Health and Safety Executive

Assessment of gas detection strategies for offshore HVAC ducts based on CFD modelling Prepared by the Health and Safety Laboratory for the Health and Safety Executive 2007

RR602 Research Report

Health and Safety Executive

Assessment of gas detection strategies for offshore HVAC ducts based on CFD modelling Dr C J Lea & Dr M Deevy Health and Safety Laboratory Harpur Hill Buxton SK17 9JN

The aim of this study has been to undertake CFD modelling to provide a basis for advice to inspectors and the industry on the effectiveness of flammable gas detection strategies for offshore HVAC ducts. CFD simulations of a high and low pressure gas release have been undertaken for idealised representations of an offshore platform, as well as a high pressure release for a more realistic geometry based on the Brae Alpha platform. In parallel with this modelling work a literature review has been carried out to build on a scoping study by Walsh et al (2005).

This report and the work it describes were funded by the Health and Safety Executive (HSE). Its contents, including any opinions and/or conclusions expressed, are those of the author alone and do not necessarily reflect HSE policy.

HSE Books

© Crown copyright 2007

First published 2007

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means (electronic, mechanical, photocopying, recording or otherwise) without the prior written permission of the copyright owner.

Applications for reproduction should be made in writing to:Licensing Division, Her Majesty’s Stationery Office,St Clements House, 2-16 Colegate, Norwich NR3 1BQor by e-mail to hmsolicensing@cabinet-office.x.gsi.gov.uk

Acknowledgements

The authors would like to thank Marathon Oil UK Ltd, Flamgard Engineering Ltd, Integrated Engineering Services Ltd, Groveley Detection Ltd and Honeywell Analytics for providing information and assistance during the course of this project. Dr Peter Walsh, HSL, has also provided helpful guidance.

ii

CONTENTS

1 INTRODUCTION 1

2 FLOW AND DISPERSION OF GAS IN A DUCT 32.1 Turbulent flow in pipes and ducts 32.2 Dispersion in pipes and ducts 52.3 Summary 8

3 INGESTION OF GAS INTO HVAC INLETS 93.1 Introduction 93.2 Scenario 1 103.3 Scenario 2 22

4 DISTRIBUTION OF GAS INSIDE AN HVAC DUCT 324.1 Overview 324.2 Results 33

5 GAS RELEASE AND INGESTION INTO HVAC DUCTS FOR A 37REALISTIC SCENARIO

5.1 Introduction 375.2 Model geometry and computational domain 375.3 Boundary conditions and flow physics 385.4 Mesh, solution numerics and convergence 435.5 Results 45

6 DISCUSSION 536.1 Review and discussion of results 536.2 Initial recommendations for gas detection strategies in HVAC ducts 56

7 CONCLUSIONS 587.1 Recommendations for further work 59

8 APPENDIX 1: EXAMPLES OF COEFFICIENT OF VARIATION 60

9 REFERENCES 61

iii

iv

EXECUTIVE SUMMARY

Objectives

The aim of this study has been to undertake CFD modelling to provide a basis for advice to

inspectors and the industry on the effectiveness of flammable gas detection strategies for

offshore HVAC ducts. CFD simulations of a high and low pressure gas release have been

undertaken for idealised representations of an offshore platform, as well as a high pressure

release for a more realistic geometry based on the Brae Alpha platform. In parallel with this

modelling work a literature review has been carried out to build on a scoping study by Walsh et

al (2005).

Main Findings

The most significant finding is that in all of the CFD simulations the distribution of gas at

HVAC inlets is non-uniform: large variations in gas concentration are present over the cross-

section of the modelled HVAC inlets. This result is consistent with the theoretical behaviour of

a high pressure gas release. Neither is there a substantial reason to believe that it is not also

consistent with the behaviour of a low pressure gas release. It means that there is the potential

for a gas release to be ‘missed’ by detection systems unless this non-uniformity in gas

concentration is anticipated in the selection and siting of gas detectors at HVAC inlets.

The CFD results also show that a variation in gas concentration over a duct cross-section only

reduces slowly with distance along a straight duct. This finding is consistent with a body of

relevant literature stemming from the sampling of gas distributions in the exhaust ducts of

nuclear stacks.

This literature highlights that purpose-designed mixing elements and bends in a duct can be

effective in creating well-mixed conditions in a duct, but at the cost of increased pressure drop.

It also suggests that relatively small-scale obstructions such as louvres and fire dampers are

unlikely to significantly enhance mixing. This is borne out by CFD modelling of such

obstructions in this study.

The implications of the modelling work, substantiated by the literature, are that in the absence

of purpose-designed mixing elements or a series of bends upstream from gas detectors no

significant benefit would be gained from siting detectors a significant distance downstream

from an HVAC inlet. Also, no significant benefit can be expected to be gained from siting

detectors inside an HVAC duct compared to locating them immediately outside the HVAC

inlet.

The CFD results have been post-processed to gain further insight into the likely effectiveness of

point and infra-red beam or aspirated detector systems for HVAC ducts. The resulting output

has been combined with the findings of the scoping study by Walsh et al (2005), and the

outcomes of the literature review and CFD modelling as reported above, to provide a basis for

the following initial recommendations on flammable gas detection strategies:

x� Detector alarm levels should be set as low as reasonably practical: 10% LEL or less.

x� Point catalytic, point infra-red, extended path point infra-red, cross-duct beam infra-red

and aspirated point detector systems all have the potential to be effective in detecting

non-uniform distributions of flammable gas in and around HVAC ducts provided that

their sensitivity is sufficiently high (low detection limit) and that due regard is given to

the possibility that gas will be distributed non-uniformly.

v

x� Extended path point infra-red detector systems currently appear to offer the greatest

sensitivity, but multiple detectors should be used and sited so as to anticipate non­

uniform mixing.

x� Cross-duct beam infra-red, extended path or aspirated point detector systems should be

based on two approximately orthogonal beams or lines of aspirated point probes.

x� No significant benefit can be expected to be gained from siting detectors inside an

HVAC duct compared to locating them immediately outside the HVAC inlet.

x� In the absence of purpose-designed mixing elements or a series of bends upstream from

gas detectors no significant benefit is to be gained from siting detectors a significant

distance downstream from an HVAC inlet.

x� Mixing elements have the potential to reduce any non-uniformity in the distribution of

gas in a duct but their effectiveness should be proven by physical tests.

Recommendations

It is recommended that large-scale physical tests using commercial detectors are undertaken to

validate the findings of this study and to support and refine the initial recommendations above.

It is also recommended that the minimum practical alarm levels for a range of commercial

detector types, especially cross-duct beam infra-red systems, are clarified.

Reference

Walsh P, Johnson A, Ivings M (2005) Effectiveness of gas detection in HVAC ducts: Scoping

study. Health and Safety Laboratory Report FM/04/11.

vi

1 INTRODUCTION

The accidental release of flammable gas on offshore installations can potentially lead to the

build-up of an explosive mixture. Natural or forced ventilation can help to mitigate such

incidents and gas detection systems play a key role in reducing the risks from releases by

enabling early detection and subsequent interventions. The provision and siting of gas detectors

for open areas and gas turbine enclosures has been studied over a number of years and is

comparatively well documented (HSE 1993, Kelsey et al 2000, Ivings et al 2002, HSE 2003,

Kelsey et al 2005). However there is much less information available on gas detection systems

for HVAC ducts supplying air to accommodation modules, temporary refuges or process areas

on an installation. The information which is available on offshore HVAC ducts tends to address

other aspects such as mechanical design with minimal information on gas detection systems (BS

EN ISO 15138:2001, Chin & Lam 2004).

Following an incident on the Brae Alpha platform in November 2004 in which there was a

delay in confirmed detection and shutdown of the HVAC system, despite gas being ingested

into the HVAC inlets, it became clear that there was a need for a study to establish the current

state of the art. This led to the initiation of the scoping study by Walsh et al (2005) into the

effectiveness of gas detection systems for HVAC ducts on offshore installations. This was

carried out in response to concerns raised by HSE offshore safety inspectors on the positioning

of flammable gas detectors in or around HVAC inlets.

Walsh et al (2005) reviewed national standards, industry guidance, published papers and reports

to obtain information on the siting of flammable gas detection systems for offshore HVAC

inlets. Some industry guidance was obtained on performance targets for HVAC gas detection

systems specific to a particular platform, stating that flammable gas at concentrations above

20% LEL (Lower Explosive Limit) at ventilation intake ducts should lead to an alarm and that

concentrations above 60% should lead to isolation of the intake by stopping fans and closing

dampers (Micropack, 2000). Other industry guidance advised that either point or beam detectors

may be installed inside or outside of HVAC ducts and that for point detectors three should

normally be installed to maintain voting logic (Shell UK, 1995). Guidance has also been

obtained from a gas detector company advising on the appropriate proximity of point detectors

to features such as bends or diffusers in gas turbine ductwork.

Walsh et al (2005) surveyed commercial gas detector systems for HVAC applications. They

found that a number of differing systems are available: point catalytic, point infra-red, beam

infra-red or aspirated systems. In general the operating range of such systems was found to be

up to 100% LEL, with a minimum alarm level of 20% LEL. However, Walsh et al noted that

recent developments in HVAC gas detection technology have led to the availability of extended

closed path infrared point detectors which measure an average concentration over a path length

of typically around 1 m. Such systems have increased sensitivity and typically allow for

minimum alarm levels of 5% LEL. Recent developments in catalytic point sensors for the

industrial gas turbine industry also apparently allow for minimum alarm levels of 5% LEL.

However, Walsh et al (2005) found essentially no guidance specific to the siting of gas detectors

in and around HVAC inlets or on the effectiveness of differing siting strategies. They

recommended that further work be undertaken to remedy this situation. In consultation with the

Offshore Safety Division (OSD) in HSE they proposed that this further work should initially be

based on Computational Fluid Dynamics (CFD) modelling, recognising that CFD can guide any

potential future experimental work by helping identify effective and ineffective gas detection

systems. This provides the basis and motivation for the project which is documented in the

present report.

1

The overall aim of this project is therefore to apply CFD modelling to provide a basis for advice

to inspectors and the industry on the effectiveness of flammable gas detection strategies for

offshore HVAC ducts. It builds upon the scoping study of Walsh et al (2005). The primary

focus of the project is on siting strategies for gas detectors located in or around HVAC inlets for

circumstances in which gas is not ingested uniformly over the entire inlet. The present project

and the earlier scoping study have both been funded by OSD.

In Section 2 of this report the behaviour of flow and dispersion of gas in a duct is discussed.

This discussion is supported by a literature review that encompasses sources of information

which compliment that from the field of gas detection in offshore HVAC ducts.

Section 3 addresses circumstances by which a non-uniform distribution of gas could be present

immediately outside or inside of an HVAC inlet. This is of particular relevance to the present

project. Thus, if a well-mixed and uniform distribution of gas and air is present then the

positioning of detectors should have little impact on the performance of a system in detecting

gas; instead the performance will be governed by other factors such as the sensitivity, speed of

response and reliability of the detector. Such information is available from gas detector

companies supported by experience of use. Therefore in this section the outcome of CFD

modelling of gas releases is presented for two differing idealised scenarios showing how a non­

uniform distribution of gas and air may be present in and around an HVAC inlet.

Section 4 reports the mixing and distribution of gas inside an HVAC duct as simulated using

CFD. The modelling extends that presented in Section 3 by focusing on the detail of the

geometry and associated flow features likely to be found in an actual HVAC inlet.

Section 5 presents CFD results for a more realistic scenario based on the release from the Brae

Alpha platform in November 2004. Although the most significant elements of this incident are

represented, the scenario is not a detailed replication of the incident; this would be very time-

consuming to undertake and may not be practical. The value of these results instead lies in the

more general lessons which they provide on the behaviour of a release in a complex geometrical

environment and which is subsequently ingested into HVAC ducts of varying size.

The results from the CFD modelling are discussed in Section 6. Their implications for gas

detection strategies for HVAC ducts are highlighted. Initial recommendations on the effective

siting of flammable gas detectors in and around HVAC inlets are also provided. Conclusions

and recommendations for further experimentally-based work are given in Section 7. Appendix 1

provides illustrative values of a parameter which can be used to characterise the degree of

mixing in a duct. References are provided in Section 9.

All of the CFD modelling has been undertaken using ANSYS CFX 10, a commercial software

package.

2

2 FLOW AND DISPERSION OF GAS IN A DUCT

2.1 TURBULENT FLOW IN PIPES AND DUCTS

An introduction to the flow regime in HVAC ducts is provided by Walsh et al (2005). They

show that for offshore HVAC ducts the Reynolds number will be sufficiently high that the flow

will be fully turbulent. This is based on a typical duct velocity of 5 m/s as outlined in BS EN

ISO 15138 and a range of duct hydraulic diameters from 0.5 to 5 m, giving Reynolds numbers 6in the range of approximately 105 to 10 .

Turbulence is often assumed to be synonymous with good mixing. Although this is generally

the case, turbulent mixing does not occur instantaneously. It would be wrong to assume that

because the flow in HVAC ducts is turbulent any non-uniformity in a distribution of gas at an

HVAC inlet will very quickly be dispersed to give well-mixed uniform conditions. This has a

significant impact on the siting of gas detectors in HVAC ducts.

An indication that this is the case is given by experimental data on turbulent developing flow in

a pipe. Sufficiently far downstream from the entrance to a long straight pipe the flow eventually

reaches a condition in which it no longer changes. It is then said to be fully-developed.

