Sie sind auf Seite 1von 10

.

.. . .

@6iKJ

IADCLSPE 27528

Tracking Stuck Pipe Probability Wh”ile Drilling


J.A. Howard, Enertech Engineering & Research Co., and S.S. Glover, Enertech Europe
SPE Members
.
Copmlgh! i ?94, LADCLSPEDfii!lw CmiE!B.!@. ..: ;: . . . ... x. ,. .

This oabW w c.m-aiea f., W@entatlon al the 15S4 lALXiSPE Dr81(ngCm fe,mcs held 8“ Dallas, 7?=, 1>! 8 FebGw 199+.

ABSTRACT _: . . Databa$e sssembly

TWOstuck pipe datibases have been developed with data from The data ccdlectkm strategy was detine$l after a number of
mWe than 106O””W1Sdrilled in the Gidf of Mexico aid the meetings tith the proj@.&partiiipanti. Each cpmp~y collect:
North Sea. S@sti6ial tecbiiqires WCI.Sused to “&efop stuck ed about 50 data s& For the Gulf of Mexico each data set
pipe”prdlctiv: “rnodils. When cl.assifyiig in tbiee categories - contains tbme days Of infoi-nii~tm, wbich was extend&d to five
diffeimtialIy~ “rneihanically ck “i%tsmik - an “a&mcy of75 % days for the North Sea database. Each ccinpimy . “w& to
is achieved, risiiig tb” 80 % “foi’iiwo-class (stucldiot stick) submit halT stuck welk snd half”ri6t stuck’wells. The databsse
prediction. Using these models, graphical tracking of the was to include welh drilfed in the previous five y~s. The
probab~ty” of stuck @pe OCCui-ri@II@y”“be un&taken~ tO- variabks to be ticlu&_l were &elect&!duxiig discussions ti”tb
monitor drifling” operstioti ‘FOI stuck pipe avoidance. ,A participants, followings literature reiiew and consultation @b
similar statisti& model W8Sd&eloped to @let t& probabllJ- a}knoil~~~ experts. Spe&I atteution was diwti tevaird
ty of f$?riiig=fipe 6u&3uik. Comparisoti are aisb made tie twenty” vdablti “tiggested by Kiggsborough er al ‘]] in
with two published pipe-freeing” rn”deIs. The &“t.i were 1985. “’”:. .“ .
further analysed to identify ‘zinY’tifids which may provide
operational guidelin= for” stuck pipe avoidaiwe or freeiig it “w&“obse.R-d thats6R of the stiller” compahies may not
operations. “ . . have 25 stuck weIls 0;50 total welfs drilled since 19s3.. The
larger companies were asked to submit a bigber prop+m of
INTRODUCTION stuck wells, say 35-40 stuck and 10-15”not stuck. This, would
allow the final database to have the required mix of half stuck
Stuck pipe i{ n m?jor diillhig trouble cost in the petroleum snd half not stwk. Selection of wills was requested in reverse
industry. Histmiiilly, one out of “thp wells drilled in the clyonological order, ie.., the ,rnost recently drill,d wells
Gulf of Mexico and the “Nortl”-S& eipefi’~ie ““”kuck””pipe reported first.
problems, at a tetsl estimated cost of $250 &iK1on annually.
~mm. ~9g8 ~d 1992;. ~o, ~OEt. .tiaG@ ~mjaE ;* ‘Stuck’ was defied w the pipe””imgble to be moved ~d for
&Ck’@J~~@tiOm (eg. j~g, spotting fi”id p@@)
undertaken to assemble and amdyse data relating to stuck pipe. had commenced. For stuck wells, the ‘Day 1* data wss from
A total of 20 til operating companie%Were “tivolved, seveiil the last morning i+ofi”prior 6 beco~g siwk,’and “Days 2-.
of them in boih projects. Two daiabasei were assern”bled, orie 5. tbe four dSyS priOI to thd for a to~l of five co~~u!lv?. _. ._
relating to weifs chilled iii ffii ‘Gulf of M&icO &a, the’otbii @+ysrecorded (for the North Sea - three days for the Gulf of “‘
relathig to wL1ls drilled in the European North Sea. Mexice). The other wells were selected at random snd
included 3” or 5 drilliig days (not .n.%.e&cjIy ‘JOnS+c@e),
seleded below a depth of 3500’ tl. It is possible that a not
., . .
—- .-— stuck..well,w US@.@.rnDWt~. oqe_da@.sgt.,if”a“partlc]pant
References, tables and fi@es at end of paper lacked sufficient Wellq the actual days are not dupliciti?d~

“901
2 .eTRACK3NG STUCK PEW ~OBABILITY WIDLE DRILLING SP3%IL4DC
.2752s

Most of .~e Gidf Of Mexico wells (47 1 out of 503) are located defie a well as mechanicall~, differentially or not stuck. TWO
in latitude ranges from 260 to 32”N and longitude ranges from class analysis will defie it fi stuck or no; stuck. The models
87” to 99! W, and mast oftb.5se in the North Sea (503 out of were developwl using a well known proprietary statistical
523.) are Iocated in latitude ranges from 50° to 66”N and amdysis program.
longitude ranges tlom 7.5”W to 106E. Partitipmits unable to
complete a full set of 50 wells from the North Sea were Separate models were developed specifically for wells drilled
rquested 03 select weUs with similar geological traits to the using water base ‘mud (WBM) and oiI base mud (OBM).
North Sea. However, the number of wells drilled with OBM recorded in
the Gulf of Mexico database, was insufficiitit to ensuie ik
Database content statistical validity.

