GB2433137A

GB2433137A - Method for the early warning of severe slugging

Info

Publication number: GB2433137A
Application number: GB0525180A
Authority: GB
Inventors: Peter Howard Knight; Andrew William Pike
Original assignee: Alstom Technology AG
Current assignee: General Electric Technology GmbH
Priority date: 2005-12-10
Filing date: 2005-12-10
Publication date: 2007-06-13
Also published as: GB0525180D0

Abstract

The present invention provides a method of monitoring the flow of a multiphase fluid through a pipeline 2 and has particular application as a method for the early warning of severe slugging. The method uses at least one signal corresponding to a physical characteristic of the fluid such as pressure or density measured by a sensor 8. This signal is then analysed using a computer processing unit 10 to see how it varies with time in order to identify characteristics that indicate a future change in the flow regime of the fluid. In other words, the method can make a real-time prediction that the fluid will experience severe slugging several minutes before this actually occurs. The analysis can he carried out by using the signal to estimate the parameters of an adaptive dynamic model. A characteristic of the adaptive dynamic model is then monitored and when this reaches a predetermined threshold then a future change in the flow regime of the fluid is predicted.

Description

TITLE

Method for the early warning of severe slugging

DESCRIPTION

Teclmical Field

The present invention provides a method that gives accurate real-time prediction of the imminent onset of severe slugging in multiphase pipelines. The method uses on-line dynamic modelling where the model is updated automatically so that it can adapt to changes in the flow through the pipeline.

Background Art

The mechanics of fluid flow in multiphase pipelines is extremely complex. It is particularly complex in oil production pipelines that contain multiphase fluid that is a mixture of oil, gas and water. Normally the fluid flow within such pipelines is one of a finite number of equilibrium steady-state flow regimes e.g. stratified flow, bubble flow, oscillatory flow, slug flow. These flow regimes are differentiated by strongly characterised physical features which are well known and widely described within the literature. It is desirable that the pipeline operates under a safe steady-state flow regime in which there is a relatively constant fluid flow rate, for example stratified flow. However, over time, variations in the conditions -e.g. increased gas to liquid ratio, reduced well-head pressure or separator pressure -may cause the operating flow regime to change from one steady-state flow regime to another. Up until now the occurrence of such variations in flow regime have been impossible to predict. This has been a particular problem for the offshore industry as the development of some flow regimes is inherently undesirable.

Specifically, the development of large liquid slugs is a major problem in multiphase pipelines. There are two main types of "severe slugging", namely terrain-induced slugging and riser-induced slugging. They are both characterised by the accumulation of the liquid phase at low points in the pipeline, either because of dips in the terrain over which the pipeline is laid or because the pipeline includes a riser. The development of severe slugging can have significant impacts on downstream separation operations. The region of the pipeline where the severe slugging takes place (such as the riser, for example) will also be subject to potentially damaging cyclic high amplitude mechanical stresses.

Severe slugging is an understood flow regime and is produced by a well-known mechanism. Typically the pipeline will be slightly inclined downwards towards a low point of the pipeline or the base of the riser. The liquid phase therefore tends to collect in this part of the pipeline until the pipeline cross-section is filled. The upstream gas phase is compressed until the pressure in the pipeline increases to a sufficient level to expel the liquid in the pipeline or riser, thereby creating a long liquid slug that is pushed in front of the expanding gas phase. This release causes the pressure in the pipeline to reduce to normal levels and the cycle repeats.

Numerous methods have been proposed for trying to control or reduce the effects of severe slugging. Preferably these methods would only be applied when it is known that severe slug flow is imminent.

Existing pipeline monitoring methods are not able to predict the onset of severe slug flow before it happens. Instead they merely try to detect when severe slugging has commenced. They tend to achieve this by monitoring the flow through the multiphase pipeline so that the change from normal flow to severe slugging can be detected.

United States Patent Application 2003/0225533 describes a method of detecting a boundary of a fluid flowing through a pipeline. Gamma ray densitometers are used to provide a time-varying signal that corresponds to changes in the density of the fluid flowing through the pipeline. The signal is analysed in the tinie domain a using time-encoded signal processing and recognition (TESPAR) process to produce a processed signal. The processed signal is then monitored for a predetermined feature so that the location of a boundary representing the start of a liquid slug can be found.

