DEVICE AND METHOD FOR COMPUTER TOMOGRAPHIC MEASUREMENT OF WORKPIECES WITH ROTATING TARGET CARRIER

Description

[0001] The subject of an independent invention is a device and a method for examining, preferably dimensionally measuring, features on a workpiece using computed tomography, in which the target is rotated.

[0002] A challenge in computed tomography (CT) measurement of workpieces is target wear and the resulting measurement deviations due to changes in focal spot size or position. The target typically consists of a substrate with a coating that generates X-rays when bombarded with electrons. This coating is worn away due to thermal stress. To maintain accurate measurements, the target must be replaced, a complex process requiring the vacuum tube of the X-ray source to be opened. In tubes with a reflection target, the focal spot can also be directed to a different area of ​​the target coating by rotating the target eccentrically to the electron beam. This is possible because the target and target carrier are completely enclosed within the vacuum tube and do not seal it. Therefore, the rotating bearing can be constructed separately from the vacuum tube's sealing system.However, devices and methods known in the art only provide for the continuous rotation of the target during measurement. This is intended to enable measurements at higher X-ray powers with the same focal spot size and is made possible by the lower heat input per unit of time due to the continuous rotation. The rotation of the target is specifically not intended to be incremental, nor is it intended to continue rotating after a predetermined target wear threshold, as this would result in excessive heat input and cause the target to melt.

[0003] This solution is not currently available for X-ray tubes with transmission targets, as the target or target carrier seals the vacuum tube with a gasket and is not designed to rotate during operation. To enable further rotation, the target carrier must currently be loosened and manually turned. This is time-consuming and cannot be automated. In particular, the tube must be switched off for this process. Another possibility is to direct the electron beam to a different area of ​​the target coating by deflecting it accordingly, as described in the applicant's DE102014103439.2. However, this has the disadvantage that the focal spot is shifted relative to the workpiece being measured and the sensor (detector). This requires complex correction to ensure accurate measurements.

[0004] From DE69736368T2 a rotating anode device and a corresponding method for computed tomography are known, wherein the electron beam irradiates different regions of the rotating target, wherein the axis of the target carrier does not coincide with the electron beam and the rotating means are arranged outside the tube.

[0005] DE102017100594A1 relates to a device and a method for computed tomography of workpieces, wherein the electron beam strikes the rotatable target eccentrically, the rotating means being arranged inside the tube. The target carrier is a carrier flange that closes the tube.

[0006] EP3214636 describes the use of an X-ray window closing the vacuum tube as a multi-target carrier; the targeting of specific regions of the target carrier, which carry targets made of different materials, is achieved by appropriately tilting the window.

[0007] A further object of the present invention is therefore to enable high-precision computed tomography measurements, particularly when using an X-ray tube with transmission target, in which measuring with a worn target is avoided using simple and automatable means.

[0008] The invention provides a solution by using different target areas for the focal spot, in that the target is rotatable by means arranged outside the vacuum tube.

[0009] An apparatus according to the invention provides a solution for the computed tomographic examination of workpieces, preferably for the dimensional measurement of features on the workpieces, comprising a radiation source (source) such as an X-ray source for a computed tomography scanner, wherein the source at least comprises a base body (tube head) from which a vacuum tube containing a device for focusing an electron beam onto a target extends, a target carrier enclosing the vacuum tube, from which the target extends, wherein the target has an area that generates X-rays when irradiated with electrons, which is formed by a coating of a substrate, and wherein the target carrier is rotatable about an axis that penetrates the target offset from the point of impact (focal spot) of the electron beam on the target, and wherein the target is particularly preferably designed as a transmission target.which is characterized in that the device has means for rotating the target carrier arranged outside the vacuum tube (rotating means), wherein the means extend from the base body (tube head) itself and / or from a beam guiding body (beam stop) extending from the base body (tube head) and arranged in the X-ray beam.

[0010] As an example, the target carrier is made of metal, the substrate of diamond, and the coating of tungsten. The target carrier seals the vacuum tube via a gasket. This gasket is designed to allow rotation of the target carrier and thus of the target itself. In a particular embodiment, the target carrier is coupled to the gasket, and therefore to the vacuum tube, by means of spring force. The spring force can also be adjustable via an adjusting screw. The spring can run between the rotatable beam diaphragm and the rotatable target carrier, thus rotating with it, or it can be fixed between a part rigidly connected to the tube head and an optional holder, such as an eccentric ring, for receiving the target carrier.

[0011] The beam diaphragm is, for example, fixedly or rotatably mounted on a flange that forms part of or surrounds the tube head and is, for example, made of steel. The beam diaphragm is preferably made of tungsten and has a central opening to limit the emitted X-rays to a selected area. For this purpose, the beam diaphragm is sufficiently thick to absorb the X-rays outside the opening. In a particular embodiment, the outer surface of the beam diaphragm is not made of tungsten but of another material, for example, steel, to save costs. According to a preferred embodiment, this outer surface is used to provide the coupling to the rotating elements. If the beam diaphragm is rotatable, the cross-section of the opening is circular.

[0012] To ensure that the electron beam strikes different areas of the coated substrate of the target when the target carrier is rotated, thus forming the focal spot, the rotation occurs around an axis that penetrates the target offset from the focal spot. This causes the focal spot to move along a circular path on the target as the target carrier rotates.

[0013] However, the invention is not limited to X-ray tubes with transmission target, but also aims to enable highly accurate computed tomography measurements when using an X-ray tube with a reflection target, avoiding the need to measure with a worn target using simple and automatable means.

[0014] To solve this problem, the invention therefore provides for the use of different target areas for the focal spot, by making the target rotatable by means arranged within the vacuum tube. In particular, this is intended to enable incremental and / or wear-dependent rotation.

[0015] An apparatus according to the invention provides a solution for the computed tomographic examination of workpieces, preferably for the dimensional measurement of features on the workpieces, comprising a radiation source (source) such as an X-ray source for a computed tomography scanner, wherein the source at least comprises a base body (tube head) from which a vacuum tube containing a device for focusing an electron beam onto a target extends, a target carrier arranged inside the vacuum tube from which the target extends, wherein the target has an area that generates X-rays when irradiated with electrons, which is formed by a coating of a substrate, and wherein the target carrier is rotatable about an axis that penetrates the target offset from the point of impact (focal spot) of the electron beam on the target, and wherein the target is particularly preferably designed as a reflection target.which is characterized by the fact that the device has means for rotating the target carrier arranged within the vacuum tube (rotating means).

[0016] In particular, the invention is characterized by the fact that the rotating means have a drive such as a motor or lifting magnet, which is designed and controlled for automatic rotation.

[0017] The control system for operating the motor is integrated, for example, in the computed tomography scanner or the coordinate measuring machine (CMM) containing it, and allows the automatic operation of the rotary devices.

[0018] Preferably, the target carrier is provided to originate from an eccentric ring in which it is rotatably mounted about an axis that penetrates the target offset from the focal spot.

