You are on page 1of 12

1

Javier Trelles, Ph.D., Craig L. Beyler, Ph.D., Jason E. Floyd, Ph.D., Joseph L. Scheffey,
P.E. (Hughes Associates, Inc.), and Kim W. Yee (Naval Surface Warfare Center)

Fire and Smoke Spread Modeling to Support Damage
Control Assessment and Decision Making in
Shipboard Environments


ABSTRACT

Predictive modeling of the spread of fire, smoke,
radiological, chemical, and biological agents is
needed to reduce the manning requirements of
future ships. Physics-based models for fire
hazards do exist but they are not integrated with
the ships sensors and they are not designed to
work as faster-than-real-time advisory tools.
Current generation damage decision and
assessment system uses the Fire and Smoke
Spread Simulator (FSSIM), a physics-based
network model, to pre-compute fire & smoke
scenarios for the systems heuristic fire and
smoke federate. The next generation will
incorporate predictive federates. FSSIMShip is
being developed as the real-time, deterministic
fire and smoke spread model. The objective is
to test and demonstrate a successful prototype
capable of sensor driven predictions of
shipboard conditions operating at or faster than
real time. The results will forecast fire & smoke
conditions in order to support the damage
control decision making process. The effort
involves removing inefficiencies in the source
code and underlying algorithms as well as
incorporating multi-processor ready solvers into
FSSIM. A sensor-driven element is being added
to the model, allowing for continual verification
and adjustment of the fire and smoke spread
prediction outputs. Use of the model in a multi-
processor shipboard computing environment is
envisioned. As such, the algorithms are being
updated for parallel processing. This paper
provides an update on this effort.

INTRODUCTION

In an effort to reduce life-cycle cost, the United
States Navy (USN) currently has a reduced
manning initiative with the goal of cutting
assigned tasks by means of automation and
computer-based decisions. A key to the success
of this initiative is to determine the right balance
of automated control systems capable of
supplementing manning for both normal and
damage control operations. During casualty
operations, manning can be reduced if the ships
automation is capable of determining the extent
of current damage, the likely progression of the
damage over time, and recommend and/or take
action to mitigate the damage quickly and
efficiently. Since damage can progress rapidly,
automation for assessing damage must occur fast
enough that mitigation can have a positive
outcome.

Hence on future naval ships, crew members will
need to concentrate on their assigned tasks or
specialties such as piloting the ship, telemetry,
and weapons command. In a casualty event,
crew members skilled in firefighting will be in
limited supply as compared to current CG-47 or
DDG-51 class ships. The crew will be
complimented by an advisory system that
accurately provides situational awareness and
decision aids about damage control operations.

The present effort supports a small business
innovation and research (SBIR) topic to develop
a shipboard computational titled Model to
Support Damage Control Assessment and
Decision-Making in Shipboard Environments.
Its goal is to reduce the amount of time required
to process damage control information and
estimate casualty spread in order to support
automated damage control systems in real time.
Currently, the Fire and Smoke Spread Simulator
(FSSIM) is the physics-based network model
used to pre-compute scenarios in the heuristic

2
fire and smoke federate of the damage decision
and assessment system (DDAS). FSSIM (Floyd
et al. 2004) presently lacks real-time
performance and sensor interaction. The next
generation of this predictive software,
FSSIMShip, will be developed to provide
shipboard sensor driven, faster than real time
simulations.

The software will be integrated with the ships
sensor and alarm networks. This will ground the
advisory system with the latest situational
updates. In its new incarnation the software will
also have similarities with the Sensor Driven
Fire Model, a fire fighter tool proposed by the
National Institute of Standards and Technology
(NIST) (Davis, Cleary, Donnelly, & Hellerman,
2003; Davis & Forney, 2001) but yet to be
realized in practice.

Speed and the ability to accurately model large,
complexly interconnected spaces equipped with
ventilation, sensors, and suppression systems are
the performance requirements for the model.
Processing requirements, however, should also
be kept to a minimum. While the processing
power planned for future naval ships is
considerable, there are many computational
needs to be supported for both damage control
functions and normal ship operations functions.
The prudent approach, and the one embraced in
this effort, is to minimize the computational
footprint of advisory systems such as
FSSIMShip.

FSSIMShips capability can be envisioned by a
weapons-induced damage scenario onboard a
ship. Immediately after a weapon hit there will
be a period of automatic reconfiguration of
electronic and hydraulic systems in an attempt to
route system capabilities around damaged
portions of the ship. These automated actions
will be largely predetermined based doctrine and
may or may not represent the optimal response.
Following this initial response, further damage
control (DC) activities may be relatively time
consuming. First, information is collected by
the damage detection and control systems. Then
the information is analyzed. Finally, potential
responses are determined and compared to
optimize the response. Any manned response
will involve the times to communicate the task,
complete dressing out, acquire equipment,
transit to the desired location, and gain access.
Any response involving the reconfiguration of
systems will require sufficient time to perform
the reconfiguration and to monitor the results.

