Method for Determining the Geometry of a Defect and for Determining a Load Limit

Abstract

Method for determining the geometry of multiple defects in a magnetizable object using a reference data record of the object, comprising determining an initial defect geometry as starting defect geometry, determining a first MFL prediction data record as starting prediction data record on the basis of the starting defect geometry, and iteratively matching the starting defect geometry to the geometry of the real defect(s) by means of the EDP unit and by means of multiple expert routines (11) running in competition and preferably in parallel with one another.

Claims

1-15. (canceled)

16. A method for determining the geometry of one or more real, examined defects in a magnetizable object, in particular a pipe or a tank, using a reference data record of the object, which is produced on the basis of one or more MFL measurements, preferably comprising at least partially representing the object by means of an EDP unit, in particular on or by means of an at least three-dimensional object grid, and comprising determining an initial defect geometry as starting defect geometry, in particular on the object grid or an at least two-dimensional defect grid (5), determining a first MFL prediction data record as starting prediction data record on the basis of the starting defect geometry, in particular by simulating an MFL measurement or assigning an MFL data record, and iteratively matching the starting defect geometry to the geometry of the real defect(s) by means of the EDP unit and by means of multiple expert routines (11) running in competition and preferably in parallel with one another, wherein a respective expert defect geometry is produced in respective expert routines (11) by means of at least one dedicated algorithm and on the basis of the starting defect geometry, a respective expert prediction data record is determined on the basis of the respective expert defect geometry, in particular by simulating an MFL measurement or assigning an MFL data record, and the expert defect geometry on which the respective expert prediction data record is based is then made available to at least one, preferably multiple, in particular all of the expert routines (11) as a new starting defect geometry for further matching to the geometry of the real defect(s) if the expert prediction data record is more similar to the reference data record than the starting prediction data record, and subsequently the expert prediction data record associated with the new starting defect geometry is used as new starting prediction data record, wherein the iterative matching by means of the expert routines takes place until a stop criterion is met.

17. The method of claim 16, wherein the expert routines (11) run in competition with one another such that the resources of the EDP unit are distributed to a respective expert routine, in particular in the form of computing time, preferably CPU and/or GPU time, on the basis of a success rate, for which in particular the number of starting defect geometries calculated by this expert routine and made available for one or more other expert routines (11) is taken into consideration, and/or on the basis of a reduction in a fitness function, for which in particular the number of expert prediction data records produced for the reduction is taken into consideration.

18. The method of claim 17, wherein the fitness function is used as a measure of the similarity of the expert prediction and reference data records.

19. The method of claim 16, wherein the geometry of the defect(s) is determined by additionally using a further reference data record that is linearly independent of the first reference data record in respect of the magnetization, and the starting defect geometry is taken as a basis for determining a further starting prediction data record, in particular by means of a further MFL simulation that takes into consideration the linear independence, and an expert defect geometry is used as starting defect geometry only when the associated expert prediction data records determined for both independent magnetizations are more similar to the respective reference data records than the starting prediction data records determined for the two magnetizations and/or a fitness function taking into consideration both expert prediction data records is improved.

20. The method of claim 19, wherein the first reference data record was produced by means of an MFL measurement with axial magnetization and the second reference data record was produced by means of an MFL measurement with magnetization in the circumferential direction of the pipe.

21. The method of claim 16, wherein starting and/or expert prediction data records are produced on the basis of a forward model for simulating the MFL measurement and in particular by means of a finite element model.

22. The method of claim 16, wherein the initial defect geometry is produced by means of a lookup table, by one of the expert routines (LM) and/or by a machine learning algorithm.

23. The method of claim 16, wherein the defect grid (5) is refined in regions in which the depth of the simulated defect(s) exceeds a threshold value.

24. The method of claim 16, wherein the object grid and/or the defect grid (5) is refined before a respective expert prediction data record is calculated.

