use cone calorimeter at a heat flux level of 25
kW/m
2
. In many cases however materials do not ignite at this flux level or the
experiments show poor repeatability. For this reason it was investigated to adapt the data
so that it could be used at higher heat flux levels.
In order to be able to use the model also with cone results different from the preset value
of 25 kW/m
2
a correction was introduced for both the ignition time and the HRR level.
The correction is based on fine-tuning the results for the HRR and assuming materials are
semi-infinite for ignition properties:

t
ignCorr
= t
ignCone
.* (ConeFlux / RCTFlux)^
2

HRR
Corr
= HRR
Cone
(RCTFlux / ConeFlux)^
0.33

Where:
t
ignCorr:
Corrected ignition time used in the model
t
ignCone
Ignition time in the cone calorimeter test
HRR
Corr
Corrected heat release rate
HRR
Cone
Heat release rate in the cone calorimeter test
ConeFlux Flux level in the cone calorimeter test
RCTFlux Corresponding reference flux for the cone-SBI model being 25 kW/m
2

34

6.2.2 Description of test protocol in ISO 5660 used for the
simulations

In the simulation of the room corner test all materials were tested according the ISO 5660
without any specific change in sample preparation. The materials were wrapped in
Aluminum foil and placed on ceramic wool. All tests were conducted with the retainer
frame in place and without grid. The tests were conducted at 50 kW/m
2
.

6.2.3 Comparison and discussion of simulation results

In Table 6 the results are given for the simulation of the SBI RR materials and a number
of the Eurefic test data. All simulations were made with a cone calorimeter test results
obtained at 50 kW/m
2
. For the Eurefic data a number of test results were not taken into
the table, as the cone calorimeter data were not of sufficient quality (too few data points).
The results show that the prediction level is of the same order of success as previously
been shown with the model with input data at 25 kW/m
2
. This shows that the model has
been adapted successfully for use with cone calorimeter data at 50 kW/m
2
or even other
flux levels could be used. This is important as it gives the user a larger flexibility.
The marked results are those were the model does not give the correct class. In two cases
it concerns thermoplastics with heavy melting behavior (M2 and M7). In two other cases
it concerns sandwich panels where it can be expected that joint behavior etc are more
important and where the model has its limitation (M21 and Eurefic 9). In the other two
cases it concerns materials going to flashover at 300 kW (M14 and M15). Here it maybe
should be investigated in the future whether even higher heat flux levels should be used
as input. For the Euroclass of a material in the room corner the procedure outlined in the
EN standard for classification of construction products and building elements has been
used. This means the following classification:
Class B or higher: No flashover
Class C Flashover time larger than 10 minutes
Class D Flashover time between 2 and 10 minutes
Class E Flashover less than 2 minutes

35

Table 6 Summary of simulation results for the cone room-corner model
Material Euroclass
according
to
simulation
Euroclass
according
to test
result
M01 B B
M02 D B
M03 E E
M04 E E
M05 E E
M06 B B
M07 D B
M08 B B
M09 B B
M10 C C
M11 B B
M12 D D
M13 B B
M14 B C
M15 B C
M16 D D
M19 B B
M20 D D
M21 B C
M22 D D
M23 D D
M24 D D
M25 D D
M26 E E
M27 B B
M28 B B
M29 B B
M30 E E
Eurefic1 B B
Eurefic2 D D
Eurefic3 C C
Eurefic4 B B
Eurefic5 B B
Eurefic6 C C
Eurefic9 E D
Eurefic10 C C
Eurefic11 E E

36

7 Development of models for Smoke
production

7.1.1 Statistical method

It has earlier been shown that statistical information from Cone Calorimeter tests can be
used to predict time to flashover in the Room Corner test
10
. Smoke production in the SBI
test and in the Room Corner test has also been predicted with good precision using
multivariate statistical methods
11,7
. The same techniques are used in this project, but on a
more extensive set of test results than in earlier projects. This may have led to some
differences between the calculation models developed here and the models presented in
the earlier studies. However, as the models presented here are built on information from a
larger number of results from fire tests, we believe that the present models will have a
broader range of validity than the previously published models.

Multivariate statistical methods may find links between different variables that are not
obvious to the investigator. In a single Cone Calorimeter test of a product several
variables are recorded, like time to ignition, smoke gas concentrations, heat release rate,
specimen mass loss and optical smoke density. Other parameters are used to describe the
product before testing, like density and thickness. Since a test in the Cone Calorimeter
clearly is a multivariate phenomenon, the test results should be well suited for a
multivariate statistical analysis. Applied on a set of Cone Calorimeter test results a
multivariate statistical analysis may be able to find ways of combining relevant variables
that could be used to predict large-scale fire behaviour.

Multiple discriminant function analysis, abbreviated MDA, is a multivariate statistical
method used to classify cases into groups
12,13
. One case is in our analysis defined as
results from one Cone Calorimeter test. The groups are determined based on a categorical
dependent variable, i.e. a variable that shows discrete values that can be assigned to
discrete classes.

MDA can be used to
classify cases into groups
investigate differences between groups
detect variables that are important for distinguishing between groups
discard variables that are irrelevant for group distinctions

When a relation between groups and variables exist, MDA will find the simplest way of
assigning cases to a set of predetermined groups. The classification is then governed by
functions, which include only the variables that are most strongly related to the group
distinction.

A discriminant function is analogous to multiple regression in that it creates a linear
function between the latent variable L and the n different independent variables x
1
,
x
2
,,x
n
that are found to be relevant for distinguishing between groups:

L = b
1
x
1
+ b
2
x
2
+ + b
n
x
n
+ c (1)

where c is a constant. If there are g different predetermined groups and p different
discriminating variables, a set of the lesser of (g-1) or p functions will be evaluated. Each
37

function is orthogonal to the others, i.e. all functions are independent of the other
functions.

