Method and device for inspecting 3-dimensional objects

Abstract

The invention relates to methods for inspecting 3-dimensional objects, for example pouch cells (11, 31, 111), and a corresponding device (1, 101), wherein each object has a substantially cushion-shaped or cuboid-shaped housing with a top and a bottom, wherein the upper surface of the housing (12, 112) is composed of at least one upper surface section (13, 113) and a plurality of lateral surface sections (17, 18, 21, 22, 23, 24, 117, 118, 121, 122, 123, 124, 125, 126, 127, 128, 129) which are inclined, parallel or perpendicular to the at least one upper surface section (13, 113) or Represent corner sections (125), wherein, for each object, matrix-wise image information is acquired by means of a matrix camera from light reflected from the top surface of an area lighting device in a rest state of the object to be inspected and transmitted to a data processing device (70), wherein the matrix-wise acquired image information includes light that was reflected from the lateral surface sections, wherein the matrix-wise acquired image information is further processed as a first overall matrix by means of the data processing device, the following steps are then carried out using the data processing equipment (70): • Segmentation of the first overall matrix • Determining the type of error of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object.

Description

[0001] The invention relates to a method for inspecting 3-dimensional objects, in particular so-called pouch battery cells (hereinafter referred to as pouch cells), and a corresponding device.

[0002] Pouch cells are a type of battery, primarily used for lithium-ion batteries. A pouch cell typically consists of a pillow-shaped casing or packaging formed by a plastic-coated metal foil (e.g., aluminum foil). Therefore, such a cell is also referred to as a polymer battery. The casing is designed as a flexible, flat, and lightweight, externally sealed pouch or pillow. Inside the casing, there is usually a stack of stacked electrode layers, active layers, and separator layers. The terminals are formed as two tabs (connection lugs) that protrude from the pillow-shaped casing on one side, adjacent to each other, or on opposite sides. Pouch cells are known for their high energy density, compact design, and flexibility, making them suitable for various applications, including electric vehicles.Pouch cells can be easily enlarged or reduced in size to meet the specific requirements of different electric vehicle models. Their flat and flexible design also allows for easier integration into various areas of the vehicle, resulting in more efficient packaging and improved space utilization. A disadvantage of the pouch cell design is that, due to their construction, they are generally susceptible to mechanical damage. This can easily lead to the release of gases or electrolyte, or to significant swelling of the cells or internal short circuits.

[0003] It is therefore desirable to thoroughly inspect such and other 3-dimensional objects during quality control in order to detect damaged objects at an early stage.

[0004] Several methods for quality control of flat objects such as battery cells have already been disclosed. For example, document US 2022 / 0390387A1 discloses a method using optical coherence tomography (OCT) to inspect a gap between a lead foil and a connecting tab of a pouch cell. This allows for an assessment of the pouch cell's closure quality, but this assessment has very limited significance regarding the overall quality of the pouch cell. Document EP 4117081A1 describes a very complex inspection system comprising a thickness measuring unit, an electrical properties measurement unit, a printing unit, a tab cutting unit, a weighing unit, a tab testing unit, and a defect selection unit.The thickness measuring unit measures the thickness of the pouch cell, and the printing unit is used to print information about the pouch cell onto its surface. The tab inspection unit determines the length and shape of the tab using vision inspection. Defective pouch cells are sorted out into designated magazines by the defect selection unit. Document DE 10 2019 109 703 A1 shows and describes an arrangement for quality control of a battery cell whose transparent outer shell encloses an interior space. Inside this interior space, i.e., beneath the outer shell, is an (additional) glass rod or a lithium-metal plate which, in the presence of a predetermined concentration of hydrogen fluoride, alters the optical appearance of this space.Accordingly, this glass rod or these lithium-metal sheets are analyzed in a complex optoelectronic manner to determine the hydrogen fluoride concentration and thus the quality of the battery cell. Finally, document EP 3 869 603 A1 specifies a method for verifying the quality of laminated electrode-separator assemblies and batteries with electrode-separator assemblies, suitable for mass production and ensuring the secure and reliable bonding of the layers. The verification involves capturing at least a portion of the surface of the electrode-separator assembly using a scanning device to generate a measurement result and evaluating the measurement result. The scanning device is particularly suitable for capturing the surface topography, surface temperature, and / or surface color.This can be achieved using an optical sensor, a camera, and / or a camera. The detection device can include at least one illumination unit that can emit light onto the surface of the electrode-separator assembly to be detected. The evaluation process can include image processing and / or image analysis.

