An Expert System to Assess Fire Safety
in Dwellings

H. A. DONEGAN, I. R. TAYLOR and R. T. MEEHAN
1nstitute of 1nformatics
University of Ulster
Shore Road, Newtownabbey BT37 OOB, UK

ABSTRACT

This paper describes the application of an expert system to the evaluation
of fire safety in dwellings based on the body of knowledge developed by the
Fire Engineering Research Group at the University of Ulster. The
background and philosophy of the evaluation procedure together with the
associated reasoning with respect to the choice of system and its
implementation are outlined in some detail. This demonstration syat.em is
intended as a pilot study for a more ambitious programme. A discussion
relating to problems with the system and future developments concludes the
paper.

KEYWORDS: Fire safety evaluation, expert system, Delphi, points scheme.

INTRODUCTION

In recent years the notion of fire safety evaluation with respect to
buildings has become inextricably bound up with the perception of the
prioritisation of those entities which taken together comprise the fire
safety components of a specific building type. The prioritisations, often
but not exclusively the result of expert opinion, are used in the
development of points schemes. These are essentially research into
practice devices which facilitate the economical allocation of scarce
resources in the design and refurbishment of buildings. The philosophy is
clearly extendible to any form of shelter, e.g. ships, planes and trains,
where life safety and property protection are paramount. This paper will
focus on dwelling fire safety with specific reference to the determination
of a fire safety quality measure which can be optimised interactively using
the expert system shell Xi Plus in a PC environment. The level of
optimisation achieved is a function of the user's requirements and
resources given the existence of acceptable norms.

The chronology of events leading to this work began with the study by
Nelson and Shibe [ll who produced a system for the Fire Safety Evaluation

FIRE SAFETY SCIENCE-PROCEEDINGS OF THE THIRD INTERNATIONAL SYMPOSIUM, pp. 485-494 485

 
 
Copyright © International Association for Fire Safety Science



of health care facilities in the USA. Following these developments, the
Department of Health and Social Services (DHSS) sponsored the Fire
Engineering Department at the University of Edinburgh to produce a fire
safety evaluation points scheme for patient areas within hospitals. The
work was conducted by Marchant [2] and completed in 1982. Stollard [3]
reported on the development of the points scheme at Edinburgh and outlined
the procedures which were used. In 1983 a pilot study [4] based on the
methodology of Marchant was undertaken to consider the application of the
DHSS scheme to dwellings. This led to a programme of work in 1984 at the
University of Ulster, funded by the Science and Engineering Research
Council. The theoretical consequences [5] and practical considerations [6]
which followed provide the immediate environment for the present expert
system.

The motivation for the expert system derives from the complexity of
the total fire safety scheme and the need to create a 'what if' environment
for the user. The ideas in [2] and [4] are directed towards the production
of a weighted ranking, referred to as a'priority vector, of fire safety
components. Typical components might be management, occupants and
visitors, contents, fire brigade, detection systems, fire fighting
equipment and so on. The weighted rankings for n such components
(w

1,w2,w 3
' " .w

n)
emerge from the consensus of expert opinion generally

obtained through an application of the Delphi process [7]. Notwithstanding
certain shortcomings of the latter [8] and assuming that a priority vector
can be arrived at, the fundamental and as yet unanswered question is not
with regard to its existence, but with regard to its significance in the
field of application. More particularly its optimal effectiveness - can it
be used to minimize safety maintenance costs or to allocate scare resources
in the pursuit of maximum safety? The authors would contend that such
issues .i.n the environment of fire safety analysis are the driving force
behind research into practice. On the premise that total safety while
being a desirable objective is something which can only be approximated
within technological development periods, the approximation is presumably
the stability of professional expert opinion within each time period.

Fire safety history provides a wealth of examples of connected time
periods, each new period arising as a result of at least one catastrophe at
the conclusion of or within a previous time period. The overall effect has
been the production of more and more prescriptive legislation whi'ch
inhibits an engineering approach to the solution of many problems. At the
present time, within the constraints of existing legislation, consideration
is being given to engineering strategies which will facilitate a degree of
trade-off between passive and active fire safety measures. This is where
the application of an expert system capable of logical repeatability has a
distinct advantage in that it provides a basis for uniformity in the
decision making process. It is clear from Figure 1 that an expert system
which provides the user with a 'what if' environment fits comfortably as
part of a generalisation of the simple fire safety framework proposed in
[5J.

