Leak Detection(Chapter 17)
Content
Chapter 17 - Leak Detection
Pipe Line Leak Detection Techniques
• • • • • • • • • •
Causes and economical aspects of leaks Simple leak detection systems Pig-based monitoring systems Computer-based monitoring systems Pipe line leak phenomena Background philosophy of pipe line modeling Basic pipe line modeling equations Impact of instrument accuracy System design aspects and guidelines Development of pipe line monitoring systems
Pipe Line Leak Detection Techniques
R. A. Furness, Cranfield Institute of echnology, Cranfield, Bedford, !", and J. D. van Rett, Scientific Soft#are Intercomp, $ouston, e%as, !S& Summar Pipe lines are no# an integral part of the #orld's economic structure and literally billions of dollars #orth of products are no# moved annually in pipe lines( Both economic and environmental factors are influential in pipe line operation, and therefore integrity monitoring is vitally important in the control and operation of comple% systems( )eak detection systems range from simple, visual and #alking and checking to comple% arrangements of hard#are and soft#are( *o one method is universally applicable and operating requirements dictate #hich method is the most cost effective( he aim of the paper is to revie# the basic techniques of leak detection that are currently in use( he advantages and disadvantages of each method are discussed and some indications of applicability are outlined( +odern pipe line computer modeling and control is then revie#ed in detail( hese systems are the most fle%ible and versatile and are steadily being adopted( he influence of instrument accuracy on system design and performance is discussed and the basic modeling equations are revie#ed( !ntro"uction ,ur need to transport fluids from the point of production to the area of end use has led to a rapid increase in the number of pipe lines being designed and constructed( +any of these carry to%ic or ha-ardous products, often close to centers of high population or through areas of high environmental sensitivity( .ith the need to safeguard these lines, on-line monitoring is becoming routine and in some cases /0 hour surveillance is mandatory( .ith the increase in #orld terrorism, the need for rapid and reliable pipe line measurement and control systems #ill increase( he revie# begins #ith a summary of causes of leaks and the implications of the failure to detect them( Basic techniques are covered and the features of each are briefly discussed( he bulk of the paper deals #ith modern computer-based techniques( Basic flo# equations are covered and the online dynamic calculations required are listed together #ith the input data required to enable the monitoring to be accomplished #ith the minimum of do#ntime( he latest systems are capable of resolving do#n to 12 of the ma%imum rated flo# #ith a response time of a fe# minutes( Practical e%perience verifies this figure but the total costs of such a system could be high( he system therefore becomes a compromise bet#een response, performance, and alarm availability, and choice of instrumentation is crucial( he integration of good quality instruments #ith advanced real time models seems to be the current trend and the paper closes #ith some personal thoughts on future trends(
Causes and Economic Aspects of Leaks
here are four main categories of pipe line failures( hese are3
• • • •
Pipe line corrosion and #ear ,peration outside design limits !nintentional third party damage Intentional damage
+any pipe lines are operated for a number of years #ith no regard to any possible mechanical changes occurring in the line( Some of the products may be corrosive, the pipe line may be left partially full for periods of time, or atmospheric effects may cause e%ternal damage( hese three reasons are responsible for pipe line corrosion and this may give rise to corrosion 4pits4 developing along the line( hese are small in nature and could be responsible for material imbalances over a period of time( 5ery accurate flo# metering can be used to detect this as discussed in the ne%t section( &brasive fluids or dust-laden gas streams can give rise to pipe line #ear( &gain, this is a slo# process, but should a #eak spot develop 6more often than not close to a change in direction or section7 then a pipe break may occur very rapidly and totally une%pectedly( ,peration outside design guidelines is more common than is reali-ed, as operators seek to use the line for as many fluids as possible( If the line is designed for a certain ma%imum temperature and pressure, then operation at higher pressure and8or higher temperature could lead to spontaneous failure( he problem could be compounded if the line has a large but unkno#n amount of corrosion( !nintentional third party damage may occur if e%cavation or building occurs near buried lines( +ore often than not the right-of-#ays are not clearly marked and lines are sometimes broken by bulldo-ers or similar plant machinery possibly #ith fatal results( &n e%ample occurred in .est 5irginia in 19:0 #hen a /;-in( natural gas line #as punctured by an e%cavator and product leaked slo#ly into a nearby supermarket during the night( he building #as totally destroyed early ne%t morning by a massive e%plosion caused by a staff member lighting a cigarette( Such an occurrence could have been avoided by some form of leak detection on the line( Intentional damage unfortunately is on the increase and pipe lines carrying flammable or high value products make ideal targets( &larm systems linked to block valves can help to minimi-e the amount of product released as a result of sabotage, so again certain lines are instrumented #ith the intention of reducing the effects of planned terrorism( he costs of failure to detect leaks also fall into four main areas3
• • • •
)oss of life and property Direct cost of lost product and line do#ntime <nvironmental cleanup costs Possible fines and legal suits
hese are all self e%planatory, #ith the most costly of these being the last, although any of the four areas could be very e%pensive( he si-e of claims can run into many millions of dollars, so the cost of fitting and operating leak detection systems is often insignificant compared #ith the costs of failure of the line( It is this background that is causing operators and designers to turn to on-line integrity monitoring systems and the ne%t section of the paper looks at the more basic methods(
Simple Leak Detection Systems
he most basic method of leak detection involves either #alking, driving, or flying the pipe line right-of-#ay to look for evidence of discoloration of vegetation near the line or actually hear or see the leak( ,ften 4unofficial4 pipe line monitoring is performed by people living nearby #ho can inform the operator of a problem #ith the line( he most cost effective #ay to detect leaks in non-flammable products is to simply add an odorant to the fluid( his requires some care in selection, as frequently the odorant has to be removed before the transported fluid can be used( ,rganic compound make the most useful odori-ers, especially #hen the fluid being carried has no natural smell of its o#n( & good e%ample is carbon mono%ide, a highly to%ic but odorless gas #hich is often pumped in large quantities bet#een chemical plants( Chemicals such as mercaptans 6rotten egg smell7 or rimethylamine 6rotten fish smell7 can be added in small quantities to enable any leak to be located by smell( he disadvantage of such a method is that if the leak occurs in an area of no population the leak #ill go undetected unless the line is #alked regularly by pipe line surveillance cre#s carrying suitable 4sniffer4 detectors( hus to the apparent lo# costs of this method have to be added the costs of removing the odorant and maintain staff to check the line at frequent intervals( he location of a leak is also dependent on prevailing #eather conditions( Strong #inds may disperse the smell and atmospheric inversions may given an incorrect location of the leak and the uncertainty of relying on this method alone is high( *evertheless, it is a useful method if used in con=unction #ith other techniques( Simple line flo# balances are frequently used to check for gross imbalances over hourly or daily bases( his method may identify that a leak is present but flo#meters at each end of the line #ill not identify the leak location( & line pressure measurement system #ill be required in con=unction #ith the flo#meters to establish that the pressure gradient has changed from the no-leak situation( he method is useful, ho#ever, in identifying the e%istence of corrosion pits as the outputs of the flo#meters at each end of the line #ill consistently diverge if flo# in the line is maintained constant( If line flo# rate varies #ith time, then imbalances are more difficult to detect, since the flo#meter outputs may vary nonlinearly #ith flo# rate or may have different flo# characteristics from each other( & loss of product #ill be identified simply as the different bet#een the steady state inventory of the system and the instantaneous inlet and outlet flo#s( +athematically this is3
his last term can be calculated as the average of the integrated inlet and outlet flo#s in simple systems, but as #ill be seen later in this paper, the value of this term can be calculated more accurately and easily in real time as a function of several variables(
¬her method is based on detecting the noise associated #ith or generated by a leak( here are many instances #here fluid flo# can generate vibrations at frequencies in e%cess of /; k$-( hese frequencies are in the ultrasonic range but can be made portable so that pipe line cre#s can clamp a transducer at any point along the line to check for noise( By noting the signal strength, the source of the leak can be pinpointed( & similar technique, though based on a different principle, is the acoustic 4#avealert4 monitor, more correctly called a negative pressure #ave detector( his is a pie-oelectric sensor that gives an output #hen dynamically stressed( .hen a leak occurs there is a sudden drop in pressure at the leak follo#ed by rapid line repressuri-ation a fe# milliseconds later( he lo# pressure #ave moves a#ay from the leak in both directions at the speed of sound( he pipe #alls act as a #aveguide so that this rarefaction #ave can travel for great distances, attenuating in amplitude as it does so( Sensors placed at distances along the line can be triggered as the #ave passes and the location of the leak can be calculated from the line conditions and the internal timing devices in the instrument( Such devices are particularly useful in identifying large breaks in lines very rapidly, since the transient #ave typically moves 1 mile in > seconds in gases and almost 1 miles per second in liquids( ?esponse is therefore on the order of a fe# seconds depending on the positioning of the transducers( @igure 1 sho#s one adaptation #here the instrument can be made to cancel line noise and use the full measuring capability of the sensor for signal detection( here is the problem setting the background threshold correctly, as this may be affected by the location of the instrument in relation to bends, valves, pumps, regulators, etc( <%perience has sho#n that the installation is also critical to reliable performance, and there is also a dependence on the ?eynolds number of the flo#( Both of these affect the number of 4false alarms4 from the instrument(
Fi#ure 1. *egative pressure #ave detector 6acoustic monitor7
Pig Based Monitoring Systems
