Method for determining geometry of defect and method for determining load capacity limit

Abstract

The subject matter of the application is a method of determining the geometry of a defect and a method of determining a load capacity limit. The invention relates to a method for determining a geometry of a plurality of defects of a magnetisable object using a reference data set of the object, the method comprising determining an initial defect geometry as an initial defect geometry; determining a first MFL prediction data set as an initial prediction data set based on the initial defect geometry; and iteratively matching the starting defect geometry to the geometry of the one or more actual defects by means of the EDV unit and a plurality of expert routines (11) running competing with each other and preferably running in parallel with each other.

Description

[0001] This application is a divisional application of the invention patent application with international application number PCT / EP2018 / 076514, international application date September 28, 2018, application number 201880078530.4, and invention name “Method for determining the geometric structure of a defect and method for determining the load capacity limit”, which entered the Chinese national phase on June 4, 2020. Technical Field

[0002] The invention relates to a method for determining the geometry of a defect and to a method for determining the load-bearing capacity limit of an object which is subjected to pressure at least during operation. Background Art

[0003] One of the main purposes of pipeline inspection, in particular with so-called smart pipeline cleaning machines, is to predict safe operating conditions, which are derived from the state of the pipeline. In particular, pipeline operators are interested in the state of welds that may be present and the number and size of defects. Defects are, for example, areas with metal loss caused by corrosion, cracks or other weakening of the wall of the object, which is in particular intended for the storage or transportation of liquid or gaseous media. Objects include, for example, pipelines, pipes or tanks.

[0004] For the adjustable operating pressure in the pipeline, it is important to know the maximum pressure applicable to the pipeline ("maximum burst pressure"), in the event of which the pipeline is damaged. Accordingly, it is important to accurately predict this pressure. Currently, for the calculation of this limit value, defects are only approximated in terms of their length, width and depth and are therefore considered as boxes. In particular for metal losses caused by corrosion on the outside or inside of the pipeline, this necessarily conservative approach is disadvantageous, since the simplified geometry necessarily overestimates the current structure of the defect. This leads to an underestimation of the maximum burst pressure of the object and therefore an underestimation of the permitted operating pressure. Objects such as pipelines or gas tanks that can be operated at higher pressures can be operated in a significantly more economical manner.

[0005] The prior art for measuring corrosion of an object is to evaluate a scan of magnetic flux leakage data (MFL data) by specially trained personnel based on magnetic flux leakage measurement (MFL measurement) in order to determine the size of the (corroded) defects. The signals displayed in the scan are parameterized and evaluated in a box. On the one hand, the assumptions required to evaluate the measurement results are proprietary, and this evaluation of the measurement results is also called sizing. On the other hand, the interpretation of the measurement results is strongly influenced by the experience of the evaluator. Finally, the quality of the prediction can only be ensured with the help of on-site inspections of the pipeline. This in turn brings high costs to the operator. The most widely used industrial standard API 1163 describes the disadvantages of this simplified approach. It has been well documented that the quality of this approach is highly dependent on the knowledge of the inspector. The approach implemented in practice is always an interpretation of the data obtained by the MFL measurement process, which is influenced by subjective factors.

[0006] Furthermore, from scientific, i.e. theoretical considerations, approaches are known in which the aim is to simulate the measured signal as accurately as possible by means of a forward model by means of a continuous variation of the assumed defect geometry and an iterative method. In this case, neural networks are used, for example. In theory, these approaches can lead to solutions based on the resulting defect geometry, but these solutions are not necessarily practical. This applies in particular to complex data sets, which in the case of certain inverse problems lead to undesirable, special and erroneous solutions. Although such scientific models lead to solutions to the problems in the form of defect geometries in precisely defined and restricted test scenarios, this has not been successful so far for actual measurement data with multiple interfering influences. Summary of the invention

[0007] The object of the present invention is therefore to specify a robust method for accurately reconstructing the actual defect from an MFL measurement data set of the defect and for carrying out a calculation of the load capacity limit of an object affected by corrosion that is as accurate as possible.

[0008] This object is achieved by a method according to claim 1, and the determination of the load capacity limit is achieved by a method according to claim 15. Advantageous embodiments of the invention can be gathered from the dependent claims which refer back to these claims and from the subsequent description.

[0009] According to the invention, provision is made for the determination of the geometry of one or more defects to be carried out in parallel, in particular on an EDV unit, by means of a plurality of expert routines competing with one another, each having at least one dedicated search strategy or at least one dedicated algorithm.

[0010] This method according to the invention is used for determining the geometry of one or more actual defects of a magnetizable object, in particular a pipeline or a tank, being inspected, using a reference data set of the object, generated on the basis of one or more MFL measurements, the method preferably comprising at least partially representing the object by means of an EDV unit, in particular on or by an at least three-dimensional object grid; and the method further comprises determining, in particular generating, an initial defect geometry as a starting defect geometry, in particular on the object grid or on an at least two-dimensional defect grid;

[0011] determining, in particular generating, a first MFL prediction data set as a starting prediction data set based on the starting defect geometry, in particular by simulating an MFL measurement or assigning an MFL data set; and

[0012] The starting defect geometry is iteratively matched to the geometry of one or more actual defects with the aid of an EDV unit, the matching being performed with the aid of a plurality of expert routines which compete with one another and preferably run in parallel with one another, wherein in the respective expert routine a respective expert defect geometry is generated with the aid of at least one dedicated algorithm or a dedicated search strategy and based on the starting defect geometry.