At the point in time when either a manned
response is on scene and ready to commence or
a successful system reconfiguration has
occurred, the action that was taken needs to still
be valid. If the recommended action is negated
by further fire and smoke spread, then some
fraction of the time and effort expended on the
action will have been wasted. The ability to
look ahead far enough to avoid recommending
an unsuitable response will greatly improve the
overall effectiveness of the DC effort and
minimize the consumption of ship personnel and
equipment resources. The prediction from
FSSIMShip will complete rapidly enough to
allow the damage control assistant (DCA) to
formulate a response before the prognostic time
window expires. The goal of FSSIMShip is a 30
minute prediction horizon in under 20 seconds.

Based on current sensor data and crew reports,
the onboard software will locate actively
burning compartments. Those spaces that are
untenable would be identified so that response
crews would thus avoid these areas. For areas
where sensor data is lost/unavailable, the
software will fill in the information gap. In the
remaining spaces, the software will coordinate
with the functional sensors. All logged crew
reports would be imported by the model. The
continuously updating 30-min projection gives
the DCA a sense of not just the current situation
but also what the near term consequences of the
involved spaces will be.

FSSIMShip will not only predict fire and smoke
spread but will be able to adjust actions based on
effects of deployed automatic fire protection
systems. For example, on future naval ships fire
suppression systems consist primarily of water
mist systems. The quasi-equilibrium
evaporation model of (Back, Beyler, & Hanssen,
2000) is used to rapidly predict the anticipated
temperature drop following water mist
activation. The model uses the provided

3
compartment temperature, estimated fire size
(based on prediction or sensor data), and the
compartment ventilation based on heating,
ventilation, and air conditioning (HVAC) system
settings and portal status to predict the
effectiveness of the water mist system. Hence
the DCA will acquire a sense of what
compartments will need a manned response.

The goal planned for FSSIMship is to provide
the Navy with a real-time fire and smoke engine
to replace the pre-computated fire and smoke
database in support of a damage decision tool
termed Damage Decision & Assessment (DDA).

Overview of the FSSIMShip Network
Model

FSSIMShip is a network fire model written to
simulate the spread of fire and smoke in a naval
vessel. However, there is nothing in the model
to preclude its use for other types of
compartmented structures such as a building. In
FSSIMShip each compartment in a structure is
represented as a single node with surfaces (e.g.,
bulkheads, decks, and overheads) and vent
openings (e.g., doors, hatches, etc.) represented
as node connections.

FSSIMShip encompasses the following
capabilities:

- 1D flow model including friction losses and
temperature-dependent specific heat.
- 1D multiple-layer, temperature-dependent
heat transfer.
- N-surface, gray-gas radiation heat transfer,
including radiation streaming through
openings.
- Bidirectional flow through horizontal
(hatches) and vertical (doors) flow
connections.
- Combustion product species tracking.
- Oxygen and fuel limited combustion.
- Multiple fires and fire spread via
compartment-to-compartment heat transfer.
- HVAC systems including ducts, dampers,
chillers, and fans with forward and reverse
flow losses and multiple fan models.
- Fire detection via heat, smoke, and fire
detection
- Fire spread by compartment-specific
criteria.
- Fire suppression via sprinklers, water mist,
gaseous agents, aerosol agents, and foam.
- Fire spread prevention via boundary cooling.
- Binary control structures used to link
operation of equipment to sensors or times.
- Faster-than-real-time execution speed.
- Both coordinated and complimentary
interaction with shipboard sensors and/or
fire alarm system.

PHYSICS

The core physics of FSSIMShip is little changed
from that of FSSIM. These are briefly reviewed
in the ensuing subsections.

Compartment Flows

FSSIMShip solves lumped, time-dependent
conservation equations for mass, momentum and
energy. Energy and mass are conserved
explicitly, whereas momentum is conserved
implicitly. Energy and mass conservation use a
control-volume approach, where the control
volume is either a single compartment or a
ventilation system node. Momentum is
implicitly solved for at vent connections or in
ducts. The set of equations used is taken from
MELCOR (Gauntt et al. 2000) as implemented
in FSSIM (Floyd et al. 2004). This set of
equations is solved numerically using ordinary
differential equations (ODE) and differential-
algebraic equations (DAE) solvers.

