Application of Probabilistic Reliability

Methods to Tubular Designs

Mike L Payne, SPE, and .John D. Swanson, SPE, Area Oil & Gas Co.
seE 19556
Summa..,. Standard tubular design methods are based on a specific margin (i.e., "safety factor") between the maximum ~ticipated
field load an~ the pu.blished rating q~ the tu~ular. This technique minimizes the risk of failure but promotes overdesign because of
the conservatJveness m the tubular ratm.g ~d m the assumed high-load case. To quantify the safety of a particular design, we developed
a new method that accounts for the vanatlOn of field loadings and tubular performance. Probability distributions for load and capacity
ar~ ~veloped o~ th~ ~asis ?f as~umed field-load histories and actual test results on tubular performance. Reliability design methods,
ongmally used .m CIvil engmeermg, are used to combine these distributions and to quantify the probability of failure. Results show
~at current ~eslgn factors do not pr?vide an effective reliability measurement. As a result of these limitations, true-cost/reliability deci-
sIons on desIgns cannot be made ~Ith conventional techniques. If reliability is quantified with the proposed method, decisions can be
made that pro~rly balanc~ economIcs, safety, and uncertainty. This paper should assist engineers who·need alternative design approaches

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
and managers mterested m a better understanding of how existing tubular design methods relate to risk exposure and optimal costs.

Background interval. Burst tests were not performed. Therefore, the 1.33 burst-
Total domestic drilling costs have averaged over $22 billion per design factor was not changed. This field testing of design factors
year during the past 10 years. Tubular goods represented about $3.5 resulted in a uniform lowering of previous design margins.
billion, 16% of these average yearly expenditures, and accounted In March 1955, Moody4 surveyed 38 companies and summar-
for the second largest percentage of total costs on almost all drill- ized the casing design factors being used by the industry. The sur-
ing projects. J Because of the magnitude of these costs, the need vey showed that about 70 % of the collapse designs were based on
to refme design methods is critical, but casing design methods re- design factors of 1.125, while about 17 % were based on factors
main controversial and inexact. Design factors and loading con- of 1.00. Tension design factors ranged between 2.00 and 1.60, and
siderations vary greatly within the industry. A historical review of burst design factors ranged between 1.33 and 1.00. In 1978,
design factors demonstrates an ongoing trend toward more exact Greenips cited 1.125 for collapse, 1.80 for tension, and 1.l0 for
designs and less conservativeness. burst. In 1986, Bourgoyne et aI. 6 cited 1.10 for both burst and
Before 1939, collapse designs were based on average failure pres- collapse and 1.60 for tensien. Thus, although some significant
sures and a design factor of 1.50. The source of the 1.50 design changes have occurred in isolated instances, design factors have
factor is unknown, but the API published setting-depth tables with remained essentially unchanged for many years.
the average collapse rating, the design factor of 1.50, and a stan- The premise of all casing design methods is to balance casing
dard collapse gradient of 0.5 psi/ft. Analysis of collapse data indi- costs with design reliability-i.e., the correct casing design for a
cated that minimum collapse pressure occurred at about 75% of given situation is not only reliable, but also economical. The void
average. Hence, when API revised the collapse ratings from aver- in current practice is the quantitative balance of design reliability
age to minimum, engineers reduced collapse design factors by 75 % and cost. The most economical casing design is achieved by find-
