lADC/SPE

35038

Applying Quantitative Risk Assessment 1.D. Keilty, British Gas Research & Technology, and

to Casing Design
H. Rabia, SPE, British Gas E&P

Copyr[ght 19S6 IADCISPE Drllllng Conference Th!s paper was prepared for pmsentabon at the 19S6 IAOCISPE Dr!lhng Conference held m New Orleans, Loumana, 12-t5 March 19S6 Th!s paper was selected fm pmsentabon by the lACC/SPE Program Commdtee following rewew of information contained m an abstract subm!ned by the author(s) Contents of the paper as presented, have not been reviewed by the SGaety of Petroleum Engineers or the Internatmnal Assocnatlon of Drdllng Contractors and are subject to mrrecbon by the author(s) The material as presented, does not necessarily reflect any pasit]on ot the IADC or SPE, their offjcers, or members Papers presented at IACCISPE meebngs are s.b)ect to publ,catm renew by Edltonal Commttee of the IAOC and SPE Permmsmn to copy IS restricted to an abslract of not more than 300 wvrds Illustrations may not be cop!ed The abstract should contain conspicuous acknowledgement of where and by whom the papet was presented Write L[braflan, SPE P O Box 833%36, Richardson TX 75083-3836 U S A

answer the fundamental question: How often will a failure occur?. Quantitative Risk Assessment (QRA) is a probabilistic tool that addresses this question, laying the foundations for increased safety awareness whilst maintaining cost efflcieney, The purpose of probabilistic design is to replace over- or under-engineering with consistent design that is fit-for-purpose. QRA is a method through which the risks of failure inherent in an engineering design can be determined, to compare different designs, and gain an overall idea of the safety and adequacy of the design. The implementation of a probabilistic methodology also disciplines the designer into considering the real situation, as opposed to a series of intuitive assumptions that may have sufficed in the past. QRA has a number of components that require detailed consideration: Data, The acquisition, quality cent rol and subsequent use of all types of data is central for the adequate execution of a Quantitative Risk Assessment, British Gas decided some time ago to join a Joint Industry Project on QRA, and this exercise has been very instrumental in improving data quality and quantity for subsequent QRA Methodology. There are a number of mathematical approaches suitable for reliability based design, each having a series of benefits and drawbacks. This paper will critically evaluate these approaches, and outline the QRA methodology adopted by British Gas. Implementation, QRA design can be a powerful design tool and the potential for its usc (or abuse) is large Care must be taken when setting up such a tool. Applications, QRA does not replace conventional design tools, and does not solve all the problems of deterministic design A way must be found to take advantage of both methods.

Abstract There are three kinds of lies:

In 1993 British Gas decided that it would be advantageous to explore lhc possibility of developing a new casing design methodology, based on probabilistic concepts and historical data. This bccamc ncccssary in order to align drilling design practice with that of other disciplines within the E&P group For example, it was considered uneconomic to design a WC1lto last for ever when platform Iifc-spans were of the order of twrn[y years,. To date, British Gas has been encouraged by its findings, and the potential benefit from such an approach, dcmonslrakd through initial studies into current casing designs. Introduction Two of the most Important factors in any design are safety and cost. and throughout industry reliability-based design methods have a long pedigree in the control and optimisation of these factors. This paper concerns the experiences of British Gas during its ongoing implementation of a reliability-based casing design tool within its drilling function Traditional casing designs usc a deterministic approach, where minimum cxpectcd strength is designed to exceed maximum possible loading, usually with a safety factor added as a contingency. In the majority of cases, however, [his worst case scenario will not be realised. Conversely, there will bc rare instances when load dots exceed strength, and the casing fails. This deterministic approach does not 61