Before performing a discriminant analysis, some assumptions concerning the cases in the
data set must be validated through statistical examination. Population sizes should not
differ too much, and all cases should be independent. The variables are assumed to follow
a multivariate normal distribution, and within-group distributions should be symmetric.
Different populations should have about equal spread of variance for each variable.
Transformation of variables may be necessary to improve normality, stabilise variance
and make distributions more symmetric. However, MDA is said to be relatively robust
against modest violations of these assumptions.

MDA was applied to the data set, to evaluate if this tool could be used to predict the
levels of smoke production in the SBI test and in the Room Corner test with satisfactory
accuracy. The software program SPSS 9.0
14
gives the option to choose development of
Fishers linear discriminant functions for classification of cases. The result of this
analysis is a set of g linear functions, one for each of the g predetermined groups. The
Fishers linear discriminant functions are combinations of the p variables z
1
z
p
that are
found to discriminate between the groups. The functions are expressed in the format

F
i
= a
i
z
1
+ b
i
z
2
++ c
i
z
p
+ constant
i ,
(i=1,..,g) (2)

All functions from F
1
to F
g
are calculated for each case. A new case will be associated to
the group which classification function obtains the highest value.

About 20 different variables assumed to be important for predicting smoke production in
large scale was calculated from the Cone Calorimeter test results. These variables gave
information connected to smoke production, production of CO, HRR, time to ignition,
time to extinction etc. Some of the variables had to be transformed to make them fulfil the
criteria set to variables suited for an MDA. The software program SPSS gives the option
to choose an automatic stepwise selection of variables that are able to distinguish between
the predetermined groups. The variables are chosen based on their statistical significance
as discriminators in the actual analysis.

7.2 Test results from ISO 5660 used in the prediction
models

The Cone Calorimeter test results are recorded until 2 minutes after extinguishment. For
test specimens with a long burning period, only test results from the first 15 minutes are
used in the calculations, otherwise results from the complete test are used.

The statistical models use the recorded arrays of heat release rate, HRR [kW/m
2
] and
smoke production rate, SPR [m
2
/s] as a base for the calculation of the following variables:

Time to ignition, t
ign
[s], which is determined as the moment when HRR exceeds 50
kW/m
2
. This value has to be compared to the observed time to ignition, to avoid noise
influencing the determination of the parameter.
SPR
max
[m
2
/s], which is the maximum value of SPR.
HRR
max
[kW/m
2
], which is the maximum value of HRR.
THR
300s
[MJ/m
2
], which is the total heat released during 300 s after time to ignition.
38

TSP [m
2
], which is the total smoke production calculated over the total test period, or
alternatively over 15 minutes of testing time for tests with longer duration.
SMOGRA
cc
[m
2
/s
2
], which is the maximum value of the ratio between SPR and time
when SPR was measured.
FIGRA
cc
[kW/m
2
s], which is the maximum value of the ratio between HRR and time
when HRR was measured.

The mean density of the outer 10 mm of the tested product,
mean
[kg/m
3
], is also an
important parameter used in the models.

7.3 Prediction of smoke production in the SBI test

7.3.1 General

In the new European system for testing and classification of reaction-to-fire properties
15
,
products are categorised into 3 subclasses according to their smoke production in the SBI
test. The criteria for the additional classifications s1, s2 and s3 are based on the
SMOGRA index and on TSP
600s
as presented in Table 7.

Table 7 Criteria to parameters related to smoke production in the SBI test in
the European classification system
15
.
Smoke class SMOGRA [m
2
/s
2
] TSP
600s
[m
2
]
s1 30 50
s2 180 200
s3 - -

From Table 4 we see that test results of both SMOGRA and TSP
600s
can be divided into
three groups in this system. Our aim is to be able to predict the correct level of both
SMOGRA and TSP
600s
,

and to predict the final level of smoke classification.

The set of products that the prediction model development is based on is collected from
different research projects, and are described in references
10
,
11
,
16
, and
17
. Three additional
products are included in this study, namely the PIR with aluminium facing (Insulation
material 1), PIR with glass fibre facing (Insulation material 2) and polystyrene (Insulation
material 3), all products described elsewhere in this report. 35 products are used for
development of the SBI smoke prediction models, and a number of 116 single Cone
Calorimeter tests have been analysed. For the development of the Room Corner models,
results from a total of 152 Cone Calorimeter tests of 53 different products have been the
base. New Cone Calorimeter test data have been included in this project compared to the
previously published studies
10,11,
.

A point that should be mentioned is that the SBI smoke measurement system was
modified after the SBI round robin was finished. This was done to avoid problems with
sooting of the lenses. A possible result of the sooting may have been that a higher smoke
density than actually present in the exhaust system was measured for some of the SBI
round robin products. The improvement in the smoke measurement system may lead to a
change in the obtained smoke classification for some products, most likely an improved
result. Future work with validation and refinement of the models will take account of the
possible effects the modifications may have had on the products smoke performance in
the SBI apparatus. However, the distribution of smoke performance in the available data
39

set is rather skewed, with the majority of products in class s1, and we therefore believe
that the changes in the SBI apparatus will have negligible effect on the SBI smoke
prediction models presented here.

The smoke classification for all of the products tested in the SBI round robin was
determined by the value of TSP
600s
15
. The same is also the case for 8 of the 9 additional
products used in this analysis. The only case where SMOGRA was governing for the final
smoke class was for the product Insulation material 1, where the SMOGRA value
exceeded the s2 limit with a few units, leading to a final s3 classification of smoke.
However, this conclusion was drawn from a single SBI test only. Because this single test
result would affect the final models to a very high degree, we have chosen to omit it from
this analysis. We do not exclude the possibility that such results may occur from SBI tests
of other products as well, the event is, however, assumed to be rare.