Pipe line pigs are frequently used for pipe line commissioning, cleaning, filling, de#a%ing, batching, and more recently pipe line monitoring( his last type of pig can be designed to carry a #ide range of surveillance and monitoring equipment and can be used at regular intervals to check internal conditions rather than continuously monitoring the line( Data, ho#ever, can be built up over a period of time to provide a history of the line( his information can be used to predict or estimate #hen maintenance, line cleaning, or repairs are required( If a leak is detected, for e%ample, by flo# meter imbalance, the location can be found by using a pig #ith acoustic equipment on board( his #ill alarm #hen the detection equipment output reaches a ma%imum and the precise location of the pig can be confirmed by radio transmitters also mounted on board( Pigs require tracking because they may become stuck, at a point of debris build-up, for e%ample( Pigging should be carried out at a steady speed, but occasionally the pig may stop and start, particularly in smaller lines( Information on #hen and #here the pig stops is therefore important in interpreting the inspection records( Pig tracking is not ne# and many such proprietary systems e%ist( In the best systems, ho#ever, a picture of the line is often programmed in so that outputs from =unctions, valves, cross-overs, and other geometries act as an aid to location( Pig tracking can make use of the acoustic methods discussed earlier( .hen the sealing cups at the front of the pig encounter a #eld, vibrational or acoustic signals are generated( <ach pipe line therefore has its characteristics sound pattern( #hen a crack occurs this pattern changes form the no-leak case and the location can be found from direct comparison( he technology has become so advanced that information on dents, buckles, ovality, #eld penetration, e%pansion, and pipe line footage can be generated( he equipment is often simple, consisting of sensor, conditioning, and amplifier circuits and suitable output and recording devices( Such a device developed by British Aas is sho#n in @igure /( he range of detection is dependent on the pipe line diameter and the type of pig( ,perational data have sho#n that light pigs in a /;; mm line can be detected at a range of : km, increasing to :; km for a heavy pig in a 9;; mm line( &s the signals travel at acoustic velocity this means a signal from a pig at :; km range #ill take 19; seconds to be picked up( Such technology is no# becoming routine in both offshore gas and onshore liquid lines(
Fi#ure $. 4Intelligent4 pipe line monitoring pig( Courtesy British Aas Corp(
Computer !ased Monitoring Systems
It is in computer-based systems that the greatest amount of data can be gathered, processed, analy-ed, and acted upon in the shortest period of time( Programs can calculate the inventory of the line at any time and compare this #ith accurate measurements at any section in the system( he effects of pressure and temperature on line dimensions, for e%ample, can be calculated to provide an accurate estimate of the mass of fluid in the line( Data from a #ide range of instruments can be transmitted by telemetry, radio, or phone links to a central computer #hich monitors the 4health4 of the line continuously( By changing programs and subroutines, a vast amount of functions and tasks can be accomplished very easily and cost effectively( he many functions that can be performed by computer-based systems include not only leak detection but also3
• • • • •
Pig tracking Batch tracking of fluids Inventory accounting ,n-line flo# compensation Instrument data and malfunction checking, etc(
Such systems have very rapid response and have the advantage of multiple inputs being re#ired before leaks are declared( hus some systems can run for short outage periods #ith no loss of integrity( hey are often comple% and costly to install but once the initial capital investment has been made running costs are lo#( he first section in the detailed revie# of such systems looks at the phenomena that need to be modeled #hen a leak occurs(
Pipe Line Leak Phenomena
.hen a leak occurs in a pipe line the measured pressure do#nstream of the leak falls but the pressure at the same location is predicted to rise( he first is not difficult to understand as the line is depressuri-ing as mass leaves through the leak( he second effect can be e%plained as follo#s( he equations presented later in the paper predict pressure based on measured flo# or flo# based on measured pressure( &s mass leaves the system through the leak hole, a reduced flo# at the do#nstream end is compared to the inlet flo#( he may not have chanted and so to balance the system, the equations predict a do#nstream presser rise( In physical terms the model thinks the line is 4packing4 and total system inventory is increasing( here is therefore a divergence bet#een measured and modeled pressure( he same is true of flo# changes, but here the inlet flo# could increase due to lo#er pipe flo# resistance bet#een meter and leak #hile the section outlet flo# #ill fall as mass leave through the leak instead of passing through the meter( hus a real imbalance #ill result( he model ho#ever #ill sho# an inconsistency since the pressure comparison #ill indicate line packing and the flo# comparison a line unpacking( If selected pressure and flo# imbalance limits are e%ceeded than leak is declared( he magnitude of the leak is predicted from the flo# imbalance and the location from the pressure profile imbalance and the flo# leak indicators( he impact of instrument accuracy on the predicted location is discussed later in the paper( It is vitally important to good leak si-ing and location to have the best pipe line instrumentation possible to minimi-e uncertainty(
Background Philosophy of Pipe Line Modeling
?eal time modeling is a technique that uses the full data gathering capabilities of modern digital systems and the computational po#er of small computers to give accurate 4snapshots4 of the pipe line( he #hole system is under the control of a SC&D& package of programs, #hich poll the data stations on the line, process the data, control the running of the transient pipe line model and activate the alarm and leak location routines( In addition to these basic soft#are modules, more comple% systems might include a predictive model to analy-e 4#hat if4 operating scenarios, provide an optimi-ation route for least-cost operating strategies or include a separate man8machine interface for the model system( he SC&D& interface is responsible for acquiring the data from the SC&D& system and relating them to the model representation of the line( &s a point in the system #here t#o large and independently developed systems =oin, this can be the source of many problems in the implementation of the real time modeling( ,nce the measurement data have been obtained, noise filtering and plausibility checking can be performed prior to running of the model( he model is the mathematical representation of the pipe line and #ill include such features as elevation data, diameters, valve and pump locations, changes of direction and the location or cross-overs and =unctions( he model provides data on the flo# conditions #ithin the line at intervals bet#een seconds and minutes, depending on operational needs( .ith the data available from both the measurement system and the pipe line model, the real time applications modules are run( hese are the leak detection and location routines in the conte%t of integrity monitoring( he leak detection module functions by computing the difference bet#een the modeled flo#s and pressures and the measured values at tall points #here measurements not already used as boundary conditions are located( Because the model accounts for normal transient operations, these differences #ill be small under normal conditions( .hen a leak is present, the differences become larger since the model system does not account for leakage( .hen these differences e%ceed preselected values, a leak alarm is declared( Sophisticated voting schemes #hich require multiple leak indicators to be in alarm for several consecutive time intervals are used to reduce false alarms #hile maintaining lo# thresholds( ,ften a simple pipe line balance of the type discussed earlier is used as a back-up to verify the transient model( ?esponse characteristics are, ho#ever, much slo#er than the real time model( ,nce the leak detection module declares a leak, the location routine is activated( he location is calculated from the magnitude and distribution of the leak indicators( &s an e%ample, in a straight pipe line #ith an upstream flo# discrepancy and a do#nstream pressure discrepancy as leak indicators, it is an easy calculation to determine #here the leak must be such that the leak flo# #hen added to the modeled flo# #ill produce the additional pressure drop observed at the do#nstream end( Solutions for pipe net#orks are more complicated and unique locations do not al#ays e%ist( his might be the case #ith parallel looped lines, for e%ample( In this case all the calculated locations should be checked(
The components of the real time modeling system "ork together to reduce the large #olumes of ra" data from the data acquisition system to a much smaller num!er of parameters and alarms that are more meaningful to pipe line operations$ %n the case of integrity monitoring& this means leak e#ents that could not !e detected !y inspection of the measured data can !e found and isolated quickly and relia!ly$ Basic Pipe Line Modeling Equations
he transient pipe line flo# model is the heart of a pipe line modeling system( he model computes the state of the pipe line at each time interval for #hich data are available( he state of the pipe line is defined as a set of pressures, temperatures, flo#s, and densities that described the fluids being transported at all points #ithin the system( hese quantities are found as the solution to a set of equations #hich describe the behavior of the pipe line system( hese basic equations are the continuity equation, the +omentum equation, the <nergy equation, and an equation of state( he continuity equation enforces the conservation of mass principle( Simply stated, it requires that the difference in mass flo# into and out of a section of pipe line is equal to the rate of change of mass #ithin the section( his can be e%pressed mathematically by the relation3
he momentum equation describes the force balance on the fluid #ithin a section of pipe line( It requires that any unbalanced force result in an acceleration of the fluid element( In mathematical form, this is3
he energy equation states that the difference in the energy flo# into and out of a section equals the rate of change of energy #ithin the section( he equation is3
hese three are the basic one dimensional pipeflo# equations are present in one form or another in all transient pipe models( .hat is needed to solve them, ho#ever, is a relation bet#een the pressure, density, and temperature for the fluid - an equation of state(
he state equation depends on the type of fluid being modeled, as no one equation fully describes the variety of products that are shipped in pipe lines( Some of the forms in use include a bulk modulus type of relation of the form3
his is normally used for liquids that can be regarded as incompressible( he bulk modulus B and the thermal e%pansion coefficient a can be constant or functions of temperature and8or pressure depending on the application( @or light hydrocarbon gases a basic equation such as P B r % ? % C % is appropriate, #here C 6the compressibility7 is a kno#n function of temperature and pressure( @or reasonable ranges of temperature and pressure a function of the form3
