Does Computational Fluid Dynamics ‘Improve Safety’ of Fixed Gas Detection Systems?
Author : Randall Williams, Proeon Systems
19 April 2022
This article addresses the concerns of CFD being used to design gas detector locations for fixed gas detection systems as it relies on assumptions the likes of which will likely not be accurate on the day of a given release, which is not good practice when designing a system to protect the occupants of the site.
The Petrochemical industry uses CFD when designing structures – Image: Shutterstock
(Click here to view article in digital edition)
Computational Fluid Dynamics (CFD), also known as dispersion modelling and Scenario, Dispersion or Risk Based Gas Mapping, uses computer algorithms to predict how fluids or atmospheres will react to predetermined conditions. This impressive technology has helped forecast air flow around vehicles and how air turbulence will affect buildings. However, in recent years it has also been used to model toxic and flammable gas for the purpose of designing gas detector locations. The claim that CFD ‘improves safety’ is based on predicting how the gas will migrate based on a fixed number of assumptions.
This article is addressing the concerns of CFD being used to design gas detector locations for fixed gas detection systems as it relies on assumptions the likes of which will likely not be accurate on the day of a given release, which is not good practice when designing a system to protect the occupants of the site. To support this argument, the article will define the necessary terms and then highlight each of the potentially mistaken assumptions and oversights.
Scenario Coverage compared to Geographic Coverage
There is an important difference between Scenario and Geographic coverage. The British Standard 60080 (Explosive and Toxic Atmospheres Hazard Detection Mapping – 2020 section 7.5) provides guidance on placing the gas detectors for a system solution and outlines three design methods:
- Prescriptive - gas detectors are placed based on proven designs or a ruleset-based approach by a design engineer, locating where the presence of gas is likely to cause harm.
- Volumetric Based Method - assesses the ability of fixed gas detection to detect gas clouds of a certain size in a pre-defined area of risk (volume). Software packages can verify the percentage of the area covered.
- Scenario-Based Method - considers the frequency and consequence to quantify the risk. To do this effectively requires a Quantitative Risk Assessment (QRA). This is where CFD software is often deployed.
The International Society of Automation (ISA) guidelines (TR84.00.07) summarises that Geographic Coverage measures the percentage of coverage for the defined area whilst Scenario Coverage measures the percentage of coverage for a set of scenarios. However, there is a widespread misunderstanding about the place of Scenario Coverage in the British Standard.
It was suggested that the British Standard approached the three methods as simplest right through to the more complicated (i.e., Prescriptive through to Scenario-Based) to reflect the level of complication of the project. The basis of this is that the more complicated projects would require the more complicated approach. It is the authors opinion, having been on the drafting committee for the Standard itself, that is not the case. This misunderstanding often justifies the use of technology which appears to be more complex: CFD. Designing for complicated sites, the number of unknown variables increases at the same time as known variables. Crucially, experienced fire and gas engineers are concerned that CFD can not sufficiently account for unknown variables, potentially providing an underdeveloped solution. Two further definitions are necessary to explain the case.
What is Computational Fluid Dynamics (CFD)?
CFD uses algorithms (mathematical modelling) of physical phenomenon to predict how fluids or atmospheres should react to predetermined conditions. The petrochemical industry uses CFD when designing structures. It determines the impact of an explosion on a structure, with consideration of environmental factors (flame speed, volume of gas, congestion, etc.), in order to prevent escalation. Furthermore, drawing on knowledge of the environment of onshore facilities in which a loss of containment (LOC) occurs, CFD models can predict how LOC can impact the surrounding installations and environment. They often form the basis of COMAH Reports, which aim to manage land use around hazardous sites. These reports use software and factor in the properties of the gas alongside wind diffusion effects, including the weight of the gas (passive, buoyant, dense), different types of release and evaporation. Importantly, when CFD is used for COMAH Reports, algorithms are used to predict how a gas may behave if there is a loss of containment.
Image (below) - CFD designs can be based on ‘average conditions’, however incidents often arise from conditions that are ‘out of the ordinary’, such as the Buncefield explosion where windspeed was very low, or zero, which facilitated the growth of a significant vapour cloud – Image: Wikimedia
As the availability of high-performance computing and improving algorithms has increased, CFD has started to be used to model for toxic and flammable gas detector location design, including to meet regulatory requirements and new design standards. The software allows for pre-determined conditions set out by the design engineer, such as leak points facing different directions, sizes of holes, frequency of leaks with wind/ventilation detection and speed. It provides models predicting how the gas will migrate and accumulate to unacceptable levels. However, to the detriment of CFD, these models are based on the assumption that the pre-determined variables are correct. This article will show that this is not good practice for designing systems designed for the protection of life.