5 variables were found to be able to distinguish between both the three levels of
SMOGRA-values and between the three smoke classes. The selected parameters were:

z
1
= ) ln(
max
max
HRR
SPR

z
2
= TSP
z
3
= ln(SMOGRA
cc
)
z
4
=
mean
[kg/m
3
]
z
5
= t
ign

The data set was divided into one test set containing 89 cases for building the functions,
and one validation set containing 27 cases for testing the model precision afterwards.

40

7.3.2 Predicting the level of SMOGRA

The three classification functions are expressed as follows:

F
SMOGRA1
= - 26.300z
1
+ 3.851z
2
- 11.006z
3
+ 0.004352z
4
+ 0.05247z
5
- 173.937
F
SMOGRA2
= - 21.501z
1
+ 5.801z
2
- 8.629z
3
+ 0.003791z
4
+ 0.03838z
5
- 119.595
F
SMOGRA3
= - 21.566z
1
+ 6.845z
2
- 11.745z
3
- 0.006831z
4
+ 0.04458z
5
- 140.686
(3)

All functions F
SMOGRA1
, F
SMOGRA2
and F
SMOGRA3
are calculated for the case to be predicted.
The predicted SMOGRA level is determined as the level, which associated Fishers
function, gives the highest result. If e.g. F
SMOGRA3
gives a higher result than both F
SMOGRA1

and F
SMOGRA2
, the case is predicted to obtain a SMOGRA level s3 in the SBI.

Figure 10 shows how the model described by equation (3) is able to discriminate between
members of the three SMOGRA levels.

Canonical Discriminant Functions
Function 1
6 4 2 0 -2 -4
F
u
n
c
t
i
o
n

2
3
2
1
0
-1
-2
-3
-4
SMOGRA level
Group Centroids
3
2
1
3
2
1

Figure 10 The statistical classification models ability to separate cases
belonging to different levels of SMOGRA measured in the SBI test.
The diagram shows the scores for each of the 89 cases in the test
set for the two canonical discriminant functions. The circular dark
spots show the group centroids, i.e. the centre of gravity for each
cluster.

41

In Figure 11 the predictability for the different levels of SMOGRA for cases in the test set
and cases in the validation set are presented in confusion tables
18
.

Test set (n=89) Validation set (n=27 )
Predicted level Predicted level
1 2 3 1 2 3
1
65
(97%)
2 0 1
18
(95%)
3 0
Actual
level
2 0
11
(69%)
5
Actual
level
2 0
2
(67%)
1
3 1 0
5
(83%)
3 0 1
2
(67%)

Figure 11 Confusion tables
18
for prediction of SMOGRA level in the SBI test.

The results in Figure 10 and Figure 11 show that the members of levels 1 and 2 of
SMOGRA can be separated quite well based on combinations of these parameters, while
SMOGRA level 3 (only 6 cases in the test set) is more difficult to sort out by this model.

7.3.3 Predicting the smoke classification s1, s2 or s3

The three classification functions are expressed as follows:

F
s1
= - 47.981z
1
+ 13.874z
2
- 12.564z
3
+ 0.005609z
4
- 0.191z
5
- 289.047
F
s2
= - 38.197z
1
+ 12.385z
2
- 10.345z
3
+ 0.004679z
4
- 0.133z
5
- 189.621
F
s3
= - 39.698z
1
+ 20.439z
2
- 13.200z
3
- 0.007048z
4
- 0.156z
5
- 233.990
(4)

All functions F
s1
, F
s2
and F
s3
are calculated for the case to be predicted. The predicted
result is determined as the class, which associated Fishers function gives the highest
result. If e.g. F
s3
gives a higher result than both F
s1
and F
s2
, the case is predicted to obtain
smoke classification s3 in the SBI.

members of the three smoke classes.

42

Function 1
6 4 2 0 -2 -4 -6
F
u
n
c
t
i
o
n

2
6
4
2
0
-2
-4
-6
Smoke class
Group Centroids
s3
s2
s1
3
2
1

Figure 12 The statistical classification models ability to separate cases
belonging to the different smoke classes determined from SBI test
results. The diagram shows the scores for each of the 89 cases in
the test set for the two canonical discriminant functions. The
circular dark spots show the group centroids, i.e. the centre of
gravity for each cluster.

In Figure 13 the predictability for the different smoke classes s1, s2 and s3 for cases in
the test set and cases in the validation set are presented in confusion tables.

Test set (n=89) Validation set (n=27 )
Predicted class Predicted class
s1 s2 s3 s1 s2 s3
s1
56
(100%)
0 0 s1
16
(94%)
0 1
Actual
class
s2 0
19
(100%)
0
Actual
class
s2 0
5
(100%)
0
s3 1 0
13
(93%)
s3 0 1
4
(80%)

Figure 13 Confusion tables for prediction of the smoke classification
determined from SBI test results.

The results in Figure 12 and Figure 13 show that the members of s1, s2 and s3 can be
separated quite well based on combinations of the parameters in equation (4).

43

7.4 Prediction of smoke production in the Room
Corner test method

7.4.1 Flashover is determining for the smoke production

It has earlier been shown that the event of flashover is crucial for the resulting smoke
production in the Room Corner test
11,19,20,21
. The products are therefore grouped according
to in which time interval the flashover, if any, occurs. We have chosen to name these
possible groups FO-categories
10
, and the separation criteria are as follows:

FO-category 1: products not reaching flashover during 1200 s of testing time
FO-category 2: 600 s t
FO
<1200 s
FO-category 3: 120 s t
FO
< 600 s
FO-category 4: t
FO
< 120 s

Before the statistical models for smoke prediction in the Room Corner test can be applied,
the most probable FO-category must be predicted. There are several efficient models
available for this purpose
10
.