25. The method of claim 16, wherein the starting defect geometry or a pointer referring thereto is stored in a memory area (12) of the EDP unit that is accessible to all expert routines (11).

26. The method of claim 16, wherein the stop criterion assumed is a substantial change in the starting defect geometry, and/or in the geometry of the object and/or defect grid (5) and/or the starting prediction data record and/or at least one expert prediction data record, that fails to materialize after a plurality of iterations.

27. The method of claim 16, wherein the stop criterion used is a comparison of the variation of the expert prediction data record with the measurement dispersion of the real data record.

28. The method of claim 16, wherein an expert routine (11) is assigned multiple algorithms for matching the expert defect geometry comprising machine learning, stochastic optimization, empirical and/or numerical model functions.

29. The method of claim 28, wherein an algorithm is selected and/or changed in an expert routine (11) on the basis of random number generation or by means of a selection function.

30. A method for determining a load limit for an object that is subject to pressure loading at least during operation and, in particular, is in the form of an oil, gas or water pipeline, in which a data record that describes one or more defect(s) is used as input data record in a calculation of the load limit, wherein the input data record is initially determined using the method of claim 16.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Further advantages and details of the invention can be taken from the description of the figures below, in which, schematically:

[0048] FIG. 1 shows a defect determination based on the prior art,

[0049] FIG. 2 shows a schematic depiction of a method according to the invention,

[0050] FIG. 3 shows a more detailed explanation of a portion of FIG. 2,

[0051] FIG. 4 shows comparison of a result of a method according to the invention with measurement data,

[0052] FIG. 5 shows a schematic depiction of a grid refinement as part of a method according to the invention,

[0053] FIG. 6 shows a result of a method according to the invention.

DETAILED DESCRIPTION

[0054] Individual features of the exemplary embodiments described below can, in combination with the features of the independent claims, also result in developments according to the invention.

[0055] The prior art involves the evaluation of MFL data of a pipe as shown in FIG. 1 being performed by means of the in particular also empirically based definition of boxes. The boxes depicted in the figure have respective length, width and depth dimensions. The x and y axes are represented in metre units ([m]). A check on the actual defect geometry on which this evaluation is based by means of laser scanning, i.e. by means of a direct measurement, revealed that the maximum burst pressure determinable on the basis of the defect geometry assumed as a result of the MFL data evaluation is, at 4744.69 kPa, only 55.2% of the maxium burst pressure calculated on the basis of the actual geometry. From the prior art, the operating pressure for safe operation of the pipeline, which is revealed as 3621.29 kPa on the basis of the empirically-based evaluation, is distinctly below a possible safe operating pressure.

[0056] In the method according to the invention, one exemplary embodiment involves the surface of a pipe being represented by a 2D mesh surface. The defect geometry can be described as a vector of depth values D located on a defect grid 5 (FIG. 5). This defect geometry is compared with the starting defect geometry on the basis of a result of a fitness function F(D) that takes into consideration the MFL fields associated with the respective geometries. In so doing, it is assumed that the lower the value of a fitness function, the closer the assumed expert defect geometry to the real geometry:

[00001]$F\left(D\right)=\underset{M}{.Math.}.Math..Math.H_{\mathrm{cal}}\left(D\right)-H_m.Math.+R\left(D\right)$

[0057] Here, M is the number of data records that can be handled at the same time (real MFL data records), H.sub.cal is the result of a simulation of the MFL measurement, H.sub.m are the measured data from the MFL measurement (reference data record) and R (D) is a regularization term, which can be estimated as follows:

R(D)=|D|,

where is a scaling term.

[0058] At least sections of the method sequence according to the invention are described below in accordance with FIG. 2, a plurality of the expert routines 11 that are in parallel and in competition being described just with one block 14.

