SILy

Functional Safety for beginners ...


It's all about the trips !

This is only intended for Managers and "Capable" Amateurs to increase awareness - for real guidance on the topic contact a Competent Professional (and be amazed by the rates they quote ☺).

In a nutshell (rather than a bombshell), it can be summarized like this:
  1. Define and understand the Process.
  2. Plan for execution and support using Competent personnel.
  3. Identify Hazards and evaluate Risks.
  4. Take credit for Protection measures and compare risk to Tolerable levels.
  5. Determine Integrity level for Safety Instrumented Function if required.
  6. Clearly, concisely and completely document all Requirements.
  7. Develop design to meet Reliability.
  8. Confirm functionality and performance before putting into service.
  9. Operate, Maintain and Test in accordance with specification.
  10. Evaluate demands, failure and changes and Revise accordingly. 
But for a more detailed (interesting) opinion, read on...

If you feel the need for a Safety Instrumented Function (SIF), Safety Instrumented System (SIS), Instrumented Protection Function or System or anything using electrons (E/E/PE = Electrical/Electronic/Programmable Electronic) then you've probably made a mess of your Process Design because your pipework isn't strong enough or your vessels aren't big enough.

WHAT ?

OK - you've probably inherited an old plant when design envelopes were what you used instead of email or you've got a really bad-ass process that's too big or fast to be contained so you can only play with the cards that you've been dealt.

But your SIS/SIF doesn't have to do all the work, there is often a number of other Protection Measure which Prevent (reduce Likelihood) or Mitigate (reduce Consequences) the Risks associated with each Hazard.




Swiss Cheese Model

You should know by now that SIL 1 is good and SIL 4 is excellent (but not perfect), but don't just dive in and select a Safety Integrity Level based purely on Severity alone - there is a process, a Lifecycle, to follow that starts with a familiar activity - Hazard Identification.


Functional Safety Lifecycle (Simplified)
HAZARD IDENTIFICATION
You may know this as HAZOP or PHA - whatever you call it, you need a systematic method to identify the hazards.

For each hazard, there may be a number of potential Causes all leading to the same Consequence and you may be tempted to focus on the most likely one, but don't forget that (for example) if you are crossing a road and there are 10 different colored cars that may pass, each of these going by once every hour then there is a threat of being hit by any car once every 6 mins (crudely) - you don't care what color it is.

Don't get bamboozled by the maths (math) just yet - all I'm saying is that many infrequent events can lead to an overall frequent occurrence.

Remember that a Safety Integrity Level relates to Safety i.e. the absence of HARM which is a human effect and that should be our primary focus. There may also be other effects on the environment (which can also experience harm) as well as equipment, output and image (these will strictly not be harmed but damaged).


Bow Tie Diagram
So SIL only applies to people and not the other receptors although some companies may refer to EIL (Environmental Integrity Level) and CIL (Commercial Integrity Level) or FIL (Financial Integrity Level) but basically they are consistent methods to score the criticality (reliability) of the protection measures.

RISK ASSESSMENT
Before you evaluate how much risk each hazard presents, you must have a consistent reference point about how much risk you can actually accept.

Remember that Zero risk is a noble concept but is like Absolute Zero (-273.15 ºC or -459.67 ºF if you are using Victorian/Colonial units) or Full Vacuum i.e. it can never actually be attained.

There is always an element of risk present, in fact you are probably at risk right now as you read this (the threat of your boss catching you or being chastised for not playing with the kids) - Stay ALERT ! (Your Country Needs LERTS).

You can find lots of references out there but it's really not your call about how much risk is TOLERABLE - we all have a different perception of risk e.g. driving, foreign food, gambling etc and therefore we must leave it to those who will typically go to court (or jail) if something bad happens. In other words your tolerable risk should be set at a Corporate level (based on their appetite for risk - ultimately a commercial balancing act) to ensure a consistent approach across your organisation.

In simple (but perhaps distasteful) terms, the "easiest" way to measure consequences are to consider deaths (since you are either dead or not dead - everything else is just grades of injury). I know some countries don't like to think/talk this way (either legally or culturally) but humor me and go along with the concept just now.

A major hazard scenario typically has the potential to cause one of more fatalities so we can measure this as Deaths/Event.

The likelihood (frequency) of these fatalities actually occurring is measured in Events/Year.