Combustion

Upon activation, FSSIMShip will set up fires
according to inputs from sensors and from the
DCA. These fires will spread according to local
conditions. Ignition of additional fires is
determined at the beginning of each time step.
Each compartment can have a use-type
designation, which denotes a fuel loading and
fuel classification. Separate temperature
ignition criteria are specified for surfaces,
temperature of incoming vent flows, and

4
compartment temperature (Back et al. 2003).
Overhead surfaces have different ignition
temperatures from other surfaces.

Pyrolysis is based on a composite function that
includes
2
t growth, continuous burn according
to local conditions, and polynomial or
exponential decay. Growth is limited by
specification of a maximum pyrolysis rate. The
calculated pyrolysis rate can be reduced by
various mechanisms. If the fire is being
suppressed by an agent, the pyrolysis will be
reduced by a suppression factor. If the fire has
become oxygen limited, then the pyrolysis rate
is determined by a linear function of
temperature. The maximum pyrolysis rate after
oxygen-limiting conditions are reached is set at
the point where the fire became oxygen limited;
the compartment temperature for that point is
also stored. The oxygen-limited pyrolysis rate is
then calculated by the ratio of the current
temperature to the stored temperature not to
exceed the maximum rate. The actual heat
release rate is adjusted to use the calculated
amount of available oxygen.

Species are generated based on the yields for
each fuel being burned. These yields represent
the mass of combustion products formed for a
unit mass of fuel burned.

\Heat Transfer

Convection heat transfer is calculated at each
surface for each time step. Two sets of
convection heat transfer correlations are used.
The first is used for spaces where either no fire
exists or the volume of the space is small in
comparison to the anticipated fire size. The
second set of correlations applies a plume and
ceiling jet model to improve computation of heat
transfer to the overhead. If a surface has been
specified as transparent to radiation, no
convection heat transfer computation is
performed.

Radiation heat transfer is computed on a
compartment-by-compartment basis using the
beginning of time step surface temperatures,
compartment temperature, compartment gas
composition, and compartment heat release rate.
The heat transfer is calculated using a modified
gray-gas, n -surface net radiation method
(Forney 1991).

Conduction heat transfer is computed from a
one-dimensional (1D) heat equation discretized
with central differences (Strauss 1992) to arrive
at a set of ODEs for each wall. Each 1D
partition/bulkhead is divided into N -1 cells or
nodes with N boundaries. Boundary conditions
are obtained from the convection and radiation
calculations. Each wall can have multiple layers
of materials and temperature-dependent specific
heats and conductivities.

SENSITIVITY AND
UNCERTAINTY

The remaining sections detail new features in
FSSIMShip. Because the DCA is using the data
from FSSIMShip to address an ongoing
situation, it is desirable to give the DCA an
assessment of the accuracy of FSSIMShip
predictions. Hence automatic estimation of
sensitivity and uncertainty are part of every
FSSIMship prediction cycle. Two variables are
currently singled out for this treatment: the
temperature, T , and the spread distance, x .
Hence, if x is the uncertainty in the spread,
the prediction can be written as

. x x (1)




Figure 1. Some standard distributions associated
with measurements.

5
Formally, the squares of the uncertainties are the
variances of the predictions. Each datum has a
probability distribution. In Figure 1 the
uncertainty is shown as the standard deviation of
the symmetric distribution. Skew distributions
are possible as well. For the present analysis,
the top hat distribution is assumed. As Figure 1
shows, it is symmetric about its center and all
values within the standard deviation bounds are
equally probable.

If the uncertainty in a set of parameters is taken
as a given, a formal methodology can be
followed to determined how this is manifested in
variables calculated from this set (ASTM 1998).
The compartment temperature plus the spatially
discretized heat transfer equations in FSSIMShip
form a system of differential algebraic equations
(DAEs). The uncertainty is quantified by
solving the DAE plus sensitivity system for each
compartment and each computational cell. The
uncertainty in the boundary conditions appear as
extra algebraic terms. The sensitivity
coefficients are then used to quantify the
uncertainty in the temperature prediction in each
compartment. The solvers DASPAK 3.0,
CVODES, and IDAS (Hindmarsh 2000;
Hindmarsh 2005; Li & Petzold 1999; Li &
Petzold 2000; Petzold et al. 2006) have the
requisite numerical machinery (such as the
evaluation of the partial derivatives) built into
them to solve these types of sensitivity
problems.