United States Patent 5148405 describes a different method where acoustic emissions of the pipeline are monitored by ultrasonic sound transducers to generate a time-varying signal. The transient characteristics of the time-varying signal (such as amplitude, for example) are then used to distinguish between normal flow where the liquid and gas phases are well mixed and slugging flow.

Summary of the invention

The invention provides a method of monitoring the flow of multiphase fluid through a pipeline comprising the steps of (a) obtaining at least one signal corresponding to a physical characteristic of a multiphase fluid flowing through the pipeline, and (b) analysing how the at least one signal varies with time in order to identify characteristics indicative of a future change iii the flow regime of the fluid.

It is to be undcrstood that a multiphase fluid in the context of this invention could be any mixture of gas and liquid phase fluids. For example, the multiphase fluid in oil production pipelines is a mixture of oil, water and gas.

As discussed above, it is already known to analyse a signal corresponding to the physical characteristics of a multiphase fluid flowing through a pipeline to distinguish which Ilow regime is operating at that moment in time. The method of the present invention improves upon the prior art as it can actively predict if a change in flow regime will take place at some point in the future. This is a critical improvement because it allows steps to be taken to control the problem of severe slugging well in advance, in some instances, steps may even be taken to prevent the onset of severe slugging by employing pre-emptive measures such as foaming, choking, riser base pressure control, flow rate control and gas lift, for example.

A further distinguishing feature of the current invention in relation to the prior art is that the properties of an adaptive dynamic model (updated in real-time) are monitored and it is the variation of these properties as the characteristics of the flow alter that provides a means for flow regime prediction.

Although the method of the present invention will typically be used to provide an advance warning of a future change from an acceptable flow regime such as stratified flow, for example, to an unacceptable flow regime such as severe slugging, it will be readily appreciated that the method is not limited to such use and can provide an indication of a future change between any two flow regimes of the fluid.

Physical characteristics of the fluid that can be used by the method of the present invention include inter a/ia the pressure of the fluid at one point within the pipeline, the difference in pressure of the fluid at two spaced points within the pipeline, and the density of the fluid within the pipeline. If the pressure or density of the fluid within the pipeline is only monitored at one point then this is preferably at or near the base of the riser because this has been found from experiment to provide the earliest indication of a transition from one flow regime to another. If the difference in pressure of the fluid in the pipeline is measured at two spaced points and the pipeline includes a riser (of any geometry e.g. free hanging catenary, vertical or S' shaped) then it is preferable that one of the spaced points is at or near the base of the riser and the other one of the spaced points is near the top of the riser. Even if the pipeline does not include a riser, the spaced points within the pipeline are required to be at least a few metres apart in order to attain a pressure difference with an adequate signal to noise ratio. However, at the other extreme it is also perfectly feasible to place the transducers at the extremes (i.e. the welihead and the separator) of the flowline riser system, with a typical separation distance of several kilometres.

The pipeline can include a separator (optionally a mini-separator located at or near the top of a riser and upstream of the main separator) and in this case a physical characteristic of the fluid can include the outflow of the gas and/or liquid phase from the separator.

The physical characteristics can be monitored singularly or in combination. In the case where two or more of the physical characteristics are monitored simultaneously then a number of individual signals (each one corresponding to one of the physical characteristics) will be analysed to determine how they vary with time. This gives the option of analysing the individual signals separately and thus looking for characteristics that might indicate a future change in the flow regime of the fluid from each of the physical characteristics separately. Alternatively, the individual signals

--

representing the different physical characteristics of the fluid flowing through the pipeline can be combined to form a single signal that is then analysed to see how it varies with time to identify characteristics indicative of a future change in the flow regime of the fluid.