[0059] By way of example, multiple passes by the same MFL pipeline pig can be combined as input data records as per box 2. Both data records 1 can be filtered and aligned with one another beforehand to improve combination (method step 3), for example in order to reduce any artefacts or background noise. Furthermore, a further data record 4, produced on the basis of a linearly independent, further magnetization and likewise filtered for the purpose of alignment with identical grid structures, can be used, so that, as per method section 6, two aligned reference data records obtained on the basis of measurement passes provided with linearly independent magnetizations are available.

[0060] Exactly aligned data records can be handled jointly, the method according to the invention implementing the simultaneous handling of the data records by using a fitness function that takes into consideration the combined data records.

[0061] In step 7, a first of the reference data records available in step 6 is selected for further handling. In addition to this, step 8 first of all involves an initial defect geometry being assumed, in particular generated in the present case, as starting defect geometry, said starting defect geometry being based for example on a normalized measurement signal S(x,y)/(max S). By way of example, the defect geometry can be derived from a threshold value function that takes into consideration the amplitude at grid points at which the signal is greater than a specific limit value l (e.g. 0.2):

[00002]$G\left(x,y\right)=\{\begin{array}{c}0,\mathrm{if}.Math..Math.\frac{S\left(x,y\right)}{\max .Math..Math.S}<l\\ \frac{S\left(x,y\right)}{\max .Math..Math.S}.Math.,\mathrm{else}\end{array}.$

[0062] The above approximation leads to a number of N defect depth values at the respective grid points:

D.sub.i=i wt/N*G,

with wt as the thickness of the wall of the pipe. i is the index also used as a value for determining the defect depth value. For a defect geometry of this kind, the fitness function is calculated and the profile having the lowest function value is used as initial solution:

D.sub.init=arg min F(D.sub.i)

[0063] This initial solution is then made available as starting defect geometry for the individual expert modules. To begin with, the number of parameter values (elements of the vector D) that describe the defect geometry can be kept as low as possible with the aim of reducing computing time. This is achieved by means of a dynamic grid adaptation, in particular. Since the number of depth values corresponds to the number of nodes in the defect grid 5, the number of nodes is at the same time also the number of defect parameters. Beginning with a comparatively coarse grid, it is progressively refined in relevant regions.

[0064] By way of example, given a prescribed node spacing of 14 mm, for example, an accompanying grid cell size of 14 mm14 mm and defect limit values of 30%, 50% and 80% of the wall thickness, it is possible to achieve the refinement depicted in FIG. 5 in the relevant grid region, those cells that exceed the depth values above being progressively divided. The grid deformation then correlates with the assumed defect geometry, i.e. there is a larger number of grid points in regions having high gradients.

[0065] The EDP unit is used to simulate the sequence of the workflow of a group of expert routines 11 that are in competition with one another using the method according to the invention. To this end, the program can have various modules that, independently of one another and in particular not in sync with one another, can put data into specific areas of the EDP unit so that said data are processed further therein. This takes place in particular under the supervision of a monitoring routine 9 (FIG. 3a). A plurality of expert routines 11 therefore take the success defined above, i.e. for example the number of starting defect geometries written to a common memory area 12, as a basis for keeping a number of computing slots 13 in order to produce respective expert defect geometries and/or to be able to perform associated MFL simulations or to have them performed in the case of an independent MFL simulation module. This corresponds to block 14 according to FIG. 2, said block serving as an example of multiple expert routines 11 (FIG. 3a). On the basis of the individual computing slots 13, the present exemplary embodiment involves the MFL simulations of the individual expert defect geometries likewise being performed under the supervision of the monitoring routine 9 in the simulation modules 16 for the purpose of creating the expert prediction data records. The more slots 13 are available for an expert routine, the greater the proportion of EDP resources for this expert routine. Preferably, the number of program modules required for performing MFL simulations is equal to the number of slots. The monitoring routine 9 monitors the number of iterations and the resultant changes in the starting defect geometry and, furthermore, monitors whether an associated stop criterion is reached. Subsequently, the result is output in accordance with block 17.

