Software-Based Leak Detection-Presentation - PPSX

You might also like

Download as ppsx, pdf, or txt
Download as ppsx, pdf, or txt
You are on page 1of 51

Software - Based Pipeline Leak Detection

Presented by:
James Akingbola
David Mannel
Elijah Odusina
Overview

S
H o
Intr ar
ft Generaliz
odu Methods d ed Simulation

ctio
of leak w Likelihoo
Procedure &
detection w Results
n ar a d Ratio

e r
e
Introduction
 Pipelines used as bulk carriers of crude oil
and natural gas
 Used in water distribution systems
 Results of Leakages include:
 Loss of product
 Environmental hazards
 Loss of life
Introduction

Gas pipeline explosion in the Ukrainian village of Luka south of the capital Kiev May 7, 2007
Pipeline Configurations
Methods of leak detection
 Hardware Method:
This method is used to detect leaks with instrumentation
outside the pipeline.

 Software Method:
This method uses different instruments to monitor internal
parameters (such as pressure, flow, temperature) that infer
product release.
It is also known as computational pipeline monitoring
(CPM).
Hardware Method
 Acoustic Emissions

 Fiber Optic Sensing

 Vapor Monitoring

 Multi-hop Sensor Wireless Network*

* Method developed by OU faculty: Professor Sridhar Radhakrishnan


Hardware Method - Acoustic Emissions

* http://www.wavealert.com/Pages/sys1.html
Hardware Method – Fiber Optic

.
Hardware Method – Vapor Sensing Method
Pipeline Monitoring: Multi-hop Sensor Wireless Network
Professor Sridhar Radhakrishnan
Solution
Problem: Develop continuous real-time Multi-hop wireless sensor network with
monitoring of pipelines to determine leak and appropriate sensor fusion technologies.
other structural damages.

Importance: Failure of gas pipelines will result


in both human, property, and environmental
damage.

Current Solution: Low flying aircraft, visual


inspection, and use of pigs for internal
monitoring.

Weaknesses: Expensive, non-continuous Research Issues


monitoring, fail-first, fix-later solutions. • Communication in the presence of unreliable
sensors
• Power aware strategies
• Optimal sensor configuration and data fusion
School of Computer Science
Hardware Methods
 Advantages
 Good sensitivity to leak
 Very accurate in leak locating

 Disadvantages
 System costs are usually high
 High complexity of installation
Software Methods
 Balancing Systems
 Uses the principle of mass conservation
. . dML
M I (t)  MO (t) 
dt
 Types of Balancing systems
 Volume Balance
 Compensated Mass Balance
 Model Compensated Mass Balance
Software Methods- Balancing Systems
 Advantage
 It is simple to implement
 Cost effective

 Disadvantage
 Can be affected by instrument error
 High rate of false alarms
Software Method - Pressure Analysis
 Gradient Intersection method
 This method uses SCADA* values to calculate the theoretical
hydraulic profile or baseline of the pipe
 The presence of a leak can be determined from a specific
deviation or combinations of several deviations from the baseline
Software Method – Gradient Intersection
 Advantages
 Estimation of leak location can be determined
 Disadvantages
 Dependent on instrument sensitivity.
 Smaller leaks typically take longer time to
detect.
 Does not give magnitude of leak
Software Method – Gradient Intersection

Xleak
Software Method - RTTM
 Real Time Transient Modeling (RTTM)
 Involves computer simulation of pipeline conditions using
advanced fluid mechanics and hydraulic modeling.
 Basic equations used in RTTM are:
d v
 0
 Continuity equation: dt x
dv 1 p
 .  fD  0
dt  x
 Momentum Equation:
dh 1 dp
 .  lL  0
dt  dt
 Energy equation:
Software Method - RTTM
 Advantages
 It takes into account the configuration of the pipe as
well as the product characteristics.
 Very fast in detection and location
 Disadvantage
 It is a complex way of leak detection
 The RTTM method costs much more than the other
methods
 Requires many instruments, controller training and
maintenance.
 Errors in instrument calibration could raise false alarms
Generalized Likelihood Ratio
 This is a statistical method modeled after
the flow conditions in the pipeline
 A mathematical model that describes
effects of leaks and biases on the flow
process is used.
 Can detect leaks in pipeline branch,
location in the branch and magnitude of
the leak.
 Can identify various types of gross errors
GLR for Gross Error Identification
Process Model Process leak Model
Steady state model without leak A mass flow leak in process unit (node) j
z xv of unknown magnitude b can be modeled
by;
z is a measurement vector
Ax  bm j  0
x is the true value of state variables
the elements of vector m j corresponds to
v is the vector of random error
the total mass flow constraint associated
Ax  0 with node j