The pressure of the fluid caii be obtained using at least one pressure transducer situated within the pipeline. If tile difference in pressure between two points is being measured either two pressure transducers or a single differential pressure transducer can be used. Tile density of tile fluid can he obtained using a gamma ray densitometer situated outside the pipeiine. It will he readily appreciated that any other suitable pressure or density sensors or sensing devices can be used. More than one pressure transducer or gamma ray densitometer can he used at each point in the pipeline to provide a degree of redundancy in case of equipment failure or to obtain an average pressure or density reading.

The at least one signal corresponding to the physical characteristic of the fluid flowing through the pipeline is preferably analysed by a computer running suitable real-time adaptive dynamic modelling software. Preferably the analysis of how the at least one signal varies with time is carried out on a discrete signal. In other words, the value of each of the physical characteristics is preferahiy measured at set intervals of time. For example, the pressure of the fluid at a point within the pipeline may be measured once every second thereby giving a signal of the form y(1). This may be achieved by logging a continuous signal output from a pressure sensor at discrete time intervals via an analogue to digital conversion device. Alternatively, the pressure sensor can he adapted or controlled to measure tile pressure at discrete time intervals so that ii outputs a series of individual signals indicative of tile pressure at tile time when the measurement was taken.

Each of the discrete signals is preferably analysed using a suitable second or higher order autoregressive (AR) iinear model. One of the features of such a model is tile location of its poles in tile discrete frequency plane. The poie locations correspond to an associated natural frequency of tile flow regime at that point in time. When a discrete signal corresponding to any of the physical characteristics mentioned above is analysed using a second or higher order AR model, the poles in the discrete frequency plane will occupy particular locations or regions that are indicative of the flow in the pipeline at that instant in time. If there is a steady-state flow regime operating the poles will occupy a region particular to that flow regime. If the flow is undergoing a transition from one flow regime to another, the poles will be located between the regions associated with the two flow regimes. Furthermore during such transitions the location of each of the poles will migrate over a period of time along a particular well-defined trajectory or path from the regions corresponding to the current flow regime to the regions corresponding to the future flow regime. These trajectories are termed the migration paths.

The transitions between flow regimes are not instantaneous and therefore the migration of the poles between regions is not immediate either. It can often take several minutes for the poles to complete their movement along the migration paths. It is by monitoring the location of the poles and observing when they are progressing along a particular migration path towards a particular future pole location that is indicative of a specific flow regime that future flow regimes can be predicted. It is important to note that it is not possible to produce such predictions merely by directly observing the physical characteristics of the flow of fluid within the pipeline. If these characteristics are observed directly then changes may be noticed during the transitional periods but it would be impossible to predict from these changes what the future flow regime will be. For example, using such methods severe slugging can only be detected once it has begun.

Before the method of the present invention can be used to predict a future change in the flow regime of the fluid it is first necessary to identify the regions in the discrete frequency plane that are occupied by the poles for each of the various flow regimes.

This can be achieved by monitoring the location of the poles over a period of time while the pipeline is operating under controlled conditions. For example, if while the pipeline is operating the fluid is known to be experiencing a particular flow regime then the location of the poles for each of the discrete signals obtained during this particular flow regime can be mapped to define regions of the discrete frequency plane associated with that particular flow regime. The region associated with one flow regime (such as stratified flow, for example) will be different and spaced apart from the region associated with a different flow regime (such as severe slugging, for example). Slight differences in the flow characteristics of the fluid within the same flow regime mean that it is likely that the pole locations of the adaptive model will not always be precisely coincident with each other over time. Therefore, in practice, the various pole locations will be smeared out to define a region or area of the discrete frequency plane that may be identified with a particular flow regime.

Typical migration paths along which the poles traverse between the regions of the discrete frequency plane can also be mapped while the pipeline is operating under commonly encountered changes in operating conditions (e.g. increasing gas flow, reducing liquid flow, reducing wellhead pressure etc.). However, it is understood that there are, in principle, an infinite number of migration paths between one region and another. Therefore ii may be preferable that the prediction of the future flow regime is realised via a migration path tracking algorithm which generates an extrapolation of the current migration path on the discrete frequency plane.

After the regions of the discrete frequency plane associated with each flow regime have been mapped, at least one threshold curve is calculated and placed on the discrete frequency plane so that at least one boundary between areas of acceptable and unacceptable flow regimes is defined. The threshold curves will necessarily extend across all the possible migration paths that define a transition between an acceptable and unacceptable flow regime.