may be adequate( @or conditions #here fluids are transported at or near the critical point, a more sophisticated correlation is required to obtain the required accuracy but there is still a large uncertainty in the true density under these operating conditions and they should be avoided #herever possible( +any real time systems have been installed on lines carrying ethylene, butane, propane, and )*A8)PA products and have used the Benedict-.ebb-?ubin correlation as modified by Starling 6commonly called the B.?S equation7 #ith reasonable results( <ernatively, tables such as *D-19 or special correlations such as the *BS ethylene equation can be used, but this increases the comple%ity of the programming( & further complication arises if the product is not uniform throughout the system( his can occur due to batching of fluids or from varying inlet conditions( he first is more common #here different products are shipped in a common line( he properties are essentially discontinuous across the interface of the t#o fluids, but can be considered as uniform #ithin batches( he basic problem here is to keep tack of the location of the interface( Systems #ith continuous variations in inlet conditions occur in both liquid and gas systems( he variations can result from mi%ing of fluids of slightly different composition or from large variations in supply conditions( he governing equations presented are non-linear partial differential equations #hich are not suitable for machine computation( hey have to be solved by implicit or e%plicit finite difference techniques or the method or characteristics( ,f these the implicit method seems the most appropriate, as the other t#o methods could give rise to mathematical instabilities if the #rong timestep or distance interval is used( In order for the transient pipe line model to compute the state along the line at the end of each time interval, a set of initial conditions and a set of boundary conditions are required( he initial conditions specify the state at the beginning of the time interval and are normally the last set of data from the model( & steady state model must be used to generate an initial state #hen the model is started from rest, a so-called 4cold start(4 In this case a period of time must elapse before the pipe line model truly represents the actual state of the line( his time period allo#s any transient
conditions present and not represented by the steady state model to die out( Aenerally, less compressible systems #ill cold start faster than the more compressible ones, but the actual time for the transient model to be activated depends on the application( his may typically be on the order of E; minutes for a gas pipe line( he boundary conditions required by the model are taken from measured data along the line( @or each point #here fluids enter the system, its temperature, fluid type or composition, and either a flo# or a pressure is required( @or any equipment in the system that affects or controls the line, a suitable boundary condition must also be given( @or a gas compressor for e%ample, either its suction pressure, flo# rate or discharge pressure must be specified( &dditional measurements are used by the applications modules, generally by comparing their values to the corresponding model calculations(
Basic Pipe Line Modeling Equations
he transient pipe line flo# model is the heart of a pipe line modeling system( he model computes the state of the pipe line at each time interval for #hich data are available( he state of the pipe line is defined as a set of pressures, temperatures, flo#s, and densities that described the fluids being transported at all points #ithin the system( hese quantities are found as the solution to a set of equations #hich describe the behavior of the pipe line system( hese basic equations are the continuity equation, the +omentum equation, the <nergy equation, and an equation of state( he continuity equation enforces the conservation of mass principle( Simply stated, it requires that the difference in mass flo# into and out of a section of pipe line is equal to the rate of change of mass #ithin the section( his can be e%pressed mathematically by the relation3
he momentum equation describes the force balance on the fluid #ithin a section of pipe line( It requires that any unbalanced force result in an acceleration of the fluid element( In mathematical form, this is3
he energy equation states that the difference in the energy flo# into and out of a section equals the rate of change of energy #ithin the section( he equation is3
hese three are the basic one dimensional pipeflo# equations are present in one form or another in all transient pipe models( .hat is needed to solve them, ho#ever, is a relation bet#een the pressure, density, and temperature for the fluid - an equation of state( he state equation depends on the type of fluid being modeled, as no one equation fully describes the variety of products that are shipped in pipe lines( Some of the forms in use include a bulk modulus type of relation of the form3
his is normally used for liquids that can be regarded as incompressible( he bulk modulus B and the thermal e%pansion coefficient a can be constant or functions of temperature and8or pressure depending on the application( @or light hydrocarbon gases a basic equation such as P B r % ? % C % is appropriate, #here C 6the compressibility7 is a kno#n function of temperature and pressure( @or reasonable ranges of temperature and pressure a function of the form3
may be adequate( @or conditions #here fluids are transported at or near the critical point, a more sophisticated correlation is required to obtain the required accuracy but there is still a large uncertainty in the true density under these operating conditions and they should be avoided #herever possible( +any real time systems have been installed on lines carrying ethylene, butane, propane, and )*A8)PA products and have used the Benedict-.ebb-?ubin correlation as modified by Starling 6commonly called the B.?S equation7 #ith reasonable results( <ernatively, tables such as *D-19 or special correlations such as the *BS ethylene equation can be used, but this increases the comple%ity of the programming( & further complication arises if the product is not uniform throughout the system( his can occur due to batching of fluids or from varying inlet conditions( he first is more common #here different products are shipped in a common line( he properties are essentially discontinuous across the interface of the t#o fluids, but can be considered as uniform #ithin batches( he basic problem here is to keep tack of the location of the interface( Systems #ith continuous variations in inlet conditions occur in both liquid and gas systems( he variations can result from mi%ing of fluids of slightly different composition or from large variations in supply conditions( he governing equations presented are non-linear partial differential equations #hich are not suitable for machine computation( hey have to be solved by implicit or e%plicit finite difference
techniques or the method or characteristics( ,f these the implicit method seems the most appropriate, as the other t#o methods could give rise to mathematical instabilities if the #rong timestep or distance interval is used( In order for the transient pipe line model to compute the state along the line at the end of each time interval, a set of initial conditions and a set of boundary conditions are required( he initial conditions specify the state at the beginning of the time interval and are normally the last set of data from the model( & steady state model must be used to generate an initial state #hen the model is started from rest, a so-called 4cold start(4 In this case a period of time must elapse before the pipe line model truly represents the actual state of the line( his time period allo#s any transient conditions present and not represented by the steady state model to die out( Aenerally, less compressible systems #ill cold start faster than the more compressible ones, but the actual time for the transient model to be activated depends on the application( his may typically be on the order of E; minutes for a gas pipe line( he boundary conditions required by the model are taken from measured data along the line( @or each point #here fluids enter the system, its temperature, fluid type or composition, and either a flo# or a pressure is required( @or any equipment in the system that affects or controls the line, a suitable boundary condition must also be given( @or a gas compressor for e%ample, either its suction pressure, flo# rate or discharge pressure must be specified( &dditional measurements are used by the applications modules, generally by comparing their values to the corresponding model calculations(
%mpact of %nstrument Accuracy
he performance of real time pipe line monitoring systems is limited primarily by the accuracy of the instrumentation installed on the line( o estimate the performance of the leak detection and location routines, it is important to under the effect of measurement uncertainty on the model and the real time applications module( +easurement uncertainty is composed of bias and random components( he first is usually a fi%ed error bet#een the indicated and true values but this could change #ith time as components #ear( he second is temporal and possibly spatial fluctuation of the output about its mean value( @ortunately there are techniques that can be employed #ithin the soft#are to largely mitigate the effects of bias, but care should be e%ercised as this is not al#ays the case( )eak detection and location both use differences bet#een the measured values and the modeled values to discern leak characteristics( he measured values could contain both bias and8or random errors as discussed( he model values, because they are driven by the measured values as boundary conditions, also include error terms #hich are less obvious( Because of these errors the difference in the model and measured values, or leak indicators, #ill not normally be -ero but #ill fluctuate about some non--ero mean( his mean value determined by observation during periods