A = constraint matrix
Procedure for single gross error
Measurement bias Model
z  x  v  bei r  Az
When there is no gross error;
b is the bias of unknown magnitude in
instrument I
E r   0
e i =is a vector with unity in position i
Cov  r   V  AQA '

S. Narasimhan and R.S.H. Mah. "Generalized Likelihood Ratio Method for Gross Error Identification." AIChe Journal
33, No.9(1987): 1514-1519.
GLR for Gross Error Identification
If a gross error due to a bias of magnitude let μ be the unknown expected value of r,
b is present in measurement I, then; we can formulate the hypotheses for gross
error detection as
E r   b Aei H 0 :  0
If a gross error due to process leak in H 1 :   b fi
magnitude b is present in node j, then;
Ho: is the null hypothesis that no gross
E r   bmi errors are present and
H1: is the alternative hypothesis that
either a leak or a measurement bias is
When a gross error due to a bias or present.
process leak is present;
b and fi are unknown parameters. b can
E r   b f be any real number and fi will be referred
i to as a gross error vectors from the set F

F   Aei , m j :i  1...n, j  1...m


where
 Ae i For a bias in measurement i
fi 
m j For a process leak in node j
GLR for Gross Error Identification
We will use the likelihood ratio test The maximum likelihood estimate b :
statistics to test the hypothesis by:
bf V   f V r
1 1 1
fi
Pr r H1 i i
  sup
Pr r H 0 
Substituting b in the test statistics
equation and denoting T by Ti:

 sup
 
exp  0.5 r  b f i V   r  b f 
' 1
di2
exp 0.5r V r 
i
' 1 Ti 
bi f i Ci
Where: ' 1
di  f i V r
The expression on the right hand side is
always positive. The calculation can be ' 1
simplified by the calculation by the test
Ci  f i V f i
statistics, T as:
This calculation is performed for every
1

T  2ln  supr V r  r b f i V r b f i
b, f j
'
 
' 1
 vector fi in set F and the test statistics T
is:
T  sup Ti
i
GLR for Gross Error Identification
NOTE

The gross error that corresponds to vector f* is identified as


the gross error and its magnitude b^ can be estimated

Performance Measures
The overall power of the method to identify gross errors is
given by:

Overall
power
 Number of gross errors correctly identified
Number of gross errors simulated
GLR for Gross Error Identification
 Results & Discussion
For the Recycle process network

1 2 3 5 7

6
GLR for Gross Error Identification
Constraint Matrix for the process Network

Streams
Type of
Constraint Pipe 1 2 3 4 5 6 7
Mass
Balance A 1 -1 0 1 0 1 0
Mass
Balance
B 0 1 -1 0 0 0 0
Mass
Balance
C 0 0 1 -1 -1 0 0
Mass
Balance
D 0 0 0 0 1 -1 -1
GLR for Gross Error Identification
Bias in sensor 1
Simulations (Ti)
Sensors
1 2 3 4 5 6 7 8 9 10
1 2110 719 1190 431 1496 2457 1394 2386 1404 1056
2 61 729 16 0.18 181 930 1 34 189 491
3 82 631 227 141 491 712 521 10 535 4
4 371 5 159 196 99 19 751 10 114 528
5 1144 379 270 86 283 1008 217 32 21 39
317 550 29 583 72 926 127 10 33 261
6
919 39 1538 0.18 975 523 755 2659 1052 336
7
2110 729 1538 583 1496 2457 1394 2659 1404 1056
T
-58 34 -50 -35 49 -63 47 66 48 -41
b^

-50 -35 35 -33 32 -60 45 -55 40 -50


b

Overall power = 0.6


GLR for Gross Error Identification
Leaks in node B and C

Overall power= 0.8


Advanced Pipe Network
Error vs. Simulated Magnitude (1% Meter Variance)
300.0000%

250.0000%

200.0000%
Error

150.0000%
TRUE
FALSE

100.0000%

50.0000%

0.0000%
0 20 40 60 80 100 120 140

Simulated Magnitude
Error vs Simulated Magnitude (3% Meter Variance)
300.0000%

250.0000%

200.0000%
Error

150.0000%
TRUE
FALSE

100.0000%

50.0000%

0.0000%
-250 -200 -150 -100 -50 0 50 100 150 200 250

Simulated Magnitude
Error vs Simulated Magnitude (5% Meter Variance)
300.0000%

250.0000%

200.0000%
Error

150.0000%
TRUE
FALSE

100.0000%

50.0000%

0.0000%
0 50 100 150 200 250 300 350 400

Simulated Magnitude
Overall Power vs. Simulated Magnitude (1% Process
Variance)
1.2

0.8
Overall Power

0.6
TRUE
FALSE

0.4

0.2

0
0 20 40 60 80 100 120 140

Simulated Magnitude
Overall Power vs. Simulated Magnitude (3% Meter
Variance)

