Download as pdf or txt
Download as pdf or txt
You are on page 1of 41

Pipeline Integrity Assessment Reliability

ILI Performance Verification and


Validation

Corrosion Threat: Three Dimensions of


Shape and a Fourth Dimension of Time

Webinar June 2011


Agenda

Introduction by Organizers
Audience Participation
Poll Questions directed to audience
Audience questions to Applus RTD or Organizers
Applus RTD
Martin Fingerhut- Manager AIES
Rick McNealy- Principle Engineer
In-Line Inspection Verification and Validation
Why its important
Role of Technology
Managing In
In-Ditch
Ditch Errors
Value of True ILI Performance
Questions
Applus RTD

ApplusRTD: Leading provider of Asset Integrity


Management, Inspection and specialist NDT
services
Over 70 years at the forefront of
t h l i l development.
technological d l t
80 offices in 29 countries

Employs around 3000 personnel with experience spanning materials,


i
inspection
ti engineering
i i and d integrity
i t it management,
t pre or in-service
i i
inspection; advanced and conventional NDT.
Supplies own specialist equipment world-wide
Member of PRCI

|3
ILI Verification and Validation

Why is this an issue?


How can ILI Performance affect Pipeline Integrity?
Conservative vs Non-conservative Response
Safety and Resources
Fact is: When corrosion is present
50% to 90% of ILI indications are re-coats
Rare occasions; unexpected safety conditions missed
Pburst ILI>> Pburst Actual
How do we verify and validate performance?
Comparison of ILI with Direct Exam NDE; API 1163
Tools and Techniques
State of Art?
Problem Definition

MFL In-Line Inspection Performance Claims


Not an direct wall measurement technology
Expert System that must be trained (i.e. API 1163)
Does not claim to detect everything
B31.8S: ILI has limited discrimination capability
Fundamental Integrity Management Errors to Consider
Falsely accepting ILI indications as safe
Consequence: Unexpected Failure
Falsely rejecting ILI indications as repairs
Consequences
Both errors can result in unexpected failures
Failure to Respond; Never made it to dig list
Failure to Respond
p in Time: Prioritization and Resources
FAQ 224 (timely response)
Solutions Presented and Discussed

Sample Size
Technologies to gather large, significant samples
In-Ditch NDE Performance (all the errors)
Nil error beneficial
Reveals True ILI error
Digital data allows for accurate correlation
Matching complex corrosion
Sorting out false call metal loss
Managing ILI filters
ILI Response Management
Acceptance
A off log
l based
b d on large
l scale
l tests
Log rejection; rejection of ILI performance (should be rare)
Is Tool in-spec for what it is designed to see?
Calibration of response based on true ILI tool error
Considers the actual metal loss
Addressing ILI Performance

No problem with
what you dig
But recoats vs
repairs cost
resources
cost of
business?
Failures very, Improve Adjust
very rare
What is left in No
ground ?
Y
Yes
Confidence? Confidently
Accept ILI
Re-inspection Indications,
No Integrity

Failures have
Threat?
ILI Performance
happened: rare
- Exceptional conditions Repair? Or Recoat?
Exceptional Conditions

Magnetic Flux Leakage MFL


Very reliable for metal loss and what it claims to detect and discriminate
MFL does not claim to see and measure ALL metal loss
80% of the time +/-10% based on POF shapes
Fact is;
It happens
pp
Rare unexpected failures
Past/Recent MFL
Pburst ILI> 1.1*Pburst Actual
Reluctance to admit it
Exceptional Conditions
Sensitivity to Corrosion Growth
Challenge
What is important
Is tool in spec?; expected performance
True
T ILI performance
f for
f actual
t l corrosion?
i ?
Conservative Response
Effective Response
Response Management (Can I trust the pig?)

API 1163 guidance for Field Comparisons


Practical ?
No
Needed ?
Pieces
Dont trust the
Parts
Sensors pig
Speed
Find
Yes Measure
Matter
Yes

Find
Predictions; Any System Measure
Data? Matter
Find? No Dig
Pull Tests
Measure?
Do they Matter?
Analog Yes
Pipelines?