This formal procedure works fine for the
temperature but not for the extent of the affected
areas because spread is determined by the
satisfaction of certain criteria. For example,
when a compartment exceeds an ignition
temperature for a certain amount of time it is
designated as containing a fire. For the extent of
the affected area, use is made of the
characteristic distance (Trelles & Pagni 1997),

( )
,
1
5 / 2
(
(

g p
Q
x
max
c


(2)

where
c
x is the characteristic length scale,
max
Q


is the highest heat release rate from all the fires
that are currently burning, is the ratio of the
specific heats,

p is the ambient pressure, and


g is the acceleration of gravity. The distance
uncertainty as a function of the heat release rate
is

( )
,
1
5
2

5 / 3
5 / 2
max
max
Q
Q
g p
x

(
(


(3)

or, in terms of relative uncertainty,

.
5
2
5
2
Q
Q
Q
x
x
x
max c

= = = (4)
Along the border of the involved compartments,
distance is measured from the origin,
0
x

, to the
edge, x

. The uncertainty in the spread is


therefore

.
0
x x x x

= (5)

The characteristic length in Eq. 2 formally
applies to an unconstrained plume. In a ship, the
plume and fire would be affected by many
factors such as passageway geometry and
ventilation. However, since the plume is
fundamental even in a compartment, it is a
rational starting point. (Hamins & McGrattan
2007) give the example of the flame height
based on the Heskestad expression. Even
though flame height is a distance, because of its
dependency on heat release rate and pool fire
diameter, the asymptotic relative uncertainty of
the flame height scales as Q L

2
1
= . This
is greater than the 2/5 coefficient in Eq. 4. In
general, the complications associated with
potentially multiple fires that have flames
impinging on the rocking overhead of a ship
indicate increased uncertainty. For these reasons
it is believed that the spread uncertainty in Eq. 5,
which was based on Eqs. 2 and 4, is a lower
bound.





6
INVERSE METHODS

A key part of the damage control software for
future naval ships is the integration of sensor
data and crew reports into modeling tools and
damage decision aids such as FSSIMShip. On
the most basic level, FSSIMShip accepts sensor
data as a means of model initialization. Another
opportunity afforded by the sensor and crew
inputs is the extrapolation of these reports to
determine the heat release rate (HRR) of a fire.
The meteorological, geophysical, and
engineering disciplines have developed a variety
of inverse methods in order to obtain model
inputs from sparse readings (Backus 1971;
Menke 1989). The robustness of inverse
methods was improved with the advent of
regularization techniques (Tikhanov & Arsenin
1977).

In the realm of heat transfer, previous studies
(zisik & Orlande 2000) have shown how to
arrive at heat fluxes and heat sources from
temperature measurements made in laboratory
settings. These efforts benefited from the ability
to put as many sensors as would be feasibly
desired within the laboratory sample. The same
can be said for the early sensor-driven fire
model (Davis & Forney 2001) and other fire
protection engineering investigations of
reconstructing fire physics from sensor
measurements (Kramer et al 2003; Padakannaya,
Richards, & Plumb 1994; Richards, Munk, &
Plumb 1997; Richards et al. 1997). This may
not be an option with future naval ships.

The algorithms developed for FSSIMShip use
nonlinear parameter estimation to analyze sensor
inputs from local and nearby compartments in
order to estimate the heat release rate (HRR).
Although inverse methods (Beck & Arnold
1977; Tikhanov & Arsenin 1977) have been
widely explored, they tend to be computationally
expensive. This computational load needs to be
minimized for FSSIMShip. Since FSSIMShip
will be constantly updating itself, previous
sensor results will also be utilized for this HRR
calculation. Inverse methods will only be
possible for compartments with working sensors
that provide numerical readings. They can only
be used from the start of the incident up to the
present time. Once FSSIMShip goes beyond the
present time inverse methods are not possible
due to the lack of sensor input. The next two
subsections detail inverse methods that can be
used to improve the prediction of the heat
release rate associated with a spreading set of
fires.

To recap, in FSSIMShip measurements are used
to set conditions in compartments from the time
of the incident to the present. This sensor data is
utilized in the inverse methods to improve
estimates for key parameters. During this
incident to present period, simulations are still
used to determine conditions in compartments
which (1) do not have sensors, (2) have damaged
sensors, (3) are not reporting sensor readings for
whatever reason, or (4) have sensors that only
provide an alarm of some type, i.e., no data.
From the present to the future FSSIMShip only
simulates conditions with, wherever possible,
the sensors providing initial conditions and
improved estimates for key model parameters.
Compartments lacking this information use the
last set of calculated conditions.

Global Inverse Methods

The classic inverse method is global in scope.
This means that it takes data from throughout
the platform in order to arrive at its estimates.
The general approach is now outlined. Assume
that the heat release rate can be modeled as

( ) t C q
P
N
j
j j f
=
=
1
(6)

where
j
are the parameters and ( ) t C
j
are
functions (typically polynomials of some type),
both with
P
N elements. For example, fire
growth is typically described as

2
t q
f
o = (7)

Here
P
N = 1,
1
= o , and ( ) t C
1
=
2
t . The
goal is to use sensor data to arrive at better

7
estimates of
1
. Even if the simple model of
Eq. 7 is adopted, having a better handle on fire
growth can significantly improve model
predictions. A quadratic objective function, S ,
is defined to indicate the level of agreement
between measured and predicted temperatures.
In order to minimize the difference, the gradient
of S with respect to the parameters is taken and
the result is set to zero. This is analogous to
taking the derivative of a scalar function to
determine its extrema.