7.4.2 How predictable is the Room Corner SMOGRA value?

That the Room Corner test has status as the reference scenario for the SBI test implies
that ranking of materials according to test results from the two methods should be more or
less equivalent. A good correlation between ranking order has been found for heat release
results, using FIGRA from the SBI test and FIGRA

from the Room Corner test. No
obvious and simple correlation has been found between SMOGRA values from the two
methods, and the ranking based on smoke production is very different from ranking based
on heat release. There is at the time being no classification system based on SMOGRA
calculated from Room Corner test results.

When SMOGRA is calculated from Room Corner test results for the products included in
this study, the values cover a broad range, from values near zero to values in the order of
magnitude 10
2
. Smoke production during the very first minutes of the test will be
determining for the final SMOGRA value. We therefore suggest that, like in the
SMOGRA calculations from SBI test results, threshold values should be applied in the
Room Corner calculations. Threshold values could be defined to prevent calculation of
SMOGRA before SPR exceeds a predefined value and the total smoke production is
above a certain level. Because of the uncertainty with regard to the calculation and
applicability of the Room Corner SMOGRA value, we have concentrated on prediction of
other smoke related parameters measured in the Room Corner test, namely the maximum
and average smoke production rate, SPR
max
and SPR
avg
.

44

7.4.3 The EUREFIC classification system

A products performance in the Room Corner test apparatus can be evaluated according
to the classification system proposed through the EUREFIC programme
17
. Time to
flashover, maximum heat release rate (HRR
max
)

and average heat release rate (HRR
avg
)
form the basis for the EUREFIC-classes, while the classification of smoke production is
based on maximum smoke production rate (SPR
max
)

and average smoke production rate
(SPR
avg
).

The EUREFIC-classes and requirements for heat release- and smoke production
parameters are shown in Table 8 below.

Table 8 Classification criteria for smoke production, together with the
corresponding EUREFIC-classes and requirements for heat release
parameters
17
.
EUREFIC
class
Minimum t
FO
1)

[s]
HRR
max
2)
[kW]
HRR
avg
2)
[kW]
SPR
max
3)

[m
2
/s]
SPR
avg
3)
[m
2
/s]
A

1200 300 50 2.3 0.7
B

1200 700 100 16.1 1.2
C

720 700 100 16.1 1.2
D

600 900 100 16.1 1.2
E 120 900 No
requirement
16.1 No
requirement
1)
t
FO
: time to flashover
2)
Heat release rate from burner not included
3)
Smoke production rate from burner not included
Based on this system, both SPR
max
and SPR
avg
can be divided into three levels:

SPR
max

[m
2
/s] : SPR
avg
[m
2
/s] :

Level 1: SPR
max
2.3 Level 1: SPR
avg
0.7
Level 2: 2.3 < SPR
max
16.1 Level 2: 0.7 < SPR
max
1.2
Level 3: SPR
max
16.1 Level 3: SPR
max
1.2

Each set of data in the FO-categories was divided into a test set and a validation set, and
MDA was then performed separately on each group. For FO-category 2, the number of
cases was too small to allow for any separation into subsets. The validation of the
prediction rules for FO-category 2 products is therefore only made through cross-
validation
18
(also called jack-knifing).

45

5 variables were found to be able to distinguish between the three levels of both SPR
max

and SPR
avg
. The selected parameters were:

w
1
=
mean
[kg/m
3
]
w
2
= THR
300s
[MJ/m
2
]
w
3
= ln(t
ign
)
w
4
= ln(FIGRA
cc
)
w
5
= ) ln(
max
max
HRR
SPR

7.4.4 Prediction of maximum smoke production rate in the
Room Corner test

The variables w
1
,,w
5
are used to build sets of Fishers discrimination functions, one set
for each FO-category. We have chosen to name the functions F
i-max_k
. i is a reference to
the level of smoke performance (1, 2 or 3), max refers to prediction of maximum SPR
and k is a reference to the FO-category.

The three classification functions for FO-category 1 are expressed as follows:

F
1-max1
= - 0.004348w
1
+1.070w
2
+6.230w
3
+12.940w
4
-15.094w
5
-87.792
F
2-max1
= - 0.009168w
1
+ 0.846w
2
+9.014w
3
+ 12.513w
4
-12.376w
5
-69.726
F
3-max1
= - 0.003952w
1
+ 1.432w
2
+17.014w
3
+20.660w
4
-13.653w
5
-110.025
(5)

All functions F
1-max1
, F
1-max3
and F
1-max3
are calculated for the case to be predicted. (The
case must, of course, first be predicted to not reaching flashover). The predicted level of
SPR
max
is then determined as the level, which associated Fishers function gives the
highest result.


F
1-max2
= No products in FO-category 2 in the test set belonged to level 1 of SPR
max
.
F
2-max2
= 0.04986w
1
+ 1.326w
2
-7.169w
3
+ 0.294w
4
-14.104w
5
-75.944
F
3-max2
= 0.04096w
1
+ 1.297w
2
-6.775w
3
-1.175w
4
-13.913w
5
-68.389
(6)

The prediction of SPR
max
level is then performed as for the FO-category 1 cases.

It was not possible to obtain any good prediction models for products in FO-category 3.
However, all of the cases in this category in the available data set belonged to either level
2 or level 3 regarding SPR
max
.

members of the three smoke classes for non-flashover (i.e. FO-category 1) products.

