4 views

Original Title: Bachelor Thesis -Ahmedmsahli_ar-final (1)

Uploaded by Waleed Barakat Maria

- Proceedings_2013.pdf
- Genetic Algorithm with SRM SVM Classifier for Face Verification
- 54
- entropy-15-00416-v2
- rapidminer-4.3-guimanual
- Key for Dm II Mid 2017 May Set A
- 1-s2.0-S0957417414003455-main
- prt%3A978-0-387-74759-0%2F13
- Short Term Load Forecasting Using a Hybrid Model Based on Support Vector Regression
- 05601721
- 7
- Drilling Mudd
- Limpieza en Huecos Direccionales y Horizontales-new Drilling Technology
- classification3
- FreyReport
- Trapped Annular Pressure Gonzales 4th 2008
- Acoustic Signal Based Fault Detection
- 1969_6_1
- Lost Circulation Assessment and Planning Program[1]
- ModEco Manual

You are on page 1of 61

Pipe Sticking

Causes & Preventions

A Bachelor Thesis

in the

Mining University of Leoben

Leoben, Austria

Mai 2016

by

Msahli Ahmed

Supervisors

MSc. Roohi Abbas

Ahmed Msahli 1

Pipe Sticking 2016

Abstract

Studying and researching the roots of problems is the only way to better understand processes and

move to solutions.

In order to have better successful drilling operations, the elimination and mitigation of risks and

drilling problem is necessary. And in order to achieve that, a much deeper understanding of the

roots and physical reasons behind these problems in needed.

Having an understanding of the problem will eventually lead to the establishment of solutions and

practices that improve the overall productivity of a drilling operation. Furthermore, Using

mathematical models and methods will lead to having more sophisticated and useful results

The first objective behind this study is to establish a better understanding to pipe sticking and the

physical and mechanical reasons behind every occurrence. Then, to know the possible means of

detection and prevention of the phenomenon based on case studies and realistic results.

Later, this study aim to discuss the most efficient methods of detection and risk estimation via

introducing Artificial Neural Networks and Support Vector Machine, Evaluate their performance and

potential as well as comparing them based on the case study results.

And finally, the study aim to have an idea about further development of the studies and the

upcoming possibilities.

Ahmed Msahli 2

Pipe Sticking 2016

Table of Contents

Abstract ................................................................................................................................................... 2

Objectives of this study ............................................................................ Error! Bookmark not defined.

List of Figures .......................................................................................................................................... 5

List of Tables ........................................................................................................................................... 6

CHAPTER 1: INTRODUCTION: .................................................................................................................. 7

WHAT IS PIPE STICKING?.......................................................................... Error! Bookmark not defined.

Size of the problem .................................................................................. Error! Bookmark not defined.

CHAPTER 2 : MECHANICAL STICKING : .................................................................................................... 7

2.1 Types of Mechanical Sticking ............................................................................................................ 8

2.2 Freeing Methods for Mechanical Sticking ...................................................................................... 10

CHAPTER 3 : DIFFERENTIAL STICKING : ................................................................................................. 11

3.1 Theory behind DPS .......................................................................................................................... 11

3.2 Elimination of DPS ........................................................................................................................... 13

3.2.1 Practices that help Minimizing contact area ............................................................................... 14

3.2.2 Minimize Still-Pipe time ............................................................................................................... 17

3.2.3 Cake morphology and fluids design ............................................................................................. 18

3.2.4 Modeling ...................................................................................................................................... 21

3.2.5 Results .......................................................................................................................................... 25

CHAPTER 4: SIDE FACTORS CONTRIBUITING IN PIPE STICKING: ........................................................... 27

4.1 Human Factor.................................................................................................................................. 27

4.2 Hole Angle ....................................................................................................................................... 27

4.3 Hole Diameter ................................................................................................................................. 28

CHAPTER 5: CASE STUDY 1: REAL TIME FRICTION ANALYSIS: .............................................................. 29

5.1 Wellbore Friction Modeling ............................................................................................................ 29

5.2 Buoyancy Effect............................................................................................................................... 30

5.3 Contact surface Effect ..................................................................................................................... 31

5.4 Results and Discussion .................................................................................................................... 31

CHAPTER 6: ARTIFICIAL NEURAL NETWORKS: ..................................................................................... 39

6.1 Concept of ANN .............................................................................................................................. 39

Ahmed Msahli 3

Pipe Sticking 2016

6.3 Applied Neural Networks ................................................................................................................ 41

6.4 Training and Learning...................................................................................................................... 42

6.5 Evaluation criteria and Limitations ................................................................................................. 44

6.6 Apply Production Datasets.............................................................................................................. 45

6.7 Sensitivity Analysis .......................................................................................................................... 46

CHAPTER 7 : SUPPORT VECTOR MACHINE : .......................................................................................... 48

7.1 Concept of Support Vector Machine .............................................................................................. 48

7.2 Predictor Performance Criteria ....................................................................................................... 51

7.3 Results and Discussion .................................................................................................................... 52

Conclusions ........................................................................................................................................... 57

Bibliography .......................................................................................................................................... 59

Ahmed Msahli 4

Pipe Sticking 2016

List of Figures

Figure 2: Cuttings accumulation ............................................................................................................. 8

Figure 3: Borehole Instability .................................................................................................................. 9

Figure 4: Key Seating ............................................................................................................................. 10

Figure 5: Embedment by differential pressure ..................................................................................... 12

Figure 6: Differential pipe sticking ........................................................................................................ 13

Figure 7: Stress and shear development for DPS.................................................................................. 13

Figure 8: Comparison of the contact areas in DP and HWDP ............................................................... 16

Figure 9: Example of wear groove that frequently appears in the 3D images processed from high-

angle wells............................................................................................................................................. 16

Figure 10: Effect of Centralizer position on probability of success ...................................................... 22

Figure 11: FE model schematic apparatus ............................................................................................ 23

Figure 12: Pullout Force between WBM and NAF models ................................................................... 24

Figure 13: Differential Axial stress vs. Axial & Radial strain% ............................................................... 24

Figure 14: Occurrence of DPS between 2004 and 2008 ....................................................................... 25

Figure 15: Freeing success rate vs. hole angle ...................................................................................... 27

Figure 16: Stuck pipe occurrence vs. hole angle ................................................................................... 28

Figure 17:Studied well geometry .......................................................................................................... 32

Figure 18: Friction coefficient over the BHA movement ...................................................................... 32

Figure 19: Buoyancy factor vs. measured depth for drilling, tripping in and tripping out ................... 33

Figure 20: Mud pressure over the wellbore ......................................................................................... 34

Figure 21: Hook load vs. measured depth for tripping in and tripping out .......................................... 35

Figure 22: Frictional force during the tripping out process .................................................................. 35

Figure 23: Overall friction coefficient vs. measured depth during tripping out operation .................. 36

Figure 24: Frictional force during the tripping in process ..................................................................... 37

Figure 25: Overall friction coefficient vs. measured depth during tripping in operation ..................... 37

Figure 26: General structure of ANNs ................................................................................................... 40

Figure 27: Simplified structure of ANNs ............................................................................................... 42

Figure 28: MSE versus a default number of epochs for MLP model..................................................... 43

Figure 29: MSE versus a default number of epochs for RBF model ..................................................... 44

Figure 30: Relative importance of drilling parameters for MLP ........................................................... 46

Figure 31: Relative importance of drilling parameters for RBF ............................................................ 47

Figure 32: Maximum-margin hyperplanes for SVM trained with examples from two classes............. 49

Figure 33: Mapping input space into high-dimentional feature space................................................. 50

Figure 34: Comparison of SVM results , FFBP results and real data from field .................................... 56

Ahmed Msahli 5

Pipe Sticking 2016

List of Tables

Table 2: Freeing stastics in the studied wells........................................................................................ 25

Table 3: Applied datasets example ....................................................................................................... 40

Table 4: Error of training and cross validation for MLP ........................................................................ 44

Table 5: Error of training and cross validation for RBF ......................................................................... 44

Table 6: Accuracy measurements ......................................................................................................... 45

Table 7: Well data before sticking day .................................................................................................. 45

Table 8: RBF and MLP predictions ........................................................................................................ 46

Table 9: Well parameters used for the study ....................................................................................... 52

Table 10 : Performance of transfer functions ....................................................................................... 53

Table 11: Performance evaluation for Gaussian Kernel function ......................................................... 54

Table 12: Performance evaluation for polynomial Kernel function ..................................................... 54

Table 13: Comparison of Performance for FFBP and SVM ................................................................... 55

Table 14: difference in the efficiency between the best developed SVM and ANN models ................ 55

Table 15: Comparison of SVM and FFBP predictions with real cases ................................................... 56

Ahmed Msahli 6

Pipe Sticking 2016

CHAPTER 1: INTRODUCTION

During drilling many problems would occur, understanding the physics behind these problems will

help avoiding their occurrence in future operation and thus minimize the costs of the operations and

minimize the drilling time which is beside money costing will decide if projects can stay on schedule.

It could occur due to many reasons and consists of having the drilling pipe stuck literally in the

wellbore without being possibly rotated or translated or both in most cases.

The pipe is in this case considered stuck if it cannot be freed without being damaged or without

exceeding the Maximum allowed rig hook load.

Mainly pipe sticking is classified into differential sticking and mechanical sticking. Each of have many

different causes and occurrence scenarios.

BUDDY J. DOMANGUE , Texaco USA : “ After well control , stuck pipe is probably our biggest concern

in the Gulf of Mexico , Between 1981 and 1990 , Texaco drilled 383 wells and experienced 105

incidents of stuck pipe in 95 wells _ a 27.4% frequency . In that period, stuck pipe in GoM cost

Texaco $65.8 million _ an average of about $626.000 per incident. We lost 1122 days of rig time _ an

equivalent of having a rig sit idle for 3.1 years at a daily cost of $58.000.”