The resulting equations form a system of
nonlinear equations. In order to solve it, the
gradient equations are fed to a nonlinear
equations solver such as KINSOL (Collier at al.
2006). KINSOL in turn calls the appropriate
subroutines in FSSIMShip to obtain the
sensitivity Jacobian. KINSOL will work to
satisfy an error tolerance (if possible). These
basic types of algorithms tend to suffer from
difficulties with numerical stability. As an
example of why this is the case, remember the
relationship between the flame height and the
heat release rate uncertainties given above,
Q L

5 . 0 = . For the inverse problem the
heat release rate would be determined from the
flame height. For this case then L Q 2 =

.
This is typical of inverse problems:
measurement uncertainties become amplified.
For numerical methods, the truncation of real
numbers to the n -byte rational numbers that are
available on digital computers can introduce
enough error into the problem to result in an
unstable algorithm. Regularization techniques
add terms to the definition of the objective
function in order to improve the robustness of
the overall inverse method.

Local Inverse Methods

For a ship section modeled as a lumped mass of
compartments, techniques such as deconvolution
(Casey & Walnut 1994) can be used to obtain
the physically meaningful variables from all of
the sensor readings obtained at different
sampling rates. This approach has the advantage
that it is local in scope. It makes use of sensor
data at different sampling rates. The method
works formally for linear problems so the
current set of equations would have to be
appropriately linearized. Much of this work has
already been done in FSSIM, though. The
challenge for FSSIMShip is to arrive at suitable
convolution integrals for the bulk of the
equations that describe compartment
temperature. The applicability of this method as
a substitute for the other inverse methods, which
are global in scope, will then be assessed.

Start
Register Event
Start Time
t
S
No
All
Clear?
Yes
Monitor
Monitor
Sensors
DCA Inputs
Register
Event?
Yes
No
Forecast
Forecast
Report
Record Current
Time t
C
Set Cycle
Final Time
t
F
= t
C
+ 30 min
Review
Predict
Output to
Consoles

Figure 2. Flowchart showing the basic software
architecture of FSSIMShip.

OVERALL SOLUTION SCHEME

FSSIMShip is configured to continuously march
out to a specified time interval beyond the
present. The main loop calls the forcast loop.
(Refer to Figure 2.) This loop has two sections.
The first part is for events from the beginning of
the emergency to the present time. The
information processed in this loop will consist of
new sensor data, new calculations, and

8
previously stored readings and calculations. The
second part of the time marching loop will go
from the present into the future. The state
determined in the first part of the time marching
loop will be used to initialize the deterministic
models. The solution of the governing equations
is realized by the use of DAE solvers. An outer
loop is set up to march through time. (See
Figure 3.) An inner loop is then initiated for
node states at the current time. The solvers are
called for the compartment conditions. They are
then called to solve the heat transfer equations.
If the exit code indicates a successful return then
this inner loop runs for only one count.
Otherwise measures are taken according to the
exit status flag.

Review
Gather
Inputs from
All Sensors
Determine
Compartment
Fire Statuses
Inverse
Methods
Set HRR
Parameters
Predict
Predict
Set
t = t
C
+ At
Determine
Reduced
Computational
Domain
Solve in
Reduced
Computational
Domain
t > t
F
?
Report
Yes
No

Figure 3. Flowchart showing the sections of the
FSSIMShip software architecture that interact
with sensors and predict the impact of current
fires.

REAL-TIME COMPUTATION

Two techniques will be used to improve run
time performance. Currently FSSIMShip
simulates the entire ship. In general this will not
be necessary for supporting damage control.
Regions of the ship remote from the weapon hit
should not see a significant fire and smoke
spread impact. The time for fire and smoke to
spread to these regions will provide time for
damage control activities to halt the spread of
secondary damage to these regions.
Autonomous suppression systems in remote
regions should be functional (or regain
functionality prior to spread). The candidate
approaches include strict limiting of the
geometric extent being computed, a multi-region
approach where computational fidelity varies
with remoteness of the region (e.g., a hybrid
heuristics-physics model), and multi-block
approach where remote spaces and their
associated ventilation are combined.