[0019] In particular, the invention is characterized in that the rotating means originate from the tube head and are designed to rotate the beam aperture, i.e., are coupled to it, wherein at least one driver extends from the beam aperture, which transmits the rotational movement to the target carrier, wherein preferably the driver(s) are pins, the target carrier has blind holes, preferably radially extended elongated holes, and the pins are inserted into the blind holes.

[0020] The rotation of the motor is transmitted to the beam aperture, for example, by a grinding wheel, toothed belt, worm drive, or other gearing. Preferably, the rotational movement of the motor is transmitted to the outer surface of the beam aperture. The beam aperture is preferably rotatably mounted about an axis that passes through the focal spot. In a particularly preferred embodiment, the target carrier has at least three, preferably four, blind holes distributed radially around the target, and the driver extending from the beam aperture has the corresponding number of pins. The pins engage in an area around the target that is sufficiently far outside the target to allow for cooling with respect to the very high temperatures directly on the target. The engagement area is preferably electrically insulated from the pins so that the electrons striking the target are not conducted away via the driver.To allow the drive pins, which rotate with the drive mechanism, to move radially within the blind holes of the target carrier, which rotates about an eccentric axis of rotation, the blind holes are preferably designed as radially extended elongated slots. The rotational position of the target carrier is preferably determined by the eccentric rotatable mounting of the target carrier in an eccentric ring, which presses the target carrier against the seal. This also makes it possible to use different paths on the target for the focal spot by employing differently eccentric eccentric rings, i.e., circular paths of different diameters. An additional or alternative embodiment also provides for setting a different circular path by directing the electron beam onto the target at a corresponding offset.

[0021] Preferably, the invention provides that the rotating means originate from the tube head and are designed to rotate the target carrier, i.e., are coupled to it, wherein coupling with the target carrier is realized by blind holes formed in the target carrier, into which at least one driver engages, wherein the blind holes are preferably designed as radially extended elongated holes, or by engagement at the circumference of the target carrier.

[0022] The transmission of the motor's rotation to the target carrier is again achieved, for example, by the aforementioned drive gears or, if accessible, by a slip wheel, toothed belt, worm drive, or other gear system that engages the circumference of the target carrier. In both cases, the axis of rotation of the target carrier is determined by its mounting in the eccentric ring.

[0023] It should also be emphasized that, in order to generate a contact force between the target carrier and the vacuum tube, at least one contact element such as a spring and / or adjusting screw is provided, which preferably runs between on the one hand the tube head or a holder extending from the tube head and on the other hand the eccentric ring and / or between on the one hand the beam diaphragm and on the other hand the target carrier.

[0024] Pressing the target carrier against the vacuum tube via the seal is necessary to maintain the vacuum within the tube. However, the contact force must be sufficient to allow the target carrier to rotate relative to the vacuum tube. A defined force must be set and maintained for this purpose. The contact elements can be, for example, a spring, an adjusting screw, or a combination of both, such as a spring force adjustable via an adjusting screw. Two preferred solutions are provided regarding the relationship between which the contact force is generated. The first solution involves the contact elements running between the two non-rotating elements: the tube head and the eccentric ring. The second solution involves the contact elements running between the jointly rotatable elements: the beam diaphragm and the target carrier.

[0025] The invention is also characterized in that at least one encoder is provided for monitoring and controlling the rotational position of the target carrier, which consists of at least the two components scale and read head, wherein one of the components, preferably the scale, originates from the target carrier or the beam aperture and the other component, preferably the read head, originates from the tube head.

[0026] In particular, the invention is characterized in that the device is designed to accommodate eccentric rings of different eccentricities, preferably in that the different eccentric rings have identical outer diameters.

[0027] As previously described, this allows for the use of different paths on the target for the focal spot, i.e., circular paths of different diameters. To accommodate the different eccentric rings, the outer diameter of the various eccentric rings is designed to be identical.

[0028] One particularly noteworthy proposal envisages that the target coating be segmentally made of different materials such as tungsten, copper or beryllium and / or that the target coating be segmentally or continuously of varying thickness.

[0029] The invention is also characterized in that the drive is arranged inside the vacuum tube and connecting cables are led from the vacuum tube to a control or regulation system that is connected to or identical with the control or regulation system of the computed tomography scanner.

[0030] Another object of the present invention is to eliminate the need for manual replacement of eccentric rings in order to realize a path of the focal spot on the target on a changed radius.

[0031] To solve this problem, the invention provides for a drive, such as a piezoelectric drive or stepper motor, which is used instead of the eccentric ring to change the radius of the electron beam's circular path on the rotatable target. For this purpose, the drive is used to move the target perpendicular to the electron beam. This is preferably done together with the target carrier. It must be ensured that the axis around which the target and the target carrier are rotated also moves perpendicular to the electron beam. Otherwise, the resulting path would be non-centric around the center of the target, and the different circular paths could intersect.

[0032] The invention provides for a solution an independent device or a device combinable with the aforementioned devices, which is characterized by a drive, for example a piezo drive or stepper motor, for moving the target of an X-ray source for a computed tomography scanner perpendicular to the direction of the electron beam of the X-ray source, wherein the target carrier receiving the target is particularly preferably moved together with the target perpendicular to the electron beams.

[0033] Another objective of the invention is to design the target on the various circular paths in such a way that measurements with different accelerating voltages and thus an optimization of the radiation intensity and focal spot size is possible.

[0034] The invention provides a solution by varying the thickness and / or material of the target coating in a radial step pattern. This allows the optimal target thickness to be achieved for different voltages and, if necessary, using different target coating materials. For example, the target can be thicker in a more central area, for instance, to produce large focal spots, and thinner towards the outer edges. The width of the steps is tailored to the focal spot size expected for the corresponding accelerating voltage and associated power.

[0035] It is also planned that the thickness of the target will be varied radially continuously or according to a continuous or discontinuous function.

[0036] In particular, the invention is characterized by the fact that the target coating is formed in a radial direction in stages from different materials such as tungsten, copper or beryllium and / or of different thicknesses.

[0037] The invention also provides a method for the computed tomographic examination of workpieces, preferably for the dimensional measurement of features on workpieces, comprising at least the following steps: recording at least one set of several radiographic images (radiographic image set) of the workpiece arranged on the rotary table in several rotational positions (rotational steps) relative to a radiation source (source) such as an X-ray source and a planar detector such as an X-ray detector, with the detector, reconstructing a voxel volume, which has voxels with associated voxel gray values, from the radiographic image set and preferably determining surface measurement points from the voxel volume by means of a surface extraction method, with a computed tomography scanner at least comprising a previously described device according to the invention, which is characterized by the fact thatthat different impact points of the electron beam on the target (focal spot positions) are successively set, lying on at least one circular path, by bringing the target carrier receiving the target into different rotational positions, wherein preferably the different focal spots on the target do not overlap, preferably at least 90, particularly preferably at least 180 different focal spot positions are set distributed on the circumference of the circular path.

[0038] The number of focal spot positions that can be distributed around the circumference depends on the set focal spot size, i.e., the focusing of the focal spot, determined by the tube power used and the resulting heat to be dissipated, and the diameter of the circular path, i.e., the eccentricity of the target holder's clamping in, for example, the eccentric ring. Offset circular paths can be achieved by using different eccentric rings or by different orientations of the electron beam.