For turbulent flow in a smooth pipe the fully-developed mean velocity profile approximately

follows a 1/n power law (Hinze, 1975): / 1 n

§¨

U 2x · ©

¸¹U d max

where U is the velocity at distance x from the wall of a pipe of diameter d and Umax is the peak 5velocity on the axis of the pipe. The constant, n, is 7 for a Reynolds number of 10 and

approximately 9 for a Reynolds number of 106. For rough pipes, n increases to between 4 and 5.

The turbulent mean velocity profile is therefore rather flat across much of the cross-section of a

pipe, dropping away steeply at the walls. Turbulent flow in a pipe is therefore often

characterised as consisting of a relatively thin near-wall region and a central core.

The distance before fully-developed conditions are established in a pipe is known as the

development length. The development length depends on a number of factors: the Reynolds

number, geometry of the inlet and flow conditions upstream of the inlet. For flow from

quiescent surroundings into a straight pipe with rounded edges at its inlet the development

length for fully-developed turbulent flow is given by White (1987):

L e Re 4.4 6 / 1

d

where Le is the development length and d is the pipe diameter. The development length is 6approximately 30 d to 40 d for Reynolds numbers in the range of interest, i.e. 105 to 10 . For

laminar flow (Reynolds number below about 2000 to 4000) the development lengths are much

greater.

The above expression applies to idealised conditions in which the flow is from quiescent

surroundings and the pipe is straight and rounded at its inlet. In these circumstances the flow

which enters the pipe is initially uniform and laminar. A laminar boundary layer grows on the

surface of the pipe and then rapidly becomes turbulent. The turbulent boundary layer grows in

3

thickness until eventually it extends to the axis of the pipe. Fully-developed flow conditions are

reached shortly afterwards.

Klein (1981) correlates data on developing flow in pipes from a range of sources, which include

sharp-edged inlets. These data show that fully-developed flow is not established until greater

than about 50 d downstream from the inlet. Other workers (Laws et al, 1979) have indicated that

in the presence of significant obstructions at the inlet to a pipe, fully-developed flow is not

established until significantly further downstream, possibly greater than 100 d. It should be

noted that these studies are seeking flow profiles which are exactly the same with distance from

the pipe inlet and so are rather rigorous in their requirements. As a practical rule of thumb Hinze

(1975) recommends that turbulent fully developed flow can be assumed to occur in straight

pipes after a minimum development length of 40 d.

The turbulent developing flow in a straight square or rectangular duct behaves in a broadly

similar manner, in that the distance to fully-developed flow is not short. Melling & Whitelaw

(1976) present data which show that turbulent fully-developed flow in a square duct is reached

at about 25 duct widths from the inlet. Other measurements, by Gessner et al (1977), show that

fully-developed flow is not reached until much further downstream, at between 40 to 84 duct

widths downstream from a rounded inlet. The development of the flow occurs in much the same

manner as that for a circular pipe, with the gradual growth of boundary layers on the duct walls

until they occupy the entire section of the duct.

There is a further feature of flow in straight square or rectangular ducts which should be

mentioned. This is the existence of a turbulence-driven secondary flow in the cross-stream

direction. The secondary flow is of a very much smaller magnitude than the primary flow along

the axis of the duct (Demuren & Rodi, 1984). Its effect is to slowly transport material from the

centre of the duct towards the walls and back again. Overall this leads to an enhancement in

turbulent mixing.

This secondary flow presents a difficulty for CFD models. Simulations based on the commonly-

used k-H turbulence model (Launder & Spalding, 1972) completely fail to capture this secondary

flow (Speziale, 1996). More advanced turbulence models are available (Gatski & Speziale 1993,

Gatski & Rumsey 2002) which do capture the secondary flow but they often tend to be more

problematic to apply and require more computational resources. However, the secondary flow is

not fully established immediately the flow enters a duct; in fact it is only completely established

when fully developed conditions are met far downstream. In practice, offshore gas detectors for

HVAC ducts are nearly always located close to the inlets. This is partly to minimise the volume

of gas ingested following action to isolate an HVAC system upon detection of gas, and partly to

provide access for maintenance and calibration. Therefore, although a standard k-H turbulence

model cannot capture these secondary flows, the resulting error is unlikely to be significant in

the few duct widths immediately downstream from an HVAC inlet, i.e. the region in which gas

detectors are commonly found.

The literature on developing and fully-developed flow in pipes and ducts gives a useful

indication that equilibrium in the turbulent diffusion of momentum – responsible for the

creation of fully-developed flow profiles – only occurs at a considerable distance downstream

from the inlet to a straight pipe or duct. Of course, the offshore environment is somewhat

different from the idealised experiments upon which this literature is based: offshore HVAC

inlets do not have rounded edges; often they are sharp-edged and have obstructions such as

grilles or weather louvres across their entire face. Figure 1 shows some examples of such

features.

4

2.2

Figure 1: Examples of HVAC inlets

Usually, dampers will be encountered a short distance inside an HVAC duct; to provide

isolation from fire and gas in the event of an incident. The dampers will normally be open and

parallel to the main flow direction, but they still present an obstruction to the flow, as does their

supporting structure. Overall, the gross effect of obstructions such as grilles, louvres and

dampers will be to generate turbulence which will lead to enhanced mixing.

DISPERSION IN PIPES AND DUCTS

The above discussion has concentrated on the characteristics of turbulent flow in straight pipes

and ducts. Our main interest is in the turbulent transfer of mass, or more specifically, in the

turbulent diffusion of a contaminant such as natural gas at a relatively low concentration. It

should be noted that at low concentrations in the flammable range and for the velocities

typically encountered in offshore HVAC ducts, a natural gas mixture can be regarded as a

passive contaminant, i.e. it is transported by, but does not directly influence the flow. This is

because the Richardson number (Simpson, 1997) is likely to be at least an order of magnitude

too low for any turbulence-modifying effects of a slightly buoyant gas to dominate over shear-

induced turbulence.

The turbulent diffusion of a contaminant which is essentially passive takes place in an

analogous manner to that of the turbulent diffusion of momentum. Both are the result of the

same turbulent flow-field. It can therefore be expected that the dispersion of a passive

contaminant in a duct will follow roughly the same behaviour as that of the development of the

velocity profile: if the development length is long then the distance to uniformly mixed

conditions will also be long.

A review of the literature was undertaken to obtain more information on the development of an

initially non-uniform distribution of passive gas in a duct or pipe, in the presence or absence of

obstructions. Initially this review focused on a search for papers in the field of gas dispersion, or

detection, in ducts or HVAC systems. This produced little useful information on the distribution

and mixing of gas in ducts. A search for papers in the field of sampling, rather than detection or

dispersion, proved much more fruitful.

Hence a number of papers were obtained on the sampling of exhaust duct stacks in the nuclear

industry (Hampl et al 1986, Rodgers et al 1996, McFarland et al 1999a & 1999b, Anand et al

2003 and Seo et al 2006). These papers report research into the mixing of a passive tracer in

circular, square and rectangular ducts in the presence or absence of bends and mixing elements.

5

In most cases the tracer was released from a single location on the axis of a duct. Multiple point

concentration measurements were then made across the entire cross-section of the duct at a

number of axial locations. This body of research was largely undertaken to support an

improvement and updating of American standards on the sampling of gaseous radionuclide

emissions from stacks. Sampling had historically been required at multiple points over the

cross-section of a duct, with as many as 20 points being required for large ducts (Rodgers et al,

1996) to ensure that peaks in the concentration distribution were not missed. The sample plane

was required to be no closer than eight duct diameters from the nearest upstream flow

disturbance and two duct diameters from the nearest downstream disturbance. However, Hampl

(1986) suggested that up to 50 duct diameters may be needed for near-uniform mixing of a

passive tracer in a straight pipe, but less than two duct diameters for acceptable mixing

downstream of two elbows. Proposals for a new approach were therefore made, based on a

single point sample taken at a location where acceptable well-mixed conditions could be

assured.

To characterise the degree of mixing, a parameter known as the Coefficient of Variation (COV)

was introduced. This is defined as:

N

i

meani CC N

COV

¦ � � 1

2)( 1

1

C mean

where N is the number of samples at a particular downstream location, Ci is the concentration of

the ith sample and Cmean is the mean concentration over all samples at that location, defined as:

1 N

C ¦C mean iN i 1

As an example, if the concentration distribution is such that across one half of a duct the

concentration is a uniform 30% LEL whilst across the other half the concentration is a uniform

10% LEL, then the COV would be 50%. Appendix 1 provides examples of concentration

distributions for a range of COV between 10% and 150%. The updated American standards

(ANSI/HPS N13.1-1999) allow for single point sampling of gaseous contaminants in a duct if

the COV for both velocity and concentration of a tracer gas are less than 20% over the central

two-thirds of a duct. We do not suggest or comment on the practicality or appropriateness of

these criteria for offshore HVAC ducts. However, the notion of a COV is helpful in quantifying

the uniformity of mixing in a duct: it is readily computed from CFD results. In addition, this

body of research on stack sampling provides much useful data on how the COV is affected by a

range of configurations; this can be used to inform the present research.

Thus McFarland et al (1999a) showed that for a tracer released at a single point on the axis of a

duct the COV can readily be reduced to ~5% following passage through a compact mixing

chamber installed in the path of a duct. However, the pressure loss for this mixing chamber was

not small (non-dimensional pressure coefficient of approximately 4.5, equivalent to a head loss

of about 70 Pa for an average velocity of 5 m/s) and, more significantly, it would require re-

engineering of existing HVAC installations.

Measurements of the COV in straight circular pipes by Anand et al (2003) show that the

distance from a point release of a tracer to that at which the COV is less than 20% depends on

the upstream turbulence intensity. For a low turbulence intensity of 1.5% the COV falls very

slowly with distance and is still over 100% at 30 duct diameters downstream from the point of

6

release. Even with a high turbulence intensity of 10% - generated by passing the flow through

an array of thick rods – a COV of 20% was still not reached after 25 duct diameters

downstream.

McFarland et al (1999b) investigated the effect of bends and static mixing elements on the

COV. They show that a single smooth 90o bend in a circular duct still requires a distance of nine

diameters downstream from the bend before the 20% COV criterion is met. The performance of

the static mixing elements was very variable: all resulted in the 20% COV criterion being met

within nine diameters downstream, but the most effective mixers were able to meet the criterion

within three duct diameters of the mixing element. The most simple and effective mixing

elements consisted of two large flow deflectors attached to opposite walls of a duct, leading to a

slot-like opening in the centre of the duct. Two or more of these mixing elements were used in

series. The common characteristic of the most effective mixing elements appears to be the

introduction of cross-stream mixing from one side of a duct to another due to the introduction of

large turbulent eddies (Seo et al, 2006). Mixing elements which only introduced flow swirl were

less effective. One disadvantage of these simple deflector mixing elements is a relatively large

non-dimensional pressure coefficient (5.0) for two elements in series.

Seo et al (2006) examine the behaviour of the COV in square and rectangular ducts (aspect ratio

of 3:1) with and without bends. They report that the COV is similar for circular and square-

section ducts at large distances downstream from the tracer release point, both with and without

bends. This implies that the main findings of the previous work on circular-section ducts,

discussed above, are likely to largely carry over to square-section ducts. However, a significant

difference was found when the COV for the square and rectangular-section ducts were

compared: typically the COV was much higher for the rectangular duct at any given distance

downstream from the point of release, by about a factor of four. The COV for rectangular ducts

with bends are also consistently higher than the same flow configuration in a square duct. Seo et

al speculate that this is because turbulent eddies in a wide duct have less opportunity to

effectively transfer mass and momentum from one side of a duct to another.

The effect of grilles on the turbulence in a duct is relatively well understood. Laws & Livesey

(1978) explain that the effect is either to suppress, or enhance turbulence, and this depends on

the geometry of the grille. Thus a very fine grille, or mesh, will tend to suppress turbulence.

Any turbulence which is introduced by the mesh decays quickly due to its small scale. A grid of

relatively large diameter rods will enhance turbulence, although Laws & Livesey state that it is

difficult to achieve a turbulence intensity of much higher than 10%. It is not clear whether

grilles typically used to cover HVAC inlets will suppress or enhance turbulence. However, even

if turbulence is significantly enhanced, Anand et al (2003) show that the COV will remain high

for long distances downstream. For an inlet turbulence intensity of 10% created by an array of

rods they found that the COV was still over 50% at 15 duct diameters downstream from the rod

array. The reason for the relative ineffectiveness of such devices on mixing is that they

introduce turbulence on too small a length-scale.

Much of the above work is based on a flow which is well-controlled at the inlet to a duct, for

example by use of a rounded entrance or other flow control devices. In the case of offshore

HVAC installations this will not be the case: turbulence can be generated by large-scale flow

separation at the sharp-edged entrance to a duct or is already present in the ambient flow outside

of the duct. McFarland et al (1999b) note that the earlier work of Hampl et al (1986) was based

on a sharp-edged inlet and in comparison to their later research with a well-controlled approach

flow the COV were found to be reduced by between a factor of two to three: flow separation at

the inlet enhances mixing inside a duct. Nevertheless, it should be recalled that Hampl et al still

suggest that up to 50 duct diameters may be needed for near-uniform mixing of a passive tracer

in a straight pipe, even with a sharp-edged inlet. It is difficult to know to what extent the effect

7

of turbulence which is present in the ambient flow affects the mixing inside a duct. However,

the gross effects of pre-existing turbulence in the outside ambient flow on mixing in a duct

should be represented, albeit crudely, by the CFD modelling which follows in Sections 3, 4, and

5.