Set Openings /
Closings
Identify New /
Extinguished
Fires
Regenerate
Computational
Domain
Update Solver
RHS Function
Initialize
Solver
Call Solver
Any Error
Flags?
Continue
Corrective
Measures
Predict
Set
t = t
C
+ At
Determine
Reduced
Computational
Domain
Solve in
Reduced
Computational
Domain
t > t
F
?
Report
Yes
No
No
Yes

Figure 4. Flowchart showing the section of the
FSSIMShip code that solves the governing
equations. The dashed rectangles on the right
diagram are expansions of the corresponding left
diagram actions.

The bulk of the computation time within FSSIM
lies in matrix-based solvers. Incorporating more
efficient solvers and code structures (Drummond
& Marques 2005) in FSSIMShip will result in
faster turn-around. FSSIM is currently

9
structured as a single processor software tool.
Many portions of the code are easily amenable
to high performance (multiple-processor)
computing (Pacheco 1997). FSSIMShip uses
the Message Passing Interface (MPI) (Pacheco
1997; Gropp, Lusk, & Skjellum 1999; Snir
1998). This allows distributing the calculation
on a set of computers/processors in the data
center. MPI-capable solvers, such as those
available with Argonne National Laboratorys
Portable, Extensible Toolkit for Scientific
Computation (PETSc) (Balay et al. 1997; Balay
et al. 2004), will be employed. The goal is to
perform as many implicit solves in parallel as
possible. The appropriate number of processors
needed to achieve faster than real time goals will
be explored. Refer to Figure 4 for the
envisioned software architecture.

SENSORS AND DETECTORS

For state-of-the-art ships of the future, advisory
software developers will have input on the
number and types of sensors to be installed.
This will allow FSSIMShip to work
cooperatively with sensors. However, older
generation naval vessels are typically equipped
with a fire alarm system which has a limited set
of sensors. There is no guarantee that the
readings from the sensors will be available at a
ship console. More likely than not the fire alarm
system provides just that: an alarm and an
annunciated location and no further information.
For these cases, FSSIMShip will compliment
sensors, filling in the gap afforded by the alarm
interface with simulation data on the conditions
in affected areas. Refer to Figure 4 for the
software architecture that shows how sensor data
will be incorporated into FSSIMShip.

Once sensor algorithms are defined, there
remain significant implementation issues to
optimize their performance with FSSIMShip.
Whenever possible FSSIMShip will be
interfaced with the sensor network. Typical
interfaces are TCP/IP and MODBUS. Since
FSSIMShip needs to address the problem of
incomplete sensor readings, some damage
assessment (Fritzen, Jennewein, & Kiefer 1998)
is necessary within FSSIMShip. In addition,
there are challenges to be addressed in
identifying and implementing event type data
into the simulation. These may include events
like creation of an opening via damage or crew
action, as well as fire mitigation via crew action.
To aid in limiting the range of events that needs
to be considered, it is envisioned that high level
knowledge will be provided to the model from
the DCA or other systems. This is expected to
include clarifications such as fire due to hostile
weapons (for known types) and other
information about the incident that is developed
via the various systems and crew observations.

DATABASE

Because FSSIMShip is an autonomous system,
there will be no need to create an input file as
was the case with FSSIM. However,
FSSIMShip will need access to a database or a
set of databases that contain crucial information
such as ship information and physical constants.
It remains to be determined whether this
information will be contained within the
compiled program or externally. It is currently
expected to be a combination thereof: material
properties can be contained within FSSIMShip
but ship specifics will be contained in data files.
The following is a list of important information
to be contained in the databases.

- Connections based on the ship layout;
- Compartment volumes;
- Wall areas;
- Resistance coefficients for standard objects
such as doors, hatches, ducts, etc;
- Thermal properties for common materials
such as steel and insulation;
- Suppression systems data such as activation
temperatures and response time indices; and
- Detector data such as threshold values and
response lags;
- Sensor locations.

Smart product model development and
implementation could be leveraged to provide
this ship specific data.




10
CONCLUSIONS

The basic framework for shipboard FSS
established FSSIM as an incident modeling and
investigation tool. FSSIM is currently being
transitioned to the real-time, onboard damage
control tool known as FSSIMShip. To
accomplish this, the following changes are in the
process of being made to FSSIM:

- Parallel computation;
- Estimation of uncertainty and sensitivity;
- Creation of ship- and physics-specific
databases;
- Integration with shipboard sensors and/or
fire alarm system; and
- Use of sensor data to improve key model
inputs such as the heat release rate.

Approaches to accomplish this transition have
been detailed above. They will be developed
and refined as the project moves forward.

REFERENCES

ASTM, Standard Guide for Evaluating the
Predictive Capability of Deterministic Fire
Models, E 1355, New York: American National
Standards Institute, 1998.