Ahmed Msahli 7

Pipe Sticking 2016

Mechanical sticking occurs when the pipe is in motion and goes into many scenarios defined based

on the direct cause that gets the pipe to be stuck in the wellbore.

Drilled Cuttings

Because of improper wellbore cleaning during drilling, cuttings may be accumulating in the annular

space, which will end up by causing the drilling pipe to be stuck. Unless the cutting bed is flushed

before tripping out this problem is noticed in deviated well by an increase in torque, drag and

circulating pressure.

Ahmed Msahli 8

Pipe Sticking 2016

Borehole Instability

The well borehole can be unstable due to many reasons, but basically the problem has been noticed

during drilling in shale formations.

Shale sloughing or flowing inward will trap the pipe and cause it sticking and raising the circulation

pressure and probably keep fluids from returning to the surface.

In less common rate, the same problems are caused by drilling with a low weight mud which will

lead to borehole collapse and salt flow inward in case of overburden.

The sticking in this case is indicated by a rise in circulating drill-pipe-pressure, increase in torque and

in some cases no fluid return.

Key Seating

In the case of key seating, the drill string rotation when combined with a lateral (side) force acting on

it will be pushed to create a groove that consists of a small hole on the side of a full gauge hole .

The main conditions to create key seating are either doglegs or undetected ledges. And the

occurrence could be indicated by being able to pull out several stands of pipe before the pipe gets

stuck.

Long bit runs are usually known of causing key seats so as a way to avoid this problem making wiper

trips is recommended.

(1)

Where θ.dl represents the Dogleg severity and T represents the tensile force.

Ahmed Msahli 9

Pipe Sticking 2016

For each of the problems causing the pipe to get mechanically stuck an appropriate solution has

been engineered.

To get through cuttings accumulation and hole slaughtering it is suggested to alternate between

rotating and reciprocating the drill string with increasing the flow rate without exceeding the

equivalent circulation density ECD.

As for narrowing , if it is a result of plastic shale behavior then increasing the mud weight would

solve the problem and if it caused by salt when drilling salt domes as an example the use of fresh

water for drilling is recommended .

For the key-seat area, it is recommended to back off below the key seat and go back into the hole

with an opener to drill out the key section.

And finally for fishing operation it would depend on the choice of the operating company, because

choosing between fishing and backing off then plugging is an economical issue basically.

Ahmed Msahli 10

Pipe Sticking 2016

We have differential pipe sticking when the drill string, wireline or other surfaces are held against

the borehole wall by forces that develop in an area of contact with permeable formations.

If

Mud Pressure Pm is greater than Formation Pressure Pff

Then

The Pipe is differentially stuck

DPS became a major concern when the rate of sticking increased with the tendency of the industry

to move into abnormal pressure and high overbalance and directional drilling.

When the pipe become stationary, the pressure within the contact area begins to decline

immediately then this continues as long as there is sufficient differential pressure between the cake

and formation to extract filtrate from the cake. When flow from the cake stops, the pressure within

it will be close to that of formation. As the fluid pressure declines, the differential force across the

pipe is transferred to the solids in the cake. The stress between the solids is the effective stress.

Even the dominant force is usually associated with the pressure differential between the borehole

and formation in the contact area, adhesion and cohesion may also contribute some resistance to

pipe movement.

The general properties of the drilling fluid and mud cake, BHA, still pipe time and differential

pressure between the formation and the drilling mud should be considered to provide a concept of

how they are affecting the DPS.

To free the pipe, two theories exist. The first one is pulling the pipe hard enough to overcome the

shear resistance that exists across the entire contact area and the second is that the effective stress

causes a high sliding resistance and thus overcoming the friction existing across the entire contact

area is needed.

In both theories, the contact area is the dominant factor controlling the BHA, but based on

Laboratory tests shear within the cake is believed to be the dominant mechanism in the field.

Ahmed Msahli 11

Pipe Sticking 2016

Pm is the pressure acting on the outsider wall the pipe, and this pressure is usually higher than Pff

except for the case of underbalanced drilling.

The differential Pressure acting on the stuck portion of the drill pipe is given simply by:

Ac , Area of contact

Ahmed Msahli 12

Pipe Sticking 2016

The pullout pressure is believed to be dependent on cake shear strength, then the sticking tendency

can be reduced by practices that reduce this strength.

Ahmed Msahli 13

Pipe Sticking 2016

As it is seen in the previous graph the pressure decline is time dependent that means the effective

stress and the shear strength that develops from this must be also time dependent. In other words,

even some sticking force develops immediately, time is required for enough filtrate loss to occur

sufficiently for shear strength and pullout force to become high.

Other result of the relationship between filtrate loss, strength development and time is that thin

cakes may develop shear strength much faster than thick ones because the less time required to lose

the internal pressure while thicker cakes tend to have less shear strength despite having a larger

contact area.

Basically the safe pipe stationary time depends on the pressure differential, contact area and

properties of the cake itself, while it does not depend on the permeability of the formation against

which it is lying since the permeability will only affect the rate of decline of the cake’s internal

pressure to end up at the same level eventually.

Initially the studies concerning DPS were in the objective of avoiding differential sticking . Now that

this phenomenon is better understood, the philosophy of the studies is now tending to ensure that

conditions are maintained at all times that allow pulling force to exceed sticking force. This new goal

is proven to be feasible virtually in all operations by field experiences , and with that a modeling

process has been developed and calibrated to allow design of contact areas and BHAs to ensure that

the sticking force can be overcome . If key designs are applied consistently such specific modeling is

used for extreme conditions, unique fluids and unique BHA and completions designs such as swell

packers.

Large diameter drill collars (DC) have been used to provide WOB and their stiffness was designed

that the prevent buckling. Being thick-walled, their weight per foot is high and very little length must

be put into compression to achieve the desired WOB . The DCs are prone to sticking , since they have

a very large contact areas , first because the ratio of their curvature may be close to the curvature of

the borehole wall so that the contact area rapidly increase as the cake thickness increases and

secondly because their OD in uniform they make contact along their length 30ft .

From this came the option of using HWDP in compression to provide the WOB, an option able to

reduce the contact area from 30ft to around 6ft per joint and the design allows a wear pad section in

the middle to prevent contact along the tube body. Thus, any use of non-supported DC in directional

drilling is prohibited by operators for the fact that only the contact area of one unsupported DC is

enough to prevent the pipe from being pulled out in many situations.

This is observed in high-angle wells, knowing that it may be impossible to deliver the required force

to the stuck point if even a small amount of DC contact is allowed, regardless of the tensile pull

capacity at the surface.

Ahmed Msahli 14

Pipe Sticking 2016

There is factor that affects the compression of HWDP is that polycrystalline-diamond bits are often

running on at a low WOB. Generally, bit weights in soft formations are in the range of 5000 to 20000

lbf while a typical BHA assembly get the first 15000 to 25000 lbf of WOB from the DCs, Logging While

Drilling (LWD) tools and other stabilized members in compression before further slackoff would

place the HWDP into compression, in this case the HWDP is never actually compressed.

Generally for this practice the design should lead to a minimum number of stabilized DCs and run

the compression in the HWDP up to the point of helical buckling.

Slick assemblies are common in the industry especially in hard formations drilled with bent-housing

motors, and for many reasons directional drillers prefer slick assemblies first that they believe the

stabilizers hang up and reduce the ability to slide the motor when drilling and secondly because with

changing the WOB they become able to adjust the build rate to allow them to catch up with the

planned trajectory if they get behind the desired rate or to slow down in case ahead, but this

variability in WOB is resulting in larger doglegs and more bit whirl.

Even they are common, slick assemblies are not recommended in any type of application. Instead all

BHAs must be fully stabilized in a way that the spacing and the number of the stabilizers are ensuring

no contact between the DCs and formation wall.

First Guidelines allowed on or two DCs above the top stabilizer for vertical wells since no significant

wall contact in the first 60ft was assumed, but sticking events occurred anyway. In recent years this

allowance is reduced to maximum one DC. And in directional well it is recommended that the HWDP

be made up directly to the top stabilizer so there are not unsupported DCs.

Though HWDP has significantly less contact area than DC, the 6ft borehole contact length is reduced

to less than 3ft per feet with the use of conventional drillpipe (DP) with shorted tool joints as shown

in the figure below, especially used for intermediate and high-angle wells.

Conventional drillpipe can be used at low angles but with a limited reach because of the low force at

which it buckles helically. However at intermediate to higher angles buckling is supported and a

significantly greater compression can be out into the thinner tube body of the conventional pipe

(Dawson and Paslay 1984).

For this the use of conventional DP instead of HWDP in higher angles will help reducing the contact

area while being buckling risk-free.

Ahmed Msahli 15

Pipe Sticking 2016

A single jar has been found as the root cause in many sticking events. Thus a drilling jar poses the

same risk in terms of differential sticking potential as an unsupported DC.

For this standoff subs are recommended on all jars to prevent wall contact, the design of the subs

depends on the vendor.

It has been thought that the progressive sticking of the tool joints during the time the pipe did not

move is the cause of pipe sticking at high angles while the stuck point is defined in the drillpipe

above the BHA. This assumption did not relate to the speed and the resistance associated with some

of these sticking events.

After 3D high-resolution images became available, they showed the development of a groove in the

bottom on high-angle wells, created by the rotation of the tool joints or the tube body against the

bottom of the hole as shown in the figure below.

Figure 9: Example of wear groove that frequently appears in the 3D images processed from high-angle wells (3)

Ahmed Msahli 16

Pipe Sticking 2016

The severity would depend on the rock hardness, normal force, string rotation, and roughness of the

tool joints. The groove persists for thousands of feet with no mechanical resistance that has been

noted.

