Explosion Loading on Equipment from CFD Simulations

Olav R. Hansen1, Malte T. Kjellander1, and Jan A.Pappas2

1
Lloyd's Register Consulting, Bergen, Norway, Email: olav.hansen@lr.org
2
Lloyd's Register Consulting, Sandvika, Norway

Abstract:
Explosion studies using computational fluid dynamics (CFD) are performed on daily basis
among safety consultants all over the world. The purpose of the explosion studies is usually to
give guidance on required design strength of equipment, piping, blast walls or buildings. One
key element is to translate the results from an explosion simulation, into actual forces on
equipment. Major weaknesses exist in the current approaches for estimation of loads on small
and medium sized equipment. Hansen et al. (2014) demonstrated how loads on rectangular
equipment could be estimated by combining free field form drag and pressure gradient. In the
current work it is discussed how best to estimate loads onto other types of equipment including
cylindrical pipes, pipe bundles and other shapes, and some comparisons and validation against
large scale experiments are also included. The main findings of this work are the more accurate
guidance on how best to extract explosion loads from simulations, including discussons on
appropriate drag coefficients.

1 Introduction
Explosion studies are performed to give guidance on required design strength or to confirm
actual design for a facility or nearby structures. For offshore oil and gas installations a common
approach (NORSOK Z-013, 2010, and ISO 19901-3, 2010) is to demonstrate that the non-
tolerable explosion that will give uncontrolled escalation/damage has a return frequency lower
than e.g. 10-4/year, and when possible robustness against even less frequent high consequence
scenarios should be demonstrated. Another common approach is to design to withstand a worst-
case credible explosion (API RP-752, 2009). For most studies on offshore petrochemical
installations, and also frequently for onshore facilities, Computational Fluid Dynamics (CFD)
simulations with FLACS are performed to predict the nature of the explosion loads of concern
(Hansen et al., 1999 and Hansen et al., 2013).

In a transient flow field during an explosion pressure differences will build up around objects,
and by integrating the pressure over the surface of an object, the explosion loading (forces onto
the object) can be obtained. For objects properly resolved on the simulation grid in a CFD-
simulation, like blast walls, decks and large objects with a minimum of 2-3 grid cells across,
explosion loads can be well estimated from differential pressures. Typical grid cell size for
explosion studies at a petrochemical facility is 0.5m to 1.0m. For smaller objects estimates of
differential pressures will be less accurate, and may not be possible to extract if diameter is less
than 1.0-1.5 grid cells. For such objects the load may be estimated from a general drag formula
(Sand, 1999):

(1)

The first term is the form drag, the second term is primarily accelerated mass, third term is forces
due to density changes during combustion, fourth term is from pressure waves in the flow field
and the last term is a hydro-elastic term potentially increasing the drag force due to oscillations.

Load estimates for pipes and smaller equipment are usually only considering the form drag, with
increasing size of object the second (accelerated mass) and fourth terms (pressure waves) of
equation (1) become more important.

The guidance from (FABIG TN-08) on how to estimate loads on objects based on CFD-
explosion calculations is as follows:

Object size: Proposed method to estimate loads:

Diameter > 2m Direct load measurement (DLM) to report pressure on surfaces

(PDF = pressure distribution factor used if maximum local P is used):

F= P  A  PDF PDF = 1.0 (box) or 2/ =0.64 (cylinder)

Diameter < 0.3m Use form drag formula with conservative drag coefficients from (Baker,
1983)

F=  v2  Cd  A Cd = 2.0 (box) or 1.2 (cylinder)

0.3m < D < 2.0m Form drag formula for objects up to D~0.5-1.0m, no suggestion for 1.0-
2.0m.

The Lloyd’s Register Guidance Notes for the Calculation of Probabilistic Explosion Loads
(Lloyd’s Register, 2015) suggest 3 alternative methods for load estimation for objects 0.3m to
2.0m, these are:

a) local grid refinement and use of DLM

b) estimation methods including load contributions 2 and 4 of the net reaction force (1)
c) interpolation between loads on objects resolved on the grid (DLM) and drag load

Due to the Cartesian grid of FLACS a grid refinement around a target will in most cases have
implication on the grid size/shape and the time step settings, which can change the predicted
explosion loads significantly. Expert users can limit the impact of local grid refinement by
instructing the time stepping routines of FLACS to ignore local grid refinement. Still, this
method should be used with care and primarily for one or a few simulations only.

(Hansen et al., 2014) described estimation methods including load contributions from accelerated
mass and pressure waves and demonstrated the ability to predict directional, transient loads onto
various box shaped objects from weak and strong explosions with good precision using the
formula:

Force(t) / Area = Cd x Drag (t) + PDF x [Ps(t) - Ps(t- t)] (2)

With Drag = ½ u2 and Ps being the static (side-on) pressure in the flowfield near the object. t
is the estimated time for the pressure waves to equalize around the object, which is estimated
based on the size and shape of the object and the fluid temperature (Csound ~ T-½ , i.e. t is almost
3 times higher prior to flame arrival than after). PDF is a pressure distribution factor (PDF=1.0
for box shapes, PDF=2/ =0.64 for cylinders). In Figure 1 this method is illustrated for a 1m box
shaped target 43m outside an explosion with the drag (blue dotted curve) and side-on pressure
(red dotted curve) used to estimate the force on the object (black solid curve is estimate, black
dotted curve is actual load from DLM). This is illustrating a frequent challenge in explosion
studies, where reported drag loads and pressure loads are to be used to estimate loads on e.g. life-
boats, crane supports and other critical equipment outside the explosions. From Figure 1 it
should be obvious that neither the drag nor the pressure loads are representative for the actual
load as both the magnitudes and durations are very wrong. Using relation (2), however, the
transient load curve, including magnitude and duration, is estimated with good precision.

