You are on page 1of 12

.

WI m

I Society
ofPetroleum
Enginears
I

SPE 28298

Quantifying Stuck Pipe Risk in Gulf of Mexico Oil and Gas Drilling
A.P. Wisnie, * Conoco Inc., and Zhiwei Zhu, U. of Southwestern Louisiana
●SPE Member
A
~hl lW.SooktY
d Petrokum Engineer%k.
TM pepw wae pmperedtor pmeemetionat the SPE SSthAnnueiTwhnkal Conferenceand Exhibitionheld In New Orleens, LA, U.S.A., 25-2S September1SS4.

ThLepeper wee wleoted for prewntetkm by en SPE ProgmmCommitteeMOWIng reviewot Intormetkmoontelnedin en ebetmcteubmHtedby the euthof(s).Contentsot the peper,
M-~. b notL-I fevkvd by the SOC+BO ot PeIrokum Engheere end are eubjwt to wrrwtbn by the author(s).The meteriel,es presented,doee not nweeeerily reflwt
--~~~~~kumEngi~mtti*,m~ . Pepem preeentedat SPE meetlngeare subieotto Publkationreviewby Edhxiel CommHteesot the SooietY
ofPetroleumEngineem.Permidon tO*k-titOen ~dmnt~~tie.liMr*- mytib~. ~~tiMtin~~m
ot where end by whc+nthe peper ie pmaented.WrHeLibrerien,SPE, P.O. Sox SSSS3S,Rkhardwn, TX 7WSS-3SW, U.S.A. Telex, 1SS24SSPEUT.

RBSTRACT return and low risk. While this concept has


worked very well for the revenue side of the
Stuck drillpipe is one of the most costly NPV calculation, little work has been done to
risks of exploitation drilling. However, assess the variability of the cost side of
few, if any companies incorporate this risk the equation. Risk in exploitation drilling
into their pro ject evaluation economics. is becoming increasingly significant as
This paper describes a method to quantify reserves are becoming more scarce and
such risks. The method involves gathering technically demanding to access.
data from over 311 cases drilled in the Gulf
of Mexico. From this data, Logistic One of the greatest risk factors in
Regression was used to build stuck pipe exploitation drilling is the occurrence of
probabilityy models for both water-based and stuck drillpipe. Stuck drillpipe occurs when
oil-based drilling mud. These models, which the drillstring cannot be moved, either in or
measure both the probability of getting stuck out of the hole. If a stuck drillstring
and freeing the drillpipe once it becomes cannot be freed, the hole is plugged back
stuck, are incorporated into a Monte Carlo above the trouble spot, and the lost section
.
simulation, which yields a log-normal redrilled. stuck “ occurrences
distribution of costs for a hole section of ~~nerally make up the la~g~st portion of
a given design. This distribution of costs unanticipated drilling costs for oil and gas
can be incorporated with exploration risk operators. As an example, through the years
methods for a project evaluation technique of 1985-1988, British Petroleum incurred an
that incorporates cost (mechanical) risks as average stuck pi e cost of over $30 million
well as revenue (exploration) risks. dollars per year F.

INTRODUCTION The purpose of this study is to analyze the


occurrence of stuck pipe from a risked cost
The oil and gas industry has used the concept perspective in order to improve oil and gas
of risk economics for several years’=. This exploitation investment decisions. The study
concept entails assigning a probability of will begin by evaluating a stuck pipe
success to finding hydrocarbons and database from a statistical perspective.
discounting the expected revenue stream by an With this data, the purpose is to form a
according amount. This concept has helped model which predicts both the probability of
companies in ranking high potential, high stuck pipe and the ability to free stuck pipe
risk projects, with those that are moderate from a given set of technical parameters.

References and illustrations at end of paper


.