#hen no leak is present and is attributed to bias errors in the measuring system( By subtracting this from the leak indicators, the bias component can be eliminated( he leak detection then #ill be a function of the measurement precision errors( Problems #ith this technique arise #hen dealing #ith pipe lines #hose operations change substantially from time to time( &n e%ample #ould be a liquid line operating intermittently( By monitoring differences in this #ay, the instruments contributing to the error are not identified( Because fluid flo# in a pipe line is governed by highly non-linear relationships, fi%ed errors in the boundary conditions can cause variable differences in the leak indicators #hen the pipe line operation changes( &s an e%ample, consider a steady state pipe line that is driven by pressure difference bet#een upstream and do#nstream boundary points( he leak indicator is the difference in the measured and modeled flo#( 6*ote that a transient model #ould have a flo# difference at each end of the pipe line(7 he true value of the data for this pipe line is an upstream pressure of 1,;;; units #ith a pressure drop of 1;; units for a flo# of 1;; units( he model of this system is then described by the equation3 P1 - P/ B ;(;1 F/ If a 12 pressure error is introduced into the upstream pressure, the modeled flo# becomes 1;> and a difference of > units bet#een measured and modeled flo# #ould sho# up in the leak indicators( &s long as the flo# stays near this value, the error in the leak indicator #ill remain nearly constant( @or instance, an actual flo# of :; units #ould result in a modeled flo# of :G units( hus the effects of the bias error in the pressure measurement can be substantially mitigated by subtracting > units from the leak indicator( his is termed the leak indicator 4offset4( *o# assume the line is shut do#n( he 1; unit pressure measurement error causes the flo# to compute a flo# of E1 units( &fter applying the offset, the value of the leak indicator still remains at />( In general, a line that undergoes large and rapid changes in operation #ill be affected by instrument bias errors( he more common case of lines that operate #ithin relatively narro# bounds, or that change operations slo#ly so that the offset can be automatically ad=usted, #ill only be affected by the precision error of the measurements( +easurement errors impact leak detection by limiting the si-e of the leak that can be detected by the monitoring system( he problem comes in finding a threshold value for each alarm in the system( @or a simple system that operates #ithin narro# bounds, this can be as simple as the offset previously discussed( he values of the leak indicators 6no# after the offset has been removed7 can be observed during normal operation and the appropriate alarm values set( @or more complicated systems, the thresholds can be set in a more rigorous #ay( he pipe line hydraulic equations can be used to determine the sensitivity of each leak indicator to the measurement at each boundary point( he error, #hether bias or precision, of each boundary point can be estimated from kno#ledge of the transducers and the data gathering equipment installed on the pipe line( he error in the boundary instruments times the sensitivity of the leak indicator to the boundary point #ill give the threshold required to prevent normal noise in the measurement value from being interpreted as a leak( !sing the root sum square as a result of combining the threshold for each boundary point #ith the error for the leak indicator's comparison measurement, a threshold for the leak indicator can be calculated on-line( his 4auto tuning4 of the leak indicator is found in the more advanced systems commercially available(
&s an e%ample, consider the steady state pipe line used in the earlier paragraph( he sensitivity of the modeled flo# to the upstream and do#nstream pressures area3 dF 8 dP1 B -6dF 8 dP/7 B >; 8 F hus, at a flo# of 1;; units, the sensitivity of the flo# to either pressure #ould be ;(>, #ith a decrease in the do#nstream pressure being equal to an increase in the upstream pressure( @or a 12, or 1; unit error in the pressure, a > unit error in the flo# #ould be e%pected, #hich is consistent #ith the previous results, he threshold required for this system #ould then be3
his sho#s that the required threshold for the system increases #ith decreasing flo#, #ith leak detection being impossible at -ero flo#( he overly stringent requirement at -ero flo# is due to the simplified model used( ,ther than that, the results are representative of the manner in #hich the leak thresholds must be ad=usted #hen large flo# variations occur in a pipe line( +easurement uncertainty affects leak location by increasing the uncertainty in the calculated leak position( )eaks are located in a line by discovering #here a leak of a given si-e #ould need to be located to best match the observed discrepancies in the leak indicators( Consider our steady state model again( If there is a leak of /; units half#ay do#n the line so that the flo# is 1/; units before and 1;; after the leak, then the pressure drop #ould be 1// units( his pressure drop #ould correspond to a modeled flo# of 11;( his #ould result in discrepancies in the leak indicator at both ends of the pipe at 1;( he leak location for a pressure-pressure boundary condition is given by3 D B ) % 6F/ 8 F1 H F/7 @or the condition given above, the correct location of half the pipe length is obtained( If, ho#ever the flo#meters have a / unit error such that the upstream leak indicator is 1/ and the do#nstream indicator is :, the leak is located 0;2 do#n the pipe, and not half#ay( In a 1;; mile line this is an error of 1; miles #hich is very significant( .hen evaluating the effect of measurement uncertainty on leak detection or location, it is useful to compare the uncertainty to the magnitude of the hydraulic events of interest( hus a pressure transducer that is 12 accurate over a 1,>;; psi span is only E;2 accurate #hen the pressure drop bet#een t#o closely spaced valve sites if >; psi( his is because the flo# is governed by pressure differences and not absolute pressures( )ike#ise a flo# meter that is /2 accurate in comparison to its span is I;2 accurate for si-ing a E2 leak(
System Design Aspects and 'uidelines
he availability of leak alarm uptime depends heavily on the system design and the choice of hard#are( Aenerally the more comple% the system, the greater the risk of leak indicator loss, but the more accurate the location of the leaks( Design is therefore a compromise bet#een cost, performance, and reliability( Consider a section of line sho#n in @igure E( @or the three simple alarms of flo# imbalance, pressure imbalance, and acoustic alarms bet#een stations & and B or B and C, the follo#ing components are needed3 +ass flo#3 / flo#meters, / pressure sensors, / temperature sensors, / ? !s, / communication links, 1 computer 611 elements7 Pressure3 0 pressure sensors, 0 ? !s, 0 communications links, and 1 computer 61E elements7 &coustic3 / acoustic monitors, / ? !s, / communications links, and 1 computer 6I elements7
Fi#ure %. Schematic overvie# of pipe line computer-based monitoring system( If combined hybrid alarms of flo#8acoustic, flo#8pressure, or pressure8acoustic are used then the number of components in the chain is increased( By summing the component availabilities for each element, an uptake for each alarm can be estimated( @or e%ample, a flo#meter #ith a failure of once in E years #ith a repair time of 0 hours has an availability factor of ;(999:> 6or 99(9:>27( @rom such an analysis, the conclusion can be dra#n that the system should be made as simple as possible or instruments should be installed in duplicate to ma%imi-e alarm uptake( Section 9 also sho#ed that instruments should be selected on performance and not on economic grounds( It is better to install fe#er high performance instruments than numerous poor ones( Digital conversion also requires attention( ,ften 1/ or 10 bit conversion is required to give the necessary accuracy of data processing and usually double precision computation is also required( he use of standard outputs should be made #herever possible( Custom designed electronics invariably lead to problems( he best guideline, ho#ever, is to seek users of pipe line monitoring systems to ask their advice and e%perience #ith instruments and system components( Independent validation of
all information should be made #herever possible( Companies that supply such complete systems usually have a client list and it is #orth spending time talking to these clients before the final design specification is fi%ed( .ith regard to instrumentation, flo#meters #ith the highest accuracy are required for mass balance functions( Suitable types include turbine and displacement meters #ith pulse outputs( ,rifice meters are not really suitable, since the best accuracy that can be obtained from a #ell maintained system is around J12 of full scale( he correct choice of turbine by comparison is J;(/>2 of reading or better(
De#elopment of Pipe Line Monitoring Systems
he speed of instrumentation development generally is rather frightening( he impact of microelectronics is still being felt some ten years after they first appeared, and ne# and improved transducers #ith on-board 4intelligence4 are being sold in increasing numbers( &t the same time the si-e of computers is decreasing and the computating capability is increasing( Soft#are is also advancing rapidly and the performance of modern flo# monitoring systems is becoming dependent on the accuracy of the modeling equations( <quations of state and the behavior of hydrocarbon mi%tures are not particularly advanced or #ell understood, and fundamental research is required before the ne%t advance in this type of technology can proceed( &ll of these points indicate that computer-based monitoring systems #ill become the standard technique of operating and controlling pipe lines in the future( Control algorithms can be integrated #ith the applications modules to produce a semi-intelligent complete integrity monitoring scheme( &s e%perience in the design and operation of such systems gro#s they #ill be applied #ith increasing confidence( @uture systems #ill use a combination of the ne# technology discussed( his could include internal monitoring pigs and advanced pipe line models, both run from a central control room( hus the internal and e%ternal state of the line could be checked simultaneously( Such technology can enable safer and more economic operation of pipe lines to be carried out( Conc&usion he paper cannot do =ustice in such a short space, to the comple% and diverse sub=ect of leak detection( Such systems have been in operation in many forms all over the #orld, but it is only recently that environmental as #ell as economic factors have influenced their development( +odern digital systems are transforming operation and design, #ith many parallel functions being possible #ith a single system( here is no# a clear need to ensure complete integration of all components in the system to guarantee safer and more accurate pipe line management( Instrument selection is critical, as is the need to develop better thermodynamic models, for the ne%t generation of systems to become more reliable and accurate( Re'erences
1( @urness, ?( &(, 4+odern Pipe )ine +onitoring echniques,4 Pipes and Pipelines International, +ay-Kune, I-11 and September-,ctober, 10-1:, 19:>( /( +aillou%, ?( )( and van ?eet, K( D(, 4?eal ime ransient @lo# +odeling &pplications,4 PSIA &nnual +eeting , 19:E( E( Covington, +( (, 4Pipe )ine ?upture Detection and Control,4 &S+< Paper :-P< ->0, 19I:( 0( Bernard, A( A(, 4<nergy Balance Derivation,4 C?C Bethany Internal ?eport, 19:/( >( &non, 4Pinpointing pigs in pipe lines,4 R & D Digest, :, 10-1>, British Aas Corporation, 19:G( G( @u=imori, *( and Sugaya, S(, 4& study of a leak detection based on in-out flo# difference method,4 Proceedings of I+<", Symposium on @lo# +easurement and Control, pp( /;>/;9, okyo, Kapan, 19I9( I( Ste#art, ( )(, 4,perating e%perience using a computer model for pipe line leak detection,4 Journal of Pipelines, E, /EE-/EI, 19:E( (otation & - cross-sectional area B - fluid bulk modulus c - specific heat at constant volume D - pipe line diameter <P1 - uncertainty in upstream pressure measurement <P/ - uncertainty in do#nstream pressure measurement <F - uncertainty in flo# measurement g - gravitational acceleration h - elevation ) - pipe length P - pressure P; - reference or base pressure P1 - section upstream pressure P/ - pipe line flo# rate F - pipe line flo# rate F1 - upstream measured vs( modeled flo# discrepancy F/ - do#nstream measured vs( modeled flo# discrepancy ? - gas constant - temperature g - ground temperature ) - leak detection threshold level o - reference or base temperature t - time ! - heat transfer coefficient 5 - velocity D - leak location % - incremental distance along the pipe line y - C factor correlation coefficient C - gas compressibility factor alpha - coefficient of thermal e%pansion p - density po - reference or base density