(from 1.50 to 1.125). This collapse design factor remains an in- ing the casing string that most closely exceeds a company's or an
dustry standard. In 1951, Hills 2 reviewed casing design practices engineer's specified design factors. Unfortunately, the basis for the
and conducted a survey of operator philosophies. He promoted the design factor is often not well-founded because design factors have
1.125 collapse design factor and cited tension design factors of 1.50 been handed down through the years and only rarely subjected to
for pipe-body yield strength (pBYS) and 2.00 for ultimate joint field or analytical verification.
strength. An average burst design factor of 1.50 with a range of Current design practice (see Fig. 1) shows the perceived relia-
1.l0 to 1.75 was also specified. bility of the string design vs. the design factor for an example set
In 1954, Saye and Richardson 3 discussed the field testing of of design factors. This figure shows the perception that any design
casing strings with lower-than-normal design factors. Results factor less than the minimum is completely unacceptable and has
no chance for success while all design factors at or above the mini-
showed that design factors could be substantially reduced. Before
mum are very reliable with a 100% probability of success. To gener-
this study in the Elk City field in Oklahoma, Saye and Richardson
ate such a drastic breakover between reliability and failure, the
used design factors of 1.60 in tension, 1.00 in collapse, and 1.33
probability distributions for field load and tubular capacity shown
in burst. They experimented with tension design factors by run-
in Fig. 2 would have to be assumed. Fig. 2 shows an assumed 100%
ning eight strings with design factors as low as 1.40. The strings
probability of occurrence at maximum field load and minimwn tubu-
designed with the 1.40 tension-design factor did not fail, but tests
lar capacity. This simplistic view of design variables, however, is
with lower design factors were not run and the 1.40 tension-design
inadequate. Consider the following dilemmas.
factor was adopted. This is even more remarkable when measured
I. A design factor of 1.20 must be more reliable than a design
reciprocating loads were added to buoyant string weight, reducing
factor of 1.10 and so forth, but no means to quantify how reliabili-
actual design factors from 1.40 to 1.28.
ty is improved exists. Also, if a design factor of 1.01 is as accept-
Forty strings were run with this lower design factor without in-
able as a design factor of 1.00, then how unacceptable is a design
cident to validate the results of the program. Collapse tests involved
factor of 0.99 or 0.98?
drilling additional rathole and then casing it off to allow applica-
2. Design factors :s 1.00 are used for collapse in certain circum-
tion of evacuation pressures in the well using packer assemblies
stances. A design factor of 1:00 implies no safety margin because
below the productive zones. Five wells were tested at collapse pres-
the rating is matched to the field load. However, even a design factor
sures greater than the rating until failure occurred or equipment
of 1.00 includes a safety margin because of the conservativeness
limitations were reached. Test results indicated that casing with de-
built into the rating and the load. If designs are based on hidden
sign factors 2:0.90 in the uncemented interval and 2:0.75 in the
margins, shouldn't those margins be quantified and explicitly han-
cemented interval sustained imposed loads without failure. These
dled in the string design?
results prompted the lowering of collapse design factors to 0.85
3. High design factors, such as a 1.8 for tension, are frequently
in the cemented interval while maintaining 1.00 in the uncemented said to account for effects not included in the load calculation, such
as bending, dynamic loads, triaxial considerations, wear, cement-
Copyright 1990 SocIety of Petroleum Engl.-IS ing loads, and wellbore drag. But if these additional loads are ana-