The initial selediii of variables for inclusion in the dstabasei Variable sdedibn
was accomplished using engine&ing judgement. The process
. .
commenced with a review of “the causes of stuck pipe. b“ To develop a satisfactory pidictive model, It m necessary to
addhion to reii&vin.g the published literature []-’3, the subjecl select those variables from the &tabase that best discriminate
was discussed with experienced operational and research between the classes. Di$ciiti”a.nt functions can” then be
personnel. Eleven stuck pipe categmies are shown in Table 1; derived, using the variables selected, to calculate the probabil-
the lists of related variables are given as examples, and are by ity of”a well being in any stuck condition. The canonical
no means exhaustive. analysis simplified identification of stuck condition by !howing
the three classes on a two-dimensional plot. This is pitduced
BP Exploration have since developed a Kig$ik””Stuck Pipe by seItiting a plane in the mukidlmerisiotial space such that
Prevention Course [’O. They report additional causes of stuck the projection of all data points onto the plane shows the best
pipe - such as undergauge hole, junk in hole, cement relsted disctiating chamckristics ie. such that the mean of each
problems, and collapsed casing - which do not have related class plots as far as possible from that of the others “7.Z].
variables in the current database.. They al%o.identifi.$d two
variables related to human errors that appear to affect the The pd-iitive rnodeli were” developed using datab+es that
incidence “oTsluck”“pipe. These are ti’me in hours ‘stice tOUr &intaiii 224 variablti for each of 503 wells (Gulf of Mexico),’
iharige, and days s:@i crew ccge. ~: “WOrk ~ PUb- and .544iiwiibles for each of 523 wells (North Sea), collected
lisbed tw late for fic~usion ir:the Gulf 6f_Mexico database, by the project pmticipmts. “After assembling the database,. a
but the variables were ihcl”ded in the North Sea shidy. end stige engine@#lit&iture rev~ew rw”ulted,jqa list of
40 variables (for W“M) for “-ch area which were considered
There is i total 0f<03 “w611sin the Guff of Mexim &b base, to be”“potentially sig@c&t ii pri&cting stiick pipe. Th&e
with i possible 224 variables for each well, and a to.t?.of .!iZ3 Iiit.”of variables were diff>rtit ~etwq- the two gtigrapbical”
tieUs in &e North Sea daki base, with a possible 544 v’ii-iables areas. This was ih part due to changes made in the data
for .eiachwelL General in forniaticm about these w.411sis shown giihered for the North Sti &taba&, as a result of Gulf of
iii Table 2. An addkioiid 25 Gulf of Mexico” wells were Mexico experience. S.5me of these 4Q were derived from
submitted by a hate entrant to the project, . and were not othm, variables, to. fomi diieniionkss or physi~ groups.
included in the full database. These wells were resewed as an
independ@ database f6r model verification. ” For each database, three subsets of the 40 variables were
.
selected using the smtistical techniques of forward, backward
STATISTICAL MfJDELL3NG “ “. and stepwise elimination. These resulted in classification
accuracies’ of 59-62% for the Gulf of Mexico and 72 % for the
Anafysisrtecbnique .- . F@h Sea. For the North Sea database,, ,all threg .elimimtion
tdmiques selected the-&me 13 variables. fn & effort to.
M“ltivariate Discriniiliant ‘=Analysis “(MD~) w~l .US@ to ifirove &e ‘wc@y further, ”a, ‘riimiber of other variable
develop stuck pipe prd”icdve” models for wells drilled in the groups were tested.
Gulf of Mexicom:d. North SW .meas. The purpos: of MDA
is to classify ce~in outc.omei by plottiri.g linear ~mbi@fo:s For the Gulf of Mexico, seven groups of variables, listed in
of “variables in’ multidiinensionid space. The point iii *ace Table 3, and for the North Sea eight groups (Table 4), were
plotted by each set of variables defines the outcome for that@ selected for testing. T’& awuracy of classification depends on
of variables. FOL tbe best dkcriminaiion”it is “desirable that. tlii”variables selectd” and the database size. where a seleckd
each class of outcome- ploti in--a discrete regitm. viiiable does not have i vahie “&ported in a particular dataset,
that dataset cannot be used in model development. This.till
For tii intilysis, the Wataset for tich well is a s@kd group tend to limit the number of wells available for model develop-
of viaiiabies}- tifi%Zd for-accuracy of cla?.sificatkm, and the
,% 1Clamifimticm
outcome clatse.s are the stuck c-mdit~o”m. Two classificaticifi- acc..op ~.
ty@s are possible for-this tie; Three class analysis id Numberof walls m be d,%itkd

902
- —.