[0013] An expert routine has a dedicated algorithm if at least one of the algorithms available to the expert routine for matching the defect geometry differs at least in part from the algorithms of another expert routine. Preferably, a random process can be used to distinguish the algorithms of different expert routines. Each expert routine has at least one algorithm for matching the defect geometry, preferably at least one expert routine has a plurality of algorithms available. Likewise, the selection of an algorithm can be made or specified within an expert routine based on a random process.

[0014] According to the invention, it is also provided that a corresponding expert prediction data set is determined based on the corresponding expert defect geometry, in particular by simulating MFL measurements or assigning MFL data sets, wherein if the expert prediction data set is more similar to the reference data set than the starting prediction data set, the expert defect geometry on which the corresponding expert prediction data set is based is provided to at least one, preferably a plurality, and in particular all of the expert routines as a new starting defect geometry for further matching to the geometry of one or more actual defects. Subsequently, i.e. for the next comparison of the corresponding expert defect geometry with the new starting defect geometry, the expert prediction data set belonging to the new starting defect geometry is used as the new starting prediction data set. A measure of similarity can be formed by a fitness function. The matching is iterated with the aid of the expert routines until a stop criterion is met. Both the expert prediction data set and the first MFL prediction data set show MFL fields belonging to the correspondingly assumed defect geometry. These MFL fields can be calculated or simulated.

[0015] The expert prediction data set can be determined within the workflow of the expert routine and / or by a program module which is separately controlled by the monitoring routine.

[0016] Furthermore, in particular if a sufficiently large database with calculated or measured MFL data for the corresponding defect geometry is not yet available by means of the simulation of the MFL measurement described further below, an expert prediction data set is determined based on the corresponding expert defect geometry. Alternatively, the expert defect geometry can also be provided with an MFL data set from a sufficiently large database. Also possible is a combined method in which the database is first searched for already existing MFL data and a simulation is subsequently performed in the event of a negative search result. Overall, this can lead to a rapid determination of the corresponding expert prediction data set.

[0017] The method according to the invention is carried out completely and in particular automatically on an EDV unit which may be composed of a plurality of computers if necessary. The associated computer program may be a single program or a program package comprising a plurality of program modules which are run, for example, in a distributed manner depending on resources on different EDV systems or EDV subunits and which can be stored on corresponding EDV media at the different EDV systems or EDV subunits. In particular, the computer has typical means of a data processing unit such as one or more processors, at least a temporary memory (RAM), a data communication device, a display unit and / or an input unit. The selection of the reference data set can preferably be carried out in a user-controlled manner, while the determination of the defect geometry is automatically carried out during the iteration. Preferably, program parameters for selecting an algorithm available for the expert routine, for determining the initial defect geometry, for determining a first prediction data set and / or an expert prediction data set, respectively showing the MFL field, can also be set before the actual iteration. For example, it can be set whether the determination of the starting prediction data set should be created by simulating an MFL measurement based on a grid representing the object with the defect, or whether the determination of the starting prediction data set should be loaded from a database by regression. In particular, parameters required for the comparison with a reference data set, such as magnetization direction, magnetization intensity, distance of the sensor from the object surface and / or speed of the measuring device, can be set in the case of a simulated MFL field.

[0018] In particular, the problem of finding isolated solutions present in the theoretical approaches of the prior art is avoided by expert routines that compete for the resources of the EDV unit, and which, with their dedicated search strategies, respectively search for a dedicated solution for determining the geometry of the actual defect. The solutions found are not only significantly better than those found by manual evaluation of the data set, but are also better understood and justified. In this way, singular solutions are avoided, in which no algorithm of any kind can proceed and the defect geometry is still not realistically reproduced.

[0019] For the simulation of magnetic flux leakage data associated with geometric structures, it is usually necessary to represent the object on or by a three-dimensional grid. The representation is realized on this three-dimensional grid, at least partially in the sense that at least a part of the object with one or more defects and preferably adjacent areas are represented by or on the object grid. Alternatively, the determination of magnetic flux leakage data can also be carried out by means of a database query, for example by means of a regression function.

[0020] Although the method according to the invention can be carried out simultaneously in a reference data set to determine one or more defects, for the sake of simplicity only one defect will mostly be discussed below.

[0021] For the MFL simulation, in the case of a cleverly chosen representation of the object, in particular via or on an at least three-dimensional object mesh, the defect geometry can be assigned as values ​​to mesh elements or mesh points. Depending on the geometry of the respective mesh, an interpolation or mesh adaptation is required for this purpose.

[0022] In particular, the defect geometries are defined on the grid nodes of the two-dimensional defect grid by, for example, the defect depth representing the corrosion depth of the object surface, which are the basis for determining the associated (MFL or expert) prediction data set. With the two-dimensional representation of the defect, the expert routine can run significantly faster than performing the matching of the defect geometry on a three-dimensional grid.