The 3rd method, interpolation, has been used frequently in the past. One challenge with the
method is that while there may be reported drag values all over the domain, there may only be a
few larger objects which can be used to report DLMs, and it is not obvious that an estimated load
on a 1.5m diameter object will be particularly accurate if this is based on interpolation between a
local drag value and a direct load measurement on the nearest 4m sized object that could be 20m
away. As a consequence in studies performed by Lloyd’s Register Consulting in recent years,
reported pressures on local wall and deck panels (3m x 3m) spread all over the modules, have
been used for the interpolation as an estimate for local DLM across medium sized objects.
Figure 1 Improved method for load estimation from explosions, the pressure load
(black curves - dotted=actual, solid=estimated) on an object equals the
dynamic pressure x drag coefficient (blue curve x Cd) plus P (pressure
change in time t, red curve)

The character of explosion loading will vary significantly with explosion strength, and whether
the target is inside or near the explosion, or in the far-field. Inside the explosion the duration of
the loads will often be rather long, often 100-200 ms, similar to the duration of the local pressure
peak, and a triangular approximation of the load with Tduration = 2 x Trisetime can often be
representative. Some distance outside the explosion the loads may be dominated by pressure
waves, usually shockwaves from explosions ~0.5 bar and higher, and the duration of the load
onto a 1-2m sized object tends to be signficantly shorter, typically around 5-15 ms while the
blast wave itself may have a duration of 20-40 ms.

Chapter 2 will highlight important elements of the modelling that must be right before proper
explosion loads can be predicted. Chapter 3 will go through how best to extract loads from
explosion simulations for different types of equipment. Chapter 4 will discuss proper drag
coefficients, and include a validation example from BFETS full scale tests, while Chapter 5 and
Chapter 6 will discuss elements around the application of explosion loads in design.
2 Accurate Simulations Required for Accurate Load Estimates

Before accurate estimates of explosion loads can be extracted it is important that the competent
modeller uses a properly validated CFD model, following the best practice user guidelines,
and models relevant and representative scenarios.

A competent modeller should know the CFD tool very well, be aware of limitations and validity,
and follow relevant user guidelines. It is also important that the modeller understands well the
phenomena being modelled, as well as the study philosophy/approach of the explosion study.

For gas explosion modelling FLACS is with little doubt the best validated and globally most
used CFD model towards the petrochemical industry. The FLACS development from 1982 until
2000 at CMI/CMR was primarily done with support from major oil and gas industry through
joint industry projects. The development teams were thereafter transferred to GexCon which has
continued the development with license maintenance payments from commercial users being the
main funding. Validation against experiments has been a key activity in the FLACS
development, see e.g. (Hansen et al. 1999), and best practice user guidelines have been
established based on this.

Best practice user guidelines for FLACS include several different topics, and much of this can be
found in the user manual (GexCon, 2015), but useful information is to be found also in e.g.
(NORSOK Z-013, 2010) and (Lloyd’s Register, 2015). The most important topics for explosion
loading may be:
 Accurate 3D geometry model is required, with proper confinement (walls, decks, vent
areas) and a representative congestion level, i.e. as much as possible details of piping and
equipment included, and with fine piping and equipment added as anticipated congestion
(AC) to obtain expected reference levels of congestion (pipelength/volume and
area/volume). With the FLACS interfaces selected geometry elements from other as-built
geometry models, can be efficiently copied and manipulated to develop realistic AC.
 Grid guidelines must be followed strictly in particular the requirements for sufficient
resolution with cubical grid cells of gas cloud/congested region (10-13 grid cells in
shortest direction), and also proper grid resolution towards far-field targets (2-4 times
coarser grid cells may be acceptable).
 Time resolution must be as specified in the FLACS guidelines during the flame
acceleration and pressure generation (CFLC=5 and CFLV=0.5). For strong explosions
with source pressure higher than 0.5 barg, shock waves can be expected in the far-field.
In order to avoid numerical smearing and underprediction of loads, Hansen and Johnson
(2015) concluded that a finer time step must be used after the blast waves have separated
from the flame front (CFLC=0.1). For weaker explosions standard time step settings can
be kept, but time step control key “KEEP_LOW” must be specified to avoid increased
time steps and blast smearing.
 Scenario parameters must also be defined accurately, including gas composition. If a Q9
equivalent stoichiometric volume is defined the gas concentration should be at slightly
fuel-rich concentrations (ER~1.05). For far-field loading estimates it is essential to use
the non-reflecting boundary condition “PLANE_WAVE”. In most cases there will be
 some air movement and turbulence in the gas cloud, and some pre-ignition turbulence
should be defined (Hansen et al. 2013, Lloyd’s Register, 2015, NORSOK Z-013, 2010).
 The loading targets (walls, decks, equipment and other) should be properly instrumented
with pressure panels or drag/pressure sensors appropriately placed, see discussion in
following chapters.

Finally, when representative loads shall be predicted it is important that relevant scenarios are
simulated and evaluated. If a worst-case or realistic worst-case approach is used, it is essential
that the selected scenarios really do represent the worst events that may be expected, and as the
distribution of explosion loads may be directional for a scenario, the worst-case scenarios for one
target may be different from the worst-case scenario for another target. If a probabilistic
assessment is performed, it is important that the distribution of scenarios is representative, with a
proper distribution of cloud sizes, locations and ignition locations and frequencies.

FLACS is a deflagration model, but has functionality to predict if deflagration-to-detonation may

take place, and with certain settings it may also be possible to predict the detonation
consequences (Hansen and Johnson, 2015). Inside the exploding cloud, the consequences given
DDT will usually be far too severe to design against, for far-field pressures outside the gas cloud,
however, it may be of interest to predict the loads.

3 How to extract loading on different types of targets

In the following it will be discussed how loads can be extracted for various types of objects. The
shape and duration of the overpressure pulses onto a target will depend strongly on where the
target is located relative to the exploding gas cloud, and of course the character/strength of the
explosion.