Find
No Measure
Matter

Get Performance Data:


Compare Predictions
with Actual
Sample Size

API 1163 Appendix E Table 8


/ 10%wt
Assume 80% +/-
Sample Size of 5
4 X 5 (80%) have to be outside of
+/ 10%wt to reject p
+/- p=0.8
08
If 2 X 5 (40%) in-spec cannot
reject +/-10%wt but larger
sample could still reject +/
+/-
10%wt.
Therefore Cannot reject does
not mean Accept
Consistent w/in API 1163 is
misleading. Revision
Larger Samples required to
confidently ACCEPT a
performance
So Large Samples are Better

Allow for determination of the bounds between what we


can confidentlyy reject
j and confidentlyy accept
p
Converges on the true population proportion (x/n) for
large samples
i.e.
i 80% off theh time
i
Large is 75 to 200 comparisons
Tools
oo s and
a d processes
p o esses required
equ ed that
a capture
ap u e large
a ge sa
samples
p es
Cannot forget about random samples
Timely and Cost Effective Execution
In-Ditch Measurement Error and True Error

errorILI = (D/t)ILI (D/t)actual


= [(D/t)ILI (D/t)Field]+ [(D/t)Field (D/t)Actual]
= [(D/t)ILI (D/t)Field
Fi ld] + errorField
Fi ld

This equation shows why it is important to minimize the error in field


measurement and matching error.
*
Set Z = (D/t)ILI (D/t)Field so that
errorILI = Z + errorField
Z = errorILI errorField

* IPC2010-31269 ILI Performance III Effect of In-ditch Errors

With nil NDE error (< 3%wt), observed performance is TRUE ILI error
No Correction Needed | 12
How do we compare ILI to Actual Conditions?

Direct Measurements
Conventional NDE
Pit gauges
Straight beam UT
Advanced NDE
Laser Profilometry
UT scanners
Matching of ILI predictions with actual conditions
Manual matching
Pattern matching
Metal Loss Depth Error

Both cases measured with same tool, but


Different Metal Loss Shapes, different error
Error due to locating within corrosion

100% 100%
%

90% 90%

80% 80%

70% 70%
PitGauge,d/t

60%

PitGauge,d/t
60%

50%
50%

40%
40%
Rick
30%
Chris(III) 30%
Ham(Eng) AMS(Eng)
20% DZK(Eng)
John(III) 20% SL(Mgr)
Ben RWA(LvlIII)
10%
JT(LvlIII)
unity 10% MvK(Eng)
0% unity
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Validation,d/t
Validation,d/t
Nil Depth Error +/- 10% wt Error
Conventional NDE

Manual Validation
Adequate Confidence and Accuracy
Resolution
R l ti d
depends
d on corrosion
i ffrequency
Accuracy depends on corrosion morphology
But you have to have base metal for reference
Where is the deepest point within each grid?
API 1163:How much data for statistically significant results?

Pit Gauge Probe UT Pen Probe


Wide Area Corrosion Error

Pipe may not be straight nor round


When metal loss is extensive
Local reference can be source of error
Reconstructing a surface is key
Advanced NDE solves this issue
Depth Performance Example; MFL

MFL Prediction: 50%wt Depth


Actual Depth Laser Measurement: 79%wt
2mm
2 wide
id att maximum
i depth
d th
10 mm wide at 50%wt depth
ILI vendors have improved algorithms to improve performance
Advanced NDE- Addresses the Issues

Minimizes In-Ditch Measurement Errors


Laser Profilometry (LPIT)
AUT Scanners
Better than +/- 2%wt
Digital Source of Large Samples !
Laser Profile Inspection Tool

LPIT

Launch
Handyscan 3D

Benefits of Handyscan Application for


Corrosion
The hand-held 3D laser scanner--- Handyscan, integrated with Applus RTD Remaining Strength Analysis
software---CorroPro, for in-the-ditch use, and real time burst pressure determination.

The only truly portable, light weight 3D scanner for external corrosion
mapping on pipelines (diameters >/= 4)
Fast and accurate in-the-ditch assessment of external corrosion
ILI Verification, Correlation & Permanent Reference Pieces for Subsequent
ILI Runs- External Corrosion AND Mechanical Damage

Launch
Handyscan 3D

PROCEDURE

1. Prepare
p Scan Surface to NACE 2 finish

2. Apply Randomly Spaced and removable


referencing target (Dots)