SPE Drilling Engineering, December 1990 299

 100 .. Certainty Maximwn FIeld Load

- -1.... ,._._._._._.-. l.ikeUboodof

i 0a:wTen<e
(Probahility Demity)
i "Safety"" or Deoign Factor
iI
i
i
__._._._._._._._._._._._._..!..~_._._._._._.Ji ...... -
Load or Capacity
I
CIrreoi ...... _
F< = 1.1 Fig. 2-Requlred load and rating probabilities to support cur-
Fb=l.l
Ft = l.6 rent design views.
01<.

Fig. 1-Success and failure Implications of current design

factors. _=_[L*>~J

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
lyzed with comprehensive tubular design techniques, 7 the design Load
factors should be reduced proportionately. In fact, load and stress
analysis have improved greatly in the past several years, but de-
sign factors have remained largely unchanged.

-
The assumed load and capacity distributions and the associated
design reliability illustrated in Figs. 1 and 2 are clearly inadequate.
The dilemmas presented here pose valid engineering problems that
cannot be addressed without a quantification of the reliability and
risks associated with a given casing design.
Fig. 3-Actualload and rating probabilities and definition of
Probabilistic DeSign Concepts risk.
Some level of risk is assumed in all engineering designs. One prin-
cipal goal of all designed systems is to ensure performance within cause it is based on initial ID yield instead of actual burst failure,
the budget. Designs are often formulated without complete or ac- the burst pressure resistance of casing is statistical in nature, long
curate information. Therefore, it is never practical or economical casing strings have a higher probability of failure than short casing
to claim absolute design dependability. The reliability of any de- strings, and significant wear can be tolerated before the probabili-
sign realistically is ensured only in terms of the probability that ty of failure becomes a concern.
the weakest point in the system will be adequate to withstand the Probabilistic analysis applied to casing performance recognized
maximum load. the influence of length on probability for string failure and showed
Traditional casing design practices compare an estimated abso- that burst designs are conservative because of the nature of the MIYP
lute maximum load to an assumed minimum capacity (see Fig. 2). burst rating. Limitations of Bradley's work were the application
A preassigned safety margin is maintained between the two values of burst load as a deterministic (fully known) quantity when the
to ensure the reliability of the design. This approach does not ac- load itself varies according to some frequency distribution, the use
count for the uncertainty of the governing design variables. Also, of constant burst loading along the string even though variation in
benefits from increased safety margins cannot be quantified. loading with depth provides a more general case, and the applica-
To model the true design problem more accurately, we realize tion of the technique to burst only.
that the expected load and given tubular rating for a particular de-
sign are defmed in terms of two random variables, c andl, where Probabilistic Design Application
c is the capacity of the selected tubular and l is the maximum load To apply reliability methods to casing design, appropriate proba-
imposed on this design during the well's life. The objective of the bility distributions for casing properties and expected load must be
design is to ensure that c> l at all times. This assurance is quanti- developed. In 1963, API tested about 2,900 casing samples ofvar-
fied only in terms of the probability of success, P s' or P(c> l). ious sizes, weights, and grades to determine a statistical distribu-
This probability represents a realistic measure of the design's relia- tion of collapse failures. 13 The collapse values were determined
bility. Conversely, P(c< l) defines the probability of failure, PI' by testing each sample to failure. These data were used to develop
for the specific design. Fig. 3 illustrates this probability of failure. API collapse ratings and are used here. More recent test programs
(c and l are assumed to be normal probability distributions.) Anal- have generated additional data that should be considered in future
ysis of Fig. 3 reveals that the potential for failure depends on the work. Current API ratings are based on a 0.50% probability of
means and variances of these distributions. failure, which means that the collapse rating is calculated at the
point on the test-data distribution where 99.5% of all specimens
Previous Probabilistic Work exceed the rating. Note that" although standard design methods do
Bradley 8-12 examined the probability of failure of worn casing un- not include probability considerations, the source of the rating is
der internal pressure. Data initially generated by the API13 were defined in probabilistic terms.
used to develop the probability distribution of failure for single joints The collapse-capacity probability distribution can be calculated
of 7-in., 26-lbm/ft, N-80 casing under various internal pressure directly from these data for the casings used in the testing program.
loads. The assumptions of Bradley's work were that mechanical Burst-capacity and tensile-strength distributions were also deter-
properties were isotropic and constant in each joint. He concluded mined. These two properties were calculated for each sample with
that as the casing string was lengthened, the probability of failure measured dimensions and material-strength values obtained from
for the string increased owing to the greater chance of including the testing program. To defme individual probability for failure dis-
a weak joint in the system. The probability of failure of a casing tributions, means and standard deviations were then computed for
string was then calculated from the probability of a single joint the 35 different sizes, weights, and grades evaluated (see Table 1
failure by use of the binomial distribution to account for the num- of Ref. 15). These data are then used to determine the probability
ber of joints in the string. Specific conclusions from Bradley's work of casing failure under load. Fig. 4 shows the probability of failure
included that the API MIYpI4 burst rating is very conservative be- for a joint of7-in., 26-lbm/ft, N-80 casing, assuming constant in-

300 SPE Drilling Engineering, December 1990

ternal (burst) and external (collapse) loads. Similar distributions
can be determined for other sizes, weights, and grades. 25
A distribution for the expected field-loading conditions must also
be considered. Current casing design philosophies use certainty of Probability
oIFaDure
the maximum load occurring during the well's life. We propose

/
(%)
a more practical method that accounts for the uncertainty of the
service conditions (see Fig. 5). The maximum load and its proba- 20

bility of actually occurring should first be estimated on the basis

of expected operations and historical data. The average load should C~LoadIIIa I
Bunt LoadIna
then be predicted as a percentage of the maximum load. These three I
I
variables define the probability distribution of the expected maxi- 15 i

mum load in the casing string. This distribution is then redefined

for any point in the well, creating a 3D load distribution.
As Fig. 3 shows, the overlapping of the two distributions for load
i
and capacity represents a qualitative measure of Pf, which depends 10
on the form of the two distributions, their relative positions, and
the associated degree of dispersion. This probability is calculated
I