Shape and duration of loads inside the flame region

In a gas explosion (deflagration) the flame will normally start slowly, with initial flame speeds of
the order 1 to 10 m/s. Depending on turbulence generation ahead of the flame, confinement,
shape and reactivity of the flammable cloud, the flames may accelerate and flame speeds for
explosions of relevance to structural strength will generally end up in the range 100-1000 m/s.
During this acceleration process the overpressure will be increasing to a maximum, and
thereafter fall to zero and below. The typical pressure curve shape inside the flame region can
often be represented with an idealized triangular pulse with a rise time of 50-100 ms, and often a
duration around twice the rise time, i.e. 100-200 ms. The underpressure following the initial
pressure pulse may typically be around 20% of the maximum overpressure, and usually within
the range 10-30% for typical explosion scenarios with overpressures 0.25 – 2.0 bar, with a
duration comparable to the positive pressure pulse duration. Within an area there may be
signficant variations in maximum overpressures, while the underpressure from the surge after the
explosion may be more evenly distributed, which means that locations with a relatively low
overpressure may see an underpressure as high, or even higher than the overpressure. The
durations of all peaks, positive and negative, will generally be reduced when the strength of the
explosion increases.
The triangular pulse shape assumption may often be representative for pressure and drag loads
inside the flame zone. From this simplified shape assumption there may be lots of deviations
depending on the location the pressure is reported, in particular for increasing complexity in the
gas cloud shape and the confining wall structures, there may be reflections and oscillations
occurring, giving double and even triple positive peaks in the overpressure, which may be more
questionable to represent with the simplified triangular profile.

For very strong explosion there is a possibility of DDT, and when this happens there will be a
strong discontinuity (shock) in the pressure waves, with an immediate jump to 10-30 barg, and if
there are significant unburnt pockets of gas remaining, these will likely detonate within the next
few milliseconds. The load pattern will be dominated by extreme loads with zero rise time of
pressure, and due to the very intense combustion in the flame front, the main drag loads will act
backwards (and possibly sideways/upwards out of the flame) behind the flame.

With increasing confinement and gas clouds filling a significant fraction of the area, pulse
durations can increase, and for the extreme with a strong, confined room or vessel filled with
gas, it may be heat losses to walls that reduce the overpressure after the explosion and the
duration of pressure may be of several seconds.

Shape and duration of loads into the far-field

Outside a surge region around the flame zone the explosion loads will gradually get the character
of pressure waves. For explosions with source overpressures of the order 0.5 bar or higher, the
pressure waves will tend gradually to develop into shockwaves with zero rise time, and with a
typical duration around half compared to inside the flame zone, slightly increasing with distance.
Underpressures may have comparable duration, but strength is often of the order 10 to 30% of
the peak overpressures. For weaker source explosions the rise time can be long and the
underpressure sometimes of comparable strength and duration as the positive pulse.

Within a large offshore module with partial confinement (e.g. deck, ceiling and one or more blast
walls) in which the flame zone may be only a fraction of the module, the loads some distance
away from the flame zone may have a character between what has been characterized as far-field
and near-field loads, as the confining structures may still ensure that the flame surge (drag loads)
is influencing the load pattern.

In Figure 2 some examples of typical flame zone and far-field pressures are shown to illustrate
variations in pressure duration, overpressure distribution and shape and underpressure. All tests
are performed at DNV GL test site Spadeadam. Some description of experiments and previous
CFD modelling work can be found in Hansen et al. (2013) and Hansen and Johnson (2014).

The first test BFETS-7 is the end ignited explosion test with 1600m3 stoichiometric natural gas
mixture (100% of tunnel shaped module), with maximum overpressures centrally in the module
around 2 barg and duration more than 100 ms, and above 4 barg with multiple peaks and 50-80
ms duration where the flames exit the module. Blast pressures shown 30m and 50m diagonally
outside the end of the module have the character of shock waves with a duration around 25-
30ms.
The second example is from Phase 3B project and is the first of three 2600m3 stoichiometric gas
clouds base case reference tests with ignition near center of gas cloud and 12m x 8m vent
openings on three sides (North, East and West). Maximum explosion pressures above 4 barg
were recorded near the East vent opening (maximum flame travel distance), and somewhat lower
pressures 1-2 barg towards North and West. Pressures at external explosion sensors towards
North (6m, 12m and 24m outside vent opening) and East (48m outside opening) are shown.
While the nearest sensors to the North may be inside the vented flame ball and have similar
characteristics as internal pressure waves (triangular shape with 100 ms duration), it can be seen
that the blast duration of the sensor 24m North is reduced to 60-70 ms, and for the sensor 48m to
the West the blast pressure has the character of a shock wave.

The third tests is test 8 among the ignited jet releases from the Phase 3B experiments, this is
performed with the same geometry configuration as the previous example with venting in three
directions. The scenario is an 11.6 kg/s natural gas high pressure release centrally in the module
directed towards the blast wall and ignited centrally when approaching steady-state cloud size.
Maximum pressures around 700 mbarg were reported towards East and West with duration 60-
70 ms, towards North, pressures around 400 mbarg with duration 100 ms were reported. Again
near triangular external pressure curves with 100 ms duration are reported outside to the North
(6m, 12m and 24m away), while the blast pressure 48m to the East is in the process of
developing into a shockwave with duration below 40 ms.

The last example shown is from the post-Buncefield DDT test in tree congestion. In this
experiment flames underwent deflagration-to-detonation transition within 10m from ignition, and
the rest of the 50m x 4m x 3m propane cloud detonated. Internal pressures shown were recorded
towards the end of the congested tree array, while the external blast pressures were 13m and 37m
beyond the end of the gas cloud. The duration of the detonation pressure is of the order 5 ms,
while the duration increases somewhat in the far-field to 10-20 ms.
Figure 2 Examples of experimental overpressures inside flame zone (left) and blast
pressures outside the explosion in the near-field/far-field (right) for same
explosion scenario, BFETS-7 (1st row), Phase 3B base case explosion test1
(2nd row), Phase 3B realistic release test 8 (3rd row) and Buncefield tree test 2
(4th row).
Loading onto walls and decks
Explosion loads onto decks, blastwalls, equipment casings/hoods and other walls are usually
extracted by defining measurement panels in FLACS. Usually local panels (3m x 3m) are used to
report localized pressures that may cause local damage, while larger global/total panels, either
covering the entire deck/wall or significant fractions of these, are used to evaluate whether the
structural beams can resist the accidental loads. The pressure panels will report detailed transient
pressure time histories across each wall/deck element, which can be used for anything from
simplified assessments against nominal wall/deck resistance to detailed non-linear finite-element
analysis in which transient pressure curves from each wall element are fed into a finite element
model to calculate whether the maximum deflection may lead to escalation. As both the walls
and the panels will be moved to the nearest grid cell layer, it can be important to check that the
panel actually reports pressure at the same grid layer as the wall is placed. “Inactive panels” used
for pressure reporting are not physical objects, but area sensors which report the average
pressures across. Parts of the panel which are not on surfaces (full porosity on both sides, for
instance if there is a door through a panel) or are fully blocked on both sides, will not be included
when the average pressure is estimated, thus the reporting of pressure is relatively robust.