[0039] Of particular note is that different circular orbits are set by using eccentric rings of different eccentricities and / or by directing the electron beam radially to different positions around the axis of rotation of the target carrier.

[0040] Furthermore, the invention is characterized in that the adjustment of a changed focal spot position is optionally carried out per measurement, per rotational position, per image in the case of several images per rotational position and / or per grid position in a raster tomography, preferably automatically based on the monitoring of the operating hours of the X-ray tube and / or a predetermined time period and / or based on the result of a cyclically repeated measurement of the wear of the target at the respective current focal spot position, or by manual rotation by the operator.

[0041] Scanning tomography involves measuring sections of the workpiece one after the other in several offset positions of the workpiece relative to the source and the detector, i.e., several sets of radiographic images are acquired.

[0042] In particular, the invention is characterized in that the wear measurement is carried out by determining the brightness or intensity of the emitted X-ray radiation and / or by determining the contrast in at least one transmission image recorded with the detector when a changed point of impact of the electron beam on the target is briefly set to a previously unused area of ​​the target.

[0043] According to a particularly noteworthy proposal, the rotation of the target holder is carried out continuously, preferably at a low speed of preferably less than 50 µm per hour, and particularly preferably less than 10 µm per hour, in order to avoid measurement deviations due to, for example, inaccuracies.

[0044] The continuous rotation of the target serves to reduce target wear and is not intended to improve heat dissipation for higher source power settings. Therefore, the rotation is slow enough to ensure that the next unobstructed focal spot position is reached before the currently used focal spot area becomes worn. Even at very high power, this occurs after at least approximately one hour. Assuming a focal spot size of 10 µm to 50 µm in diameter, the previously mentioned maximum required speeds result, which must be converted into angular velocities depending on the radius of the circular path. While faster rotation is possible, it offers no advantage here. Slower rotation is intended for use with lower tube powers.The time required for the incremental rotation can also be derived from the relationships presented here. Therefore, the rotation will continue by 10 µm to 50 µm after at least one hour.

[0045] According to an independent concept, the invention also provides for setting higher speeds for rotating the target to enable better heat dissipation and the use of higher power sources. However, the seal must be designed accordingly. One possible solution provided by the invention is to adjust the contact force of the target carrier against the vacuum tube to compensate for the wear of the seal.

[0046] Another independent inventive method, or one that can be combined with the aforementioned ideas, provides that the focal spot position on a target, in particular on a transmission or reflection target, is changed stepwise and / or depending on the wear of the target, i.e., in particular, is not changed continuously.

[0047] In particular, the invention is characterized in that the adjustment of a changed point of impact of the electron beam on the target of an X-ray source (focal spot position), preferably an X-ray source for a computed tomography scanner for the particularly preferably dimensional measurement of features on workpieces, is carried out stepwise and / or only when a set wear limit of the target is reached or exceeded, particularly preferably by bringing the target or a target carrier receiving the target into different rotational positions and / or by using and rotating eccentric rings of different eccentricities to receive the target or target carrier and / or by directing the electron beam to different positions on the target.

[0048] In a particular embodiment of the invention, it is also provided that the X-ray generating area of ​​the target originates not from a substrate, but from a holder designed in some other form, or that the substrate is dispensed with entirely. The aforementioned embodiments are accordingly provided in combination with such a holder or without a holder as an inventive solution.

[0049] The subject of an independent invention is a method for correcting measurement data, in particular from a computed tomography scanner.

[0050] Known methods for determining and correcting measurement deviations that arise during dimensional measurement or inspection with a computed tomography scanner are based on the existence of an analytical model of the measurement process, or at least of subprocesses, from which expected deviations are calculated or simulated. The calculated or simulated deviations are then used in various ways. For example, they serve to correct measured data. If simulated data from different boundary conditions (measurement conditions) are compared, this can also be used to decide which measurement conditions should later be used for the actual measurement, for example, at which workpiece orientation the fewest scattered radiation artifacts occur.

[0051] Measurement deviations arise from various physical processes. In the case of computed tomography, to which the invention primarily relates but is not limited, these include, for example, the attenuation and scattering of the measurement radiation by the workpiece under investigation. However, deviations of the detector, such as distortion, deviations in the position and angular orientation of the detector, source, and workpiece or the rotating table holding the workpiece relative to each other (referred to here as "geometry" or "geometry data" and incorporated into the reconstruction in the form of so-called SOUV vectors), deviations in the rotation of the rotating table, deviations in the position of the source's focal spot emitting the measurement radiation (referred to here as "focal spot drift data"), and several other effects also lead to measurement deviations.

[0052] Calculating or simulating these measurement deviations is sometimes a very complex and time-consuming process. Furthermore, appropriate analytical models are required. Particularly for computed tomography, models accurate enough to predict the measurement deviations (artifacts) arising from various physical processes, some of which are statistical in nature, are only partially available and are generally very computationally intensive and therefore time-consuming. The same applies to other measurement methods, especially imaging techniques, such as image processing, where influences from lighting and workpiece properties, particularly the workpiece surface, can lead to inaccuracies in the position of a structure, especially an edge, in the image captured by a camera.

[0053] A further object of the present invention is therefore to provide a precise and rapid determination of deviations, particularly with regard to measurement data and / or simulation data from computed tomography or other imaging techniques such as image processing. It is also an object to apply the determined deviations to correct measured or simulated data. Preferably, the invention aims to perform the determination and correction without requiring an analytical model of the respective measurement process or sub-process.

[0054] The invention provides a solution whereby the determination and correction of deviations is carried out by an artificial intelligence (AI), such as a neural network, which is trained on the deviations to be determined and, if necessary, corrected. The AI ​​thus replaces the analytical calculation or simulation that would otherwise be required to determine and correct certain deviations. For this purpose, the AI ​​is trained on the corresponding function, whereby one or a sequence of several mathematical operations, referred to here as filter operations, preferably convolutions with a convolution kernel, are determined.

[0055] However, the invention is not limited to convolution-based networks; fully connected networks that are not convolution-based are also provided. In general, the sequence of several mathematical operations can also be described as a nonlinear transformation. Therefore, whenever filtering or convolution with a convolution kernel is mentioned below, the invention includes any nonlinear transformation.

[0056] This training process uses multiple sets of source and target data. The filter operation that best transforms these sets is determined. To assess the success of this process, a functional relationship (objective function) is established, such as a cost function, whose result fulfills or optimizes a specific criterion. The filter operation thus determined is then suitable for modifying other measured or simulated data (referred to here as actual data) in the same way as changes occurred between the source and target data used for training. For example, if the source data contained artifacts (e.g., due to scattered radiation) and the target data were artifact-free or artifact-reduced (e.g., without scattered radiation), then the actual data will be corrected by applying the trained filter operation to precisely account for the differences in the artifacts (i.e., the scattered radiation).This approach can also be reversed through appropriate training. A filter operation can therefore also add artifacts or other effects. This makes it possible, for example, to apply a previously trained scattering beam calculation to different workpiece orientations or positions by applying the corresponding filter operation, without having to perform a new, complex scattering beam simulation. Fundamentally, any conceivable difference between source and target data can be represented by a filter operation. With appropriate training, the AI ​​can thus handle every mathematical step in the measurement chain, including any necessary corrections; the source and target data used for training simply need to exhibit the corresponding differences.

