DE102024110510A1 - Method for operating a computer tomograph and device for carrying out - Google Patents

Abstract

The invention relates to a device and a method for correcting artifacts in measurement data that have been determined by means of computer tomography or laminography measurement (computer tomography) of at least one workpiece or workpiece area, wherein the measurement data are radiographic image data and/or volume data reconstructed therefrom and/or surface data determined therefrom, such as STL data. In order to ensure that meaningfully effective correction methods for artifacts are used in a computer tomography measurement of workpieces, it is provided that by means of a computer-implemented method (identification software) on the basis of prior knowledge of the effect of artifacts in measurement data, the measurement data are examined for the presence, and preferably the strength, of one or more types of artifact (identification of the artifacts), and the result of the examination is made available for further processing, for example to recommend one or more corrective measures, preferably for applying the recommended corrective measures.

Description

The subject matter of an independent invention is a method for correcting computed tomographic measurements and a device for carrying out the method.

In a computer tomography (with a computer tomograph), the workpiece to be examined is placed between a radiation source, in the case of X-ray computer tomography, an X-ray source, and a radiation detector, so that radiographic images can be recorded with the detector in different rotational positions of the workpiece in relation to the source and detector and reconstructed into a volume data set (voxel volume, often just referred to as "volume"). The workpiece is usually placed on a turntable in order to realize the different rotational positions by rotating it around a rotation axis. Alternatively, the workpiece is stationary and the source and detector rotate around the workpiece around a rotation axis. In laminography, extended trajectories (also referred to as recording geometry, with correspondingly characterizing recording geometry parameters) are provided for the movement of workpiece, source and detector relative to one another, in particular in order to be able to measure workpieces that cannot be tomographed using a 360° rotation, for example because otherwise a collision of the workpiece with the source or detector would occur, or in order to examine workpieces more quickly than with a computer tomography by recording radiographic images in a smaller number of rotational positions. The methods described in the present invention with regard to a computer tomographic measurement (computer tomography), in particular correction methods, can also be used analogously for laminography.

The detector used to capture the 2D radiographic images is usually flat, i.e. designed as a 2D detector, and consists of a scintillator layer in which the X-rays are converted into light, tubular light channels located behind it in the direction of radiation, which reduce crosstalk between different detector areas, and a fixed pixel matrix directly behind it, i.e. a flat matrix camera, such as a CCD or CMOS camera. Alternatively, line detectors with one or a few lines are also known. In order to capture large measuring areas, however, the workpiece must be moved across the line direction, which is time-consuming, and irradiated several times. Here, individual layers are also reconstructed separately. The invention is therefore advantageously based on flat detectors or detectors with a flat detection surface such as a flat scintillator layer.

More cost-effective detectors can be realized if the scintillator layer and the matrix camera are constructed separately, with imaging optics arranged between the two and often a deflection of the beam path to protect the camera (so-called mirror detectors), as is the case in the US 7,400,704 and the DE102020102849 described.

In order to be able to measure dimensionally, surface points are generated from the volume data, for example in STL format (STL - Standard Triangulation Language), and linked to dimensions or similar that describe the geometric properties of a workpiece. The computer tomograph is then designed as a coordinate measuring machine or coordinate measuring system. The invention also provides for this for laminography processes that are carried out with the computer tomograph, for example. In order to be able to measure precisely, one of the necessary conditions is that a high resolution of the structures of the workpiece (structural resolution) is guaranteed in the spatial area, which usually results from the respective measuring task. In addition to various others, two dominant effects overlap here: the focal spot size of the focal spot (focus) of the X-ray tube that emits the X-ray radiation and the pixel resolution of the detector used. The structural resolution is to be considered in the workpiece plane or position and is referred to here as effective structural resolution, but can also be transformed into the detector plane and viewed and vice versa, taking into account the CT geometry (position of source, workpiece or axis of rotation and detector in relation to one another, also referred to as recording geometry), in particular the geometric imaging scale A, which results from the ratio FDD (focus to detector distance) to FOD (focus to object distance), i.e. the position of the workpiece between the source and detector, more precisely between the focal spot (focus) of the source emitting the measuring radiation and the detector.

In the known methods for correcting artifacts present in computed tomography measurement data, the application of the appropriate correction methods is left to the user. This requires an extremely high level of expertise on the part of the user in recognizing which artifacts or type of artifact is present in the measurement data. This often leads to sensible artifact correction methods not being applied or ineffective corrections being applied, which unnecessarily extend the total time of the measurement process including the correction. This will be explained in more detail below.

Voxel volumes from computed tomography (CT) measurements generally show many image artifacts that have different causes and can be corrected using different methods (beam hardening, scattered radiation, ring artifacts, etc.). Voxel volumes from a CT measurement also contain a superposition of different artifacts, each of which is of varying severity depending on the properties of the workpiece, the measurement parameters and the device parameters. In a heavily artifacted voxel volume, it is often difficult to assign the artifacts to the artifact origins, since the different artifacts have a similar appearance in the voxel volume and/or the surface model calculated from it. It is therefore not clear to the user per se which corrections must be applied to sufficiently reduce the artifacts in a CT measurement. It can therefore happen that artifacts were not sufficiently reduced due to ignorance or that unnecessary artifact corrections were calculated, which is associated with increased measurement time, evaluation time and/or unnecessary computing power, which then have a negligible influence on the measurement result.