3. Rebuilding Original Pipe Surface, and


Calculate (corrosion pit depth)
3D Handyscan probes the pipe corrosion
surface and record its profile data.
Custom Algorithm applied to rebuild the
Original pipe surface, independent of
any pipe distortion (bend/oval)
Real time comparison of the rebuilt pipe
surface with corroded surface to
calculate corrosion pit depth/remaining
pipe wall thickness.
Automatically extract selected corrosion
areas to
t feed
f d CorroPro
C P f
for R
Remaining
i i
strength analysis; B31G, ModB31G,
Rstreng, FEA
Laser Accuracy
Laser Profilometry
Depth 80% +/- 3%wt applied to pipe
Handy Scan and LPIT surface rebuild algorithm
Performance Qualified, multiple diameters 100

HandyScan

Apparent Comparison from Ditch


90

80

Reflects the True ILI error 70

HandyScanMLDepth,%wt
No correction needed for in-ditch error 60

50

40

y=1.0769x 4.8922
30

100

LPIT 20

90
10

80
0
0 10 20 30 40 50 60 70 80 90 100
70
UTMLDepth,%wt
p ,
LPITMLDepth,%wt

60
y=1.0865x 1.9665

50

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

UTMLDepth,%wt
Surface Algorithm is the Key

Performance Qualification Tests


Handy Scan
8 inch Pipe 100

+/- 1%wt 90

80

70

Depth,%wt
60

50
LaserMLD
40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

ActualMLDepth,%wt
When Expectations and Reality Correlate

We can respond and manage risk respecting errors


Random
System
But realityy is seldom p
perfect
What are the ranges for correlation

Correlation depends on measurement AND matching


Both Errors affect degree of correlation
Useful Performance Correlations

Accurate measurements AND Accurate Correlations


Must consider what is important to prove (rules needed)
ILI in spec for what it is supposed to detect and measure?, or
TRUE performance given the actual metal loss conditions?

Non-conservative depth predictions can become evident


Generally contribute to non-conservative Pburst
Next step is to understand why??
Take
T k Action;
A ti Adjust
Adj t RResponse or R
Re-analyze
l ILI data
d t

DepthCorrelation BurstPressureCorrelation
80 10000

70
8000
60

ILIEAARstrreng,(kPa)
ILIDepthh,(%wt)

50
6000
40

30 4000

20
2000
10

0 0
0 10 20 30 40 50 60 70 80 0 2000 4000 6000 8000 10000

LaserValidation Depth,(%wt) LaserValidation RSTRENG Pburst,(kPa)


Matching ILI to In-Ditch Data

Often manual correlate is adequate


Often straightforward but
Can you get enough data
Statistically confident?
Matching ILI to In-Ditch Data

Sometimes a better level of understanding is required


ILI location conventions; top-bottom-or center of clusters?
Make up of ILI boxes could help?
Here is where pattern matching can increase understanding.

Why MFL and Corrosion may not


correlate ?
Effect of ML shape on Flux Leakage

Location and Measurement


Error
ILI boxes may not locate or
measure actual ML shape
ILI boxes may not be actual
ML
Magnetic false calls
Laser Sure

DepthCorrelation
80

70

60
pth,(%wt)

50

40
ILIDep

BurstPressureCorrelation
30 10000

20 8000
ILIEAARsstreng,(kPa)

10 6000

0 4000
0 10 20 30 40 50 60 70 80
2000
LaserValidation Depth,(%wt)
0
0 2000 4000 6000 8000 10000

LaserValidation RSTRENG Pburst,(kPa)


New Technique: Pattern Matching

High Res MFL against


Laser NDE
Manual Single event
correlation very difficult
Remember MFL not the
same as UT- not direct
measurement
Software solutions to ILI
align patterns
Pits and Clusters
Laser Sure
Compensating for
magnetic anomalies
and ILI boxing
boxing

Align Patterns Dig 5


Scan 01
Laser Sure: Managing All the Errors
Location Area Match: Note Box Density Difference
Depth Error ILI Calcs Pburst from Box Profiles
Box Edit Error (related to LOD) Laser Calcs Pburst from Continuous Profile
Laser Boxes for location match

ILI Laser= Actual

Laser CLS/MFL CLS Overlay


Conservatism and Accuracy

Reduced Conservatism associated with more accuracy


p y reduced safetyy
does not imply
Requires reliable estimates of defect size and corrosion
growth rates
If performance cant support safety
- Adjust the dig response
- Adjust the ILI

* M Rosenfeld PHMSA
Public Meeting
g June 2009
Burst Pressure Validation Example

High Res NDE (Laser)