46

Function 1
8 6 4 2 0 -2 -4
F
u
n
c
t
i
o
n

2
4
3
2
1
0
-1
-2
-3
-4
SPRmax
Group Centroids
3
2
1
3
2
1

Figure 14 The statistical classification models ability to separate cases in FO-
category 1 belonging to the three different levels of SPR
max
. The
diagram shows the scores for each of the 39 cases in the test set for
the two canonical discriminant functions. The circular dark spots
show the group centroids, i.e. the centre of gravity for each cluster.
47

In Figure 15 the predictability of the levels of SPR
max
for FO-categories 1 and 2 for cases
in the test sets and cases in the validation sets are presented.

FO-category =1:
Test set (n=39) Validation set (n=10)
1 2 3 1 2 3
1
22
(100%)
0 0 1
6
(100%)
0 0
Actual
level
2 3
10
(77%)
0
Actual
level
2 1
3
(75%)
0
3 0 0
4
(100%)
3 - - -

FO-category =2:
Test set (n=25) Test set (n=25, cross validated)
1 2 3 1 2 3
1 0 0 1 0 0
Actual
level
2 10
(100%)
0 Actual
level
2 10
(100%)
0
3 3 12
(80%)
3 4 11
(73%)

FO-category =3: No prediction model for SPR
max
available.

Figure 15 Confusion tables for Classification rules 1and 2 for prediction of
SPR
max
level in the Room Corner test. The grey shaded areas
indicate the levels where no classification rules apply.

48

7.4.5 Prediction of average smoke production rate in the
Room Corner test

The variables w
1
,,w
5
are used to build sets of Fishers discrimination functions, one set
for each FO-category. We have chosen to name the functions F
i-avg_k
. i is a reference to
the level of smoke performance (1, 2 or 3), avg refers to prediction of average SPR and k
is a reference to the FO-category.


F
1-avg1
= 0.008004w
1
+0.07154w
2
-0.227w
3
+9.976w
4
-18.308w
5
-94.359
F
2-avg1
= 0.01445w
1
+0.09239w
2
+1.611w
3
+ 10.318w
4
-14.280w
5
-71.270
F
3-avg1
= 0.002022w
1
+0.01573w
2
+4.466w
3
+9.735w
4
-13.128w
5
-62.885
(7)

All functions F
1-avg1
, F
1-avg3
and F
1-avg3
are calculated for the case to be predicted. (The
case must, of course, first be predicted to not reaching flashover). The predicted level of
SPR
avg
is then determined as the level, which associated Fishers function gives the
highest result.


F
1-avg2
= 0.02123w
1
+0.158w
2
-3.567w
3
+7.893w
4
-23.272w
5
-116.940
F
2-avg2
= No products in FO-category 2 in the test set belonged to level 2 of SPR
avg
.
F
3-avg2
= 0.002587w
1
+0.599w
2
-4.577w
3
+3.910w
4
-19.343w
5
-84.499
(8)

The prediction of SPR
avg
level is then performed as for the FO-category 1 cases.

All products in FO-category 3 belonged to SPR
avg
level 3.

members of the three smoke classes for non-flashover (i.e. FO-category 1) products.
49

Function 1
4 2 0 -2 -4
F
u
n
c
t
i
o
n

2
4
3
2
1
0
-1
-2
-3
SPRavg
Group Centroids
3
2
1
3
2
1

Figure 16 The statistical classification models ability to separate cases in FO-
category 1 belonging to the three different levels of SPR
avg
. The
diagram shows the scores for each of the 39 cases in the test set for
the two canonical discriminant functions. The circular dark spots
show the group centroids, i.e. the centre of gravity for each cluster.

In Figure 17 the predictability of the levels of SPR
avg
for FO-categories 1 and 2 for cases
in the test sets and cases in the validation sets are presented.
50

FO-category =1:
Test set (n=39) Validation set (n=10)
1 2 3 1 2 3
1
18
(82%)
3 0 1
6
(100%)
0 0
Actual
level
2 4
1
(8%)
0
Actual
level
2 0
1
(25%)
0
3 0 9
4
(100%)
3 0 3 0

FO-category =2:
Test set (n=25) Test set (n=25, cross validated )
1 2 3 1 2 3
1 10
(100%)
0 0 1 10
(100%)
0 0
Actual
level
2 0 0 0 Actual
level
2 0 0

0
3 2 0 13
(87%)
3 2 0 13
(87%)

FO-category =3: All cases at SPR
avg
level 3.

Figure 17 Confusion tables for Classification rules 1and 2 for prediction of
SPR
avg
level in the Room Corner test. The grey shaded areas
indicate the levels where no classification rules apply.

7.5 Implementation of the prediction models

The predictions of smoke production are easily calculated on an ordinary PC. The
statistical classification models have been implemented as simple calculation formulas in
an Excel worksheet, and may also be implemented as an algorithm of a computer
program.

51

8 Software development

Within the project it was envisaged to develop a user-friendly software package, which
would calculate the HRR part of the SBI and Room corner models. At a later stage the
smoke model could be incorporated. In the next chapter the different options of the model
are explained which can be used as manual for the programme.

8.1 Principle

The software package is based a Visual Basic programme written for use under the
Windows environment. The outlook of the programme is given in Figure 18 and Figure
18. Figure 18 shows the menu fields while Figure 19 gives the screen when the
programme has started from the windows system. Both a number of scroll down menus
and menu buttons are available which are explained in the following chapters.