Figure 3 Example of distribution of pressure panels to report loads onto decks and
walls of an oil platform (left) with a combination of 3m x 3m panels (local),
8m x 8m panels and panels covering entire walls and decks (global). The
right picture shows a possible distribution of 10-4/year local pressure loads
onto a blast wall.
Loading onto rectangular equipment resolved on the grid
For equipment with dimensions larger than 2-3 grid cells (normally 1-2m) in each direction
direct load measurement (DLM) is recommended. This is normally achieved by placing pressure
panels onto the object and extracting maximum local loads onto a surface, which may give local
damage or the differential pressure across the object, which may challenge its supports. As the
directionality of explosion loads can depend on the module layout, directional forces with
duration can be estimated.

Loading onto cylinders and objects of other shapes partially resolved on the grid
For objects like separators and coalescers, major structural beams and other larger objects with
non-rectangular shape, it has been more of a challenge to estimate the loads from explosions
properly. Best practice up to now has been to define an array of pressure point monitors around
the cylindrical object and from that try to estimate maximum pressure differences, either
transient or maximum value, across the object, and thereafter multiply this with a pressure
distribution factor (PDF). FABIG TN-8 has proposed to use PDF=2/ =0.64 for a cylindrical
object, which may be appropriate. There are however challenges using this method, the first is to
ensure that the pressure monitoring points are located sufficiently close to the cylindrical object
to report proper stagnation pressures which is often not the case, it is also questionable to what
extent a few selected points will be representative for the average load onto the cylinder, and
thus there may be a significant uncertainty in the predicted loads. To conclude, the current
methods applied for single cylindrical objects are cumbersome to use, and has a significant
uncertainty, and for a group of objects, such an approach would not be feasible.
In Figure 4 an improved approach for direct load measurement onto cylindrical or non-
rectangular objects and groups of such objects partially resolved on the grid is illustrated. By
defining an array of pressure panels going entirely through all the objects at each layer of grid
cells, it is possible to report accurately the transient directional pressure loading onto the object
or group of objects, without the need for pressure distribution factors or other assumptions. This
method will integrate directional load contributions over the entire surface of all objects within
the bundle with good precision provided each object blocks at least one grid cell. In Figure 4 an
illustration of a shockwave from the left is shown after being reflected at the front of the pipe
bundle, at the same time there is a chaotic pressure reflection pattern between the pipes, and the
2nd row of pipes are further shielded by the front row. Using the traditional approach it would be
very challenging to estimate loads onto the pipe bundle, or even one of the pipes. With the new
proposed approach in which we define an array of pressure panels to calculate the integrated,
directional load onto pipes or pipe bundles, the loads onto the pipe bundle can be estimated. In
lower right plot of Figure 4 the total load on the four pipes, as well as the load on a single pipe
(without the other three pipes present) estimated with the new method is shown. It can be
observed that the peak load onto the pipe bundle is only 10% lower than it would be with four
individual pipes not influencing each other, i.e. the shielding effect is very moderate. The
duration of the load is also significantly longer for the bundle than for a single cylinder, due to a
quicker equalization of pressure across one cylinder than across the bundle.

Loading onto equipment not resolved on the grid

In the introduction section an improved, proposed new method (Hansen and Johnson, 2015) was
described for accurate load estimates onto objects not resolved on the grid (diameter ~2 grid cells
or less). This approach takes into account both loading due to the dynamic pressure (drag) of the
flow but also load contributions due to pressure differences in the flow field due to accelerated
flow field around the object or pressure waves acting onto the object, ref (2).

Force(t) / Area = Cd x Drag (t) + PDF x [Ps(t) - Ps(t- t)]

For box shaped object Hansen and Johnson (2015) applied the method to rectangular objects and
by comparison to direct load measurements (DLM) demonstrated that transient, directional loads
could be estimated with good precision if drag coefficient Cd ~ 2.0 and PDF=1.0 were used. A
similar validation effort has not yet been performed for cylindrical objects or objects of other
shapes. But provided proper directional drag coefficients and PDF are chosen, the approach
should work equally well for these objects, and can easily be adapted also to estimate loads onto
rotated objects not aligned to the Carthesian axes. For cylindrical objects, and some other shapes,
one should be aware that the drag coefficient may be Mach- and Reynolds-number dependent.
While drag coefficients of 1.2, 1.0 or even lower are frequently applied and justified by some, it
is a fact that the drag coefficient of pipes may also be higher (Cd~2.0) in a strong explosion.
Until a better understanding has been developed in this field, it is proposed to apply a drag-
coefficient of 1.2 for cylindrical objects, and PDF=0.64, and potentially even higher, Cd=2.0, for
loads in situations with Mach-numbers approaching 1.0. This will be discussed in the next
chapter.
For precise load estimates care must be taken to detect the directions of the accelerated flow or
pressure waves (whose load contributions could in theory act in different directions at the same
time) in order to keep track of cancelling load contributions or orientation of the loads.

Figure 4 Illustration of method used to report accurate loads from cylindrical tanks
and other non-rectangular shapes partially resolved on the grid. In the upper
plot the pressure field is shown as a shockwave hits a pipebundle. By defining
FLACS pressure panels through every grid plane through a cylinder or
bundle (lower left), the detailed transient load received at one cylinder, or the
bundle as a group of objects, can be predicted with good precision.
Loading onto grated decks and porous objects
For partially porous objects, i.e. louvered walls or grated deck areas, the methods described
above will not work. While a pressure panel defined across a deck area which is e.g. 40% open
and 60% plated will report a pressure load which is representative for the plated area, and ignore
the open areas, a panel defined across a 60% area porosity grated area will instead report the
average load over the entire area. The best way to report the loading onto the grated deck area
will thus be to define a panel across the entire area covered by the grated deck, which is assumed
to have porosity somewhat lower than the grating porosity due to beams and equipment. The
load onto the deck area will then be the panel pressure multiplied by the total area (deck porosity
should not be considered when estimating the total forces to the deck). If there are large H-beams
below the deck receiving some of the loads, or elevated areas above the deck (e.g. skid
fundaments), there may be a need to combine this panel approach with the approach discussed
for objects partially resolved on the grid (as shown in Figure 4), in order to estimate the total
load onto the deck.