Manual Match
Correlation errors
EAA 12000

12 ILI Predicted
P di t d Repairs
R i
67% False (Recoat) 10000

0.85dL 8000
SF

ILIPburstPredicted,kPaa
43 ILI Predicted Repairs 6000
EAA
0.85dL

88% False Call (Recoat) Linear(EAA)


Linear(Psafe1)

Non-conservative
Linear(Psafe2)
4000

predictions

SF
2000

Requires more digs


0
Based on Pb variance 0 2000 4000 6000 8000 10000 12000

InFieldPburstRSTRENG,kPa
False Acceptance
When Things Are As-Expected

Both ILI Tool and Corrosion; in-spec


Depth: Accept 80% +/- 10% wt, no bias
Nil NDE and Nil Correlation Error (Laser Sure)
Decreased variance in Pburst error
No Need for More Conservative Dig Response
Repair rate will be as good as RSTRENG allows
PburstUnity
14000

12000

10000
urst,kPa

8000
1.25MAOP
ILIEAAPbu

6000
MAOP

4000

1.25MAOP
OP
MAO
2000

0
0 2000 4000 6000 8000 10000 12000 14000

LaserValidatedRSTRENGPburst,kPa
So What?...

Effect of True Tool Error on POE (MAOP)


D=20 in, t=.219 in, SMYS=52000 psi, MAOP=820 psi
Higher accuracy ILI assessment sensitive to ILI error
Example of adjustment to response based on error
Understanding of bias and variance basis for action

POE vs ILI Tool Depth Tolerance, Nil Length


Error
ILI Tool Depth Tolerance, +/- %wt
5 10 15 20 25
1.0E+00
1 0E 01
1.0E-01
1.0E-02
1.0E-03
1.0E-04
1.0E-05 B31G
1.0E-06
ModB31G
1.0E-07
POE

1.0E-08 RSTRENG
1.0E-09
1 0E 10
1.0E-10
1.0E-11
1.0E-12
1.0E-13
1.0E-14
1.0E-15
1.0E-16
Maximize Value in the Data

Corrective lens for Validation Effort


Same corrosion
Same ILI performance (not +/-10%) but bias only evident when
Data set is larger and pattern matching used
Digging any way, get something you can work with
FICLMatchingLowestILIRPRwith DepthCorrelation
LaserFeatures Depths 80
80.00
70
70.00
60
60 00
60.00

ILIDepth,(%wt)
50
ILIDepth,%wt

50.00
40
40.00
30
30.00

20 00
20.00
20

10.00 10

0.00 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80

LaserValidation Depth,%wt LaserValidation Depth,(%wt)

High Resolution NDE but High Resolution NDE but


Manual Correlation Pattern Matched Data
Pburst is the rest of the story

Laser Sure Matched; Std Dev less than half of manual

Manual Matched Laser Sure Matched


How much conservatism is enough ?
Validation Justifies ILI Response Schedule
Validation provides error population (Actual-ILI)
Sample from the total feature population
Assume errors are normally distributed
Mean indicating bias from unity
Standard deviation
Determine likelihood for ILI features accepted
p at S.F. having
g actual
dimensions exceeding limit state
97%, 99%, 99.999% of values within limit state?
Useful when managing finite resources
Opens up possibilities for data calibration

Laser Sure minimizes all the errors


Extracting all possible value from Effective Area, highest confidence
The money is in the excavation
Maintaining safety
Immediate
Future (re-inspection interval)
Challenge

Implement processes to recognize exceptional corrosion


when it occurs
Role for ILI Verification Processes
e.g. API 1163 Figure 4 approach
Historical Reliabilityy
Integrity Sensitivity
Role for Field Comparison or ILI Validation
Direct examination (see IPC-31269)
IPC 31269)
Appropriate NDE technology, conventional or advanced
Appropriate samples
Processes that can respond in time
Provide actionable data in time to support Discovery
Dig when needed
Dont dig when it is not needed
Why Advanced NDE and Laser Sure?

Data Population
Records and Considers ALL the data
3 dimensional shape data
Lots of data in minimum number of digs
In-Ditch Accuracy
High Resolution Laser Profilometry or AUT
Nil error or minimum operator variability
THE answer when you cant get to native surface
Matching error controlled
Manage high res MFL false call boxes
Assurance of TRUE ILI Performance
Actionable data and conclusions
Insure Safety and Optimization of Resources
Thank You

Questions

You might also like