MƏQALƏ VƏ TEZİSLƏRİN TƏRTİBİ QAYDALARI
THE PREDICTION OF CASING RUNNING DRAG LOADS USING
MACHINE LEARNING TECHNIQUES.
Vusal Iskanda o Aki khan 1
i. [email protected]
1Aze baijan S a e Oil and Indus y Uni e si y
ABSTRACT
Key wo ds: D illing, Machine Lea ning, To que and D ag, D illing Op imiza ion.
The objec i e o his s udy was o build a model in o de o p edic ipping in weigh om he
eal ime da a. The s udy is aiming o suppo comp ehensi e o que and d ag analysis o casing
unning ope a ions and u he op imize he modelling job by cu ing he ime equi ed o analyse
he eal ime da a. This s udy in ol es se e al machine lea ning algo i hms so as o build he
model using eal ime da a. The eal ime da a om o sho e ields loca ed in Sou h Caspian
Basin was applied o ain he model. The di e en machine lea ning echniques, such as linea
and non-linea machine lea ning and deep a i icial neu al ne wo ks, ained model. The
e alua ion me ic o aining is Roo Mean Squa e E o , howe e , he pe o mances o he
eg essions a e e alua ed on he da a using R-squa ed o hei compa ison.
The need o build he model o op imize casing unning o que and d ag simula ions was aised
when he side- ack ope a ions we e commenced in he ma u e ield whe e i was equi ed o un
4.5” line in slim-hole condi ions wi h challenging well ajec o y and se e e dogleg se e i ies. A
couple o ailu es had al eady been occu ed in o se wells, in which he 4.5” line go s uck in
slim open hole sec ion. Also, in some cases, e en hough he line was success ully un o he
equi ed dep h, cemen ing ope a ions we e poo ly pe o med due o poo cen aliza ion in he
well. The e o e, as one o he imp o emen s o manage he isks and op imize he ope a ions,
comp ehensi e model is de e mined o be buil o u he op imize and p ecisely p edic o que
and d ag loads o casing unning ope a ions.
The pape will look a in dep h look a how model was buil o he i s well, hen calib a ed
wi h d illing da a, and how he ic ion ac o s oge he wi h pos job ic ion ac o s we e easily
analyzed. This pape also co e s why his s udy is impo an , how i is di e en om he exis ing
wo k lows and how machine lea ning can help o au oma e he ime-consuming p ocess in
d illing enginee ing.
GİRİŞ (In oduc ion)
Sou h Wa e Gunashli is an o sho e oil ield loca ed in Sou h Caspian Basin. The s a o he
p oduc ion om he ield was da ed back o 1982 and boomed since (Leonid , Geo ge , & F ed ,
2001). E en hough he ield de elopmen s a ed mo e han 40 yea s ago, he i s side acked
in- ill p oduc ion well was d illed in 2023. Du ing he d illing ope a ion, a lo o di icul ies had
been happened. Du ing he side- ack ope a ion planning and execu ion, i was becoming
appa en ha special enginee ing calcula ions and ca e should be conside ed o d illing and
casing unning ope a ions as he side- acked wells will ha e slim-hole sizes because o he old
well design.
In many cases, planned wells need 4 ½” p oduc ion line ac oss side acked hole sec ion, and his
c ea ed a lo o challenges ega ding o que and d ag o ces and ECD limi s in bo h line unning
and cemen ing ope a ions. A he beginning o he d illing campaigns, hese challenges had
esul ed in line s uck ac oss open hole, e en in one case, line go s uck app oxima ely 500 m
below casing exi window. This has igge ed d illing enginee ing and well planning eam o be
mo e p e-cau ious abou casing unning.
The goal o his esea ch is o sha e a case s udy o new enginee ing ool ha helps o o ecas
line unning d ag alues in eal- ime om Leuza 2 mud-logging sys em which was p elimina y
used in he d illing ope a ions.
The machine lea ning echniques a e equen ly being used o add ess challenging and ime-
consuming wo k p ocess in he indus y nowadays. One o he impo an bu ime-consuming
wo ks ha d illing enginee s do is analysing eal ime da a o u he op imize o que and d ag
simula ions. Usually, he d illing enginee s build o que and d ag model in ad anced so wa e,
such as WellPlan, D illBench e c, using essen ial inpu pa ame e s: wellbo e in o ma ion, open
hole size, casing es ic ions, well ajec o y, casing s ing da a, line unning ool in o ma ion,
line hange es ic ions, casing unning equipmen da a, ig limi a ions, cen alize da a and
cen alize placemen . To co ela e he model o ac ual expec ed weigh da a, one o he mos
essen ial pa ame e s is “ ic ion ac o ”. The e m “ ic ion ac o ”, is ob ained om ic ion, an
impo an sou ce o ene gy loss. Since he ubula s and downhole equipmen a e being mo ed in
and ou du ing he casing unning ope a ions, he esis i e o ce – ic ion o ce is obse ed in
he o m o d ag. Du ing he casing unning ope a ions, a ious ope a ions ha e ollowed each
o he – ipping in, ipping ou , ipping in o ou wi h and wi hou o a ion. These ob iously
complica ed he calcula ions, as a esul a ious ic ion ac o s a e used o de e mine hose
weigh s (Samuel, F ic ion ac o s: Wha a e hey o o que, d ag, ib a ion, bo om hole
assembly and ansien su ge/swab analyses?, 2010).