The grooves eventually create a high differential-sticking risk because of the very close curvature to

the tool joints that created it and this increase in contact area result in a dramatic increase in the

pullout force. For example, a 5 ½’’ tool joint with a common 7’’ OD might normally have a contact

arc of 1’’ to 2’’. If the tool joint is lying in the groove and has 60° of contact, such as the figure on

top, the arc increases to more than 6’’.

Beside the contact area, the differential sticking may occur more quickly and the shear strength of

the cake may be higher. If the pipe is rotated in place the tool joint may wear the cake down until it

is thinner or completely gone, also reaming the stands allows the tool joints above the BHA assembly

to wear down the groove just before the pipe stops to make a connection, even this is unproven, it

explains two facts, the speed of strength development and the high pullout force.

When multiple tool joints are stuck, mechanical shear strength in the wall cake may grow

immediately above the tool joint and this additional resistance make it a combination of differential

and mechanical sticking.

Yet the groove cannot be eliminated but the impact minimization is possible, by reducing the vertical

load, wear on the bottom of the hole and the groove development rate by replacing HWDP by

conventional pipe and also by using shorter tool joints to reduce the contact area along the groove

by 50 to 70% . Beside these solutions , other measures exists such as higher drill rates , lower whirl

levels ( detrimental condition where a bit bites into a part of the hole off center and forms a pivot

point that creates impact of the bit and some of the string with the borehole part ) and less

backreaming on connections .

Actual strength development versus time cannot be predicted, but the time dependence of

differential sticking has been reported by many researchers. As the figure (one) * showed, the cake

has little shear strength until fluid is lost and its internal pressure drops.

Field methods have been developed to alert the crew when changes in the fluid properties occurred

and increasing the risk of sticking (Reid et al. 1996) and a real-time surveillance process has been

developed to address it . This test; Progressive sticking test, is conducted before making a

connection in case of high risk. It consists of letting the pipe sit still and measuring the force required

to move it back again. The still-pipe time is increased to enough to make a connection (5 to 10

minutes) and if the trend in the pullout force is acceptable then the crew proceeds to making

connection.

The progressive sticking test is required when there is potential for tool-joint sticking , then if the

pullout force is seen to increase it is assumed because of the engagement of tool-joint contact area

above BHA with the possibility of have wear groove.

Ahmed Msahli 17

Pipe Sticking 2016

There are two desired designs in the filter cake, first that the cake should be thin in order to

minimize the contact area and the second is to have a slow rate of filtrate loss to the formation to

allow greater still-pipe time.

Effective cakes have both blocking solids to prevent other solids from entering the formation pore

throats and filtration control to prevent the fluid phase from passing through the blocking solids.

Barite is the primary blocking solid in the majority of fluids, but barite particles may be too small to

effectively block pore throats in higher permeability which may lead to both solids and filtrate to

enter the formation. Low-weight fluids may have very little barite presence but in either case the

cake thickens and the sticking risk increases. Thus, the cake quality is both dependent blocking solids

and filtration-control design. (Fisk et al. 1990)

Recommended practices 13B1 and 13B2 by the American Petroleum Institute (API) describe testing

protocols for fluid loss (FL) in WBF and OBF and include low pressure FL tests, high-

temperature/high-pressure (HT/HP) tests and particle-plugging tests (PPTs). The filtration test can

achieve real low values of 1 to 2 μm in pore size which correspond to sandstone with less than 5 md

of permeability while the majority DS risk is in higher permeability range, usually 500 to 7,000 md. As

a consequence relying on these tests resulted in high treatment costs because field personnel had to

continuously add filtration control material to drive down the FL values, as the continued to

experience pulls on connections or high drag while tripping, while the actual issue is lack of solids

and that does not show up in FL or HT/HP tests.

Thus, it is recommended to use the PPTs that simulates the local permeability and expected

differential pressure and temperature through a filtration medium.

Depending on the PPTs results other blocking solids in addition to barite are used in cases where

low-weight muds are used, permeability is above 1 or 2 d or where field experience shows cake

regrowth occurrence and not just in order to block pore throats but also to block the interstitial

spaces in the cake itself, these particles are from ranges of 5 to 50 μm and most commonly used

ones are calcium carbonate and the D50.

As example of additional bridging particles, Safe-Carb is preferred since it is generally harder and

purer than limestone and this purity provides a nearly complete acid solubility, more than 98% in

15% HCL at 24.4°C (76° F).

As for size ranges for D50 are shown in the table below

Ahmed Msahli 18

Pipe Sticking 2016

Stabilizers serve an important role in cake conditioning beside in minimizing contact area. Initial cake

quality is very poor and would contain quantities of undesired particles in all cases, if it is built by

drill solids in lightweight fluids and high drill rate or even if it is built by barite in high mud weight

and low drill rate.

While rotating, the stabilizers shear the cake to a diameter equal to the diameter of the stabilizer

which is nearly equal to the diameter of the hole. As this shear remove the larger particles, with time

the size of the blocking solids in the cake surface will become finer so the remaining gaps are more

easily scaled by filtration-control material. This leads the permeability of the surface of the cake to

decline progressively.

The degree of conditioning achieved by rotating stabilizers depends on many factors. One factor is

the drill rate, as it increases any given foot will be wiped fewer times. Another factor, the number of

stabilizers directly affect the conditioning efficiency as it is interpreted as the number of blades to

the point that if a slick assembly is used the only conditioning will be the one done by the gauge area

of the polycrystalline-diamond-compact bit. Else, cakes built from lightweight fluids require more

conditioning because of lack in barite.

If unfavorable conditions such as cake regrowth in a tight hole on trips, the interval may be reamed

while clean mud is circulated before making each connection.

After gas prices increase in the mid-90s, deeper reservoirs became economic which increased the

wells drilled through the severely depleted reservoirs. As a result pressure of 2,000 to 5,000 psi

became common in regions of Texas and the Gulf of Mexico. With these well, obtaining pressure

Ahmed Msahli 19

Pipe Sticking 2016

D&S was developed back then to reduce the occurrence of wireline differential sticking, but its use

expended to eliminate drillstring sticking. In this process stabilizers ream the original cake in the

presence of a pill that is rich in the appropriate blocking solids for the given formation as well as the

needed filtration-control material.

The D&S procedure consists of pumping the pill with timing it to arrive at the bit as the next stand of

drillpipe is drilled down. As the pill enters the annulus, the pump rate is reduced to a very low rate,

and the string is reciprocated and rotated as the pill is pumped up the annulus. As the stabilizers

rotate, they strip the original cake and the rich content of the D&S pill accelerates the fine-particle-

selection process at the re-exposed cake surface. At the end, a low-porosity, low-permeability cake

is formed in a short period of time and after the pill has passed across the BHA the connection is

made and the drilling is resumed.

One pill can treat the length of the stabilized BHA plus the stroke in the derrick while reciprocating

(150 to 180 ft).

After proven to be a successful process, D&S process is now a routine across the global organization

when overbalance is expected to exceed 2,000 psi or where there has been a history of chronic cake

regrowth. Also, the process is considered economical for some situations where instead of treating

the entire mud system with blocking solids and higher concentrations of filtration-controls additives

it became possible to treat a short interval.

The cases where the D&S process did not prevent sticking, the sticking point was found in the HWDP

above the stabilized zone and usually suspected the presence of a wear groove.

Since the stuck-pipe-avoidance practices were implemented, there have been no DS events in water-

baser mud as the recommended practices were followed. All the sticking events with compliant BHA

designs have been in nonaqueous fluids (NAF), referred to as oil-based mud. While NAF is known to

reduce sticking having the facts that NAF filter cakes are thin to minimize the contact area and they

also have relatively low shear strength under a given effective stress. So this can be explained that in

each case of sticking in NAF the stuck point was found to be in HWDP and that the cumulative-

contact-area wear groove may have been the root cause.

The effect of additives such as surface-active agents, polymers and emulsified oil was also reported

beside other early studies on the pullout force associated with various water-based muds. However,

none of these results has been incorporated as a standard component in the sticking-avoidance

practices.

So, wells with highest sticking risk are usually drilled with NAF. Else, a combination of minimized

contact area, the use of appropriate blocking solids and D&S operations in specific sands with severe

overbalance is used for challenges with water-based mud.

Ahmed Msahli 20

Pipe Sticking 2016

necessary because the mitigations differ. With drilling the thickness of a previously established cake

continues to grow and the shear strength increases as the cake loses fluid and the effective stress

increases. If this cake has significant thickness, drag will be observed when the top stabilizer in the

BHA arrives at the permeable formation while tripping out and if the driller keeps on increasing the

pull, the stabilizer will shear farther into the cake and eventually become stuck. This will end up with

the stabilizer being differentially stuck but the event main cause is the mechanical resistance of the

high shear strength cake.

Mechanical cake sticking occurs while the pipe is moving while differential sticking occurs while the

pipe is still. But both events are affected by cake and fluid design thus any practice that reduces the

cake thickness and shear strength will always be beneficial.

Mechanical sticking is easily avoided by stopping when only a low level of drag is observed, lowering

the string, engaging the top drive and reaming slowly up through the sand to remove and condition

the filter cake. That is why training the drillers is essential in order not to pull too far into the cake

before stopping and to know the potential trouble zones.

Regrowth suggests that the blocking solids or filtration control is not properly designed as the cake

was not there after the initial drilling. There, the full fluid system may be modified, or a D&S

treatment may be used if only one area is proven to be problematic.

Reduce Overbalance

If the total shear resistance can be reduced below the force that be delivered to the stuck point, the

pipe may be pulled free. As overbalance is reduced, the effective stress in the contact area declines

and the shear strength declines also. But reducing the MW is not linear so there is no value in

reducing the MW unless it can be reduced to the critical level required.