SPE/IADC 27528 J.A. HOWARD AND S-B. GLCWER 3

ment as the number of variables required for the model sample size for each stuck condition is gieater than the
mcreasei: For example, ii the Gidf of Mexico database, there miniiiuim @CkRl, and the total number of wells is signifi:” - ““” ““”
are 217 wells for which all 15 viriables in the first (sK) group cantly increased.
have values reported, but only 115 welh for the seventh (G2)
group’s .38 vaiiables. The variable set for the combmed WBM/OBM mcdel was
,.
d5@10p@ from a combination’ of “Ihe WBM and the OBM
In order to compare the accuracy of cl~sifi”caiion be~~q ke models by eliiit-mg iariibles that are specific to only one
varibus groups iii analysis wis conducted for the Gulf of mud type. Several sets, which include those variabl= that are
Mexico with a “fixed” database. This was chosen as the 16o common to the WBM aad OBM models, were selected and
wells which conti” “all the variabl~ in group G 1; this is the tested for accuracy of classification. A group of 27 variables
smallest well set which siKsti6s the statisdcal r~uirements. wti selected to devdop the stuck pipe predictive mcdel. The
Table 5 showi” the over@ accuracy ofclas:ifiiti,cm f~,,@ combination model has .a higher number bf wells (252).
group, for both the Kdl “ad tii; fix6d databases. Most groups miilabfe ““for analy& and the” accumcy “of cl+ssifibatiori is
showed a slight fmprove”rnerit ii a“tiuracy,” but group G I comparable with that of either the water base mud model or
remained the most accurate. Four additional groups were the oil base mud model.
tested, without improving-the accuracy of classitiiatiiin: These
gm”ps were selected by a co@tijatiou of statisti@ and STliK PIPE P@3DICTION
mgineefig” selection. from a sub-set of 34 variables. I@
accuracies of” classifkation for iach of the eight groups of Three class modkls
vaiiablti foi”fkrth Sea WBhf wells ire also listed in Table 5.
In three class andwis, a well IMY be classified as mecbanical-
A@’ from overall accuracy of clasiifititibn, a nu.rnber of Iy, differ@ialIy & not stuck. “~e discriminant .functioni
other parametem - number of variables used, number of wells (linti”c&nbinatiom of “the selected variables) are evalu&d.
available, accuracy in each stuck class, the distribution of The dkcrimihint” tiiiiitioi with the highest. numtical value
wells in the canonical plot, etc. - were used to select the predicts the stuck condition for that well. fn addition, two
“best” discri”ti-tit””~roiip:”” The BASECASE group of 27 canonical functions””may “be deftied to give a “griphiial
variables .~d .~e NSE@ASE group of 18 variables were presentation in a two-dimensional plot, showing three regions
selected fir divelop”med of the water btie mud models for the projected from the @&-dimensional” space plotted by the
~G .. . . .
Gulf of fJeii@ and North Sea [+p~tiieiy.” et f&tiO@@i@e 1)., Table 9 and Figure 2 shOW,
the resulti fo~-25 wells “collected by”C+i@ny H in the Nmtb
The selection gf vaiiiabks for wells dril@4 wjth oil base .mqd Sea, that have all the IS w+bles included in the stuck pipe
(OBM) was accomplish” by a si~lar procedi~ “to that predctive mcfdel for water base mud. Taking well No. 1 as an
outlined for the WBM model. For the Gulf of Mexico, of the exsunplc ,=.
503 wells reported, only 66 w“ire drilled with OBM. This was
insuffi?iEiit @ tie a statistical selection of the important lb three disc riminant functions are WDl =. 1125.21,
variable groups. Therefore, the variable set for the model was WD2” = “1131.99, WD3 =“ 1124:10. WD2 has the
selmtd by eriattig &x wwi{tilis”~.orn ihe water baie.rnid highest numerical value so the model predicts this well to
model. There are 49 welk drilled using OBM in the database be mechanically stuck,.ie. P~DICT = 2.. The mnoo+al
for which values were reported for all 21 variables. The timctious (WCl = 3.Z37, WCZ = -2.088) plot a poiit ii
overall accur~y of pre++ion is”92 %; however, +! s+mple the mechanically stuck region (see Figure 2).
“size is tco small to bi–statistica13y v~id.
For the North Sea, the overall accuracy of classification is
FOI the North Sea, variables that a~ specific to watei’ base 82%, 75.% and 73% for the WBM, OBM and combinafioii
mud were qdaced by those specific to oil base nmd.’ Table 6 models respectively the accuracies of classification, including
shows tie selected variable groups Wd the_ accim.cy of $iase for the. “ii@iyidualpirticipa~@g mJrnp-ti&S, a~” sumiiua- -
classificatiti foi-&ch. The ac&mcy and the nu-mberof w6!Js rized in Table “9.
in each class of stuck condition” (ie. differeatiilly, ”me&ani@ly
and not stuck) are shown in Table 7. The “best” group of For the Gulf~f Mexico, the “rno~elhas bi-sn v@Kd using a
variables is designated BIXf23, and this set was us?d to. separate datibase providd by a late entrant to the proje$t.
develop the stuck pipe predictive”model for wells drilled with This database contains 25 well datasets, but only 1g of”tbese
oil base mud. contain all 27 variables used in the model. The overall
aw”tmicy of predktioti is 72% for the three-class model; this
Both the WSM model and the OBM model have fewer than compares well with the 75 %““accuracy tisiig the ori@d
the statistically desirable number of wells in the difheniiilly &iaba& (Thble 9].””—
stuck categorf:’ the”minimum-sa.niple size should-be 301. “h”ai
. . .. . ..—. —
attempt to overcome this hnutatlon, a modeI was developed The statillity otlbe Gulf of Mexico modei wa.smko inv@igat-
cmnbtig wells drilled with WBM and with OBM.., The ed using a ‘rolling test’ technique. A rolliiig kst ii coriductcxi