Figure 18 Menus in the cone tools software package

Figure 19 Screen after start-up of the programme

52

8.2 Different Button menus

8.2.1 Open file menu

The open file menu, see Figure 20, allows the user to import a cone calorimeter file for
processing within the programme. The browsing function allows selecting the cone
calorimeter. At the same type the user can choose the type of input file. A number of
standard input file types are given such as FDMS export file and a CSV file adapted to
the FTT software. Besides that the user can defines its own type of import file provided
the files are vector files. For sequential data files only the FDMS format is possible. The
user can define is own type of file under the types menu button, see 8.2.7. The load cone
data menu is given in

Figure 20 Open file dialog box
From the moment a file is selected, the cone calorimeter data is imported in the software
and the screen will shown the HRR curve of the cone calorimeter test. Here the user can
investigate e.g. the HRR threshold for ignition. The user can also investigate the quality
of the cone calorimeter data (noise, drift etc.).
53

Figure 21 Screen after import of cone calorimeter data

8.2.2 Calculate menu

Once a cone calorimeter file is imported the calculate menu button can perform
calculations. First the heat flux level has to be introduced. If the input file is an FDMS file
this can be taken from the file immediately. Else the user types in the heat flux level in
the cone calorimeter test. Then the user should indicate whether a HRR threshold or a
visual ignition time is used for the ignition properties. Finally the user can choose the
simulate SBI or room corner test results or both. Clicking the calculate button starts the
calculation and brings the user back into the main menu with the results as shown in
Figure 23. Left of the graph the scalar simulation results can be seen.
54

Figure 22 Calculate dialog box

Figure 23 Result screen

8.2.3 Select what to plot menu

By selecting the scroll menu select what to plot the user can examine the results on the
screen and show what vector data to plot on the screen e.g. simulated HRR in the SBI, see
Figure 24
55

Figure 24 Example of selection of what to plot

8.2.4 Compare menu

Figure 25 Compare dialog box
56

Figure 26 Example of a comparison of two cone calorimeter data sets

With the compare button it is possible to compare e.g. two cone calorimeter data sets. The
user should indicate which vector the HRR data contains and whether some rows have to
skipped (offset). The dialog menu is given in Figure 25
The results of a comparison are shown in Figure 26.

8.2.5 Save SBI and SBI RCT menu

The results from either a SBI or a RCT simulation can be saved as a vector data. The
vector data file contains also a number of scalars such as FIGRA, THR etc. The dialog
box is shown in Figure 27. The data is saved as a comma separated file or the separator
defined in windows under the national settings. In certain countries this can be a semi-
colon.
57

Figure 27 Save data dialog box

8.2.6 Print menu

Figure 28 Print menu

With the print button the actual chosen graphs and the scalar data left of the graph are
printed on the standard printer configured for your computer.
58

8.2.7 Types menu

An important powerful menu in the programme is the types menu. In this menu you can
define the type of file which is to be used for data import. The dialog box is given in
Figure 29. Either a new type can be added by using the add new type button or a
selected type can be deleted by using the delete this type button.

Figure 29 Select type dialog box

The dialog box create new file type, see Figure 30, allows the user to define, within
certain limites, his own file type. First the name of the file type you want is given. Then
the format of your file type is defined by given the first and last row of the data. This is
especially important if header are used in the file. Then the column for time and HRR
vectors are given together with the used units.
The next items are the column separator and the decimal separator used.
Finally the user can define the position of heat flux level and ignition time if necessary.
The creation of a new file type assumes that your data is in a vector format and not in a
sequential format such as FDMS. For FDMS the FDMS file type should be used. In this
case the so called export files should be used (EXP extension).

59

Figure 30 Create new file type dialog box

8.2.8 Help menu

In the help menu a short summary of the different function of the model are included. The
dialog box in the help menu is shown in Figure 31.

Figure 31 Help screen
60

8.2.9 Exit menu

To exit the conetools programme. The user should be aware to have saved all possible
runs.

8.3 Different Scroll menus

8.3.1 File menu

The file menu contains the following scroll items:

1. Open cone file, see 8.2.1
2. Edit file type, see 8.2.7
3. Save SBI data, see 8.2.5
4. Save RCT data, see 8.2.5
5. Print data and diagram, see 8.2.6

8.3.2 Simulation menu

The simulation menu contains the following scroll items:

1. Calculate, see 8.2.2
2. Plot comparison, see 8.2.4

8.3.3 Help menu

The help menu contains the following scroll items:

1. Help, see 8.2.8
2. About conetools: information on the version of the programme and the system
information

61

9 Conclusions

In this project a screening protocol for prediction the SBI and room corner test results
have been developed. The major achievements are:
1. A cone-SBI numerical calculation model for prediction the main HRR
classification parameters of the SBI test methods, namely FIGRA and THR. As
input cone calorimeter data at only one user-defined heat flux level is necessary.
This report also quotes a number of advice for the sample preparation of the cone
calorimeter test results
2. A statistical calculation model for the smoke parameters in SBI and room corner
test by means of the cone calorimeter test results.
3. The extension of the original conetool model (cone-room corner model) by
introducing a correction of the data for heat flux levels other than the original 25
kW/m
2
.
4. Development of a user-friendly software programme in Windows environment.

The major implementation of the results of this project is its use for cost effective product
development and quality control. This cost effective tool is combined with a considerable
reduction of scale and exhaust of combustion gases

62

Annex A Simulation results for cone-RCT model
for HRR

The Eurefic project
Eurefic1
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Painted gypsum plaster board
Eurefic2
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)

Conetools
data

Ordinary Birch Plywood
63

Eurefic3
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Textile wall covering on gypsum paper plasterboard
Eurefic4
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Melamine faced high density non-combustible board
64

Eurefic5
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Plastic faced steel sheet on mineral wool

Eurefic6
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetool
FR particle board type B1

65

Eurefic9
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Polyurethane foam covered with steel sheets

Eurefic10
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data Conetools
PVC wall carpet on gypsum plasterboard
66

Eurefic11
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data

FR polystyrene foam
67

The SBI research project
Plasterboard
M02
0
100
200
300
400
500
600
700
800
900
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
FR PVC
M01
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
68