[0023] For the simulation of the MFL measurement of a new defect geometry, a particularly two-dimensional defect grid is preferably interpolated onto the grid points of the object grid, wherein the surface of the object to be represented is matched to the defect depth of the defect geometry. The simulation is then calculated on the particularly three-dimensional object grid. Alternatively, the magnetic flux leakage simulation can also be performed on a two-dimensional grid or with the aid of a regression model, which is constructed on a database with MFL data sets based on finite element method simulations and / or MFL measurements.

[0024] An initial defect geometry is obtained or specified, for example by a table lookup, a database comparison or in particular by executing an expert routine, on the basis of which a first MFL prediction data set is determined as a starting prediction data set, in particular by simulating an MFL measurement. For example, the simulation of the MFL measurement is performed in a forward calculation with the aid of a finite element model. In the simulation of the magnetic flux leakage measurement, the parameters required for this are set according to the actual measurement. In particular, this relates to the magnetization direction, the magnetization intensity and / or the distance of the sensor from the object surface. Then, based on the initial defect geometry, a starting prediction data set is derived as a simulated magnetic flux leakage measurement. This data set can already be compared with a reference data set of the object, but this usually does not lead to a convincing solution at the beginning of the iteration.

[0025] A separate routine for setting the initial defect geometry as the starting defect geometry is not absolutely necessary, but reduces the required computing time in the subsequent program flow. Alternatively, the initial defect geometry can already be the result of running the expert routine. Alternatively, the initial defect geometry can also be specified, for example, by a completely flat, so-called defect-free geometry.

[0026] The initial defect geometry is introduced as the starting defect geometry into the iterative approximation process of the competing expert routines. The expert routines themselves are independent of each other, for example as dedicated program modules that do not directly interact with each other and can be equipped with resources, especially computing time, depending on the monitoring routine or the main module.

[0027] Based on the expert defect geometries developed in the respective expert modules, an expert prediction data set is then determined for in particular each of these geometries, in particular by simulating an MFL measurement. Thus, for each expert defect geometry present, in particular as a 2D data set of depth values ​​of the defect, an expert prediction data set is generated as a simulated MFL measurement. Based on the aforementioned calculation of the starting prediction data set, the MFL field is simulated based on the respective expert defect geometry. In particular, the calculation is performed in parallel based on the respective expert defect geometry. This can be accompanied by the construction of a database in which the magnetic flux leakage fields belonging to the respective defect are stored, in order to save later calculation time for other similar data.

[0028] Before the expert prediction data set is generated, it is advantageous to adapt, in particular partially refine, the mesh on which the expert defect geometry calculation is based, in particular the defect mesh and, if necessary, the object mesh. In particular, mesh deformation techniques can be used for this purpose, in which the object mesh or defect mesh is refined by moving and / or dividing mesh points, in particular in areas with large gradients, in order to achieve a more accurate determination of the geometry or a subsequent more accurate simulation. In other areas with smaller gradients, the mesh can be coarser in order to save computing time. The applied mesh is thus automatically adapted to optimally evaluate the defect geometry. At the same time, a significant reduction in the number of unknowns is thereby achieved, thereby further saving computing time.

[0029] If the comparison between the reference data set and the expert prediction data set of an expert routine results that the expert prediction data set is closer to the reference data set than the previous starting prediction data set, the associated expert defect geometry is provided as starting defect geometry for the further expert routine and the corresponding expert routine. Based on this solution, the further expert routine can start from this geometry in the next iteration step, unless the further expert routine finds a better solution, for example during a still running dedicated defect geometry determination, which solution is then provided to the further or all expert routines.

[0030] Among the expert routines that compete with each other, preferably an expert routine is preferred in terms of the available resources of the EDV unit, which, as described below, is more successful in approximating the actual measurement data than other expert routines that are in competition. In particular, the resources of the EDV unit are the priority in the allocation of CPU time or GPU time and / or memory.

[0031] Advantageously, the expert routines (on the EDV unit) are run in competition with one another, so that resources of the EDV unit, in particular in the form of computing time, are allocated to the respective expert routines according to a success rate and / or according to a reduction in a fitness function, in particular taking into account the number of starting defect geometries calculated by the expert routine and available to one or more other expert routines, and in particular taking into account the number of expert prediction data sets generated for this reduction. In particular, if the respective expert routine is more successful than the other expert routines, more resources, in particular in the form of computing time, preferably CPU time and / or GPU time, are allocated to the respective expert routine by the program part designed as a monitoring routine, thus resulting in a competition among the expert routines. An expert routine is successful if it finds a defect geometry with a reference data set, for example a simulated magnetic leakage measurement, which is more suitable and available to other expert routines.

[0032] For example, this can result in a particularly successful individual expert routine obtaining more than 50% of the total available computing time, which significantly reduces the overall duration of the method according to the invention. At the same time, it can be provided on the program side that no expert routine or individual expert routines of the expert routines do not fall below a certain percentage of the computing time, in order to avoid problems with idiosyncratic and special defect geometries or results from individual routines. Thus, a way out of a blocking situation that would otherwise occur in the prior art can be found in the case where a previously successful expert routine has only found a local solution and not a global solution.