903
4 TRACKING STUCK PIPE PROBABILITY WHILE DR3LL3NG SPE/IADC 27528

by removiig a number of wells, selected at random, from the these assumptions may not be valid, resulting in inaccurate
&tabase of .kdid wells, and developing anew predictive model predictions.
using the remaining tiells. The wells removed are then
:~~sified yi~ t@ model _wd the predktions compmed with The accuracy of prdlction” drops signifi~tlj. fi tie periOd
the database records, &h new”p~ictive model’ W8.Salso prior to sticking. However, it is interesting to note that 6V.ir
used to classify the additional set Of“dataprovided by the la: 50% of wells that became stuck are predicted correctly. This
entrant. tie iesults ‘of the rolling tests ire Suqw&@d in result mtiy s=e’~ “a.@ninE before actilY getting sm.
Table IO. ” For each roOing test tbe top line is the results of Table 11 shows the accur.wy “of prediction agakist days prior
tie tbrw-clasi prediction. to sticking. F@r*”2-5%how the disti’ibution of m@,anicaUyr
stuck wells h the stuck pipe prexhctive plot for. the Gulf of
For Rollbig Tests 1 and 2, the results= gene~[lY ~n~~ent. Mexiw Model at 4S and. 24 hours before stic~g, and on the
Rolliig T&ts 3 and 4 show lower accugcy “In predicting the daythe well became stuck. Apparently, wells tend to &g-
20 tieUs set aside. Howevei, “the tbrez-elms analysis is.,still yerge on the mechanically stuck region over time.
substantially better @w tbe..?~ ? acW~Y. .Wd?m ‘el.Vti,:n.
~a&. inda
should achieve. Table 10 also shows the residti of applying
the model developed fkom the 160 well &tabase to the four
sets of rolliig test wells. These reds are genendly excdent, Multivariaie. Discriti-tit .4@+is was dso used to develop
with the exception of the set for Rolling Test 3. models to predict whelber a stuck drill string cag be freed.
The Gulf of Mexi6u databass. records 78 wells for which the
Two +ss modefs 9 iariabIes identified as being most impoa”t “were rEPoit@d.
For the North Sea, the same variables were used, with the
In addition’~ Oie .@ ck+.ss_~ysis described above, it is excepti6iiof one. The database records 59 Sells for which@
possible to develop models which simply classi$ i “wETIw S variables were reported.
tither stuck or not stuck. .F_Asfigle Smear combmatkn of the
sele+~ variables is. tised, which, in this study, bas been The Freeing Index @l) is a liiear mmbinatioii’ of th+ese
described as the Sticking Index. The classifi”mtion depends on variables+ if “tie value of the FI is less than 4“.7353 (Gulf of
whether, when evaluated for a well, its value is above or Mexim) 010,2745 (North Sea), the m:d?l pr~lc~_ ~ he
b@ow a tiedold v#ue.:The Stiig Jgdex can alsa be ukd @u& driu stig-cm be freed. The FI corfectly predicti 29
to calculate. the pmbabilityof “awell I&g cla.ssitkd as stuck, (83 %) freed wells, and 35 (81%) not freed for”tbe Gi.lf of =
as descr+d in ~e .~n=~ple at +e ,md of the text. Mexico, aid 34 (90%). freed wells, and 18 (86%) not ti”f:f
“,
the North Sea. However, the small sample siti may UQ!be
The tio+lass modeii for North Sea weili gave ;Xcies” Of statistically valid. “.
classification of 87 %, 81% and “83% for the WBM, OBM and
cornbiri~ models respectively. The+ I&td@, along wi~ those TLWFI M akc.be used to talc.l,qe ,@i probability ‘+fr-~~g..
for each participant, ‘iie also summarised in Table 9.. a smik well, as explained k“ the example at the end of the
text. The late entrant’s wells were used to verify tie Gulf6f
For the Gulf of Mexiw” theoveil accuracy was 78 % (Table Mexko modefi all 9 variables required were recorded for.Only
9). The model was again tested using the ‘independent date- 6 wells, of which 4 (67 %) were correctly cl=”Sifi.Xl. .
base and rolling tests. Tim& rixuks are ?&I? ti .&e @WeI

Iine fdr &ch” tS.$””iF”Table10. _lNG ANALYSIS

History tracking
.,.
For stuck wells, Gulf Of Mexiti &tiwas c~llect&i Tm tijI.&
consecutive days, and North Sea data for five consecutive
A review of the literature ‘m] reveals the informiwion which
operators require when attempting to remedy stuck pipe. An
estitite of the probability of freeing the pipe and operational
~ldelines to optimii.e the freeing process are given priority:
I
days, culminating. the day the drillsttig bec”m stick. Tb& The review suggests which data may affect the probability” of
models c@loped have beem used to predct the drill string freeing. -
stuck conditio’ri from dnts collected in the peritid prior to
Stickiog. Database trends ““’”

Some variables, were not explicitly reported before stickin~ Table 12 shows the results of the general data SWULXsome
occurred. These wem ekher estkatexj (eg. true vertical depth guidelines may be deduced from these data. The general
and b]t depth when siuik), based on reasonable assumptions, trends were consistent between the Gulf of Meiifi and North
or assumed, to be the same as at tbe time sticking o~curd. Sea databases. In the following discussion, the percentages
Variables s&h as bit size, top of stabilizer, drill COUWOD, quded are for the Gulf of Mexico first, the North Sea second
average hole angle, ma.iimum dogleg severity, ”differential (GoM/NS).
pressure, and annular gap are examples of these. Some of
.-+- . . . .
904
,,-

SP3VIADC275~ J.A. HOW.iRi3 AND KB. “GLOVER 5

The propcmtion of wells freed for each stuck mechanism is Comparison with published modefs
shown first in Table 12. Mdmnically stuck wells have a
higher probability of fReing than diff&entiidfy SLUCIG
46 % vs. The Stickims Facto@ (SF) was calculated for the 208 Gulf’
37% for the Gulf of Mexico: 74 % vs. 60% for the North Sea. of Mexico and 2L4 North Sea stuck weu data.sets which
contain all five vtiables, and is plotted against probability of’
The M5imi&onin the datiib=”” sugg&k” that ibe total time freAng in Figuie 6. For the Gulf of Mexico, the database
stuck significantly affwts the probability of freeing. This is shows a lower probability of freeiiig stiik pipe up to a SF of
delined as the time idirhl stm’tiig” whea the pipe became 3, and a higher probability for SF values higher than “3,
stick, and ending when the pipe was freed. If the pipe was whereas for the North Sea, the results show good agreement
not fried, the”‘ad time is’iikn the timent plug was sit to” with Exxon’s published data up to a SF of 3. For SF values
sidetrack or abandon the well. The histogram shows that higher than 3, the current database shows a higher probability
beyond the first day, the propoK!on of wells freed is only of freeing stick pipe. However, the data is sparse in this
about 33% for the Gulf of Mexi60 and 56% for the North Sea. region of the plot.
One iite@.-E2ti0u of”tis ii”tbat most opemtcm itop attempting
to free the pipe after one day. where attempLs continued The TexacoFreeing Index (TFI) ranges from 86”to 130 for the
beyond one day, Me proportion .@ed falls to ~.% WC! 17%, 218 North S= s~ck wells for which values c&dd be calculat-
respwtively ai%er“he fifth day. - ~ ed. Thwev-du& are higker oiemll” h those reported by
Shivers &d Domangue ‘]. The TFI was grouped in three
The suggestion that the probability of freeing dezres.ses as ~g~: 8p-1~.”1~-l 15 ad 115-130 (Table 13) ~“~~~Ond
stuck time”iikr~”ii is SUppOrt-&h the Iik”iiiti”ie. Keller # al apprO”fi~tely Mtb tbe ~g~ (ie.. 0-45; 45-55 ~d 55+)
WI ~kt~ ~ ~~g4 that fi=fi=g @titiom should be StOPPed published. Compared w“tb Texaco’s data, North Sea wells
tier eight days. Recently, Shivers and Domungue 1~]rwom- with TFI values in the middle and low ranges apparently have
mendexi that attempts to frek stuck pipe should not be initiitd a bigher probability of freeiig, whilst those in the high range
or continued more than four days atle.r becoming stuck. appear less likely to be freed.