APPLYING QUANTITATIVE

RISK ASSESSMENT TECHNIQUES TO CASING DESIGN

SPE 35038

QRA: The Basics. This seetion will describe the basic tenets of reliability theory used by QRA3, and contrast them with deterministic design. Conventional Design. Conventional casing design techniques developed over the last 50 years or so relied on the use of poorly estimated subsurface load data, including geopressured, mud backups, influx behaviour and rock formation loads such as salt squeezing. Material properties were predicted from equations derived from a limited set of experimental data$, and the data themselves referred to a limited number of grades and did not address special grades or duplex steels, Due to the above uncertainties, it became common practice when designing casing to assume maximum well loading and minimum material resistance. This assumption ignores Lhe majority of loads and resistance combinations that occur most of the time by assuming a worst case. Obviously, it would be beneficial to quantifi and make use of this redundancy in the design process without compromising safety, Recently, designers using a deterministic approach have recognised the inadequacy of the SF,, and attempt to rectify this by including as many of the load variables as can be predicted, together with the operational consequences of each, The latter is particularly significant in designs based on given kick volume, where the kick detection time is taken into consideration. QRA offers an explicit and scientific way of carrying out this operation. If a deterministic design is executed correctly, the resistance will exceed the load, and the relative magnitude of the difference between the two values will represent the safety factor (see Figure 1). However, this safety factor is not a direct measure of safety or risk. The satety factor as a stand-alone design basis is becoming increasingly outmoded in the structural industry More designers are choosing (or are forced to adopt) risk-based design criteria. For example, a passenger would not board an aeroplane if told:
Ladies and Gentlemen, we are proud to announce that this plane has a safety factor of 1,5!.

Randomly Distributed Variables. The real case of many variables is one whereby there is no absolute and simple way of predicting the outcome in any single case, Taking yield stress as an example again, if a tensile test was performed on one steel sample, a single value for yield strength would be produced (see Figure 3.). However, if the same test was performed on an ostensibly identical specimen, it would yield at a different value. If this process were to be repeated, a large range of values would be obtained, In the case of yield stress, this would produce a distribution of possible yield strengths, each one having a different frequency of occurrence. This behaviour characterises an underlying Normal Distribution (see Figure 3), This is an example of a Probability Density Function (PDF), A PDF shows which values are more likely to occur by representing that higher probability through a larger area under the graph over that range. Thus, the median yield strength would have a higher frequency of occurrence than the SMYS, and therefore a higher probability density (a higher point on a PDF). Each of the input variables in a casing design has a PDF associated with it. For example, on the load side, there is a pore pressure (predicted against actual) distribution, a kick size and intensity distribution. Also, when predicting casing resistance, Diameter, thickness and yield strength are not single-valued quantities, but also have distributions associated with thcm, QRA Using Randomly Distributed Variables, Once the input PDFs have been defined, it is then necessary to begin the design process, This, perhaps surprisingly, can be carried out in almost the same way as a standard design. The same equations may be used: the crucial difference is that the input variables have changes from assumed nominal values to more realistic random distributions, QRA uscs probabilistic mathematics and statistics to factor together the load and resistance variables into two distributions (see Figures 2 and 4.). The first defines all the possible values that a load case can have, For a kick load case, this would cover the surface pressure experienced from a 1-2 bbl condensate kick, right through to a very large gas kick at surface, Each different severity level will also have a probability of occurrence attached to it, The second distribution will govern casing resistance. The data space will cover the range in which all the input variables conspire to produce, for example, a very low collapse value (reduced yield strength, thinner wall thickness, ovality, etc.) to the equally unlikely situation where all those variables combine to give the casing a very high collapse resistance. The result will be two opposing distributions: load and resistance,

In reality, that maximum load is seldom reached (e.g. entire casing filled with gas). It is also a relatively rare occurrence for the yield strength of the casing to fall as low as its Specified Minimum Yield Stress (SMYS). Equally, however, there will be occasions on which the yield strength falls below the SMYS.

62

SPE35038

KEILTY, RABIA

Figure 2 illustrates the point that despitea safety factor and good design practices, there will be a few times in a large number of applications of a particular design, in which load exceeds resistance. This is almost inevitable, but rather than ignoring the possibility, it is far more sensible tocngineer thedesign toensurc that this failure rate isreduced to an economic level. This can be done by examining the casing parameters governing resistance, and adjusting them to reduce the proportion of the data space for which the load exceeds the resistance. This relative area will determine failure rate (e.g. a value of 104would mean one failure in every 10,000 applications) Tolerable Risk Levels (TRLs). A failure rate can be used in two ways, It is possible to execute a QRA on a particular casing, designed in the standard way, and then accept or reject it, based on the failure rate obtained. A more consistent method is 10 set a Tolerable Risk Level (i.e. maximum acceptable failure rate) and choose the casing parameters to ensure that the failure rate is below this Icvel.
Setting of TILLs. The Cullcn report was issued after the Piper Alpha disaster in the United Kingdom, and it recommended the principle of ALARP for UK offshore operations. This means that the risk of failure of any engineering design should be As I.OW As is Reasonably Practicable. This principle requires that every engineering design must have an ALARP failure rate, given current lCVCIS of technology and knowledge, as well as all reasonable expenditure to make the design safe.