[0033] The adaptation is performed with the aid of an expert routine until a stop criterion is met. Here, the stop criterion is, for example, a residual of the measured measurement data and the simulated measurement data. The stop criterion can also be an external stop criterion, for example based on the available running time, or a particularly predefinable number of iterations, or a particularly predefinable running time, or a predefined running time, or a running time determined from the available running time.

[0034] It has been found that the accuracy of the defect determination is qualitatively improved by the method according to the invention. The resulting calculation of the maximum load capacity shows, for example, that the pipeline can be operated for a significantly longer time. The accuracy of the defect determination is significantly improved. The maximum operating pressure resulting from the simulated defect geometry according to the method described above and below can be estimated up to 50% higher, which significantly reduces the maintenance and servicing costs for the operation of the pipeline and its operator. From now on, for the first time, the ASME B31G-2012 Class 2 solution ("river bottom profile") for the "residual strength algorithm" can be appropriately determined for the MFL data set.

[0035] Preferably, the resources of the EDV unit, in particular in the form of CPU time, are allocated to the respective expert routines as a function of the number of starting defect geometries provided by the respective expert routine for all expert routines. This can be, for example, the number of slots for calculating the expert prediction data set in the form of a simulated MFL data set, the number of processor cores for parallel processing of computing tasks, etc. It can also be provided in the scope of a computer program product for implementing the method according to the invention that the computer program product is adapted to the resources present in the EDV unit in the form of processor cores, memory space, memory architecture, graphics card, etc. By giving priority to particularly preferred expert routines and their algorithms, the actual defect geometry is detected significantly faster.

[0036] In order to further minimize the problem of singular solutions, local solutions, it is particularly provided that a further reference data set is additionally used which is linearly independent of the first reference data set to determine the geometry of one or more defects. If the data sets are generated by MFL measurements with mutually angled magnetizations of the object, these data sets are linearly independent. If the corresponding average induced magnetic field strengths do not run parallel or equally in the examined area, the magnetizations are angled with respect to each other. In particular, the angle is between 40° and 140°, preferably between 80° and 100°, and particularly preferably 90°. Based on the starting defect geometry, a further starting prediction data set is determined, in particular by means of a further MFL simulation which takes into account linear independence, i.e. in particular takes into account different magnetizations; and the expert defect geometry is used as the starting defect geometry only if the associated expert prediction data set determined for the two independent magnetizations is more similar to the corresponding reference data set than the starting prediction data set determined for the two magnetizations, and / or the fitness function taking into account the two expert prediction data sets is improved. The risk of singularities is further reduced by the parallel or concomitant processing of two linearly independent data sets and the use of a starting defect geometry whose simulated measured data must be generally good in terms of similarity or fitness functions. At the same time, the quality of the starting prediction data set available to all expert routines improves. As a result, the number of iterations can be further reduced.

[0037] In particular, a first reference data set is generated by an MFL measurement with magnetization in the axial direction, and a second reference data set is generated by an MFL measurement with magnetization in the circumferential direction of the pipeline. Here, the respective magnetizations of the pipeline or object are at right angles to each other, so that a maximum amount of information can be obtained from the magnetic flux leakage measurement, which is fully available during the calculation by simultaneously considering the corresponding reference data set and the simulated expert prediction data set. The method steps described below take the above into account, similarly to when using two reference data sets generated based on linearly independent magnetizations.

[0038] By using a starting prediction data set and / or an expert prediction data set based on a forward model for simulating the MFL measurement, in particular by means of a finite element model, the MFL measurement simulation is quickly performed. The simulation of the magnetic flux leakage measurement based on the expert defect geometry is realized by a dedicated program module, which is in particular controlled and / or monitored by the monitoring routine and is called individually by the respective expert routine. This can also be a plurality of modules, which are available to the respective expert routine in a distributed manner on the respective computing units.

[0039] Preferably, the initial defect geometry is generated by one of the expert routines and / or by a machine learning algorithm with the aid of a look-up table, which, as described above, improves the overall computation time, in particular when a mesh matching has been additionally performed.

[0040] In particular, a refinement of the object mesh and / or the defect mesh can be performed in regions where the depth of one or more simulated defects exceeds a predeterminable threshold value, so that only gradients exceeding a certain size lead to a change in the mesh. For this refinement, in particular the total number of gradients of the new expert defect geometry can be taken into account in order to achieve a balance between the adaptation of the corresponding mesh, in particular the object mesh, and the subsequent computational operations.

[0041] The mesh refinement for the purpose of reducing the computation time can be performed not only based on the initial reference data set, but also before calculating the expert prediction data set. For this purpose, a separate program module or individual submodules of the corresponding expert routine can also be provided.

[0042] In particular, refinement of the object mesh and / or defect mesh is particularly advantageous in reducing the required CPU time by significantly reducing the number of independent variables used to simulate MFL measurements in the forward algorithm through grid point movement and / or grid point division. Grid point movement can also be used for matching of the object mesh or defect mesh.

[0043] Preferably, a fitness function is used as a measure of the similarity of the expert prediction data set to the reference data set in order to compare the simulated data set to the measured data set based on a standard routine and suitably quickly, ie while saving computing time.

[0044] In particular, the starting defect geometry in the form of a two-dimensional data set or a pointer to the starting defect geometry is stored in a memory area of ​​the EDV unit accessible to all expert routines. In particular, this memory area is also controlled by the monitoring routine, so that the individual expert routines can also be prioritized in this regard.