1.2

0.8
Overall Power

0.6
TRUE
FALSE

0.4

0.2

0
-250 -200 -150 -100 -50 0 50 100 150 200 250

Simulated Magnitude
Overall Power vs. Simulated Magnitude (5% Meter
Variance)
1.2

0.8
Overall Power

0.6
TRUE
FALSE

0.4

0.2

0
0 50 100 150 200 250 300 350 400

Simulated Magnitude
Simulation Procedure
Energy balance
P1  P2  ( P1  P1e )  ( P1e  P2 )
without leak

P1  P2  f (G ) The pressure drop becomes:


P1  P2  f (G, b, lb )
where
In the presence of leak of b  leak magnitude
magnitude b and location x from
lb  leak location
the head of the branch

Miguel J. Bagajewicz and Emmanuel Cabrera. "Data Reconciliation in Gas Pipeline Systems." Ind.
Eng. Chem. Res 42, No.22(2003): 1-11
Simulation Procedure
Problem formulation With Error:
Without Error: ~ ~

~ ~
Min (Gi  Gi ) * S  ( Pi  Pi )2 * S Pi1
2 1
Gi
Min (Gi  Gi ) * S  ( Pi  Pi )2 * S Pi1
2 1
Gi
i

i
Subject to:
Subject to:
Gi ,in  Gi ,out  b  0
Gi ,in  Gi ,out  0 So:

Pi,in  Pi,out  f (G, b, lb )


Pi ,in  Pi ,out  f (G)
Simulation Procedure
Leak detection procedure:
 Hypothesize leak in every branch and solve data reconciliation
problem

 Obtain GLR test statistic for each branch objno_leak –objwith_leak_k

 Determine the maximum test statistic objno_leak - objwith_leak_k

 We compare the max test statistic with the chosen threshold


value: Max{objno_leak – objwith_leak_k}> threshold value: leak is
identified and located in the branch corresponding to the
maximum test statistic

NOTE: Assuming only one possible error


Why pressure measurements?
 Stream 1 and 2 measured
 Discrepancy in flow measurement

Sensor 1 Leak Sensor 2

Case 1 0.4 0 0
Case 2 0 0.4 0
Case 3 0 0 -0.4
Sample Pipeline Network
Simulation Procedure - Leak in Pipe 1
Calculator Optimizer

Leak induced in Pipe 1


Simulation Results- Leak in Pipe 1
Leak Simulated Best Objective
Pipe
function
Pipe 1   1 15.9834

Location(m) 4000 2 18.0199


3 60.4256
Magnitude(kg/s) 4.915
4 60.7056
Measured Flow 15.482 5 21.3695
Measured Pressure 6 16.8630
2420.3
(KPa)
7 78.6864
Estimated
4.640
Magnitude(kg/s) 8 81.0650
Estimated
4048 9 123.2020
Location(m)
Simulation Procedure - Leak in Pipe 8
Leak induced in Pipe 8
Simulation Results- Leak in Pipe 8
Leak Simulated Best Objective
Pipe
function
Pipe 8   1 126.678
Location(m) 450 2 97.438
3 101.864
Magnitude(kg/s) 2.611
4 123.710
Measured Flow 4.946 5 126.447
Measured Pressure 6
2160.1 126.447
(kPa)
7 63.294
Estimated
2.609
Magnitude(kg/s) 8 0.151
Estimated
450 9 159.922
Location(m)
Simulation Procedure - Case Study
 Case Study
 Perfect measurements
 Introduce noise in measurements
 Different leak magnitudes

 Error
 The error is calculated as follows:
 Error = (Calculated Value-True Value)/(True
Value)
Simulation Procedure-Case Study
Error vs. Leak Magnitude (Perfect Measurment)
1.00%
The system gives perfect results when
perfect measurements are taken.

0.50%
Error

0.00%
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

-0.50%

-1.00%

Leak Magnitude
Simulation Procedure-Case Study
Error VS. Leak Simulated
As the simulated leak magnitude decreases, the error in
70.00%
the estimated magnitude increases.
The Simulated magnitude has little effect on the location.
60.00%
There is no trend in identification of True/False values.
50.00%

40.00% Location True


Magnitude True
Error

Location False
30.00%
Magnitude False

20.00%

10.00%

0.00%
0 0.5 1 1.5 2 2.5 3 3.5

-10.00%

Leak Simulated
Simulation Procedure-Case Study
Overall Power VS. Leak Simulated
70

60

50
Overall Power

40
Overall Power True
Overall Power False
30
As the simulated leak magnitude
increases, the overall power increases.
20

10

0
0 0.5 1 1.5 2 2.5 3 3.5

Leak Simulated
Future Research
Run more simulations
Explore multiple leak detection
Acknowledgements
 Miguel Bagajewicz
 Quang Nguyen
 Roman Voronov
 Rufei Lu
Questions

You might also like