Since Risk is the product of Consequence & Likelihood, if we multiply these together we get a risk measured in Deaths/Year.

Every hazard will have its risk and from this you can estimate (or calculate) how far you are away from your tolerable risk target. The Risk Reduction Factor (RRF) is a measure of how much greater the Process Risk is compared to the Tolerable Risk.

In the example below, the Unmitigated (i.e.without any protection measures - if you can imagine such a thing) Process Risk is estimated/predicted to be 1 death every 100 years which is 10,000 times greater that the Tolerable Risk target of 1 death every 1,000,000 years so therefore the risk has to be reduced by a factor of 10,000.
Risk Reduction Overview
Your aim is NOT to find the "best" SIF for the job i.e. enough to take on the whole risk reduction burden (e.g. mathematically a SIL 4 SIF would appear to do the job), but to evaluate all your protection measures and consider what risk reduction contribution they can "confidently" make (I've avoided the word reliable just now to limit confusing terms).

In the above overview, you will see Non-SIS IPL listed which are Independent Protection Layers (i.e. they don't rely on other measures for their performance) seperate to the SIF you are considering for this hazard.

These may include the following Prevention measures:
  • Basic Process Control Systems (BPCS) aka DCS, PLC, SCADA
  • Alarms (not implemented in the BPCS)
  • Mechanical protection (e.g. safety/relief valves/discs or overspeed bolts)
You may also may want to consider Mitigation measures, such as:
  • Control of Ignition Sources
  • Occupancy
  • Avoidance
  • Vulnerability
And these lead us neatly onto the next topic

SIL DETERMINATION
At the start of this post I mentioned that if you need a SIF (or SIS) then your design is probably inadequate. OK - a bit harsh but don't be fooled that a SIF/SIS is a panacea for all hazard. Sure, the modern ones are "intelligent" but they require a significant investment in time & money to Analyse, Specify, Design, Procure, Install, Commission, Maintain and Test so be careful how you use them.

The Functional Safety Standards do NOT require you to implement SIF/SIS for ever hazard, they fully respect non-SIS Protection Measures as components within a hierarchy of risk reduction and recognise that SIF/SIS may only be required where there is a shortfall in risk reduction when all other appropriate (capable and independent) measures have been considered.

There are generally 2 types of SIL Determination methods in common use:
  • Risk Matrix of Graph
    • Qualitative - target SIL is selected based on a number of parameters.
  • Layer of Protection Analysis (LOPA)
    • Quantitative - target RRF is calculated based on a number of values.
Note that Fault Tree Analysis (FTA) may also be used where LOPA is not capable of evaluating complex scenarios where independence between protection measures is not clear.

Whichever method is used, they must still be calibrated to reflect the Corporate Tolerable Risk criteria.
DO NOT JUST TAKE THE DIAGRAMS FROM THE STANDARD AND USE THAT WITHOUT UNDERSTANDING WHAT THEIR BASELINE (TOLERABLE RISK) IS !
I'm not going to teach you how to use these methods (somebody else can charge you for the pleasure), however I will emphasis that whichever method is used, relevant experience and significant data are vital i.e. avoid the lazy approach of looking up a book to get failure frequencies or probability of failure rather than examine your own plant/site/company records.

OK - in many cases, such data is scarce and not always complete but make sure whatever source you cite, record it in your assessment to help future reviewers (which may include Regulatory Inspectors) understand why you selected that parameter or value.

Both Risk Graph & LOPA consider mitigation (post release) measures as potential risk reduction contributors. Do not get these mixed up.
  • Occupancy
    • The probability that somebody will be in the hazard effect ("kill") zone. May also be know as Exposure.
    • Ideally you want this to be very low number (close to 0) and will be affected by the work patterns (shift cycle) as well as the proximity of a permanent (e.g. office) or transient (e.g. public) population.
  • Avoidance
    • The probability that (given that somebody is in the "kill" zone) they can either escape or intervene.
    • Ideally you want this to be a very high number (close to 1) and will be affected by the measures available to warn of an imminent hazard, the escape routes available and the escalation time of the hazard.
  • Vulnerability
    • The probability that (given that somebody is in the "kill" zone and is unable to avoid the hazard) a fatality will result. This is the opposite of Survivability.
    • Ideally you want this to be a very low number (close to 0) and will be affected by the nature of the hazard e.g. Vulnerability is higher (~ 1) for an explosion compared to a release of toxic liquid (~ 0.1).
The last thing I'm going to say in this topic is don't get confused between Probability and Frequency.