It is also significant that both Anand et al (2003) and Seo et al (2006) note that the COV is little-

affected by the Reynolds number. Seo et al state that for a square duct there is only a small

effect of the Reynolds number on COV over a range from 25,000 to 150,000. For a rectangular

duct with a 3:1 aspect ratio there is a significant dependence on the COV for a Reynolds number

below 50,000, but relatively little effect at higher Reynolds number. Both Anand et al (2003)

and Seo et al (2006) conclude that, for fully turbulent flow, mixing is primarily dependent on

geometry.

2.3 SUMMARY

The main findings of relevance to the present project from the above literature review are as

follows:

x� The Reynolds number for offshore HVAC ducts is sufficiently high that the flow will

be fully turbulent.

x� A flow development region exists downstream from the inlet to a duct; in the absence of

obstructions and for a smoothly rounded inlet the length of this region can be assumed

to be a minimum of 40 duct diameters for a straight duct – whereupon the flow is said to

be fully-developed and does not change further downstream.

x� In the development region the flow may not be turbulent across the entire section of a

duct: in this region it is more likely to be turbulent in the presence of a sharp-edged

entrance to the duct and where the duct contains obstructions such as louvres.

x� A weak secondary flow exists in square and rectangular ducts which will enhance

mixing. In CFD modelling this secondary flow is not captured by the most commonly-

used k-H turbulence model. However, the secondary flow is not completely established

until the flow is fully developed, so associated errors from use of the k-H model are

unlikely to be significant close to an HVAC inlet.

x� A natural gas mixture which is ingested into offshore HVAC ducts at concentrations in

the flammable range can generally be treated as a passive contaminant.

x� The turbulent diffusion of a passive contaminant in a duct will take place in a broadly

analogous manner to that of the turbulent diffusion of momentum; long development

lengths imply similarly long distances before well-mixed conditions are obtained.

x� The degree of mixing in a duct can be quantified by a Coefficient of Variation, COV.

x� Uniform mixing of a point release of a passive tracer in a duct typically requires 50 duct

diameters downstream from the release for a straight circular duct. The presence of

bends and mixing elements can significantly reduce this distance.

x� The turbulent mixing in a rectangular duct with a large aspect ratio (3:1 and above) is

significantly less effective than that in a square or circular duct.

x� Effective mixing in a duct is driven by large-scale turbulent eddies which are able to

transfer mass and momentum across the duct.

x� The most effective mixing elements are those which introduce large turbulent eddies.

x� Grilles at the entrance to HVAC ducts are likely to be ineffective in introducing large-

scale turbulent eddies and so will do little to enhance mixing.

x� For fully turbulent flow, mixing in a duct is primarily dependent on geometry.

8

3 INGESTION OF GAS INTO HVAC INLETS

3.1 INTRODUCTION

The circumstances by which a non-uniform distribution of gas could be present immediately

inside or outside of an HVAC inlet are of particular interest in the present project. If an HVAC

inlet ingests a non-uniform distribution of gas then there is the possibility that this could be

missed by the detection system.

To understand how it might be possible that a non-uniform distribution of gas could be present

at an HVAC inlet, it is useful to briefly examine the characteristics of a gas jet. For simplicity

we examine a free round jet, although in practice a release is likely to be more complex.

Nevertheless, this simple case does give a broad indication of the flow behaviour which might

be expected of some more complex releases.

Far downstream from a high pressure release from a round hole, in the region where the Mach

number is low enough for the flow to be treated as incompressible (Ma < 0.3), the

concentration on the centre-line of the resulting jet is inversely proportional to distance (Rodi,

1982):

T D m

T 194.0 z o

� where Tm is the concentration on the centre-line at distance z from a release of diameter D with

an initial uniform concentration TR. Far downstream, the concentration varies continuously

across the radius of a round jet and can be described by the following function (Rodi, 1982):

z r

126.0 2�T T e

�� m

where T� is the concentration at radius r.

An example is helpful to indicate the radial variation in concentration which is implied by the

expression immediately above. Consider a release of pure methane such that the centre-line

concentration of the resulting jet is equal to LEL 10 m downstream from an initial diameter D

(to be defined). Concentrations of 100%, 50%, 20% and 10% LEL would be obtained at radii of

0 m, 1.05 m, 1.6 m and 1.9 m.

In fact a jet of these dimensions could be expected from a high pressure release of methane at a

stagnation pressure and temperature of 100 bar and 40oC, from a hole of approximately 12 mm

diameter, respectively. This is a large, but credible, release. Note that the hole diameter of 12

mm is not D in the expression above. D is the diameter of the jet when it has expanded to

atmospheric pressure, 85 mm in this instance - considerably larger than the release hole. The

expressions by which this fully expanded diameter has been calculated are given in Section 5.

This example shows that for a large, but credible, high pressure jet release the radial

concentration at 10 m downstream can vary from a maximum of 100% LEL at the jet centre-line

to just 10% LEL at a radius of 1.9 m. This distance, over which concentration varies by a factor

of ten, is broadly comparable to the dimensions of typical offshore HVAC inlets. If such a

release were ingested into an HVAC inlet, then significant non-uniformity in gas concentration

could be expected outside and inside of the HVAC duct.

9

In practice a release could be affected by many factors such as the wind flow in and around a

module, the presence of obstructions, impingement on nearby objects, etc. The release could

also be at lower pressure and more significantly influenced by buoyancy. The effects of these

factors, and others, are difficult or impossible to predict from theory alone, but in principle can

be captured by CFD modelling. Therefore two scenarios have been simulated using ANSYS

CFX 10 to illustrate and examine circumstances in which a non-uniform distribution of gas and

air may be ingested into HVAC inlets. The scenarios are idealised but still include many of the

complicating factors mentioned above.

Scenario 1 is a high pressure release from a riser on the underside of a platform. The release is

directed horizontally underneath the platform towards an HVAC duct which faces vertically

down and whose inlet protrudes from the underside of the platform. Scenario 2 is a low pressure

release from within a module. The release is directed upwards and out of the module to pass up

the downstream face of a platform. A horizontal HVAC duct protrudes from this downstream

face of the platform and is intersected by the release. Full details of these idealised scenarios,

including a description of the CFD modelling and results, are given below

3.2 SCENARIO 1

3.2.1 Model geometry and computational domain

The model geometry is shown in Figures 2 and 3. Although it is a very simplified representation

of an offshore platform this geometry presents approximately the same obstruction to the

approaching wind-field as a real platform, but without the inclusion of any platform-specific

geometrical elements such as flare-stacks, helipad, etc. The modelled geometry consists of a

cube of side 30 m whose base is located 25 m above sea level. The release is located 3 m below

the platform, 6 m inboard from the platform edges, and is initially angled at 30o across the

underside of the platform towards its downstream edge. The release is also oriented at a very

shallow angle upwards such that its axis as it exits from the underside of the platform is

approximately 2 m below the base of the platform. The location and initial direction of the

release are indicated in Figures 2 and 3. A single HVAC duct of external dimensions 3 m x 1.5

m and internal dimensions 2.84 m x 1.34 m is located 4 m from the downstream face of the

platform and 12 m from the nearest edge of the platform. The size of the duct was guided by

information supplied by Integrated Engineering Services Ltd (specialist designer and supplier of

offshore HVAC systems). The axis of the duct is vertical, 20.9 m long and protrudes 2 m below

the underside of the platform. The HVAC duct can be seen in Figures 2 and 3.

Figure 2: Scenario 1, geometry (HVAC duct in blue, gas release in red)

10

a) Plan view

b) Vertical elevation

Figure 3: Scenario 1, dimensions (m)

An extensive region of the atmosphere that surrounds the platform is included in the CFD

model. This is to ensure that the simulated flow around the platform is not constrained by the

presence of the nearby boundaries of the computational domain. In particular it allows for the

presence of large-scale time-dependent flow features to develop naturally as a consequence of

flow over the bluff body of the platform. The resulting computational domain is 120 m wide,

11

115 m high, 240 m long and is illustrated in Figure 4. It includes the entire length of the HVAC

duct. Note that the platform has been oriented at 20o to the oncoming wind direction.

Figure 4: Scenario 1, computational domain

3.2.2 Boundary conditions

The platform is located in the atmospheric boundary layer. Therefore at the upwind face of the

computational domain a neutral stability atmospheric boundary layer profile was prescribed

(Hargreaves & Wright, 2007). The wind speed was specified to be 1.5 m/s at 25 m above sea

level. Although this is a low wind speed the results show that it is still sufficient to deflect the

trajectory of the jet towards the HVAC duct. At the upper boundary of the computational

domain a constant shear stress was applied by imposing a boundary velocity equal to that which

would be expected at 115 m above sea level. At the downwind and two side boundaries in the

atmosphere a condition was applied which allows for flow to leave or enter the domain.

At sea level a rough wall boundary condition was applied. The sides, base and top of the

platform were also specified as rough walls, with a surface roughness appropriate for corrugated

metal (White, 1987).

A mass flow rate of 19.2 kg/s was imposed at the upper end of the HVAC duct. This type of

boundary condition allows the simulation to calculate an appropriate flow profile, rather than

requiring a flow profile to be prescribed. The mass flow rate is equivalent to a uniform velocity

of 4 m/s. Simulations have also been undertaken with a mass flow rate which is equivalent to a

uniform velocity of 6 m/s and the results are qualitatively very similar.

The release was assumed to be from a high pressure riser. Note that the riser is not explicitly

included in the model since it will have negligible effect on the flow. In reality a gas release

from a high pressure riser will initially be under-expanded and supersonic, i.e. it exhausts from

the riser, accelerates to high Mach number then relaxes to atmospheric pressure by expansion

through a complex shock structure. It is not practical or necessary to explicitly simulate this

phase of a release. Instead, it is appropriate for such a release to be modelled from a location at

12

which expansion to atmospheric pressure has occurred and the flow is just sonic (Ivings et al,

2003). This has two advantages: it avoids the need to model the high Mach number region –

which can be problematic; and the dimension of the jet after expansion is much greater than that

of the release hole and so is more readily resolved by the CFD mesh.

In this scenario the release is modelled from the point at which expansion to atmospheric

pressure has occurred. The release is sized and specified using the expressions in Section 3.1 for

a free round jet as a rough guide. In this case the release is fixed at approximately 23 m from the

HVAC duct. A peak concentration of 50% LEL was decided upon as a target concentration at

this location since this is comparable to current action levels upon detection of gas at 50% to

60% LEL (Walsh et al, 2005). Using the expressions in Section 3.1 implies that after expansion

to atmospheric pressure a jet which meets these target conditions should be specified as being

0.1 m in diameter. Note that for a high pressure release this actually corresponds to a much

smaller hole size. For example, if the riser pressure and temperature were 100 bar and 40oC,

then the hole size would be approximately 14 mm diameter for a release of 100% methane.

Section 5 provides expressions by which this effective hole size can be calculated.

The release was therefore modelled as a sonic source of 100% methane over a nominal diameter

of 0.1 m. The temperature of gas in the riser was assumed to be 40oC which gives a sonic

velocity of 428 m/s. The release rate is approximately 2.5 kg/s.

3.2.3 Flow physics

The flow over a bluff body such as an offshore platform is known to be inherently transient.

Large vortices are typically shed from a bluff body to form an oscillating wake. In fact initial

attempts to compute this scenario as steady-state failed to converge. Therefore, all the

simulations undertaken and presented in this report are time-dependent.

The flow around the platform and in the HVAC duct will be turbulent. The commonly-used k-H turbulence model (Launder & Spalding, 1972) has been used throughout this work.

Wall functions were used at all solid boundaries, appropriately modified to account for surface

roughness.

The physical properties for methane were taken from Santon et al (2002), including its LEL of

4.4% by volume.

3.2.4 Mesh

An unstructured mesh was used to resolve the geometry. The mesh was refined in regions where

high gradients of velocity and concentration were expected; across the methane jet, the HVAC

inlet and duct. A total of 22 mesh elements were used to resolve the release source. On the walls

of the platform four layers of prismatic mesh elements were used to aid resolution of the near-

wall flow and help ensure that the first mesh node correctly falls within the log-law region for a

turbulent boundary layer. A mesh with a total of 317,000 nodes (control volumes) was

generated, constructed from 1.7 million mesh elements. The mesh is illustrated in Figures 5 to

7, overleaf.

13

Figure 5: Scenario 1, mesh - vertical elevation coincident with the release

Figure 6: Scenario 1, mesh – plan view coincident with the release

14

Figure 7: Scenario 1, mesh – detail near the HVAC inlet

3.2.5 Solution numerics and convergence

All simulations in this work have been undertaken using high-order-accuracy spatial and

temporal discretisation schemes in an effort to reduce numerical errors.

Time-dependent simulations were undertaken with a short time-step of 1 s to aid convergence

and also to help resolve the main time-dependent flow features. A non-dimensional frequency

known as the Strouhal number, S, can be identified for the time-dependent flow over a sharp-

edged building:

S fW /U

where f is frequency, W is width in the cross-wind direction and U is mean velocity. For a cube

of side 30 m in a 1.5 m/s flow the Strouhal number has a value of approximately 0.1 (Schetz and

Fuhs, 1996), giving a frequency of 0.005 Hz, equivalent to a time-period of 200 s. Hence, a

time-step of 1 s is short in comparison to this long time-period. The transient simulation was run

until the flow appeared periodic. A total of 1600 time-steps were computed.

The residuals for all transport equations decreased by between at least two and usually three

orders of magnitude, to very small rms values, and the global imbalances of mass, momentum,

energy and gas mass fraction all reduced to very much less than 0.05% of reference values. This

indicates that the solution is well-converged.