Once the method of the present invention is put into operation then, if the poles derived from the analysis of the discrete signal are located in the regions that were mapped when the fluid in the pipeline was known to have stratified flow, the method can make a positive determination that the flow regime within the pipeline is stratified flow and is therefore acceptable. Similarly, if the poles derived from the analysis of the discrete signal are located in the regions that were mapped when the pipeline was known to he experiencing severe slugging then the method can make a positive determination that severe slugging is taking place.

When a transition of flow regime begins, the location of the each of the poles will start to move along a particular migration path in the discrete frequency plane. If the transition that occurs is from an acceptable flow regime to an unacceptable flow regime then at some point in time the poles will cross the threshold curve. At that point the method can make a positive determination that a change to an unacceptable flow regime is likely to take place in the future (that is when the poles have finally completed their movement along the particular migration path being followed and are located in the regions associated with the future flow regime). The positive determination can typically be made a few minutes before the transition in the flow regime actually takes place but this will depend on the operating conditions of the pipeline and the mapping of the threshold curves.

The early warning of the future onset of an undesirable flow regime can be further enhanced by the application of a suitable migration path tracking algorithm. Such an algorithm would extrapolate the current migration path into the future to anticipate the moment the poles will cross the threshold curve and/or the moment the poles will arrive at the undesirable flow regime region. It could also estimate the time interval before the occurrence of these events.

An example of a suitable migration path tracking algorithm is as follows. The natural frequency values of each of the poles for at least a finite number of preceding time intervals would be stored. A polynomial function would then be fitted to these data points to produce a time varying polynomial function. This lunction could then be used to predict the future behaviour of the poles. Specifically, it could give a prediction of whether the poles will cross a threshold curve within a first predetermined future period and a determination of future regime change can be made. The function may also be used to predict whether the poles will reach an undesirable flow regime region within a second predetermined future period and which flow regime they may reach. 1-lowever it is to he understood that many other tracking algorithms would also be apparent to the man skilled in the art and these may also be applied to this invention in the same manner. Furthermore it is important to note that the operation of the invention does not rely upon applying an additional tracking algorithm as described above, but a tracking algorithm may enhance the early warning performance of the invention via extrapolation of the observed migration paths and in this manner greatly improved performance may be achieved.

As an alternative to determining the model parameters by operating the pipeline under controlled conditions the predetermined threshold curves, typical migration paths, the regions indicative of the current flow regime and the regions indicative of the future flow regime could be determined using multiphase flow modelling software to predict the dynamic behaviour of signals thereby eliminating the need to experience undesirable flow regimes offshore.

The method of the invention preferably further comprises the step (c) of producing a warning signal when characteristics indicative of a future change in the flow regime of the fluid are found in the at least one signal. The warning signal may be a simple siren or other signal for alerting an operator. Alternatively, the warning signal may also initiate one or more severe slugging flow control methods to prevent the onset of severe slugging, or may be used by an apparatus to reduce severe slugging within the pipeline once it occurs.

If the transition between different flow regimes is not completed (in other words if the poles move back to the regions associated with the current flow regime after passing the threshold curves) then the method can cancel the warning signal or produce an "all clear" signal to tell the operator that a change of flow regime is not going to take place after all.

The present invention further provides an apparatus for monitoring a pipeline using the method described above, the apparatus comprising a sensor for measuring a physical characteristic of the fluid flowing through the pipeline and producing at least one time-varying signal corresponding to the physical characteristic Drawings Figure 1 shows a schematic of an apparatus that uses the method of the present invention to monitor a pipeline; Figure 2 shows an algorithm that may be used to provide an early warning of severe slugging; Figure 3 shows the variation in the parameters of a second-order autoregressive (AR) model as the flow regime in the pipeline of Figure 1 varies; Figure 4 shows a plot of the pole locations in the discrete frequency plane for the different flow regimes of the pipeline of Figure 1; Figure 5 shows a plot of the variation of pole locations in the discrete frequency plane for the pipeline of Figure 1 over a period of time; and Figure 6 shows a log of pressure data against time for the apparatus of Figure 1 when the pipeline undergoes a transition between oscillatory flow and severe slugging.