Probability is a dimensionless number between 0 (impossible) and 1 (guaranteed) and
Frequency is a number with units of time (per hour, per year etc)

So remember, you can:
  • Multiply Probabilities together (still dimensionless and between 0 and 1)
  • Multiply Probabilities and Frequencies (in same units as Frequency)
  • Add Frequencies together (still in time units)
But you can't:
  • Multiply Frequencies together (you get time²)
  • Add Probabilities together (you'll probably get a number > 1)
Now you've determined your target SIL (or RRF), how do you get somebody to make it happen ?

SAFETY REQUIREMENT SPECIFICATION

Remember this classic cartoon ?
TREE SWING
It is founded on really simple principles - know what you NEED and ensure that whoever is providing it UNDERSTAND what you need and then DELIVER it.

The HSE publication Out of Control highlighted that 44% of accident involving Control Systems were attributed to poor Specification and roughly 60% of incidents are already inherent (waiting to happen) before the system is even installed and commissioned.


We expect that Specifications are written by educated and informed people, however from the above analysis it is apparent that:
  • Incidents can and do happen at any time throughout the lifetime of a system.
  • The majority of incidents are caused by not understanding WHAT and WHY.
    • Poor Specification
    • Uncontrolled Changes
  • Specification is typically done on somebody else’s behalf i.e. for the USER.
  • Changes are often made without due consideration of the impact.
  • Plants are operated & maintained for decades, system implementation is brief.
So, remember that your Safety Requirement Specification must include (at the very least):

Functional Requirements
  • What setting or condition will initiate the trip.
  • What the required action/response is when a deviation is detected (demand occurs)
  • What the require action/response is when power or air/hydraulics is lost
  • What the required action/response is when a fault is detected (before a demand occurs)
  • What limits there are on testing (max or min proof test interval)
Performance Requirements
  • How reliable the function (or system) should be (SIL or Probability of Failure)
  • How quickly the function should operate (response time < process time)
And should be documented in a way that is:
  • CLEAR
  • CONSISTENT
  • CONCISE
  • COMPLETE
  • CORRECT
Then your Supplier will have no excuse for not giving you the tree swing you really needed.


SIL VERIFICATION
How reliable does it need to be - please jump down to Death & Taxes to discuss this further.

SIL SUSTAIN
This is deliberately called Sustain and not Maintain as there is more to this than just maintenance.

Safety Systems are not just "Fit & Forget" they need to be looked after so they look after you and this is not solely the responsibility of the Maintenance folks, to inspect and test the trips.

The ability of a SIF to operate as required as quickly as required will depend on a number of factors.
  • Management of Change (including trip settings)
  • Management of Overrides (including maintenance friggs and startup/commissioning)
  • Environmental conditions (both internal and external)
Of course regular and rigorous Proof Testing is a key element in ensuring that your SIF/SIS continues to perform as intended but are you testing the complete SIF from Sensor all the way to the Final Element or are you only injecting/measuring signals where it is convenient (warm and dry) to do so.

Remember sensors can block and final elements can be stiff or jam - but have you checked for this ?

Finally (and thanks for reading this far down)...

Functional Safety is not an exact science
[Insert your Country Name] Roulette
Predictions are made of process and equipment performance based on Historical Evidence and Theoretical Calculations.

Functional Safety Requirements (target) are estimated based on the risk (and risk reduction) associated with a particular process hazard.

Increasing the Likelihood and/or the exacerbating (making worse) the Consequences will increase the risk and therefore the Risk Reduction requirements on that SIF.

So if experience shows that the hazard occurs more frequently and/or the effects are worse than expected, then the SIF may no longer be adequate to provide the (now increased) risk reduction.

Similarly (but conversely) the Functional Safety Achievements (actual) are expected based on the instrument/element failure rate and the proof testing regime.

So if the SIF components fail more often than the manufacturer declared and/or testing is not as often or stringent (demanding) as necessary then the SIF is also unlikely to be able to operate as required as quickly as required.

All estimates, assumptions and predictions must be regularly reviewed and changes made where necessary to ensure that the appropriate level of risk reduction is provided.