15

3.2.6 Results

The flow-field around the platform is shown in Figure 8 for two time-steps separated by 60 s.

Although the gross flow is similar – consisting of massive flow separation in the wake of the

platform – the detail of the flow in the wake region is dissimilar.

a) time t

b) time t + 60 s

Figure 8: Scenario 1, velocity field in plan view at mid-height of the platform

16

The flow-field in a vertical elevation is shown in Figure 9. Again, the separated region in the

wake of the platform can be seen.

Figure 9: Scenario 1, velocity field shown on a vertical plane

The time-dependent wake flow has a minor effect on the trajectory of the gas jet, but sufficient

to affect the concentration distribution at the HVAC inlet and inside the duct. An iso-surface of

gas concentration at 50% of LEL is shown at two times in Figures 10 and 11. These times

represent the two extreme states in the quasi-periodic time-dependent flow.

a) time t b) time t + 60s

Figure 10: Scenario 1, iso-surface at 50% LEL, plan view

17

a) time t b) time t + 60s

Figure 11: Scenario 1, iso-surface at 50% LEL, iso-metric view

The very small difference in the trajectory of the jet at these two time-steps can be seen to result

in a more significant influence on the flow within the HVAC duct. This is especially evident in

Figure 11, in which a more prominent tongue of gas can be seen in the duct at t + 60 s.

The full extent of the gas jet is shown in Figure 12, below. Gas concentrations of 10% LEL and

above are evident a significant way downstream from the platform.

Figure 12: Scenario 1, gas concentration in % LEL, plan view 0.5 m below the HVAC inlet

The distribution of gas at 0.5 m below the HVAC inlet, 1 m and 3 m inside the HVAC duct, is

shown in Figure 13 at time t and t + 60 s. There is a large gradient in concentration across the

duct in each case, with concentration differing by a factor of approximately five or six across

the breadth of the duct. However, there is only a small difference in the concentration

distribution between planes located immediately outside and inside of the duct. Although gas

concentration is somewhat higher in the HVAC duct at time t + 60 s, the overall flow behaviour

18

in the duct remains the same for both time-steps. Hence, the results in Figures 14 to 17, which

follow below, are all presented at time t.

a) time t b) time t + 60s

Figure 13: Scenario 1, gas concentration in % LEL at the HVAC inlet

The flow behaviour at the HVAC inlet is shown in Figure 14. The jet impacts one side of the

HVAC duct and creates a high velocity flow up one side. At the other side of the duct the flow

separates and recirculates. This recirculation region probably contributes to the somewhat better

mixed conditions at this side of the duct.

Figure 14: Scenario 1, flow-field at the HVAC inlet, velocity vectors shown on a vertical plane which intersects the HVAC duct

19

The distribution of gas around and inside the HVAC duct is shown in Figure 15. The location of

the gas jet just outside of the duct is clearly seen, with peak concentrations greater than 100%

LEL. This is higher than the target concentration of 50% obtained from the round jet

expressions in Section 3.1. The difference is no doubt a result of the additional complexity

which occurs for a release in a cross-flow and whose spreading is constrained by the under-side

of the platform. Figure 15 indicates that well-mixed conditions are not achieved at the duct

outlet, almost 21 m distant from the inlet. The presence of a weak swirling flow in the duct is

implied by the concentration distribution in Figure 15, as the location of the peak gas

concentration at a cross-section is gradually shifted to the upwind face of the duct.

The distribution of gas at two sample lines across the breadth and width of the HVAC duct, 1 m

above the inlet is shown overleaf; Figures 16 and Figure 17.

Figure 15: Scenario 1, gas concentration in % LEL in the HVAC duct

20

Figure 16: Scenario 1, location of sample lines 1 m inside the HVAC inlet

100 100

90 90

80 80

70 70

60 60

% LEL 50 % LEL 50

40 40

30 30

20 20

10 10

0 0

0

5

0

5

0

5

0

5

0

5

9

4

9

4

9

4

9

4

9

4

0

7

4

1

8

5

2

9

6

3

1

8

5

2

9

6

3

00 1 3 4 6 7 9 0 2 3 4 6 7 9 0 2 3 5 6 8 0 0 1 2 2 3 4 4 5 6 7 7 8 9 9 0 1 2

0.

0.

0.

0.

0.

0.

0.

1.

1.

1.

1.

1.

1.

1.

2.

2.

2.

2.

2.

2.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

1.

1.

1. . .

Distance across breadth of duct, m Distance across width of duct, m

a) Across the breadth of the duct b) Across the width of the duct

Figure 17: Scenario 1, distribution of gas 1 m inside the HVAC inlet

Line integrals of gas concentration have been calculated at the locations shown in Figure 16 to

model the output of cross-duct beam infra-red detectors. Values of these line integrals are

shown in Table 1. The average concentration of gas in the duct is 27% LEL.

Table 1: Scenario 1, Line integrals of gas concentration inside the HVAC duct

Location of line integral Length of Concentration

Orientation Distance inside line LEL.m %LEL

inlet (m) integral (m) (m) per m

Mid-plane across breadth 1.0 2.84 0.78 27

Mid-plane across width 1.0 1.34 0.27 20

An indication of the effectiveness of mixing in the HVAC duct can be given by post-processing

the CFD results to obtain the Coefficient of Variation (COV) – as defined in Section 2.2 - at a

number of cross-sections. The COV falls very slowly over the first three metres of the duct:

57%, 56% and 52% immediately outside, at 1 m and 3 m inside, respectively. 10 m inside the

duct it has only fallen to 39%. At the outlet of the duct the COV falls to 19%. It can be seen

from the examples in Appendix 1 that COV values of 50% and greater imply a very high degree

of non-uniformity in concentration distribution, whilst even at a COV of 20% there is still a

significant difference between maximum and minimum values of concentration.

21

127

1

34

3.3 SCENARIO 2

3.3.1 Model geometry and computational domain

The model geometry is shown in Figures 18 and 19. Again, this is a simplified representation of

an offshore platform. The modelled geometry is retained as a cube of side 30 m whose base is

located 25 m above sea level. A horizontal HVAC duct of internal dimensions 2.8 m x 1.8 m is

located 23 m above the base of the platform. It is offset to one side of the platform and opens

onto its downstream face, i.e. the wake region. The duct protrudes 0.5 m from the face of the

platform. The size of this duct was also guided by information supplied by Integrated

Engineering Services Ltd. The duct is 23 m long and is shown in Figures 18 and 19.

The release is located towards the downstream end of a module which is located at low level in

the platform. The module is assumed to be 5 m high and its floor is 5 m above the base of the

platform. The module has two solid walls and is open at its upwind and downwind faces. A

congested region occupies the upwind half of the module. The release is sited 0.46 m inside the

module, 2 m below its roof and is directed upwards at an angle of 70o to the horizontal. The

release is not directly below the HVAC duct, but is offset towards the centre of the module. The

location and initial direction of the release are indicated in Figures 18 and 19.

Figure 18: Scenario 2, geometry (HVAC duct in blue, gas release in red)

22

a) Plan view

b) Vertical elevation

Figure 19: Scenario 2, dimensions (m)

Once again an extensive region of the atmosphere that surrounds the platform is included in the

CFD model. The computational domain is as per Scenario 1: 120 m wide, 115 m high, 240 m

long. It includes the entire length of the HVAC duct. Note that, as per Scenario 1, the platform

has been oriented at 20o to the oncoming wind direction.

3.3.2 Boundary conditions

The boundary conditions at the six outer faces of the computational domain were prescribed to

be the same as those of Scenario 1, except that the wind speed was slightly higher; 2 m/s at 25

m above sea level.

A mass flow rate of 37.7 kg/s was imposed at the end of the HVAC duct, equivalent to a

uniform velocity of 6 m/s.

In this scenario the gas release was assumed to be at low pressure and velocity, but input over a

larger area than that of scenario 1. Hence, the release was specified to be 100% methane at

23

ambient temperature of 10oC and an initial velocity of 12.3 m/s over a circular area of nominal

diameter 0.29 m. The mass release rate is 22% that of scenario 1, being 0.55 kg/s. It is

postulated that a release of this type and magnitude could possibly result from a break in a

pressurised line or vessel which is surrounded by sufficient lagging or insulation to significantly

reduce but not entirely destroy its initial momentum.

3.3.3 Flow physics

The flow was again modelled as time-dependent. A standard k-H turbulence model (Launder &

Spalding, 1972) was also used. Wall functions were employed at all solid boundaries, modified

to account for surface roughness. Physical properties were as per Scenario 1.

The congested region which occupies the upwind half of the open module was modelled as a

region of isotropic resistance with a volume porosity of approximately 0.8. It has the effect of

reducing, but not eliminating, the cross-flow through the module.

3.3.4 Mesh

An unstructured mesh was generated. The mesh was again refined in regions where high

gradients of velocity and concentration were expected. On the walls of the platform and inside

the module four layers of prismatic mesh elements were used to aid resolution of the near-wall

flow and help ensure that the first mesh node correctly falls within the log-law region for a

turbulent boundary layer. A mesh with a total of 378,000 nodes was generated, constructed from

2.0 million mesh elements. The mesh is illustrated in Figures 20 to 22.

Figure 20: Scenario 2, mesh - vertical elevation coincident with the release

24

Figure 21: Scenario 2, mesh –plan view through the HVAC duct

Figure 22: Scenario 2, mesh – detail near the HVAC inlet

3.3.5 Solution numerics and convergence

Transient simulations were undertaken using high-order-accuracy spatial and temporal

discretisation schemes with a time-step of 1 s. The simulations were carried out until the flow

appeared quasi-periodic. Over 1300 time-steps were computed. The residuals for all transport

equations decreased by three or more orders of magnitude, to very small rms values, and the

25

global imbalances of mass, momentum, energy and gas mass fraction all reduced to very much

less than 0.05% of reference values. This indicates that the solution is well-converged.

3.3.6 Results

The flow-field around the platform is very similar to that of Scenario 1, shown in Figures 8 and

9, except for the existence of a weak flow through the open module. The gross flow again

consists of massive flow separation in the wake of the platform.

In this scenario the gas release has far lower momentum than the high pressure release of

Scenario 1. Thus, it is more influenced by the time-dependent flow in the wake of the platform.

In addition, the release is also partly driven by its buoyancy. These effects interact in a complex

manner and their net result is that the gas plume meanders significantly in the platform wake.

This can be seen in Figure 23, overleaf, which shows iso-surfaces at 50% LEL at 40 s time

intervals. Essentially the plume oscillates from side to side. At the extremity of this oscillation it

encounters the HVAC duct and gas is ingested, see Figures 23e and 23f.

26

a) time t b) time t + 40s

c) time t + 80s d) time t + 120s

e) time t + 160s f) time t + 200s

Figure 23: Scenario 2, iso-surface at 50% LEL, iso-metric view, at 40 s intervals

27

All of the results which follow are presented at the extremity of the meandering motion of the

plume at time t + 200 s, when gas is being ingested into the HVAC duct as per Figure 23f.

Figure 24 shows that the plume attaches to the sheltered face of the platform towards its upper

end. The distribution of gas on cross-sections at 0.5 m outside the HVAC inlet, 0.5 m, 2 m and

10 m inside the HVAC duct, as well as at the end of the duct, is shown in Figure 25, together

with the concentration on a horizontal plane down the middle of the duct. Figure 26, overleaf

provides a close-up view of concentration at the HVAC inlet.

Figure 24: Scenario 2, gas concentration in % LEL, elevation through the gas plume

Figure 25: Scenario 2, gas concentration in % LEL in the HVAC duct

28

Figure 26: Scenario 2, gas concentration in % LEL at 0.5 m outside, 0.5 m and 2.0 m inside the HVAC inlet

Figures 25 and 26 can be compared to Figures 13 and 15, showing that there is an even more

pronounced concentration gradient across the duct than in Scenario 1. In one corner of the duct,

gas which is very close to its LEL is being ingested. However, in the opposite corner the

concentration is close to zero. Again, well-mixed conditions are not achieved at the duct outlet.

Also, the presence of a weak swirling flow in the duct is again implied by the gradual shift in

location of the peak gas concentration at a cross-section.

Figure 27 shows that the flow stays largely attached to the walls of the duct as it enters the

HVAC inlet, with little sign of any significant flow separation.

Figure 27: Scenario 2, flow-field at the HVAC inlet, velocity vectors shown on a vertical plane which intersects the HVAC duct

29

The distribution of gas at four sample lines across the breadth and height of the HVAC duct, 0.5

m and 2 m inside the inlet – see Figure 28 below - is shown in Figure 29.

Figure 28: Scenario 2, location of sample lines 0.5 m and 2 m inside the HVAC inlet

100 100

90 90

80 80

70 70

60

0.5m

inside

2.0m

inside

60

% LEL 50 % LEL 50

40 40

30 30

20 20

10 10

00

0.5m

inside

2.0m

inside

00. 051. 092. 044. 095. 047. 088. 130. 181. 133. 174. 126. 177. 129. 260. 212. 263. 215. 256. 208.

0 9 9 8 8 70 0 1 2 3 4. . . . . .0 0 0 0 0 0 07 65 6. .0 067. 058. 059. 140. 141. 132. 133. 124. 125. 116. 117. 108.0

Distance across breadth of duct, m Distance across height of duct, m

a) Across the breadth of the duct b) Across the height of the duct

Figure 29: Scenario 2, distribution of gas 0.5 m and 2 m inside the HVAC inlet

Line integrals of gas concentration have again been calculated, at the locations shown in Figure

28, to model the output of cross-duct beam infra-red detectors. Values of these line integrals are

shown in Table 2. The average concentration of gas in the duct is 22% LEL.