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
by defining the safety margin for a given design, m, as c-t. 16 As-
suming that c and t are normal, the probability distribution of the
safety margin, m, is also normal and is defmed by a mean margin of
API RatIng- 5240 API RatIng - 7240 /
M=C-I ........................................ (1)
and by a variance (standard deviation squared) of
um2 =ue 2 +u/ 2 ................................... (2)
o +--+~. ./ -----,-,-+ ""F- -,--/J
........ --r----1

5000 6000 7000 8000 9000

for statistically independent c and t. We can then derive that the
probability of failure for the design is
Pf =4>( -M/um) = 1-4>(M/um), ....................... (3)
Fig. 4-Probablllty of failure for single Joint of 7-ln., 26·lbm/ft,
while the reliability (probability of success) is N-80 casing under Increasing pressure loads.
P s =I-Pf =4>(M/um), .............................. (4)
where 4>~cumulative distribution function for a normal random
variable with a mean of zero and a standard deviation of one.

- A __- ~
Assuming that the probability distribution for the casing proper- I .... ~ w.... JIIIId .... w.... _ ....
.. .. ..........- _. ---------"----
ties is constant for each joint and that the probability distribution
for load is approximated at each joint as a function of depth, we
~
can calculate the total reliability of the design by multiplying the ... _~c
probability of success for all joints:
ps(string) =P s(1)P s (2)P s(3)P s (4) .. . Ps(n) ............... (5a)
ne.-. (J
n
or Ps(string) = lIps (i); i=l, .......................... (5b)
........ ~

where i=individual joint and n=total number of joints.

Fig. 5-Superposltlon of actual load and capacity distributions
with current pressure design methods.
Example Reliability Calculations
Collapse. Full internal evacuation at some point in the well's life
is often assumed for the expected maximum collapse load for a
casing-string design. For example, this condition is frequently used
for surface and protective pipe because of catastrophic lost circu-
lation during drilling. Experience, however, indicates that these con-
tingency events occur rarely, if ever. For this example, the
maximum load is conservatively estimated to occur 2 % of the time
(in one out of 50 wells). Average collapse loads are normally ex-
perienced during cementing operations as a result of density differ-
ences between the cement in the armulus and the displacing fluid.
This load can vary, but is commonly 20 to 60 % of the expected
maximum load. Conservatively assuming that the mean collapse
load is 60% of the maximum design load and using the estimated
2 % maximum probability generates a normal probability distribu-
tion for collapse loading. This approximated load distribution is
defined by a mean of
Ie =0.6L emax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)

and a standard deviation of

ulc =O.4L cmax /z, ................................... (7)
where Lemax = specified maximum collapse load and z=standard
Fig. 6-Probabliity densities for collapse load and capacity
normal deviate (number of standard deviations between the mean along wellbore.
and maximum loads). z is defined by the occurrence of the maxi-

SPE Drilling Engineering, December 1990 301

 mum load (2% probability in this case). With the stated assump-
ReIIabIIity(%) tions, the maximum load completely defines the normal distribu-
tion for load.
100 Once a single probability distribution for collapse load is esti-
/' __":~:=//::: C_,,~~.-.~.~~.~::-.::::= mated, distributions are calculated for each joint in the string to

. //Q: ..-~. . . . . . .-
,I / 1 % ProbabilIty of Maximum Load OcClUTence
reflect the reduction in collapse load from total depth (TD) to sur-
face resulting from the decreasing hydrostatic gradient. The load
distribution is assumed to be constant throughout each joint and
equal to the calculated distribution at the center for manipulative
ease. The result of generating collapse-load distributions as a func-
tion of depth is a 3D load distribution (Fig. 6). The product of the
; /' 5 % Probability of Maximum Load OcclUTeoce
9Z ! reliabilities of the individual joints in the string is then computed
i to estimate the total design reliability as stated in Eq. 5.
.... Fig. 7 shows the calculated reliabilities for a 1O,OOO-ft design
i of 7-in., 26-lbm/ft, N-80 casing vs. the standard collapse design
88
t
/ factor. Mean and standard deviation values for capacity (7,245 and
737 psi, respectively) and the API collapse rating of 5,410 psi were