4 How to select drag coefficients – example from large scale explosion test

The estimated explosion loading for small objects becomes directly proportional to the chosen
drag coefficient. FABIG TN-08 guidance note recommends using conservative drag coefficients
for load estimates on piping and small equipment, which means Cd=1.2 for pipes, and likely
Cd~2.0 for rectangular profiles. Still, some standards recommend lower drag coefficients, e.g.
Cd=1.0 for pipes (API-RP-2FB, 2006 and DNV-RP-D101, 2008), and to limit design costs it may
be no surprise that many companies choose the lower values. Based on simulations of gas
explosions Hansen and Johnson (2015) concluded that conservative drag coefficients (Cd~2.0 for
box shapes) were required to obtain accurate predictions of loading, and further expecting that
Cd~1.2 would be appropriate for cylindrical objects. A validation effort for cylindrical objects
similar to that performed for rectangular objects, identifying the drag coefficients required to
estimate the reported DLMs using free-flow parameters (drag, pressure and density) has not yet
been performed.

Steady-state flow experiments have shown that while the drag coefficient of Cd=1.2 may be
appropriate for low Re, with increasing Re (beyond 105) the drag-crisis may give a narrower
wake, with the result that the drag coefficient drops to less than half, before increasing again to
0.7-0.8 at Re=107. The drag-crisis is observed for developed, steady-state flow scenarios around
a single cylinder, and it is highly questionable whether this is relevant for the flow-field inside a
turbulent explosion at its maximum strength. It is assumed that upstream wake effects and
accelerated flow fields will prevent the developed flow situation of the drag-crisis, and therefore
recommended to use conservative drag coefficients for cylinders (Cd=1.2 and even 2.0 for Ma
approaching 1.0). In the following a validation example will illustrate why.

Drag coefficients – example case & validation

During the BFETS full-scale explosion experiments at Spadeadam (SCI, 1998) an instrumented
cylinder was used to measure received explosion load for some of the explosion tests. The actual
cylinder was a 2.0m long 0.5m diameter vertical pipe located at floor level within one meter of
the East exit of the full-scale rig, see FLACS-model and photo in Figure 5. The cylinder was
placed on top of a load cell which would report the torque along and across the axis of the
module, which will be the main flow direction. In addition to the load cells at the base of the
cylinder, there were also four pressure sensors at mid-height of the cylinder, which could be used
to get a good estimate of the pressure load across the cylinder along and across the main flow
direction. There were also 25 other pressure sensors inside the module which would report
transient pressure.

Figure 5 FLACS model of BFETS module with instrumented load cell cylinder (left),
and photo (SCI, 1998) of actual cylinder (right)
The case used for this example is the BFETS test 12, which was a center ignited, slightly
overstoichiometric natural gas explosion experiment in the base-case 25.6m x 8m x 8m tunnel
geometry with venting towards East and West. The maximum overpressures reported in this
experiment were generally near 2.5 barg at the centre of the module, and around 1.0-1.5 barg
near the ends, with the exception of one sensor near West exit (load cell was at East) which
reported a pressure peak around 3.1 barg with a short duration. In the left plot of Figure 6
maximum pressures at two sensors at the center of the module is shown as reported in the
experiment (green) and as simulated with FLACS using coarse (1.0m-blue) and fine (0.5m-red)
grid resolution. The predicted overpressure at the ground next to the instrumented cylinder is
also shown for all three cases. In general the pressure predictions seem very good for most of the
25 locations, both for the fine grid and the coarse grid simulations, however it can be seen that
the pressure next to the cylinder is somewhat high (almost 2.0 barg vs 1.5 bar) for the 0.5m grid.
In the right plot of Figure 6 the flow Mach number next to the cylinder can be seen at four
different heights, peaking at Ma=0.6 prior to flame arrival, and between Ma=0.7 (lower part of
cylinder) and Ma=0.9 (upper part of cylinder) after the flame has passed. Due to the higher
temperature in the flame zone, the actual flow velocity is around 4 times higher after the flame
has passed.

With a grid resolution of 0.5 or 1.0m is it not feasible to extract any reliable stagnation pressures
onto the load cylinder, however, by refining the grid to 0.10m locally around the cylinder, it is
possible to extract pressures around the cylinder with reasonable precision. This approach is
mentioned in (Lloyd’s Register, 2015), but requires a special setting to make the time stepping
routine in FLACS ignore the grid refinement to deliver reasonable predictions, and the approach
is not practical to use for more comprehensive studies including a large number of simulations.
In Figure 7 the differential pressures at mid-height across the cylinder in the direction of the
module axis can be seen for the experiment (red) and for the two FLACS simulations with 0.5m
(purple) and 1.0m (green) base grid resolution. The first pressure peak (prior to flame arrival)
reached 0.75 barg in the simulations, however, only 0.5 barg in the experiment while the second
and main peak reached around 1.8 barg in the experiment and around 1.5 barg in the simulations.
Except for some mismatch in the shape of the first peak, and some noise and termination on the
reported differential pressure peak from experiment, the shape and duration of the main
differential pressure peak corresponds very well to the simulations. It should be observed that the
duration of the main peak seems close to 100 ms. In the right plot of Figure 7 the maximum
pressures reported at object surfaces can be seen for the fine grid simulation (0.5m base grid with
refinement to 0.10m).