Different QRA Levels and Methods There are several different levels and methodologies available when adopting a QRA approach] b8, and this section will give a brief qualitative description of each. Revised Single Factors. Historical data and improved design equations can be used to update an operators current safety factors through a one-off QRA. The analysis is also likely to highlight the inadequacy of using a single Safety Factor for each load case, no matter what the well conditions QRA will demonstrate that the probability response will not bc flat across the full range of wells designed for. For example, it is common sense that a wildcat well may require more redundancy than a development well that is more of a known quantity. This kind of contingency built in to cope with the unexpected will almost certainly bc designed in by experienced designers (e.g. through more conservative load cases), However a QRA analysis process would ensure that any conservatism were explicit, and consummate with whatever increased risks might bc present in critical wells. It would also support the case for using different safety factors for different well t3-pes in order to achieve a flat risk response, i.e. equal risk levels. However, the benefits from such an approach would be limited. It would be diftlcult to winkle out the variables on which it would be possible to reduce conservatism, due to the need to maintain safety levels and contingency on other variables (e.g. pore pressures). Also, the system of safety factors would rapidly bccomc sub-optimal, as drilling technology advances (e.g. a ncw kick detection system) or more data bccomcs available. Partial Factors. This approach involves assigning safety factors to each variable before input into the design equations. This atlords the designer more control over the design process. In this way, over-conservatisms can be more easily screened out, while uncertainties (e.g. pore pressures) can be accounted for adequately. The procedure involves an escalation of complexity, as a greater number of safety factors arc required, but the design process remains broadly the same. The same ditXculty exists as with single S.F, design, whereby the one-off QRA assessment means that once this stage is over, the designer is again designing blind to risk. Full QRA methodologies. The next stage of development is to consider a full QRA implementation for every WCIIdesign, or at least to have the capability of doing SC, The benefits of this are t~vofold: it is a tool by which the designer can quickly and objectively evaluate casing designs on a risk basis, It is also useful in that it allows the regular rc-validation of the two approaches outlined above, maintaining accuracy in the light of fresh information or changes in conditions. The following section provides a very brief description, and presents some of the pros and cons of each technique.

Risk in an cngincermg sense can be detincd as:

CONSEQUENCES

( 1)

The TRL thresholds identified by the report range from 104to 106.The higher bound is broadly acceptable for most operations, whereas a risk Ievcl towards the 10-4end of the band must be investigated and, if possible, Iowcred. British Gas have not yet assigned Tolerable Risk Levels to any casing failure modes. This will depend principally on the guidance provided by the UK regulatory and legislative authorities. Thus, QRA involved not only an assessment of the likely probability of failure, but also consideration of ihc effects (severity) of each type of failure. For example, a casing collapse at depth may not involve the cost or safety risks of a blowout. It may, therefore, be acceptable to assign a lower TRL to this failure type - say 104- than for a blowout. Non safety-risk events could be left to the discretion of the company, and based on the cost implications for this t}-pc of failure

63

APPLYING QUANTITATIVE

RISK ASSESSMENT TECHNIQUES TO CASING DESIGN

SPE 35038

Gaussian Linearisation. This proccdurc takes the mathematical formulation for the various input PDFs, and converts each to an expanded series of terms, which can then be Imearly combined 10 produce the desired formulation of .tf Resistance -[mad . . (2)

Implementation Issues In order to make the most of a Quantitative Risk Assessment implementation, a company should bc willing to do more than simply re-jig their design factors, A change in design procedures presents a number of potentially valuable opportunities to improve the whole design process, but also a number of challenges that must be met, Acceptance of the possibility of failure. In any cnginccrlng design the possibility of failure exists, It can be masked by safety factors, but the underlying possibility of a perfectly good design failing can be minimised, but never eliminated. The basic philosophy of a reliability-based approach requires the acceptance of the possibility of a failure. Once this has been achieved, it can bc a positive rather than a negative action. It disciplines the company into better contingency planning and mitigation strategies, and a comprehensive assessment of what sort of failures they really can and cannot afford, This can often pose diftlcultics for cnginccrs, management and governmental bodies Data Capture, Storage and Maintenance. If a decision is made to implement a full QRA methodology, then the quality of data is extremely important. It is certainly a truism to say that one only gets out what one puts into a Quantitative Risk Assessment, and appropriate systems must bc put in place in order to ensure that the QRA has the best quality and the Iargcst quantity of data possible. This may involve a closer relationship with steel suppliers. It may also mean ncw data rcquircmcnts from scrvicc companies for different data during and after drilling (c.g, a record of surface tcmpcraturcs, ECDS and logging information),
Changes in Design Philosophy. As previously stated, any consideration of a change in design practice should initiate a process of challenging and reviewing current practice, in order to improve the cftlcacy of the design, For example, questions like given current equipment capability and operating procedure, is casing burst through complctc evacuation in the wellbore still a realistic - and thcrcforc an dominant - load case? should be asked. However, a move to a new way of casing design must respect the Icssons learned during years of deterministic design, This ncccssitatcs not straying too far from conventional wisdom, and a series of incrcmcntal, technically justiticd changes. In practice, this means a great deal of field testing and practical evaluation of a QRA methodology. No changes to design practice will be carried out without Icgislativc approval.