[0066] The number of computing slots 13 available for an expert routine 11 and the subsequently available simulation routines can vary such that a first expert routine can utilize up to 50%, for example, of the total computing time available for the computing slots and simulation routines.

[0067] The memory area 12 is used to store the starting defect geometries as depicted. It can be a memory area accessible to the expert routines 11. It can likewise be used to store log files of the expert routines 11 and monitoring routine 9 and also instructions to the expert routines 11 that are then implemented by them independently. By way of example, these can be an interrupt command that is applied when the stop criterion is reached.

[0068] Preferably, the expert routines 11 are independent program modules that produce new expert defect geometries and put them into the simulation routines 16. Furthermore, the fitness function described at the outset can be produced in the expert routines 11 on the basis of the expert prediction data record and can be compared with the starting prediction data record stored in the area 12. If the expert prediction data record is more similar overall to the reference data record, or, in the case of linearly independent measurement data records, then accordingly to the two reference data records, than the data record stored in the area 12, this expert prediction data record is then used as new starting prediction data record.

[0069] By way of example, a new defect geometry is produced in the expert routines 11 on a random number basis. This can be accomplished by using machine learning algorithms or empirical rules. Advantageously, however, additionally improved convergence of the solutions is accomplished by providing for the implementation of at least two base expert routines as described below.

[0070] These search strategies, which are preferably always implemented for a method according to the invention, are based on an assumed probability distribution p(x,y) of grid points, the depth value of which results in a maximum reduction in the fitness function. The probability function is used to identify N grid points (x.sub.n,y.sub.n). At each of the points considered, the depth function, which describes the depth of the corrosion at the grid location, for example, is changed by D, the arithmetic sign of the change being distributed on the basis of random number generation. The number of selected points N can also be chosen on a random number basis:

[00003]$D_{\mathrm{new}}\left(x,y\right)=\{\begin{array}{c}D.Math.\left(x_n,y_n\right).Math..Math.D,\mathrm{for}.Math..Math.\mathrm{selected}.Math..Math.\mathrm{points}\\ D\left(x,y\right),\mathrm{else}\end{array}$

[0071] A selection of the probability function p (x,y) can be used to implement different expert strategies, for example:

[00004]$p\left(x,y\right)=\frac{D\left(x,y\right)}{.Math.D\left(x,y\right).Math.}$

[0072] This algorithm produces a variation in the defect depths in which the grid points having the greatest depth are favoured. Another strategy can have the following appearance:

[00005]$p\left(x,y\right)=\frac{H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right)}{.Math.H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right).Math.}$

[0073] Such an algorithm varies the defect geometry at positions at which the simulated measurement signal for the best known solution H.sub.the best has the greatest difference from the measured signal.

[0074] On that basis, variations in the number of grid points to be considered and in the D allow different expert routines, or the algorithms thereof, to be set up. By way of example, the six expert routines below can be used:

[00006]$\begin{array}{cc}.Math.p\left(x,y\right)=\frac{D\left(x,y\right)}{.Math.D\left(x,y\right).Math.},N=1.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=1.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 1.\\ .Math.p\left(x,y\right)=\frac{D\left(x,y\right)}{.Math.D\left(x,y\right).Math.},N=2.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=5.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 2.\\ .Math.p\left(x,y\right)=\frac{D\left(x,y\right)}{.Math.D\left(x,y\right).Math.},N=3.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=5.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 3.\\ p\left(x,y\right)=\frac{H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right)}{.Math.H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right).Math.},N=1.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=1.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 4.\\ p\left(x,y\right)=\frac{H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right)}{.Math.H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right).Math.},N=2.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=5.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 5.\\ p\left(x,y\right)=\frac{H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right)}{.Math.H_{\mathrm{the}.Math..Math.\mathrm{best}}\left(x,y\right)-H_m\left(x,y\right).Math.},N=3.Math..Math.\mathrm{and}.Math..Math..Math..Math.D=5.Math.\%.Math..Math.\mathrm{wall}.Math..Math.\mathrm{thickness}& 6.\end{array}$