A further object of the present invention is therefore to provide a method and a device for applying the method, which ensures that meaningfully effective correction methods for artifacts can be used in a computer tomography measurement of workpieces, and in particular that correction methods that are not or hardly effective are not used. Effective is to be understood here as meaning that the image quality in the radiographic image and volume data and/or the accuracy of the metrological evaluations based on them is significantly increased, for example, is improved by at least 10%, preferably at least 20%.

To solve this problem, the invention provides that in order to correct artifacts in computed tomography measurement data of workpieces (a CT measurement) by means of a computer-implemented method (hereinafter referred to as identification software) on the basis of prior knowledge of the effects of artifacts in the measurement data, the measurement data are examined for the presence, and preferably the strength, of one or more types of artifacts (identification of the artifacts), and the result of the examination is made available for further processing, for example to recommend one or more corrective measures, preferably for applying the recommended corrective measures.

In a preferred embodiment, the invention provides that a learning-based algorithm, i.e. an artificial intelligence (AI), preferably a neural network, in particular a deep neural network, e.g. a convolution-based neural network, is trained to determine the causes of artifacts and the respective artifact strengths in a CT measurement and to provide the user with an optimal correction suggestion. This correction suggestion can provide both software-based corrections (those that are integrated in a measurement software or not), as well as hardware-based corrections (e.g. measurement with a slit diaphragm, alternative clamping of the workpiece or adjustment of the magnification) or measurement parameter-specific changes (different tube voltage, power, etc.). In addition, the correction suggestion can be optimized taking into account a known, specific measurement task (e.g. optimal image quality only required locally, or maximum measurement time or correction duration). In contrast to direct artifact correction with neural networks, this approach has the advantage that result data is not generated directly by a neural network (“black box”) and that the neural network does not produce visually good-looking data that could be incorrect in terms of measurement technology, i.e. in particular could lead to incorrect measurements of the geometric properties of the workpiece. The neural network is only used here for a user recommendation. This means that the training data does not have to cover the application parameter space exactly (approximate coverage is probably sufficient for this purpose), which is why simulations can also be used that show certain differences to real measurements.

Using training data (e.g. generated through simulations), a neural network is trained to estimate which artifacts have what influence on the measurement result and ultimately suggests which corrections are useful for the measurement in question. A sufficiently trained neural network can then be applied to any measurement and provide suggestions for corrections to be made.

This can be achieved as follows: Neural networks (especially convolutional neural networks) do not necessarily generate image data as output data. They can also simply deliver one or more values or converge on one of two values (e.g. "yes-no" decision). If the training data set is now simulated, artifact-free volume data is directly available for supervised training and individual artifact causes can be switched on and off (it is also possible to targeted series of measurements can be carried out, but this would be very time-consuming. This gives an impression of the effects of different artifacts. A measure is now defined which determines which artifacts in the training data are disturbing. For example, a threshold-based segmentation into the basic materials of the object is planned for this. The number of incorrectly segmented voxels can serve as a quality measure, with a criterion (e.g. more than 0.1% of the voxels incorrectly segmented due to the simulated artifact) deciding whether a correction is applied ("yes-no" scenario). This defines a so-called label for the learning objective for this training data set; the neural network with the task of determining whether a correction should be carried out for a certain artifact therefore has a clear target value (yes or no). Different networks could be trained for the various artifacts, each of which provides a "yes/no" answer. Preferably, a single neural network is trained, which, for example, directly predicts the proportion of incorrectly segmented voxels for the data set with regard to a certain artifact as an output value (e.g. 94% due to beam hardening, 5% due to scattered radiation, 1% due to ring artifacts). This reduces the application to a single neural network and the user knows which artifact corrections are most effective. In addition, the neural network can provide a "yes/no" answer to correct certain artifacts based on this output and, if applicable, the defined measurement task.

A method according to the invention for correcting artifacts in measurement data that have been determined by means of computed tomography or laminographic measurement (computed tomography) of at least one workpiece or workpiece area, wherein the measurement data are radiographic image data and/or volume data reconstructed therefrom and/or surface data determined therefrom such as STL data, is characterized in that by means of a computer-implemented method (identification software) on the basis of prior knowledge of the effect of artifacts in measurement data, the measurement data are examined for the presence, and preferably the strength, of one or more types of artifacts (identification of the artifacts), and the result of the examination is made available for further processing, for example to recommend one or more corrective measures, preferably for applying the recommended corrective measures.

To identify the artifacts, it is intended that the radiographic image data and/or volume data reconstructed from them and/or surface data determined from them, such as STL data, are examined as measurement data or measurement data types. The choice of measurement data to be examined depends on which of the measurement data the respective artifact can be most reliably identified in. In the case of identification by a learning-based algorithm, the choice of measurement data to be examined and, if necessary, the use of several measurement data types arises from the training. If several measurement data types are made available during training, the algorithm automatically learns in which measurement data types which artifacts can be recognized in which form. Which of the measurement data is then corrected when correcting the recognized artifact depends on the correction methods assigned to the respective artifact types, which are part of the prior knowledge, i.e. are already stored in advance and are state of the art.

In particular, the invention is characterized in that expected measurement data (test data) are made available as prior knowledge or as part of the prior knowledge, wherein several sets of test data are made available which differ at least in that individual types of artifacts are present or not present or are present in different degrees of intensity.

Preferably, the sets of test data with existing artifact types are assigned corrective measures that the identification software suggests when identifying the corresponding artifact type.

It should also be emphasized that the identification software for identifying the artifacts compares the measurement data or parts thereof with at least one set of test data that does not have at least one of the artifact types or has this artifact type to a lesser extent, and that the deviations between the measurement data and the respective test data are evaluated, preferably for the presence of a systematic pattern, particularly preferably by determining whether a specified threshold value has been exceeded, the result of the evaluation being the presence of the respective artifact type and preferably its severity.