Back, G., Beyler, C., and Hanssen, R., A
Quasi-Steady-State Model for Predicting Fire
Suppression in Spaces Protected by Water Mist
Systems, Fire Safety Journal, 35, pp 327362,
2000.

Back, G., E. Mack, M. Peatross, J. Scheffey, D.
White, F. Williams, J. Farley, and D. Satterfield,
A Methodology for Predicting Fire and Smoke
Spread Following a Weapon Hit, NRL Ltr Rpt
Ser 6180/0014, Naval Research Laboratory,
Washington, DC, 2003.

Backus, G.E., Inference from Inadequate and
Inaccurate Data, in Reid, W. H., (ed.),
Mathematical Problems in the Geophysical
Sciences, Providence, RI, American
Mathematical Society, pp. 1106, 1971.

Balay, S., Gropp, W.D., McInnes, L.C., &
Smith, B.F., Efficient Management of
Parallelism in Object Oriented Numerical
Software Libraries, in Arge, E., Bruaset, A.M.
and Langtangen, H.P. (eds.), Modern Software
Tools in Scientific Computing, Boston, MA:
Birkhuser, pp. 163202, 1997.

Balay, S., Buschelman, K., Eijkhout, V., Gropp,
W.D., Kaushik, D., Knepley, M.G., McInnes,
L.C., Smith, B.F., and Zhang, H., PETSc Users
Manual, Argonne National Laboratory,
ANL-95/11 - Revision 2.2.0, 2004.

Beck, J.V. and Arnold, K.J., Parameter
Estimation in Engineering and Science, New
York: Wiley, 1977.

Casey, S., and Walnut, D., Systems of
Convolution Equations, Deconvolution,
Shannon Sampling, and the Wavelet and Gabor
Transforms, SIAM Review, 36:4, pp. 537577,
1994.

Collier, A.M., Hindmarsh, A.C., Serban, R., &
Woodward, C.S., User Documentation for
KINSOL v2.5.0, UCRL-SM-208116,
Livermore, CA: Lawrence Livermore National
Laboratory, November 6, 2006.
Davis, W. D., and Forney, G. P., A Sensor-
Driven Fire Model, Version 1.1, National
Institute of Standards and Technology,
Gaithersburg, MD, NISTIR 6705; January 2001.

Drummond, L.A. and Marques, O.A., An
Overview of the Advanced CompuTational
Software (ACTS) Collection, ACM
Transactions on Mathematical Software, 31:3,
pp. 282301, 2005.

Floyd, J., S. Hunt, F. Williams, and P. Tatem,
Fire + Smoke Simulator Version 1 (FSSIM)
Theory Manual, NRL, Washington, DC, 2004.

Forney, G., Computing Radiative Heat Transfer
Occurring in a Zone Fire Model, NISTIR 4709,
National Institute of Standards and Technology,
Gaithersburg, MD, 1991.

Fritzen, C.-P., Jennewein, D., & Kiefer, T.,
"Damage Detection Based on Model Updating

11
Methods," Mechanical Systems and Signal
Processing, 12:1, January 1998, pp. 163186.

Gauntt, R., R. Cole, C. Erickson, R. Gido, R.
Gasser, S. Rodriguez, M. Young, MELCOR
Computer Code Manuals: Reference Manuals
Version 1.8.5, NUREG/CR-6119, Volume 2,
Rev. 2, U.S. Nuclear Regulatory Commission,
Washington, DC, 2000.

Gropp, W., Lusk, E., and Skjellum, A., Using
MPI: Portable Parallel Programming with the
Message-Passing Interface, 2nd Edition,
Cambridge, MA, MIT Press, 1999.

Hamins, A., and McGrattan, K. B., Verification
and Validation of Selected Fire Models for
Nuclear Power Plant Applications, Volume 2:
Experimental Uncertainty, U.S. Nuclear
Regulatory Commission, Office of Nuclear
Regulatory Research (RES), Rockville, MD, and
Electric Power Research Institute (EPRI), Palo
Alto, CA. NUREG-1824 and EPRI 1011999,
May 2007.

Hindmarsh, A.C., The PVODE and IDA
Algorithms, LLNL Technical Report UCRL-
ID-141558, December 2000.

Hindmarsh, A.C., Brown, P.N., Grant, K.E.,
Lee, S.L., Serban, R., Shumaker, D.E., and
Woodward, C.S., SUNDIALS: Suite of
Nonlinear and Differential/Algebraic Equation
Solvers, ACM Transactions on Mathematical
Software, 31:3, pp. 363396, 2005.

Li, S., and Petzold, L., Design of New DASPK
for Sensitivity Analysis, Technical Report, Dept.
of Computer Science, Technical Report UCSB,
University of California at Santa Barbara, 1999.