Pipe Line Leak Detection Techniques
• • • • • • • • • •
Causes and economical aspects of leaks Simple leak detection systems Pig-based monitoring systems Computer-based monitoring systems Pipe line leak phenomena Background philosophy of pipe line modeling Basic pipe line modeling equations Impact of instrument accuracy System design aspects and guidelines Development of pipe line monitoring systems
Pipe Line Leak Detection Techniques
R. A. Furness, Cranfield Institute of echnology, Cranfield, Bedford, !", and J. D. van Rett, Scientific Soft#are Intercomp, $ouston, e%as, !S& Summar Pipe lines are no# an integral part of the #orld's economic structure and literally billions of dollars #orth of products are no# moved annually in pipe lines( Both economic and environmental factors are influential in pipe line operation, and therefore integrity monitoring is vitally important in the control and operation of comple% systems( )eak detection systems range from simple, visual and #alking and checking to comple% arrangements of hard#are and soft#are( *o one method is universally applicable and operating requirements dictate #hich method is the most cost effective( he aim of the paper is to revie# the basic techniques of leak detection that are currently in use( he advantages and disadvantages of each method are discussed and some indications of applicability are outlined( +odern pipe line computer modeling and control is then revie#ed in detail( hese systems are the most fle%ible and versatile and are steadily being adopted( he influence of instrument accuracy on system design and performance is discussed and the basic modeling equations are revie#ed( !ntro"uction ,ur need to transport fluids from the point of production to the area of end use has led to a rapid increase in the number of pipe lines being designed and constructed( +any of these carry to%ic or ha-ardous products, often close to centers of high population or through areas of high environmental sensitivity( .ith the need to safeguard these lines, on-line monitoring is becoming routine and in some cases /0 hour surveillance is mandatory( .ith the increase in #orld terrorism, the need for rapid and reliable pipe line measurement and control systems #ill increase( he revie# begins #ith a summary of causes of leaks and the implications of the failure to detect them( Basic techniques are covered and the features of each are briefly discussed( he bulk of the paper deals #ith modern computer-based techniques( Basic flo# equations are covered and the online dynamic calculations required are listed together #ith the input data required to enable the monitoring to be accomplished #ith the minimum of do#ntime( he latest systems are capable of resolving do#n to 12 of the ma%imum rated flo# #ith a response time of a fe# minutes( Practical e%perience verifies this figure but the total costs of such a system could be high( he system therefore becomes a compromise bet#een response, performance, and alarm availability, and choice of instrumentation is crucial( he integration of good quality instruments #ith advanced real time models seems to be the current trend and the paper closes #ith some personal thoughts on future trends(
Causes and Economic Aspects of Leaks
here are four main categories of pipe line failures( hese are3
• • • •
Pipe line corrosion and #ear ,peration outside design limits !nintentional third party damage Intentional damage
+any pipe lines are operated for a number of years #ith no regard to any possible mechanical changes occurring in the line( Some of the products may be corrosive, the pipe line may be left partially full for periods of time, or atmospheric effects may cause e%ternal damage( hese three reasons are responsible for pipe line corrosion and this may give rise to corrosion 4pits4 developing along the line( hese are small in nature and could be responsible for material imbalances over a period of time( 5ery accurate flo# metering can be used to detect this as discussed in the ne%t section( &brasive fluids or dust-laden gas streams can give rise to pipe line #ear( &gain, this is a slo# process, but should a #eak spot develop 6more often than not close to a change in direction or section7 then a pipe break may occur very rapidly and totally une%pectedly( ,peration outside design guidelines is more common than is reali-ed, as operators seek to use the line for as many fluids as possible( If the line is designed for a certain ma%imum temperature and pressure, then operation at higher pressure and8or higher temperature could lead to spontaneous failure( he problem could be compounded if the line has a large but unkno#n amount of corrosion( !nintentional third party damage may occur if e%cavation or building occurs near buried lines( +ore often than not the right-of-#ays are not clearly marked and lines are sometimes broken by bulldo-ers or similar plant machinery possibly #ith fatal results( &n e%ample occurred in .est 5irginia in 19:0 #hen a /;-in( natural gas line #as punctured by an e%cavator and product leaked slo#ly into a nearby supermarket during the night( he building #as totally destroyed early ne%t morning by a massive e%plosion caused by a staff member lighting a cigarette( Such an occurrence could have been avoided by some form of leak detection on the line( Intentional damage unfortunately is on the increase and pipe lines carrying flammable or high value products make ideal targets( &larm systems linked to block valves can help to minimi-e the amount of product released as a result of sabotage, so again certain lines are instrumented #ith the intention of reducing the effects of planned terrorism( he costs of failure to detect leaks also fall into four main areas3
• • • •
)oss of life and property Direct cost of lost product and line do#ntime <nvironmental cleanup costs Possible fines and legal suits
hese are all self e%planatory, #ith the most costly of these being the last, although any of the four areas could be very e%pensive( he si-e of claims can run into many millions of dollars, so the cost of fitting and operating leak detection systems is often insignificant compared #ith the costs of failure of the line( It is this background that is causing operators and designers to turn to on-line integrity monitoring systems and the ne%t section of the paper looks at the more basic methods(
Simple Leak Detection Systems
he most basic method of leak detection involves either #alking, driving, or flying the pipe line right-of-#ay to look for evidence of discoloration of vegetation near the line or actually hear or see the leak( ,ften 4unofficial4 pipe line monitoring is performed by people living nearby #ho can inform the operator of a problem #ith the line( he most cost effective #ay to detect leaks in non-flammable products is to simply add an odorant to the fluid( his requires some care in selection, as frequently the odorant has to be removed before the transported fluid can be used( ,rganic compound make the most useful odori-ers, especially #hen the fluid being carried has no natural smell of its o#n( & good e%ample is carbon mono%ide, a highly to%ic but odorless gas #hich is often pumped in large quantities bet#een chemical plants( Chemicals such as mercaptans 6rotten egg smell7 or rimethylamine 6rotten fish smell7 can be added in small quantities to enable any leak to be located by smell( he disadvantage of such a method is that if the leak occurs in an area of no population the leak #ill go undetected unless the line is #alked regularly by pipe line surveillance cre#s carrying suitable 4sniffer4 detectors( hus to the apparent lo# costs of this method have to be added the costs of removing the odorant and maintain staff to check the line at frequent intervals( he location of a leak is also dependent on prevailing #eather conditions( Strong #inds may disperse the smell and atmospheric inversions may given an incorrect location of the leak and the uncertainty of relying on this method alone is high( *evertheless, it is a useful method if used in con=unction #ith other techniques( Simple line flo# balances are frequently used to check for gross imbalances over hourly or daily bases( his method may identify that a leak is present but flo#meters at each end of the line #ill not identify the leak location( & line pressure measurement system #ill be required in con=unction #ith the flo#meters to establish that the pressure gradient has changed from the no-leak situation( he method is useful, ho#ever, in identifying the e%istence of corrosion pits as the outputs of the flo#meters at each end of the line #ill consistently diverge if flo# in the line is maintained constant( If line flo# rate varies #ith time, then imbalances are more difficult to detect, since the flo#meter outputs may vary nonlinearly #ith flo# rate or may have different flo# characteristics from each other( & loss of product #ill be identified simply as the different bet#een the steady state inventory of the system and the instantaneous inlet and outlet flo#s( +athematically this is3
his last term can be calculated as the average of the integrated inlet and outlet flo#s in simple systems, but as #ill be seen later in this paper, the value of this term can be calculated more accurately and easily in real time as a function of several variables(
¬her method is based on detecting the noise associated #ith or generated by a leak( here are many instances #here fluid flo# can generate vibrations at frequencies in e%cess of /; k$-( hese frequencies are in the ultrasonic range but can be made portable so that pipe line cre#s can clamp a transducer at any point along the line to check for noise( By noting the signal strength, the source of the leak can be pinpointed( & similar technique, though based on a different principle, is the acoustic 4#avealert4 monitor, more correctly called a negative pressure #ave detector( his is a pie-oelectric sensor that gives an output #hen dynamically stressed( .hen a leak occurs there is a sudden drop in pressure at the leak follo#ed by rapid line repressuri-ation a fe# milliseconds later( he lo# pressure #ave moves a#ay from the leak in both directions at the speed of sound( he pipe #alls act as a #aveguide so that this rarefaction #ave can travel for great distances, attenuating in amplitude as it does so( Sensors placed at distances along the line can be triggered as the #ave passes and the location of the leak can be calculated from the line conditions and the internal timing devices in the instrument( Such devices are particularly useful in identifying large breaks in lines very rapidly, since the transient #ave typically moves 1 mile in > seconds in gases and almost 1 miles per second in liquids( ?esponse is therefore on the order of a fe# seconds depending on the positioning of the transducers( @igure 1 sho#s one adaptation #here the instrument can be made to cancel line noise and use the full measuring capability of the sensor for signal detection( here is the problem setting the background threshold correctly, as this may be affected by the location of the instrument in relation to bends, valves, pumps, regulators, etc( <%perience has sho#n that the installation is also critical to reliable performance, and there is also a dependence on the ?eynolds number of the flo#( Both of these affect the number of 4false alarms4 from the instrument(
Fi#ure 1. *egative pressure #ave detector 6acoustic monitor7
Pig Based Monitoring Systems