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
,I used to develop the plot. The collapse design factor is defined con-
i ventionally as the API collapse rating divided by the maximum load
I

84
i at TD. In addition to the base case (2 % probability of maximum
i load occurrence), Fig. 7 shows two sensitivity cases, one at 5%
!
I
probability and one at 1 % probability. If the maximum load (full
evacuation) occurs on only 1 % of the wells, Fig. 7 indicates that
f a collapse design factor of 0.89 provides a string reliability of
80 i
±97%. If the maximum load probability increases to 5%, how-
ever, the design factor would need to remain > 1.05 to maintain
I comparable reliability.
76~j____-.____-.______.-__~ Fig. 8 further illustrates the relationship between the standard
collapse design factor and actual string reliability according to this
0.9 1.1 1.1 1.3 model, showing the improvement in reliability that results from in-
creasing the standard design factor. The results indicate that at the
Collapse DesIgn Factor
low design factors of 0.8 and 0.85, significant reliability improve-
ments are achieved by moving to the next highest design factor.
Fig. 7-Collapse design reliability for various probabilities of At design factors of 0.9 to 1.05 (depending on maximum load fre-
maximum load occurrence.
quency), however, no significant benefits are achieved by further
increases. Additional refinement of the assumed load distributions
are required to define specific recommendations for changes to de-
sign factors. However, because results from this example analysis
Pereent Increase
in ReUability
indicate high reliability at design factors < 1.00, current collapse
design methods need to be re-evaluated.
60

Burst. Load and rating distributions are also generated for burst
\ loadings as a function of depth in the well. Fig. 9 shows the 3D
\ distributions for this example under a burst design load. In this case,
\ loading is held constant with depth, which is indicative of a pro-
\

45 \ duction casing design based on a tubing leak on top of a packer

\\ fluid of density equivalent to the pore-pressure backup. Again, 7-

in., 26-lbm/ft, N-80 casing is used with the load distributions gener-
\ ated by the 60% relationship between mean and maximum load and
assumed probabilities of maximum load occurrence of 1, 2, and

30
\\
\
5 %. For burst capacity, Bradley's distributions for three different
ratings are used.
Barlow ID Yield.
\\ p=2hSy lde . . ..................................... (8)
\ Barlow ID Ultimate.
\ . . . . - 5 % Probability of MaxImum Load OcClUTence

\ \ p=2hSu lde • . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)

\,. ~ 1 % Probability of MaxImum Load OcClUTence
15 ASME Ultimate.

\ ,. V\ ~\
1 % Probability of Maximum Load OcClUTence
p=2hSu l(de -0.8h). . ............................. (10)
Other estimates for failure pressure are available, 17 but the above

,
O+--------r-----=~~~--_.Ir====---.I
\~:"~~"-~---- estimates of burst failure pressure are adequate to illustrate the in-
herent conservativeness of burst design.
The three rating distributions are defined by a mean and stan-
dard deviation of 9,294 and 609 psi, respectively, for Barlow ID
0.75 0.875 lollS US Yield; 11,770 and 894.psi, respectively, for Barlow ID Ultimate;
and 12,275 and 935 psi, respectively, for the ASME Ultimate. 15
Collapse Design Factor
The API MIYP rating for standard design purposes is 7,240 psi. 14
With maximum burst load defined by the API MIYP rating divid-
Fig. 8-Dlmlnlshed Improvement In collapse reliability at In-
creasing design factors. ed by the burst design factor and load distributions generated from
that maximum, burst reliability can be calculated as a function of
302 SPE Drilling Engineering, December 1990
 Rtliability(%)
100

80 ~::;~:=:=
/ ' .. __ ........... t.ooo_
60

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
Fig. 9-30 probability densities for burst load and capacity 40
along wellbore.

RtUabWty (~)
20
100

80
o -t----,I---,----.---.----,
0.750 0.875 1.000 1.125 1.250 1.375

Burst Design Factor

60
Fig. 11-Burst design reliability using the Barlow 10 Ultimate
pressure rating.

40 criterion and the assumed minimum wall tolerance of -12.5% is

again demonstrated.