Table 2: Scenario 2, Line integrals of gas concentration inside the HVAC duct

Location of line integral Length of Concentration

Orientation Distance inside line LEL.m %LEL

inlet (m) integral (m) (m) per m

Mid-plane across breadth 0.5 2.8 0.5 18

Mid-plane across breadth 2.0 2.8 0.45 16

Mid-plane across height 0.5 1.8 0.25 14

Mid-plane across height 2.0 1.8 0.33 19

The COV is given in Table 3, overleaf, showing that there is a very high degree of non-

uniformity in concentration distribution throughout the entire length of the duct.

30

Table 3: Scenario 2, COV at various distances inside the HVAC duct

Distance inside the HVAC duct, m COV, %

-0.5 136

0.5 104

2.0 93

10.0 72

23.0 67

31

4 DISTRIBUTION OF GAS INSIDE AN HVAC DUCT

4.1 OVERVIEW

The model geometry of Scenario 2 has been refined to include obstructions which represent

louvres and fire dampers at the HVAC inlet. All other aspects of the modelling are identical to

Scenario 2. This approach allows the effect of these obstructions on flow and dispersion in the

duct to be isolated and examined. This case is referred to as Scenario 3.

A set of 24 louvres is represented at the inlet plane of the duct. The geometry of these louvres

was based on a GDL Ltd datasheet. GDL are manufacturers and suppliers of air distribution

products. The complex ribbed geometry of the louvres has been simplified for modelling

purposes, whilst retaining their overall dimensions. Each modelled louvre is 10 mm thick,

inclined at approximately 45o, with a centre-to-centre spacing of 75 mm. The louvres occupy the

first 80 mm of the duct.

Fire dampers, assumed to be fully open, are simply represented as an array of rectangular

obstructions located 1 m inside the duct. Again, the complex geometry of each damper has been

simplified, but their overall dimensions and position has been retained. Data on fire dampers

was obtained from Flamgard Engineering Ltd, upon which this modelled geometry has been

based. The dampers are represented as being 50 mm thick and 300 mm wide, with partial-width

dampers of 195 mm being included at locations nearest to the duct wall and supporting

assembly. This supporting assembly has also been included in the modelled geometry.

The modelled geometry and mesh at the HVAC inlet is shown in Figure 30.

Figure 30: Scenario 3, detail of the geometry and mesh at the HVAC inlet

32

The mesh is as used in Scenario 2 except that it was selectively refined at the HVAC inlet and in

the HVAC duct. Layers of prismatic mesh elements were added at all of the duct walls, to better

represent the near-wall flow. The mesh refinement and the layers of inflated mesh elements can

be seen in Figure 30. An additional 186,000 mesh nodes were added at the HVAC inlet and

inside the duct, to give a mesh with a total of 564,000 nodes constructed from 2.9 million mesh

elements.

Transient simulations were again undertaken using high-order-accuracy spatial and temporal

discretisation schemes, with a time-step of 1 s. The simulations were started from an initial

solution obtained for Scenario 2 at time t + 20 s, i.e. intermediate between that shown in Figures

23a and 23b. The simulation was then advanced for a total of 180 s, at which time the trajectory

of the gas plume had reached the extremity of its oscillation and was being ingested into the

HVAC duct in just the same manner as Scenario 2 at the equivalent time of t + 200 s. Results

presented below are for time t + 200 s and can be compared directly with those for Scenario 2.

The flow outside of the HVAC inlet is essentially identical at this time for both Scenarios.

The convergence history was near-identical to that of Scenario 2, i.e. the solution is well-

converged.

4.2 RESULTS

The overall flow-field around the platform, as well as the trajectory of the gas plume, are

indistinguishable from Scenario 2 and therefore are not presented.

The distribution of gas on cross-sections at 0.5 m outside the HVAC inlet, 0.5 m, 2 m and 10 m

inside the HVAC duct, as well as at the end of the duct, is shown in Figure 31, together with the

concentration on a horizontal plane down the middle of the duct. This can be compared against

Figure 25, for Scenario 2. Close to the HVAC inlet the gas distribution is qualitatively similar in

both scenarios, with a very marked non-uniformity in concentration over duct cross-sections,

which occurs because gas is essentially ingested at just one side of the duct. However, further

down the duct the region of highest gas concentration occurs at this same side of the duct, in

contrast to that of Scenario 2 in which the highest concentrations occur on the opposite side of

the duct. This suggests that the louvres and dampers have all but eliminated the weakly swirling

flow implied in the results of Scenario 2. In both scenarios, well-mixed conditions are not

achieved at the duct outlet.

A more detailed view of the gas concentration distribution at the HVAC inlet is shown in Figure

32. Again, this can be compared to Scenario 2, Figure 26. Although the distributions are

qualitatively similar it would appear that the flow is no better mixed by the presence of the

louvres and dampers. In fact, somewhat higher concentrations are evident at 2 m inside the inlet

in Scenario 3, than in Scenario 2. Examination of the velocity field in the duct, Figures 33 and

34, reveals that this behaviour is caused by the louvres. These have the effect of deflecting the

flow towards the upper part of the duct, leading to an extensive region of separated recirculating

flow in the lower region of the duct. Gas is transported with the deflected flow, resulting in

higher concentrations at the roof of the duct compared to Scenario 2.

Whilst it may initially appear surprising that the louvres and dampers have had no substantial

effect on mixing, this result is consistent with the findings of the literature review: it is only

when large obstructions are present which are capable of generating turbulence at large length-

scales of the order of the duct size that mixing is likely to be significantly enhanced.

33

Figure 31: Scenario 3, gas concentration in % LEL in the HVAC duct

Figure 32: Scenario 3, gas concentration in % LEL at 0.5 m outside, 0.5 m and 2.0 m inside the HVAC inlet

34

Figure 33: Scenario 3, flow-field at the HVAC inlet, velocity vectors shown on a vertical plane which intersects the HVAC duct

Figure 34: Scenario 3, details of flow-field at the HVAC inlet, velocity vectors shown on a vertical plane which intersects the HVAC duct

The distribution of gas at four sample lines across the breadth and height of the HVAC duct,

0.5m and 2 m inside the inlet – see Figure 28 - is shown in Figure 35. There are no very

substantial differences from Scenario 2, see Figure 29, although in general the concentration

distributions are slightly skewed towards the gas plume side and upper face of the HVAC duct.

35

100 100

90 90

80 80

70 70

60

0.5m

inside

2.0m

inside

60

% LEL 50 % LEL 50

40 40

30 30

20 20

10 10

00

0.5m

inside

2.0m

inside

0 5 90 1 2. . .0 0 044. 09 45 7. .0 088. 13 8 3 70 1 3 4. . . .1 1 1 126. 17 2 6 1 6 1 5 0

7 9 0 2 3 5 6 8. . . . . . . .1 2 2 2 2 2 2

00. 090. 091. 082. 083. 074. 075. 066. 067. 058. 059. 140. 141. 132. 133. 124. 125. 116. 117. 108.0 0

Distance across breadth of duct, m Distance across height of duct, m

a) Across the breadth of the duct b) Across the height of the duct

Figure 35: Scenario 3, distribution of gas 0.5 m and 2m inside the HVAC inlet

Line integrals of gas concentration have also been calculated, at the locations shown in Figure

28, to model the output of cross-duct beam infra-red detectors. Values of these line integrals are

shown in Table 4. The average concentration of gas in the duct is 21% LEL.

Table 4: Scenario 3, Line integrals of gas concentration inside the HVAC duct

Location of line integral Length of Concentration

Orientation Distance inside line LEL.m %LEL

inlet (m) integral (m) (m) per m

Mid-plane across breadth 0.5 2.8 0.38 14

Mid-plane across breadth 2.0 2.8 0.19 6.6

Mid-plane across height 0.5 1.8 0.18 10

Mid-plane across height 2.0 1.8 0.17 9.3

The COV is given in Table 5 which shows that, as per Scenario 2, there is still a very high

degree of non-uniformity in concentration distribution throughout the entire length of the duct.

Indeed, the COV is generally slightly higher than in Scenario 2.

Table 5: Scenario 3, COV at various distances inside the HVAC duct

Distance inside the HVAC duct, m COV, %

-0.5 127

0.5 113

2.0 97

10.0 81

23.0 67

Although not shown, the computed non-dimensional near-wall distance in the duct, y+, is well

within acceptable bounds throughout almost the entire length of the HVAC duct. This indicates

that the wall boundary condition in the duct is being appropriately applied.

36

5 GAS RELEASE AND INGESTION INTO HVAC DUCTS FOR A REALISTIC SCENARIO

5.1 INTRODUCTION

Sections 3 and 4 strongly indicate that both high and low pressure releases can result in non­

uniform gas distributions at HVAC inlets. However, these scenarios are geometrically rather

idealised. Whilst simplified geometrical arrangements are very useful in indicating how a non­

uniform gas distribution could be present at an HVAC inlet, the increased mixing and

channelling due to the passage of a gas release through congested areas is not represented. It is

therefore important to model a more realistic scenario, based on a gas release from an existing

offshore platform, to establish whether similar behaviour occurs with increased geometrical

complexity. Here, the scenario considered is a high pressure release resulting from the failure of

a riser on the Brae Alpha platform, operated by Marathon Oil UK Ltd.

thIn fact on November 27 2004 a riser did fail on Brae Alpha, resulting in a high pressure gas

release and subsequent ingestion of gas into Hazardous Modules via the Hazardous HVAC inlet

duct. A subsequent investigation (OSD, 2006) reports that gas appeared to pass across only part

of the duct for several minutes, leading to a delay in confirmed detection and shutdown of the

HVAC system. There is no suggestion that the detectors were not operating correctly.

The aim of the present scenario is not to replicate this incident. This would be very time-

consuming, would require the collation of much detailed information and may be impractical.

The aim is instead to draw upon some elements of this platform and incident to devise a more

realistic scenario than the idealised configurations presented in Sections 3 and 4, so enabling

more general conclusions on the interaction of gas and HVAC inlets and ducts.

5.2 MODEL GEOMETRY AND COMPUTATIONAL DOMAIN

Figure 36 shows the modelled geometry and indicates the location of the release.

Figure 36: Modelled geometry

37

5.3

A very simplified representation of the Brae Alpha platform has been used as the basis of this

more realistic scenario. The overall dimensions of the platform and plans of the most significant

geometrical features at the module support frame (MSF) level were provided by Marathon Oil.

These were used to guide the size, shape and location of major obstructions to the wind field

and gas dispersion at MSF level. Note, that during the 2004 incident gas was released from the

top of a riser caisson and subsequently passed below and through the MSF level. Only those

platform support legs closest to the release have been modelled.

There are three HVAC ducts which penetrate the floor of the MSF: Hazardous Supply Plant

Room duct, Safe Supply Plant Room duct, Permanent Living Quarters (PLQ) duct. Figure 37

shows these ducts. The Hazardous Supply duct is essentially open and free from major

obstructions, apart from a walkway (not represented) at approximately mid-height of the duct.

The other two ducts have also been modelled as being free from obstructions. More detail on the

modelling of the HVAC ducts and the release geometry is given in Section 5.3.

Figure 37: Modelled geometry showing HVAC ducts

BOUNDARY CONDITIONS AND FLOW PHYSICS

As with the idealised scenarios, a large region of the atmosphere surrounding the platform was

modelled. Identical boundary conditions to those imposed in Scenario 1 were applied, except

that the wind speed is much higher; 12.3 m/s at the top of the platform (65 m above sea-level).

This appears to be a plausible value for wind speeds on the day of the release in 2004. The wind

direction was reported as being from 45q East of platform North (OSD, 2006). However, initial

CFD simulations with this wind direction showed that the gas cloud would disperse away from

the platform and that the gas concentration in the HVAC ducts would be negligible. Hence, an

alternative wind direction was chosen, 5q West of platform North, such that the gas cloud would

be directed across the three HVAC ducts. This wind direction was set by fixing the platform at

5q to the main axis of the computational domain. The majority of the results presented in

Section 5.5 relate to this wind direction. The computational domain is illustrated in Figure 38.

38

Figure 38: CFD domain and boundary conditions

The size and location of the HVAC ducts is illustrated in Figure 39. Volume flow rates through

the HVAC ducts were based on information provided by Marathon Oil:

x� Hazardous supply HVAC: 148 m3/s.

x� Safe supply HVAC: 63 m3/s.

x� PLQ HVAC: 32 m3/s.

39

a.) Plan view of MSF

c.) Source dimensions

b.) Source location

Figure 39: Plan view of the modelled MSF, also showing the modelled release source

40

These volume flow-rates were imposed as a uniform out-flow velocity on a downstream face

within the ducts. The out-flow boundary condition was artificially located at the ends of the

HVAC supply rooms in an attempt to reduce the influence of this assumed boundary condition

on the flow detail in the HVAC ducts. No attempt was made to model features such as fire

dampers which exist between the HVAC ducts and supply plant rooms. Figure 40 shows the

position of the out-flow boundary conditions.

Figure 40: Position of HVAC duct outlet boundary conditions

thThe November 27 2004 gas release on Brae Alpha resulted from the failure of a riser inside a

caisson. The pressurisation of the caisson resulted in the vertical displacement of one-half of a

two-part seal around an oil production riser. The resulting gas release from the caisson occurred

over the gap between the oil production riser and the caisson top on its South side, extending

over half its circumference.