Figure 6 Simulated pressure curves (near centre of module) with 0.5m grid and 1.0m
grid compared to BFETS 12 experiment (left), predicted Ma-numbers in the
flowfield next to the instrumented cylinder (right) using 0.5m grid resolution.
The comparison of pressure curves from simulations and experiment at various locations inside
the module, and the differential pressure across the instrumented cylinder, confirms that the
simulations can reproduce the dynamics of the explosion test with good precision. It is therefore
reason to believe also that the Mach-numbers reported, as well as dynamic pressures (Drag)
presented in Figure 8 at different elevations next to the pipe are predicted with good accuracy.

In Figure 8 to the right the reported load cell torque (Force x length) received at the cylinder is
plotted. As the distance between the load cells is 0.67m this corresponds to a maximum average
force per area received over the projected cylinder area (0.5m x 2.0m=1m2) of 120 kPa, while
from Figure 7 a maximum pressure difference P across the cylinder at mid-height was reported
to be around 1.8 bar (180 kPa), while Figure 8 shows a maximum drag load around 60 kPa. For
the moment of maximum load we may then coarsely estimate

Cd = Load / Drag ~ 120 kPa / 60 kPa = 2.0

PDF = Load / DP ~ 120 kPa / 180 kPa = 0.67

which confirms that the proposed PDF=2/ =0.64 from FABIG TN-08 seems reasonable.

In the right plot of Figure 8 the load cell received torque is plotted together with an estimated
torque resulting from the drag forces for different heights above the cylinder base for the full
duration of the explosion. In order for the received and estimated torques to match, a drag
coefficient of around 0.6 would be appropriate prior to flame arrival (first peak of load curve),
however after flame arrival a drag coefficient as high as 2.0 seems to be required, as estimated
above. Thus it seems as if very high drag coefficients for pipes can be seen already at drag values
of 50-60 kPa and for Mach numbers around 0.8. Comparing the load cell reported torque with
the estimated load due to drag forces in Figure 8 indicates that the load cell cylinder enter
oscillations with a period around 50 ms after reaching the maximum displacement.

Figure 7 Differential pressure across load cylinder from pressure sensors mounted at
front and back. Simulations using local grid refinement and 1.0m and 0.5m
base grid size versus experimental measurement (left), maximum pressure
field (right)
To conclude this section, this example demonstrates that cylinder drag coefficients may be both
lower (0.6 prior to flame arrival, at Ma~0.6) and higher (2.0 after flame arrival at Ma~0.8) than
typically used drag coefficients (1.2 or 1.0) used for piping design. While more relevant cases
should be studied before concluding, industry should ensure to use sufficiently conservative drag
coefficients for design.

5 Response – filtering loads using Dynamic Load/Amplification Factor (DLF/DAF)

The loading from an explosion scenario will be a transiently changing pressure distribution
acting onto all objects, walls and decks. For a given object this can give deformation of the
surface (local pressures) or threaten the integrity (global or differential pressures). For blast
walls, structures or piping systems this may give deflections or deformations and if the
deformations are significant it may lead to failure and loss of containment of hydrocarbon piping
or equipment, and a potentially uncontrolled escalation.

The degree of damage depends on the magnitude of the overpressure or pressure difference, the
area this is acting upon, and the duration of the pressure loads (or integrated pressure impulse).
While all the necessary detailed information may be available from a CFD explosion study, it is
common that the load is simplified into one single number, the maximum overpressure, or at best
a maximum overpressure averaged across a smaller or larger surface, while the time dependency
is often ignored.
Figure 8 Predicted drag around cylinder (upper left), and load cell output versus force
predicted based on drag curves using drag coefficient Cd=0.6 prior to flame
arrival and Cd=2.0 after flame arrival (upper right). In the lower picture the
estimated received loading kN (=kPa over 1m2 projected area) onto the
cylinder (black curve assuming Cd=2.0) is shown together with the static
side-on pressure (blue) and drag pressure (red curve) using the methodology
of Hansen and Johnson (2015)
The duration of pressure loads, in particularly some distance outside the explosion, will tend to
be short, often only 10-30 ms. For heavy structures e.g. blast walls or separators, this may be
much too short to accelerate the structure, and such short loads will barely be felt by the objects.
In a probabilistic study consisting of several hundred different explosion scenarios, the received
loads onto an object may partly be due to explosions surrounding the object with load duration
100-200 ms and partly due to explosions some distance away with shorter load duration onto the
target. In a probabilistic assessment such loads are normally sorted after increasing severity into
a load frequency exceedance curve. For simplicity the maximum overpressure is used to rank the
severity, even if some of the high pressure loads have a too short duration to be of any threat to
the structure. From the probabilistic assessment a pressure with e.g. a 10-4/year return frequency
for a given object or worst-case across a group of object, would be selected as dimensioning. If a
static design would be applied, this estimated overpressure would be multiplied by a worst-case
dynamic amplification factor (DAF), typically of 1.5 for a triangular load pulse shape, and the
structure is thereafter designed for this static strength. Such an approach may be much too
conservative, since some of the loads classified as severe due to high maximum pressure would
actually be less of a problem due to the short duration of pressure pulse.

With all the information available from an explosion study, it is possible to take the load duration
into account. By applying a dynamic load factor (DLF) to estimate a characteristic static load for
each pressure curve on a target, a more accurate severity scaling may be obtained. In this
example the elastic response load function in Figure 9 is applied (FABIG, TN-08). For a load
duration td/T, between 0.5 and 1.5 (dynamic regime), the load will be multiplied by a factor of
1.2 to 1.6, while for shorter load durations (impulsive regime) the severity of the load will be
proportional to its impulse, not its maximum pressure. For loads with a longer duration than 1.5
times the eigenperiod (quasistatic regime) the load should be multiplied by a factor around 1.0-
1.2. With this approach a more accurate ranking of severity is achieved.

In practice DAF and DLF are the same and in (HSE, 1999) both are defined to be “the ratio of
the deflection at any time to the deflection which would have resulted from a static application of
the load, which is used in specifying the load time variation”.