[0005] Further methods for inspecting objects or for defect detection or tab recognition are known from the documents WANG, Xu ; CHENG, Pan: “Deep learning-based visual defect inspection system for pouch battery packs”, Cognitive computing - ICCC 2022: 6th international conference ; held as part of the services conference federation, SCF 2022 ; Honolulu, HI, USA, December 10-14, 2022 ; proceedings; Cham, Switzerland; Springer, 2022 (Lecture notes in computer science), WO 2023 / 284 712 A1, EP 4 166 935 A1 and DE 10 2021 002 363 B3.

[0006] The known methods described above are either comparatively complex or allow only a very limited assessment of the quality of a three-dimensional object, particularly a pouch cell. Therefore, the object of the present invention is to provide a fast and informative method for inspecting an object that allows for a comprehensive evaluation of its quality. Similarly, the object of the invention is to provide a suitable device for inspection.

[0007] The above problem is solved by the method for inspecting a flat object, in particular a battery cell, preferably in the form of a pouch cell, with the features of claim 1 and a corresponding inspection device with the features of claim 12.

[0008] In particular, the problem is solved by methods for inspecting 3-dimensional, optionally flat objects, for example pouch cells, wherein each object has a substantially cushion-shaped or cuboid casing with a top and a bottom, wherein the top of the housing is composed of at least one upper surface section and a plurality of lateral surface sections which run parallel to the at least one upper surface section but not at the same level as the upper surface section, obliquely or perpendicularly, or which represent corner sections, wherein the underside of the housing is composed of at least one lower surface section and a plurality of lateral surface sections which run parallel to the at least one lower surface section but not at the same level as the lower surface section, are oblique or perpendicular to it, or represent corner sections, wherein, for each object, matrix-wise image information is generated by means of a matrix camera from light reflected from the top surface of an area lighting device and, if applicable, from the upper tab surface of the first tab and / or the second tab in a resting state of the object to be inspected, and transmitted to the data processing device, wherein the matrix-wise image information includes light reflected from the lateral surface sections, wherein the matrix-wise image information is further processed as a first overall matrix by means of the data processing device, the following steps are then carried out using the data processing equipment: • Segmentation of the first overall matrix ◯ in a first image information section encompassing the image information of the upper surface section and ◯ into at least one second image information section, wherein every second image information section includes the image information of at least one predefined section of the lateral surface sections (for example, without a corner section) and / or at least one predefined corner section, • Decomposition of the first image information section into a multitude of individual patches, • Determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on ◯ a separate determination for each patch of the multitude of patches as to whether the respective patch of the first image information section has one or more anomalies, using a correspondingly trained first NN algorithm, whereby an error is detected if an anomaly is present, and ◯ a detection of whether an error is present in at least one second image information section, and a corresponding classification of the respective second image information section using a correspondingly trained, second NN algorithm that is different from the first NN algorithm.

[0009] The data processing unit can output the result of the detection or determination at a predefined interface to make it accessible to a user. A display device can be connected to this interface, showing the output result. Furthermore, the underside of the object can be inspected analogously to the above and following procedure steps; that is, the type of a detected defect, the severity of a detected defect, and / or a quality indicator of the object can be determined. The underside of the housing consists of at least one lower surface section and a plurality of lateral surface sections that run obliquely, parallel, or perpendicular to the at least one lower surface section, or that represent corner sections.