The invention is also characterized in that the identification software is based on a learning-based algorithm, preferably one or more neural networks, in particular a convolution-based neural network, which has been trained with prior knowledge of the effects of artifacts of one or more artifact types, in particular trained with the multiple sets of test data.

In particular, the invention is characterized in that the identification software is based on analytical software that uses prior knowledge of the effects of artifacts of one or more of several artifact types, comparing systematic deviations between sets of test data with the measured data or a subset of the measured data.

For example, the analytical software looks for a systematic pattern in the measurement data that is known to promote artifacts. The exact procedure is different for different artifacts. To detect scattered radiation, values above 1 in the standardized intensity image are searched for in certain detector areas; to detect ring artifacts, strong pixel variances are searched for during the measurement time; and to detect beam hardening, very dark areas in the projections are searched for.

Preferably, it is provided that, in order to generate the prior knowledge, in particular the multiple sets of test data, by means of simulation of the computer tomographic measurements of one or more workpieces that differ in terms of geometry and/or material, simulation data without and simulation data with one or more artifact types of one or more thicknesses are determined, wherein preferably the deviations between the simulation data without and with the respective artifact of the respective thickness are determined and assigned to the respective artifact type.

The identification of artifacts works most reliably when the test data is as similar as possible to the measurement data of the workpieces to be measured later in terms of material and geometry. Nevertheless, slight differences are acceptable because the basic systematic influences of artifacts are very similar, e.g. in the case of scattered radiation, the intensity in air is much greater than in a bright image (or simulation without scattered radiation). In particular, when using a learning-based algorithm, the training data does not have to cover the application parameter space exactly (approximate coverage is sufficient for this purpose), which is why simulations can also be used to generate the test data that show certain differences to real measurement data. The advantage of having many different test data for training is that the identification of artifacts works more reliably even with unknown workpieces.

It should also be emphasized that in order to generate the prior knowledge, in particular the multiple sets of test data, reference measurement data without and reference measurement data with one or more artifact types of one or more thicknesses are determined by means of computer tomography measurements of one or more workpieces that differ in terms of geometry and/or material, wherein preferably the deviations between the reference measurement data without and with the respective artifact of the respective thickness are determined and assigned to the respective artifact type.

• - Computertomografie mit geringerem Kegelstrahlwinkel
• - Computertomografie mit helixförmiger Trajektorie (Helix-Computertomografie)
• - Computertomografie mit vergrößertem Drehwinkel
• - Computertomografie mit erhöhtem Abbildungsmaßstab
• - Computertomografie mit veränderter Aufspannung (Lage) des Werkstücks auf dem Drehtisch, so dass sich kürzere Durchstrahlungslängen ergeben, insbesondere eine möglichst geringe maximale Durchstrahlungslänge
• - Erhöhung der Anzahl von Drehstellungen, in denen bei der Computertomografie Durchstrahlungsbilder aufgenommen werden
• - Erhöhung der Belichtungszeit je aufgenommenem Durchstrahlungsbild
• - Erhöhung der Anzahl der Bildüberlagerungen je Durchstrahlungsbild
• - Verringerung der Bandbreite der zur Abbildung verwendeten Strahlung, beispielsweise durch Änderung der Beschleunigungsspannung und/oder Leistung der Strahlungsquelle oder durch den Einsatz eines Strahlfilters
• - Verringerung der Drehgeschwindigkeit bei einer Computertomografie mit Bildaufnahme während der Drehbewegung (OnTheFly-CT),
• - Korrektur von Maßen an geometrischen Eigenschaften durch Referenzmessung mit einem anderen, genaueren Sensor, beispielsweise in einem Koordinatenmessgerät mit taktilem, taktil-optischen oder optischen Sensor

Examples of ways to generate artifact-free or artifact-reduced measurements include the following:

• - Computed tomography with smaller cone beam angle
• - Computed tomography with helical trajectory (helix computed tomography)
• - Computed tomography with increased angle of rotation
• - Computer tomography with increased image scale
• - Computer tomography with a modified clamping (position) of the workpiece on the rotary table, resulting in shorter radiography lengths, in particular a maximum radiography length that is as short as possible
• - Increasing the number of rotational positions in which radiographic images are taken during computed tomography
• - Increasing the exposure time per radiographic image taken
• - Increasing the number of image overlays per radiograph
• - Reducing the bandwidth of the radiation used for imaging, for example by changing the accelerating voltage and/or power of the radiation source or by using a beam filter
• - Reduction of the rotation speed during a computed tomography scan with image acquisition during rotation (OnTheFly-CT),
• - Correction of dimensions of geometric properties by reference measurement with another, more accurate sensor, for example in a coordinate measuring machine with tactile, tactile-optical or optical sensor

• - Solldaten wie CAD-Daten des einer oder der mehreren Werkstücke,
• - Analytisches Modell der Entstehung der Durchstrahlungsbilddaten mit und ohne einzelne Artefaktarten bzw. Artefakte in unterschiedlicher Stärke,
• - Analytisches Modell der Rekonstruktion der Volumendaten aus den Durchstrahlungsbilddaten,
• - Analytisches Modell der Ermittlung von Oberflächendaten aus Volumendaten und/oder
• - Analytisches Modell der Überlagerung zumindest zwei unterschiedlicher Artefakte.