Gas detection: toxic versus flammable
For the sake of context, it is important to point out the difference between toxic gas and flammable gas detection. Toxic gas detection aims to alert operatives of a toxic gas leak because it is solely about the protection of life. Whereas flammable gas detection is designed to prevent a gas cloud forming large enough to cause harm if ignited; the more built up or congested the area the more important it is to detect the gas cloud prior to ignition. In both cases, the gas detection system is required because something has failed and people are at risk and the system design must provide a margin of safety to protect the occupants on site. The moment when gas detection is required is when equipment has failed causing an unacceptable risk to the operatives. With such high stakes, fire and gas engineers are concerned that CFD reliance on limited assumptions and variables, which may not reflect the reality of the ‘real life’ event.
Protection of life
As outlined, CFD relies on the assumption that the pre-determined variables are correct. Highlighted below will demonstrate the danger in assuming values for these variables in advance because so many factors can influence them and there are a number of conditions which CFD does not take into account.
Correct 3D CAD models
On new projects, the site 3D CAD design (model) changes constantly from the FEED stage to detailed design to final production. The design is not fully finalised until the project is finished. By using inaccurate 3D models, the design can contribute to incorrect results from the software. Even after completion, the CAD Model may not be fully accurate until ‘As-Built’ drawings. Sites also change over time, through repair and renovation, and the leak points and pressures are likely to change. With CFD, change management requires additional layers of complexity for the entire life of the site and gas detection system. With older projects, accurate 3D CAD Models are difficult to obtain. Most Toxic Gas Detection systems are fitted in structures built long before 3D CAD was available. In these cases, CAD models are drafted by an engineer, but it would be impractical for them to capture every key structural element and leak point without significant investment. Thus, there CFD makes several assumptions which could easily be wrong, increasing the risk of an inaccurate predictive model.
Figure 1 – Sample of a 'Wind Rose' providing a graphical interpretation of the prevailing wind conditions. (Wind rose plot for LaGuardia Airport, New York, 2008)
Wind direction and speed / ventilation
External Applications - For external applications the software uses a set number of wind speed and direction from information obtained by meteorological sources. Importantly, for each wind speed definition the amount of processing time required increases. Developers have advised to perform initial dispersion cases (presumably with fewer inputs) to determine sizes of gas clouds produced for various wind conditions. This information can then be used to justify a smaller set of wind directions and speed for the project. The design is, therefore, based on layers of assumption that are based on ‘average conditions.’ Such assumptions are problematic, however, because investigations into incidents often reveal a chain of faults and conditions were ‘out of the ordinary.’ For example, the primary cause of the explosion at Buncefield in 2005 was a combination of mechanical and human failure, however the ‘the windspeed was very low, or zero, and a significant vapour cloud was able to form.’1 It may also be worth continuing to note that many of the worst vapour cloud explosions (more than 70%) which fixed gas detectors are there to defend against, occur on days where the wind speed is very low/nil. Graham Atkinson (Health & Safety Executive) states: “The majority of very large vapour cloud explosions occur in such low wind speeds that vapour flow is gravity-driven.”2 Surely we shouldn’t base our designs on dispersion through meteorological effects as this presents an optimistic view of a release.
Internal applications – Some argue that a higher degree of control can be achieved for internal designs, thus making CFD’s underlying assumptions more reliable. However, many assumptions are being made about the airflow and room layout not changing. BS60080 sets out that if the mechanical ventilation is not monitored it cannot be considered a mitigation for toxic gas.3 Consider the example of a site that had flammable and toxic risk being managed by a 20-year-old extract system. When the extract system failed to operate correctly the toxic and flammable gas was no longer behaving as predicted. If the ventilation was to be inputted to a software package to determine the location of the gas detectors, the ventilation would require to be monitored to be ‘fail safe or a high level of integrity (Safety Integrity Level). Once a ventilation system reaches that level of design the installation and long-term maintenance of the system can prove expensive in the long run.
Image (right) - CFD models designed for certain structures don’t always consider changes and updates, such as scaffolding, and so the gas detection design will require updating – Image: Shutterstock
Assumptions on leak points
New sites – Once again with new sites the design is constantly changing. For each leak point the engineer will need to assume location, the size of the hole, pressure, temperature (of the gas), frequency, etc. If the design changes, then the original model is no longer accurate. The assumption is also that the engineer plotting the leak points is suitably qualified and experienced.
Old sites – again, accurate 3D Models are difficult to obtain. The engineer must also consider how well the site has been maintained. For example, a site may have pressurised pipes supplying ammonia across a large building producing a food product. The pipework is all at high-level and is not tested or inspected as set out in the Pressure Systems Safety Regulations 2000 (PSSR). Ammonia is corrosive and the pipework could be beginning to fail. Consequently, a leak point may occur at any point along the ammonia supply pipe.
Ambient or pressurised – CFD Software in the domain discussed in this article is primarily designed for leaks from sealed pressurised systems suffering a loss of containment. Often the source of the gas is from an ambient source. An open tank or cabinet may be used as part of the process and gas may escape via the top or sides of the unit.
All these cases demonstrate that the assumptions CFD makes about leak points are far from complete, thus increasing the risk of an inaccurate predictive model.
Unaccounted environmental conditions
Having already highlighted varying wind conditions, other environmental factors do not seem to have been considered in the CFD Software.