Prior to the introduction of the expert system the procedure in using
a points scheme involved taking the scalar product of the priority vector
v: (w

1,w 2,w 3
' " .w

n)
with each of two corresponding sets of component

486



(DELPHI INFORMATION I

ANALYTICAL
MODEL

r--------------- ----------------,
I I
I I
I I
I I
I I E

I NORM SURVEY : ;
I VECTOR VECTOR I E
I I R
I ~ ~ IT

I ~ II I
I I S
I I Y
I NORM CHARACTERISTIC I S
: SCORE Q SCORE S i ~
: i M
I I
I IE
I IN
I I v
I I I
I I R
I 1 0
I I N
I 1 M

: I WHAT IF SURVEY VECTOR VC CHANGES I i ~
1 ~ : T
I I
I OUTPUT I
I I
I ANALYSIS I
I I
I IL J

1'1C.i GENERALISED FRAMEWORK FOR FIRE SAFETY EVALUATiON

487



points score allocations - a norm set v
N

(Q1,Q2,q3'" .Qo) for a specific
c

dwelling category and a survey set V (sl,s2,s3'" .so) for an actual

dwelling in the category (4). The corresponding products, the norm score Q
and characteristic score S were then compared.

The advantage of the above generalisation became apparent when it was
realised that it was quite feasible for a dwelling to satisfy the overall
criterion, viz: S > Q and yet fail in one or more subsets of the
components. Given that there are 2

0
- 1 mathematically possible subsets

for the n components this can lead to a large number of possibilities for
large n . However in this case with n = 11, (6), it was decided to cluster
the components from a fire engineering point of view as follows:

A: Human measures
B: Building specific or passive measures and
C: Supportive or active measures.

Table 1 exemplifies the categorisation used in this study.

TABLE 1. Fire Safety Component Clustering

A:Human B:Passive C:Active

Occupants and Internal Design Fire Brigade
Visitors Survey Volume Detection Systems

Contents Means of Escape First Aid Fire
Management Hazard Protection Fighting Equipment

External Envelope

An important feature of the conceptualisation was to distinguish the
notions of passive/active safety from passive/active measures. Literature
on safety (9J defines active safety as accident prevention and passive
safety as accident protection, with the proviso that every critical event
is a series of casual events within which it is possible to practice
accident prevention (active safety). Once the critical event occurs it is
only possible to practice accident protection (passive safety). Given the
situation of a critical event (an unwanted fire), fire engineering
literature refers to active and passive measures designed to combat the
event. It is clear that trade-off between active and passive measures is
possible but there is no potential for trade-off between active and passive
safety. Figure 2 shows the fundamental relationship between these
variables.

From this diagram it is possible to see that total fire safety will
only arise as a result of complete active safety or total prevention
whereas within a technological time period passive safety is the limiting
state of prevailing expert opinion. The implicit assumption is that as
technological time periods progress there should be a statistically
observable decrease in fatalities and in property destruction. The
efficiency of any system can only be measured in the future. Meanwhile the
present approach is a contribution to the organised thinking which is
essential for any level of improved safety. The next sections will describe
the choice of application and its implementation.

488



HUMAN
MEASURES

FIGURE 2 Passive/Active Safety and Passive/Active Measures

CHOICE OF SYSTEM

Bearing in mind the above considerations the resultant expert system
must be capable of:
1 - deciding overall if a dwelling is adequate for fire safety,
2 - allowing some trade off between active and passive measures,
3 - suggesting acceptable requirements on each of the measures A,B and C

and of
4 - allowing changes in the dwelling specification to test 'what .if'

questions.
It was decided to investigate if an' expert system could meet the above
requirements with the possibility that such a system could also answer
"why" questions, justifying its conclusions. Such a computer system should
also be consistent in its answers and allow a variety of people in
different locations to access its expertise. By building in help
information at various levels, the system can be employed by more users,
allowing them to learn from the inbuilt expertise. To aid such use,
information about the dwelling under consideration must be input, with the
system asking appropriate questions and using default values if no
information is available.

To test its feasibility the authors have applied the expert system to
the study of the previously mentioned points system for evaluating fire
safety in dwellings, [4,6,10]. The eleven components of fire safety, as
identified by the Delphi technique, are clustered into the A, Band C
categories described above. The expertise of the system includes
identification of the eleven components, their relative weighting as given
by the vector V' and the norm score Q for which a particular kind of
dwelling would be deemed to satisfy fire safety requirements. An expert
system requires a knowledge base within which such expertise is
represented, an inference engine to process the knowledge and an interface
to users and developers. Within an expert system shell the latter two
facilities are already provided allowing the user to concentrate on the
development of the knowledge base. Since the shell is 'domain-free' and so
contains no information on fire safety, the knowledge base must contain all

489



the expertise of the system, which as well as rules and facts includes
appropriate questions and help for the user. For thig investigation the PC
based shell xi Plus was chosen, in which the knowledge is represented in
a rule-based form. The shell allows links to, for example, databases,
spreadsheets and C programs and this was utilised for some of the data.
The shell provides a comprehensive set of tools for developing and testing
the knowledge base, within a user friendly environment. While it does not
allow for probabilities within the data this was not required in the
present project.