The invention is also characterized by the fact that at least one of the following pieces of information is included in the prior knowledge:

• - Target data such as CAD data of one or more workpieces,
• - Analytical model of the generation of radiographic image data with and without individual types of artifacts or artifacts of varying strengths,
• - Analytical model of reconstruction of volume data from radiographic image data,
• - Analytical model for determining surface data from volume data and/or
• - Analytical model of the superposition of at least two different artifacts.

• - Strahlaufhärtung,
• - Streustrahlung,
• - Ringartefakte,
• - Kegelstrahlartefakte,
• - Bewegungsartefakte,
• - Sampling-Artefakte,
• - Geometriefehler,
• - Extrafokalstrahlungsartefakte,
• - Trunkierungsartefakte und/oder
• - Limited-Angle-Artefakte.

In particular, the invention is characterized in that the presence of one or more of the following artifact types is examined, in particular test data with and without these artifact types and/or with different strengths of these artifact types are made available:

• - beam hardening,
• - scattered radiation,
• - ring artifacts,
• - cone beam artifacts,
• - motion artifacts,
• - sampling artifacts,
• - geometry errors,
• - extrafocal radiation artifacts,
• - Truncation artifacts and/or
• - Limited-angle artifacts.

Preferably, it is provided that one or more corrective measures are proposed from the result of the examination for the presence of one or more types of artifact, and preferably their respective severity, wherein in particular software-based measures for correcting the measurement data and/or adapted measurement parameters for a new computed tomography measurement (setting parameters) for generating modified measurement data are proposed.

• - Korrektur der Strahlaufhärtung, beispielsweise mittels empirischer Verfahren und/oder iterativer Verfahren,
• - Korrektur der Streustrahlung, beispielsweise durch Simulationsverfahren, vorzugsweise Monte-Carlo-Simulation,
• - Korrektur von Ringartefakten, beispielsweise durch Simulationsverfahren,
• - Korrektur von Kegelstrahlartefakte, beispielsweise durch Simulationsverfahren,
• - Korrektur von Bewegungsartefakten, beispielsweise durch Interpolationsverfahren,
• - Korrektur von Sampling-Artefakten, beispielsweise durch Interpolationsverfahren,
• - Korrektur von Geometriefehlern, beispielsweise durch Interpolationsverfahren,
• - Korrektur von Extrafokalstrahlungsartefakten, beispielsweise durch Filterverfahren,
• - Korrektur von Trunkierungsartefakten, beispielsweise durch Interpolationsverfahren und/oder
• - Korrektur von Limited-Angle-Artefakten, beispielsweise durch Extrapolationsverfahren oder Filterverfahren,

besonders bevorzugt, dass optimale Korrekturparameter für die Korrekturmaßnahmen vorgeschlagen werden.It should also be emphasized that one or more of the following corrective measures are proposed as software-based measures to correct the measurement data:

• - Correction of beam hardening, for example by means of empirical methods and/or iterative methods,
• - Correction of scattered radiation, for example by simulation methods, preferably Monte Carlo simulation,
• - Correction of ring artifacts, for example by simulation methods,
• - Correction of cone beam artifacts, for example by simulation methods,
• - Correction of motion artifacts, for example by interpolation methods,
• - Correction of sampling artifacts, for example by interpolation methods,
• - Correction of geometric errors, for example by interpolation methods,
• - Correction of extrafocal radiation artifacts, for example by filtering methods,
• - Correction of truncation artifacts, for example by interpolation methods and/or
• - Correction of limited-angle artifacts, for example by extrapolation or filtering methods,

It is particularly preferred that optimal correction parameters for the corrective measures are proposed.

• - Anpassung des für die computertomografische Abbildung verwendeten Kegelstrahlwinkels, insbesondere Verringerung des Kegelstrahlwinkels, vorzugsweise durch Vergrößerung des Abstandes zwischen Strahlungsquelle und Detektor des Computertomografen und/oder Vergrößerung des Abstandes zwischen Werkstück und Strahlungsquelle und/oder durch Verwendung einer Schlitzblende zwischen Strahlungsquelle und Werkstück,
• - Anpassung der Messtrajektorie, vorzugsweise Durchführung einer Helix-Computertomografie oder einer Computertomografie mit vergrößertem Drehwinkel,
• - Anpassung des Abbildungsmaßstabs, vorzugsweise Erhöhung des Abbildungsmaßstabs,
• - Anpassung der Lage des Werkstücks auf dem Drehtisch des Computertomografen, beispielsweise zur Verringerung der maximal auftretenden Durchstrahlungslänge,
• - Anpassung der Anzahl der Drehstellungen, in denen Durchstrahlungsbilder aufgenommen werden, vorzugsweise Erhöhung der Anzahl,
• - Anpassung der Belichtungszeit je aufgenommenem Durchstrahlungsbild und/oder Anpassung der Anzahl der Bildüberlagerungen je Durchstrahlungsbild, vorzugsweise Erhöhung,
• - Anpassung der Bandbreite der zur Abbildung verwendeten Strahlung, vorzugsweise durch Anpassung der Beschleunigungsspannung und/oder Leistung der Strahlungsquelle wie Röntgenröhre, insbesondere Verringerung, und/oder Anpassung des Strahlfilters, und/oder
• - Anpassung der Drehgeschwindigkeit bei einer Computertomografie mit Bildaufnahme während der Drehbewegung (OnTheFly-CT), insbesondere Verringerung,

besonders bevorzugt, dass optimale Messparameter für die erneute computertomografische Messung vorgeschlagen werden.The invention is also characterized in that one or more of the following measurement parameters are proposed as adapted measurement parameters for a new computed tomographic measurement (setting parameters) for generating modified measurement data:

• - adaptation of the cone beam angle used for the computed tomographic imaging, in particular reduction of the cone beam angle, preferably by increasing the distance between the radiation source and the detector of the computed tomograph and/or increasing the distance between the workpiece and the radiation source and/or by using a slit diaphragm between the radiation source and the workpiece,
• - Adjustment of the measurement trajectory, preferably by performing a helix computed tomography or a computed tomography with an increased angle of rotation,
• - Adjustment of the image scale, preferably increase of the image scale,
• - Adjustment of the position of the workpiece on the rotary table of the CT scanner, for example to reduce the maximum radiographic length,
• - Adjustment of the number of rotational positions in which radiographic images are taken, preferably increasing the number,
• - Adjustment of the exposure time per radiographic image taken and/or adjustment of the number of image overlays per radiographic image, preferably increase,
• - Adjustment of the bandwidth of the radiation used for imaging, preferably by Adjustment of the acceleration voltage and/or power of the radiation source such as X-ray tube, in particular reduction, and/or adjustment of the beam filter, and/or
• - Adjustment of the rotation speed in a computed tomography with image acquisition during the rotational movement (OnTheFly-CT), in particular reduction,

It is particularly preferred that optimal measurement parameters are suggested for the repeated computed tomographic measurement.

In particular, the invention is characterized in that the proposed corrective measures are applied automatically or are displayed to the operator for selection on a display means, wherein the proposed corrective measures are preferably restricted to the corrective measures available in the respective computer tomograph.

Preferably, the learning-based algorithm provides a "yes/no" decision and/or a key figure for the strength of the presence in relation to one or more artifact types as a result of the investigation, wherein the key figure is preferably the relative proportion of the strength of the respective artifact type to all recognized artifact types, wherein the "yes/no" decision in relation to an artifact type is particularly preferably dependent on the key figure, for example a "yes" decision is made if a previously defined threshold value for the key figure is exceeded.

Different networks could be trained for the different types of artifact, each of which provides a yes/no answer and, if necessary, also indicates the strength of the respective type of artifact, for example relative to all detected artifacts in interaction with the multiple networks. A particularly preferred solution, however, only includes a single network that quantitatively determines the influences of the different types of artifact, for example how many incorrectly segmented voxels come from which type of artifact. This is advantageous in that it provides a direct statement as to which correction is most useful to apply.

The aforementioned incorrect segmentation of the voxels represents an exemplary embodiment of the invention in relation to the assessment of whether and to what extent a type of artifact is present in the measurement data. To this end, it is assessed whether an artifact type or an artifact type of a certain strength detected in the measurement data or activated in the test data during training is disturbing the measurement data or test data by performing a threshold-based segmentation in the base materials of the workpiece and determining a quality measure based on the evaluation of the proportion or number of voxels that were incorrectly segmented. If a criterion is exceeded (e.g. more than 0.1% of the voxels were incorrectly segmented due to the detected or simulated artifact), the presence of the artifact is determined and a correction recommendation (as a label for the learning goal) is assigned in the sense of a yes-no scenario.

It should also be emphasized that the learning-based algorithm comprises a single network for all artifact types or several networks for the different artifact types and is trained accordingly, in particular each network is trained on one artifact type at a time.

The invention is also characterized in that the learning-based algorithm is trained with several sets of test data which differ in that individual artifact types are present or not present or are present in different degrees of severity, and that the sets of test data with present artifact types are assigned corrective measures which the identification software suggests when the corresponding artifact is present.

In particular, the invention is characterized in that the identification software processes at least one radiographic image from the radiographic image data and/or at least one sectional image through the volume data and/or at least one characteristic value for the acquisition geometry of the computed tomographic measurement and/or at least one setting parameter of the computed tomographic measurement as input data from the measurement data.

Examples of acquisition geometry parameters are the position of the radiation source such as the X-ray source, the detector and the rotation axis of the turntable relative to one another, in particular the source-detector distance FDD (focus to detector distance), the source-object distance FOD (focus to object distance; where FOD generally indicates the distance to the rotation axis around which the workpiece is rotated), the magnification FDD / FOD and the cone angle of the part of the source's radiation received by the detector. In the case of more complex trajectories, i.e. movements of the components source, detector and workpiece (or object) relative to one another during computed tomography, additional acquisition geometry parameters can be added that characterize this movement.

Preferably, the training of the neural network corresponds to a supervised training and input and target data (template data) are simulation data of the computer-aided mographic measurement and/or input and target data are measured computed tomographic measurement data (reference measurement data).

It should also be emphasized that the training of the neural network corresponds to an unsupervised training, for example in the context of an adversarial network, and the input and target data (template data) are simulation data of the computed tomographic measurement and/or the input and target data are measured computed tomographic measurement data (reference measurement data), whereby these do not necessarily have to be paired.

For unsupervised training, approaches are provided in which a network does not run based on "Input A -> Generate Output A -> Compare Output A with Target A -> Optimize Network Parameters" (this corresponds to supervised training). There are also approaches in which a network is generated from Input A -> Output A, and this is trained with the help of a second network that evaluates whether Output A is good. The second network has only been given template data B, C, D, which have nothing to do with Input A, but which it can use to determine whether, for example, computed tomography volume data is free of artifacts.