Overbalance is required to control pore pressure and borehole stability and by drilling with a lower

MW the risk may be shifted from a problem to another. As a result, it is more practical to reduce

contact area than to drill with a lighter MW since contact area is easier to manipulate with a little

shifting of risk. In a typical high-angle well in low-strength formation, 2 to 3 lbm/gal of overbalance is

required to maintain stability and just to eliminate the effect of the unsupported DCs in many

assemblies it might come to reducing MW by more than 2 lbm/gal.

So as a result when the pipe gets stuck, very little reduction in the MW may be required to reduce

the pullout force to below the critical level and the greater the contact area, the more MW

reduction will be but without exceeding the limits of reduction.

3.2.4 Modeling

To better understand the influence of different variables on DPS and then to become able to predict

the sticking force for a particular design, a numerical model was developed. This dynamic FE model

incorporates coupled deformation and fluid-flow elastoplastic filter-cake behavior, BHA and

Ahmed Msahli 21

Pipe Sticking 2016

drillstring bending, torque and drag, and models pressure-drop and shear strength development of

the filter cake. Then, applying incremental axial displacements at the pipe wall, the force per unit

length of embedded pipe required to free the drillstring is calculated.

The BHA and drillstring-bending model is used to estimate potential contact area, and the torque-

drag model in used to estimate available downhole pulling force at the stuck point. Then the results

of this analysis are presented as probability of success as a function of MW.

Below in the figure, and example from stochastic model is showing the improvement of success

probability in centralizers are moved from adjacent to a midpoint position related to the casing joint

from 75% to almost a 100%.

The apparatus shown the figure below, define the mechanical properties and time-dependent

behaviors of various fluids for use in the FE model for small-scale DPS and pullout tests.

Ahmed Msahli 22

Pipe Sticking 2016

The test apparatus consists of a chamber that accommodates a 4in diameter cylindrical core with a

2in hole made from sandstone or ceramic of known permeability.

Once a steady-state filter cake has been established, the aluminum rod situated within the simulated

wellbore is embedded into the filter cake and the DPS force is allowed to develop and then recorded

after the desired stationary time. Pressure transducers on the rod allow the pressure drop in the

filter cake between the rod and the wellbore wall to be mentioned and recorded so that the

relationship between effective stress and strength development can be characterized as an input to

the FE model. At the end of the test, the filter-cake thickness and the location of the shear failure

within the cake are determined.

A comparison of FE-model-predicted pullout force vs. actual tests performed using 13.0-lbm/gal

water-based and NAF mud is shown the figure below. In these tests, the geometry of the test

apparatus was modeled to be the same as in real application and the dimensions and boundary

conditions for the actual wellbore are used.

Ahmed Msahli 23

Pipe Sticking 2016

Figure 12: Pullout Force between WBM and NAF models (3)

The results of triaxial tests on filter-cake material that were performed to determine the mechanical

properties of cakes built from various fluids for input into the FE model are shown below in the

figure. This figure illustrates, through the percentage of the axial and radial strain to the differential

axial stress, the elastoplastic behavior of the filter cake that must be modeled.

Figure 13: Differential Axial stress vs. Axial & Radial strain% (3)

Ahmed Msahli 24

Pipe Sticking 2016

3.2.5 Results

In the period from 2004 to 2008, according to ExxonMobil, after the training on these stuck-pipe-

avoidance practices started in 2003, the operators drilled 3.476 wells. The wells were drilled by 20

geographically independent drilling teams. The statics in the table below were recorded.

Stuck but Freed Stuck not Freed Stuck but Freed Stuck not

Freed

Number of events 0 3 14 3

There were only three stuck-pipe events in wells with BHA designs that complied with the practices

and 17 incidents with non-compliant BHAs that only three of which were not freed.

Practices such as D&S are critical to success but are only applied when appropriate for the specific

situation while recommendations for the allowed length of unsupported DCs apply to all wells

because it is believed to be the dominant risk factor.

For the non-compliant BHAs, they tended to have only one to two unsupported collars beyond the

recommendations which led to having a limited contact area that is believed to contribute in the

events where the pipe was freed. As for the case where the pipe was not freed, they were all in NAF

and the stuck point was determined in the HWDP which highly suggest that wear groove may have

been a significant factor.

Over the same period of time, the histogram in the figure below shows the variation in stuck pipe

events.

Ahmed Msahli 25

Pipe Sticking 2016

The low incident rate in the year 2008 is believed to be reflecting the increased operational

awareness of the risks associated with even small increases in the length of unsupported DCs.

Ahmed Msahli 26

Pipe Sticking 2016

Beside the Physical reasons mentioned previously as the direct cause of different types of pipe

sticking, studies Identified various factors that are connected to the occurrence of pipe sticking.

When switching crews usually one drilling supervisor in ready for days off while the other is just give

few minutes’ worth of information by the supervisor being relieved.

A strategy, that have shown efficiency and been used by Texaco and then considered by other

companies such as BP, consist of changing the pusher a couple of days after the rest of the crew.

Some Individuals play a key role and must be released in different times in order to assure the

continuity of information between all, such as the lead pusher, drilling foreman, mud hand and the

cementer.

In a study conducted by Texaco USA, 91% of the 105 well that had pipe sticking incident were

directional wells and freeing the pipes was a 66% for straight holes facing 45 % free of the directional

incidents.

So basically, straight holes contain less risk and are more promising in freeing operations.

Ahmed Msahli 27

Pipe Sticking 2016

Pipes with different diameters are specific for certain using conditions, which will make certain pipes

venerable to more sticking than others.

In the case of the drillings in northern Italy done by AGIP Italy, since the used 8 ½’’ pipes for depleted

reservoirs or reservoirs with high pressure gradient 59% of the sticking cases where in this range of

size holes while 27% where for 12 ¼’’.The following figure shows

Ahmed Msahli 28

Pipe Sticking 2016

This case study consists of a real-time wellbore friction analysis to detect onset of drillstring sticking

during extended reach well drilling. As mentioned before for reducing the still pipe time, it is

necessary to understand the behavior of the well while drilling. Wellbore friction modeling is an

important assessment to aid real-time drilling analysis and predicting drilling troubles such as onset

of pipe sticking.

In extended reach drilling, surface measurement of weight on the bit and torque differ from

downhole measurement due to the friction between the drill string and the wellbore. This friction

force can be used for overall friction coefficient measurement which is a well used indicator during

different drilling operations.

This study provided an opportunity to examine the validity of a newly-developed analytical model to

estimate friction coefficient versus measured depth in a deviated well drilled in western Canada

where the drillstring got stuck during drilling operations.

For this analysis, an element of the drill string in the wellbore is considered and filled with drilling

fluids. The forces acting on the pipe are buoyed weight, axial tension, friction force, FN,

perpendicular to the contact surface of the wellbore.

The friction force is defined as an acting force against the pipe movement which is equal to the

friction coefficient multiplied by the normal force as shown in the equation (1) , for that calculating

the normal force FN is the first step.

(1)

In straight inclined and horizontal sections the normal force in equal only to the weight of the

element while for curved section as build-up, drop-off, side bends or their combinations, the normal

force depends more on the tension at the bottom end of pipe element than on its weight [ Aadnoy,

B.S. and Anderson, K. 2001 ], The following general equation (2) defines the tension at the top of

each element .

(2)

The plus and minus sign the equation (2) are for the pipe movement either up or down.

The weight and friction force of each element should be calculated and added up from bottom to

the surface. The weight on bit can affect the value of the friction force in the curved section therefor

should be considered during analysis [ Fazaelizadeh et. Al. 2010].

For a straight inclined element, the normal force is weight dominated and is not dependant on axial

Ahmed Msahli 29

Pipe Sticking 2016

tension at the bottom of the element. The coulomb friction model can be used for friction

calculation as shown in the equation (3).

(3)

Aadnoy et. al. (2010) did the following derivation for a curved section. They assumed that the pipe is

weightless when the friction force was computed, but added the weight at the end of the bend.

They also used the concept of dogleg angle θ in the derivation which depends on both the wellbore

inclination and azimuth. Because the pipe will contact either the high side or the low side of the

wellbore, and the contact surface is given by the dogleg plane.

So, the dogleg is the absolute change of direction and can be determined by equation (4).

(4)

For build-up, drop-off, side bends or their combination, the axial force becomes:

(5)

With the shown calculations, Overall friction for any wellbore shape can be computed by dividing

the well into straight and curved elements. The forces are then summed up starting from the bottom

of the well to the surface.

In buoyancy calculations, the principle of Archimedes law is used. The principle states that the

buoyancy force, when a body is submerged into a fluid, is equal to the weight of the displaced fluid.

The drillstring tension in a wellbore filled with drilling fluid is the unit weight of pipe w multiplied by

the buoyancy factor β. For both vertical and deviated borehole, the equation (6) is valid if the inside

and outside if the pipe are submerged into the same fluid.

(6)

The equation (6) gives an estimation of the buoyancy factor when the fluids are incompressible and

temperature effect and the cutting concentration are ignored. Basically this equation can be used for

all overbalanced drilling operations unless tripping.

If there is a density difference between the inside of the string and the annulus like in cementation

and tripping operations, the following equation (7) can be used.

(7)

The equation (7) can calculate the buoyancy locally for each element. In tripping in operations, the

annulus is completely full while drilling fluid level varies inside drillstring. The reason is that the

Ahmed Msahli 30

Pipe Sticking 2016

After considering the collapse pressure of the drillstring and risk of fluid loss and flow, the drillstring

will be partially filled for a period of time and then will be completely full after filling the inside every

few hundred of meters of pipe ran by the drill crew.

When the drillstring is full of mud the equation (7) will turn into the equation (6). Else, the equation

(7) will be used to calculate the buoyancy factor calculation of each element.

This approach was integrated into the tripping in friction analysis in the case study presented.

The correction factor Cs represents an effect of the contact surface between the pipe and the

wellbore due to larger curvature surface contact. To include the contact surface effect in the friction

equations it should be multiplied by friction coefficient as shown in the equation (8) as there is

contact.