[0045] Instead of using a starting defect geometry, for example stored in a central storage area accessible to all expert routines, each time a new iteration is started, at least one expert routine can adapt its dedicated expert defect geometry without adopting the starting defect geometry when a new iteration is started. To this end, the expert routines can have functional rules in which, for example, opposite strategies are consciously selected depending on the search strategies used in other expert routines. In such a case, the expert routines can indirectly influence each other. In particular, this method can be advantageous if it is found that a routine that has been successful so far prefers unrealistic solutions. This can be detected, for example, based on values ​​that are not allowed with respect to the defect depth. If an expert routine that does not adopt the starting defect geometry does not provide an improved solution, this expert routine is automatically downgraded so that less and less computing time is available to this expert routine.

[0046] Preferably, the absence of what can be called a substantial change in the starting defect geometry, or more generally the geometry of the defect mesh and / or the geometry of the object mesh after a number of iterations can be used as a stopping criterion or convergence criterion. The solution found up to that point is the best solution. Preferably, the stopping criterion is selected such that the changes observed in the magnetic flux leakage simulation, which lead to the refinement of the object mesh or defect mesh, are substantially smaller, for example by a factor of less than 10, in particular by a factor of 2, than the following changes, which arise from individual measurement deviations, in particular measurement errors, and which are individually specified, for example, based on the so-called "Essential Variables" of the API 1163 standard. It is thereby achieved that the accuracy of the final model is within the range of accuracy specified by the measurement itself. Accordingly, a comparison of the changes of the expert prediction data set with the measured deviations of the actual data set is preferably used as a stopping criterion. The stopping criterion that causes the program to be stopped and in particular the output or storage of the starting defect geometry calculated so far can preferably be specified by a program parameter that can be set in advance.

[0047] In particular, a plurality of algorithms for matching the expert defect geometry are available to the expert routine. These algorithms can be approaches from the field of machine learning, stochastic optimization, empirical model functions and / or numerical model functions. In particular, the experience of the evaluator can also be used in the expert routine. Preferably, in the expert routine, different algorithms are generated randomly or selected by a selection function. Thus, a sufficiently diverse approach is generated, with which all solutions can be considered in a targeted manner under competitive conditions.

[0048] The object set out above can also be achieved by a method for determining the load capacity limit of an object which is loaded with pressure at least during operation and is designed in particular as an oil pipe, gas pipe or water pipe, wherein in the method a data set describing one or more defects is used as an input data set in the calculation of the load capacity limit, which is designed in particular as forward modeling, wherein the input data set is first generated according to the method for determining the defect geometry described above or below. The advantageous representation of the defect geometry, in particular as an unparameterized real three-dimensional geometry or as a two-dimensional surface with corresponding depth values, makes simplifications that have hitherto been considered necessary in the industry superfluous, so that for this reason an increase in the accuracy of the defect determination is ensured in a way that has not been possible before.

[0049] Whereas previously the accuracy was limited to the indication of the maximum depth point of the defect, now all characteristics are determined with high accuracy. Typically, the accuracy of the maximum depth is increased to the extent that can be achieved due to the measurement accuracy, namely to approximately ±5% of the wall thickness, compared to approximately ±10% of the wall thickness in the prior art previously used for dimensioning. However, the prediction of the load capacity limit based on the defect geometry, especially for critical cases, has achieved an increase in accuracy from, for example, ±50% previously to ±5% now. The advantage according to the invention is therefore that for the first time a suitable representation of the defect geometry is made possible, which for the first time makes this increase in accuracy possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Further advantages and details of the present invention can be obtained from the following description of the drawings, in which:

[0051] Figure 1 Schematically illustrates defect determination according to the prior art;

[0052] Figure 2 Schematically shows a schematic diagram of the method according to the present invention;

[0053] Figure 3 Schematically shows Figure 2 Further explanation of part of;

[0054] Figure 4 Schematically shows a comparison of the results of the method according to the invention with measured data;

[0055] Figure 5 schematically shows a schematic diagram of mesh refinement as part of the method according to the present invention;

[0056] Figure 6 The result of the method according to the invention is schematically shown. DETAILED DESCRIPTION

[0057] Individual features of the exemplary embodiments described below can be combined with the features of the independent claims to produce improvements according to the invention.

[0058] In the prior art, according to Figure 1 The MFL data of the pipeline are evaluated with the aid of a special, also empirically based definition of a box. The box shown in the figure has corresponding dimensions in length, width and depth. The x-axis and the y-axis are shown in meters ([m]). An examination of the actual defect geometry on which the evaluation is based by means of laser scanning, i.e. by means of direct measurement, shows that the maximum bursting pressure that can be determined based on the defect geometry assumed by evaluating the MFL data is 4744.69 kPa, which is only 55.2% of the maximum bursting pressure calculated based on the actual geometry. With the prior art, the operating pressure for safe operation of the pipeline is 3621.29 kPa according to an empirically based evaluation, which is significantly lower than the possible safe operating pressure.