Relatively little work on the application of expert systems within
fire safety has been pub Li s he d , with most of it on the compliance of
buildings with fire regulations. The best known example is the commercial
program BRIGADE [11], a system with 4000 rules based on the shell LevelS,
while there are discussions of ongoing work in [12].

IMPLEMENTATION

In developing an expert system the main problems are obtaining the
expertise and then deciding how to represent it within the knowledge base.
In basing this expert system on the work of [4,6], the acquisition of
knowledge using the Delphi technique has been performed and it remains to
represent it within the shell. The components of Table 1 are structured as
illustrated as in figure 3, with appropriate questions, on-line help and
corresponding rules for each component.

I FIRE SAFETY I

I
HUMAN

I
I PASSIVE I

I
ACTIVE

IMEASURES MEASURES MEASURES

I I I I
Occupants Management Contents Fire Detection Fire
and Brigade Systems Fighting
Visitors Equipment

I I
I
Int~rnall I Survey I [Means ofl IHazardl IExternal I
Deslgn Volume Egress Protection Envelope

FIGURE 3 External Knowledge Structure

490



In the system as described in [6], worksheets were used to assess each
of the eleven components with a rating on a scale from 0-5. Each component
set was then weighted by the priority vector V' to give an overall score, S
in a scale 0-500. Expert opinion indicates that for the house type
considered in the investigation an overall score of 375 or 75% would be
sufficient for the dwelling to satisfy fire safety standards. This equates
with norm scores for each measure of (A, 154) , (B, 154) and (C, 67), each
being 75% of the corresponding maximum allocations of (A,205), (B,205) and
(C,90). To allow for trade-off between the different measures the selected
norm scores for A, Band C must be less than 75% of the overall maximum
allocation. This is illustrated in the Results section below.

Within the expert system, separate knowledge bases were constructed
for the three measures A, Band C, aiding separate development and testing.
The information required is obtained by prompting the user with various
questions. The answers are tested by the inbuilt rules of the knowledge
base and along with other information stored as data the program decides on
an overall score for the dwelling.

Unlike the manual system, the expert system can allow different levels
of trade-off between A, Band C by setting separate norms for these
measures. For example in obtaining a rating for "occupants and visitors"
the information considered is the number of occupants compared with the
number of bedspaces, the number of floors and the ratio of adults to
dependents. A screen of the form of figure 4 is used to request
information, with help available if required. Within this component there
are rules such as

IF
THEN

IF
THEN

number of occupants
occupant score = 1.7

dwelling is two storey
survey score = 1.3

bed spaces

with a total of 17 rules for the first component. Under A for human
measures there are 46 rules, in B on building specific measures there are
80 rules with 34 rules in C. To test a system with a total of 160 rules,
as many as possible paths through the system must be tried. A set of.test
cases representing standard and extreme situations was developed, to probe
the system for potential limitations and weaknesses.

RESULTS

Applying the program to an actual dwelling, typical output might be

A: Human Measures
B: Building Specific Measures
C: Supportive Measures

OVERALL

491

actual
score

170
165

50
385

selected
norm score

141
135

58
375

pass/fail
p
p
f

P



Application: Fire Safety
Knowledgebase: Human Measures

,-----occupants and visi tors------------------------,

How many occupants in dwelling? 3

How many bed spaces in dwelling? 2

single storey
What size is dwelling? *two storey

greater than two storey

How many able adults in dwelling? 1

How many dependants in dwelling? 2

PRESS Fl FOR HELP

Tab/Backtab next/previous field
Esc cancel I I CTRL+Rtn end I I F3 why II
FIGURE 4: Information requested on occupants and visitors, within human
measures.

To allow trade-off the selected norm scores in this example are (A, 141),
(B,135) and (C, 58). The overall required score is of course 375 (75% of
500) . Trade-off is occuring be t vee n C, below standard, and A and B to
give an overall pass for the dVlelling.