[0057] In this context, "data" (data type) refers to measured or simulated data at various points in the measurement chain. The invention preferably relates to radiographic image data in computed tomography. However, data can also include volume data (voxel volume) reconstructed from the radiographic image data, layer-by-layer cross-sectional images of the volume data, or surface data (surface measurement points) derived from the volume data, or properties such as dimensions of geometries derived therefrom. The properties or parameters of the measuring device considered for determining or processing the data, for example, the parameters to be used for reconstruction, such as SOUV vectors or detector distortion, are also to be understood as data. It is also intended that the AI ​​intervenes at several points in the measurement or simulation chain, i.e., processes several data types.

[0058] To generate, partially generate, or avoid generating data with various deviations (e.g., artifacts) that need to be identified or corrected during training, several options are available. Measurement data containing artifacts can be generated through appropriate measurements or simulations that take the corresponding deviations into account. Measurement data with reduced artifacts can be generated through modified measurements, for example, by using beam apertures that reduce scattering, or by including artifacts only to a certain extent in the simulation. Artifact-free measurement data can be generated through appropriate simulations, but also through modified measurements, for example, with different measurement parameters, such as a particularly long integration time for image acquisition, and other measures.

[0059] The invention provides a method for investigating one or more workpieces and / or workpiece areas, comprising at least the acquisition, in particular measurement, and / or calculation, in particular simulation, of actual data of the workpiece, preferably using an imaging method such as computed tomography or camera-based image acquisition methods such as image processing, particularly preferably a method for dimensional measurement of features on workpieces and / or for the inspection of workpieces, characterized in that the actual data and / or the properties or parameters of the measuring device taken into account for determining or processing the actual data are processed, in particular corrected, by means of an artificial intelligence (AI) such as a neural network, preferably a deep network, convolutional neural network, deep convolutional network, particularly preferably a U-network.

[0060] The U-Network or U-Net is a special form of an encoder-decoder network. It is not limited to, for example, a fixed number of stages in the network, and this is also provided for according to the invention.

[0061] In particular, the invention is characterized in that the AI ​​or neural network for processing or correcting the actual data or for generating the correction data required for the correction is a mathematical operation or a chain of mathematical operations, particularly preferably a filter operation, which contains a single filter operation or a fixed or variable sequence of several fixed or variable filter operations, wherein the mathematical operation results from prior or continuously supplemented training of the AI, in particular a neural network, and wherein at least one, preferably each, filter operation preferably includes a convolution with a convolution kernel.

[0062] Preferably, the neural network is trained by functionally relating recorded or simulated source data and recorded or simulated target data (goal data), wherein the target function optimizes a criterion in particular, and wherein the data to be related preferably refer to the same or identical workpiece or calibrated workpiece and / or to identical parameters of the measuring device, in particular acquisition parameters of a computed tomography scanner such as image scale or geometry and / or voltage and / or current and / or spectrum of the source and / or trajectory of the movement between workpiece, source and detector and / or number of rotation steps and / or image acquisition time per image and / or number of images superimposed per rotation step.

[0063] The criterion can be to minimize the deviation between corrected source and target data.

[0064] In particular, it is intended that the data to be correlated relate to the same workpiece, with target data being based on target data such as CAD data and source data being based on target data or on measurement data from a real measurement, for example, computed tomography. Preferably, the acquisition parameters known for the measurement are used for the simulation. These can be the target acquisition parameters, but also, in part, actual acquisition parameters, insofar as these are determined.

[0065] In particular, the invention is characterized in that actual data, source data and target data correspond to one of the following data types or a mathematical combination of several of these, preferably the difference of several of these: one or more sets of radiographic images generated or simulated by means of computed tomography (radiographic image set or actual radiographic image set), in particular using the gray values ​​determined from the radiographic images, projection values ​​P (normalized and logarithmic gray values ​​of the radiographic images), radiographic image values ​​e^-P or P * e^-P, voxel volumes reconstructed from radiographic image sets, in particular actual voxel volumes reconstructed from actual radiographic image sets, cross-sectional image data sets generated from voxel volumes by layer-by-layer cutting, in particular actual cross-sectional image data sets generated from actual voxel volumes by layer-by-layer cutting, surface measurement points generated from voxel volumes by means of surface extraction methods, in particular surface measurement points generated from actual voxel volumes by means of surface extraction methods (actual surface points),Images (actual image) captured by at least one camera in transmitted and / or reflected light, in particular brightfield reflected light and / or darkfield reflected light, measurement parameters of the measuring device, in particular of a computed tomography scanner, such as detector distortion correction data, geometry data (SOUV vectors), turntable data and / or focal spot drift data, , wherein the AI ​​preferably processes one or more types of data, in particular making corrections successively to several types of data of the same measurement or simulation.

[0066] The projection values ​​P represent the attenuation, which results from the irradiated length L and the attenuation coefficient µ. P is determined from the grayscale values ​​of the transmission image by normalization and logarithmization. The term e^-P, referred to here as the transmission image value, represents a normalized intensity. P * e^-P is preferred as a particularly suitable estimator for forward scatter.

[0067] A further part of the invention involves obtaining the target data by the difference between a measurement and a simulation, without considering certain artifacts. This can be achieved, for example, through a monochromatic and / or polychromatic simulation of the radiographs based on a CAD model or a computed tomography measurement of the object. This difference is optionally low-pass filtered to enable the targeted correction of low-frequency artifacts. This provides an estimate of interfering artifacts, which can be subtracted from the artifact-laden actual data to obtain an artifact-corrected dataset.

[0068] The invention claims to use this target data directly for correction, or alternatively to train an artificial intelligence on this target data.

[0069] It is also worth emphasizing, according to the invention, that the source data used to train the neural network corresponds to a stack of radiographs of different rotational positions in order to have information from other rotational positions available for correct mapping to the target data, as will be explained in more detail later under the term subset.

[0070] The correction by AI described here is also provided according to the invention as a pre-correction before the merging of at least two actual data sets. Preferably, this refers to the merging of actual data sets that were acquired with different lighting conditions, in particular different exposure intensities in a camera-based image acquisition method or different X-ray spectra in a computed tomography method, as will be described in more detail later.

[0071] Furthermore, the invention expressly includes the separate correction (pre-correction) of the individual data sets (actual data sets) of different spectra using the methods mentioned above or below, preferably the correction of scattered radiation artifacts, before these data sets are fused (merged) into a single data set. In this way, artifact-pre-corrected data are merged, which reduces the artifacts in the fused data set.

[0072] However, the pre-correction is not limited to this type of merging, but is also intended, for example, for other mostly well-known artifact correction methods such as the empirical characteristic curve-based artifact correction, which mainly corrects beam hardening, or artifact corrections of, for example, scattered radiation based on Monte Carlo simulations.

[0073] Preferably, the invention provides that in the respective training and the associated correction, the source, target and actual data are data of the same type.