2 QUANTIFYING STUCK PIPE RISKS IN GULF OF


~E~~CO GIL ~~~ GAS mQTT.T.TN~-
“*.*-.Z--A. - SPE 28298

Once this model is constructed, work will Clark et al also used operational variables
~ontinue ~GwaE~= fGEiRiil~ a
P.*1 -
~~~bt= -=-AU
a,,mh
----- g= ~~e ~Qrn~e~ Df short trips and the
simulation of a given drilling situation. twenty-four hour rate of penetration in their
The cost distribution from the Monte Carlo models.
simulation can then be combined with the
exploratory risk distribution for a more Howard and Glover16 recently published their
accurate project evaluation. work for the Drilling Engineering Association
(DEA) . This work compiled information from
REVIEW OF THE LITERATURE several different operators in an attempt to
check previous research and create new oil
Several studies have been performed to and water based models. The model building
identify those parameters that are associated methodology is similar to that of
with the occurrence of stuck piwo%67&%lo It Kingsborough et al.
was not until fairly recently, however, that
authors began to use statistical analysis to Most recently, Biegler and Kuhni’ presented
quantify stuck pipe decisions. Harrisonll their work which builds upon the work by
first demonstrated that a quantitative Kingsborough et a2. Through the creation of
approach could be applied to stuck pipe engineering parameters relating to the
fishing decisions with the use of subjective different causes of stuck pipe (ie iieie
---k-k4 1:+:-- m,,-u=e&G*”*,
pJ.wuuu&AAb&ea *“A A--; m4mm +Vea =m=l,,=+a
*.-- u..-*J”&”. cleaning: differential sticking~ etc)? the
authors use discriminant analysis to create
Lovelz was the first to use historical data canonical functions for mechanically stuckc
to create a model that predicted the success differentially stuck, and non-stuck
rate for freeing stuck pipe. Love’s model situations. Through the use of Bayesian
was built using a trial and error method of statistics, the authors define isoprobability
maximizing the correlation coefficient curves over the two canonical function plane.
between a group of predictor variables (the They further illustrate the use of their
stickiness factor) and the success rate of model in optimizing various parameters such
freeing stuck pipe. Although an excellent as mud weight and flow rate to minimize stuck
study for predicting the ability to free pipe pipe probability.
once stuck, Love emphasized that this
“stickiness factor” would not necessarily METHODOLOGY
predict the onset of stuck pipe.
It was the authors’ initial intent to use
More recently, Shivers and Domangue13 previously published stuck pipe probability
published Texaco’s version of a stuck pipe models in quantifying the risk of stuck pipe.
fishing decision methodology. Although this However, this was not done for several
!nQdel is rnaEhernaEically less rigorous than reasons.
Love’s work, it incorporates the placement of
spotting fluids into the analysis, which is First, with the exception of Shivers and
felt by many to be a very important factor Domangue’s and Love’s work, none of the
for pipe-freeing success. actual predictive models have been published
in the publicly available literature.
The first work tO actually attempt to predict
the onset of stuck pipe was completed by Second, almost all of the works have relied
Kingsborough et all’ (later published as on some form of discriminant analysis which
Hempkins et ails). This study incorporates is more of a tool for classification than for
data from 131 stuck pipe cases (all in water predicting probabilities. According to
based mud) plus daily drilling data from at Hosmer and Lemeshowl*, discriminant analysis
least three days prior to the incident. is sensitive to the assumption of normality
Discriminant Analysis was used to distinguish and will especially tend to overestimate ths
the cases as not stuck, mechanically stuck, magnitude of association for dichotomous
and differentially stuck. independent variables (of which there are
several in this analysis).
Clark et al” used a similar data collection
procedure with different statistical methods Finally, the method of data collection and
to build a model for oil-based mud. Their the results of the models are oriented
procedure used a stepwise discriminant towards the monitoring of stuck pips
analysis procedure for variable selection and probability on a day-to-day basis. While
logistic -regression for the final model. this may be of great help to drilling

“ Clark, D., Kubena, E ., Carter, T., and Bible, H.E., “Stuck Pipe Models for South Pass 75
and Ewing Bank 305”, CONOCO RESEARCH REPORT # 104-421-1-89, May 1989.

70
.

SPE 28298 A. P. Wisnie and Z. Zhu 3

personnel during the execution of a drilling the mathematical result of other independent
project, it becomes cumbersome when modelling variables (ie the differential sticking
the potential cost outcomes prior to the coefficients ).
drilling project. Specifically, the use of
day-to-day predictive models would require Altogether, data was gathered for over 311