Tension. The standard tension design technique is to calculate

20 buoyant string weight and then use a tension design factor of 1.5
to 1.8 against the weaker of either PBYS or joint strength. Unlike
the burst and collapse design loads, there is great certainty that this
1001d condition will occur. In fact, there is virtual certainty that
buoyant weight will be exceeded because of friction loads between
0.750 0.875 1.000 1.125 1.250 1.375
the casing and wellbore, viscous drag from mud. etc. This situa-
Burst Design F.nor tion demonstrates that standard tension designs are based on mini-
mum field loading (Fig. 13). This assumed condition will be
Fig. 10-Burst design reliability using the Barlow 10 Yield exceeded consistently in field operations, which conflicts with burst
pressure rating. and collapse methods and results in the design factor being blindly
weighed against an unknown load distribution. Given this unusual
the standard burst design factor. In the case of assumed constant but standard tension design method, we can still develop a reliabil-
loading with depth, the single-joint success probability is constant ity calculation.
and the product function in Eq. 5 becomes a single probability raised The rating distribution for the tension example is based on PBYS
to a power equal to the total number of joints. for the example 7-in., 26-lbm/ft, N-80 casing string. The PBYS
Figs. 10 through 12 show the results from the calculations for distribution is defmed by a mean of 704,354Ibf and a standard devi-
the rating criteria of the Barlow ID Yield, Barlow ID Ultimate, ation of 53,934lbf. API's rating based on minimum yield is 604,000
Ibf.14
and ASME Ultimate pressure ratings, respectively. Each figure
The tension design load distribution is developed by first divid-
shows the sensitivity to 1,2, and 5% maximum load occurrence
ing the API PBYS by a designated tension design factor. This es-
probability. Fig. 10, based on the ID yield rating, shows that even
tablishes the minimum on the distribution curve shown in Fig. 13.
with a 5 % probability of maximum load occurrence, a burst de-
For the maximum load on this distribution, a constant factor is
sign factor of 1.22 provides 97 % reliability, while for a 1 % chance assumed for the ratio of maximum to minimum load. For exam-
of maximum occurrence, a design factor of about 1.02 would do ple, with a tension design factor of 2.0, the minimum design load
the same. These results are enlightening because ID yield is an ex- is found to be 302,000 lbf. With a factor of 20 % for wellbore drag,
tremely conservative estimate of actual burst failure. Use of the the maximum load is assumed to be 362,400 lbf. The mean for this
ASME rating, which is intended to estimate actual failure pressure distribution is assumed to occur halfway between the minimum and
more closely, and a 1 % maximum load probability results in burst maximum (in this case at 332,200 Ibf). The last parameter required
design factors between 0.75 and 0.80 that provide 97 % design relia- to define the distribution is the frequency of loads outside the max-
bility (Fig. 12). In summary, Bradley's results were refined by in- imum or minimum bounds. It is assumed that no more than 5 %
clusion of a probability distribution for the burst loading. Moreover, of the loads are greater than the maximum and that no more than
Bradley's observation of the extreme conservativeness in the API 5% are less than the minimum. This defmes a standard normal devi-
MIYP rating that resulted from the use of ID yield as the rating ate, z. of 1.645 and results in a standard deviation for load of 18,358

SPE Drilling Engineering. December 1990 303

 Reliability (%)
100

~~,O:"~~M~~~
80
/ ~ 2 % Probability of Maximum Load Occurrence

/ 5 % Probability of Maximum Load Occurrence

60 ,I

-.-ofTensioD~
tor eadt Joinl

40

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
Fig. 14-30 probability densities for tension load and capac-
Ity along wellbore.

20 Reliability (%)
100 r//"/'·'/'·············..· ......····· ..· ..· ..·----·--

90
O+------r----~I------~I----~I------,~ !
0.750 0.875 1.000 1.125
Burst Design Factor
1.250

Fig. 12-Burst design reliability using the ASME Ultimate pres-

1.375
80

!
!