[0059] In the method according to the invention, the surface of the pipeline is represented according to one embodiment by a 2D mesh surface. The defect geometry can be described as a vector of depth values ​​D on the defect grid 5 ( Figure 5 ). Based on the result of the fitness function F(D) taking into account the MFL field belonging to the corresponding geometry, the defect geometry is compared with the starting defect geometry. It is assumed here that the smaller the value of the fitness function, the closer the assumed expert defect geometry is to the actual geometry:

[0060]

[0061] Here, M is the number of datasets that can be processed simultaneously (actual MFL datasets), and H is the cal is the simulation result of MFL measurement, H m is the measured data from the MFL measurement (reference data set), and R(D) is the regularization term, which can be estimated as follows:

[0062]

[0063] where α is a scaling term.

[0064] According to the following Figure 2 At least a part of the method sequence according to the invention is described, wherein only one block 14 is used to describe a plurality of expert routines 11 which are parallel and compete with one another.

[0065] For example, multiple processes of the same MFL pipe cleaning machine can be fused as input data sets according to block 2. For better fusion, the two data sets 1 can be filtered in advance and adapted to each other (method step 3) in order to reduce possible artifacts or background noise, for example. It is also possible to use another data set 4, which is generated based on another linearly independent magnetization and is likewise filtered for the purpose of adapting to the same grid structure, so that according to method step 6, there are two adapted reference data sets, which are obtained based on measurement processes with linearly independent magnetizations.

[0066] Data sets that completely match each other can be processed together, wherein the method according to the present invention enables simultaneous processing of the data sets by using a fitness function that takes into account the fused data set.

[0067] In step 7, the first reference data set of the reference data sets present in step 6 is selected for further processing. Furthermore, in step 8, an initial defect geometry is first assumed, in particular generated here, as a starting defect geometry, which is based on the normalized measurement signal S(x,y) / (max S), for example. For example, the defect geometry can be derived from a threshold function that takes into account the amplitude at the grid points where the signal is greater than a certain limit value l (for example 0.2):

[0068]

[0069] The above approximation results in N defect depth values ​​at the corresponding grid points:

[0070] D i =i wt / N*G

[0071] Where wt is the thickness of the pipe wall. i is an index that is also used as a value for determining the defect depth value. The fitness function is calculated for such defect geometry and the feature with the smallest function value is used as the initial solution:

[0072] D init = arg min F(D i )

[0073] This initial solution is then available to the various expert modules as a starting defect geometry. Beforehand, in order to reduce the computation time, the number of parameter values ​​(elements of the vector D) describing the defect geometry can be kept as small as possible. In particular, this is achieved by dynamic mesh matching. Since the number of depth values ​​corresponds to the number of nodes in the defect mesh 5, the number of nodes is also the number of defect parameters at the same time. Starting from a relatively coarse mesh, the mesh is gradually refined in the relevant area.

[0074] By way of example, for a preset node distance of, for example, 14 mm, an associated grid cell size of 14 mm×14 mm and defect limit values ​​of 30%, 50% and 80% of the wall thickness, it is possible to achieve in the relevant grid area Figure 5 The refinement shown in , where cells above the above depth value are gradually subdivided. The mesh deformation is related to the assumed defect geometry, i.e. there is a larger number of mesh points in the regions of larger gradients.

[0075] The workflow of a group of competing expert routines 11 is simulated on the EDV unit according to the method according to the invention. For this purpose, the program can have different modules which are independent of one another and which can in particular place data in a manner asynchronous with one another into specific areas of the EDV unit so that these data are further processed in these specific areas. This is carried out in particular under the supervision of the monitoring routine 9 ( Figure 3 ). Thus, depending on the success defined above, i.e. depending on the number of starting defect geometries written, for example, into the common storage area 12, the plurality of expert routines 11 maintain a plurality of computation slots 13 in order to respectively generate expert defect geometries and / or to be able to perform associated MFL simulations or, in the case of separate MFL simulation modules, to perform associated MFL simulations. This corresponds to the number of expert routines 11 being able to perform associated MFL simulations in accordance with the above definition of success, i.e. depending on the number of starting defect geometries written, for example, into the common storage area 12. Figure 2 14, wherein the frame exemplarily represents a plurality of expert routines 11 ( Figure 3 ). According to this embodiment, an MFL simulation of each expert defect geometry is performed in a simulation module 16 based on each calculation slot 13, also under the supervision of the monitoring routine 9, in order to create an expert prediction data set. The more slots 13 are available to an expert routine, the greater the proportion of EDV resources used for this expert routine. Preferably, the number of program modules performing the MFL simulation is equal to the number of slots. The monitoring routine 9 monitors the number of iterations and the resulting changes in the starting defect geometry, and also monitors whether the associated stop criteria are reached. Subsequently, the result is output according to box 17.

[0076] The number of computing slots 13 available to the expert routine 11 and the simulation routines subsequently available can vary, so that the first expert routine can, by way of example, utilize up to 50% of the total computing time available for computing slots and simulation routines.

[0077] As shown, the starting defect geometry is stored in a storage area 12. In this case, the storage area 12 can be a storage area accessible to the expert routine 11. In such a storage area, log files of the expert routine 11 and the monitoring routine 9 as well as instructions for the expert routine 11 can also be stored, which are then independently executed by these expert routines. In this case, these instructions can be, for example, interrupt commands that are set when a stop criterion is reached.