The invention is also characterized in that the correction proposal is made taking into account the restriction to one or more selected areas or features of the workpiece and/or one or more measurement tasks on the workpiece, for example optimal image quality for a local area or maximum measurement time of the new computed tomographic measurement or maximum duration for the application of the recommended corrective measure is taken into account or optimized with regard to this.

To achieve this, the invention also provides a device for carrying out the method, which is characterized in that the device has a correspondingly designed computing unit.

In particular, the invention is characterized in that the computing unit is connected to the control unit for a computer tomograph and/or that the computing unit is connected to a device such as an offline PC for creating measurement programs for a computer tomograph and/or for evaluating measurement data, in particular correcting measurement data from a computer tomograph.

The subject matter of an independent invention is a device for generating X-rays and a method for operating the device, in particular in a computer tomograph or computer tomography sensor (CT sensor), particularly preferably in a coordinate measuring machine (CMM) with a computer tomograph or computer tomography sensor (CT sensor).

• - Einschaltvorgang: Hier muss das thermische Gleichgewicht erst abgewartet werden. Erst dann nimmt der Brennfleck die Position wie beim Einmessen des CT-Sensors ein. Weitere Drift ist dann nur noch durch Fluktuation der eingestellten Leistung (Rauschen) oder der Kühlung zu erwarten.
• - Änderung der eingestellten Leistung der Röhre: Nach Abwarten des thermischem Gleichgewichts ergibt sich ggf. eine andere Brennfleckposition, da eine andere Leistung vorliegt als beim Einmessen des CT-Sensors. Zur Vermeidung dieses Effektes wäre ein Einmessen bei unterschiedlichen Leistungen notwendig. Dies wäre sehr aufwändig.
• - Fluktuation der eingestellten Leistung (Rauschen) oder der Kühlung beim Betrieb der Röhre

X-ray tubes (tubes) with reflection targets or transmission targets are known as devices for generating X-rays. For the use of these tubes in CT sensors, for example for material inspection, or in CMMs with CT sensors, for example for material inspection and/or dimensional measurement of workpiece geometries, the spatial drift of the position of the focal spot on the target that emits the X-rays plays a decisive role in the achievable measurement accuracy. This drift is triggered by the heating of the anode of the tube that is not constant over time. As a result, the target and thus the focal spot shift, for example in the direction in which the electrons are bombarded, which create the focal spot when they hit the target. If this impact occurs on an inclined target (reflection target), the displacement of the target shifts the focal spot and thus the resulting X-ray beam perpendicular to the direction of reflection. When the tube is in operation, the focal spot shifts and measurement deviations occur. Examples of changes in the thermal situation of the anode are:

• - Switch-on process: Here, thermal equilibrium must first be reached. Only then does the focal spot assume the position it was in when the CT sensor was calibrated. Further drift is then only to be expected due to fluctuations in the set power (noise) or cooling.
• - Changing the set power of the tube: After waiting for thermal equilibrium, a different focal spot position may result because the power is different from that when the CT sensor was calibrated. To avoid this effect, calibration would be necessary at different power levels. This would be very time-consuming.
• - Fluctuation of the set power (noise) or cooling during operation of the tube

• - Abwarten des thermischen Gleichgewichts nach Veränderung der thermischen Situation.
• - Nachstellen der Lage der Anode oder des Auftreffpunktes des Elektronenstrahls auf Basis einer wiederholten Messung der Lage einer Referenz (z.B. Kugel) auf dem Detektor des CT-Sensors (ggf. auch CT der Kugel, um Lageverschiebung des Brennflecks in Strahlrichtung zu ermitteln)
• - Diverse Verfahren zur Korrektur der Brennfleckverschiebung durch Mitmessen einer Referenz (Driftkugel) oder Messung der Referenz vor und nach der Messung oder vorab Schnell-CT zur Ermittlung von Stützstellen zur nachträglichen Korrektur der eigentlichen CT-Messung. Diese Korrekturverfahren sind Inhalt folgender Schriften der Anmelderin: DE102010000473 , DE102015101378 , DE102015121127 , DE102017100594

The following methods are known to correct these drift effects:

• - Waiting for thermal equilibrium after a change in the thermal situation.
• - Adjustment of the position of the anode or the point of impact of the electron beam based on a repeated measurement of the position of a reference (e.g. sphere) on the detector of the CT sensor (if necessary also CT of the sphere to determine the position shift of the focal spot in the beam direction)
• - Various methods for correcting the focal spot shift by measuring a reference (drift sphere) or measuring the reference before and after the measurement or by conducting a rapid CT beforehand to determine support points for subsequent correction of the actual CT measurement. These correction methods are contained in the following documents by the applicant: DE102010000473 , DE102015101378 , DE102015121127 , DE102017100594

The disadvantage of all these methods, however, is that you have to wait for the tube to warm up and then measure the focal spot drift. For an accurate real-time correction of the focal spot drift, a spatially stable reference must be measured during the measurement, which limits the measurement range. These methods are complex, usually time-consuming, and the remaining drift depends on the accuracy of the correction measures.

A further object of the present invention is therefore to provide simpler and more precise methods or devices suitable for this purpose compared to the prior art, with which real-time compensation of the focal spot drift of an X-ray tube of a CT sensor is made possible, which is triggered by thermally induced changes within the X-ray tube, in particular the position of the anode. Preferably, this compensation should also be usable for the warm-up process. It is also the object to enable faster, more precise measurement with the CT sensor.

To solve this problem, the invention provides an embodiment of the anode and/or a holding device receiving it by a design in which the thermally induced change in the position of the focal spot is compensated by an additionally introduced region, for example by using a bimetallic effect, in particular the embodiment of the anode itself is realized as a bimetal.