[0074] It should also be emphasized that the previously described methods are used for the computed tomographic examination of one or more workpieces and / or workpiece areas, comprising at least the following steps: Acquisition of at least one first set of several transmission images (actual transmission image set) of the workpiece arranged on a rotary table in several rotational positions (rotational steps) relative to a radiation source (source) such as an X-ray source and at least one detector, preferably a planar detector such as an X-ray detector, with the detector; reconstruction of a voxel volume (actual voxel volume), which has voxels with associated voxel gray values, from a set of transmission images; preferably determination of surface measurement points (actual surface points) from the voxel volume by means of a surface extraction method, wherein actual data resulting from the measurement,Particularly preferably, an actual radiographic image set and / or actual voxel volume and / or actual surface points, and / or properties or parameters of the computed tomography scanner, preferably the geometry and / or detector distortion, are corrected by means of artificial intelligence.

[0075] The invention is also characterized in that previously described methods are applied for the computed tomographic examination of one or more workpieces and / or workpiece areas, comprising at least the following steps: simulation of at least one first set of several radiographic images (actual radiographic image set) of the workpiece arranged on a rotary table in several rotational positions (rotational steps) relative to a radiation source (source) such as an X-ray source and at least one detector, preferably a planar detector such as an X-ray detector, with the detector, preferably reconstruction of a voxel volume (actual voxel volume) which has voxels with associated voxel gray values, from a set of radiographic images, preferably determination of surface measurement points (actual surface points) from the voxel volume by means of a surface extraction method, wherein actual data resulting from the simulation,Particularly preferably, an actual radiographic image set and / or actual voxel volume and / or actual surface points, and / or properties or parameters of the computed tomography scanner during the simulation, preferably the geometry and / or detector distortion, are corrected by means of artificial intelligence.

[0076] In particular, the invention is characterized in that source data, in particular source radiograph set, and target data, in particular target radiograph set, are each acquired by measurement, i.e. by irradiating the workpiece and / or by simulating the irradiation of the workpiece, preferably using as target data the low-pass filtered difference between the simulated forward projection of the workpiece and the measurement of the workpiece.

[0077] In a particularly preferred embodiment, the invention also provides that the training is performed on the low-pass filtered difference between the simulated forward projection of the workpiece and the measurement of the workpiece, i.e., this low-pass filtered difference represents the target data.

[0078] According to a particularly noteworthy proposal, the simulation is based on the geometric description of the workpiece, in particular target data such as preferably CAD data and / or actual data from the measurement of the workpiece, especially a master part and / or calibration data of the workpiece.

[0079] It is particularly noteworthy that the measurement and the simulation are each carried out with complete (artifact-laden), partial (artifact-reduced) or no (artifact-free or artifact-corrected) consideration of one or more artifacts, in particular some artifacts are considered or partially considered, and other artifacts are not considered.

[0080] Furthermore, the invention is characterized in that measurement and / or simulation is initially carried out without certain artifacts and certain artifacts are additionally added completely or partially by mathematical methods, such as a separate scattering beam simulation, or that measurement and / or simulation already partially or completely contains certain artifacts.

[0081] In particular, the invention is characterized in that the addition of artifacts is carried out by simulation, such as scatter radiation simulation, preferably according to a Monte Carlo method, or from a separate measurement, for example of scatter radiation, or by measurement that already contains artifacts.

[0082] During simulation, it's easy to vary the extent to which, and even whether, certain artifacts are considered by including them in the simulation or not. During training measurements, the influence of at least some artifacts can be avoided or reduced by performing specific measurements or using specific measurement parameters. Compared to the measurement that will later be corrected, training measurements can be taken that result in fewer artifacts, such as: Measurement using one or more slit apertures or scattering gratings; measurement with a modified beam filter; measurement with a longer image acquisition time (integration time); multi-spectrum tomography with different beam spectra; application of a different artifact correction method.

[0083] These artifact-reduced measurement data, or artifact-free or artifact-reduced simulation data, can be used as target data, e.g., target radiograph sets, during training, with artifact-containing measurement or simulation data serving as source data, e.g., source radiograph sets. It is also intended that the target data, such as the target radiograph set, may still contain specific artifacts, so that the trained filter operation corrects only selected artifacts while leaving others uncorrected. In general, a filter operation is trained in this way that later corrects the artifacts, or more generally, the measurement deviations, on the actual data to be corrected, such as the actual radiograph set, according to how these deviations differ between the source and target data.

[0084] For example, a complex scattering simulation, once trained, can be replaced by a filter operation of the neural network during the subsequent actual measurement of the workpiece. Another example is that artifacts are deliberately accepted during the subsequent measurement, for example by omitting the scattering grating, using different beam filters, shortening the image acquisition time, and / or performing a single tomography scan with only one beam spectrum, and these are then corrected by a suitably trained filter operation.

[0085] Preferably, artifact-free or artifact-reduced measurement compared to the acquisition of actual data such as actual radiographic image sets is achieved by: Measurement using one or more slit apertures or scattering gratings, measurement with a modified beam filter, measurement with a longer image acquisition time, multi-spectrum tomography with different beam spectra and / or application of a different artifact correction method.

[0086] Training with different beam spectra makes it possible to estimate the beam spectrum used to generate the actual data to be corrected later, and thus predict the artifacts that need to be corrected. This results in a filter operation that depends on the estimated beam spectrum.

[0087] Based on knowledge and / or estimation of the spectrum used, taking into account the expected geometry of the workpiece (e.g., known from the workpiece's target data), the expected artifacts can be at least roughly estimated. This allows conclusions to be drawn as to whether the artifacts are high-frequency, such as beam hardening, which is more clearly visible in the reconstructed volume data, or low-frequency, such as scattered radiation, which may not always be clearly visible in the radiographic image data. Filter operations are then applied for correction, which have been trained on either volume data or radiographic image data.

[0088] In particular, the invention is characterized by the fact that during training the relationships, especially spatial orientation, between radiographic images of different rotational positions are taken into account, i.e. preferably correction of radiographic images taken in adjacent rotational positions is homogenized or smoothed.

[0089] It is also worth emphasizing that the source data used to train the neural network corresponds to a stack of radiographs taken from different rotational positions, in order to have information from other rotational positions available for correct mapping to the target data. This will be explained in more detail below in connection with subsets.

[0090] Preferably, the invention provides that several simulations are carried out during training, with one or more of the following parameters being varied: Included artifacts, such as scattered radiation, beam hardening, off-focal radiation, backscatter from the detector or housing, photon starvation, noise, drift of the focal spot generating the X-rays and / or image disturbances from miscalibrations of the computed tomography scanner and / or detector deviations such as bad pixels, distortion, tilt, nonlinearities and / or sensitivity fluctuations, etc., alignment and position of the workpiece between source and detector, in particular image scale, and / or alignment and / or position of source and detector (geometry) and / or trajectory for the relative movements between source, detector and workpiece and / or acquisition parameters used such as the voltage, current and / or spectrum of the source, number of rotation steps, image acquisition time per image and / or number of images superimposed per rotation step.