Available Gases: Many sites are part of a large complex where other gases may be available but are not toxic. However, during the loss of containment the errant gas may combine with another gas changing its behaviour.
Additional Unknown Variables: Other variables have been pointed out that do not seem to be considered, for example:
- Hours of sunlight can affect the temperature of the surface which can change the atmospheric conditions
- Organic surfaces store less heat than harder surfaces (concrete)
- Changes beyond the border of the site changing the turbulence of the incoming wind
This list is certainly not exhaustive and could be expanded further, but it illustrates that CFD does not account for a significant number of unknown variables which increases the risk of an inaccurate predictive model.
Unplanned structural changes
Randall Williams, Proeon Systems
Every site undergoes changes and updates. For an external example, scaffolding might be required for an urgent repair at a refinery. The CFD Model was designed for the structure without the scaffolding and so the gas detection design will require to be updated to include the scaffolding. For an internal example, a laboratory may have a model to protect against a toxic gas using CFD modelling. However, because of the recent epidemic, screens were required to be set up between occupants to protect them from infection, consequently the gas detection system will require to be updated to include the screens. These changes are often urgent and not accounted for in the ‘Change Management’ process, thus increasing the risk of an inaccurate predictive model. Scenario based design is only relevant if the ‘as built’ site is identical, and the conditions on the day of release are the same as the model.
Measure, don’t guess: Geographic Coverage modelling
The doubt’s highlighted here demonstrate CFD Models make several assumptions which may or may not be complete and neglect a number of unknown factors in determining the placement of gas detectors. In other words, many assumptions are being made about how the gas might behave without considering the multiple unknown variables and incomplete suppositions. There is a crucial guiding principle in Fire & Gas Safety Engineering: ‘We don’t know what we don’t know.’
Regardless how clever the people are in the room, regardless of the sophistication of CFD technology, there will be unknown elements of risk. An additional ‘unknown’ is if the algorithms employed by the CFD software are appropriate, suitable or sufficient. BS60080 has listed six disadvantages of scenario-based mapping, a task often performed with CFD. If we cannot rely on CFD models as they risk an inaccurate predictive model, what then is the alternative?
Geographic modelling is a method which measures the coverage of the target gas cloud. Admittedly, BS60080 has also listed weaknesses in the approach. But one of its many advantages, according to BS60080, is that it is ‘easily audited.’ Consequently, through a good Gas Detection Risk Assessment, defined areas of occupation can be defined and protected. The gas detection system can then provide coverage that can then be verified by appropriate software. If the engineer limits the CFD simulation to place gas detectors only in the predicted locations, the final design may be more conservative. However, increasing the number of variables will provide a greater number of predicted locations than a geographic approach. The advantage of geographic modelling is placing detectors to detect the cloud of concern. No more, no less. Therefore, an engineered solution. The coverage can be measured and will provide confidence that the gas detectors are correctly located to protect the occupants.
Conclusion
Lord Kelvin famously said: “When you can measure what you are speaking about, and express it in numbers, you know something about it.” To measure is to know, to estimate is to guess intelligently, and to predict is to guess with data. With fire & gas systems designed to protect people’s safety, we should strive to know that the gas detector locations are correct. CFD Software is clever and powerful, it predicts how air and gas will flow. It may be ideal for designing race cars or airplanes, but to answer the original question, ‘Does Computational Fluid Dynamics Improve Safety of Fixed Gas Detection Systems’ as previously discussed, the only way to prove (measure) if the gas detection is in correct locations is to define the areas that require protection (via a Gas Detection Risk Assessment)4 and Geographic Modelling to verify the coverage. Predicting where gas might migrate for Life Protection Systems is not good practice. Consequently, there perhaps should be hesitation before looking to employ a Scenario-based package to design a system to protect human life. Fire Risk Assessors, Fire Safety/Health & Safety Inspectors would be wise to question CFD Modelling for environments with toxic or flammable gas.
References
1 Buncefield Major Incident Investigation, Buncefield Major Incident Investigation Board, Page 37
2 Gravity–Driven Flammable Vapour Clouds, Graham Atkinson, Health and Safety Executive, Harpur Hill, Buxton, U.K. 2017
3 BS60080 Explosive and Toxic Atmospheres Hazard Detection Mapping – 2020 (Section 7.6.2- Table 1) Page 41
4 BS60080 Explosive and Toxic Atmospheres Hazard Detection Mapping – 2020 (Section 7.6.4) Page 42
About the author:
Randall Williams (BSc, GIFireE) has been working in the fire safety industry since 1988 as a Fire Alarm Engineer and Fire Risk Assessor. Randall was a member of the BSI Committee which drafted BS 60080 and he leads the Fire & Gas Team at Proeon Systems Limited in Norwich, (UK). Randall is an experienced Fire Risk Assessor specialising in Industrial Fire and Gas Risk Management. He is a Graduate of the Institute of Fire Engineers (GIFireE) in 2013 and graduated with BSc Fire Protection Management & Technology from California State University, Los Angeles, California in 1986.
Contact Details and Archive...