In the case of a reflection target, it is intended that the anode is made of a bimetallic strip or that the holding device has a bimetallic strip whose longitudinal direction runs parallel to the electron beam.

Even if the proposed solution is primarily intended for reflection targets, an application for transmission targets is also part of the invention. The solution is provided for the holding device to have at least one bimetallic strip, the longitudinal direction of which is arranged offset parallel (off-center) to the electron beam. In particular, a temperature-related tilting of the target and the resulting focal spot shift can be compensated for in this way.

The aim is that the expansion or displacement of the anode, and thus of the target, which occurs during heating, has no influence on the position of the focal spot by designing it as a bimetal. This means that the focal spot remains stationary. The target, preferably together with the anode, is preferably displaced and/or arranged at an angle in the cold state compared to the desired warm operating state so that when the bimetal effect takes effect, i.e. when the target is warm, the target is at the desired angle to the electron beam, for example 45° in the case of a reflection target, so that the desired main radiation direction of the X-ray beam is achieved.

To implement the method for operating the device according to the invention and to dimension the device, it is intended that an analytical model of the heating of the anode or the holding device is carried out and/or the displacement during heating is measured. To do this, the case of how the anode or the holding device and thus the focal spot shifts without the bimetal design is first examined. The bimetal is dimensioned on this basis, in particular the selection of suitable materials and the length of the components of the bimetal strip. The device with integrated bimetal strip is then subjected to a metrological test, whereby the remaining focal spot drift is determined when the anode is heated at different power levels. This should be as low as possible. If necessary, it is also intended to adapt the dimensioning on the basis of these investigations. It is also intended that any remaining drift influences, for example statistical drift, are corrected using the known correction methods.

A device according to the invention for generating X-rays (X-ray tube), comprising at least one electron source (cathode) in a vacuum chamber and a target designed to generate X-rays, emanating from an anode, on which a focal spot emitting X-rays is formed when irradiated with the electrons (electron beam) from the cathode, wherein the anode and the target emanate from a holding device which is arranged inside or outside the vacuum chamber of the X-ray tube and is connected to it directly or indirectly via a connecting element (16), is characterized in that the anode and/or the holding device has at least one region which compensates for the displacement of the focal spot in at least one given direction by moving the target when the temperature changes.

In particular, the invention is characterized in that the region of the anode and/or the holding device has at least two strips of different material which form a bimetallic strip, preferably the anode consists of the bimetallic strip.

Preferably, the target is designed as a reflection target and the longitudinal direction of the bimetallic strip runs parallel to the direction of the electron beam.

It should also be emphasized that the target is designed as a transmission target and the longitudinal direction of the bimetallic strip runs parallel to the direction of the electron beam.

The invention is also characterized in that the angular position of the target in the cold state is adjusted in comparison to the desired position in the warm state during operation of the X-ray tube in such a way that the desired position is assumed by the tilting due to the bimetal effect in the warm state.

According to a particularly noteworthy proposal, the X-ray tube is intended to be used in a computer tomography scanner or in a coordinate measuring machine with a computer tomography sensor (CT sensor), preferably a multi-sensor coordinate measuring machine together with other sensors such as tactile, tactile-optical and/or optical sensors.

To achieve this, the invention also provides a method for operating the device according to the invention for generating X-rays, which is characterized in that the X-ray tube is operated in a computer tomograph or in a coordinate measuring machine with a computer tomography sensor (CT sensor) in order to measure workpieces, in particular in order to dimensionally determine features in the interior of the workpiece, such as inclusions, cavities and/or fibers, and/or geometric features.

In particular, the invention is characterized in that the area for compensating the focal spot drift is designed on the basis of a thermal model of the temperature-related expansion of the anode and/or the holding device or by measuring this expansion.

Preferably, remaining focal spot drift is corrected by known methods.

Further details, advantages and features of the invention emerge not only from the claims, the features to be derived therefrom - individually and/or in combination - but also from the following description of the figures.

• 1 eine Ausgestaltung des erfindungsgemäßen Verfahrens zur Korrektur von Artefakten in Messdaten einer Computertomografie mittels eines auf Vorwissen basierenden computerimplementierten Verfahrens und
• 2 eine Ausgestaltung der erfindungsgemäßen Vorrichtung zur Kompensation der Brennfleckdrift einer Röntgenröhre.

They show:

• 1 an embodiment of the method according to the invention for correcting artifacts in measurement data of a computer tomography by means of a computer-implemented method based on prior knowledge and
• 2 an embodiment of the device according to the invention for compensating the focal spot drift of an X-ray tube.

Based on the 1 the invention is shown by way of example. In order to identify and correct the artifacts (artifact 1, artifact 2, ..., artifact n) in the measurement data of a computer tomography measurement (CT measurement) of a workpiece (workpiece) whose presence and strength are initially unknown (marked by "?" in the figure), a learning-based method (AI) is trained (trained AI). To train the AI, the CT measurement is simulated for a large number of workpieces (workpiece 1 to workpiece n) (CT simulation 1, CT simulation 2, ..., CT simulation i). The simulations take into account different artifacts of different strengths. For example, in simulation 1 only artifact type 1 (artifact 1) is included, in simulation 2 artifact types 1 and 2 (artifact 1, artifact 2), in simulation 3 only artifact type 2, etc. Here, for example, all combinations of artifacts 1 to n in simulations 1 to i (CT simulation 1 to CT simulation i) are simulated. The trained AI is thus able to determine the existing combination of artifacts 1 to n in the measurement data of the CT measurement of the workpiece (apply AI) and makes a yes/no decision for each artifact type (artifact 1, artifact 2, ..., artifact n) for the recommendation of the correction, i.e. a corrective measure. The respective corrective measure itself is assigned to the respective artifact or artifact type within the framework of the prior knowledge. Not shown is the additional output of the strength of the influence of the respective artifact type, which results from training the AI with artifacts of different strengths.