Table 12. also shows the effed of other variables on” the TIM TFI for wells from the Gulf of Meiiti database does not
propofiion of “Wellsfreed The’ importance of gettiig $otfing ‘show good agreement with the published data (Table 13).
fluid in plati as soon = possible is emphasised. Of the wells “Wells “with TFI vzilues in the low range apparently have a
in WbiCh SgOttilig fluid W= plw”ed within four hOllW of higher probability of fi’iiii, whilst those in the high range
stichig, 70% (Gutf of Mexic@ and 76% (NoA” Sea) Wert- appear less likeIy to be .fkd. The curve does not appear to
freed. where spotting fluid was positiciriexfbetween 4 and 24 have the same character as the published data
hours of becoming stuck, these figtires “fell ta 41 % and 43%
respectively. Although the data is sparse, it appeari that CONFUSIONS
jarring within one hour of placing a spotiiig fluid frees a
higher propmtion of stuck wells than does jarring” after two 1. Mukivmiate discriminant analysis appears to be an
hours of soaking (43% vs 27 % for the Gulf of Mexico, 67% effective tool for developing a predictive model that can
vs. 53% for &e North Sea). It is also apparent that if jai’iing cli.ssify drilled wells into three categories, namely dMer-
is not successful within 4 hours, the chance of free”mgis 1ss entially stuck, mechanically stuck and not stuck.
than 37% (Gulf of M.exicm)and 50% (North Sea).
,..
2. The accuracy of the predictive niodel depends on the size
The well profile apparently affects the probability of frieing” of database and the variables selected for.analysis. ~
once stuck. Table 12 shows that a higher proportion
(53 %/87%)tif straight Iiolii Were “=” thin build ind hold or 3. The stuck pipe predictive plot (such as F@e 1) provides
S-type profile wells (39 %/62”%=& 44%/77% respectively). a potential means for inonitoriig a well during drilling.
For the No@ Sea, average hole angle also ip~rs to .afftit For instance, a well can be monitorEd by plotting values
..— .
the probability of freeing. 76% of wells with hole angle las of the two canonical timclions at successive pouls m tune
than 40” were freed, whereas mdy 56 % of weUs in the range and noting whether the well’s position is moiiig from the
of 50.-60” were freed. not stuck region on the left toward the stuck regions’ on
the right.
Finally, the review of the freeing database for the North Sea
suggests that, there is a critical overpull, which indkates the 5. Multivariate diicriminaiit amdysis may also be used to .:
likelihood of fF&i4n~.””For wells subjecte+l to less than 150 develop models to predict whether a stuck well can be
klbs oveqndl during freeti”g’opiiatfins, 87% were frwxl. If freed, such as the Freeing Index..
overptdls higher than 150 klbs were applied, the success rate
fell to 49 %. This snggests that if a well remains stuck when 6. The zeneral trends recated above show consistence
pulled to 150 klbs more than string weight, there is only a betw~ the two regio~s studied, and should be nod
49% probability that it till &pond to higher Owpdk. . when considering operations to free stuck pipe.