M03
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
FR extruded polystyrene board

M04
0
100
200
300
400
500
600
700
800
900
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
PUR foam panel with alu foil faces
69

M05
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data

Varnished mass timber, pine

M06
0
100
200
300
400
500
600
700
800
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
FR chip board
70

M07
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
FR polycarbonate panel 3

M08
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Painted plasterboard

71

M09
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Paper wall covering on plasterboard

M10
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
72

M11
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Plastic-faced steel sheet on mineral wool
M12
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Unvarnished mass timber

73

M13
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Plasterboard on polystyrene

M14
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Phenolic foam

74

M15
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Intumescing coating on particleboard

M16
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Melamine faced MDF board

75

M19
0
50
100
150
200
250
300
350
400
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Unfaced rockwool

M20
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Melamine faced particleboard

76

M21
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Steel clad expanded polystyrene sandwich panel

M22
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Ordinary particle board

77

M23
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Ordinary plywood (birch)

M24
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Paper wall covering on particleboard

78

M25
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Medium density fiberboard
M26
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Low density fiberboard

79

M27
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
Conetools
data
Plasterboard/FR PUR foam core

M28
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

8
k
W
)
Conetools
data
Acoustic mineral fiber tiles

80

M29
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools
Textile wall covering on calcium silicate board

M30
0
200
400
600
800
1000
0 2 4 6 8 10 12 14 16 18 20
time (min)
h
r
r

(
k
W
)
data
Conetools

Paper-faced glass wool

81

Annex B Simulation results for Cone-SBI HRR
model

Eurefic data
Eurefic3-SBI
0
5
10
15
20
25
30
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Textile wall covering on gypsum plaster board
82

SBI RR materials
M01-SBI
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Plasterboard
m02-SBI
0
5
10
15
20
25
30
35
40
45
50
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
FR PVC
83

M03-SBI
0
50
100
150
200
250
0 50 100 150 200 250 300
time (s)
h
r
r

(
k
W
)
data
Conetools

FR extruded polystyrene board
M04-SBI
0
20
40
60
80
100
120
140
160
180
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
PUR foam panel with alu-foil faces
84

M05-SBI
0
10
20
30
40
50
60
70
80
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Varnished mass timber, pine

m06-SBI
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
FR chip board
85

M07-SBI
0
20
40
60
80
100
120
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
FR polycarbonate panel

M08-SBI
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Painted plasterboard
86

M09-SBI
0
2
4
6
8
10
12
14
16
18
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Paper wall covering on plasterboard

M10-SBI
0
5
10
15
20
25
30
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

87

M11-SBI
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Plastic-faced steel sheet on mineral wool

M12-SBI
0
10
20
30
40
50
60
70
80
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data

Unvarnished mass timber

88

M13-SBI
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Plasterboard on polystyrene

M14-SBI
0
1
2
3
4
5
6
7
8
9
10
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
Phenolic foam

89

m15-SBI
0
2
4
6
8
10
12
0 100 200 300 400 500 600
time (s)
h
r
r
data
Conetools

Intumescing coating on particleboard

M16-SBI
0
10
20
30
40
50
60
70
80
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Melamine faced MDF board

90

M19-SBI
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Unfaced rockwool

M20-SBI
0
20
40
60
80
100
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
Melamine faced particleboard

91

SBI-21
0
0.5
1
1.5
2
2.5
3
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
Steel clad expanded polystyrene sandwich panel

M22-SBI
0
10
20
30
40
50
60
70
80
90
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Ordinary particleboard

92

M23-SBI
0
20
40
60
80
100
120
140
160
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data
Ordinary plywood (Birch)

M24-SBI
0
10
20
30
40
50
60
70
80
90
0 100 200 300 400 500 600 700
time (s)
h
r
r

(
k
W
)
Conetools
data
Paper wall covering on particle board
93

M25-SBI
0
20
40
60
80
100
120
140
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
Conetools
data

Medium density fibreboard

M26-SBI
0
20
40
60
80
100
120
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Low density fiberboard
94

M27-SBI
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
Plasterboard/FR PUR foam core

M28-SBI
0
1
2
3
4
5
6
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Acoustic mineral fiber tiles
95

M29-SBI
0
2
4
6
8
10
12
14
16
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)data
Conetools

Textile wall covering on calcium silicate board

M30-SBI
0
20
40
60
80
100
120
140
160
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools
Paper-faced glass wool
96

Additional data
CP-1
0
1
2
3
4
5
6
7
8
9
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Ceiling panel 1
CP-2
0
1
2
3
4
5
6
7
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Ceiling panel 2
97

CP-3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Ceiling panel 3

CP-4
0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Ceiling panel 4

98

CP-5
0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600
time (s)
h
r
r

(
k
W
)
data
Conetools

Ceiling panel 5
99

Annex C Generic description of SBI RR materials

Code

Product name Thickness
(mm)
Density
(kg/m
3
)
Surface
weight
(g/m
2
)
M01 Plasterboard 13 700 -
M02 FR PVC 3 1180 -
M03 FR extruded polystyrene board 40 32 -
M04 PUR foam panel with alu foil
faces
40 PUR:40 -
M05 Varnished mass timber, pine 10 380 -
M06 FR chip board 12 780 -
M07 FR polycarbonate panel 3
layered
16 175 -
M08 Painted plasterboard 13 700 Paint:145
M09 Paper wall covering on
plasterboard
13 700 Paper:200
M10 PVC wall carpet on
plasterboard
13 700 PVC: 1500
M11 Plastic-faced steel sheet on
mineral wool
0,15 + 1 +50 Min.wool:
160
-
M12 Unvarnished mass timber,
spruce
10 450 -
M13 Plasterboard on polystyrene 13 + 100 700/20 -
M14 Phenolic foam 40 - -
100