[0078] Preferably, the expert routine 11 is an independent program module which generates new expert defect geometries and places the expert defect geometries into the simulation routine 16. The fitness function shown above can also be generated in the expert routine 11 based on the expert prediction data set and compared with the starting prediction data set stored in the area 12. If the expert prediction data set is more similar to the reference data set overall than the data set stored in the area 12, or to both reference data sets respectively in the case of linearly independent measurement data sets, then the expert prediction data set is used as the new starting prediction data set.

[0079] For example, new defect geometries are randomly generated in the expert routine 11. For this purpose, machine learning algorithms or empirical rules can be used. Advantageously, at least two implementations of the basic expert routines are provided as described below in order to further improve the convergence of the solution.

[0080] These search strategies, which are preferably always implemented in the method according to the invention, are based on an assumed probability distribution p(x,y) of the grid points whose depth values ​​produce the largest reduction in the fitness function. The probability function is used to identify N grid points (x n ,y n ). At each of the considered points, for example, a depth function describing the corrosion depth at the grid position changes ΔD, where the sign of the change is distributed in a randomly generated manner. The number of selected points N can also be randomly selected:

[0081]

[0082] By choosing the probability function p(x,y), different expert strategies can be implemented, for example:

[0083]

[0084] This algorithm implements a variation of the defect depth, where the grid point with the largest depth is preferred. Another strategy may look as follows:

[0085]

[0086] Such an algorithm changes the defect geometry at a location where the best known solution H the best The simulated measurement signal has a maximum difference from the measured signal.

[0087] On this basis, different expert routines or algorithms of expert routines can be formed by changing the number of grid points to be considered and the change of ΔD. Exemplarily, the following six expert routines can be used:

[0088] 1. N = 1 and ΔD = 1% wall thickness

[0089] 2. N = 2 and ΔD = 5% wall thickness

[0090] 3. N = 3 and ΔD = 5% wall thickness

[0091] 4. N = 1 and ΔD = 1% wall thickness

[0092] 5. N = 2 and ΔD = 5% wall thickness

[0093] 6. N = 3 and ΔD = 5% wall thickness

[0094] As described, Figure 3 The monitoring routine 9 shown in has two functions in particular: the first is to check whether the stopping criteria have been reached, and the second is to manage the resources of the EDV unit between the various experts based on their success. The measure of success is:

[0095]

[0096] where ΔF is the decrease in the fitness function F caused by the results of the corresponding expert routine, and N is the number of simulations required for this. The evaluation of n expert routines can be assumed to be:

[0097]

[0098] The number of computational slots used by the expert routine in an iteration is N S for:

[0099] N S = int(R n N all )

[0100] Where N all is the number of all slots available.

[0101] The MFL measurement for the expert defect geometry is simulated in the simulation routine 16. The expert routine may iterate until the expert routine finds a solution whose expert prediction data set is better than the starting prediction data set stored in area 12. If this is the case, the expert routine 11 may process another linearly independent data set or attempt to achieve another better solution based on the already improved solution.

[0102] If the expert routine iteratively calculates a plurality of data sets from different iterations that cannot be adapted consistently, the resulting geometries can also be automatically superimposed within the scope of carrying out the method according to the invention, with the maximum depth being used as a conservative estimate at the individual grid points:

[0103]

[0104] For n=1...N, where N is the number of data sets to be processed in sequence. Based on the depth characteristics generated on this superposition of the defect geometry, the MFL signal can be simulated again. The resulting error can be derived from the error of the corresponding data set in the individual calculations:

[0105] E=||H cal (D)-H m ||

[0106] In order to demonstrate the efficiency of the proposed method, several test scenarios were performed, in which the following Figure 4 Data from two MFL detection processes performed with mutually linearly independent magnetizations were used.

[0107] Figure 4 The pictures numbered 21 show actual MFL measurements with magnetization running in the axial direction, while FIG. 22 results from measurements in the circumferential direction. The result of the defect geometry calculation obtained according to the method according to the invention described above is indicated by 23. The contour lines evenly divide the area between 0 and 60% of the metal loss depth as in picture 24. FIG. 24 shows the actual scanned and thus directly measured outer surface of the pipe section corresponding to FIGS. 21 and 22. The laser scanning measurement is in very high agreement with the solution achieved by the method according to the invention. This is significantly better than the solution according to the evaluation known from the prior art. It can be assumed here that the differences between the results according to the method according to the invention and the laser scanning measurement exist mainly due to technical tolerances.

[0108] Based on the prior art established and Figure 1 The conventional consideration of the determination of the defect geometry shown in the results leads to the mentioned maximum bursting pressure of 4744.69 kPa. Figure 1 The MFL dataset based on Figure 6 The defect geometry shown in (contour at a depth of 2 mm) and based on this a maximum bursting pressure of 8543.46 kPa is obtained. Here, the maximum bursting pressure is close to 99.4% of the maximum bursting pressure determined based on the actual defect geometry determined by the laser scan. Therefore, the pipeline inspected according to the method according to the invention can be operated at a safe operating pressure of 6520.53 kPa. Compared with the evaluation based on the prior art ( Figure 1 ) of 3621.29 kPa, which brings significant advantages to the pipeline operator. Based on the method according to the invention, the state of the pipeline can be given significantly more realistically, and thus the pressure that can be given for the safe operation of the pipeline can be given significantly more realistically, while the safety of operation is ensured as always. By means of the method according to the invention with the expert routine of the resources of the competing EDV unit, such results can be provided to the pipeline operator more quickly or at least in the same evaluation time as in the prior art.