905
6 TRACKJNG STUCK PJPJli?ROBABILPYY W31JLE DRJLLING SPE/IADC 27528

EXANLTLE 1957) 132.
11. Adam, N.: “How to Control Differetitial Pipe Sticking”,
Calculating Probabiht~ of Sffckin~ and FmQJJS ._ ..... . . j-” P.etrOleum Engineer (Sept. 1977) 32. ..
.12.. Hopkin, E. A.: “Factors Affecting Cutting Removal
fn addition tO initial classification, the discriminant finction: During Rotary Drilling”, paper SPE 1697 presented at the
reported in thk paper may be used to calculate the probability SPE Third Conference On Drilling and Rock Mechanics
of a will being placed in any one category. The method is (Jan. 1967).
described below: 13. Adams. N.: “A Field Study of Differential-Pressure Pipe
Sticking”, paper SPE 6716, (1977).
1. Calculate the numerical value of each discriminant -“14. Dunbar, M. E., Warren, T.M. and Kadaster, A’.G.:... .: ~~~
function using the relevant variable values. “Them-y and Solutions to Bit Sticking Caused by Borehole
2:. Calculate the exponential function if each discriniinaut Deformation”, paper SPE 14179 (1985).
function value. 15. Courteille, J.M. and Zurdo, C: “A New Approxch to
3. Sum the values calculated in Step 2. “’. Differential Sticking”, paper SPE 14244 (1985).
4. The probability of a well bei.n~ classiti~ in any class is 16. Rir-Site Stuck Pine Prevention Course Workbook, BP.
obtained bv. ..dividine the exmmential function of that class Exploration (1991).
(calculated in Step Z).by tie sum calculated in Step 3. 17. Bo~, G. E. P.: H~ter, W.G. and Hunter, J. S.: Statistics
for Exmrimenteti, John Wiley& Sons, Inc. (1978).
FRSEI and FREEO”arethe gk.mimitint Iinctions which 18. James, M.: Classificad6ri Algorithms, John Wiley &
Cm@. damify.a pipe as freed or not freed and FREDICT is Ox Sons, In.. (1985).
pwlkted outcome(1 = kyd, O = not freed). PROBABI
md FROBABO ,* lhe probabilitiesof Owe 0U1CO171eS.
19. Jobmcm, R.A. and Wlcher, D. W.: Amdisd Mukivariate
Takhg an examplewhex FRSEOiszmalcrtim FREEI,du Statistical .Awalwis, 2nd cd., Prentice Hall, Inc. (1988).
well is .Iamiliedas not tied (TREDP3 = O), ‘l%. pmba- 20. Ott, L.: AII Introduction to Statistical Methods and Data.
V,!ity of lhk predktim ii: . .= e -:.. Ad@, ~rd d.,, PWS-KeIZf Publishing Co. (1988).
1. The nmmicd vabm of the two diwfmi”a.t functionsati; 21. Chambers, J. M., Cleveland, W. S., Kleiner, B. and
FREEO= 15.844K;-*”EI = 13.2019.
2.? EXP@’REEOI = EXP(15.8440 ~ 7.607x 106; Turkey, P.A ~.
EXP(3=REEI) = 6xP(13.2019)- 0,5416X 10’ wadsworth.& Books/ColePublishing Co; “(19.83).
~:
SUM = EXF@REEO) + FXP(FRSE1) 22. Cbatterjee, S. aid Price, B.: Re,mssion Analysis by
= (7..507 10.5.414).X 10’ = 8.148x 10’ @iOIE!S, JOfUIWIIey & SOns, Inc (1977)t
4. PmbAility of classitieatimas not freed:
23.. Keller, P. S., Brinkmann, P.E. and Taneja, P.K.:. “Ecxz:..
&XP(FREEO) 1.sIIW=7.60718.14S = 93.4%
Probabilityof ch~,ficalionas”~reed:
nomic and SWistical Analysis of Time Liqimtions for
EXPfFREEl) / SUM = 0.5414/8.148 = 6.6% Spotiiig Fluids and Fishing Operations”, paper OTC 4792
(1984),
REF’EmN&ES . ..; : . .- :., 24. hbive~, R.M. and DOmmI-P, R. L:. “Operational Deci- :
~ sion Making for Stuck Pipe IncidenS in the. Gulf of
1. Kimzsbomugh, R. H., Lohec, W. E., Hempkins,W.B.” and Mexicc A Risk Economics Approach”, paper SPE/IADC..
Nhz;, C.J.:- “Multivariatii ~tatistirnl h-alysis of Stuck 21998 (1991).
Drilfpipe Sfh!atYo=ii”,paper SPE 14131 (1985). 25. Love, T. E.: “Stickiness Factor -A New Way of Looking
2. Short, J. A.: Drilline and Casine ODeratioiiz. PemWell at Stuck Pipe”, paper IADC/SPE 113.83 (1983).
Books,” PennWell publishing .Co. (1982). 26. Ayers, R.C. and O’Reilly, J. E.: “Offshore operators
3. Moore, P.: Drillirm Practice Manual. _: Cbmmittee .Gulf of Mexico spotting Fluid Survey”, paper
4. Short, J. A.: Casingind Fishinz C!Jeratiofis >PE@C lS@3 (1989). ..=, ..
5. Warren, J. E.: “Causes, Preventions, ”and Recnvery of 27. Perry, G..W. and Ruhe, R. W.: “To Fish, Or N@ ~o
Stuck Drill Pipe”, presented at Spring meEting, South- Fish”, 0!1 and Gas Journal (July 24, 1961) 80. ”
western Distiict, Division of Production, Houston, March 2S. Harrison, C.,G.: :Fisbing Decisions Und& Uncertainty”,
1940. “._JPT. (Feb. 19S2) 299.. “., . .. ... . . . . . .
6. McGhee, E.: “Gulf Coast Drillers whip the Wall-sticking 29.. Adams, N.: .“HQw to: Control Diff&entiiJ Pipe Sticldng -
Problem”, Oil and Gas Journal (Feb. 27, 1961) 100. Part 4 zEconomi& Methods to Avoid or Free Stuck Pipeti,
7. Am& M.R. and Monaehan. P. H.: ‘Diffeiefitfal Pressure Pelroleum Engineer (Jan. 1978) 30. . -- . .“. ..
.%ic~g -”Laboratory ~tidi~ of Friction Between Steel
and Mud Filter Caken, JPT (May 19.32) 537.
8. Outmans, H. D.: “Spot Fluid Quic~y to Free Differetdial-
Iy Stuck Pipe”, 011 and Gas Journal (July 15, 1974) 65.
9. Browse, M.: ‘H6w””””to””Haiid16- Stuck PIu& ind” Fisiiim
Problems”, World OiJ (Nov. 1982) 103. “.. .... .... -.
IO. Hdmick, W.E. and Longley, A:J.: “Pressur-Differential.
Sticking of Drill Pipe”, Oil and Gas Journal (June 17,

906
,. ..

J.% HOWmb~ S.?. GLOVER 7


SPEAADC 27528

Table k C3 usrs of stuck Dipe-and related variab!m . . ... .Takde.??.~W UDSof variables for Gulf of Maico w~k . .
.,6.,”., n B~wsTcOCHOIGZ
DEVIATION hole male, rn.asured and vertical depth
DOGLEGS rna~. dogleg severity, hole size

KEYSEATS drill collar OD, annular gap


PROBLEM FORMATION measured & vertical deplh, Ik.ca:ion,
(FRACTURE, SHALE) wellbore stresses
DRIUSTRING top mbiliser depth, no. of stabilisers,
[TORQUEIORAG) torque, drag, drill collar Iengrh & OD
MUD PRoPERTIES solids, cake, PV, YP, funnel vi.oositV, PH
SALT WATER INFLUX mud ch!ocides
HOLE CLEANING. am.!m velocity, flow rate, mud gel
strength, cutdmas carryiw capacity,
PV.YP product
WEUBORE STASILITY time inllength of open hole, stuck poinr
OIFFERENTSALSTICKING differential pressure, mud weight
HUMAN ERRORS
x
x
,
,
Table Z @l~ e ut the datsbns~ x