Figure 1 shows an apparatus for monitoring the flow of a mixture of oil, gas and water through a pipeline 2. The pipeline 2 includes an S' shaped riser 4 that feeds into a mini-separator 6. I'he output of the mini-separator 6 feeds to the main separator (not shown). The direction of flow of the oil, gas and water mixture is represented by the arrows.

The apparatus includes a pressure transducer 8 located at the base of the riser 4 and a computer processing unit 10 that is situated away from the pipeline 2 and which receives a signal from the pressure transducer 8. More particularly, the pressure transducer 8 measures the pressure of the oil, gas and water mixture at the base of the riser 4 and sends this information as an analogue electrical signal to the computer processor unit 10 where it is converted to a digital signal via an analogue to digital converter and high-pass filtered to produce the signal y(t). The signal is also high-pass filtered to make it a zero-mean signal. This is because gradual changes in mean level due to, for example, well-head pressure reduction that occurs as the well fluids are depleted, are of no utility for the flow regime detection method. In fact a slowly changing mean may introduce an undesired bias on the model parameters leading to greater variation in the pole locations for a given flow regime. This in turn would result in an increased area of the region within the discrete frequency plane associated with a particular flow regime and would consequently reduce the sensitivity of the flow regime transition detection algorithm. Furthermore the linear model used in this embodiment of the invention (equation 1) is a zero mean stochastic process (i.e. provided that the model is stable the model poles lie inside the unit circle on the discrete frequency plane). Therefore the model is only suitable for application to zero-mean signals, which is ensured in this case via high-pass filtering.

The filtered signal y(t) is used to update in real-time the parameters of a second-order autoregressive (AR) linear model of the form: A(q)y(i) = e(t) y(t)= A(q)'e(t) (1) = y(t)= iI(q)e(t) {i.e. H(q) A(q) } where: et) is a white noise disturbance; and A(q) is a discrete time polynomial in the delay operator q'; where qy(t) = y(1 -1) Note although this embodiment of the invention is described using a second-order AR linear model, the use of models of higher order is also possible and may be preferable in some cases).

The discrete time domain model (I) can be represented in the discrete complex frequency domain by using the z transform to get: Y(z)= IJ(z)E(z) The second-order model used to analyse the data is defined in the z domain is: -12 -A(z)=l+a1z +a2z2 =H(z)' A Recursive Least Squares (RLS) algorithm is used to calculate the model coefficients {a1,a, }. The algorithm uses the input taken at that time (time t) and the previous two inputs taken (time t-1 and t-2) to determine what the coefficients should be at time t. The RLS algorithm used is shown in the top half of Figure 2. Specifically this algorithm is a recursive least squares module which computes a value for {a1, a2} at each sample time using a recursive least squares update method with a forget factor 2.

Figure 2 also shows, in its lower half, a flow classification module which uses standard linear model theory to compute: 1) the discrete model poles and 2) the associated natural frequency. The computed natural frequency is then compared to a threshold frequency for the purpose of determining whether a warning of the imminent onset of severe slugging flow should be generated.

The model coefficients {a1, a2} are indicative of the flow regime that is operating at a given moment in time. This can be seen clearly in Figure 3 which shows the variations in {a1, a} as the flow regime operating within a pipeline varies from severe slugging to oscillatory flow, then to bubble flow, then to slug flow, then back to bubble flow and finally back to severe slugging. However it is quite difficult to identify the operating flow regime directly from the model coeflicients. The poles of the model in the z domain give a much more pronounced differentiation between flow regimes.

The poles of the model at a given point in time, i.e. for the values of {a1, a2} at that time, are found from the roots of A(z) = 0. That is, the poles will be the two values of z at which A(z) = 0 and these may be mapped in the discrete frequency complex plane (z domain). The positions of these poles in the discrete frequency complex plane are indicative of the flow regime that is operating at a given moment in time as previously discussed. l'he reason for this is the pole locations can be related to the frequency at -13 -which the sampled signal for a particular flow regime possesses maximum power.