modeling every drilling day. Although hole sections, 150 water-based cases and 161
attractive, the computing time and power oil-based cases. Over 36 variables were
necessary to generate a sufficient data set used in the analysis of the water based mud
would put this problem beyond the capability model, while 40 where used in analyzing the
of PC’9. oil based model.

The approach taken for this paper was to As a preliminary step, a univaritate logistic
gather data for entire hole sections as regression fit of each of the variables was
opposed to gathering data on a daily basis. made. The results of this analysis are shown
The methodology was to choose data from wells in tables 1 and 2 for the water- and oil-
drilled within the last ten years. A data based models, respectively. The general form
case was collected for two conditions: a for the logistic regression equation is:
stuck pipe incident or a successfully drilled
hole section. For example, if a hole was 1
drilled without a stuck pipe incident, one
data point was gathered indicating conditions
at TD. If pipe became stuck on two separate
occasions prior to reaching TD, then three
data points were gathered for that particular where:
hole section, one on each of the stuck pipe
days and one at the completion of the hole ~=intercept+coefficent*variable
section. On instances where TD was not
successfully reached (ie the well was stopped Thus, a negative coefficient indicates that
short of TD for mechanical reasons or the
increasing the value of the independent
well was junked), a “success” data point was variable will increase the forecasted odds of
not gathered. getting stuck. The opposite holds for a
variable with a positive coefficient. As can
A binary dependant variable (the logit) was be seen from Tables 1 and 2, the sign of the
indicative of the success of the hole coefficients agree with generally held
section. A “O” indicated a successfully TD’d
principles as to the causes of stuck pipe.
hole section, while a “1” indicated a stuck
pipe condition. Similarly, for the stuck
The last column in Tables 1 and 2 is the chi-
pipe freeing model, a “1” was tabulated for
squared probability relationship of the
a freed drillstring, while a “o” was dependant and independent variable.
indicative of a case where the drillpipe
Statistically, this represents the
could not be freed.
significance of the variable in the
univariate model. In layman’s terms, one
The independent variables were included in
might interpret this number as the
the analysis on two conditions. First, the
probability that the association between the
variable had to have an apparent causal
dependent and independent variable is due
relationship with sticking or freeing strictly to chance as opposed to a causal
drillpipe. Second, the variable had to be
relationship.
available in every case in order to solve the
multivariate problem. Thus, potentially
The next step involved building a
significant variables such as torque and
multivariate model. Three criteria were set
PPAm were not included simply because they
in determining the best possible model. In
were not available on consistent basis.
order of importance, it was judged that 1)
the model had to make sense from an
Initially, it was hoped that one model could
engineering perspective, 2) it would attempt
be created which would combine all water- and
to account for as many of the different
oil-based cases. However, since many of the
causes of stuck pipe while at the same time
properties for oil- and water-based mud are
staying within reason from a statistical
unique to their respective type, it was
perspective, and 3) it would attempt to
necessary to create separate models for each.
maximize the predictive accuracy of the
models as indicated by the goodness-of-fit to
Some of the variables were thought to have a
the current data. - -
multiplicative effect with other variables.
Others were thought to have a non-linear
In order for a model to be acceptable from an
relationship to the dependant variable. For
engineering perspective, the effect of the
these reasons, some independent variables are
variables on the model should emu 1ate

71
.

QUANTIFYING STUCK PIPE RISKS IN GULF OF


MEXICO OIL AND GAS DRILLING SPE 28298

conventional wisdom. Looking at the the scope of this paper, it suffices to know
univariate results, one would not anticipate that a perfect fit between model and data
a problem, as almost all of the coefficients would yield 100% concordant pairs.
have the right sign and are of reasonable
magnitude. With the interaction of A stepwise logistic regression procedure was
variables, however, these coefficients will employed. Limits were set for a less than .2
change in multivariate context. Those chi-squared p value to warrant inclusion into
variables that appeared counter to the model and .3 chi-squared p value to
conventional wisdom would be subject to justify elimination from the model. These
scrutiny and most likely eliminated if a screening values are based upon works b
plausible statistical and engineering %endel and Afifi21 and Michey and Greenland x