Based on the 2 The device according to the invention for compensating the focal spot drift is shown as an example. For the sake of simplicity, only the output of the X-ray tube according to the invention is shown here. sectional view showing the anode 10 and its attachment via a connecting element 16 to the other components of an X-ray tube, as well as the cathode 11. The cathode 11 is located in the vacuum chamber of the X-ray tube, whereby the vacuum chamber is also not shown for the sake of simplicity.

In the upper part of the 2 is a representation of an anode 10 with reflection target 12 in the cold state. The electron beam 17 emanates from the cathode 11, which is directed at the target 12 and forms the focal spot 13 upon impact, which emits the X-rays 14. The target 12 emanates from a holding device 15 which has a bimetal (15a, 15b). The two strips 15a and 15b are made of metals with different thermal expansion coefficients. If the anode 10, in particular the holding device 15 according to the prior art, were made of a single material, the expansion of the holding device 15 when heated would result in the focal spot 13 in the figure only being displaced to the left.

In the lower part of the 2 the warm anode 10 is shown in the operating state. Due to the design of the holding device 15 as a bimetal (15a, 15b), the holding device 15 bends upwards at the left end in addition to being shifted to the left (image exaggerated), whereby the position of the focal spot 13 remains unchanged (indicated by the horizontal dashed line in the figure.

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions.

Cited patent literature

• US 7,400,704 [0004]
• DE102020102849 [0004]
• DE102010000473 [0046]
• DE102015101378 [0046]
• DE102015121127 [0046]
• DE102017100594 [0046]

Claims (10)

Method for correcting artifacts in measurement data which have been determined by means of computed tomography or laminographic measurement (computed tomography) of at least one workpiece or workpiece area, wherein the measurement data are radiographic image data and/or volume data reconstructed therefrom and/or surface data determined therefrom such as STL data, characterized in that by means of a computer-implemented method (identification software) on the basis of prior knowledge of the effect of artifacts in measurement data, the measurement data are examined for the presence, and preferably the strength, of one or more types of artifact (identification of the artifacts), and the result of the examination is made available for further processing, for example to recommend one or more corrective measures, preferably for applying the recommended corrective measures. procedure according to claim 1 , characterized in that expected measurement data (test data) are made available as prior knowledge or as part of the prior knowledge, wherein several sets of test data are made available which differ at least in that individual types of artifact are present or not present or are present in different degrees of strength. procedure according to claim 2 , characterized in that the sets of test data with existing artifact types are assigned corrective measures which the identification software suggests when identifying the corresponding artifact type. Method according to at least one of the preceding Claims 1 until 3 , characterized in that the identification software for identifying the artifacts compares the measurement data or parts thereof with at least one set of test data which does not have at least one of the artifact types or has this artifact type to a lesser extent, and that the deviations between the measurement data and the respective test data are evaluated, preferably for the presence of a systematic pattern, particularly preferably by determining whether a specified threshold value has been exceeded, the presence of the respective artifact type and preferably its severity being determined as a result of the evaluation. Method according to at least one of the preceding Claims 1 until 4 , characterized in that the identification software is based on a learning-based algorithm, preferably one or more neural networks, in particular a convolution-based neural network, which has been trained with prior knowledge of the effects of artifacts of one or more artifact types, in particular trained with the multiple sets of test data. Device for carrying out the method according to at least one of the preceding Claims 1 until 5 , which has a correspondingly designed computing unit. device according to claim 6 , characterized in that the computing unit is connected to the control unit for a computer tomograph and/or that the computing unit is connected to a device such as an offline PC for creating measurement programs for a computer tomograph and/or for evaluating measurement data, in particular correcting measurement data from a computer tomograph. Device for generating X-rays (X-ray tube), comprising at least one electron source (cathode) (11) in a vacuum chamber and a target (12) designed to generate X-rays (14) and emanating from an anode (10), on which a focal spot (13) emitting the X-rays (14) is formed upon irradiation with the electrons (electron beam) (17) from the cathode (11), wherein the anode (10) with the target (12) emanate from a holding device (15) which is arranged inside or outside the vacuum chamber of the X-ray tube and is connected to it directly or indirectly via a connecting element (16), characterized in that the anode (10) and/or the holding device (15) has at least one region (15a, 15b) which compensates for the displacement of the focal spot (13) in at least one predetermined direction due to the temperature change of the anode (10) and/or the holding device (15) by displacement of the target (12) compensated for temperature changes. device according to claim 8 , characterized in that the region (15a, 15b) of the anode (10) and/or the holding device (15) has at least two strips (15a, 15b) of different material which form a bimetallic strip (15a, 15b), preferably the anode (10) consists of the bimetallic strip (15a, 15b). Method for operating the device for generating X-rays, according to claim 8 or 9 , characterized in that the X-ray tube is operated in a computer tomograph or in a coordinate measuring machine with a computer tomography sensor (CT sensor) in order to measure workpieces, in particular to measure features in the interior of the workpiece such as inclusions, To dimensionally determine voids and/or fibers, and/or geometric features.