(8)

The correction factor varies between 1 and 4/ , depending on the contact surface angle γ which

varies itself between 0° and 90°.

(9)

As discussed by Maidla and Wojtanowicz (1987), because of wellbore swelling, thick mud cake or

cutting accumulation etc. there is a possible reduction in hole size in some portions of the well,

which would show higher friction value when BHA passes through. The increase depends on the

degree of severity of those tight spots. In such cases, it could be required to do some remedial action

such as drillstring rotation to ream the tight area.

The well drilled in western Canada to illustrate the friction analysis has the following geometry in the

figure.

The Kick off point is at 800m and the heavier build section initiated at 2287m. The well reached 30°

of inclination at approximately 2700M and drilled with the same degree to 3231m of total depth.

At that depth the drillstring was tripped out for the purpose of replacing the measuring while drilling

(MWD) tool. Then the drilling process was continued after the tripping in procedure for only 20m.

And even that the drillstring configuration was not changed from 2200m to 3251m, the drillstring

got stuck at the depth of 3251m.

Ahmed Msahli 31

Pipe Sticking 2016

The BHA had a length of 290m from the bit. The BHA has a large effect on the value of friction force

during tripping in and out. As shown in the figure below, in smaller diameter wellbore intervals,

applies a higher friction force against the BHA movement.

Ahmed Msahli 32

Pipe Sticking 2016

An important factor that did change and thus must be considered is the buoyancy factor. In the

figure below, the buoyancy factors of three operational modes are shown.

Figure 19: Buoyancy factor vs. measured depth for drilling, tripping in and tripping out (5)

During the drilling process, the buoyancy factor had a constant value until the depth of 2485m. At

that point, mud weight was increased as much as 300kg/m3 to prevent gas inflow which led to a

decline in the buoyancy factor.

In the tripping out process, the mud weight was kept constant and the mud level in the annulus

dropped a little. That is shown with a constant buoyancy factor in the figure.

During the tripping in process, the buoyancy factor shows a gradual decrease due to increasing

length of the non-filled pipe. And some jumps in buoyancy are shown in the figure since the drill

string is filled periodically, an increase in the level of the drilling fluid inside the drillstring leads to an

increase in the buoyancy factor.

The increase in mud weight to prevent gas influx resulted in an increase in downhole pressure,

shown in the figure below.

The increase in mud weight enhances the differential pressure between the drill spots which

temporarily and instantaneously stick to the formation and applies additional force on the pipe.

Ahmed Msahli 33

Pipe Sticking 2016

This differential pressure has an effect of larger magnitude than the decrease in the normal force on

the drillstring due to change in buoyancy factor, which increases the frictional force for the spots in

question.

In any pipe sticking problem, the first remedial action in reducing the mud weight in order to

decrease the differential pressure between drillstring and the formation. This reduction may help

releasing the drillstring from the stuck point in most cases.

The difference in hook load recorded between while tripping out and tripping in the well, were used

to the friction analysis for this case and illustrated in the figure below.

During tripping in, there are periodical shifts in hook load values due to increased buoyancy factor as

the fluid fills for the case.

During tripping out and since the buoyancy factor is almost constant due to filling annulus throught

the fill-up lines, the previous jumps are not observed.

A deviation from the hook load trend may be seen as the build-up section begins at the depth of

2287m due to the friction force increase in the build-up section.

Ahmed Msahli 34

Pipe Sticking 2016

Figure 21: Hook load vs. measured depth for tripping in and tripping out (5)

The friction force is calculated from the difference between the static weight of drillstring and the

hookload value. The Frictional force during the tripping out process is presented in the following

figure.

Figure 22: Frictional force during the tripping out process (5)

Ahmed Msahli 35

Pipe Sticking 2016

The friction in the curved section is dominated by tension which shows rapid increase in the friction

force value. As the top of the BHA reached the buildup section, around 2990m, the frictional force

increased and then declined as the BHA passes through the section.

The same effect is seen in the figure below for the overall friction coefficient. The overall pressure by

definition is a single friction coefficient for the entire wellbore.

In the figure the change in friction coefficient is the resulting effect of contact surface between the

borehole and the BHA. The effect disappears gradually as the BHA passes through the buildup

section.

Friction coefficient

Figure 23: Overall friction coefficient vs. measured depth during tripping out operation (5)

The reason that the friction coefficient should be calculated and not only the friction force, is that in

some cases the friction force increases due to change in the well geometry. And with a wellbore

friction model, distinguishing the effect of other factors from the friction coefficient increments is

possible.

For tripping in, the estimated friction force shown in the figure (1) and the overall friction shown in

the figure (2) both show a significant increase as the tight hole occurs during the process.

Ahmed Msahli 36

Pipe Sticking 2016

Figure 25: Overall friction coefficient vs. measured depth during tripping in operation (5)

Ahmed Msahli 37

Pipe Sticking 2016

While figures (7) and (6) do not convey much information on a tight hole, one could possibly detect

the onset of pipe stuck in a well using the two parameters from figures (8) and (9).

The calculated overall friction coefficient during the tripping in process may be applied as an

efficient means to detect a possibly tight hole, as a rapid increase in friction force was seen while the

well geometry had a straight holding angle.

This phenomenon was not observed during the tripping out process because the collapse and tight

hole occurred as a result of lower ECD due to swabbing when tripping out and time effects on the

formation.

Ahmed Msahli 38

Pipe Sticking 2016

Differential pipe sticking (DPS) is proven to be the most costing and traditional problem in drilling

operations, therefore early identification of the cause is crucial since every cause will call for

different measures.

Many statistical analyses have been performed in order to identify parameters acting in DPS in

order to minimize or prevent the sticking, until an application of Neural Network Methodology has

been published by Halliburton.

Artificial Neural Networks (ANN) has many advantages compared to conventional statistical methods

such as using no predictions, tolerance to errors, data-driven nature and faster computation.

The main idea of ANNs is based on the biological neural system; similar to the human minds as a

large number of highly interconnected processing elements (neurons) working in union to solve a

specific problem.

So just as the human mind, in ANNs things are learned with the use of examples. To understand and

find out relationships, they deal with input and output parameters and once successfully trained, the

network can be used to process the output according to input that are similar but not necessarily the

same as those used in the training.

The interconnections of the network have their own specific weight that influences

transmission signal. These specific weights determine the relationship between input and

target neurons.

Each neuron calculates it is total weighted input. to diversify the various processing

elements, a bias is added to the sum of weighted inputs called “net input“ and the after

passing the transfer function output is generated.

In the training process, weights and biases are adjusted on the basis of learning rules and

competing training; the fixed weights and biases act as the memory of the network

Ahmed Msahli 39

Pipe Sticking 2016

The ANN performance is very dependent on the reliability and the precision of the database, which

the collection is considered very time-consuming and challenging.

To have the DPS as an output , many parameters should be provided for input such as parameters

related to drilling and fluid-mud properties ; differential pressure , hole depth , hole size , BHA length

, plastic viscosity , yield point , initial gel strength and 10 minute get strength . In other words, two

groups of parameters needs to be identified as DPS and non-DPS parameters.

The independency of these parameters is what makes it possible to predict the occurrence

conditions leading the a result between zero and one where zero indicates the non-DPS cases and

one the DPS cases while the margin in between is for the level of risk.

In the published study, Data was collected from 63 wells drilled in the Persian Gulf as following, 32

wells that experienced DPS and 31 that didn’t. Most of the wells were either sidetracked or

horizontal and the occurring of DPS was in the reservoir layers with the use of oil based or synthetic

drilling muds.

Ahmed Msahli 40

Pipe Sticking 2016

In ANNs, to form the structure of the network the input and output must be somehow connected

and according to the connection method ANNs are divided in two major categories:

The representative graph does not contain any loop which makes it static in a way that a given input

can only produce on set of outputs and thus have no memory.

Two popular FFNs are Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF). For the

composition of the MLP, it consists of several layers. The input layer that receives inputs and

basically has no other function but to receive the input signals. The output layer provides the

response of the network to the input. And layers in between, called hidden layers and has no contact

with the external environment. The hidden layers and the output layer contain neurons that execute

the activation function and the transfer functions. The most commonly used transfer functions are

linear, linear sigmoid and hyperbolic tangent.

As for the RBF neural networks, they have a least three layers, the input layer, the output layer and a

hidden layer containing a high number of neurons performing a nonlinear transformation of the

inputs by means of radial basis functions. A radial basis function is defined as a multi-dimensional

function of the inputs and a previously defined center. One very popular RBF is the Gaussian

function which can approximate any continuous nonlinear function.

ANNs like MLP and RBF are trained using an algorithm called Back Propagation of error. Using

supervised learning, the algorithm is provided with the inputs and the outputs that it is meant to

learn and then an error is calculated. The error is the difference between actual and expected results

and the idea behind the back propagation algorithm is to reduce this error by changing weights, until

the ANN learns the training data. And initially the training starts with random weights.

The error of the network (E) is defined as follow for the output neurons ( Nj ).

∑ ( ) (1)

(2)

Ahmed Msahli 41

Pipe Sticking 2016

The representative graph contains loops in an architecture that allows memorizing the information

in the networks.

The data used is divided into training, cross validation and testing subsets respectively in this study

with the percentages of 85%, 5% and 10%. These are prepared using the tag options available in the

Neuro Solutions Software with a value of 5000 epochs that been factored into the training database.

An epoch is one sweep through all the records in the training set. Then the software starts with

random values for the weights.

As the network learns, the error will drop towards zero. But lower error does not mean a better

network. There is a possibility of over training the network. In this case, instead of a generalized

result the network memorizes the training patters and would not be able to produce reasonable

outputs for inputs not encountered in the training.

To know when to stop the training, Cross validation is one of the most powerful methods used.