explanation for their presence could not be which show that more traditional levels (such
made. aa .05) often exclude variablea known to be
important. Tine results of the StSpWiSe
With respect to the second objective, the procedures are encouraging, with a 91.6%
mode 1s would represent as many of the concordant value for water-based model and a
different stuck pipe causes while still 84.4% concordant value for the oil-based mud
maintaining a reasonable level of statistical model. The variablea and their coefficients
significance for each of the variables. This are ahown in Tablea 3 and 4 for water- and
involved some judgement and subjectivity on oil-based mud respectively.
the authors’ part. Ideally, one would like
to include all of those variablea that were The next step involved running Pearson’a
statistically significant on a univariate Correlation Matrix to check for
level in the multivariate model. However, multicolinearity. This procedure was used to
conventional multivariate regression model check the association among each of the
building wisdom holds that one should attempt independent variables in the mode 1.
to obtain the best possible predictivity with Correlation values range from 1 (for a
the fewest possible variables. The rationale perfect positive correlation), to O (for no
for this is that one reaches a point after correlation), to -1 (for a perfect negative
adding a certain number of variables where correlation ). For this study, it was deemed
additional variables contribute very little desirable that each independent variable in
to the overall predictivity of the model. At the final model have a correlation
the same time, the addition of each new coefficient between .6 and -.6 with all the
variable affects the coefficients (and the other independent variables in the model.
statistical significance) of the other
variablea in the model, thus distorting the With the water-based mud model, one variable
effects of the important variables. Another (measured depth) had fairly high correlation
concern with including a large number of coefficients with several of the variables (-
variables ia the problem of multicolinearity. .63 for hole size, -.61 for API water loss,
Multicolinearity occurs when two or more and .715 for the natural log of days in open
variables in a model correlate very closely hole) . There were no multicolinearity
with one another. One example, from a stuck problems with the oil based model.
pipe perspective, might be hole size and
flowrate. The problem occurs in that the The final step involved judging the model on
effects of one variable become an engineering basia. For the water-based
indistinguishable from the other(s), and thus mud model, it was disturbing to see that the
the process of asseasing coefficients becomes variable Differential Pressure was not among
ambiguous. Subsequently, the addition of the independent variables, as this has been
variables, whose presence in the model was cited in the literature as one of the primary
plauaible from an engineering perspective sticking mechanisms. For this reason, and
were subject to intense atatiatical scrutiny. because of the results of the correlation
The resulting variables in the model are a matrix, the differential aticking coefficient
result of the authors’ attempta to balance (ace footnote on Table 1) replaced measured
the three objectives for a coherent and depth and API water loss in the final model.
statistically sound model. Also, the drag/ft variable was replaced by
drag [pickup-alackoff wt], since the two
The third objective in the model building waa variables are very similar (.82 correlation
to maximize the “accuracy” of the model aa coefficient) and the latter leads to a more
measured by the goodneas-of-fit between the predictive model with the differential
model and the data. This was determined by sticking coefficient (89.1% concordant versus
a summary atatiatic known as the percentage 88. 4% concordant) . Finally, plastic
of eeficerdafitand discordant pairs. Although .v,is~~:iq %?3s thrmm 0L3& since it W~S 110
the derivation of this statistic is beyond longer statistically significant (p greater

72
.

SPE 28298 A. P. Wisnie and Z. Zhu 5

than .3) in the new model. decision tree analysis’’*’3 has been the
preferred tool for converting probabilities
For the oil-based mud model, there were into expected values. For the stuck pipe
prgblgrns in that the variable Open Hole problem, decision tree analysis might be used
Length had a sign that is opposite for a rough approximation of expected cost.
conventional wisdom (increasing open hole However, there is a conceptual problem in
length leads to lower odds of stuck pipe). that decision tree analysis only considers a
Referring back to the univariate analysis, it finite number of possibilities, whereas a
can be seen that this variable is stuck pipe event has an infinite number of
statistically insignificant (p > .9797), and possibilities. For example, there is no
thus the multivariate association is most limit to the number of times a drillstring