The oil production riser extends 1.5 m vertically above the caisson top before bending 90˚ such

that it is horizontal and oriented in a Southerly direction. The release would probably be

directed up this short vertical section of the riser before impinging on the horizontal section

above. The modelling of this flow is likely to be very difficult since an extremely fine mesh

would be required to resolve the resulting boundary layer that would form along the riser. In

addition, such an approach might result in numerical instabilities due to the very high near-wall

velocity gradients in that region. Instead, a simplified geometry was used to capture the main

features of the jet impingement on the horizontal section of the pipe. This simplified geometry

is shown in Figure 41.

41

Figure 41: Simplified geometry at the release source

The location of the top of the caisson was very approximately 4 metres below the MSF level,

6.5 m from the Western perimeter and 5 m from the Northern perimeter of the platform. A plan

view in Figure 39 shows the overall dimensions of the modelled MSF - including the location

and dimensions of the HVAC ducts – together with the location of the release relative to the

North and West faces of the platform.

During the November 2004 incident the release from the top of the caisson occurred over an

approximately semi-circular slot, 1 mm wide and 480 mm long (OSD, 2006). The stagnation

conditions of gas in the caisson (assumed to be pure methane) are prescribed as follows:

P0 = 20 barg, T0 = 273K

Under these conditions the flow would be choked and the resulting jet under-expanded. As

explained in Section 3.2.2, it is not practical or necessary to model such an under-expanded

release from the immediate top of the caisson. Instead, an effective source is prescribed, at the

point at which the release has expanded to atmospheric pressure but is still at sonic velocity

(Mach 1). The methodology by which this effective sonic source can be obtained is provided in

Ivings et al (2003) and is reproduced here in summary form. It is based on an assumption of

isentropic flow through the release slot, followed by expansion to sonic velocity but with the

same mass flow rate and static temperature as the original under-expanded release.

42

5.4

The following isentropic flow expressions relate conditions at the release point (subscript ‘1’) to

those from the known stagnation conditions (subscript ‘0’):

J 2 T1 J �1T T �J �1 �, P P � � , V1 JRT11 0 1 0 T0

�the mass release rate, m , is given by;

m� U V A 1 1 1

in which a discharge coefficient of unity has been assumed. U1 is given by:

U P / RT1 1 1

Ewan & Moodie (1986) have shown that an under-expanded jet can be represented by an

‘equivalent’ jet at atmospheric pressure. The equivalent, or effective, jet is assumed to be sonic

with a mass flow rate and temperature the same as the original jet, but with a new area based on

expansion of the jet to atmospheric pressure. The following effective source expressions apply:

V V1 , P T T1 , A A P / P2 2 Pambient , 2 2 1 1 ambient

where subscript ‘2’ refers to the effective sonic source.

The expressions above can be used to obtain the following boundary conditions for the effective

sonic source, assuming that the rupture in the caisson top is shaped as a semi-circular slot 1 mm

wide and 480 mm long:

V2 = 400 m/s, T2 = 237K, P2 = 101325 Pa, A2 = 5324.4 mm2

Figure 41 shows the shape and dimensions of the effective sonic source, modelled as a semi­

circular slot with the same inner radius as the release source but with a width of 10.1 mm rather

than 1 mm.

All other aspects of the modelling of the flow physics are the same as that for Scenario 1,

Section 3.2.3

MESH, SOLUTION NUMERICS AND CONVERGENCE

An unstructured mesh was generated which was refined near the source and in other regions

where high gradients of velocity and mass fraction were expected. A total of 665,000 mesh

nodes were used. Figure 42 shows the mesh on the surface of the caisson and Figure 43 shows

the mesh in a plane through the expected trajectory of the cloud.

43

a) Caisson top and source b) Close up of source

Figure 42: Mesh resolution at the source

Figure 43: Mesh resolution in a plane through the expected trajectory of the cloud

Time-dependent simulations were undertaken with a time-step of 1 s. High-order-accuracy

spatial and temporal discretisation schemes were used. The results in the region of the HVAC

ducts were essentially steady in time, although the flow over the platform was unsteady.

Equation residuals reduced by two orders of magnitude to very small rms values and the global

imbalances of transported variables were less than 0.1% of reference values, indicating good

convergence.

44

5.5 RESULTS

5.5.1 Overall flow field and dispersion

Figure 44 shows an iso-surface of 10% LEL for an initial wind direction of 45q East of platform

North. The figure shows that the gas cloud disperses away from the platform.

Figure 44: Iso-surface of 10% LEL for a wind direction of 45q East of platform North

Figure 45 shows an iso-surface of 10% LEL for a wind direction of 5q West of platform North.

All of the results which follow correspond to this wind direction. Gas can be seen to disperse

through the western half of the platform at the MSF level. In this figure, velocity vectors in a

vertical plane are also shown to give an indication of the flow around the platform (the

maximum speed shown is 15m/s). As with the more idealised scenarios in Sections 3 and 4,

there is a large wake region in the lee of the platform.

A more detailed view of the 10% LEL iso-surface is shown in Figure 46. The gas jet is very

quickly deflected in the direction of the prevailing wind, meaning that there is no impingement

on the underside of the module which is located immediately above the MSF level. As the gas

travels through the MSF it is channelled around obstructions which generate turbulence in their

wake. Qualitatively it would appear that gas dispersion is enhanced by the presence of

congestion in the MSF.

Figure 46 also indicates that gas spreads underneath the MSF level and is ingested by the

HVAC ducts. This is confirmed in Figure 47, which shows contours of gas concentration in a

horizontal plane 23.5 m above sea level (just above the MSF floor). Gas is visible in all of the

HVAC ducts, although at low concentrations in the Hazardous and Safe Supply ducts. However,

in the PLQ HVAC duct, gas at a concentration of greater than 10% LEL is visible along its

entire length. Velocity vectors on this plane are also shown to indicate the flow speed through

the MSF (the maximum speed shown is 15 m/s).

45

Figure 45: Iso-surface of 10% LEL for a wind direction of 5q West of platform North

Figure 46: Detail of 10% LEL iso-surface

Figure 47: Gas concentration contours in % LEL on a horizontal plane at 23.5 m above sea-level – just above the MSF floor.

46

5.5.2 Gas distribution in the Hazardous supply HVAC duct

Figure 48 shows contours of gas concentration across the Hazardous supply HVAC duct at

approximately the height at which gas detectors were located during the 2004 incident on Brae

Alpha. The average concentration of gas in the duct is 5.4% LEL. Whilst this is a relatively low

average concentration, gas is distributed very non-uniformly: peak concentrations are close to

10%, but with a concentration minimum of less than 1%. The COV for gas at this plane is 39%,

which as indicated in Appendix 1 represents a significant variation in values over the cross-

section.

Figure 48: Gas concentration contours (% LEL) across the hazardous HVAC duct, 23.5 m above sea-level and just above the MSF floor

Figure 49 shows locations at which point and line integrals of gas concentration in the

Hazardous supply duct have been obtained from post-processing of the CFD results. In effect,

point detectors have been located 0.6 m and 0.4 m from the sides of the duct (as shown in the

diagram), and placed at 23.5 m above sea level. In addition, horizontal lines along each diagonal

at 23.5 m are shown, along which integrals of gas concentration have been computed to

replicate the output of cross-duct beam infra-red detectors.

a) Plan view b) View from south east

Figure 49: Locations of gas detectors in Hazardous supply HVAC duct and gas concentrations at selected points

47

The gas distribution along lines 1 and 2 (Figure 49) is shown in Figure 50. Values of the line

integrals along lines 1 and 2 are shown in Table 6. %

LE

L

9

8

7

6

5

4

3

2

1

0

0 1 2 3 4 5 6 7

Line 1

Line 2

Distance along line (m)

Figure 50: Gas concentration (% LEL) in Hazardous supply HVAC duct, along lines 1 and 2

Table 6: Realistic scenario, line integrals of gas concentration inside the Hazardous supply HVAC duct

Location of line

integral

Length of line

integral (m)

Concentration

LEL.m

(m)

%LEL

per m

Line 1 7.42 0.44 5.9

Line 2 7.42 0.33 4.4

The values of gas concentration at the four point detectors (Figure 49) range from 3.1% to 8.1%

LEL, i.e. approximately a factor of 2.5 variation between minimum and maximum values. The

two line integrals, replicating the effect of cross-duct beam detectors, show much less difference

in their values and both are close to the actual average concentration of gas in the duct (5.4%

LEL).

Also of interest is the flow-field in the Hazardous supply HVAC duct, Figure 51. Regions of

flow separation and recirculation are evident close to the HVAC inlet.

Figure 51: Flow-field in the Hazardous supply HVAC duct

48

During the course of this work an alternative slightly simpler geometry was used, in which

certain blockages which represent plenums connecting the HVAC supply rooms to the module

above were not included. The results of these simulations were somewhat different from those

presented above and give an indication of the effect of relatively small changes to the modelled

geometry. Figure 52 shows the two model geometries, highlighting the plenums above the

HVAC supply rooms.

a) Geometry 1 b) Geometry 2 (as described in Section 5.2) (not including blockage due to plenums)

Figure 52: Alternative model geometry

The results for Geometry 2 showed more transient behaviour in the duct than those for

Geometry 1. Gas concentrations monitored at a number of locations in the duct suggest that the

flow in this region is time-dependent with a time period of approximately 10 s. Figure 53 shows

contours of concentration across the Hazardous supply HVAC duct for Geometry 2 at different

times.

a) t b) t + 5 s c) t + 10 s

Figure 53: Gas concentration contours (% LEL) across the hazardous HVAC duct for Geometry 2, at various times.

At time t, the average concentration across the duct at the detector height is 7.8% LEL, whilst

the COV for gas is 37%. Very similar values are obtained at later times.

49

Figure 54 shows the gas concentration along lines 1 and 2, for both Geometry 1 and 2 (at time

t). Values of the line integrals along lines 1 and 2 for Geometry 2 are shown in Table 7. %

LE

L

12

10

8

6

4

2

0

0 1 2 3 4 5 6 7

Geometry 1 - Line 1

Geometry 1 - Line 2

Geometry 2 - Line 1

Geometry 2 - Line 2

Distance along line (m)

Figure 54: Gas concentration (% LEL) in Hazardous supply HVAC duct, along lines 1 and 2

Table 7: Realistic scenario, Geometry 2, line integrals of gas concentration inside the Hazardous supply HVAC duct

Location of line

integral

Length of line

integral (m)

Concentration

LEL.m

(m)

%LEL

per m

Line 1 7.42 0.62 8.3

Line 2 7.42 0.47 6.3

Again, the two line integrals are both close to the actual average concentration of gas in the duct

(7.8% LEL).

5.5.3 Gas distribution in the PLQ HVAC duct

The PLQ HVAC duct is far smaller than the Hazardous supply HVAC duct, just 1.5 m square,

compared to 6.25 m x 4 m. It might initially be expected that, since the PLQ duct is smaller,

more uniformity in the gas concentration would be seen across the duct. In fact this is not the

case. Figure 55, overleaf, shows contours of gas concentration across the PLQ HVAC duct for

Geometry 1. There is a very marked variation in gas concentration across the duct, which

persists up the duct. The COV 2 m inside the duct, is 43%. The peak gas concentration is greater

than in the Hazardous supply duct, with an average value of 16.4% LEL.

For Geometry 2, the situation is qualitatively similar, except that the gas concentration is overall

of lower magnitude; an average of 4.1%. However, the COV at 2 m inside the duct remains

high, 56%, indicating that the flow is far from being well-mixed.

50

a) 2 m inside the duct b) 4 m inside the duct c) 6 m inside the duct

Figure 55: Gas concentration contours (% LEL) across the PLQ duct, Geometry 1. Note that platform North is to the left, East to the top of the figures, etc.

Figure 56 shows the location of horizontal sample lines across the PLQ duct, 2 m inside the duct

inlet. The concentration distribution along these sample lines is shown overleaf, in Figure 57.

Figure 56: Locations of sample lines 2 m inside the PLQ duct

51

Gas concentrations in PLQ HVAC duct %

LE

L

30

25

20

15

10

5

0

-

-

With plenums - Line 1

With plenums - Line 2

No plenums Line 1

No plenums Line 2

0 0.2 0.4 0.6 0.8 1 1.2 1.4

Distance along line (m)

Figure 54: Gas concentration (% LEL) in PLQ HVAC duct, along lines 1 and 2

Line integrals of gas concentration have been computed along lines 1 and 2 in the PLQ HVAC

duct, for Geometry 1 and 2. The results are given in Table 8.

Table 8: Realistic scenario, line integrals of gas concentration inside the PLQ HVAC duct

Geometry Location of line

integral

Length of line

integral (m)

Concentration

LEL.m

(m)

%LEL

per m

1 Line 1 1.5 0.24 16.1

1 Line 2 1.5 0.25 16.8

2 Line 1 1.5 0.06 4.0

2 Line 2 1.5 0.06 4.1

The two line integrals (replicating the effect of cross-duct beam detectors) for each geometry

give values which are very close to the actual average concentration of gas in the duct: 16.4%

and 4.1% LEL for Geometry 1 & 2, respectively. The line-integrated concentration is almost the

same for lines 1 and 2, which simply reflects the fact that the peak concentration gradient is

along a diagonal of the duct (the sample lines are across the width of the duct).