When the error in cross validation dataset increases, the training then should be stopped as the best

point of generalization is considered reached.

To test the network performance it is best to apply data that has not been seen yet. It goes by

freezing the weights after the training phase and applying the data. If the training is successful and

the network’s topology is correct, it will apply the past experience to this data and still produce a

good solution. Then the network is considered able to generalize.

Preprocessing is necessary for ensuring the convergence of neural nets. In this stage and in order to

make the process and analysis more effectively with a high degree of precision, noisy incoming data

should be removed. Even this training would cost a lot of time, the result is much better by removing

the noisy datasets.

Ahmed Msahli 42

Pipe Sticking 2016

Insufficient amount of data either for training or for testing can lead to a poor generalization. After

processing, attempts were made to vary the number of hidden layers and the number of neurons

included. The results showed that for this particular training datasets, increasing the number of

hidden layer could not increase the performance of the network while increasing the number of

hidden layer neurons could decrease the error and lead to obtain a better performance.

Finally, a network with one hidden layer with 40 neurons assigned the best weights to the input data

parameters for both MLP and RBF networks. Training curves of the best generalized MLP and RBF

networks released after several runs are shown below.

Although the training error function for MLP network in closer to zero than the training error

function for RBF network, the cross validation error for RBF network is less and this result a network

with better performance in prediction of differential sticking incidents.

Figure 28: MSE versus a default number of epochs for MLP model (7)

Ahmed Msahli 43

Pipe Sticking 2016

Figure 29: MSE versus a default number of epochs for RBF model (7)

To access the models estimation performance, two quantitative measures for estimation accuracy

and examined for training datasets, the minimum training error (MSE) and the training error at last

epoch (Final MSE).

In the two tables below, the minimum MSE and Final MSE errors for the training and cross validation

are shown.

Epoch # 5000 3401

Minimum MSE 0.010645014 0.002225183

Final MSE 0.010645014 0.005474218

Table 4: Error of training and cross validation for MLP (7)

Epoch # 6798 4264

Minimum MSE 0.011482671 0.002882395

Final MSE 0.011482671 0.004191717

Table 5: Error of training and cross validation for RBF (7)

It is seen that after 3401 epochs for MLP network and 4264 epochs for RBF network, the networks

are validated and able to approximate the error around 1%.

Ahmed Msahli 44

Pipe Sticking 2016

All quantitative accuracy measurements values for proposed networks are listed in the tables

following.

Desired Desired

Stuck 5 1 Stuck 5 0

Non Stuck 1 7 Non Stuck 1 8

MSE 0.098629911 0.11497561 MSE 0.076746819 0.08127283

NMSE 0.402738803 0.469483739 NMSE 0.313382846 0.331864056

MAE 0.192982043 0.224900055 MAE 0.170898439 0.185877963

Min Abs Error 0.013326439 0.024498626 Min Abs Error 0.002866021 0.009825315

Max Abs Error 0.870987511 0.770815895 Max Abs Error 0.714543252 0.698180125

r 0.797403844 0.762350715 r 0.838218624 0.84052662

Percent Correct 83.33333333 87.5 Percent Correct 83.33333333 100

Table 6: Accuracy measurements (7)

Shown in these tables, both proposed networks could predict correctly five out of six stuck data sets

among testing datasets. And from eight non-stuck sets the RBF network could predict correctly all

while the MLP network predicted seven sets only.

This, results for RBF model are more accurate than the MLP model.

The mean square error (MSE) may represent how well the network output fits the desired output,

but it does not necessarily reflect weather there will be sticking event or not.

In order to predict DPS, the stability of ANN is tested by preparing a production dataset with the

actual data from horizontal well F8-0H in Forouzan field. In this well the pipe was stuck at the depth

of 3126m (2826m TVD) and had oil based mud used in the reservoir section.

The data of this well been recorded for three days before beside the sticking day, and it is shown in

the table below.

(psi) (m) (cp) (cc/min) (V%) (cp) (lb/100ft²) (ln/100ft²) (lb/100ft²) Index

505 2826 55 4 15 20 15 6 12 0

505 2826 56 4 15 20 16 6 12 0

505 2826 63 4 16 15 23 8 12 0

505 2826 75 4 16 18 34 8 12 1

The results of the prediction are shown in the next table. Both proposed networks were able to

predict correctly the sticking. The MLP network did predict the probability of sticking with 100%

confidence while RBF network predicted it with 70% confidence which is more acceptable and

Ahmed Msahli 45

Pipe Sticking 2016

reflects the proper training of RBF model and having a low MSE.

And so on, selecting a suitable structure and a proper training of ANNs with enough datasets will

lead to more accurate prediction.

Desired Output Predicted Output Desired Output Predicted Output

0 1 0 0.69

0 1 0 0.69

0 1 0 0.76

1 1 1 1

This method comes after the neural network was trained and is very important. It is a method for

extracting the cause and effect relationship between the inputs and the outputs of the network.

Thus, the effect of every input becomes known.

The network learning has to be disabled during this operation so none of the weights is changed.

The sensitivity for both models, MLP and RBF, is shown the following histograms for mean values of

this study.

Ahmed Msahli 46

Pipe Sticking 2016

Ahmed Msahli 47

Pipe Sticking 2016

Even with the ANN being a very effective method of prediction and compared to other techniques

like Fuzzy Logic and genetic algorithms, it still has many shortcomings and restrictions. This chapter

studies the notion and the advantages of Support Vector Machine over the other methods.

Support Vector Machine (SVM), developed by Vapnik (1999), can enhance the efficiency of the

prediction performance while pursuing a global optimization solution and becoming a promising

alternative. The SVM is a combination of three ideas, the solution technique from optimal

hyperplanes that allows the expansion of the solution vector on support vectors, the convolution of

the dot-product to extend the solution surfaces from linear to non-linear and the notion of soft

margins to allow errors on the training set.

SVM, else than not keeping the number of features small in order to control complexity, has many

other advantages over the ANNs such as designing classifier with maximum generalization, reaching

the general optimization of the cost function, automatic determination of structure and optimized

topology for the classifier and modeling the non-linear classification functions and the dot-product

concept of Hilbert spaces (symmetric, linear and positive).

The example is composed of the vector {p1, p2 , .., pi,.., pq}, each p is a “d” dimensional vector that

has the label ti and ti { -1 , +1 }.

These two classes are separated by the use of the classification function and a hyperplane H.

(2)

Weight vector W is a perpendicular vector on the separator plane, and b is the bias value.

The data are linearly separated in a way that the boundary data of the first class are on the

hyperplane H+ and the boundary data of the second class are on the hyperplane H-.

The training data on hyperplanes H+ or H- are called support vectors. And the area in between the

two hyperplanes is called a margin. The Hyperplanes are separated by the distance of ‖ ‖

.

Designing the hyperplanes with the most margin width, which is optimal, is achieved if the pattern’s

classification is proper and the margin width is maximized, which means ‖ ‖ minimized.

Ahmed Msahli 48

Pipe Sticking 2016

Figure 32: Maximum-margin hyperplanes for SVM trained with examples from two classes (6)

The goal is to classify the training patters correctly so they lay on or out of the margin, and this is

translated in the equation;

Then the optimal design of a hyperplane classifier with margin would be as followed.

MinW ( ‖ ‖² ) (6)

W = ( W1 , W2, .. , Wd ) T ( 7)

To find solutions for this equation is found using the Lagrange function, with (αi) being the Lagrange

multiples. The Lagrange equation is expressed as following;

L (W, b, α) = W. W - ∑ (8)

In solution point, the derivation of L from W, B and α is equal to zero. This will result in the following

system.

∑

{ (9)

∑

Using the equations (8) and (9), the problem results in;

∑ ∑ ∑

{ ( 10)

∑

Ahmed Msahli 49

Pipe Sticking 2016

After solving the dual problem, the Lagrange αi will be obtained. And each pi that corresponds

with αi is called a “support vector” (SVi).

The weight vector and b are obtained by the use of the following equations;

∑

{ ∑ (11)

∑

Then the optimal plane that separates binary decision categories is as follows;

(∑ ) (12)

is the classifier , pi is the training example and ti is the class of training example

The vector p demonstrates the input data and the vector SVi is the support vector. The value of b is

the bias value of the separator hyperplane and αi is a Lagrange multiplier.

The Previous algorithm gives the margin of two completely linear separate classes. In case the

classes intersect, separating the classes by the use of the linear-decision boundary would always

result in errors. But this can be solved by mapping the data from the input space Rd into high-

dimensional feature space Rm by the use of nonlinear mapping. In the new space the classes have

less intersection with each other, as shown in the following figure.

Figure 33: Mapping input space into high-dimensional feature space (6)

Then in the new space the optimal-decision boundary is calculated by the use of equation (12) and

by substituting pi with φ (pi). φ (pi) denotes the high-dimensional feature space.

The problem now results in finding a solution to the following optimization problem.

Ahmed Msahli 50

Pipe Sticking 2016

∑ ∑ ∑

{ (13)

∑

toward finding an optimal boundary for classes with the most conflicts.

In the equation (13), instead of using φ, a kernel function is used, defined as follows.

After determining a suitable K ( pi , pj ) , the function is replaced in the equation (13) by φ(pi ) . φ(pj )

and the optimization problem is solved.

Eventually the ultimate classifier is obtained by the use of the following equation.

(∑ ) (15)

Two well-known Kernel functions that are mostly used for SVM are the Polynomial degree and the

Gaussian radial basis function.

The statistical measures of the performance of a binary classification test are Accuracy, sensitivity

and specificity according to Fawcett (2006).

In order to understand the meaning of these statistical measures, the explanation of the following

definitions is needed, true positive (TP), false positive (FP), true negative (TN) and false negative

(FN).