I~T __. . m...,._

sure rating. 70 I

I
60 I
IOO,",C"-ty
I
-L 50
I
(J
, i
I
40
I
!
Fig. 13-Superposition of tension load and capacity dIstribu- 30
I
I
tions with current design method.
I
2°i-----~----1,----.,-----1,-----,---~
lbf. A family of load-distribution curves is then developed for differ-
ent tension design factors. 1.35 1.5 1.65 1.8 1.95 2.1
The final step in generating the full load distribution is to ac- Tension Design Factor
count for the variation in tension with depth. For this example, relia-
Fig. 15-Tension design reliability using the API PBYS ten-
bility will be calculated for a 15,OOO-ft string set in 14-lbm/gal mud.
sion yield rating.
Buoyant weight is used to define the tension design load at sur-
face, and tension load is reduced down the string according to the
weight of the pipe. Fig. 14 shows the resulting load and capacity ings are often limited by the connection, not the pipe body. Data
distributions. are readily available for 8-round and buttress tension performance,
The results, shown in Fig. 15, indicate that virtually 100% relia- but scarce for leak resistance. With premium threads, ratings vary,
bility (99.8%) is achieved with a tension design factor of 1.50. This with full-strength joints rated higher than the pipe body and special-
is remarkable because the 20% increase in buoyant load is consid- clearance or flush connections rated at some fraction of the pipe
ered conservative, particularly for straight holes. Furthermore, the body in tension and pressure. Moreover, combined load ratings are
capacity distribution based on yield is also conservative compared not available for most connections.
with one based on the ultimate strength of the material, which in- Finally, certain phenomena that still defy predictive models make
dicates failure. application of reliability methods difficult. Most notably, condi-
tions that lead to drilling wear must be better understood and drill-
Future Considerations ing wear rates must be better quantified so that they can also be
The reliability methods presented here can be easily extended to accurately included in-the design process.
biaxial and triaxial analysis by use of the Monte Carlo technique
or other random-variable-generation techniques. Conclusions
Additional calculations are also required to account for connec- 1. Conservativeness is inherent in tubular design. It begins when
tion performance. With API connections, tension and pressure rat- severe, anticipated load conditions are allowed for and increases

304 SPE Drilling Engineering, December 1990

when the ratings, such as MIYP and PBYS, represent the start of Authors
material yielding and not actual material failure.
2. API collapse ratings appear to be the least conservative of the Mike L. Payne,
casing properties examined because they are based on actual failure Coordinator of
data rather than material yield. Drilling Computing
3. Both burst- and tensile-strength ratings are based on minimum at Arco 011 & Gas
yield strength and the onset of yielding. The difference between Co. 's Dallas head-
yield and failure is substantial in most cases. quarters, oversees
4. For burst and collapse designs, the key variable is the maxi- development and
mum expected load usually generated during an unlikely event. support of com-
puter systems for
However, tension designs are based on a relatively certain mini- drilling operations.
mum anticipated load. The result of this awkward design approach He previously was
is a large tension design factor that is likely to be excessive. Payne Swanson a senior drilling en-
5. The methodology presented in this paper allows engineers to gineer for special
quantify the risks inherent to alternative designs. In the past, ar- proJects at Arco's Southern Dlst. In Houston and a conSUlt-
bitrary increases in design factors have often provided "peace of ant at 011 Technology Services and Enertech Engineering &
mind" at an unknown expense. This technique provides a method Research. Payne holds an MS degree In petroleum engineer-
Ing from the U. of Houston and a BS degree In mechanical