Claims

1. A method for determining the geometry of one or more actual defects of a magnetizable object, in particular a pipeline or a tank, using a reference data set generated on the basis of one or more MFL measurements of said object, Preferably comprising at least partially representing the object by means of the EDV unit, in particular on or through an at least three-dimensional object grid; and comprising determining an initial defect geometry as a starting defect geometry, in particular on said object grid or on an at least two-dimensional defect grid (15); determining a first MFL prediction data set as a starting prediction data set based on the starting defect geometry, in particular by simulating an MFL measurement or assigning an MFL data set; as well as iteratively adapting the starting defect geometry to the geometry of the one or more actual defects by means of the EDV unit and a plurality of expert routines (11) running competitively with one another and preferably running in parallel with one another; wherein in a corresponding expert routine (11) a corresponding expert defect geometry is generated by means of at least one dedicated algorithm and based on the starting defect geometry; In particular, determining a corresponding expert prediction data set based on a corresponding expert defect geometry by simulating an MFL measurement or assigning an MFL data set; as well as If the expert prediction data set is more similar to the reference data set than the starting prediction data set, the expert defect geometry on which the corresponding expert prediction data set is based is provided to at least one, preferably a plurality, in particular all of the expert routines (11) as a new starting defect geometry for further adjustment to the geometry of the one or more actual defects, and the expert prediction data set belonging to the new starting defect geometry is subsequently used as a new starting prediction data set; wherein an iterative adjustment is performed with the aid of the expert routine until a stop criterion is met, Additionally using a further reference data set which is linearly independent from the first reference data set with respect to the magnetization to determine the geometry of the one or more defects; and based on the starting defect geometry, determining a further starting prediction data set, in particular by means of a further MFL simulation taking into account linear independence; and using the expert defect geometry as the starting defect geometry only if an associated expert prediction data set determined for two independent magnetizations is more similar to the corresponding reference data set than the starting prediction data set determined for the two magnetizations, and / or a fitness function taking into account both expert prediction data sets is improved.

2. The method according to claim 1, It is characterized in that The expert routines (11) are run in competition with one another so that the resources of the EDV unit, in particular in the form of computing time, preferably CPU time and / or GPU time, are allocated to the corresponding expert routines according to the success rate and / or according to a reduction in the fitness function, in particular taking into account the number of starting defect geometries calculated by the expert routine and available to one or more other expert routines (11), and in particular taking into account the number of expert prediction data sets generated for the reduction.

3. The method according to any one of the preceding claims, It is characterized in that The fitness function is used as a measure of the similarity between the expert prediction dataset and the reference dataset.

4. The method according to any one of the preceding claims, It is characterized in that The first reference data set is generated by MFL measurements with magnetization in an axial direction, and the second reference data set is generated by MFL measurements with magnetization in a circumferential direction of the pipeline.

5. The method according to any one of the preceding claims, It is characterized in that A starting prediction data set and / or an expert prediction data set is generated based on a forward model for simulating the MFL measurement and in particular with the aid of a finite element model.

6. The method according to any one of the preceding claims, It is characterized in that The initial defect geometry is generated by one of the expert routines (11) and / or by a machine learning algorithm with the aid of a look-up table.

7. The method according to any one of the preceding claims, It is characterized in that The defect grid is refined in regions where the depth of one or more simulated defects exceeds a threshold value (5).

8. The method according to any one of the preceding claims, It is characterized in that The object mesh and / or the defect mesh are refined before calculating the corresponding expert prediction data set (15).

9. The method according to any one of the preceding claims, It is characterized in that The starting defect geometry or a pointer indicating the starting defect geometry is stored in a memory area (12) of the EDV unit accessible to all expert routines (11).

10. The method according to any one of the preceding claims, It is characterized in that The absence of substantial changes in the starting defect geometry and / or the geometry of the object mesh and / or the geometry of the defect mesh (15) and / or the starting prediction dataset and / or at least one expert prediction dataset after a number of iterations is used as a stopping criterion.

11. The method according to claim 10, It is characterized in that A comparison of the changes in the expert predicted dataset with the measured deviations of the actual dataset is used as a stopping criterion.

12. The method according to any one of the preceding claims, It is characterized in that An expert routine (11) is assigned with a plurality of algorithms for adjusting the expert defect geometry, the plurality of algorithms comprising machine learning, stochastic optimization, empirical model functions and / or numerical model functions.

13. The method according to claim 12, It is characterized in that The algorithm is generated randomly in the expert routine (11), or selected and / or changed by a selection function.

14. A method for determining the load capacity limit of an object which is subjected to pressure at least during operation and is designed in particular as an oil pipe, gas pipe or water pipe, wherein a data set describing one or more defects is used as an input data set for calculating the load capacity limit, It is characterized in that Firstly, the input data set is determined according to the method according to one of the preceding claims.

Images