Also remember that in your Risk Assessment or SIL Determination you may have assumed that the non-SIS IPL would provide a certain amount of risk reduction. For example you may have (generically) claimed that your BPCS will fail 1 time in 10 and your relief valves will fail 1 time in 100, but unless they do perform to these levels the risk reduction requirement the SIF needs to bridge/cover may actually be higher and your SIL 1 SIF may actually need to be SIL 2.

Ensure you treat your non-SIS IPL with same/similar respect that you treat your SIF with regards to Management of Change/Overrides and Inspection & Testing. Also review their performance and contribution as part of a specific risk reduction approach.

"'In this world nothing can be said to be certain, except death and taxes." but how can you be certain that your Safety System or Function will work when it needs to ?
I'll be OK - I have lots of barriers
Enter the black art of probability - when you place a demand on your Safety Instrumented Function (SIF aka trip) then it will probably not definitely work.


So how do you get more confidence that it operates when required and as required (completely and quickly enough).

Hopefully you have enjoyed a painless and informative SIL Determination and have either identified a target Safety Integrity Level (SIL) or Risk Reduction Factor (RRF) to help address a particular hazard.

From this you (or your Supplier) need to demonstrate that this SIL/RRF can be achieved by the proposed (or actual) SIF.

This performance requirement can be determined from either the SIL or the RRF and is measured as a Probability of Failure on Demand (PFD).


Safety Integrity Level comparison table
The higher the PFD (closer to 1) the more likely it is that the SIF will fail when requested/required. The lower the PFD (closer to 0) the less likely it is that the SIF will fail when requested/required.

But don't look at these numbers and think that you can sleep soundly - remember these are predictions of probability and that "1 in a million chance"  event may actually be tomorrow or on the next demand:
Damn Lies and Statistics
Your Safety Instrumented Function will (even if you don't know it) consist of the following key components:
  • Sensor (SE)
    • To DETECT when a deviation or abnormal condition has occurred
  • Logic Solver (LS)
    • To DIAGNOSE what has gone wrong and figure out what to do about it
  • Final Element (FE)
    • To DO the required action to make the plant safe (or keep it in a safe state)
And the overall reliability of the SIF will depend on all of these components (and more)


What is the weakest link in the SIF chain ?
In simple terms, the Probability of Failure on Demand of the SIF (PFDSIF) is the sum of the PFD for the Sensor (PFDSE) plus the PFD of the Logic Solver (PFDLS) plus the PFD of the Final Element (PFDFE).


PFDSIF = PFDSE + PFDLS + PFDFE

So you'll see from this, that to achieve as low a PFDSIF as possible you need to consider the contribution of all the major elements (plus any intermediate components e.g. signal conditioning, solenoids etc which have been excluded here for simplicity) and that the overall result will be heavily influence by the highest PFD since the maths is addition rather than multiplication.

The SIF is therefore only as strong (reliable) as its weakest link and industry experience/evidence shows that typically the contribution to failure is approximately

  • Sensors = 40%
  • Logic Solvers = 10%
  • Final Elements = 50%
This is kind of obvious since both the Sensor and Final Element are exposed to the process and the environment and also the Final Element tends to have more moving parts so is prone to wear & tear compared to the Logic Solver which may actually be a electromagnetic relay (yes they still do exist and do a very good job) which moves but is typically in a controlled environment or if it is a programmable device (e.g. Safety PLC) there is little to degrade as software doesn't age - but it can be tampered with !

So think hard before you ask for dual redundant processors and I/O since that may not add a huge amount to the overall reliability if ultimate safety relies on an old. slow, rusty valve that hasn't moved in ages.

Reliability Mathematics
The maths (or math if you insist) of reliability is generally too tricky for mere mortals so I won't attempt to educate you on the likes of Differential Equations and Markov Models but I will run through the basics. If you really want to know more I can recommend some learned professionals who know much more than I do.

Nothing is perfect, everything will fail sometime based on:
  • Quality
  • Quantity
  • Testing
Failure calculations will therefore depend on:
  • How often the device fails (failure rate)
  • How many spare (redundant) or different (diverse) devices there are
  • How often testing is performed (frequency) and how well (effectiveness)
There are other factors such as Diagnostics but I'll keep this simple.