Code

Product name Thickness
(mm)
Density
(kg/m
3
)
Surface
weight
(g/m
2
)
M15 Intumescent coat on particle
board
12 700 Paint:500
M16 Melamine faced MDF board 12 MDF: 750 Mel.:120
M19 Unfaced rockwool 50 145 -
M20 Melamine faced particle board 12 - -
M21 Steel clad expanded
polystyrene sandwich panel
0,5 + 100 EPS:20 -
M22 Ordinary particle board 12 700 -
M23 Ordinary plywood ( Birch ) 12 650 -
M24 Paper wall covering on particle
board
12 700 Wallpaper:
200
M25 Medium density fibre board 12 700 -
M26 Low density fibre board 12 250 -
M27 Plasterboard/FR PUR foam
core
13 + 87 PUR:38 -
M28 Acoustic mineral fibre tiles 18 Min.wool:
220
-
M29 Textile wall paper on calcium
silicate board
CaSi.10 CaSi.875 Wallpaper:
400
M30 Paper-faced glass wool 100 18 90

COMMENTS:
Plasterboard is in all cases in this report gypsum plasterboard.
Products are not fire retardant treated unless specified with FR

101

References

1
ISO 5660; Fire Tests Reaction to Fire Rate of Heat Release from building
products, International Standards Organisation (ISO), 1991.
2
EN 13823 Reaction to fire tests for building products - Building products
excluding floorings - exposed to the thermal attack by a single burning item,
CEN, February 2002.
3
ISO 9705:1993(E), Fire Tests - Full-scale room test for surface products, ISO
1993
4
B. Messerschmidt, P. Van Hees, U. Wickstrm, Prediction of SBI (Single
burning item) test results by means of Cone Calorimeter Test results,
Interflam proceedings 1999 pp 11-22, Interscience communications Ltd,
London 1999.
5
P. Van Hees, The need for of a screening method for the major Euroclass
methods, Flame Retardants Conference proceeding 2002.
6
T. Hakkarainen, Correlation studies of SBI and Cone Calorimeter test results
Interflam proceedings 2001 pp 519-530, Interscience communications Ltd,
London 2001.
7
A. Steen Hansen, P. Hovde, Prediction of smoke production based on statistical
analyses and mathematical modelling, Interflam proceedings 2001 pp 113-
124, Interscience communications Ltd, London 2001.
8
Wickstrm U. and Gransson U, Full-Scale/Bench-Scale Correlations of Wall
and Ceiling linings, Journal of Fire and Materials, vol. 16, 1992.
9
McCaffrey B, Flame Height, The SFPE Handbook of Fire Protection
Engineering, 2
nd
edition, Chapter 2-1
10
Hansen, A S. Hovde P J. Prediction of time to flashover in the ISO 9705 Room
Corner test based on Cone Calorimeter test results. Submitted to Fire and
Materials, May 2001. Revised January 2002.
11
Hansen, A S. Hovde P J. Prediction of Smoke Production in Large and
Intermediate Scale Tests based on Bench Scale Test Results. A Multivariate
Statistical Analysis. Proceedings of Fire and Materials 2001 Conference,
January 22-24 2001, San Francisco, USA, pp 363-374.
12
Kinnear, P R. Gray, C D. SPSS for Windows made simple. Release 10.
Psychology Press Ltd, Publishers, East Sussex, UK, 2000, pp 319-331.
13
Garson D. Notes to course PA 765 Quantitative Research in Public
Administration, at website http://www2.chass.ncsu.edu/garson/pa765/mda.htm,
(February 2001) North Carolina State University, USA.
14
SPSS Inc. 1999: SPSS
Base 9.0 Applications Guide ISBN 0-13-020401-3,

Chicago, USA, 1999,
p 248
.
15
EN 13501-1 :2001 E. Fire classification of construction products and building
elements Part 1: Classification using test data from reaction to fire tests.
European Committee for Standardization (CEN), Brussels, Belgium,
February 2002.
16
stman, B A-L. (2001) Wooden facades in multi-storey buildings. Proceedings
of Fire and Materials 2001 Conference, January 22-24 2001, San Francisco,
USA, pp 185-196.
102

17
Wickstrm, U. (editor) (1991) Proceedings of the International EUREFIC
Seminar 11-12 September 1991, Copenhagen, Denmark. ISBN 0 9516320 19.
Interscience Communications Limited, London, England.
18
Johnson, R A. Wichern, D W. (1998) Applied Multivariate Statistical Analysis.
C
hapter
s 11.4-11.6. Fourth edition. ISBN 0-13-834194-X Prentice-Hall, Inc.,
New Jersey, USA.
19
stman, B. Tsantaridis, L. Stensaas, J P., Hovde, P J. (1992) Smoke Production
in the Cone Calorimeter and the Room Fire Test for Surface Products
Correlation Studies. Trtek, Report I 9208053, Stockholm 1992.
20
Heskestad, AW. Hovde, PJ. (1993) Evaluation of smoke test methods for
classification of building products. Nordtest Technical Report 220, Approved
1993-10.ISBN 82-91412-00-6. University of Trondheim, Norway.
21
Heskestad, AW. Hovde, PJ. (1999) Empirical Prediction of Smoke Production
in the ISO Room Corner Fire Test by Use of ISO Cone Calorimeter Fire Test
Data. Fire and Materials, 23, 193-199.

1479 00

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

1479 00

Transféré par

Droits d'auteur :

Formats disponibles

Patrick Van Hees

Base 9.0 Applications Guide ISBN 0-13-020401-3,

Vous aimerez peut-être aussi