This frequency (a)nax) is the frequency at which the power spectral density (PSD) is maximal.

This relationship is true because the model A(z) may be mapped to the continuous frequency complex plane (s domain) standard form H(s): fI(s) + 2çw,, + whereais the natural frequency and is the damping coefficient by using the function: z = H(s) has poles given by: s = -4?o,, jw /1 -Therefore, for a given moment in time when a1, a} have been calculated and the values of the poles of A(z) have also been calculated, by transforming the poles of A(z) to the s domain and equating them to the poles of H(s) the values of 4 andw,1 for the operating flow regime may be calculated.

The poles of H(s) for constant values of 4 andw,1 may also he mapped into the z domain in order to provide a means of defining lines of constant and w,, in the z domain unit circle, thus helping to define the different flow regime regions within the z domain unit circle. -14-

H(s) has a pcak resonant frequency (that is. the frequency at which H(s is maximal) given by: (V,. = o,4l-2 -for Therefore, as it is possible to calculate and o,, at any point in time, it is also possible to calculate the peak resonant frequency Vr. This is important as it is known that when a specific flow regime is operating, the peak resonant frequency Wr of the model H(s) is a good approximation to the power spectral density (PSD) peak power frequency w1 of that flow regime and each separate flow regime possesses a unique PSD maximal frequency. This means the poles of H(s), and therefore A(z) at a given moment in time, also relate to the how regime that is operating. In other words, as each flow regime possesses a unique PSD maximal frequency, each flow regime has a pair of poles oi the function A(z) that take up a unique location within the z-domain unit circle.

Prior to the use of the monitoring apparatus these unique pole locations are mapped in the z domain unit circle by monitoring thepipeline whilst the fluid is experiencing known flow regimes. As described above, the uniclue pole locations for repeated models derived from the same flow regime are not always precisely coincident with each other and SO it is better to consider a region that is defined by the boundary of the smeared out pole locations. ligure 4 shows the known pole locations in the z domain for oscillatory flow A and severe slugging B. The lines labelled 0.3 14, 0.628, 0.942, 1.26, 1.57, 1.88, 2.2, 2.1, 2.83 and 3.14 are lines of constant pole natural frequency o,, and the lines labelled 0.]. 0.2 to 0.9 are lines of constant pole clamping (see above). The z domain unit circle represents the boundary for stable models.

Figure 5 shows a similar diagram of the pole locations in the z domain over a period of time when the flow through the pipeline 2.inclergoes a transition from oscillatory flow to severe slugging. The known pole locations in the z domain for oscillatory flow A and severe slugging B are oined by a well defined trajectories or migration -15- paths MP. The Threshold curve is labelled T. Note that in this case the curve corresponds to a constant pole natural frequency contour. The undesired severe slugging flow regime maps to a region of the z domain corresponding to a lower natural frequency than the threshold whereas all other flow regimes map to a region with higher natural frequency than the threshold. Therefore the threshold curve divides the unit circle into two segments corresponding to acceptable and unacceptable flow conditions. As the oil, gas and water mixture flowing through the pipeline 2 starts the transition from oscillatory flow to severe slugging, the location of the poles in the z domain derived from the adaptive model will start to move from the known pole locations A for oscillatory flow along particular migration paths MP towards the known pole locations B for severe slugging. Eventually, the poles will intersect the threshold curve T that lies part of the way along the migration paths MP.

At this point, the computer processing unit 10 can output a warning signal that a transition from oscillatory flow to severe slugging will take place in the near future.