In a scenario where some wells are tested for DPS, the test outcome can be either positive where

DPS occurs or negative where there’s no sticking.

FP: Non-DPS sections of the wells incorrectly identified as DPS.

TN: Non-DPS sections of the wells correctly identified as non-DPS.

FN: DPS sections of the wells incorrectly identified as non-DPS.

Ahmed Msahli 51

Pipe Sticking 2016

Specificity of 100% means that the test recognizes all the actual negatives, so the specificity alone

does not explain how well the test recognizes positive cases (TP + FN). Sensitivity of 100% means

that the test recognizes all the actual positives (TP + FN). A high sensitivity is used to rule out the

DPS. The accuracy is the degree of measurement closeness of a quantity to its actual true value

(Taylor 1999).

So the same as ANN, the designed SVM approach follows the priorities accuracy, sensitivity and

specificity in analyzing the performance.

For this study, the same data used in the ANN (FFBP) case study in the previous chapter has been

used in order to compare the results.

The 12 parameters used as input for this study are in following table.

Differential pressure ( psi) 100 1178

Hole depth (ft) 2624 11713

Mud-filtrate viscosity (cp) 32 180

Fluid loss (cm3/min) 0.5 9

Solid content (vol%) 1.6 24

Plastic viscosity (cp) 9 65

Yield point (lbf/100ft²) 6 51

Initial get strength (lbf/100ft²) 1 23

10-minute get strength (lb/100ft²) 2 51

BHA length (ft) 630 1170

Still-pipe time (seconds) 0 300

Hole size (in) ( fixed sizes) 6.5 17.5

Table 9: Well parameters used for the study (6)

To define the structure, two most popular transfer functions tansig and logsig were tested for

hidden layer. The optimal number of neurons in a single hidden layer is shown in the next two

tables, by the use of the trial-and-error method for FFBP.

It is shown through the network-performance criteria; accuracy 82.81%, sensitivity 84.38% and

specificity 81.25%, that 18 neurons in a hidden layer with the tansig transfer function is the most

effective case in the developed FFBP.

Ahmed Msahli 52

Pipe Sticking 2016

As for the SVM, Gaussian and polynomial kernel functions have been evaluated to investigate the

performance. These analyses are listed in the following tables.

The results of these analyses show the values of 92.19% accuracy, 93.75% sensitivity and 90.63%

specificity for the SVM using Gaussian kernel function with C=1 and σ=0.5.

The results of the Gaussian function are reasonable and there is no need to try a different kernel

function.

Ahmed Msahli 53

Pipe Sticking 2016

Ahmed Msahli 54

Pipe Sticking 2016

Finally, comparing the performance criteria for all the developed FFBP and the SVM implies that

SVM with Gaussian kernel function yields better efficiency in the prediction of DPS as shown in the

table below.

Furthermore, to show the difference in the efficiency between the best developed SVM and ANN

models, the prediction results compared to the real data from field are show in the following table

with the figure to show the matches.

Table 14: difference in the efficiency between the best developed SVM and ANN models (6)

Ahmed Msahli 55

Pipe Sticking 2016

Figure 34: Comparison of SVM results , FFBP results and real data from field (6)

And finally, to more prove that SVM predictions have better agreement with actual states than

ANNs, few examples of real cases have been examined by both methods are present in the following

table.

Table 15: Comparison of SVM and FFBP predictions with real cases (6)

Ahmed Msahli 56

Pipe Sticking 2016

Conclusions

Sticking potential does not exist in all wells, but still exists in the great majority. Stuck-pipe-

avoidance practices have been developed and implemented in well that do have filtration control

and solids and the objective was not eliminate differential sticking, which is impossible, but in the

objective of maintaining conditions that ensure delivering the pullout force required to the stuck

point, which statistical experience have proven to be achievable.

3. Minimize overbalance while insuring borehole stability.

4. Use HWDP in compression for bit weight in vertical and low-angle wells.

5. Use conventional drillpipe in compression in intermediate and high-angle wells (in the limits

of buckling).

6. Use standoff subs on drilling jars above the stabilized BHA.

7. Conduct progressive pipe-sticking tests before connections in high risk wells.

8. Do not use API FL or API HT/HP test as indicator of cake quality except for low-permeability

formations.

9. Conduct API PPTs and use appropriate blocking solids to improve cake quality.

10. Conduct D&S treatments to enhance cake quality in intervals of high differential pressure or

chronic cake growth.

11. Model the differential-sticking risk quantitatively when planning operations that lie outside

of previous experience.

12. Consider the sticking risk associated with a wear groove in high-angle wells when planning

mitigations.

The differential pressure affects the friction force and that result in more forces on the drillstring

spots which instantly stick to the wellbore. This can be seen from higher values of the real-time

calculation of the friction coefficient.

In friction analysis, the buoyancy effect should be treated precisely. In case not and particularly in

tripping in, the results of overall friction coefficient estimation could be wrong.

The use of friction model instead of friction force can allow the detection of the geometry effect on

the friction force value.

While tripping in, an increase in the real-time friction coefficient indicates tight hole that happened

during tripping operations.

The collapse and tight hole in this case study might have been initiated after the drill string was

pulled out due to swabbing and time effects of the collapsed formation

Ahmed Msahli 57

Pipe Sticking 2016

Neural network has proven to be a very efficient and powerful tool that can provide better and more

accurate solutions for the problems associated with differential pipe sticking events. Prediction of

pipe sticking is possible using two different types of neural networks. Both types have tolerance to

noisy data, but RBF model is more accurate than MLP model. The performance of the network

depends on the database and parameters used for the analysis. The number of hidden layers is

critical to the network. Sensitivity analysis is used to determine the role played by each parameter.

In the study ANN and SVM methodologies to predict DPS have been compared. Both methods can

be of a great use. The performance criteria for both the FFBP and the SVM are very promising, but

the SVM yields better results in the prediction of DPS and not only in well planning but also in a real-

time drilling operation.

The existence or loss of training samples that are not support vectors does not influence the

ultimate classification. Thus, putting aside the training samples with low probability of being support

vectors leads to reducing the classifying time.

Because the SVMs are support-vector-dependent, the ultimate classification will have a better result

than conventional ANNs even if few data are available. Some parameters such as mud properties

can be optimized by SVM in a more efficient and economical way.

The use of artificial intelligence has proven to be beneficial in the understanding of the pipe sticking

but it can be further developed and applied in other aspects of drilling. With the use of lab testing

and simulating machines, research can be focused on more critical factors in the drilling process.

This development in the ANN and SVM can put to question the statement made in beginning that

the sticking can only be mitigated and not eliminated.

Ahmed Msahli 58

Pipe Sticking 2016

Bibliography

1. Oil Company Perspectives. Offshore Technology Conference. Houston, Texas : s.n., May 7, 1991.

2. Azar, J.J. Drilling Problems and Solutions. [book auth.] Robert F.Michell. Volume 2 Drilling

Engineering ; Pertoleum Engineering Handbook. Richardson, Texas : SPE, 2006.

3. Design Methodology and Operational Practices Eliminate Differential Sticking. Fred E. Dupriest,

William C. Elks Jr, and Steinar Ottesen, SPE, Exxo,Mobil Development Company. 2011, SPE 128129.

5. Real-time Wellbore Friction Analysis to Detect Onsel of Drillstring Sticking during Extended Reach

Well Drilling: Case Study. M. Fazaelizadeh, SPE, University of Calgary, G. Hareland, SPE, University

of Calgary and Z. Wu, SPE, University of Calgary and M. Tahmeen, SPE, University of Calgary. 2011,

SPE 143157.

6. Intelligent Prediction of Differential Pipe Sticking by Support Vector Machine Compared With

Conventional Artificial Neural Networks: An Example of Iranian Offshore Oil Fields. Reza

Jahanbakhshi and Reza Keshavazy, Young Researchers and Elites Club, Science and Research

Branch, Islamic Azad University, Tehran, Iran and Mahdi Aliyari Shoorehdeli, K.N. Toosi University

of Technology and Abolqasem Emamzadeh,SPE,Islamic Azad University. 2012, SPE 163062.

7. Development of Artificial Neural Networks To Predict Differential Pipe Sticking in Iranian Offshore

Oil Fields. R. Miri, Iranian Offshore Oil company, et al., et al. 2007, SPE 108500.

8. Stuck Pipe Prediction and Avoidance: A Convolutional Neural Network Approach. Siruvuri, C.,

Halliburton Digital and Consulting Solutions, Nagarakanti, S., Samuel, R. LADC/SPE Drilling

Conference , Miami, Florida : SPE, 2006. 98378.

9. A Field Case Study of Differencial Pressure Sticking. Adam, N. SPE Annual Technical Conference

and Exhibition, Denver : SPE, 1977. 6716.

10. Advances in Prediction of Stuck Pipe Using Multivariate Statistical Analysis. Biegler, M.W. and

Kuhn, G.R. SPE/LADC conference, Dallas. TX : SPE, 1994.

11. Quantifying Stuck Pipe Risk in Gulf of Mexico Oil and Gas Drilling. A.P. Wisnie, Conoco Inc, and

Zheiwei Zhu. SPE, s.l. : University Of Southwestern Louisiana. 28298.

12. A Case Study of Stuck Drillpipe Problems and Development of Statistican Models to Predict the

Probability of Getting Stuck and If Stuck, the Probablity of Getting Free. Sharif, Q.J. Lahore, Pakistan :

University of Engineering & Tech, 1997.

13. Analysis of Stuck Pipe in Deviated Boreholes. Aadnoy, B.S, Larsen K, and Berg P.C. SPE annual

Ahmed Msahli 59

Pipe Sticking 2016

14. Pressure Differential Sticking of Drill Pipe and How It Can Be Avoided or Relieved. Helmick, W, R.

and Longley, A.J. s.l. : Oil and Gas Jour, 1957. 55,132.