likely a fluke. For this reason, the can become stuck in a particular hole
variable was eliminated from selection, and section. This alone makes for an infinite
the stepwise procedure was rerun. The decision tree (see figure 1). Additionaiiy,
resulting model was plausible from both a the variables that determine stuck pipe
statistical and engineering perspective. probability and costs are not constant, but
The final water- and oil-based models are will change depending on what has occurred up
presented in tables 5 and 6. to that point.

~PP What is needed is a way to mathematically


--d-l
UIW =- +b -..l+;~lm m.ltfl-q-=
b..e,,,M*-*r*---”------ and
-.. S.eaSQEm ~heir
Once the drillpipe becomes stuck, it is of distribution. Monte Carlo simulation is a
major economic significance as to whether the tool that can be used to solve this problem.
pipe can be freed. If the pipe can be freed, Through this technique, one generates
the trouble cost associated with the stuck multiple outcomes by directing individual
pipe incident is merely the time spent to outcomes with random number generations.
---------1---.:
reGuvwL -- \uauaA*y
L1A6p&p 1..-..-11.. “,,=
m-~ +-
b“
4*amA*;
b“” “**.**..
11{m? vhwm,,mh
* *A*“-=.. s ~a~,lu $ =< ~=p. # Qp-e ~~~ ~h@Oret&C~&l~

days for an approximate cost of $60,000 to drill the hole as many times as is necessary
$120,000, depending on the cumulative daily to yield a representative population. With
rig cost). However, an inability to free a representative sample, one can use the
stuck drillpipe requires a sidetrack, which distribution in further simulating monetary
can cost hundrede of thousands of dollars, outcome or digest the distriiiution through
depending on the success of the sidetracking summary measures such as the average or
operation and the amount of hole lost. median.

Because of the quality of Love’s12 work, and D~ A ~G CAS~


in the interest of time, this model will be
used as the Pipe Freeing predictor for the A flowchart for this particular simulation
water-based model. However, since no oil- can be seen in figure 2. The following
based equivalent exists, one was constructed describes the algorithm and its assumptions:
from the data. A model-building procedure
Sifii2=Z
t~ tht Gf the St!!Ck r-r-
n~~ -~=~~~+ifiq
F--------- BQX A inv~ives inpueing data required for the
models was employed. The resulting model is models in Tables 5 (for water-based mud) and
shown in table 7. 6 (for oil-based mud). These are the assumed
conditions at total depth. From this data,
TRANSLATING TSE RBSULTS probabilities for pipe sticking through the
interval can be predicted.
As given, the results from the probability
models are somewhat useful in that they allow Box B is the heart of the simulation. In
relative comparisons. One might use them to this step a random number between O and 1 is
cmpaze +hm
. ..- =e~=~~~e “
%3s!Ss of Varinus
.----- genera~ed. If this number is less than the
wellplan scenarios or to measure the effect stuck pipe probability, the drillpipe is
of changing certain drilling parameters. assumed to be stuck in this particular cassi.
What would be most useful, however, is the If the random number is greater than the
ability to quantify stuck pipe risk from a stuck pipe probability, the hole is assumed
dollar perspective. This would allow the to have been successfully drilled and the
user to: a) incorporate mechanical risk into simulation proceeds to Box J.
overall project economics, and b)
economically justify (or condemn) investments Box C uses random number generation to
that are intended to reduce stuck pipe risk determine the depth that the pipe is stuck,
(ie lowering water loss, installing a top which in turn is used in Box D to determine
drive, making trips to condition the noie and the probability of f~~~ili~ the stiiek ~ipe.
reduce drag, etc. ).
Using input from Boxes A and C, Box D uses
In drilling operational decision-making, Love’s’z model (for water-based mud) or Table

73
.

6 QUANTIFYING STUCK PIPE RISKS IN GULF OF


MEXICO OIL AND GAS DRILLING SPE 28298

7 (for oil-based mud) to calculate the


---k-k< 1:*.. ~f 8“--<
p&wuaMAAAky
-“ ~~ae .a+,%mlc
&&--&Lsy ..”- Sa+na
r-r-” -a-
random number process similar to that of Box TO illustrate the use of the mdel, a
B is used to determine whether the pipe is “moderate” stuck-pipe potential hole section
freed or if a sidetrack is necessary. will be used to compare the unrisked cost to
the risked cost for both oil- and water-based
If the pipe is freed, Box E assigns a cost mud.
=r,~ ~~T,= ● A ~~.e ~;czh;mm
**-..**.- n-l-s+
W--------- + fin ~~=

simplicity:”it is assumed that O-3 days would Technical parameters and probability model
be required to free the drillstring. Random results are listed in table 8. The results
number generation creates the actual fishing of the simulations are shown in figures 4 and
time and cost within this range. 5.

In the case that the drillpipe is not able to It is interesting to note that although the
be freed, the algorithm proceeds to Box F. trouble-free cost for using water-based mud