Pipe line pigs are frequently used for pipe line commissioning, cleaning, filling, de#a%ing, batching, and more recently pipe line monitoring( his last type of pig can be designed to carry a #ide range of surveillance and monitoring equipment and can be used at regular intervals to check internal conditions rather than continuously monitoring the line( Data, ho#ever, can be built up over a period of time to provide a history of the line( his information can be used to predict or estimate #hen maintenance, line cleaning, or repairs are required( If a leak is detected, for e%ample, by flo# meter imbalance, the location can be found by using a pig #ith acoustic equipment on board( his #ill alarm #hen the detection equipment output reaches a ma%imum and the precise location of the pig can be confirmed by radio transmitters also mounted on board( Pigs require tracking because they may become stuck, at a point of debris build-up, for e%ample( Pigging should be carried out at a steady speed, but occasionally the pig may stop and start, particularly in smaller lines( Information on #hen and #here the pig stops is therefore important in interpreting the inspection records( Pig tracking is not ne# and many such proprietary systems e%ist( In the best systems, ho#ever, a picture of the line is often programmed in so that outputs from =unctions, valves, cross-overs, and other geometries act as an aid to location( Pig tracking can make use of the acoustic methods discussed earlier( .hen the sealing cups at the front of the pig encounter a #eld, vibrational or acoustic signals are generated( <ach pipe line therefore has its characteristics sound pattern( #hen a crack occurs this pattern changes form the no-leak case and the location can be found from direct comparison( he technology has become so advanced that information on dents, buckles, ovality, #eld penetration, e%pansion, and pipe line footage can be generated( he equipment is often simple, consisting of sensor, conditioning, and amplifier circuits and suitable output and recording devices( Such a device developed by British Aas is sho#n in @igure /( he range of detection is dependent on the pipe line diameter and the type of pig( ,perational data have sho#n that light pigs in a /;; mm line can be detected at a range of : km, increasing to :; km for a heavy pig in a 9;; mm line( &s the signals travel at acoustic velocity this means a signal from a pig at :; km range #ill take 19; seconds to be picked up( Such technology is no# becoming routine in both offshore gas and onshore liquid lines(
Fi#ure $. 4Intelligent4 pipe line monitoring pig( Courtesy British Aas Corp(
Computer !ased Monitoring Systems
It is in computer-based systems that the greatest amount of data can be gathered, processed, analy-ed, and acted upon in the shortest period of time( Programs can calculate the inventory of the line at any time and compare this #ith accurate measurements at any section in the system( he effects of pressure and temperature on line dimensions, for e%ample, can be calculated to provide an accurate estimate of the mass of fluid in the line( Data from a #ide range of instruments can be transmitted by telemetry, radio, or phone links to a central computer #hich monitors the 4health4 of the line continuously( By changing programs and subroutines, a vast amount of functions and tasks can be accomplished very easily and cost effectively( he many functions that can be performed by computer-based systems include not only leak detection but also3
• • • • •
Pig tracking Batch tracking of fluids Inventory accounting ,n-line flo# compensation Instrument data and malfunction checking, etc(
Such systems have very rapid response and have the advantage of multiple inputs being re#ired before leaks are declared( hus some systems can run for short outage periods #ith no loss of integrity( hey are often comple% and costly to install but once the initial capital investment has been made running costs are lo#( he first section in the detailed revie# of such systems looks at the phenomena that need to be modeled #hen a leak occurs(
Pipe Line Leak Phenomena
.hen a leak occurs in a pipe line the measured pressure do#nstream of the leak falls but the pressure at the same location is predicted to rise( he first is not difficult to understand as the line is depressuri-ing as mass leaves through the leak( he second effect can be e%plained as follo#s( he equations presented later in the paper predict pressure based on measured flo# or flo# based on measured pressure( &s mass leaves the system through the leak hole, a reduced flo# at the do#nstream end is compared to the inlet flo#( he may not have chanted and so to balance the system, the equations predict a do#nstream presser rise( In physical terms the model thinks the line is 4packing4 and total system inventory is increasing( here is therefore a divergence bet#een measured and modeled pressure( he same is true of flo# changes, but here the inlet flo# could increase due to lo#er pipe flo# resistance bet#een meter and leak #hile the section outlet flo# #ill fall as mass leave through the leak instead of passing through the meter( hus a real imbalance #ill result( he model ho#ever #ill sho# an inconsistency since the pressure comparison #ill indicate line packing and the flo# comparison a line unpacking( If selected pressure and flo# imbalance limits are e%ceeded than leak is declared( he magnitude of the leak is predicted from the flo# imbalance and the location from the pressure profile imbalance and the flo# leak indicators( he impact of instrument accuracy on the predicted location is discussed later in the paper( It is vitally important to good leak si-ing and location to have the best pipe line instrumentation possible to minimi-e uncertainty(
Background Philosophy of Pipe Line Modeling
?eal time modeling is a technique that uses the full data gathering capabilities of modern digital systems and the computational po#er of small computers to give accurate 4snapshots4 of the pipe line( he #hole system is under the control of a SC&D& package of programs, #hich poll the data stations on the line, process the data, control the running of the transient pipe line model and activate the alarm and leak location routines( In addition to these basic soft#are modules, more comple% systems might include a predictive model to analy-e 4#hat if4 operating scenarios, provide an optimi-ation route for least-cost operating strategies or include a separate man8machine interface for the model system( he SC&D& interface is responsible for acquiring the data from the SC&D& system and relating them to the model representation of the line( &s a point in the system #here t#o large and independently developed systems =oin, this can be the source of many problems in the implementation of the real time modeling( ,nce the measurement data have been obtained, noise filtering and plausibility checking can be performed prior to running of the model( he model is the mathematical representation of the pipe line and #ill include such features as elevation data, diameters, valve and pump locations, changes of direction and the location or cross-overs and =unctions( he model provides data on the flo# conditions #ithin the line at intervals bet#een seconds and minutes, depending on operational needs( .ith the data available from both the measurement system and the pipe line model, the real time applications modules are run( hese are the leak detection and location routines in the conte%t of integrity monitoring( he leak detection module functions by computing the difference bet#een the modeled flo#s and pressures and the measured values at tall points #here measurements not already used as boundary conditions are located( Because the model accounts for normal transient operations, these differences #ill be small under normal conditions( .hen a leak is present, the differences become larger since the model system does not account for leakage( .hen these differences e%ceed preselected values, a leak alarm is declared( Sophisticated voting schemes #hich require multiple leak indicators to be in alarm for several consecutive time intervals are used to reduce false alarms #hile maintaining lo# thresholds( ,ften a simple pipe line balance of the type discussed earlier is used as a back-up to verify the transient model( ?esponse characteristics are, ho#ever, much slo#er than the real time model( ,nce the leak detection module declares a leak, the location routine is activated( he location is calculated from the magnitude and distribution of the leak indicators( &s an e%ample, in a straight pipe line #ith an upstream flo# discrepancy and a do#nstream pressure discrepancy as leak indicators, it is an easy calculation to determine #here the leak must be such that the leak flo# #hen added to the modeled flo# #ill produce the additional pressure drop observed at the do#nstream end( Solutions for pipe net#orks are more complicated and unique locations do not al#ays e%ist( his might be the case #ith parallel looped lines, for e%ample( In this case all the calculated locations should be checked(
The components of the real time modeling system "ork together to reduce the large #olumes of ra" data from the data acquisition system to a much smaller num!er of parameters and alarms that are more meaningful to pipe line operations$ %n the case of integrity monitoring& this means leak e#ents that could not !e detected !y inspection of the measured data can !e found and isolated quickly and relia!ly$ Basic Pipe Line Modeling Equations
he transient pipe line flo# model is the heart of a pipe line modeling system( he model computes the state of the pipe line at each time interval for #hich data are available( he state of the pipe line is defined as a set of pressures, temperatures, flo#s, and densities that described the fluids being transported at all points #ithin the system( hese quantities are found as the solution to a set of equations #hich describe the behavior of the pipe line system( hese basic equations are the continuity equation, the +omentum equation, the <nergy equation, and an equation of state( he continuity equation enforces the conservation of mass principle( Simply stated, it requires that the difference in mass flo# into and out of a section of pipe line is equal to the rate of change of mass #ithin the section( his can be e%pressed mathematically by the relation3
he momentum equation describes the force balance on the fluid #ithin a section of pipe line( It requires that any unbalanced force result in an acceleration of the fluid element( In mathematical form, this is3
he energy equation states that the difference in the energy flo# into and out of a section equals the rate of change of energy #ithin the section( he equation is3
hese three are the basic one dimensional pipeflo# equations are present in one form or another in all transient pipe models( .hat is needed to solve them, ho#ever, is a relation bet#een the pressure, density, and temperature for the fluid - an equation of state(
he state equation depends on the type of fluid being modeled, as no one equation fully describes the variety of products that are shipped in pipe lines( Some of the forms in use include a bulk modulus type of relation of the form3
his is normally used for liquids that can be regarded as incompressible( he bulk modulus B and the thermal e%pansion coefficient a can be constant or functions of temperature and8or pressure depending on the application( @or light hydrocarbon gases a basic equation such as P B r % ? % C % is appropriate, #here C 6the compressibility7 is a kno#n function of temperature and pressure( @or reasonable ranges of temperature and pressure a function of the form3