52

6.1

6 DISCUSSION

REVIEW AND DISCUSSION OF RESULTS

In every Scenario the most significant feature of the CFD results is the presence of a large

variation in gas concentration across HVAC inlets. This characteristic of the flow is consistent

with the concentration field expected of a jet release at a large distance from the release source

(as described in Section 3.1). Whilst the behaviour of a buoyant plume is different in detail it is

similar in essence, therefore the low momentum release of Scenarios 2 and 3 would also be

expected to show a significant variation in concentration across the plume at a large distance

from the gas source. In this respect the CFD results are unsurprising and broadly consistent with

the established fluid mechanics of jets and plumes.

In every Scenario in which gas enters a straight HVAC duct, the rate at which the variation in

gas concentration over the duct cross-section reduces with distance along the duct is very slow.

This outcome of the CFD modelling is consistent with the body of literature reviewed in Section

2.2. The variation in gas concentration over a duct cross-section can be expressed as a

Coefficient of Variation (COV). The computed rate of decay of COV with distance along a

straight duct is qualitatively similar to that from this same body of literature, in that the COV

reduces slowly with distance along a straight duct.

Furthermore, the computed effect of louvres and fire dampers on mixing over a duct cross-

section is seen to be very small. This is evident if the values for COV are compared in Tables 3

and 4. Again, the literature reviewed in Section 2.2 strongly suggests that only purpose-

designed mixing elements, or bends in a duct, are effective in substantially enhancing mixing.

So, whilst there necessarily remains some uncertainty in the ability of the CFD modelling to

reproduce the fine details of turbulent mixing in the wake of obstacles such as louvres and

grilles (discussed in more detail at the end of this Section), overall this aspect of the CFD results

is again consistent with the findings of the literature review.

In summary, there is no substantial reason to doubt that the CFD results broadly reflect the gross

flow behaviour in and around HVAC ducts, at least for the simplified release sources and

geometries considered in this study.

One further aspect of the CFD results should be mentioned: that the presence of a large variation

in gas concentration across a HVAC duct appears to occur not only for very large duct cross-

sections with a high aspect ratio (width:breadth), but also for more modestly-sized square ducts.

The range of duct sizes examined in this study varies from a maximum of 6.25 m x 4 m, through

2.8 m x 1.8 m and 2.84 m x 1.34 m, to a minimum of 1.5 m x 1.5 m. In every case the COV at,

or close to, the duct inlet is quite large, although smaller values of COV are seen for the realistic

scenario than for the simplified cases. However, with ducts of smaller dimensions than those

examined here it can be expected that the variation in gas concentration would also be smaller,

although possibly still significant.

In principle the CFD results in this study can be used to draw some general conclusions on the

likely effectiveness of differing flammable gas detection strategies for HVAC inlets and ducts.

However, it should not be forgotten that CFD modelling is an approximation to reality. Also

there are other factors, in addition to the findings of this report, that are likely to influence the

choice of gas detection strategy such as cost, maintenance, reliability, etc. Nevertheless, with

these caveats in mind, it is possible to draw some conclusions. In due course, experimental

validation via large-scale physical tests using commercial detectors would be very useful in

reducing remaining uncertainties.

53

Hence, for large ducts (major dimension >> 1 m), it will be difficult to always ensure that peaks

in gas concentration are not missed by point-based detection systems, especially if standard

catalytic or infra-red point detectors are used and typically set to alarm at 20% LEL. For

example, examination of the results of Scenario 1, Figure 13a, at a location 1 m inside the duct,

shows that gas concentrations are less than 20% LEL over about half of the cross-section of the

duct, even though the average concentration of gas in the duct is 27% LEL, i.e. well-above a

20% alarm level. For Scenario 3, Figure 32, at 0.5 m or 2 m inside the duct, gas concentrations

are below 20% LEL over the vast majority of the duct cross-section, although the average

concentration of gas in the duct is just greater than 20% LEL. If using a point-based detection

system some mitigation for this situation would be possible if multiple detectors were used and

located towards the extremities of the duct cross-section. However, the CFD results indicate that

for a large duct up to four point detectors could be needed for systems which alarm upon two

positive detections. Additional mitigation could be gained by the use of lower alarm settings.

HSE (2006) recommend that the feasibility of alarm levels of ~10% be explored. The use of

extended path point infra red detectors could also be beneficial, since these appear to have much

higher sensitivity - with quoted minimum alarm levels of 5% - although whether they operate

reliably in an offshore environment is currently uncertain. These detector types also have the

benefit that they cover a path length of the order of 1 m, so providing greater coverage than

standard point detectors. For these reasons extended path point infra-red detectors could also be

well-suited to the monitoring of small ducts.

For large ducts, cross-duct beam infra-red detector systems could potentially be beneficial. This

is partly because the path length is relatively long - thereby maximising absorption of the beam

- and partly because they are less likely than a point detector to ‘miss’ a peak in concentration

due to their coverage of a duct. However, there is some uncertainty as to whether they have

sufficiently high sensitivity. Line integrals of gas concentration have been computed from the

CFD results in this study to replicate the output of cross-duct beam infra-red detectors. For

Scenarios 1 to 3, in which the average concentration of gas in the duct was always greater than

20%, line integrals which were equivalent to maximum and minimum output of 0.78 LEL.m to

0.17 LEL.m were obtained. It is questionable whether output as low as 0.17 LEL.m would meet

minimum alarm levels for cross-duct beam infra-red systems. The CFD results in the present

study indicate that the variation of gas concentration over a duct cross-section can occur in a

simple fashion as a gradient along the major axis of a duct (Scenario 1, Figure 13) or along its

diagonal (realistic scenario PLQ duct, Figure 55), or, in a more complex manner (Scenario 3,

Figure 32; realistic scenario hazardous supply duct, Figures 48 & 53). Hence, where cross-duct

beam infra-red detector systems are employed, two beams would be recommended to reduce the

likelihood that a peak in gas concentration could be missed. The beams would probably best be

arranged so as to be approximately orthogonal, either along the diagonals of a duct, or across the

width & breadth of a duct.

Aspirated systems could also potentially be beneficial for gas detection in a large duct, since in

principle they could also provide good coverage of a duct. If such systems are configured as a

line of sample holes, then two lines would be recommended, arranged approximately

orthogonally, for the same reasons as given above. There is again some uncertainty as to

whether the sensitivity of the detector at the end of an aspirated probe system would be

sufficiently high.

The CFD modelling in this study indicates that detector types which are based on sampling

along a line (either cross-duct beam infra-red, extended path point infra-red, or aspirated line

sample probes) tend to give an average concentration which is more approximately

representative of the average concentration of gas in a duct (see Table 1, Scenario 1; Tables 6, 7

54

and 8, realistic scenario) than might be obtained with a limited number of point detectors.

However, this is not always the case (see Table 4, Scenario 3).

On the basis of the present CFD results, and the findings of the literature review, there would

appear to be no significant increase to be gained in the likelihood of gas detection as a

consequence of siting detectors a long way down an HVAC duct - in the expectation that well-

mixed conditions will exist. Any non-uniform distribution of gas at the inlet to an HVAC duct is

very likely to persist a long way downstream inside the duct, at least as far as any fans, unless

purpose-designed mixing elements or bends are also present in the duct. This behaviour is

illustrated in Figures 15 and 31. In addition, there are other practical considerations raised by

such a strategy, such as the fact that a larger volume of gas would be ingested prior to closure of

dampers.

The CFD results indicate that the distribution of gas immediately outside of an HVAC inlet is

quite similar to that which exists a short distance inside the inlet. See Figures 13 and 32. The

literature review also suggests that unless purpose-designed mixing elements or bends are

present close to an HVAC inlet then the COV reduces slowly with distance along a duct.

Therefore, there would appear to be no significant benefit to be gained from siting detectors

inside an HVAC duct compared to locating detectors immediately outside the duct. Practical

considerations, such as ease of access to detectors for maintenance purposes, or sheltering of

detectors from the weather, could be of more importance.

The CFD results, supported by the fluid mechanics of flow around a bluff body, show that the

flow around an offshore platform is likely to be time-dependent as a consequence of flow

separation in the wake of the platform. In addition, the flow around other bluff objects on a

platform can lead to transient flow in their wake. The time-scales for such transient flow

features can cover a large range; from minutes for flow over the entire platform (see Section

3.2.5) to possibly tens of seconds or less (an indication of time-scales of the order of 10 s is

evident in the simulation of the realistic scenario, see Section 5.5.2, Figure 53). The time-

dependency of the flow and gas dispersion, and its implications on gas detector response, is not

investigated further in this study. This is in part because the simulations would probably need to

be repeated using a much more sophisticated approach to the modelling of turbulence before

firm conclusions could be reached. This would be very time-consuming and possibly

impractical. However, the likely existence of time-dependency of flow and gas dispersion on a

range of time-scales adds weight to recommendations that detection limits should be as low as

reasonably practical.

As mentioned above, it should not be forgotten that CFD modelling is an approximation to

reality. Uncertainty in the CFD results will arise primarily as a consequence of approximations

and assumptions which are implicit in the turbulence modelling. Errors in the CFD results will

arise primarily as a result of mesh resolution, although the meshes used in this study are refined

in regions of steep gradients. Also, high-order-accuracy numerical discretisation schemes are

employed. Nevertheless, numerically-generated diffusion is likely to be present. However, this

artificial diffusion will generally lead to a reduction in gas concentration gradients, such that in

reality even more significant variations in gas concentration could be expected.

Perhaps of more significance is the fact that only a few scenarios have been investigated, for a

limited range of conditions, and that these scenarios are all somewhat idealised. Even so, the

CFD results are consistent with the outcomes of the literature review and the known behaviour

of flow and gas dispersion in and around ducts. Nevertheless, the recommendations which

follow in Section 6.2, on gas detection strategies, are necessarily labelled as initial

recommendations. Experimental validation, via large-scale physical tests using commercial

detectors, is therefore recommended as a next step to reduce remaining uncertainties.

55

6.2 INITIAL RECOMMENDATIONS FOR GAS DETECTION STRATEGIES

Initial recommendations on flammable gas detection strategies for HVAC ducts are listed

below. These initial recommendations are based on the findings of the scoping study by Walsh

et al (2005), a review of the literature on flow and dispersion in ducts which is presented in

Section 2, as well as the outcomes of the CFD modelling reported in Sections 3 to 5.

(1) Detector alarm levels should be set as low as reasonably practical: 10% LEL or less.

Justification: The possibility of significant non-uniformity in the distribution of gas which is

ingested into an HVAC duct has been demonstrated by CFD modelling and is also indicated by

theoretical considerations. The literature review has highlighted that, in the absence of purpose-

designed mixing elements, an initial non-uniform distribution of gas in a duct requires a very

long downstream distance before uniformity is approached. HVAC detectors are now available

with a concentration range of 0 to 20% LEL and quoted minimum alarm levels of 5% LEL

(Walsh et al, 2005). To reduce the likelihood that detectors will ‘miss’ a non-uniform

distribution of gas ingested into an HVAC duct, it is recommended that alarm levels be set no

greater than 10% LEL. HSE have already provided information which states that although it is

common practice for gas detector alarm levels to be set at 20% LEL, duty holders should

explore the feasibility of reducing this alarm level to ~ 10% LEL (HSE, 2006). The low alarm

levels have to be balanced with the minimisation of false alarms, which arise from detector drift

and transient operational activities.

(2) Point catalytic, point infra-red, extended path point infra-red, cross-duct beam infra­

red and aspirated point detector systems all have the potential to be effective in detecting

non-uniform distributions of flammable gas in and around HVAC ducts provided that

their sensitivity is sufficiently high (low detection limit) and that due regard is given to the

possibility that gas will be distributed non-uniformly.

Justification: Walsh et al (2005) reviewed a number of different detector types. A range of

detector types are available with high sensitivity, although there is some question as to whether

all of the point and cross-duct beam infra-red systems have sufficiently-high sensitivity. Each

detector type has its benefits and limitations, demanding differing siting requirements to ensure

that a non-uniform distribution of gas is not ‘missed’.

(3) Extended path point infra-red detector systems currently appear to offer the greatest

sensitivity, but multiple detectors should be used and sited so as to anticipate non-uniform

mixing.

Justification: Walsh et al (2005) show that extended path point infra-red detector systems are

available with a concentration range of 0 to 20% LEL and quoted minimum alarm levels of 5%

LEL. In addition, specially designed point catalytic (e.g. for gas turbine enclosures) detectors

are available with a similar sensitivity (typical catalytic detectors are usually not as sensitive or

as reliable as infra-red types). However, whether either type operate reliably in the field with

minimum false alarms is currently uncertain. The CFD modelling demonstrates that there is a

possibility of significant non-uniformity in the distribution of gas inside and around an HVAC

inlet. The literature review indicates that this non-uniformity will reduce slowly with distance

downstream in a duct. It is difficult to provide firm guidance on how many point or extended

path detectors should be used since this depends on the size and shape of a duct. However, there

should be good coverage of the cross-section of the duct. For large ducts this may mean that

four detectors would be needed for systems which alarm upon two positive detections.

56

(4) Cross-duct beam infra-red, extended path or aspirated point detector systems should

be based on two approximately orthogonal beams or lines of aspirated point probes.

Justification: As stated above, the CFD modelling demonstrates that there is a possibility of

significant non-uniformity in the distribution of gas inside and around an HVAC inlet whilst the

literature review indicates that this non-uniformity will reduce slowly with distance downstream

in a duct. For these reasons there should be good coverage of the cross-section of a duct. This

can be achieved by two infra-red beams, either as open-path cross duct or extended path point

infra-red, or lines of aspirated point probes, arranged approximately orthogonally.