[0091] It should also be emphasized that a workpiece, preferably a calibrated workpiece, is used for training the neural network, whereby a filter operation is determined which is suitable for determining the respective position of source, workpiece and detector relative to each other and / or the respective imaging scale, in particular the geometry.

[0092] The invention is also characterized in that the set of recorded radiographic images (gray values) or radiographic images normalized therefrom or radiographic images normalized and logarithmized therefrom (projection values ​​P) or radiographic image values ​​e^-P or P * e^-P is used as the actual radiographic image set to be corrected.

[0093] In particular, the invention is characterized in that, during correction, one filter operator is selected from several filter operators, taking into account one or more of the following parameters: Voltage and / or current and / or spectrum of the source, beam filter arranged in front of the source, image scale, number of rotation steps, image acquisition time per image, number of images superimposed per rotation step, relative position of source and detector to each other, in particular geometry and / or trajectory of the movement of workpiece, source and detector to each other.

[0094] According to a particularly noteworthy proposal, the method is intended to be used for the dimensional measurement of features on the workpieces by determining surface measurement points from the voxel volume using surface extraction methods and / or for the inspection of workpieces, in particular the inside of the workpiece, for example, for voids, inclusions or similar features.

[0095] It is particularly noteworthy that the correction is used to reduce measurement deviations in computed tomography and preferably to reduce the measurement uncertainty of dimensional measurements of features, in particular correcting deviations resulting from scattered radiation, beam hardening, off-focal radiation, backscattering from the detector, etc.Backscatter, photon starvation, noise, drift of the focal spot generating the X-rays and / or image disturbances such as those resulting from miscalibrations of the computed tomography scanner and / or detector deviations such as bad pixels, distortion, tilt, nonlinearities and / or sensitivity fluctuations, preferably by including the deviations to be corrected in the training of the neural network, by particularly preferably taking into account the deviations to be considered in the generation of the source data such as source radiograph sets and not taking into account in the generation of the target data such as target radiograph sets.

[0096] Furthermore, the invention is characterized in that the method is used in a computed tomography scanner which is operated in an integrated manner in a coordinate measuring machine, wherein the coordinate measuring machine preferably has further tactile, optical and / or tactile-optical sensors and is designed for operation in the sense of a multi-sensor coordinate measuring machine.

[0097] In particular, the invention is characterized in that the neural network is trained in such a way that the mathematical operation (filter operation) does not cause any change in size (scaling) and / or scaling is compensated by a separate filter operation, preferably by using source data generated during training by computed tomography of a calibrated workpiece and a model generated on the basis of the calibration data for the simulation of target data, particularly preferably by determining the filter operator for compensation by determining a scaling factor between measured and simulated data of the identical workpiece.

[0098] Avoiding size changes means that the underlying image scale of the computed tomography scanner, used for evaluating and correcting the measurement data, is not indirectly altered by the AI ​​enlarging or reducing radiographic images or parts thereof during correction. This is necessary to ensure that the workpiece dimensions are measured correctly. The AI ​​must therefore not scale the data to be corrected. Instead, the AI ​​should correct local image disturbances or artifacts of various kinds itself.

[0099] However, problems can arise when training involves comparing data with even slightly different scales. This can occur when source data measured using computed tomography and target data based on a CAD model are used to simulate target data. The target data may have a different overall dimension than the actual workpiece being measured; for example, the workpiece may have shrunk due to the manufacturing process. Training with such data would cause the AI ​​to learn a scaling filter operation. To avoid this usually undesirable effect, the invention provides that instead of using only the target data from the CAD model or similar source...The simulation uses a model as input data that corresponds to the actual size of the workpiece, meaning it is identically scaled to the source data measured by computed tomography. This model must therefore have the actual workpiece dimensions, which can be obtained by calibrating the workpiece, i.e., by measuring it with a different measuring system or method. This model is then used for the simulation, specifically the forward projection, of the target data. The image scale used to reconstruct the measured CT data should be known as precisely as possible. Otherwise, the resulting source data would have a different scale compared to the target data simulated based on calibration values, and this difference would be passed on to the AI ​​for training.

[0100] Should artifacts occur during the computed tomography measurement that trigger a scaling error in the measured values, the AI ​​would still be trained to correct the scaling. This would then be a correct correction, since the artifact to be corrected is clearly causing a scaling error that distorts the measurement. Once such artifacts are known or expected, appropriate training can be performed. Simulations with and without this artifact can be used for this purpose, allowing the effect caused by the artifact to be identified by comparing these two simulations and the AI ​​to be trained for the corresponding correction. Furthermore, comparing the simulations is also suitable for determining whether this specific scaling artifact was present during the computed tomography measurement of the workpiece (in the case of training data).to be expected (in the case of actual data to be corrected) and as a result, the AI ​​can decide whether a corresponding correction is trained (in the case of training data) or necessary (in the case of actual data to be corrected).

[0101] In particular, the invention is characterized by the fact that a so-called watershed method is used for the surface extraction process to generate surface points and / or to generate measuring points at material transitions of the workpiece and / or to segment the volume data (voxel volume), preferably for voxel-accurate or subvoxel-accurate generation of surface points or measuring points at material transitions, and is used particularly in the training of the AI.

[0102] In the so-called watershed method, also known from 2D image processing, the gray values ​​of the voxels in the 3D voxel volume are interpreted as elevation data. This elevation data is then virtually filled in from the bottom up to find watershed points and / or perform segmentation. In this way, surface transitions to the surrounding medium (usually air) or to other materials (in the case of multi-material workpieces) can be effectively located and / or local areas of the different materials can be distinguished. The method initially operates with voxel-level accuracy but can also be used as an intermediate step towards a subsequent sub-voxel-level surface extraction method.

[0103] According to a particularly noteworthy proposal, it is envisaged that a segmentation of areas, in particular of the different materials of the workpiece, is carried out based on the volume data (voxel volume), in particular wherein each voxel is assigned a material or a list of the associated voxels is assigned to each material, wherein preferably subsequently the surface points and / or measurement points at material transitions of the workpiece are determined by means of edge location finding, wherein target data in the form of volume data (target volume data) and / or target data in the form of surface points (target surface data), or in the case of a multi-material workpiece, target volume data and / or target surface data for each material are used for training the AI.

[0104] Segmentation allows for the rapid identification of material transitions in multi-material workpieces and can be implemented, for example, using the aforementioned watershed method. It can be used for AI training.

[0105] In particular, the invention is also characterized by an original inventive idea or an idea that can be combined with the aforementioned ideas, which provides that the AI, through appropriate training, performs one or both of the steps of reconstruction and surface finding, wherein the AI The actual voxel volume is calculated from the actual radiographic image set, or the actual surface points are calculated from the actual voxel volume, or the actual surface points are calculated from the actual radiographic image set, or the actual surface points are calculated from the actual radiographic image set and the actual voxel volume, wherein the corresponding source data and target data of different data types are used in the corresponding training of the AI, namely radiographic image sets as source data and voxel volume as target data, or voxel volume as source data and surface points as target data, or radiographic image sets as source data and surface points as target data, or radiographic image sets and voxel volume as source data and surface points as target data.