may be adequate( @or conditions #here fluids are transported at or near the critical point, a more sophisticated correlation is required to obtain the required accuracy but there is still a large uncertainty in the true density under these operating conditions and they should be avoided #herever possible( +any real time systems have been installed on lines carrying ethylene, butane, propane, and )*A8)PA products and have used the Benedict-.ebb-?ubin correlation as modified by Starling 6commonly called the B.?S equation7 #ith reasonable results( <ernatively, tables such as *D-19 or special correlations such as the *BS ethylene equation can be used, but this increases the comple%ity of the programming( & further complication arises if the product is not uniform throughout the system( his can occur due to batching of fluids or from varying inlet conditions( he first is more common #here different products are shipped in a common line( he properties are essentially discontinuous across the interface of the t#o fluids, but can be considered as uniform #ithin batches( he basic problem here is to keep tack of the location of the interface( Systems #ith continuous variations in inlet conditions occur in both liquid and gas systems( he variations can result from mi%ing of fluids of slightly different composition or from large variations in supply conditions( he governing equations presented are non-linear partial differential equations #hich are not suitable for machine computation( hey have to be solved by implicit or e%plicit finite difference techniques or the method or characteristics( ,f these the implicit method seems the most appropriate, as the other t#o methods could give rise to mathematical instabilities if the #rong timestep or distance interval is used( In order for the transient pipe line model to compute the state along the line at the end of each time interval, a set of initial conditions and a set of boundary conditions are required( he initial conditions specify the state at the beginning of the time interval and are normally the last set of data from the model( & steady state model must be used to generate an initial state #hen the model is started from rest, a so-called 4cold start(4 In this case a period of time must elapse before the pipe line model truly represents the actual state of the line( his time period allo#s any transient
conditions present and not represented by the steady state model to die out( Aenerally, less compressible systems #ill cold start faster than the more compressible ones, but the actual time for the transient model to be activated depends on the application( his may typically be on the order of E; minutes for a gas pipe line( he boundary conditions required by the model are taken from measured data along the line( @or each point #here fluids enter the system, its temperature, fluid type or composition, and either a flo# or a pressure is required( @or any equipment in the system that affects or controls the line, a suitable boundary condition must also be given( @or a gas compressor for e%ample, either its suction pressure, flo# rate or discharge pressure must be specified( &dditional measurements are used by the applications modules, generally by comparing their values to the corresponding model calculations(
Basic Pipe Line Modeling Equations
he transient pipe line flo# model is the heart of a pipe line modeling system( he model computes the state of the pipe line at each time interval for #hich data are available( he state of the pipe line is defined as a set of pressures, temperatures, flo#s, and densities that described the fluids being transported at all points #ithin the system( hese quantities are found as the solution to a set of equations #hich describe the behavior of the pipe line system( hese basic equations are the continuity equation, the +omentum equation, the <nergy equation, and an equation of state( he continuity equation enforces the conservation of mass principle( Simply stated, it requires that the difference in mass flo# into and out of a section of pipe line is equal to the rate of change of mass #ithin the section( his can be e%pressed mathematically by the relation3
he momentum equation describes the force balance on the fluid #ithin a section of pipe line( It requires that any unbalanced force result in an acceleration of the fluid element( In mathematical form, this is3
he energy equation states that the difference in the energy flo# into and out of a section equals the rate of change of energy #ithin the section( he equation is3
hese three are the basic one dimensional pipeflo# equations are present in one form or another in all transient pipe models( .hat is needed to solve them, ho#ever, is a relation bet#een the pressure, density, and temperature for the fluid - an equation of state( he state equation depends on the type of fluid being modeled, as no one equation fully describes the variety of products that are shipped in pipe lines( Some of the forms in use include a bulk modulus type of relation of the form3
his is normally used for liquids that can be regarded as incompressible( he bulk modulus B and the thermal e%pansion coefficient a can be constant or functions of temperature and8or pressure depending on the application( @or light hydrocarbon gases a basic equation such as P B r % ? % C % is appropriate, #here C 6the compressibility7 is a kno#n function of temperature and pressure( @or reasonable ranges of temperature and pressure a function of the form3
may be adequate( @or conditions #here fluids are transported at or near the critical point, a more sophisticated correlation is required to obtain the required accuracy but there is still a large uncertainty in the true density under these operating conditions and they should be avoided #herever possible( +any real time systems have been installed on lines carrying ethylene, butane, propane, and )*A8)PA products and have used the Benedict-.ebb-?ubin correlation as modified by Starling 6commonly called the B.?S equation7 #ith reasonable results( <ernatively, tables such as *D-19 or special correlations such as the *BS ethylene equation can be used, but this increases the comple%ity of the programming( & further complication arises if the product is not uniform throughout the system( his can occur due to batching of fluids or from varying inlet conditions( he first is more common #here different products are shipped in a common line( he properties are essentially discontinuous across the interface of the t#o fluids, but can be considered as uniform #ithin batches( he basic problem here is to keep tack of the location of the interface( Systems #ith continuous variations in inlet conditions occur in both liquid and gas systems( he variations can result from mi%ing of fluids of slightly different composition or from large variations in supply conditions( he governing equations presented are non-linear partial differential equations #hich are not suitable for machine computation( hey have to be solved by implicit or e%plicit finite difference
techniques or the method or characteristics( ,f these the implicit method seems the most appropriate, as the other t#o methods could give rise to mathematical instabilities if the #rong timestep or distance interval is used( In order for the transient pipe line model to compute the state along the line at the end of each time interval, a set of initial conditions and a set of boundary conditions are required( he initial conditions specify the state at the beginning of the time interval and are normally the last set of data from the model( & steady state model must be used to generate an initial state #hen the model is started from rest, a so-called 4cold start(4 In this case a period of time must elapse before the pipe line model truly represents the actual state of the line( his time period allo#s any transient conditions present and not represented by the steady state model to die out( Aenerally, less compressible systems #ill cold start faster than the more compressible ones, but the actual time for the transient model to be activated depends on the application( his may typically be on the order of E; minutes for a gas pipe line( he boundary conditions required by the model are taken from measured data along the line( @or each point #here fluids enter the system, its temperature, fluid type or composition, and either a flo# or a pressure is required( @or any equipment in the system that affects or controls the line, a suitable boundary condition must also be given( @or a gas compressor for e%ample, either its suction pressure, flo# rate or discharge pressure must be specified( &dditional measurements are used by the applications modules, generally by comparing their values to the corresponding model calculations(
%mpact of %nstrument Accuracy
he performance of real time pipe line monitoring systems is limited primarily by the accuracy of the instrumentation installed on the line( o estimate the performance of the leak detection and location routines, it is important to under the effect of measurement uncertainty on the model and the real time applications module( +easurement uncertainty is composed of bias and random components( he first is usually a fi%ed error bet#een the indicated and true values but this could change #ith time as components #ear( he second is temporal and possibly spatial fluctuation of the output about its mean value( @ortunately there are techniques that can be employed #ithin the soft#are to largely mitigate the effects of bias, but care should be e%ercised as this is not al#ays the case( )eak detection and location both use differences bet#een the measured values and the modeled values to discern leak characteristics( he measured values could contain both bias and8or random errors as discussed( he model values, because they are driven by the measured values as boundary conditions, also include error terms #hich are less obvious( Because of these errors the difference in the model and measured values, or leak indicators, #ill not normally be -ero but #ill fluctuate about some non--ero mean( his mean value determined by observation during periods