This step assigns a sidetrack time from 2-4 was less ($900,000 vs $945,000), the median
days in a similar manner as Box E. (P50) cost was $140,000 more and the high
side estimate (P90) was $1,350,000 more!
Box G uses random number generation to pick
the amount of hole lost. With this number, The results from the simulation are also
drilling and pipe costs of the stuck pipe helpful in that they can be employed with
incident are added to the cumulative interval other project risk evaluation methods=u.
cost . Most of these methods deal with exploration
and price risks but use either an historical
Box H details the step in which the rig time distribution or fixed number for cost. The
associated with fishing and side tracking are incorporation of this method allows cost-side
added to the interval total cost. risk distribution to be based upon technical
parameters, which ultimately help to
Box I is only necessary for the water-based determine the true risk distribution of the
stuck pipe model, in which the age of the entire project.
hole is a determinate of stuck pipe
probability. This step adds the fishing, CONCLUSION
sidetracking and lost hole drilling time to
the trouble free time initially used in Box As reserves become increasing scarce and
B. The number is used to recalculate stuck -demanding to access~ the technical demands
pipe probability in Box B. placed on oil and gas drilling have increased
exponential ly. These demands have led to
At this point, the algorithm is repeated some wonderful advances in technology but
until the hole is “successfully TD*d”, which also some very costly failures. Given the
occurs when a random number generated in Box mature phase that the industry is in,
B is greater than the probability of getting companies can no longer afford a haphazard
stuck. approach to operational risk. Risk must be
quantified to assure that a proper return is
In the event of a completed case, Box J being paid for it. This paper presents a
totals all costs over the entire interval. method for quantifying stuck pipe risks in
This process occurs until 1,000 cases have Gulf of Mexico exploitation drilling. The
been generated, which was figured to be an results are based on research~ historical
acceptable sample size. data, and accepted numerical methods. In
addition to the above, we can conclude that:
The resulting distribution is ordered from
high to low. The range from high to low is 1. The recommended method for evaluating the
L—-,--- !-L- Luu
QruKen Lncw .AA equaA rangw xracs~~um.
-—.-1 —--—- E—.-AI 1-- ——-L Al.-
Cost of -&”--’-
13K.UCK ‘z--
pLp6 is t= ---’-’
IIluu= A LS8U
Thus , a range of costs from $100,000 to multiplicity of outcomes and assess the risk
$5,100,000 would be comprised of one hundred in terms of the resulting distribution.
fractile ranges of $50,000 each. The mean of
each fractile range is plotted versus the 2. Taking stuck pips into account, the
number of cases in each fractile range to resulting cost distribution for a given
portray a relative frequency plot. Upon drilling interval appears to be log-normal.
visual inspection, it is apparent that the
drilling cost frequency plot approximates a 3. Gathering data on an interval basis, as
log-normal distribution (see figures 4 and opposed to a day to day basis, sacrifices
s). little accuracy and is more manageable for
assessing stuck pipe risk.

74
SPE 28298 A. P. Wisnie and Z. Zhu 7

4. Logistic regression may offer advantages 10. Courteille, J.M. and Zurdot C.~ “A New
over discriminant analysis for predicting Approach to Differential Sticking”, SOCIETX
stuck pipe probabilities, particularly for OF PETROLEUM ENGINEER s PAPER 14244 , 198S.
non-normally distributed variables.
11. Harrison, C.G., “Fishing Decisions Under
Uncertainty” JOURNAL OF PETROLEUM TECHNOLOG Y$
ACIWOWLEDGEMBNTS February 1982, pp. 299-300.

We would like to take this time to express 12. Love, T., “Stickiness Fackor: A New Way
our gratitude towards Conoco Inc. and the of Looking at Stuck Pipe”~ ~Q
University of Southwestern Louisiana for Journal, Oct. 3, 1983, pp. 87-91.
permission to publish this paper. I would
also like to personally thank Ms. Jaye 13. Shivers, R.M. and Domangue, R.J.,
Thompson of Synergy Statistical Consulting “Operational Decision Making for Stuck Pipe
for her helpful advice. Incidents in the Gulf of Mexico: A Risk
Economics Approach”, Presented at the SPE
REFERENCES IADC Conference, Amsterdam, 11-14 March,
1991.
1. Grayson, C. J., “Bayesian Analysis - A
New Approachto Statistical Decision-Making”, 14. Kingsborough, R.H., Lohec, W.E.,
JOURNAL OF PETROLEUM TECHNOLOGY, June 1962, HemPkins, W.B.~ and Nini. C.J.. “Multivariate
pp. 603-607. Sta~istical Analysis of Stuck Drillpipe
Situation”, SOCIETY OF PETROLEU M ENGINEER s
2. Capen, E. C. “The Difficulty of Assessing PAPE R 14181, 1985.
Uncertainty”, JOURNAL OF PETROLEUM
TECHNOLOGY, August, 1967. 15. Hempkins, W.B., Kingsborough, R.H.,
Lohec, W.E., Nini, C.J., “Multivariate
3. Rose, P. R., “Dealing with Risk and statistical Analysis of Stuck drill Pipe