In contrast for another dwelling the results may be

actual selected
score norm score pass/fail

A: Human Measures 145 141 P
B: Building Specific Measures 140 135 P
C: Supportive Measures 70 58 P

OVERALL 355 375 f

so that the dwelling fails to reach the required standard even though it
passes in each of A, Band C. If trade-off is allowed, by setting the
norms for A, Band C beloVl the overall standard required, then inevitably
some dwellings will pass each aspect but fail to reach the required overall
standard, as in this example. By setting the required norm scores for A, B
and C at 154,154 and 67 respectively, ie 75% of each maximum possible score
of 205,205 and 90, no trade-off would be allowed between the measures. The
philosophy of trade-off is discussed further in [13].

If a dwelling fails to reach the overall standard, the expert system
allows a 'what if' type investigation to be performed to ascertain what
improvements may be made to bring the house up to specification. with the
addition of some financial information the, alternate cost of a range of
possible alterations to the building may be explored using the expert
system. The program can be easily adjusted to allow trade-off only between
the passive and active measures, Band C.

492



CONCLUSIONS

This paper shows that a points scheme for the fire safety evaluation
of dwellings may be computerised using an expert system. The program
illustrates the merits of using an expert system shell to encapsulate the
knowledge, especially in comparison to using a conventional programming
language. The program can be run at all stages of development with a
built-in user friendliness. On any consultation, context sensitive help is
available to the user at the level required, the user may ask the system
why some information is required and explore 'what if' type situations.
When loaded in a portable PC, the system could be employed on-site to
assess the status of a dwelling and, if required, to suggest options for
improving its level of fire safety to any required level. The program
allows trade-off between different aspects of fire safety, but the amount
of trade-off may be adjusted. The system could be developed to assess the
fire safety of for example public assembly buildings as discussed in [14].
On the basis of this pilot study the authors are convinced that subject to
appropriate funding, the expert system incorporating a points scheme could
be extended to more complex buildings where the risk factors are likely to
be much higher.

REFERENCES

1. Nelson, H.E., and Shibe, A.J.,"A system of Fire Safety Evaluation of
Health Care Facilities", Report NBSIR 78-1555-1, NBS, 1978.

2. Marchant, E.W., (ed), "Fire Safety Evaluation (points) Scheme for
Patient Areas within Hospitals", A report on its origins and
development sponsored by DHSS Department of Fire Safety Engineering,
University of Edinburgh, 1982.

3. Stollard, P., "The Development' of a Points Scheme to Assess Fire
Safety in Hospitals", Fire Safety Journal, 7, 145-153, 1984.

4. Shields, T.J. Silcock, G.W. and Bell Y., "Fi-;::e Safety Evaluation of
Dwellings", Fire Safety Journal, 10 , 29-36, 1986.

5. Donegan, H.A., Shields T.J., and Si1cock G.W., "A Mathematical
Strategy to Relate Fire Safety Evaluation and Fire Safety Policy
Formulation for Buildings" in Fire Safety Science - Proceedings of the
2nd International Symposium, Tokyo, pp. 433-441, 1988.

6. Shields, T.J, Silcock G.W. and Donegan, H.A., "Assessing Fire Risk in
Dwellings", University of Ulster, 1989.

7. Linstone, H.A. and Turoff, M., (eds), The Delphi Method, Techniques
and Applications, Addison Wesley, 1975.

8. Shields, T.J., Silcock, G.W. and Donegan, H.A., "Methodological
Problems Associated with the Use of the Delphi Technique", Fire
Technology, 23, 175-186, 1987.

9. Wilson, G.A., "Techniques of Safety and Risk Management", Proc. One
Day Seminar, Dept. of Mech. and Ind. Eng., University of Ulster and
Plant Safety, 22, 2, 1990.

10. Shields, T.J., Silcock, G.W. and Donegan, H.A., "The Development of a
Fire Safety Evaluation Points Scheme for Dwellings, Part I - Some
Theoretical Considerations", Fire Safety Journal, 15, 313-324, 1989.

11. Peregrine Expert Systems Ltd, "Brigade", Dublin, 1988.
12. Hamilton G., Harrison A.P., Pascall J.R., Directory of Research and

development of expert systems in the construction and building

493



services industries Vol III, BSRIA Ref 7177, 1983.
13. Shields T.J., Silcock G.W. and Donegan H.A., "A Philosophy for Trade-

Off", Report for Department of Environment, NI, 1989.
14. Shields, T.J., Silcock G.W. and Donegan, H.A., " Towards the

development of a fire safety systems evaluation for public assembly
buildings", Construction Management and Economics, ~, 147-158, 1990.

494