To add mo e con ex on ic ion ac o s as hey a e co e o his s udy - The F ic ion Fac o used
in To que and D ag calcula ions ep esen s he mul iplie applied o he side o ce o de e mine
he esul ing ic ional o ce. F ic ion wi hin he wellbo e a ises om a ious physical
mechanisms and is in luenced by ac o s such as he cha ac e is ics o he con ac ing su aces,
he ype o d illing luid, he wellbo e’s ajec o y, he ex en o con ac , and he p esence o
obs uc ions like d illed cu ings.
In mos scena ios, he lub ica ing p ope ies o he d illing luid play he mos signi ican ole in
de ining he app op ia e ic ion ac o . I 's common p ac ice o use di e en ic ion ac o s o
Cased Hole and Open Hole sec ions.
Conside ing he complexi ies o he simula ions o accu a ely de e mine he casing unning
weigh s, sensi i i y analysis o ic ion ac o s is usually pe o med upon any casing unning
ope a ions (Figu e 1).
Figu e 1 - To que and D ag Loads wi h a ious F ic ion Fac o s.
D illing enginee s may spend a couple o days on unning hose sensi i i ies and combining hem
in a single place. The ange o ic ion ac o s usually depends on he ield, leng h o he casing
s ing, hole size, well ajec o y, casing size and es ic ions in he wellbo e, and many o he
ac o s. The ic ion ac o s a e also a ied o open hole and cased hole. Fo complex ope a ions,
hey a e much mo e complica ed o analyze because hey depend on a ious o he ac o s,
including empe a u e, aspe i ies be ween he su ace, ype o ma e ials, sliding speed, and
olling speed in case o an objec o a ing ela i e o ano he . Also, he eal unknown is he
measu emen o he hookload a he su ace (Samuel, F ic ion ac o s: Wha a e hey o o que,
d ag, ib a ion, bo om hole assembly and ansien su ge/swab analyses?, 2010).
The sensi i i y analysis esul s o casing unning ope a ions gi es he sense o su ace
hookloads expec a ions and he ange whe e he casing unning weigh s may be expec ed (John
& Da id , 2013). The usual ange o he casing unning ic ion ac o s a e wi hin 0.15 o 0.30.
Howe e , in ex eme condi ions, hey may be well o e 0.45 as well (Rabba , 1985).
Once he casing unning ope a ions a e comple ed, he d illing enginee s collec he mud logging
ou pu s wi h all he eal- ime da a om he ig and analyse he casing unning loads (Samuel &
Jamal, D illing Enginee ing, 2007). They ma ch he ipping in, ipping ou , ipping in o ou
wi h and wi hou o a ion weigh s eading om he mud logging sys em and check hem agains
he mode ou pu s. This p ocess helps d illing enginee s and well planne s o check he ic ion
ac o s and calib a e hei models o he u u e well planning ac i i ies
MATERİAL VƏ METODLAR (Me hods)
The main concep o he ools lies on machine lea ning (ML) echniques and combina ion o ML
echniques wi h o que and d ag simula ions. Machine lea ning echniques we e analysed o
de e mine line unning o ces om he eal- ime da a se s om mud-logge sys em and hen o
calib a e i o op imum selec ion o ic ion ac o s o he nex well deli e y.
The main p oblem was unning line o equi ed dep h. I may sound easy, ye due o long open
hole sec ion, co e ing mul iple laye s, equi emen o isola e high p essu e amp and deple ed
ese oi s in a single sec ion, as well as high dog-leg se e i y ac oss casing exi , exace ba ed he
si ua ion in almos each well. As he 4 ½” line leng h was app oxima ely in ange o 1200 –
1700 m, ex ensi e enginee ing calcula ions we e equi ed o be done in well planning s age.
ƏLDƏ OLUNAN NƏTİCƏLƏR (Resul s)
The enginee ing solu ion was c ea ed and es ed agains he mul iple well da a se s. The models
a e ained using a a ie y o machine lea ning algo i hms using mud logge da a om he Leuza
2 p og am. The sugges ed app oach o e s p ecise ic ion ac o s and i s calib a ion o o que
and d ag so wa e in addi ion o help wi h d ag o ce in well planning. O all he me hods
examined, Random Fo es u ns ou o be he bes op ion because i pe o ms be e in e ms o
accu acy and compu ing e iciency. This s udy ills a signi ican gap in he ield by p o iding a
wo kable solu ion o imp o ed ope a ional planning in non- adi ional wellbo e ci cums ances.