Gulf of North
Mexico Sea

Wells classified as differendallv’ stuck 19% 12%


Wells .i..sifiod as rn..h.n~.dlv st..k 37% 33%
Wells were drilled between 1986 & 1S90 .81% 85%
Maximum water depth 2609 fl 1616fl 2609 ft ,2n Z.21
‘mm.
42% .{, s? co CH G1
Water depth from 0 to 100 ft
Water depth fr.org100 t. 4D0 V.. .=41% 5.3.4; ,x . BACKWARD EUMIN4110N
w . FORWARD SEL=IION
Water depth from 401 to 500 ft “/. 22% sc . SIE?WISE co . WM81NE0 W’ +.EK+ fl
G, - GROUP2
Ut?odwater. bas?d mud 83% 45% & = c“EwO!JS RUN 39
U . GROUP 2
vertical 31% 36%
Build and hold 5s% 48% :.
77%
~2% . . . Table & GrGims o f variables for No rtb Sea wBM we 11s‘
Rotary drilled
Top drive drilled, . .::;. - ;;, ;;;,,; : 20% 27?6
...- . .
Drilled wirh 3 cone bits 72% 55% “ x
Drilled wirh PDC bits 21.% 3s% x
50% 21% x x
Jack .“P rigs
27%. x , , x x .
Platform rik?s , 21% x x x x
Semi. submersibles 1z% 50% ,
26% 28% A , x
8.1 /2” hole x x
x x x
$-7/8” hole 24% !/? . z x r
51%
,. d? rbr.i x
r
12.1 /4” hole 22% 1a d, mdwo x x A A L
./a 14% ,,. x A A x
17-1 /2” hole d,,.,”,
x-
-9% 22% ,2, dl~” x x
Depths bmwe.m 4000 and 7000 ft x x x , x
@,, ,3. dlyp x
Depths batwee” 7000 a“d 10000 ft 23% w. dlw;el x x x x x x
41% 1s% x x
Dwxhs b,lween 10000 and 1400D ft 15. d~mkser x
x .
6% ,6. dlaPi X x
Depths greater than 1wOO R 17% , x . x
,7. .dtph z x
Mud weights between 8.4 andl 0.0 PP9 15% 23% ,& 6%111.51 r. . x z,. x x ,
40% , x x x
Mud weights belweeh 10.1 and 12.0 ppg 29% ,s. dl%ol:ds x
x
20. *.6P 1 A
Mud weights between 12.1 and 14.0 ppg 20% .30% =. x x , ,
23. ,nw=6p x
Mud weights gm.cer than 14.0 PP9 36% 7% 2. Nd x ?. ?. x
r. A x x
23. .Pe”hd. x
. .. ., ... *.+... 24. 6.1wob x x x .
i x x *
25. ypp” ,
26, cd A
, x
x
x
x
x ,
.
A A x
x , : x z
x
x
,
x
,
. .. x
,7 <8 ,8 ,6 ,9 ?+

907
.. -,

8 TRACKING STUCK PIPE PROBABILITY WHILE DRILLING SPE/L4DC 27S28

Table & ~~..


K&b!!S *_ ., ,_ .; .. . . . .. . ... ..
Cw,an ;..
Gm.P of Number Num- Accmacy of Accuracy of SW&=w ~,c.. —
variables of vari- bar of classification’ classifimtlo-nz
able. wells
GULF OF MEXICCI (503 wellsl _ _
Gnckwacd 15 .217 135/217 = 62% 103/158 = 65%
Forward 8 179 105[179 = 59% 105/158 “~ 6;% 11$/1,, . ~,)
-SEPWISE 74 7.3(7,) sq7q 33{?2)
+ M,*g!,
Stepwiw 12 175 11 4/170 = 65% 102/159 = 05%
Combined .23. 170 121/170 = 71% 114/159 = 72% W.cwm. ,7 18(75) ~. 61 [821 32(,21 ,,3,,40. @,)
+ .SIdls
Chevron 20 210 130/21 0 = 62°% 102/180 = .94%
WMPLUS 24 !7(71] 55(72) %(,,] 2041138 = (75)
GrouP 1 27 1 S0 120[1 60 = 75% 120/180 = 75%
Group 2 38 115 91/115 =78% 91/115 =79%
Now Foreach caww, IWsinumber representsnumber
ofwellscorrectly
NORTH SEA Water base mud (523 wdlsl dssslfi,dandmcondn“nfmrinPnrenltrasesqwewis Percsntag*
GOMCASE 22 158 1211158 = 77% corlecllyCrmslmd

CHEVCA8E 17 167 11 2/167 = .67%


GOMPLUS 26 129 105/129 = 81°A
9ASE34 18 153 125/1 53 = 82% –:..
NSEABA8E 18 157 130/157 = 83%
SASE16 16 158 131/158 = 83%
PHYSCASE 19 128 88/1 29 = 68%
ALL24 ~ 34 129 112/128 = 88%

1From full database


9From fixed 160 wells

Table 6: .GniuDsof variables for Nortlf Sea 013M Welk _


~:1 ~, ___6.- s 6 7 s 8
.. L -.
Variables $IW.XE;*-.”8 _ MCC”MO ,s- - -. a.,,m ,r,m

xx xx x . xx
x x x ..x-
,., . , x x ., ~ u
Table 9: A f
x x x x
x, x , x x x
xx x“ , x x x, Water base mud Oil base mud ComblnaEori
xx xx , x .x
,, x, x x , GULF OF MEXICO
x ,. ., x xx
x. xx x , xx
,, x , x, 0“.,.[1 1.S0/75/78 46/92/83
xx ~.
,, x x Independent dam 16f12@
xx x ,. x .x .-.
. . . x“. ,
r ,X x x NoRTH S&
xx x x.
xx x xx
x. x x x x. 0“.,.s11 157/83/87 177i75/31 252/73/83
x * ComPany A ~ 7/1 00/100 33/73/85 45/64/87
. x
0 , Compmy B 28/97/97 12/75/83. 47/87/64
company c 2/100/100 19/5s/58
x
x company D 6/1 00/1 00 42{81 /31 10/90/100
x. .X .x comPany E 12J75133 7/8 S/1 DO 29/52/66
x .x
, xx Company F 41/81/88 24/67/75 36i78i78
x xx
x xx ComPany G 14/50/71 24/67/75 15/80/87
x xx Company H 25172180 23/52/61
company I 14/86/36 20/70/70 22/82/91
.
CornPany J 2/50/50 20/65/80
,. *
x x
1In the above table, the numbers xx/w/zz represent
-m- ,,,,,0 - ,m. ,,Ma -, ,s,,, - ,_qa ,H.. m = .“mber of wells classifie~ yy = ,Dercenxcorm.tly classified
ACCU.CY
by three-ciass model: zz = parcent corre.dy classified by lwo-
% 723 71s 74.1 80.7 7s.8 243 74.1 75.4 74.4
.!,ss modd