#hen no leak is present and is attributed to bias errors in the measuring system( By subtracting this from the leak indicators, the bias component can be eliminated( he leak detection then #ill be a function of the measurement precision errors( Problems #ith this technique arise #hen dealing #ith pipe lines #hose operations change substantially from time to time( &n e%ample #ould be a liquid line operating intermittently( By monitoring differences in this #ay, the instruments contributing to the error are not identified( Because fluid flo# in a pipe line is governed by highly non-linear relationships, fi%ed errors in the boundary conditions can cause variable differences in the leak indicators #hen the pipe line operation changes( &s an e%ample, consider a steady state pipe line that is driven by pressure difference bet#een upstream and do#nstream boundary points( he leak indicator is the difference in the measured and modeled flo#( 6*ote that a transient model #ould have a flo# difference at each end of the pipe line(7 he true value of the data for this pipe line is an upstream pressure of 1,;;; units #ith a pressure drop of 1;; units for a flo# of 1;; units( he model of this system is then described by the equation3 P1 - P/ B ;(;1 F/ If a 12 pressure error is introduced into the upstream pressure, the modeled flo# becomes 1;> and a difference of > units bet#een measured and modeled flo# #ould sho# up in the leak indicators( &s long as the flo# stays near this value, the error in the leak indicator #ill remain nearly constant( @or instance, an actual flo# of :; units #ould result in a modeled flo# of :G units( hus the effects of the bias error in the pressure measurement can be substantially mitigated by subtracting > units from the leak indicator( his is termed the leak indicator 4offset4( *o# assume the line is shut do#n( he 1; unit pressure measurement error causes the flo# to compute a flo# of E1 units( &fter applying the offset, the value of the leak indicator still remains at />( In general, a line that undergoes large and rapid changes in operation #ill be affected by instrument bias errors( he more common case of lines that operate #ithin relatively narro# bounds, or that change operations slo#ly so that the offset can be automatically ad=usted, #ill only be affected by the precision error of the measurements( +easurement errors impact leak detection by limiting the si-e of the leak that can be detected by the monitoring system( he problem comes in finding a threshold value for each alarm in the system( @or a simple system that operates #ithin narro# bounds, this can be as simple as the offset previously discussed( he values of the leak indicators 6no# after the offset has been removed7 can be observed during normal operation and the appropriate alarm values set( @or more complicated systems, the thresholds can be set in a more rigorous #ay( he pipe line hydraulic equations can be used to determine the sensitivity of each leak indicator to the measurement at each boundary point( he error, #hether bias or precision, of each boundary point can be estimated from kno#ledge of the transducers and the data gathering equipment installed on the pipe line( he error in the boundary instruments times the sensitivity of the leak indicator to the boundary point #ill give the threshold required to prevent normal noise in the measurement value from being interpreted as a leak( !sing the root sum square as a result of combining the threshold for each boundary point #ith the error for the leak indicator's comparison measurement, a threshold for the leak indicator can be calculated on-line( his 4auto tuning4 of the leak indicator is found in the more advanced systems commercially available(
&s an e%ample, consider the steady state pipe line used in the earlier paragraph( he sensitivity of the modeled flo# to the upstream and do#nstream pressures area3 dF 8 dP1 B -6dF 8 dP/7 B >; 8 F hus, at a flo# of 1;; units, the sensitivity of the flo# to either pressure #ould be ;(>, #ith a decrease in the do#nstream pressure being equal to an increase in the upstream pressure( @or a 12, or 1; unit error in the pressure, a > unit error in the flo# #ould be e%pected, #hich is consistent #ith the previous results, he threshold required for this system #ould then be3
his sho#s that the required threshold for the system increases #ith decreasing flo#, #ith leak detection being impossible at -ero flo#( he overly stringent requirement at -ero flo# is due to the simplified model used( ,ther than that, the results are representative of the manner in #hich the leak thresholds must be ad=usted #hen large flo# variations occur in a pipe line( +easurement uncertainty affects leak location by increasing the uncertainty in the calculated leak position( )eaks are located in a line by discovering #here a leak of a given si-e #ould need to be located to best match the observed discrepancies in the leak indicators( Consider our steady state model again( If there is a leak of /; units half#ay do#n the line so that the flo# is 1/; units before and 1;; after the leak, then the pressure drop #ould be 1// units( his pressure drop #ould correspond to a modeled flo# of 11;( his #ould result in discrepancies in the leak indicator at both ends of the pipe at 1;( he leak location for a pressure-pressure boundary condition is given by3 D B ) % 6F/ 8 F1 H F/7 @or the condition given above, the correct location of half the pipe length is obtained( If, ho#ever the flo#meters have a / unit error such that the upstream leak indicator is 1/ and the do#nstream indicator is :, the leak is located 0;2 do#n the pipe, and not half#ay( In a 1;; mile line this is an error of 1; miles #hich is very significant( .hen evaluating the effect of measurement uncertainty on leak detection or location, it is useful to compare the uncertainty to the magnitude of the hydraulic events of interest( hus a pressure transducer that is 12 accurate over a 1,>;; psi span is only E;2 accurate #hen the pressure drop bet#een t#o closely spaced valve sites if >; psi( his is because the flo# is governed by pressure differences and not absolute pressures( )ike#ise a flo# meter that is /2 accurate in comparison to its span is I;2 accurate for si-ing a E2 leak(
System Design Aspects and 'uidelines
he availability of leak alarm uptime depends heavily on the system design and the choice of hard#are( Aenerally the more comple% the system, the greater the risk of leak indicator loss, but the more accurate the location of the leaks( Design is therefore a compromise bet#een cost, performance, and reliability( Consider a section of line sho#n in @igure E( @or the three simple alarms of flo# imbalance, pressure imbalance, and acoustic alarms bet#een stations & and B or B and C, the follo#ing components are needed3 +ass flo#3 / flo#meters, / pressure sensors, / temperature sensors, / ? !s, / communication links, 1 computer 611 elements7 Pressure3 0 pressure sensors, 0 ? !s, 0 communications links, and 1 computer 61E elements7 &coustic3 / acoustic monitors, / ? !s, / communications links, and 1 computer 6I elements7
Fi#ure %. Schematic overvie# of pipe line computer-based monitoring system( If combined hybrid alarms of flo#8acoustic, flo#8pressure, or pressure8acoustic are used then the number of components in the chain is increased( By summing the component availabilities for each element, an uptake for each alarm can be estimated( @or e%ample, a flo#meter #ith a failure of once in E years #ith a repair time of 0 hours has an availability factor of ;(999:> 6or 99(9:>27( @rom such an analysis, the conclusion can be dra#n that the system should be made as simple as possible or instruments should be installed in duplicate to ma%imi-e alarm uptake( Section 9 also sho#ed that instruments should be selected on performance and not on economic grounds( It is better to install fe#er high performance instruments than numerous poor ones( Digital conversion also requires attention( ,ften 1/ or 10 bit conversion is required to give the necessary accuracy of data processing and usually double precision computation is also required( he use of standard outputs should be made #herever possible( Custom designed electronics invariably lead to problems( he best guideline, ho#ever, is to seek users of pipe line monitoring systems to ask their advice and e%perience #ith instruments and system components( Independent validation of
all information should be made #herever possible( Companies that supply such complete systems usually have a client list and it is #orth spending time talking to these clients before the final design specification is fi%ed( .ith regard to instrumentation, flo#meters #ith the highest accuracy are required for mass balance functions( Suitable types include turbine and displacement meters #ith pulse outputs( ,rifice meters are not really suitable, since the best accuracy that can be obtained from a #ell maintained system is around J12 of full scale( he correct choice of turbine by comparison is J;(/>2 of reading or better(
De#elopment of Pipe Line Monitoring Systems
he speed of instrumentation development generally is rather frightening( he impact of microelectronics is still being felt some ten years after they first appeared, and ne# and improved transducers #ith on-board 4intelligence4 are being sold in increasing numbers( &t the same time the si-e of computers is decreasing and the computating capability is increasing( Soft#are is also advancing rapidly and the performance of modern flo# monitoring systems is becoming dependent on the accuracy of the modeling equations( <quations of state and the behavior of hydrocarbon mi%tures are not particularly advanced or #ell understood, and fundamental research is required before the ne%t advance in this type of technology can proceed( &ll of these points indicate that computer-based monitoring systems #ill become the standard technique of operating and controlling pipe lines in the future( Control algorithms can be integrated #ith the applications modules to produce a semi-intelligent complete integrity monitoring scheme( &s e%perience in the design and operation of such systems gro#s they #ill be applied #ith increasing confidence( @uture systems #ill use a combination of the ne# technology discussed( his could include internal monitoring pigs and advanced pipe line models, both run from a central control room( hus the internal and e%ternal state of the line could be checked simultaneously( Such technology can enable safer and more economic operation of pipe lines to be carried out( Conc&usion he paper cannot do =ustice in such a short space, to the comple% and diverse sub=ect of leak detection( Such systems have been in operation in many forms all over the #orld, but it is only recently that environmental as #ell as economic factors have influenced their development( +odern digital systems are transforming operation and design, #ith many parallel functions being possible #ith a single system( here is no# a clear need to ensure complete integration of all components in the system to guarantee safer and more accurate pipe line management( Instrument selection is critical, as is the need to develop better thermodynamic models, for the ne%t generation of systems to become more reliable and accurate( Re'erences
1( @urness, ?( &(, 4+odern Pipe )ine +onitoring echniques,4 Pipes and Pipelines International, +ay-Kune, I-11 and September-,ctober, 10-1:, 19:>( /( +aillou%, ?( )( and van ?eet, K( D(, 4?eal ime ransient @lo# +odeling &pplications,4 PSIA &nnual +eeting , 19:E( E( Covington, +( (, 4Pipe )ine ?upture Detection and Control,4 &S+< Paper :-P< ->0, 19I:( 0( Bernard, A( A(, 4<nergy Balance Derivation,4 C?C Bethany Internal ?eport, 19:/( >( &non, 4Pinpointing pigs in pipe lines,4 R & D Digest, :, 10-1>, British Aas Corporation, 19:G( G( @u=imori, *( and Sugaya, S(, 4& study of a leak detection based on in-out flo# difference method,4 Proceedings of I+<", Symposium on @lo# +easurement and Control, pp( /;>/;9, okyo, Kapan, 19I9( I( Ste#art, ( )(, 4,perating e%perience using a computer model for pipe line leak detection,4 Journal of Pipelines, E, /EE-/EI, 19:E( (otation & - cross-sectional area B - fluid bulk modulus c - specific heat at constant volume D - pipe line diameter <P1 - uncertainty in upstream pressure measurement <P/ - uncertainty in do#nstream pressure measurement <F - uncertainty in flo# measurement g - gravitational acceleration h - elevation ) - pipe length P - pressure P; - reference or base pressure P1 - section upstream pressure P/ - pipe line flo# rate F - pipe line flo# rate F1 - upstream measured vs( modeled flo# discrepancy F/ - do#nstream measured vs( modeled flo# discrepancy ? - gas constant - temperature g - ground temperature ) - leak detection threshold level o - reference or base temperature t - time ! - heat transfer coefficient 5 - velocity D - leak location % - incremental distance along the pipe line y - C factor correlation coefficient C - gas compressibility factor alpha - coefficient of thermal e%pansion p - density po - reference or base density