The time before the onset of severe slugging is estimated using a tracking algorithm that logs the poles migration from A to T and then computes a time to go' estimate based upon an extrapolation of the logged trajectory on the unit circle. l'he tracking algorithm operates in the following manner. The natural frequency of each of the poles for a finite number of preceding sample times is stored. A polynomial function of finite order is then fitted to these data points to generate a polynomial function with time as the input and the predicted natural frequency of the poles at that input time as the output. The generated polynomial function is then used to predict whether the poles will cross the threshold curve within a first predetermined future period (for example five minutes) and/or whether the poles will reach an undesirable flow regime region within a second predetermined future period. A warning signal may be generated if either of these events is predicted. This is possible as the threshold curves are lines of constant natural frequency and the flow regime regions are characterised by a finite band of natural frequencies. The function may also predict which undesirable flow regime is likely to occur. At each new sample time the polynomial function is updated and the determination is repeated. -16-

The number of values of the natural frequency of the polesthat form the basis of the prediction must be sufficient to provide enough data to make a reasonably accurate prediction of future behaviour. However the required number of arithmetic operations greatly increases with the number of values used. This means the number of values used is limited by the computing power available. In general a low order polynomial is preferred to a high order polynomial in terms of minimising extrapolation errors.

J-Iigh order polynomials can result in spurious' oscillation in the extrapolated function values. In this embodiment of the invention the best prediction has been fbund to result from fitting a cubic polynomial, using the method of linear least squares regression, to the values of the natural frequency of the poles for the ten most recent sample times. Using this algorithm, if the threshold curve T is properly selected then the computer processing unit 10 can provide a warning signal at least two or three minutes before the onset of the severe slugging.

If the oil, gas and water mixture undergoes a transition from oscillatory flow to a flow regime other than severe slugging then it will be readily appreciated that the poles derived from the model would move along entirely different migration paths. If the known pole locations at the end of the different migration paths correspond to a flow regime which is still unacceptable then a warning signal will be provided by the computer processing unit 10 as described above. However, if the flow regime is acceptable then the transition can be logged but no warning signal would be provided.

If the warning signal is used by an apparatus (not shown) to reduce severe slugging within the pipeline then it is possible that the conditions in the pipeline 2 might be changed such that the transition does not take place and the oscillatory flow is maintained or other acceptable flow regime established. In this case it will be readily appreciated that the poles derived from the model will follow a migration path back towards the known pole locations A for oscillatory flow or to pole locations corresponding to other acceptable flow regimes. The migration path followed would depend upon the conditions operating within the pipeline and the final flow regime being established.

Figure 6 shows two diagrams that illustrate the transition from oscillatory flow to severe slugging in more detail. The top diagram is a log of pressure data against time and represents the pressure transducer 8 signal transmitted to computer processing unit 10. The diagram shows that the oil, gas and water mixture is initially experiencing oscillatory flow (labelled ()SC) where the pressure varies between fairly stable limits. [he low regime of the oil, gas and water mixture eventually changes to severe slugging (labelled SSI) where the pressure varies with a characteristic shape.

Between these Iwo flow regimes there is a period of transition which can be determined from (lie movement of the poles along the particular migration path for this particular flow regime change in the z domain, but not from the actual pressure data itself As the computer processing unit 1 0 continues to analyse the pressure data using the adaptive second-order autoregressive (AR) linear model the poles will eventually intersect the threshold curve. At this point, the computer processing unit 10 triggers the severe slugging early warning flag as shown in the bottom diagram.

The warning flag is triggered at a time of 353 minutes but the onset of severe slugging does not actually take place until at least 2 minutes later. The computer processing unit 10 can there lore provide a warning signal at least two minutes befbre the severe slugging starts to occur. -18-

Claims

CLAIMS

I A method of monitoring the flow of multiphase fluid through a pipeline comprising tlic steps of: (a) obtaining at least one signal corresponding to a physical characteristic of a. multiphase fluid flowing through the pipeline; and (b) analysing how the at least one signal varies with time in order to identil.y characteristics indicative of a future change in the flow regime of the fluid.

2. A method according to claim, wherein the physical characteristic that is obtained is the pressure of the fluid within the pipeline.

3. A method according to claim
2. wherein the pressure of the fluid is obtained using at least one pressure transducer situated within the pipeline.

4. A method according to claim 2 or claim 3, wherein the pipeline includes a riser and the at least one pressure transducer is situated at or near the base of the riser.

5. A method according to claim I, wherein the physical characteristic that is obtained is the dilicrence in pressure of the fluid at two spaced points within the pipeline.