908
,,,
.~-

SPEIIADC 27528 J.A. HOWARD AND S.B. GLOVER 9

Table 10 ResuIt6 of rolliu tests for Guff of MeXiCQ Table. ~: llwtm sffecfim the grobabilitv of fre=zing
m.!2df?.
Gulf of Mexico. ~
Developing Stuck Mechanism
Model wbh Predicting Predicting
D-stuck 35/(35 + 60) ’37% 40/(40 + 27) = 60%
-140 wells 20 Wells ~a - M-stuck 86/(86 + 101) = 46% 134/(134 + 4.6) : 74%
D+M 1211282 = 43% .1741247 = 70%
Rolling 1 102/140 = 72.8”A 13;20 = ‘65Y. 10/18 = 56%
i10/140 = 78.6% 14/20”= 70% 11/18 = 61% stuck Time

Rolling 2 97/140 =.69.3% 14/’20 = 70% 11/18 =61% Hrs Free Stuck % Free Free Stuck 7. Free
104/140 = 74.3% 7%/20 = 75% 11/18 = 61% “’: - 2’4< 71 8 90% 132 2 997.
24.48 9 18 337. 15 12 56%
48-72 ~0 18 35% 9 9 50?.
Rolling 3 110/140 = 78.6% 9/20 = 45% 13/18 = 72%
72-96 3 13 19?. 6 77 26%
116/140 = 828% “11/20 = 55% 13/18 = 72%
96-120 6 23 2*% 6 10 38%
120+ 21 67 24?. 4 20 17%
Ro]llng 4 107/140” = 76.$0~ _, 9/20,= 45% 12/18 = 66-6%”
116/140 =. 8+8% 10/20 = 50% 13178 = 72%

R1 80,80
Model 120{160 = 75% R2 85,90 13118 = 727.
126/160 = 78.8% R3 55,60 14118 = 77.7%

Table 11: Acmu-acY of c1ossification in ven-od Drier to


@i@@g
R4 75:75

Ii!dIm3
23 .,, -=..
ToM .$”.k ~nw [dws)

M9M m. f 0,”2 Da” 3 Da” z 08”5 ?ime to PosiOonSpoUingFluid


“u 0, Wx!co
“o-4 22 .10 “70 39 12
W,!.,: 4-24 28 41 41 73
3*S, ,201,60 .75 8W36J -60 741!45 - 51 24+.. 4 11 26 2 3
2-s=” 124160 - 7s W31M8 - 63 .s0/!.5 - 55
Hr. Soak Bi;ore Jra,rlng

w., m , O*I 24 45 43 43 21 67%


2+ 18 48 27 9 8 53%
3<1,s6 ,28,156= 82 lJ4/’l5l
-74 !08,,4, .73 9, ,,s .,8 $41M .,0
2<,,,. ,3.,56. = ,7 122,5! .,, la,,,, . 82 ,0C,4, - ,, ,0,;,,4 .,,
Hrs Jar Fumtioned
=4 43 33 57 112 13
0!: —
-4-.24 30 52 -27 39 40
3<,***, ,,3,77 = .7s ,,;169 . ,, *51,6, . s, =;14, . s, w, 4, -55
,24-48 15 29 34 8 9
2<.s 164/,= . m TOW59 .64 SW* .,, 93”.4,.63 ,,,,4, . C2 .“ 48+ 4
7 14 33 4

Cmlblnwo”: Angle Type


-!.,s ,84,2,? .,, 1,,/24, = m ,,2.3s. ,, ,a,,ms .6, ,,amd .,,
,<1,,, 2,,,2,2= 83 ,,,,246= ti-,57i235 =67 ,62.z% = 72,4,i2LM -,9 Straight 31 52 8 87%
Build & hold 75 f: z 92 56 62%
S.type 15 ?9 44 30 9 77%

Angle
Table 13: lksco freeitw index .“ . ...
Degrees
GOM Data S20 73 19 79%
TFI Chances of Freeing Tsxaco’s Data s 30 18 9 .67%
o-45 .28Y. 5% S40 29 10 74%
45-55 38% 50% S50 20 14 597.
67% s 60 19 15 56%
55+ 93%
60+ 8 4 87%
North Sea Data oveQull
TFI Free Not Free % Free
86 s 100 31 23 57% 10’ lbs
100 s 115 41 24 63% 0-50 60 3 93%
115 s 130 81 18 62% 50-100 34 8 81%
100-150 24 4 86%
152-200 19 13 5%
200-300 22 22 50~%
300 + 15 14 52%

909
-.. , ~

10 TRACKING STUCK PIPE PROBABILITY wHILE DRILLING SPEAADC. 27S28


;-

. .=.
.===
..==
?. b...
...=
.
..
/“””””
\
\.
x .
.

.
=.
\ .
.
x
.
-

>“
~- .. .
a

=x ~ .. ,..-
3
. .=.

,. .. .. . .
z
.,
.
. .

.-
.
... +-.... ---------- . ----+---- ,-.. -.-4
——. —. — _
. . . .
7

..
\ “’” 1“
\====

.
. .

—--...+-. - .-,.—<
z-” T

910

Das könnte Ihnen auch gefallen