(5) No significant benefit can be expected to be gained from siting detectors inside an

HVAC duct compared to locating them immediately outside the HVAC inlet.

Justification: The literature review indicates that effective mixing in a duct is only achieved if

large-scale turbulent eddies are introduced via purpose-designed mixing elements or bends. The

CFD modelling indicates that louvres at the inlet to a duct or fire/gas dampers inside a duct will

not, in themselves, be sufficient to rapidly ensure that well-mixed conditions exist in a duct. In

fact CFD simulations indicate that a non-uniform distribution of gas immediately outside of an

HVAC inlet persists downstream from these geometrical elements. The literature review also

indicates that grilles at the entrance to HVAC ducts are unlikely to significantly enhance

mixing.

(6) In the absence of purpose-designed mixing elements or a series of bends upstream from

gas detectors no significant benefit is to be gained from siting detectors a significant

distance downstream from an HVAC inlet.

Justification: The literature review and the CFD modelling strongly indicate that, for a straight

duct, well-mixed conditions are only achieved a very long way downstream from an HVAC

inlet.

(7) Mixing elements have the potential to reduce any non-uniformity in the distribution of

gas in a duct but their effectiveness should be proven by physical tests.

Justification: This is supported by the literature review. It should also be noted that mixing

elements will result in an additional resistance to flow in a duct and that the resulting pressure

drop may be significant.

The above recommendations are based on evidence from CFD modelling and the published

literature. However, CFD modelling has inherent uncertainties and it is not certain that the

findings from the literature are always directly relevant. Therefore it is strongly recommended

that the above initial recommendations are substantiated by physical trials, using real detectors.

57

7 CONCLUSIONS

The aim of this study has been to undertake CFD modelling to provide a basis for advice to

inspectors and the industry on the effectiveness of flammable gas detection strategies for

offshore HVAC ducts. CFD simulations of a high and low pressure gas release have been

undertaken for idealised representations of an offshore platform, as well as a high pressure

release for a more realistic geometry based on the Brae Alpha platform. In parallel with this

modelling work a literature review has been carried out to build on the scoping study of Walsh

et al (2005).

The most significant feature of the CFD results is that in all cases the distribution of gas at

HVAC inlets was non-uniform: large variations in gas concentration were present over the

cross-section of the modelled HVAC inlets. This result was found to be consistent with the

theoretical behaviour of a high pressure gas release. Neither is there a substantial reason to

believe that it is not also consistent with the behaviour of a low pressure gas release. It means

that there is the potential for a gas release to be ‘missed’ by detection systems unless this non-

uniformity in gas concentration is anticipated in the selection and siting of gas detectors at

HVAC inlets.

The CFD results also show that a variation in gas concentration over a duct cross-section only

reduces slowly with distance along a straight duct. This finding was found to be consistent with

a body of relevant literature stemming from the sampling of gas distributions in the exhaust

ducts of nuclear stacks. This literature highlights that purpose-designed mixing elements and

bends in a duct can be effective in creating well-mixed conditions in a duct, but at the cost of

increased pressure drop. It also suggests that relatively small-scale obstructions such as louvres

and fire dampers are unlikely to significantly enhance mixing. This was borne out by CFD

modelling of such obstructions. The implications of the modelling work, substantiated by the

literature, are that in the absence of purpose-designed mixing elements or a series of bends

upstream from gas detectors no significant benefit would be gained from siting detectors a

significant distance downstream from an HVAC inlet. Also, no significant benefit can be

expected to be gained from siting detectors inside an HVAC duct compared to locating them

immediately outside the HVAC inlet.

The CFD results were post-processed to gain some further insight into the likely effectiveness

of point and infra-red beam or aspirated detector systems for HVAC ducts. The resulting output

has been combined with the findings of the scoping study by Walsh et al (2005), and the

outcomes of the literature review and CFD modelling as reported above, to provide a basis for

the following initial recommendations on flammable gas detection strategies:

x� Detector alarm levels should be set as low as reasonably practical: 10% LEL or less.

x� Point catalytic, point infra-red, extended path point infra-red, cross-duct beam infra-red

and aspirated point detector systems all have the potential to be effective in detecting

non-uniform distributions of flammable gas in and around HVAC ducts provided that

their sensitivity is sufficiently high (low detection limit) and that due regard is given to

the possibility that gas will be distributed non-uniformly.

x� Extended path point infra-red detector systems currently appear to offer the greatest

sensitivity, but multiple detectors should be used and sited so as to anticipate non­

uniform mixing.

58

7.1

x� Cross-duct beam infra-red, extended path or aspirated point detector systems should be

based on two approximately orthogonal beams or lines of aspirated point probes.

x� No significant benefit can be expected to be gained from siting detectors inside an

HVAC duct compared to locating them immediately outside the HVAC inlet.

x� In the absence of purpose-designed mixing elements or a series of bends upstream from

gas detectors no significant benefit is to be gained from siting detectors a significant

distance downstream from an HVAC inlet.

x� Mixing elements have the potential to reduce any non-uniformity in the distribution of

gas in a duct but their effectiveness should be proven by physical tests.

RECOMMENDATIONS FOR FURTHER WORK

The CFD modelling in this study is consistent with the findings of the literature review and the

theoretical behaviour of high and low pressure gas releases. However, CFD modelling is an

approximation to reality. Also, the range of scenarios which have been investigated using CFD

are relatively small, as well as being somewhat idealised. In addition there are other factors,

such as the reliability of detectors, which have not been addressed in this study. Therefore it is

recommended that large-scale physical tests, using commercial detectors, are undertaken to

validate the findings of this study and to support and refine the initial recommendations above.

It would also be useful to clarify the minimum practical alarm levels for a range of commercial

detector types, especially cross-duct beam infra-red systems.

59

8 APPENDIX 1: EXAMPLES OF COEFFICIENT OF VARIATION

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

a) 10% b) 10%

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

c) 20% d) 20%

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

e) 50% f) 50%

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

0

20

40

60

80

100

80-100

60-80

40-60

20-40

0-20

g) 100% h) 100%

Figure A1: Example distributions of a scalar on a regular 10 x 10 grid with Coefficient of Variation ranging from 10% to 100%.

60

9 REFERENCES

Anand N K, McFarland A R, Rajagopal K R (2003) Gas mixing for achieving suitable

conditions for single point aerosol sampling in a straight tube: experimental and numerical

results, Health Physics, Vol. 84, No. 1, pp 82-91.

ANSI/HPS N13.1-1999 (1999) Guide to sampling airborne radioactive materials in stacks and

ducts, American National Standards Institute/Health Physics Society.

BS EN 15138:2001 (2001) Offshore production installations – HVAC, British Standards

Institute, London.

Chin W, Lam S (2004) Guidelines provide offshore HVAC design criteria, Oil & Gas Journal,

Vol. 102, Part 28, pp 47-50, July 26 2004.

Demuren A O, Rodi W (1984) Calculation of turbulence-driven secondary motion in non-

circular ducts, J. Fluid Mechanics, Vol. 140, pp 189-222.

Ewan B C R, Moodie K (1986) Structure and velocity measurements in under-expanded jets,

Combustion Science & Technology, Vol. 45, pp 275-288.

Gatski T B, Rumsey C L (2002) Linear and non-linear eddy visocisity models, Closure

Strategies for Turbulent and Transitional Flows, Eds. Launder B E & Sandham N D, Cambridge

University Press, pp 9-46.

Gatski T B, Speziale C G (1993) On explicit algebraic stress models for complex turbulent

flows, J. Fluid Mechanics, Vol. 254, pp 59–78.

Gessner F B, Po J K, Emery A F (1977) Measurements of developing turbulent flow in a square

duct, Proc. Symp. on Turbulent Shear Flows, April 18-20 1977, Pennsylvania State University,

USA, pp 9.1-9.10.

Hampl V, Niemela R, Shulman S, Bartley D L (1986) Use of tracer gas technique for industrial

exhaust hood efficiency evaluation – where to sample? American Industrial Hygiene Assoc J.,

Vol. 45, Part 5, pp 281-287.

Hargreaves D M, Wright N G (2007) On the use of the k-H model in commercial CFD software

to model the neutral atmospheric boundary layer, J. Wind Engineering and Industrial

Aerodynamics, Vol. 95, pp 355-369.

Hinze J O (1975) Turbulence, McGraw-Hill Book Company.

HSE (1993) Offshore detector siting criterion investigation of detector spacing by Lloyds

Register, Offshore Technology Report OTO 93002, HSE, London.

HSE (2003) HSE Guidance Note PM84 - Control of safety risks at gas turbines used for power

generation, HSE Books. Sudbury.

HSE (2006) Testing regime for offshore TR-HVAC fire dampers & TR pressurisation

requirements, Offshore Information Sheet No. 1/2006.

61

Ivings M J, Azhar M, Carey C, Lea C, Ledin S et al (2003) Outstanding safety questions

concerning the use of gas turbines for power generation: Final report on the CFD modelling

programme of work, Health and Safety Laboratory Report CM/03/08.

Ivings M, Kelsey A, Hemingway MA, Walsh P (2002) Guidance note on the placement of

flammable gas detectors at high and low levels, Health and Safety Laboratory Report

FMS/02/01.

Kelsey A, Hemingway MA, Walsh PT, Connolly S (2000), Evaluation of flammable gas

detector networks based on experimental simulations of offshore, high pressure gas releases,

Trans. IChemE, 80 Part B, 78-86.

Kelsey A, Ivings MJ, Hemingway MA, Walsh PT, Connolly S (2005) Sensitivity studies of

offshore gas detector networks based on experimental simulations of high pressure gas releases,

Trans. IChemE. Process Safety and Environmental Protection, 83(B3), 262-269.

Klein A (1981) Review: Turbulent developing pipe flow, J. Fluids Engineering, Trans ASME,

Vol 103, pp 243- 249.

Laws E M, Lim E H, Livesey J L (1979) Turbulent pipe flows in development and decay, Proc nd2 Symp on Turbulent Shear Flows, July 2-4 1979, Imperial College, London, pp 4.6-4.11.

Laws E M, Livesey J L (1978) Flow through screens, Ann. Review of Fluid Mechanics, Vol.

10, pp 247-266.

Launder B E, Spalding D B (1972) Mathematical models of turbulence, Academic Press.

McFarland A R, Anand N K, Ortiz C A, Gupta R, Chandra S, McManigle A P (1999a) A

generic mixing system for achieving conditions suitable for single point representative effluent

air sampling, Health Physics, Vol. 76, No. 1, pp 17-26.

McFarland A R, Gupta R, Anand N K (1999b) Suitability of air sampling locations downstream

of bends and static mixing elements, Health Physics, Vol. 77, No. 6, pp 703-712.

Melling A, Whitelaw J H (1976) Turbulent flow in a rectangular duct, J. Fluid Mechanics, Vol.

78, Part 2, pp 289-315.

Micropack (2000) Micropack Performance Targets for BP Northern Everest.

OSD (2006) Brae A caisson failure, 27th Nov 2004, Draft report supplied by Offshore Safety

Division.

Rodgers J C, Fairchild C I, Wood G O, Ortiz C A, Muyshondt A, McFarland A R (1996) Single

point aerosol sampling: evaluation of mixing and probe performance in a nuclear stack, Health

Physics, Vol. 70, No. 1, pp 25-35.

Rodi W (1982) Turbulent buoyant jets and plumes, Pergamon Press.

Santon R C, Kidger J W, Lea C J, (2002) Safety developments in gas turbine power

applications, Proc. ASME Turbo Expo 2002, Paper GT2002-30469, 3 – 6 June, Amsterdam.

Schetz J A & Fuhs A E (1996) Handbook of fluid dynamics and fluid machinery, John Wiley &

Sons.

62

Seo Y, McFarland A R, Ortiz C A, O’Neal D L (2006) Mixing in a square and rectangular duct

regarding selection of locations for extractive sampling of gaseous contaminants, Health

Physics, Vol. 91, No. 1, pp 47-57.

Shell UK (1995) Engineering Reference Document. Code of Practice. Fire and gas detection

and alarm systems for offshore installations. EA/081. Rev : 1, Shell UK Exploration &

Production.

Simpson J E (1997) Gravity currents in the environment and the laboratory, 2nd edition,

Cambridge University Press.

Speziale C G (1996) Turbulent Transport Equations, Simulation and Modelling of Turbulent

Flows, Eds. Gatski T B, Hussaini M Y, Lumley J L, Oxford University Press, pp 185–242.

Walsh P, Johnson A, Ivings M (2005) Effectiveness of gas detection in HVAC ducts: Scoping

study, Health and Safety Laboratory Report FM/04/11.

White F M (1987) Fluid Mechanics, McGraw-Hill Book Company.

63

64

65

Published by the Health and Safety Executive 12/07

Health and Safety Executive

Assessment of gas detection strategies for offshore HVAC ducts based on CFD modelling The aim of this study has been to undertake CFD modelling to provide a basis for advice to inspectors and the industry on the effectiveness of flammable gas detection strategies for offshore HVAC ducts. CFD simulations of a high and low pressure gas release have been undertaken for idealised representations of an offshore platform, as well as a high pressure release for a more realistic geometry based on the Brae Alpha platform. In parallel with this modelling work a literature review has been carried out to build on a scoping study by Walsh et al (2005).

This report and the work it describes were funded by the Health and Safety Executive (HSE). Its contents, including any opinions and/or conclusions expressed, are those of the author alone and do not necessarily reflect HSE policy.

RR602

www.hse.gov.uk