Figure 9 If a static design is applied the predicted maximum loads are often multiplied
by a dynamic amplification factor of 1.5. In reality this factor will depend on
the eigenperiod of the object (see graph from FABIG TN-08, left). If this is
taken into account in the assessment, and the predicted load is scaled, the
load exceedance curves could be adjusted based on the eigenfrequency of the
objects.
When applying the DLF of Figure 9 (left), modified load frequency exceedance curves can be
obtained as seen in Figure 9 (right plot), which shows the maximum load onto a 1.5m diameter
object located anywhere, with any orientation inside a module. More than 500 explosion
scenarios, each with a frequency of occurrence, have been used to generate each of these curves.
While the red curve is the typical unfiltered pressure exceedance curve resulting from typical
explosion studies, the black curve will be the same data after multiplied by a dynamic
amplification factor (DAF) of 1.5. Then there are three more curves using the DLF-filters
assuming eigenperiods of 50, 100 and 200 ms of the objects. If a design frequency of 5 x 10 -
5
/year is applied, this would give required design strength of 1.5 bar without considering the
eigenperiod. By assuming 50, 100 or 200 ms eigenperiod of the structures, this would reduce the
required static design strength significantly, to 1.3, 1.2 and 0.8 bar respectively. And if also the
size, orientation and position of the critical equipment would be considered, a significantly lower
required design load might be estimated.

6 DAL-Area differentiation

One of the main purposes of an explosion study during design phase is to establish the design
strength of the facility. Depending on the philosophy this can either be designed to withstand a
credible or realistic worst-case incident, or more commonly for offshore facilities, to survive
accidental explosions for 10,000 years or more. To design for any credible scenario can be
challenging. Sometimes credible is defined by the size of the gas release, only considering the
90% lower hole sizes, but since the severity of an explosion scenario depends on so many other
factors, e.g. combination of release location, direction, wind speed and direction, ignition
location and time, and the worst-case consequences for different targets will be seen for different
scenarios, it is actually quite challenging for a safety consultant to perform a “credible worst-
case” study properly. If “non-credible” instead would be defined to be the most severe scenarios
with a return frequency less than every 10,000 years, these two approaches would converge.

A typical approach when defining dimensional accidental loads (DimAL) from an explosion
study, for instance on an oil platform, is to define a number of safety critical functions or
elements, e.g. main support frame, blast walls or decks between processing areas and safe utility
areas on an oil platform, or escape route impairment. A main job will then be to define the
necessary dimensioning strength for blast walls and decks to maintain their function separating
the different areas, or proper design strength for hydrocarbon piping, so these will avoid rupture
and severe fire escalation that may impair escape routes. Safety systems like deluge piping
should also be designed sufficiently strong, the same applies for any piece of equipment to
prevent it from becoming a projectile that could contribute to uncontrolled escalation in the event
of a significant explosion. Since defined safety critical functions or barriers can be impaired in
different ways, there is often a significant room for optimization when defining dimensioning
accidental loads. A weak design of piping will lead to a higher frequency of fire, which may
require better protection with passive fire protection or more focus on protection of escape routes
from smoke. The division between process or drilling areas and safe areas (utility, TR) may
consist of several different wall elements, and the tolerable frequency of escalation (e.g. 10 -4/y)
can be distributed between these, and the consultant can choose to apply comparable design
strength for all wall elements, so that the main risk for escalation is found in the areas with most
frequent strong explosions, or alternatively a stronger design will be applied here, so that the risk
for failure is more evenly distributed among the various areas.

Once the consultant has proposed dimensioning accidental loads the engineering company and
the asset owner will consider this input, and potential uncertainties, and thereafter choose design
accidental loads (DeAL). These must be selected so that the structures can withstand equally
high loads or higher loads than specified in the DimAL, to account for uncertainties and give
additional robustness of the facility.
For onshore siting studies the design strength of control rooms, and other buildings or areas
where people are working or sheltering people, may be more in focus.

To establish proper dimensioning loads based on a probabilistic study it is important to have

modeled a significant number of scenarios, with as representative as possible distribution of
scenarios and frequencies. In addition it is important to have clear tolerance criteria for the
targets, either individual targets like main blast wall or TR, or combination of targets like any
hydrocarbon carrying pipe or equipment inside the module that may rupture and lead to escalated
fire.

For individual targets it may be important to check that frequency of scenarios that may give
loads beyond design strength must be lower than tolerance criteria, depending on the design,
there maybe criteria for localized loads/damage (maximum pressure at any local panel) and loads
acting onto the entire structure (global panels). While it is common to develop pressure
frequency exceedance curves for one wall at a time based on all simulated scenarios that may
contribute to failure, a different approach will be required to estimate the failure properly, since
some explosion scenarios may give failure on several walls, another group of scenarios may
impair e.g. the West wall and finally a third group of scenarios may impair the North or South
walls. For a consistent handling there may thus be a need to calculate the potential for
impairment scenario by scenario on either of the walls, and to complicate the evaluation, the
different walls may also have a different design strength.

Similarly, if the strength of the blast wall local panels is verified, one should not evaluate the
frequency of each particular panel to fail, but the frequency for at least one panel to fail the
criteria. With a good communication between structural disciplines and the safety consultant, it
may be possible to go even more in detail regarding the failure criteria, e.g. considering not only
the maximum magnitude of the load, but also the duration.

When establishing the dimensioning loads for hydrocarbon piping and equipment one can often
assume that wherever there is a high drag load, there will be a hydrocarbon pipe or equipment
that may break and lead to fires. Thus, the frequency of rupture of a given pipe/segment is
therefore not particularly interesting, but instead the frequency that one or more pipes or
segments anywhere in the process area will rupture from explosion events should be established.
For this reason it is common to define one dimensioning drag load across the entire module area,
and similarly one load for each object size for intermediate objects.

For large process areas, which includes areas of different character both regarding leak sources
and degree of congestion/confinement, there may be systematic variation in loads across the
module. In the inner part of the module, both the magnitude and duration of loads may be higher
than in the less confined outskirts of the module. For such situations it may be a waste of steel
and money to apply the same dimensioning accidental loads everywhere. In Figure 10, upper
picture, an illustration shows a process module which has been split into 10 different regions,
and for each region the frequency for exceeding a typical design drag load has been calculated.
The plot clearly shows that the explosion load severity is significantly higher in the central parts
of the module close to the blast wall. In order to avoid overdesign a differentiated dimensioning
drag load across the module was proposed, with the highest dimensioning load in the most
exposed areas of the module, 20% lower resistance in the corners of the module, and 40% lower
resistance in the outskirts (cantilevers), see illustration in the lower part of Figure 10. By
assessing all the simulated scenarios against failure, it was found that this modification of
proposed dimensioning load led to only a marginal increase of the required maximum
dimensioning drag load, while a weaker design would be acceptable in large parts of the module.