This p ojec 's main goal is o c ea e and e alua e a p edic i e model o casing unning d ag
alues. Wi h he use o machine lea ning echniques and mud logge da a om he Leuza 2
p og am, he model aims o imp o e well deli e y and well in eg i y by calib a ing FFs o well
planning. This model can be used in p ac ice o p o ide accu a e ic ion ac o s o o que and
d ag so wa e as well as eal- ime moni o ing. Wi h an emphasis on e iciency, accu acy, and
compu a ional e ec i eness, he p ojec seeks o enhance ope a ional planning in uncon en ional
wellbo e si ua ions, ul ima ely ad ancing d illing echnologies.
To c ea e a p edic i e model o line unning d ag alues, his esea ch makes use o mud logge
da a om he Leuza 2 so wa e. Basically, om Leuza da a and conside ing line unning
p ocedu es, i is manually challenging o ob ain op imum unning loads and ge exac unning
loads h oughou he wellbo e. The e o e, his model was c ea ed. Fu he mo e, he esea ch da a
is ob ained om eal- ime d illing pa ame e s using a a ie y o supe ised machine-lea ning
algo i hms, including linea and non-linea models. The Random Fo es app oach is p e e ed in
a compa a i e s udy due o i s highe accu acy and compu a ional e iciency. The co ec ness o
he model is e alua ed h ough he use o he Roo Mean Squa e E o (RMSE). The model's
inal use will enable eal- ime d ag o ce moni o ing, exac ic ion ac o p o ision o o que
and d ag so wa e, and ope a ional planning op imiza ion in non-s anda d wellbo e
ci cums ances.
Figu e 2 - De e mined unning loads o line .
Op imum ic ion ac o s we e ob ained by using ML echniques om he e y i s well and
hese da a we e implemen ed in o he nex well campaign o imp o e he well cen aliza ion plan
which allowed o o enhancemen o cemen bond long and zonal isola ion.
Du ing he execu ion, his ool was qui e c ucial o keeping an eye on d ag o ces when
deploying a 4.5" line in di icul and igh wellbo e condi ions.
This esul was ini ially checked manually wi h d ille s unning loads and was con i med o be
co ec .
MÜZAKİRƏ (Discussion)
In conclusion, his s udy success ully es ablishes he Random Fo es algo i hm as an op imal
solu ion o p edic ing line unning d ag alues in slim-hole side ack ope a ions, achie ing a
ema kable a e age R-squa ed o 0.90. The p ac ical iabili y o he de eloped model in eal-
ime d ag o ce moni o ing, coupled wi h i s abili y o p o ide accu a e ic ion ac o s,
unde sco es i s signi icance o ope a ional planning in uncon en ional wellbo e scena ios.
These indings con ibu e o ad ancemen s in d illing echnologies, o e ing a eliable ool o
op imizing e iciency and p ecision in p edic ing and managing line unning d ag alues.
ƏDƏBİYYAT SİYAHISI
John , M., & Da id , W. (2013). A Wo k Me hod o Analyzing F ic ion Fac o s in To que and
D ag Modeling. SPE Uncon en ional Resou ces Con e ence Canada.
doi:h ps://doi.o g/10.2118/167172-MS
Leonid , B., Geo ge , C., & F ed , A. (2001). Pe oleum Geology o he Sou h Caspian Basin. (E.
Inc, Ed.) Gul P o essional Publishing. doi:2001
Rabba . (1985). F ic ion Coe icien o S eel on Conc e e o G ou . Jou nal o S uc u al
Enginee ing 111, 505–515. doi: h ps://doi.o g/10.1061/(ASCE)0733-9445(1985)111:3(505)
Samuel, R. (2010, Sep embe ). F ic ion ac o s: Wha a e hey o o que, d ag, ib a ion,
bo om hole assembly and ansien su ge/swab analyses? Jou nal o Pe oleum Science and
Enginee ing, Volume 73 (3-4), 258-266. doi:h ps://doi.o g/10.1016/j.pe ol.2010.07.007
Samuel, R. (Feb ua y 2010). F ic ion Fac o s: Wha a e They o To que, D ag, Vib a ion,
Bo om Hole Assembly and T ansien Su ge/Swab Analyses? IADC/SPE D illing Con e ence
and Exhibi ion. doi:h ps://doi.o g/10.2118/128059-MS
Samuel, R., & Jamal, A. (2007). D illing Enginee ing. PennWell Co p. doi:ISBN-13 : 978-
1593700720