Downloaded from http://onepetro.org/DC/article-pdf/5/04/299/2628593/spe-19556-pa.pdf by Halliburton Energy Services Group user on 03 May 2024
to approximate the associated increase in reliability.
engineering from Rice U., where he Is a PhD degree candi-
date In mechanical engineering. ,John D. Swanson Is an In-
Recommendation. ternal control consultant for Arco 011 & Gas Co. In Dallas.
1. To improve future casing design practices, better data collec- Swanson previously was a drilling engineer In Houston, Den-
tion must be incorporated to evaluate actual casing loads during ver, Midland, and Dallas. He holds a BS degree In chemical
a well's life. Quality statistics are also required to develop extreme- engineering from the U. of Colorado and Is currently pursu-
Ing an MBA degree.
load-occurrence frequencies.
2. Accurate failure cost, frequency, and cause data are needed.
Evaluating the efficiency of a casing design program is impossible 4. Moody, W .C.: "Survey Report on Casing-String Design Factors," Mid-
Continent District Study Conunittee of Casing Programs, Drill. & Prod.
without failure data retained. Also, predicting the total consequences Prac., API, Dallas (1955) 154-64.
of failure without good historical data is implausible. 5. Greenip, J.F.: "Designing and Running Pipe-Part 2, Optimum Casing
Design Program Stresses Economy," Oil & Gas J. (Oct. 16, 1978)
Nomenclature 77-86.
e = random variable describing tubular capacity 6. Bourgoyne, A.T. et al.: Applied Drilling Engineering, Textbook Ser-
ies, SPE, Richardson, TX (1986) 2, 335.
C= mean capacity
7. K1ementich, E.F. and Jellison, M.J.: "A Service-Life Model for Casing
de = OD Strings," SPEDE (April 1986) 141-52.
F b = burst design factor 8. Bradley, W.B.: "The Effect of Casing Wear on the Burst Strength of
Fc = collapse design factor Casing: Part I-Joint Leakage," paper ASME 75-PET-ll presented
F/ = tension design factor at the 1975 Annual Petroleum Mechanical Engineering Conference, Tul-
h = pipe wall thickness sa, Sept. 21-25.
9. Bradley, W.B.: "The Effect of Casing Wear on the Burst Strength of
I = random variable describing tubular load
Casing: Part 2-Statistical Burst Strength of Worn and Unworn Casing
I = mean load Strings," paper ASME 75-PET-27 presented at the 1975 Annual Pe-
Ic = mean collapse load troleum Mechanical Engineering Conference, Tulsa, Sept. 21-25.
Lcmax = maximum collapse design load 10. Bradley, W.B.: "Here's How Casing Wear Affects Joint Leakage,"
m = random variable describing design safety margin Oil & Gas J. (Dec. 29, 1975) 170-73.
M = mean safety margin II. Bradley, W.B.: "How Wear Affects Casing-Burst Strength," Oil &
Gas J. (Jan. 5, 1976) 85-88.
p = internal pressure rating
12. Bradley, W.B.: "Casing-String Length Affects Failure Probability,"
P = probability Oil & Gas J. (Jan. 12, 1976) 102-08.
PI = probability of failure for string 13. API Casing Collapse Data, API, Dallas (1963).
P s = probability of success for string 14. Bull. 5C3, Formulas and Calculations/or Casing, Tubing, Drill Pipe
Psi = probability of success for single joint and Line Pipe Properties, fifth edition, API, Dallas (July I, 1989).
Su = material ultimate strength 15. Payne, M.L. and Swanson, J.D.: "Application of Probabilistic Relia-
bility Methods to Tubular Design, " paper SPE 19556 presented at the
Sy = material yield strength
1989 SPE Annual Technical Conference and Exhibition, San Antonio,
Z = standard normal deviate Oct. 8-11.
CTc = standard deviation of capacity 16. Ang, A.H-S. and Tang, W.H.: Probability Concepts in Engineering
CTI = standard deviation of load Planning and Design, John Wiley & Sons Inc. (1984) I, 178-79.
CTlc = standard deviation of collapse load 17. Payne, M.L. and Hurst, D.M.: "Heavy-Wall Production Tubing De-
CT m = standard deviation of safety margin sign for Special-Alloy Steels," SPEPE (July 1986) 289-302; Trans.,
AIME,283.
Acknowledgment SI Metric Conversion Factors
We thank the management of Arco Oil & Gas Co., a division of ft x 3.048* E-Ol rn
the Atlantic Richfield Co., for permission to prepare and publish gal x 3.785412 E-03 m3
this paper. in. x 2.54* E+OO cm
Ibf x 4.448222 E+OO N
Reference. Ibm x 4.535924 E-Ol kg
I. "Oil and Gas Wells Drilling Expenditures," Oil & Gas J. Energy Data- psi x 6.894757 E+OO kPa
base (July 1988).
2. Hills, J.O.: "A Review of Casing-String Design Principles and Prac- • Conversion factor is exact. SPEDE
tice," Drill. & Prod. Prac., API, Dallas (1951) 91-107.
Original SPE manuscript received for review Oct. 9, 1989. Paper accepted for publication
3. Saye, J.E. and Richardson, T.W.G.: "Field Testing of Casing-String Aug. 20, 1990. Revised manuscript received Aug. 9,1990. Paper (SPE 19556) first presented
Design Factors," Drill. & Prod. Prac., API, Dallas (1954) 23-28. at the 1989 SPE Annual Technical Conference and ExhibHion held in San Antonia, Oct. 8-11.

SPE Drilling Engineering, December 1990 305