If the detailed location and orientation of larger hydrocarbon process equipment and piping is
known, such an assessment could easily be combined with an assessment of larger vessels,
separators, knock-out drum and more, in which directional loads are evaluated for the larger
equipment

Possibilities are many, and there is a lot of information available from an extensive explosion
modeling study which could be of great value to the structural assessment study. The more
information that is utilized in the structural assessment, the better job can be done. Typically the
more simplified the output from the explosion study to the structural assessment become, the
more conservatism will be required in the design.

Figure 10 Illustration of area differentiation of proposed dimensioning drag loads onto

piping. In the upper plot the annual frequency of exceeding a typical
dimensioning load (=failure) is evaluated by area. Based on this it is proposed
to keep a robust design in central areas of the module near the blast wall,
while the dimensioning loads are proposed reduced towards the periphery of
the module, see proposed load reduction factors by area in the lower plot.

7 Discussion and conclusion

The extraction of explosion loads onto equipment from CFD explosion calculations is done
every day by safety consultants. The guidance and understanding on how this should be done is
often weak, and very conservative or non-conservative results may result. In this paper several
aspects around extracting explosion loads have been discussed. First the need for accurate CFD
calculations and a representative scenario selection is discussed. Thereafter shape and duration of
explosion loads, as well as optimal ways for extracting loads on various types of targets are
handled. Among the novel suggestions in this paper is how detailed loads may be extracted for
cylindrical vessels, as well as pipe bundles, by the use of multiple pressure panels. A recently
developed approach to predict detailed transient, directional loads onto small and medium sized
equipment is also further discussed. The choice of drag coefficients is discussed, and a full scale
explosion experiment in which received loading was reported is simulated for validation and
illustration purposes. This both confirmed that the pressure distribution function,
PDF=2/ =0.64, proposed by FABIG TN-08 seems reasonable, but also that for explosions of
significant strength conservative drag coefficients should be applied for pipes. For the
experiment simulated a drag coefficient as high as Cd=2.0 was necessary to estimate loads
properly. The paper also discusses challenges with linearizing the load severity into maximum
pressure only, and illustrated the potential benefits if a dynamic load function filter, taking into
account duration of load, could be applied when linearizing the load severity. Finally there is a
discussion on establishing dimensioning accidental loads from an explosion study, in which an
example is shown where differentiated design strength for different parts of a module is
proposed. To conclude, the better the precision in the predicted accidental loads is, and the more
information that can be utilized in the structural assessment, the more optimal and cost-efficient
it is possible to design.

References

API-RP-2FB (2006), Recommended Practice for the Design of Offshore Facilities against
Fire and Blast Loading, the American Petroleum Institute

API-RP 752 (2009): Management of Hazards Associated with Location of Process Plant
Permanent Buildings, the American Petroleum Institute

DNV RP-D101 (2008), Structural analysis of piping systems, Det Norske Veritas, October 2008

FABIG TN-08, Protection of Piping Systems Subject to Fires and Explosions, The Steel
Construction Institute, Silwood Park, Ascot, Berkshire, SL5 7QN, United Kingdom

FLACS 10.4 Users Manual (2015), www.gexcon.com

Hansen, O. R., Talberg, O., & Bakke, J. R. (1999). CFD-based methodology for quantitative gas
explosion risk assessment in congested process areas: examples and validation status. In Int.
Conf. and Workshop on Modelling the Consequences of Accidental Releases of Hazardous
Materials. 28 Sept-1 Oct 1999 (pp. 457-477). San Francisco, California: AIChE, CCPS.

Hansen O.R., F. Gavelli, S.G. Davis, and P. Middha (2013), Equivalent cloud methods used for
explosion risk and consequence studies, J Loss Prev Process Ind 26 (2013), 511–527.
Hansen, O.R., Kjellander, M.T., Martini, R. and Pappas, J.A. (2014), Estimation of explosion
loading on equipment from CFD simulations, proceedings 17th Annual Int. Symp., Mary K
O'Connor Process Safety Center, College Station Oct 28-30, 2014, pp 1042-1069

Hansen, O.R., Storvik, I. & van Wingerden, K. (1999). Validation of CFD-models for gas
explosions, FLACS is used as example. Model description, experiences and
recommendations for model evaluation. Presented at the ‘European meeting on chemical
Industry and Environment', 1-3 September 1999, Krackow, Poland: 16pp.

Hansen, O.R. and Johnson, D.M. (2015), Improved far-field blast predictions from fast
deflagrations, DDTs and detonations of vapour clouds using FLACS CFD, Journal of Loss
Prevention in the Process Industries, 35 pp 293-316, May 2015.

HSE, 1999, Review of Analysis of Explosion Response, Offshore Technology Report –OTO
98174, UK Health and Safety Executive, Bootle, Merseyside, L20 3DL, UK

ISO 19901-3 (2010) Petroleum and natural gas industries - Specific requirements for offshore
structures - Part 3: Topsides structure, the Int. Org. for Standardization

Lloyd’s Register, 2015, Guidance Notes for the Calculation of Probabilistic Explosion Loads,
issued February 2015, Lloyd’s Register EMEA, 71 Fenchurch Str, London EC3M 4BS, UK

NORSOK Z-013, (2010). Risk and emergency preparedness analysis. Norsok standard.
Available from Standard Norge, Postboks 242, N-1326 Lysaker, Norway.

Sand, I.Ø. (1999). Received loading test cases. Fire and Blast Information Group (FABIG)
Technical Newsletter, Issue 24, R347: 14-18

SCI (1998), Blast and Fire Engineering for Topside Structures, Phase 2, SCI publication number
253, The Steel Construction Institute, Silwood Park, Ascot, Berkshire SL5 7QN, UK