Li, S., and Petzold, L., Software and
Algorithms for Sensitivity Analysis of Large-
Scale Differential-Algebraic Systems, Journal
of Computational and Applied Mathematics,
125, pp. 131145, 2000.

Liley, P. E., Reid, R. C., and Buck, E. (1984),
Physical and Chemical Data, in R.H. Perry,
D.W. Green, & J.O. Maloney (eds.), Perrys
Chemical Engineers Handbook, 6
th
ed., New
York, McGraw Hill, pp. 3-13-291.

Menke, W., Geophysical Data Analysis:
Discrete Inverse Theory, San Diego, CA,
Academic Press, 1989.

zisik, M.N., & Orlande, H.R.B., Inverse Heat
Transfer, New York: Taylor & Francis, 2000.

Pacheco, P., Parallel Programming with MPI,
San Francisco, CA, Morgan Kaufmann
Publishers, Inc., 1997.

Padakannaya, K., Richards, R.F., and Plumb,
O.A., Inverse Radiation Solution for Fire
Detection, NISTIR 5499, September 1994,
Gaithersburg, MD, National Institute of
Standards and Technology, 1994.

Petzold, L., Li, S., Cao, Y., and Serban, R.,
Sensitivity Analysis of Differential-Algebraic
Equations and Partial Differential Equations, to
appear, Computers and Chemical Engineering,
2006.

Richards, R.F., Munk, B.N., and Plumb, O.A.,
Fire Detection, Location and Heat Release Rate
Through Inverse Problem Solution. Part 1.
Theory, Fire Safety Journal, 28:4, pp. 323350,
1997.

Richards, R.F., Ribail, R.T., Bakkom, A.W., and
Plumb, O.A., Fire Detection, Location and Heat
Release Rate Through Inverse Problem Solution.
Part 2, Experiment, Fire Safety Journal, 28:4,
pp. 351378, 1997.

Snir, M., et al., MPI: The Complete Reference,
Cambridge, MA, MIT Press, 1998.

Strauss, W., Partial Differential Equations: An
Introduction, John Wiley & Sons, New York,
NY, 1992.

Tikhanov, A.N., and Arsenin, V.Y., Solution of
Ill-Posed Problems, Washington, DC, Winston
& Sons, 1977.

Trelles, J., and Pagni, P.J., Fire-Induced Winds
in the 20 October 1991 Oakland Hills Fire, Fire

12
Safety Science: Proceedings of the Fifth
International Symposium, March 37, 1997,
Melbourne, Victoria, Australia, Elsevier,
London, 1997, pp. 911922.


AKNOWLEDGMENTS

Scott Hill of HAI assisted with the collection of
sensor data and with the preparation of the
software design description.



Javier Trelles, Ph.D., is a senior engineer at
Hughes Associates. He has written models for a
variety of situations including blood
carboxyhemoglobin, thermal decomposition of
materials, spreading of fuel spills, and the
dispersion of smoke under the influence of
various weather conditions. He has conducted
numerous simulation studies. Dr. Trelles is the
developer of many CFD animation and
visualization programs.

Craig L. Beyler, Ph.D., is the technical director
at Hughes Associates. He has extensive
expertise in fire dynamics/fire modeling and has
been the primary editor for the Fire Dynamics
Section of the SFPE Handbook of Fire
Protection Engineering. He has been involved
in model development and use for analysis of
U.S. Navy ships for over 15 years. He is
currently the chairman of the International
Association for Fire Safety Science and the
winner of the Guise Medal, SFPEs highest
technical award.

Jason E. Floyd, Ph.D., is a senior engineer at
Hughes Associates. Dr. Floyds R&D/design
experience includes simulation tools
development, numerical modeling, and design of
large-scale experiments and experimental
apparatus. He is the lead developer of FSSIM.
Dr. Floyd has performed CFD simulations of
various types and is currently a member of the
FDS development team.

Joseph L. Scheffey, P.E., is the Director of Fire
Protection RDT&E at Hughes Associates, Inc.
Baltimore, MD. He received a BS in fire
protection engineering at the University of
Maryland and is a Fellow in the Society of Fire
Protection Engineers. He has over 25 years
experience in managing and directing full-scale
fire tests. He has directed numerous manned
firefighting and damage control tests for the
U.S. Navy.

Kim W. Yee, is the USN DDG 1000 Damage
Control Automation Lead and the author of this
SBIR topic. He has over 20 years of experience
working with the Navys Surface Combatants
HM&E Controls Systems. In his role as
DDG1000 DC Automation Controls Lead, he is
working with industry to develop a Damage
Decision & Assessment (DDA) capability that
addresses Hull stability, Hull stress and fire,
smoke, flooding, Chemicial/Biological spread.

You might also like