15. Industrial Application of Neutral Networks. Mason, I.F. Croall and J.P. s.l. : Research Report

Esprit.

16. Neural Networks for System Identification. Reynold Chu, S., Shoureshi, R,. and Tenorio, M. 10,

No. 3, pp 31-35, s.l. : IEEE Control Systems Magazine, 1990.

17. Neural Networks for intelligent multimedia processing. J-N, Kung S-Y and Hwang. s.l. : IEEE 86,

1998.

18. Drilling Fluids Optimization. James L. Lummus, JJ. Azar. Tusla, Oklahoma : Penn Well Publishing

Company, 1986.

19. Differential Pressure Sticking Laboratory Studies of Friction Between Steel and Mus Filter Cake.

Annis, M.R. and Monaghan, P.H. J Pet Technol, s.l. : SPE. 151-PA.

20. Differential Sticking Laboratory Tests Can Improve Mud Design. Bushnell-Watson, M. and

Panesar, S.S. SPE Annual Technical Conference and Exhibition, Dallas : SPE, 1991. 22549.

21. New Approach to Differential Sticking. Courteille, J.M. and Zurdo, C.A. SPE Annual Technical

Conference and Exhibition, Las Vegas Nevada : SPE, 1985. 14244.

22. Design of Oil Wells Using Analytical Friction Models. Aadnoy, B.S. and Andersen, K. s.l. : Journal

of Petroleum Science and Engineering, October 2001.

23. Application of New §-D Analytical Model for Directional Wellbore Friction. Fazaelizadeh, M.,

Hareland, G. and Aadnoy, B.S. s.l. : Journal of Modern Applied Science, February 2010.

24. A §D Analytical Model for Wellbore Friction. Aadnoy, B.S. , Fazaelizadeh, M. and Hareland, G.

s.l. : Journal of Canadian Petroleum Technology, October 2010.

25. Theor and Application of Buoyancy in Wells. Aadnoy, B.S. and Kaarstad, E. IADC/SPE Asia Pacific

Drilling Tech. Conf. and Exhibition, Bangkok, Thailand : IADC/SPE, November 2006.

26. Field Comparison of 2-D and 3-D Methods for the Borehole Friction Evaluation in Directional

Wells. Maidla, E.E. and Wojtanowicz, A.K. 62nd SPE Annual Technical Conference , Dallas, Texas :

SPE, September 1987.

27. Neural Networks: A New Tool for the Petroleum Industry? Ali, J.K. SPE, Aberbeen, UK : European

Computer Conference , 1994. 27561.

28. Hybrid Method for Porosity Classification in Carbonate Formations. Batyrshin, I. , Sheremetov, L.,

Markov, M. et al. s.l. : J. Pet. Sci. & Eng., 2005.

29. Applications of Neural-Networks in Multi-Well Field Development. Centilmen, A., Ertekin, T., and

Ahmed Msahli 60

Pipe Sticking 2016

Grader, A.S. SPE Annual Technical Conference and Exhibition, Houston, Texas : SPE, 1999. 56433.

30. Support Vector Networks. Machine Learning. Cortes, C. and Vapnik,. 1995, Vols. 20: 273-297.

31. Demuth, H. and Beale, M. Neural Networks Toolbox for use with MATLAB. In User's Guide, Fifth

Printing, Version 3. s.l. : Natick, Masachusells: Mathworks, Inc., 1998.

32. Forecasting PVT Properties of Crude Oil Systems Based on Support Vector Machines Modeling

Scheme. El-Sebakhy, E.A. s.l. : J. Pet. Sci. & Eng. , 2009.

Ahmed Msahli 61

- Proceedings_2013.pdfUploaded bysenvimjag
- Genetic Algorithm with SRM SVM Classifier for Face VerificationUploaded byAnonymous Gl4IRRjzN
- 54Uploaded byNajmul Hasan
- entropy-15-00416-v2Uploaded byHussein Razaq
- rapidminer-4.3-guimanualUploaded byKang Amuch
- Key for Dm II Mid 2017 May Set AUploaded byjyothibellaryv
- 1-s2.0-S0957417414003455-mainUploaded byKhairul
- prt%3A978-0-387-74759-0%2F13Uploaded bydebasishmee5808
- Short Term Load Forecasting Using a Hybrid Model Based on Support Vector RegressionUploaded byIJSTR Research Publication
- 05601721Uploaded byhemmoger
- 7Uploaded byNirjhor Roy
- Drilling MuddUploaded byProf Youpa
- Limpieza en Huecos Direccionales y Horizontales-new Drilling TechnologyUploaded byBrayan Alexander Galloso Terrones
- classification3Uploaded bySaad Chakkor
- FreyReportUploaded byNguyen Huu Qui
- Trapped Annular Pressure Gonzales 4th 2008Uploaded byKevin Mueller
- Acoustic Signal Based Fault DetectionUploaded byEko Riyanto
- 1969_6_1Uploaded byFelipe Pereira
- Lost Circulation Assessment and Planning Program[1]Uploaded byArturo Cordova Somellera
- ModEco ManualUploaded byfriderikos
- (16)Schedule62(315page)-3Uploaded bySarang Bonde
- Pile Group Program for Full Material ModellingUploaded byTuroy
- Sand Aging Field StudyUploaded byKen Liew
- OCTG NS-17 ABC of Hole CleaningUploaded byDon Braithwaite
- Candidate Screening FormUploaded byrdos14
- Note11_ShearStrengthOfSoilUploaded bySlavisa Prostran
- Scdot Geotechnical Design ManualUploaded byGuillermo Andres Soto Conde
- LS-G-AN3 Gravity Wall Analysis 1Uploaded byFarhan
- Real time Face Recognition using Curvelet Transform and Complete Local Binary PatternUploaded byJournalofICT
- Cho & Santamarina 2001 Clase 4Uploaded bycolisho2004

- 00086863Uploaded byWaleed Barakat Maria
- SPE-96870-MS-PUploaded byWaleed Barakat Maria
- 00000544.pdfUploaded byWaleed Barakat Maria
- Chemical Mixing Liquid and Dry GS 102Uploaded byWaleed Barakat Maria
- 00024606Uploaded byWaleed Barakat Maria
- 00013098Uploaded byWaleed Barakat Maria
- IPTC-10506-MS-PUploaded byWaleed Barakat Maria
- 00093008Uploaded byWaleed Barakat Maria
- 00053955.pdfUploaded byWaleed Barakat Maria
- 00086698Uploaded byWaleed Barakat Maria
- 00008410Uploaded byWaleed Barakat Maria
- 00086693Uploaded byWaleed Barakat Maria
- 00080513Uploaded byWaleed Barakat Maria
- Big bagUploaded byWaleed Barakat Maria
- Assessing Geo HazardsUploaded bytavis80
- IPTC-10243-MS-P.pdfUploaded byWaleed Barakat Maria
- 00069425Uploaded byWaleed Barakat Maria
- 00072366Uploaded byWaleed Barakat Maria
- 00069704Uploaded byWaleed Barakat Maria
- 00068097.pdfUploaded byWaleed Barakat Maria
- 00087261Uploaded byWaleed Barakat Maria
- 00080994Uploaded byWaleed Barakat Maria
- 00080465Uploaded byWaleed Barakat Maria
- 00090798.pdfUploaded byWaleed Barakat Maria
- 00037084Uploaded byWaleed Barakat Maria
- 00036919Uploaded byWaleed Barakat Maria
- 00081043Uploaded byWaleed Barakat Maria
- 00057716Uploaded byWaleed Barakat Maria
- 00018700Uploaded byWaleed Barakat Maria
- 00082028.pdfUploaded byWaleed Barakat Maria

- CPT for RookiesUploaded byLuiz Alves
- AADE-07-NTCE-31Uploaded byJumon Kashyap
- 30-Stabilizers.pptUploaded bynjileo
- Proxxon Micromot UkUploaded byAnghel Curty Jr.
- SAES-Q-004 Installation Piles Offshore Structures 2004Uploaded byymaseda
- SWMS 001 Drilling Soil BoresUploaded byEngr Faheem Akhtar
- 3-22 Estimating Soil Fines ContentsUploaded bynuufleming
- Pos GripUploaded byKevin Mueller
- Acid Matrix AramcoUploaded byKhalid Al-mureeh
- Pre Mantinance S1DUploaded byFelipe Kennedy
- TREX-05860Uploaded byOSDocs2012
- O NEIL REESE 1999 DRILLED SHAFTS CONSTRUCTION PROCEDURES AND DESIGN METHODS.pdfUploaded byChalko
- Sandvik Mining Grinding B1_773_1Uploaded byNeil
- Method of Boring a-bUploaded bydimaswolv
- Effective Matrix Acidizing in High-TempeUploaded bydesosan
- CE 326 Slope Design Ex Boring Logs v02 20160202Uploaded bybabsy017
- Cardwell KB 200 CUploaded bydraghiceanu
- Appendix 5 Foundation Method StatementUploaded byTAHER AMMAR
- Depre ForajUploaded byDavid Marian
- Standard Method of Measurement for Civil Engineering WorksUploaded bydave4359
- 0132368692-Ch02_ismUploaded byOnline Review Specialists
- Specification F 750 HP Trailer RigUploaded byAlexander
- A Solution Model for Stuck-pipe & Fishing Challenges in Geothermal Drilling- a Case of the Olkaria Geothermal Field(1)Uploaded byzeze_13
- To Trip Or Not To Trip.pdfUploaded byjamarti
- Transocean v. MaerskUploaded byFrank Bednarz
- h2s.docxUploaded bySaqi Khan
- Drilling RigUploaded byRaabhat Issi
- MP PHED Tubewells SOR.pdfUploaded byvikasptk
- RP-002393 Cab Comp Integral CameronUploaded byrps1977
- Shallow Gas SummaryUploaded byHerbertLansink