6. i-\ method according to claim 5. wherein the difference in pressure is obtained by comparing the signals received from at least one pressure transducer situated within the pipeline at one of the spaced points and at least one pressure transducer situated within the pipeline at the other one of the spaced points.

7. A method according to claim S or claim 6, wherein the pipeline includes a riser and one ol' the spaced points is at or near the base of the riser and the other one of the spaced poi its is at or near the top of the riser.

8. A method according to claim I, wherein the physical characteristic that is obtained is the dcniiv of the fluid within the pipeline.

9. A accord to claim 8. wherein the density of the fluid is obtained using a gamma ray densitometer situated outside the pipeline.

10. A method according to claim 8 or claim 9, wherein the pipeline includes a riser and the gamma ray densitometer is positioned at or near the base of the riser.

11. A method according to preceding claim 1, wherein the pipeline includes a separator and the physical characteristic that is obtained is the liquid outflow from the separator.

12. A method according to any preceding claim, wherein the at least one signal corresponding to the physical characteristic of a fluid flowing through the pipeline is analysed Nv a computer.

13. A method according to any preceding claim, wherein the step of analysing how the at least one signal varies with lime in order to identify characteristics indicative ola future change in the flow regime of the fluid includes the steps of: (hi) using the at least one signal to estimate the parameters of an adaptive dynamic model: (h2) nm itoring a characteristic o! the adaptive dynamic model; and (h3) predicting a future change in the flow regime of the fluid when the monitored charactcristic of the adaptive dynamic model reaches a predetermined threshold.

14. A niethod according to claim 13, wherein the adaptive dynamic model is an autoregressive ( A I) linear model of second or higher order.

1 5. A method according to claimn 14 wherein the adaptive dynamic model is a second order i\ R ear model -20 - 16. A method according to any ol claims 13 to 15, wherein the characteristic of the adaptive dynamic model that is monitored is the location of poles in the discrete complex eqiieiwy plane.

1 7. A method according to claim I 6. wherein at least one region in the discrete complex frequency plane corresponding to tlie operation of an acceptable steady state flow regime, at least one region corresponding to the operation of an unacceptable steady state flow regime and at least one threshold curve between the steady state flow regime flow regimes are predetermined as a preliminary step while the pipeline is operating under control led conditions.

18. A niethod according to claim 1 6, wherein at least one region in the discrete complex frequency plane corresponding to the operation of an acceptable steady state flow regime. at least one region corresponding to the operation of an unacceptable steady state flow regime and at least one threshold curve between the steady state flow regime flow regimes are predetermined using multi-phase flow modelling SOftware.

1 9. A method according to either claim 1 7 or 18, wherein a change in the flow regime of the fluid from a current flow regime to a future flow regime is predicted when the monitored pole locations are situated on or cross, or are predicted to be situated on or cross. a threshold curve.

20. A method according to claim I 9 wherein a. migration path tracking algorithm is used to monitor the variation of the poles over lime and thereby predict their future positions.

21. A method according to claim 20 wherein the migration path tracking algorithm is used to predict whether the poles locations will be situated on or cross a threshold curve wit hin a succeeding peIiO(I of time of predetermined length. -21 -

22. A method according to claim 20 wherein the migration path tracking algorithm is used to predict whether the poles will he situated in a region in the discrete complex IcqUeflcY plane coriespondiig to an unacceptable steady state flow regime within a 5cecdtflg period ottilne of predetermined length 23. A method ecording to any preceding claim, further comprising the step of: (c) producing a warning signal when characteristics indicative of a future change in the flow regime of the fluid arc jound in the at least one signal.

24. A method according to claim 23. wherein the warning signal is produced when characteristics indicativc ol a change iii the floW regime of the fluid are found in at least two signals.

25. A cthod according to claim 23 or 24, wherein the warning signal is used by an apparatus to reduce severe slugging within the pipeline.

26. An apparatuK br onitOri1lg a pipeline using the method according to any preceding claim, hc apparatus comprisitig a sensor for measuring a physical characteristic ol the fluid flowing through the pipeline and producing at least one jmeVaryi1g signal coirespondig to the physical characteristic.