[0010] The method is used for inspecting three-dimensional objects, for example, flat objects in the shape of a pillow or cuboid, particularly in the form of battery cells, preferably pouch cells. In one embodiment, the present invention can be used for a flat object, wherein a flat object is defined as a three-dimensional object that has a significantly smaller dimension in one spatial direction (e.g., height) than in the other two spatial directions and therefore essentially has the shape of a flat cuboid or a pillow shape, or a shape similar to these. Alternatively, the dimension in one spatial direction can also be larger, so that the object is described as essentially cuboid. Here, "essentially" means that the shape of the object is close to that of a pillow or a cuboid. For example, the cuboid can have sharply beveled edges.In many cases, such an object also has a first terminal tab (hereinafter referred to as "tab," e.g., the anode) and optionally at least one second terminal tab (hereinafter referred to as "tab," e.g., the cathode), each projecting laterally. Each object has a housing with a top surface and a bottom surface opposite the top surface, with any projecting tabs belonging to the housing. The method according to the invention can be used both for inspecting three-dimensional objects that have one or more such tabs and for inspecting three-dimensional objects without such tabs. In particular, the method is suitable for objects that have stepped or terraced sections, or optionally the aforementioned terminal tabs, especially at their edges.The object is thus considered such that one of its two largest sides forms a section of the upper surface, and the opposite, equally large side forms a section of the lower surface. With the upper surface on top and the lower surface on the bottom, the upper surface of the housing has at least one substantially horizontal upper surface section, which is the surface section of the upper surface with the largest extent. Further horizontal surface sections, parallel to the upper surface section, may be present, for example, a terraced surface section of the upper surface. The upper surface also has a multitude of lateral surface sections that run obliquely, parallel, or perpendicular to the at least one upper surface section (e.g., edges or side faces) or represent corner sections.The lateral surface sections also include sections parallel to the upper surface section or a surface of a projecting tab (tab surface). Accordingly, the underside of the housing has at least one substantially horizontal lower surface section, which is the surface section with the greatest extent. Further horizontal surface sections parallel to the upper surface section of the underside may be provided, for example, a terraced surface section of the underside. The underside further has a plurality of lateral surface sections that are oblique, parallel, or perpendicular to the at least one upper surface section (e.g., edges or side faces) or that represent corner sections.The first tab and any second tab (if present) may project from a short side and / or a long side, each having a top tab surface and a bottom tab surface. For example, the first and second tabs may project from a single short or long side, in which case they are positioned side by side. Alternatively, the first and second tabs may project from opposite short or long sides. The housing may have a substantially rectangular shape when viewed from the top or bottom (disregarding any tabs). The short side is the short side of this rectangle, and the long side is the long side of this rectangle.

[0011] To acquire image information from the 3-dimensional objects to be inspected, the objects are moved within a device used to carry out the procedure. This movement includes periods of inactivity in which the object remains stationary within the device. The movement of the 3-dimensional objects is achieved by a motion device that causes the objects to be moved relative to the line illumination device at a predetermined speed (motion state). This speed may be, for example, essentially parallel to the upper surface section, in the direction of the greatest extent of the upper surface section (length), or perpendicular to it. The predetermined speed is, for example, at least 500 mm / s or at least 800 mm / s.Furthermore, the motion device is configured such that, during the further movement of the respective object, it causes the object to remain stationary at a predetermined position for a predetermined period of time relative to a surface lighting device (stationary state). The predetermined period of the object's stationary position can be prior to or subsequent to the movement state. For example, the predetermined period in the stationary state can be at least 300 ms or at least 400 ms. In one embodiment, the motion device is implemented by a carriage that can be moved along a linear unit in a predetermined manner. The carriage has, for example, suction cups by means of which the object's housing can be attached to the carriage on its underside. The motion information for the movement of the 3-dimensional object to be inspected (i.e.,Its arrangement in motion and at rest states is recorded by a motion detection device and transmitted to the data processing unit. There, the recorded motion information is used together with the image information from the matrix camera to determine the presence of errors and / or the quality indicator.

[0012] The area lighting device illuminates the top surface (preferably the entire top surface) of the housing, optionally including the upper tab surfaces of the first and second tabs of the object to be inspected, from an oblique angle above. For example, the light from the area lighting device strikes the top surface of the housing, optionally including the upper tab surfaces of the first and second tabs, at an angle of incidence ranging from 10° to 60° with respect to the horizontal direction. This oblique illumination by the area lighting device allows for the easy detection of defects such as indentations, protrusions, scratches, folding defects, edge cracks, sealing defects, and similar topological defects. Defects in the form of absorbing defects (e.g., contamination, foreign matter on the surface) can also be detected.The area lighting device is implemented using LED spotlights or other quasi-point light sources. In one embodiment, at least one second deflecting mirror is arranged above the position of the object to be inspected in its resting state. This mirror runs perpendicular to the horizontal direction and deflects the light from the area lighting device so that it falls obliquely from above onto the top of the pocket-shaped housing, optionally equipped with tabs. This allows the overall external dimensions of the inspection device to be made smaller.