[0106] With this approach, the AI ​​does not, or not only, perform the trained correction on one of the data types, but rather one or both of the classic steps: reconstruction of radiographic image data (radiographic image set) to voxel volume and / or surface finding of surface points from the voxel volume. Accordingly, training must be performed with source and target data that differ in the steps to be performed by the AI. In one embodiment, it is also envisaged that the AI ​​uses radiographic image data and voxel volume, in particular reconstructed actual voxel volume, for the calculation of the surface data.

[0107] Furthermore, the invention is characterized in that the AI, through appropriate training, realizes the merging of at least two actual data sets into a merged actual data set for further processing, wherein the actual data sets were recorded with different lighting conditions, in particular different exposure intensities in a camera-based image acquisition method or different X-ray spectra in a computed tomography method, wherein preferably the actual data sets are corrected (pre-corrected) by means of the AI ​​before being merged.

[0108] Acquiring actual data under different lighting conditions is usually necessary for uncooperative workpiece surfaces (in the case of camera-based image acquisition methods) or multi-material workpieces (in the case of computed tomography methods) in order to achieve high image quality and accurate measurements. In the field of image processing, for example, the HDR method is used. In the field of computed tomography methods, at least two measurements with different spectra are performed and the results are combined (fused). The different spectra can be generated by operating the same X-ray tube at different energies or by using multiple X-ray tubes. Patent DE102016104582, filed by the applicant, describes a computed tomography scanner with two X-ray tubes with different spectra for measuring multi-material workpieces.

[0109] Data fusion is necessary to combine the actual data acquired under different illumination conditions. In the case of the computed tomography methods according to the invention, this data fusion includes considering the intensities measured in the radiographs and / or the respective spectrum used in the individual actual datasets, and normalizing or scaling the actual data, particularly the radiograph data, based on this, followed by combining the data into a single actual dataset. According to the invention, some or all of these steps are performed by AI. The necessary training is carried out using training data generated with known fusion methods.

[0110] As mentioned previously, the invention also provides that the actual data is pre-corrected by AI before being merged using the methods mentioned above and below.

[0111] According to a particularly noteworthy proposal, the invention is also characterized by the fact that the training of the AI, in particular the neural network, is continuously supplemented and in particular improved by measurement data, especially measurement data generated by users or end users, of workpieces, wherein preferably user data and / or existing training data are exchanged via a preferably encrypted data connection, particularly preferably user data for the improvement of a network stored in a cloud or on a server is transmitted to this, preferably encrypted, and preferably the improved network is transmitted to the user.

[0112] The purpose of encryption is to depersonalize the training data to protect customer confidentiality. Data contained in the test / validation dataset (existing training data) stored, for example, in a cloud or on a server belonging to the manufacturer, must be accessed during supplementary training to prevent overfitting. Standard encryption is suitable for this purpose. The improved, trained networks that the user receives back from the cloud or server pose no confidentiality issues. No information about the original training data can be recovered from them. Alternatively, the existing training data from the cloud could be encrypted and loaded into the user's physical storage, supplemented there, and the improved network then transmitted back to the cloud. However, this is usually not practical due to the large volume of data.

[0113] Furthermore, the invention is characterized by the fact that the network is located online in a cloud. This makes it particularly advantageous to use user-generated data directly for the ongoing training of the AI, i.e., for improving the online network.

[0114] In particular, the invention is also characterized by an inventive idea or an idea that can be combined with the aforementioned ideas, which provides that the training of the AI ​​is based on a subset (training subset) of the radiographic images belonging to a computed tomography scan, wherein the application of the network determined in this way is to a subset different from the training subset (actual subset).

[0115] A subset is understood to be a limited set of rotational positions relative to the complete computed tomography scan. This can consist of contiguous rotational angle ranges or several separate sub-ranges of rotational positions, distinct from one another with respect to their rotational angle. The basic idea is that the correction network is not entirely derived from or trained on training data originating from the identical rotational positions to be corrected. According to the invention, the two subsets can also partially overlap.

[0116] Further details, advantages and features of the invention will become apparent not only from the claims and the features to be derived therefrom - individually and / or in combination - but also from the following description of the figures.

[0117] It shows: Fig. 1 shows a schematic representation of the computed tomography arrangement according to the invention with a rotatable target, and Fig. 2 shows a second schematic representation of the computed tomography arrangement according to the invention with a rotatable target.

[0118] Based on the Figure 1 and 2The inventive device of an X-ray source 20 for a computed tomography scanner with a rotatable target 25, which here is designed as a transmission target, is illustrated. The source 20 has a base body 21, also referred to as a tube head, which contains a vacuum tube 22 and from which various attachments extend. An electron beam 24 is generated in the vacuum tube 22, which is focused or directed by the deflection device 23 onto a target 25 and forms the focal spot 31. The target 25 extends from a target carrier 26, which is pressed against the seal 40 by the eccentric ring 37 and the screws 40, thereby sealing the vacuum tube 22. On the inside of the target 25 is a coating 28 made of, for example, tungsten, which, upon impact of the focused electron beam 32a, 32b, generates X-rays 27 that are emitted outwards through the target (transmission target).The coating 28 is based on a substrate 29, which is exemplified as diamond. The target carrier 26 can, for example, be made of a metal. To limit the emitted X-ray radiation 27 to a predetermined area, for example, the X-ray detector of the computed tomography scanner, a beam diaphragm 36 is arranged in front of the target 25 on the outside of the tube head 21. According to the prior art, this beam diaphragm has a central rectangular opening. At least around the opening, the beam diaphragm 36 is made of a material that blocks the X-ray radiation 27, such as tungsten (shown here hatched). According to the invention, the areas further out are made of a more cost-effective material such as steel.

[0119] According to the invention, the target 25, together with the target carrier 26, is designed to be rotatable about the axis 30 in the eccentric ring 37, the axis 30 penetrating the target 25 offset from the focal spot 31, so that when the target carrier 26 rotates, the electron beam 32a generates the focal spot 31 on different areas of the coating 28 along a circular path. In the first embodiment of the invention shown here, the rotational movement is generated by the drive (motor) 33, 34, which is attached to the tube head 21. The rotatable part 34 of the motor is coupled to the outer surface 41 of the beam aperture 36, for example by a worm gear coupling, and rotates it. The rotational movement is transmitted to the target carrier 26 by the drivers 35, for example pins, by the pins 35 engaging in blind holes 38. The design as blind holes is necessary to maintain the vacuum in the vacuum tube 22.The blind holes 38 are also designed as radial elongated holes, allowing the drivers 35 to move radially within the target carrier 26. This enables the axes of rotation for the target carrier 26 and the drivers 35 to be arranged parallel to each other. This allows the use of different eccentric rings 37. Alternatively, the motor 33, 34 itself can be located within the beam aperture 36 and rotates only an inner ring (not shown) from which the drivers 35 extend. The rotatability of the target carrier 26 is achieved by pressing it firmly against the seal 40, allowing for rotation. The contact force is generated indirectly by the screws 37 and, if applicable, integrated springs 39, which press the eccentric ring 37 towards the vacuum tube 22 without rotating themselves.Alternatively or additionally, springs 39 are also provided in or parallel to the drivers 35, which act directly on the target carrier 26 and rotate with it. Since the beam aperture rotates according to the first embodiment shown here, its opening is circular. According to an alternative embodiment, it is also provided to use different circular paths on the coating 28 of the target 25 to generate the focal spot 31 by directing the electron beam 24 by the deflection device 23, for example, along path 32b onto the coating 28.