Uncertainty in Exploration: How Can We Situations”, SPE DRILLIN G ENGINEE RING ,
Improve?” THE AMERICAN ASSOCIATION OF September 1987.
PETROLEUM GEOL(XISTS BULLETIN, V. 71, No. 1
(Jan. 1987), pp. 1-16. 16. Howard, J.A. and Glover, S.B., “Tracking
Stuck Pipe Probability While Drilling”,
4. Bradley, W.B., Jarman, D., Wood, R.D.1 Presented at the SPE/IAD c Driw
Schofield, T.R., Auflick, R.A., and Cocking, Conferen ce, Dallas, Tx. 15-18 February 1994.
D ‘A Task Force Approach to Reducing Stuck
P~~ Costs”, SPE Paper 21999, Presented at 17. Bieglerl M.W. and Kuhn, G. R.~ “Advances
the 1991 SPE/IADC Drilling Conference, in Prediction of Stuck Pipe Using
Amsterdam, 11-14 March, 1991. Multivariate Statistical Analysis”, PresenteQ
at the SPE/IADC onfere nce in Dalks. Tx , 15-
5. Warren, J.E., “Causes, Preventions, and 18 Februarv 199:.
Recovery of Stuck Drill Pipe”, Presented at
the Spring Meeting, Southwestern District, 18. Hosmer, D.W. and Lemeshow, S., ADDlied
Division of Production, Houston, March 1940. Loaistic Regression, John Wiley & Sons, New
York, 1989, p 20.
6. McGhee, E.,”Gulf Coast Drillers Whip the
Wall-Sticking Problem”, OIL AND GAS JOURN~, 19. Millheim, K.K., and Huggins, R.L., “The
February 27, 1961, pp. 100. Engineering Simulator for Drilling (Part l)”,
SPE Paper 12075, Pr esented at the SPE Fall.
7. Annisr M.R. and Monaghan, P.H. Conference in San Fran cisco. CA. October 5-8,
“Differential Pressure Sticking - Laboratory 1983.
Studies of Friction Between Steel and Mud
Filter Cake”, JOURNAL OF PETROLEUM 20. Newhouse, C.C., “Successfully Drilling
TECHNOLOGY ( June 17, 1957, pp. 132. Severely Depleted Sands”, ~resented at the
SPEIIADC Drill ina Conference, Amsterdam. 11-
8. Hopkin, E.A., “Factors Affecting Cutting ~4 March 1991.
Removal During Rotary Drilling”, SOCIETY OF
PETR OLEUM ENGINEERS PAPER 1697, Presented at 21. Bendel, R.B., and Afifi, A.A.,
the SPE Third Conference on Drilling and Rock “Comparison of Stocmina Rules in Forward
Mechanics, Jan, 1967. Regression”, JOUW’ti- ‘OF THE RICAN
STATISTICAL ASS OCIA TION t V 72, pp. 46-53.
9. Adams, N., “A Field Study of
Differential-Pressure Pipe Sticking”, SOCIETY 22. Mickey, J., and Greenland, S., “A Study
OF PETROLEUM ENGINEERS PAPER 14244, 1985. of the Impact of Confounder-Selection
Criteria on Effect Estimation”, WERICAN

75
.

QUANTIFYING STUCK PIPE RISKS IN GULF OF


MEXICO OIL AND GAS DRILLING SPE 28298

~t “lsg~ PP. 125-137”


23. Wiggins, M.L., and Zhang, X., “Using
PC’s and Monte Carlo Simulation for Assessing
Risk in Workover Evaluations”,
SPE Petroleum Ccmmuter Conference
prleans~ La.- u lV 993.

24. Clapp, R.V. and Stibolt, R.D., ‘Useful


Measure of Exploration Performance”, JO-
~0 O’tober 1991 ~ PP”

1 St AHEWT 2nd ATTEMPT MA-PT(ETO.)


#

\\

FtGURE1. STUCKPIPEEVEWTTREE

76
.

3

Me’ow’fs-oda n

g
1 llao’mrz$-wd
0wm6ts-tdvaw
OOOWW-OM Mowns =Wvaw
Oao’aw’ls-ou
WO’C5$S-OIJ MO’OW$-OId

g . g 8
~ 8
A3NdkM Amok

Till
?

--
CNI SQD CNI SQD
VARIABLE BAMl cOSP?ICISUT STDSSROR PR VAIUASLB MIUIS COSPPICISMT STDESSOR Ps

.0001 Measured De!pth -.0000s .000048 .0009


Meaeured Depth -.00027 .000052
-.00026 .000052 .0001 ‘frutaVertical Depth -.00008 .000048 .0007
True Vertical Depth
-.0002 .000077 .0109 Open hole Length 4.45433-6 .000058 .93s7
Open hole Length
.0489 .0030 Hole Size .1417 .0754 .0604
Hole Size .1452
.00999 .0223 Angle Changle in Open Hole -.00297 .00746 .6903
Angle Change in Open Hole -.0228
-.0325 .00951 .0006 Maximum Opem Hole Angle -.0179 .0106 .0923
Maximum Open Hole Angle
.0007 Sum of Angle Changea -.00218 .00446 .6254
Sum of Angla Changes -.0262 .00772
.0715 .009 Drilling Nud Weight .1503 .1003 .1340
Drill ing Mud Weight -.1S67
.7129 .4375 .1032 Use of Top Drive Mechanism .6827 .3992 .0872
Use of Top Drive Hechaniam
.000989 .0067 Drilling Z4uldFlow Rate .00213 .00100 .0336
Drilling Nud Flow Rate .00268
Mud 10 second Gel Strength .1718 .0670 .0103 i4ud 10 seccmd Gel Strength .0680 .0385 .0778
Mud 10 minute Gel Strencth .0525 .0324 .1057 Mud 10 minute Gel Strangth .00349 .0200 .8611
Water Percentage of Mud .0379 .0172 .0275 Water Percentage of Mud -.0536 .025S .0377
Solid,a Percentage of Nud -.0514 .0191 .0071 Solids Percentage of Mud .0446 .0305 .1435
.0348 .0124 .0048 Dayw Spent in Open Hole -.0129 .0122 .2921
API Fluid Lees Reading
-.0412 .0145 .0044 Wholle Mud Chlorides Reading -.00003 9.168 E-6 .0004
Daye Spent in Open Hole
whole Mud Chloridas Reading .000027 .000031 .3921 Differential Pressure -.00021 .000131 .1121
-.00042 .000152 .0058 Mud Plastic: Viscoa ity .00778 .0211 .7125
Differential Pressure
-.0265 .0147 .0707 Dra<; (Picku~p-Slackof f wt ) -.0146 .00463 .0016
Mud P’lastic Viscosity
Drag (Pickup-Slackoff wt) -.0246 .00507 .0001 Mud Elactri.cal Stability .00113 .000548 .0387
Mud Yield Point .0279 .0320 .3s45 Mud Yield Ploint -.013s .0145 .3414
.5128 .0003 Oi 1 Water Ratio .2477 .0771 .0013
Prev Occurance of Stuck pipe ‘1.8598
Occurance of Lost Circulation -.7424 .6200 .2311 Hi Temp Hi Press Fluid Loss -.1116 .0561 .0467
-4.2987 4.2455 .3113 water Phaas, Salinity -7.23 E-6 4. 026 E-6 .0725
Low Gravity Solids in Mud