where M rcprcscnts thc safety margin between the two quantities. The failure rate can then be produced from a probabilistic form of this equation. The process is relatively simple, although it requires a ncw formulation for every design case. and introduces some degree of error when the series are truncated.
Convolution Integrafs. Although more mathematically elegant, this method is highly abstract, and requires reformulation for each case. Mathematical expressions for each of the PDFs arc combined, and then the relative size of the failure area is determined, usually through numericai A failure rate can then be or explicit integration. dctermlncd. FORMAWRM Methodology. Standing for First and Second Order Reliability method, this tcchniquc converts the PDF into a standard form, performing a transformation on each one. The normalised PDFs are then placed into Equation 2 and a failure probability can he dctcrmincd from the size of M. This value of M equates to (}, a measure of systcm reliability, the inverse of failure ralc.

This method is highly abstract, but is advantageous in that it is accurate, quick, rcpcatablc, and requires very little rearranging for each different case.
Simulation. This tcchniquc uscs a computer to randomly select a value for each input variable, using the PDF. In this way, a single value for each input variable is obtained, and can be input into the standard equations. [f this process is carried out rcpcatcdly, then a picture of systcm response for all combinations of inputs (or at least a rcprcscntativc sample) is built up, and the proportion of faihrrcs can bc directly read from the results.

This process is conceptually very simple, requiring very little additional formulation and effort. However, due to the very large number of calculation repetitions required (often of the order of several million) a power!id computer is required to perform the task in an acceptable Iimc. Also, this time-consuming proccdurc must bc repeated for every assessment.

Computerisation. For a almost a pre-requisite, as peated hand calculations Iargc, This means that at tion must bc undertaken. presented.

full QRA implementation, this is the time and effort to perform reare, in most cases, prohibitively Icast some degree of computerisaSeveral choices arc, once more,

64

SPE 35038

KEILTY, RABIA

Stand-afone impfernentation. This IT solution is perhaps best suited to a full QRA method. It would also be the simplest to create. However, such a package would not be able to make fill usc of data available during standard design using a suite of programs. Integration. Another possibility wouldbc to nest the devclopmcnl as a part of a commonly available suite of well design programs. This would enable some commonality of data to be exploited (e.g. the basic engine from a slandard casing design module) and allow greater flexibility, by quickly determining the effect of a QRA-modified casing design upon other well parameters (e.g. cementing programmc)

the explanation lies in the fact that the QRA mathematics is highly non-linear, making it impossible to infer safety lcvCISfrom the relative magnitude of current S.F. S. Instead, it is necessary to engineer a flat response (i.e. uniform failure rates) into a Safety Factor-based design procedure through a QRA approach before a designer can be confident of safety \vith optimisation.
Applications

QRA has an application potential in a variety of areas: Development Wells. The potential for optimisation through improved design and usc of appropriate safety factors for this kind of drilling is clear. A few successful QRA applications in this area should cover implementation costs. Critical/Wildcat wells. Obviously, the uncertainty associated with these wells is large, and the application of QRA in this case will almost certainly be for safety assurance purposes it will allow the application of a more powerful design tool to underwrite overall well safety through a better knowledge of the areas of increased risk. However, in this area, as in all others, the QRA tcchniquc must bc married with the cxpcricncc and judgemcnt of the operators and designers. In an imperfect world, where data is not present in the abundance or with the assurance that we would like it, there is still a strong role for experience and sound cnginccring judgement, However, QRA will still give a good, although conscrvativc, indication of the risks of failure, given the nominal input data, whereas S.Fs do not. Fitness For Purpose Assessment. If, for one reason or another, it is decided that a certain casing weight and grade must k run, and this does not meet standard SF. rcquircmcnts, it is possible to carry out a more refined assessment on the problem, using QRA to determine if the design is truly acceptable or not
Project Status