[0120] Figure 2 shows an alternative design form to Figure 1, where the reference numerals described above apply. In contrast, it is provided here that the target carrier 26 is rotated directly by the motor 33, 34 in the eccentric ring 37. In the preferred embodiment shown here, the coupling of the motor 33, 34 to the outside of the target carrier 26 is achieved, for example, by a belt, chain, or gear drive such as a worm gear. The coupling is designed to be adaptable to different eccentric rings 37 or can be exchanged with the respective eccentric ring 37. The drivers 35 are omitted, and the beam aperture 36 is designed to be non-rotatable. Alternatively, it is also provided that the motor 33, 34 is coupled to an intermediate ring or disk (not shown) arranged between the beam aperture 36 and the target carrier 26 and rotates it, with the drivers 35 extending from this intermediate ring or disk and engaging in the blind holes 38 (not shown here).

Claims

1. A device for inspecting workpieces by computer tomography, preferably by dimensional measuring of features on the workpieces, comprising a radiation source (20) for a computer tomography unit, wherein the radiation source comprises at least one basic body (21) from which extends a vacuum tube (22) containing a unit (23) for focusing an electron beam (24) on a target (25) extending from a target support (26), wherein the target (25) has a region (28) generating X-rays (27) during irradiation with electrons, said region being formed by a coating (28) of a substrate (29), and wherein the target support (26) is designed rotatable about an axis (30) using means (33, 34, 35) arranged outside the vacuum tube (22), and wherein the target (25) is preferably designed as a transmission target, characterized in that the target support (26) closes the vacuum tube (22), in that the axis (30) passes through the target (25) offset from the point of impingement (31) of the electron beam (32a, 32b) on the target (25), in that the means (33, 34, 35) extend from the basic body (21) itself or from a beam aperture (36) extending from the basic body (21) and arranged in the X-rays (27), and in that the device is designed such that point of impingement of the electron beam on the target is altered step by step.

2. The device according to claim 1, characterized in that the target support (26) extends from an eccentric ring (37) in which the target support is mounted rotatable about an axis (30).

3. The device according to claim 1, characterized in that the means (33, 34) are designed for rotating the beam aperture (36), wherein at least one driver (35) transmitting the rotary movement to the target support (26) extends from the beam aperture (36), wherein the driver(s) (35) is / are preferably pin(s), the target support (26) has blind holes (38), preferably slots extending in the radial direction, and the pins (35) are inserted into the blind holes (38).

4. The device according to claim 1, characterized in that a coupling to the target support (26) is achieved by blind holes (38) formed in the target support (26) and in which at least one driver (35) engages, wherein the blind holes (38) are designed preferably as slots extending in the radial direction, or by engagement at the circumference of the target support (26).

5. The device according to claim 1, characterized in that at least one pressing means (39, 40), such as a spring (39) and / or a setscrew (40), is provided for generating a pressing force between the target support (26) and the vacuum tube (22), said pressing means extending preferably between the tube head (21) or a holder extending from the tube head (21) on the one hand and the eccentric ring (37) on the other hand, and / or between the beam aperture (36) on the one hand and the target support (26) on the other hand.

6. The device according to at least one of the preceding claims, characterized in that at least one encoder is provided for monitoring and controlling the rotary position of the target support (26) and consists of at least the components scale and read head, wherein one of the components, preferably the scale, extends from the target support (26) or beam aperture (36), and the other component, preferably the read head, extends from the tube head (21).

7. The device according to at least one of the preceding claims, characterized in that the device is designed to receive eccentric rings (37) of differing eccentricity, preferably with the different eccentric rings having an identical outer diameter.

8. The device according to at least one of the preceding claims, characterized in that the target coating (28) is formed in segments from different materials, for example tungsten, copper or beryllium, and / or the target coating (28) is designed with a differing thickness per segment or continuously.

9. The device according to (preferably) any of the preceding claims, characterized in that a drive unit, for example a piezo drive or stepping motor, is arranged vertical to the direction of the electron beam (32a, 32b) of the X-ray source (20) for movement of the target (25), wherein particularly preferably the target support (26) receiving the target (25) is designed movable vertically to the electron beams (32a, 32b) together with the target (25).

10. The device according to at least one of the preceding claims, characterized in that the target coating (28) is formed in the radial direction in steps from different materials, for example tungsten, copper or beryllium, and / or is of differing thickness.

11. A method for inspecting workpieces by computer tomography, preferably by dimensional measuring of features on workpieces, comprising at least the following steps: recording of at least one radiographic image set made up of several radiographic images of the workpiece arranged on a rotary table in several rotary positions relative to a radiation source (20) and to a predominantly planar detector, reconstruction of a voxel volume having voxels with associated voxel gray values from the radiographic image set, and preferably determination of surface measurement points from the voxel volume by means of surface extraction methods, using a computer tomography unit at least comprising a device according to at least one of claims 1 to 10, characterized in that different points of impingement of the electron beam on the target (25), situated on at least one circular path, are set one after the other by moving the target support (26) receiving the target into different rotary positions.

12. The method according to claim 11, characterized in that the points of impingement are set such that the different points of impingement on the target (25) do not overlap, and preferably at least 90, particularly preferably at least 180 different points of impingement are set on the circumference of the circular path in a distributed manner.

13. The method according to claim 12, characterized in that different circular paths are set by using eccentric rings (37) of differing eccentricity and / or by the electron beam being aimed at different positions radially to the rotary axis about which the target support (26) is rotated.

14. The method according to claim 11, 12 or 13, characterized in that an altered point of impingement (31) is set optionally per measurement, per rotary position, per image in the case of several images per rotary position, and / or per grid position in the case of grid tomography, preferably automatically based on monitoring of the operating hours of the radiation source (20) and / or a predefined duration and / or based on the result of a cyclically repeated measurement of the wear on the target at the current point of impingement, or by further manual rotation by the operator.

15. The method according to claim 14, characterized in that the wear is measured by determining the brightness or intensity of the emitted X-rays and / or by determining the contrast in at least one radiographic image recorded using the detector during a brief setting of an altered point of impingement (31) of the electron beam (24) on the target (25) to a previously unused region of the target.

16. The method according to at least one of the preceding claims 11 to 14, characterized in that the rotary position of the target support (26) is set continuously, wherein rotation is at a low speed of preferably less than 50 µm per hour, particularly preferably less than 10 µm per hour, preferably for preventing measurement deviations, for example caused by blurring.

17. The method according to any of claims 11 to 15, characterized in that an altered point of impingement (31) of the electron beam (24) on the target (25) is set step by step and / or only when a set wear limit of the target has been reached or exceeded, particularly preferably by the target or the target support (26) receiving the target being moved into different rotary positions and / or by using eccentric rings (37) of differing eccentricity for receiving the target or target support (26) and rotating them, and / or by aiming the electron beam at different positions on the target.