-.2679 .1313 .0413 Prev Occuremce of Stuck pipe -.9982 .3959 .0117
Delta MN over interval
-.00037 .000104 .0105 Occurance of Influx (kick) -.8938 .5400 .0979
Diff Pras* (l+ain(ang) )
-.1889 .0758 .0126 Occurance of Lost Circulation -.3445 .4723 .4653
Mud Annular Velocity
Low Gravity Solids in Mud -.0894 2.8118 .9746
Above! Variable ● Yiald Pt -.00666 .00465 .1516
-.00430 .00511 .4002 Delta NW over interval .2276 .1611 .1576
Above, Variable * (COS (ang) )
.00258 .2085 Dif~! Prea* (l+ain(ang) ) -.0002 .000095 .0339
Above} Variable ● Sp Mud Denat -.00325
.0515 .0036 Mud Annular Valocity -.0372 .0919 .6854
Hole size - Orill pipe size .1497
.0041 Above Varie,ble ● Y iald Pt -.00310 .00220 .15S6
l/APl Fluid LOSS -3.1640 1.1025
.0229 Abowe Variable ● (COS (ang) ) -.00194 .00258 .4521
1/ (AP)I Fluid Lees)”2 -4.3200 1.8994
.0001 Above Varia,bla ● Sp Mud Dens! -.001 .00176 .5694
Nat Log (daya in open hole) -.9103 .2008
-94.2484 36.7012 .0102 H01c4 size - Drill ]pipe siza .1647 .0909 .0701
Drag/Measured Depth
-.0004 .0000104 .0001 l/Ai?I Fluidl L@as -3.1640 1.1025 .0041
Diff. sticking coef’
(HTiiP Fluidl LoSS) -2 -.00541 .00314 .0852
Nat Log (Day’a in open hole) -.2143 .1888 .2564
150 c~bservationa
Nat kg (Oil/Water IRatio) 1.4137 .3s53 .0002
116 Nlot Stuck Cases
Dif~! sticking coeffz -.00015 .000075 .0020
34 Stuck Caaes
161 Observa,tiona
119 Not Stulck Case)e
1 Mud density in lbs per gallon/ 8.6 lbs per gallon (unweighed
42 Stuck Casea
mud ) .

z Differential praasure sticking coefficient = differential


presfinue ● (l+sin (angle acroas max differential) )*ln (API watar
10ss ) .———
I Mud dlenaity in lbs per gal ion/ 8.6 lbs per gallon (unweighed
mud] .

2 Differential sticking coefficient = (differential pressure


● (ll+sin (angle acroas max cliff pressure) )● ln (NTNP FL)

,
TABLE 3 TABLE 5
WATER-- MUD MODEL
WATE5t-BMED MUD MODEL
STANDARD
j~ cQEm2Em EERQK &iw.bBE ~ERIMX
STANDARD R>

4.9278 2.47S3 .046s


Dc#dr .O.00043 0.000154 .0050 IMerccpi 3.1515 1.2488 .0116
size
nii o.3a14 0.16 .0605 Diff Stkmef -0.00015 0.000137 .2654
T@rive 1.033s 0.7903 .1910 Bit Size 0.1349 0.0922 .1436
AM fluid Ioiw -0.0724 0.03s1 .0393 Topdrive 1.34371 0.7355 .0755
MasticVbaelii 0.0617 0.0411 .1332 In(daysin OH) 4.95n 0.2919 .0010
Im(daysin ON) 43.9430 0.4394 .0310 Orsg (PU-so) -0.0215 o.tX371% .0050
DragIft -229.1 79s4 .0040
Conmrdmt
= 39.1%
Concordarn
= 91.6% DlscOrdsm= 10.9%
D-dam = 8.3!fi Tied = 0.1%
Tii = 0.1%

TABLE 4 TABLE 6
OIL-BASED MUD MODEL
OIL-BASED MUD MODEL
STANDARD
,yAmBLE ~- & STANDARD
&
hdtercqn -7.9291 2.4245 .0011
10’
S@ GelJ 0.0978 0.0450 .0298 Intercept -5.4m I.7644 .W22
Opealid. LmgeIr o.oat36d 0.000NM .m5 Flowrste 0.00365 O.tmldo .0224
PheIicViscdty 0.1253 0.0394 .001s 10 SW Gel StrerIgtlr 0.0523 0.0412 .2048
Orag(PU-so) 4.0232 0.00737 .0016 plasticViiity 0.1152 0.0344 .OlwE
Diff stk d 4.00043 0.0U2141 .0025 Drag (PU-SO) .0.0179 0.iKM24 .0041
Irr[oilkster ratio) 1.9795 0.5913 .aw Diff stkcoef 4xwol1 0J30C063 .rwll
In(oil/water mio) 1.7741 0.5301 .aw
COacOrdmn = S4.4%
Discordant= 15.4%
Tied = 0.2% Concordant= 82.7%
Oismrdsot= 17.1%
ia:
‘megemrdeqrudm Tied = 0.1%

PRoMBIIJ2&*.--L
1.-B
‘ DhTerentislstickingCaef?kient = diftkartid pres.surc ● (1 +Sin (mgl~ SWMS rjffe~~

pressure)“ hr(API tluid loss)


Whm p .fJfmqrf+(m@cerrt*wi@e)+ (cr@fcem*wrrjab3e)...
z Dlffermtiel StickingCOefticiem= diffrrentisl pressure* (1+sing (mgle across differential
prcssure)*In (HTMP fluid loss)

——.—
1DIffere@idStickingCoefilciti = (1+sin (sngleacrossmraxcliffpressure))● dit%retrtialpressure
FINAL OIL BASED FREEING MODEL

TABLE 7

OIL-BASED NUD FREEING MODEL

STANDARO Pr >
a..-
VARIABLE COEFFICIENT Ww

Intercept -4.3891 1.7887 .0141


Measured Depth 0.000249 0.000099 .0122
Max Hole Angle 0.0428 0.0177 .0156
10 Min Gels -0.1002 0.0394 .0109
Plaatic Viscciaity 0.0512 0.0342 .1346

TABLE 8

TECENICAL PARAMETE RS FOR CONPARIATIVE SIMULATION OF OIL- AND WATER-BASED MUD


FOR A MODERATE RISK GULF OF MBXICO WELL1

Depth 12000 ft
Open Hole length 8000 ft
Bit Size 9 7/8”
Time to drill 15 days (water-based mud)
13.5 days (oil-based mud)
Daily Rig Total cost $60,000 (water-based mud)
$70,000 (oil-based mud)
Maximum hole angle 45 degrees
Calculated Drag 100 kips (oil-based mud)
(Pick up-Slackoff) 120 kips (water-based mud)
Maximum Differential Pressure 1,560 psi
Mud Flowrate 550 gallons per minute
Mud Weight 12 ppg
API Fluid Loss 5 cc/30 min (water-based mud)
HTHP fluid Loss 4 cc/30 min (oil-based mud)
Oil/Water Ratio 4 (80% oil/20% water) (oil-based mud)
Bottom Hole assemble length 180 ft
Mud Plastic Viscosity 28 cp (oil-based mud)
Mud 10 sec Gel strength 6 cp (oil-based mud)
Mud 10 min Gel Strength 12 cp (oil-based mud)
A Topdrive rig will be utilized

Results:

Water-based mud

Odds of getting stuck 51% (from table 5)


(Assuming trouble free, 15 drilling days)
Odds of freeing pipe once stuck o% (from Love’z)
(At total depth)

oil -based mud

Odds of getting stuck 40% (from table 6)


Odds of freeing pipe once stuck 32% (from table 7)
(At total depth)

1 Alldata is reportedin the units that should be used in the models

80

You might also like