Acceptance of Method. Duc to their complex nature, and the likelihood of a computerised method, the probabilistic elements of the calculation will be all but invisible to the designer using QRA on a day-to-day basis. It is therefore important that the limitations of QRA are not ignored, and QRA results arc factored into or compared with a more standard, easily understood design approach, This will ensure continued good design, whilst making QRA more acceptable to the dcslgncr
Example of Implementation The full QRA methodology has been applied rctrospcctivcly

to a casing design, largely for bench-marking purposes, and as a means of eliminating any problems with the procedure. The well was drilled in an area where the company had a considerable knowledge of the local geology and pressure regime. The basic API design equations were used4, and some nominal variables were input, due to a lack of data. This would mean that the probabilities of failure obtained would be higher than were the equations accurate, and a correct PDF assigned to each variable. A SORM-like approach was adopted, and the safety factors were converted into reliabilities for each casing. The results presented in Table 1 refer 10 the 7 inch casing. with a yield strength of 80,000 psi. The casing was set at a depth of 8,200 feet, against a predicted pore pressure of 4100 psi, using a mudwcight of 11 ppg. The table lists the conventional safety factors (obtained through the usc of standard design proccduresg ) followed by the reliability indices and a conversion into failure rates. The results show that, despite a set of safety factors that would not bc considered abnormally high, the actual failure rates for this particular casing are exceptionally low. In fact, it was ncccssa~ to mcludc the P index in order to demonstrate how far, in relative terms, each case is away from failure, as the probabilities arc minute in two of the cases shown. It can also be seen, by comparing the burst and collapse cases, that there is no direct relationship between safety factor and failure rates. Although the burst has a slightly smaller S F. than that of the collapse case, the probability of failure is actually lower. This result is counter-intuitive, and

The company has taken the work far enough to demonstrate the potential of a fully operational QRA methodology, and is ready to take the next step into practical implementation. The project is set to go ahead into 1996. Parallel QRA-based Designs. As a way of further demonstration, and a test of robustness of methodology, several full QRA-based casing designs arc planned, to run concurrently with the standard design process, This technique will allow senior management and engineers to judge first hand the benefits and challenges created by QRA, as well as highlighting a number of changes that will no doubt bc required m the methodology along the way. The development function will work closely with the front-line drilling function, in order to transfer this technology, and also to ensure that the correct quantity and ICVCISof technology arc transferred. 65

APPLYING QUANTITATIVE

RISK ASSESSMENT TECHNIQUES TO CASING DESIGN

SPE 35038

The Future. These benefits - along with the technology and skills required to realise them - arc not yet in place, and this will be the challenge for British Gas, and the rest of the industty over the next few years, There are, however, several areas that require particular attention, possibly through a Joint Industry Forum of some kind,
Full Computerisation. In order to make a QRA tool accessible and simple to operate, routes toward a full computcriscd QRA design tool should be identified and explored. Improved design equations, To gain the fill benefit of a QRA approach, there is a need for equations that predict the actual casing behaviour accurately. Some of the API equations governing casing behaviour contain conservatism specifically factored in to take account of the statistical variation in materials4g. This is obviously a good thing, when designing using safety factors, but the question is: are these the most suitable equations for a QRA design base? It is necessary to draw out implicit conservatism wit hin equations for tubular resistance, and challenge thcl r validity, for the following reasons:

System approach, Each casing joint cannot necessarily be considered in isolation, and they are connected in terms of probabilistic response, as well as physically, The approaches discussed here focus on a consideration of casing joints as components rather than a system. This is an important distinction, and this area requires further investigation. Casing coupfingx Thus far, it is the pipe bodies that have been investigated, but the issue of casing couplings and their etTect on system reliability, should also be addressed Conclusions

A casing design method incorporating at least some use of QRA is a move that will be demanded increasingly by both regulatory bodies seeking more responsibility and accountability from operators, and the companies themselves, seeking a better understanding of the factors governing their casing designs. British Gas is committed to further exploration and development of such a methodology, in order to prepare for such regulatory pressures, and to improve its casing design. The authors have demonstrated several benefits from a QRA implementation, and foresee a number of others: A better understanding of real load and resistance behaviour in casing design, Some cost savings whilst maintaining or improving the focus on safety. Safer HPHT drilling and completion operations, Improved post-analysis and fitness-for-purpose assessment capabilities. It is worth stating that casing design is by no means the only area that impacts upon well economics and safety, and should not always be considered in isolation from operating procedures, wellhead equipment and through-life considerations. QRA is not a panacea for well safety, Like all design aids, QRA is merely a tool, and cannot in itself be blamed for failure or bad design, It must be used with caution, and as a partner - not a substitute for - sound engineering judgement. The potential benefits from using such a tool - estimated to be as high as 30% of casing costss must be considered worth the outlay and effort The potential - in the limit - of Quantitative Risk Assessment is obvious. However, upon reflection, several dangers also become apparent, and one must recall the words of Mark Twain, quoted at the beginning of this paper. Any steps the industry takes toward this method must be carefully measured, and only taken after safety has been assured.