Failure Types Failure rates for devices are initially divided into the following categories:
  • Dangerous
    • Does not do what you want it to do when you want it to do it
  • Safe
    • Does what you want it to do but not when you want it to do it
These categories can be further subdivided into:
  • Detected
    • You know about it (and hopefully can take some remedial action)
  • Undetected
    • You don't know about it (and are unable to do anything about it)
The supplier or data sheet may only provide an overall failure rate e.g. MTBF = Mean Time Between Failure but you really need to know how this is broken down as you make make an error of judgement by assuming all failure are dangerous and therefore over-engineer your protection (better safe than sorry) or assume that the balance of safe vs. dangerous failures is excessively high and therefore under-engineer your protection.

So, what you really are worried about are the Dangerous Undetected (DU) failures. OK the others are important, particularly if you want to avoid/reduce spurious failures, but lets focus on the bad boys.

To help you remember the difference, consider this scenario of a knight protecting a castle (or - if your nation is less than 300 years old - a soldier defending a fort) and the knight/soldier is the SIF and the castle/fort is the process/plant.
Who's looking after your place ?
The knight works hard (carrying all that armor and a sword is tiring) so he sleeps when he can and often on the job. He is therefore passive and only becomes active when a demand is placed on him (e.g. either he is aware of an attack or somebody else spots it and wakens him up).
  • If the knight dies in his sleep and nobody knows about it then this is a Dangerous Undetected (DU) failure - he is unable to act (fight) when required and there is no warning to allow a replacement. A lesser version of this is he becomes ill while he is awake, is unable to fight on demand and there is no opportunity to treat him i.e. to return him to service.
  • If the knight dies in his sleep and creates a noise or a smell as he expires then this is a Dangerous Detected (DD) failure - he is unable to fight but somebody knows about it and can replace him (or treat him if he only suffers a noisy/smelly illness).
  • If the knight wakens up prematurely and starts lashing about quietly with his sword before anybody attacks then this is a Safe Undetected (SU) failure - he has tried to do what he is required to do but before anybody has attacked and is therefore tiring himself out and potentially reducing his chance of either waking up when the attach occurs and/or subsequently being able to fight hard enough to protect the castle.
  • If the knight wakens up too soon, starts fighting (with nobody) and creates a commotion then this is a Safe Detected (SD) failure - his colleagues will ridicule him for being such an idiot but at least he can be rested or replaced before a real attack (demand) arrives.
Feel free to tell this story to your kids - but don't give them a taste for Game of Thrones - it will give them nightmares.

Redundancy
The formula you use to predict the failure will depend on the architecture (redundancy and diversity) of the SIF. Needless to say, the more elements you have the less likely the SIF is to fail so generally more is better but remember there is also more to go wrong (spurious trip) and to implement and test (time & money).

However you may have to have multiple elements because of the complex nature of the process/plant and you actually need all not just one or some of these to operate in order to achieve (or maintain) a safe state.

So make sure you really understand your 1oo1's from your 2oo3 and hopefully you can get by with MooN rather than NooN.

This has nothing to do with Celestial bodies or Astronomy (definitely NOT Astrology) it is simply a shorthand way to express the voting of a particular group of elements in a SIF. Be aware that each element (Sensor, Logic Solver & Final Element) may have it's own voting within the SIF.

MooN means that from a possible N elements only M of them need to operate successfully for the element group to detect/act successfully. So 1oo2 means there are 2 elements (that don't actually need to be identical - this is Diversity) and only 1 of them needs to work for the element group (e.g. transmitters or valves) to be successful. Then 2oo3 means only 2 out of a possible 3 need to work etc.

In this case , M < N however more challenging conditions occur when all elements need to work i.e. NooN e.g. 2oo2, 3oo3 etc - you really want to avoid these if you can.

Testing
I covered some of this in the SILy section - needless to say you need to look after your SIF so it can look after you. This will include regular, rigorous and representative testing with real end-to-end (sensor to final element) operation to provide confidence that it works.

But remember that a successful test is not a cast-iron guarantee that your SIF is "safe" - it's like hitting the brakes on a car. If they stop the car then that's great but it doesn't mean they'll work next time.

Also consider that you can over-test devices since every intervention is an opportunity to make an error e.g. leave an override in or some debris in a tapping line.

There is more to this than pressing a test button every year and checking that the valve closes - you have been warned !

No comments:

Post a Comment