[0075] The monitoring routine 9 depicted in FIG. 3a has two functions, in particular, as described: first, a check is performed to ascertain whether the stop criterion is reached and, second, the resources of the EDP unit are done between the individual experts on the basis of the successes thereof. A measure of the success is

[00007]$P=\frac{.Math..Math.F}{N},$

where F is the reduction in the fitness function F owing to the result of the respective expert routine and N is the number of simulations necessary therefor. A rating of the n expert routines can be assumed to be

[00008]$R_n=\frac{P_n}{.Math.P_i}.$

[0076] The number of computing slots N.sub.S for an expert routine in one iteration is then

N.sub.S=int(R.sub.nN.sub.all),

where N.sub.all is the number of all available slots.

[0077] The simulation routines 16 are used to simulate an MFL measurement for an expert defect geometry. An expert routine can iterate until it finds a solution whose expert prediction data record is better than the starting prediction data record stored in the area 12. If this is the case, the expert routine 11 can process a further linearly independent data record or set out from the already improved solution to achieve further better solutions.

[0078] If multiple data records from different iterations that cannot be concordantly aligned were worked through by the expert routines, it is likewise possible for the performance of the method according the invention to involve the geometries obtained being overlaid in automated fashion, the maximum depth at the individual grid points being taken as a conservative estimate:

[00009]$D\left(x,y\right)=\underset{N}{\max }.Math.D_n\left(x,y\right)$

for n=1 . . . N, where N is the number of data records that need to be processed in succession. A resulting depth profile obtained over such an overlay of defect geometries can in turn be taken as a starting point for simulating an MFL signal. The error obtained can be obtained from the errors of the respective data records in the individual calculations:

E=H.sub.cal(D)H.sub.m

[0079] In order to demonstrate the efficiency of the proposed method, a multiplicity of test scenarios were performed, the data of two MFL inspection passes performed using mutually linearly independent magnetizations being used below in accordance with FIG. 4.

[0080] FIG. 4 shows a depiction, labelled No 21, of a real MFL measurement with magnetization running in the axial direction, while depiction 22 results from a measurement taken in the circumferential direction. The computation result for the defect geometry, which was obtained using the method according to the invention described above, is indexed by 23. The contour lines evenly split the region between 0 and 60% metal loss depth, as in depiction 24 too. Depiction 24 shows the actually scanned and hence directly measured exterior surface of the pipe section associated with depictions 21 and 22. A very high level of concordance between the laser scan measurement and the solution achieved by means of the method according the invention is obtained. This is much better than the solution based on the evaluation known in the prior art. In this case, it can be assumed that the discrepancies between the result based on the method according invention and that of the laser scan measurement are predominantly present on account of technical tolerances.

[0081] Based on the conventional approach involving ascertainment of the defect geometry as established in the prior art and ultimately depicted in FIG. 1, the aforementioned maximum burst pressure of 4744.69 kPa is obtained. Based on the method according to the invention, the defect geometry shown in FIG. 6 (contour lines at 2 mm depth) and, based thereon, a maximum burst pressure of 8543.46 kPa are obtained for the MFL data record on which FIG. 1 is also based. In the present case, this comes to within 99.4% of the maximum burst pressure that was determined on the basis of the actual defect geometry ascertained by laser scan. Accordingly, a pipeline examined using the method according to the invention can be operated at a safe operating pressure of 6520.53 kPa. This results in significant advantages for pipeline operators in comparison with the safe operating pressure of 3621.29 kPa based on the evaluation according to the prior art (FIG. 1). The method according to the invention allows the state of a pipe and hence the pressure specifiable for safe operation of the pipeline to be specified much more realistically, while safety of operation continues to be ensured. The method according to the invention with the expert routines competing for resources of the EDP unit allows such a result to be made available to the operators of pipelines more quickly than or at least in the same evaluation time as in the prior art.