Some material properties and manufacturing processes have altered since the creation of the equations, For example, tolerances on diameter, thickness, and control over properties such as yield strength have improved drastically in recent years, In order to determine actual casing response for QRA purposes, a set of equations stripped of implicit conservatism arc required. Equations must refer to the limit states of the casing, without any statistical adjustments (e.g. the factor of 0.875 present in the API formula for casing burst). It has also been shown that the probabilistic response across the data space for the current casing design equations is not flat, and that the equations provide differing levels of redundancy for different input values (see Figure 5.), The ideal situation is, if a single equation approach is to be used, that redundancy (and therefore safety margins) arc uniform irrespective of casing dimensions etc.
Data gathering and collation. The numbr of wells drilled by the industry as a whole is fairly large, but work needs to be carried out on accessing this information and using it to reduce uncertainty in QRA. Sparsity of data in such areas as HPHT wells is an issue that must also be addressed through industry co-operation and collaboration, Methodoio~ improvements. The methods outlined for QRA have yet to be fully developed and tested for each of British Gas design load cases.

66

SPE35038
References
1.

KEILTY, RABIA

Construction Industry Research and Lrrformation Association. Rafionalisatjon of Safey and Serviceability
[~actor.~in .Ytmctural (I&es, CiRiA, 6Storey!~ Gate,

/.ondon, SWIP.?.4U
2.

(July 1977)

I.ewis, D]]., Brand, P.R. and Wlutney, W S : load and

paper OTC 7936, presented at the CMTshore Technology Conference, I iouston, ( I995).
3.

IIaugcn, ll.B and Wirschong, PIJ. PR{~BAl~ll,/,STIC;

Reprinted from:,bfAC1/INI;D ESIGN,April

17throughto June 12, 1975, Periton mc, Cleveland, 0hlo 44114 4 API Bulletin 5C1 Bulletin on tormulas and Calculations
for ~astng Tubing, Drrll Pipe and tine Pipe Properties, S111publication, (July 1989)

5 6.

API Spec 5A .YpeclficationJor

{Tasing, Tubing and Drill

Publ]catiors, (May 1987). Pipe, S111 Reeves, T B, Parfitt, S I [ 1. and Adams, A.J Casing
.S,vstem Risk ;lnalv.~is llsltrg Structural Reliabi/i~, SP] 25693, Procecdmgs of the SPWIADC Drilling Conference,

Amsterdam, (1chraary 1993)

-1 8.

Alpha Disasler,

The I Ion [.ord Odlcn: The Public lnqui~ info HMSO, I,ondon, (Nov 1990)

/he Piper

Veritas Scwm Systems: SEUM PROBAN - General

1lovik, Norway, (17 Jan. 1992).

Rabia, I{ : Fundamentals of (asmg Ile~ign, Graham & Trotman, Sterling I{ouse, 66 Wilton Rd, I.ondon SW 1V 11)}{ (1 987)

67

BURST

COLLAPSE

AXIAL

Safety Factor

T 17.36 4.86 31.38

p Value

Failure Probability

F
o
Tablel.
I I

6x1 O-7

o
S.F.'sfor7' casing

Comparison beMeen QRAandconventional

Load

Resistance

Resistance /~ Ii \ Load / I

S.F.

Figurel.

Standard deterministic design

Load > Resistafice (Failure)

Figure 2. Probabilistic design

Many Tests

Normal Distribution

Figure 3, Example of how a probability distribution is formed 68

SPE 35038

KEILIY, RABIA

Input PDFs
*A

Probability of Failure
Resistance /\

!1

, ,, /

/,
DESIGN ---EQUATIONS
,,/

/ Yield Strength -

4
,/ ,,
*

11)
Load i

t ; I I \ \

Kick Size

-Pore Pressure

Figure 4. The QRA design process

Ideal Risk response

Typical response

Increasing D/t Figure 5. Example of risk levels from casing design equations

69