[0013] A matrix camera captures the reflected light from the obliquely illuminated top surface of the housing, optionally including the upper tab surface of a stationary object to be inspected, in a matrix-like manner as image information (intensity and, in one embodiment, additionally a color value) and transmits this captured image information (image data) to the data processing unit. In one embodiment, the matrix camera can capture the entire top surface of the housing. The matrix camera is positioned, for example, above the object when the object is in its stationary state at the specified position.

[0014] The matrix camera can be implemented as a CCD or CMOS camera, for example. Within its field of view, the matrix camera detects the light intensity of numerous pixels arranged in rows and columns, i.e., in a matrix. For this purpose, the matrix camera has a light-sensitive element (e.g., a CCD or CMOS sensor) for each pixel. The size of the area captured by each light-sensitive element determines the resolution of the matrix camera. For example, the matrix camera can have a field of view of 9344 x 7000 pixels or 8192 x 8192 pixels, thus capturing, for example, an image of 805 x 603 mm matrix-wise within the field of view. The row-wise capture can, accordingly, cover an area of, for example, 16 to 128 x 1000 to 8192 pixels.The matrix camera is positioned so that it looks vertically downwards at the object to be inspected in its resting state, ensuring that this portion of the field of view is in focus. The matrix camera is focused to achieve the most uniform sharpness possible across its entire field of view. This is particularly relevant for a line of sight where image information from the object (for example, from lateral surface sections) is reflected into the matrix camera via mirrors. This is accomplished by adjusting the aperture to achieve the necessary depth of field.

[0015] The inspection method according to the invention is characterized in that, by means of segmenting the matrix-wise acquired image information ("first overall matrix of image information"), different elements of the image information, which place different demands on the analysis for the inspection, are separated and analyzed using different methods. Advantageously, the image information of the comparatively flat and, in terms of its extent, prominent area of ​​the upper surface section ("first image information section") is initially processed and analyzed separately from the image information of a lateral surface section, including the corner sections ("second image information section"), which has a smaller extent and is expected to exhibit greater unevenness.For the first image information section, it has proven advantageous in terms of processing speed to divide this section into a large number of patches and then use the first neural network algorithm, as described in more detail below, to determine whether one or more anomalies are present in each patch. If at least one anomaly is present, an error is detected. In the second image information section, a second neural network algorithm, which is explained in more detail below, can be used to detect whether an error is present. Furthermore, this error can be classified.

[0016] For the overall quality assessment, the individual analyses are combined and considered together. In particular, the type of detected defect is determined, and / or its severity is assessed, and / or a quality indicator is calculated for the object, allowing for an evaluation of the object's quality. For this purpose, the detected defect types and / or their severity are also used. As described above, it is advantageous that the respective image information sections obtained through segmentation are analyzed using different algorithms adapted to possible defect types. This can also result in time savings and improved accuracy in the evaluation of the image data.In other words, by analyzing the different sections separately, meaningful results are quickly obtained during inspection, as image data processing is adapted to the specific characteristics of the object.

[0017] Segmentation is primarily performed using so-called layout recipes. Each layout recipe defines a predefined section of the object with respect to the field of view of the matrix camera / camera. Since the object is not always exactly in the predefined ideal position when the matrix camera / camera captures the image, but may be shifted / rotated by a few pixels, a position correction is performed, for example, using predefined fixed points of the object. This means that the object is registered to the expected position, so that the first overall matrix (or, accordingly, the first n overall matrices, or the second overall matrix determined from the row-by-row analysis) is adjusted to the ideal position of the object. The aforementioned matrices containing image information are then rotated and / or shifted accordingly.Once this adjustment has been made, the desired image information sections can be reliably identified and extracted according to the given layout recipe.

[0018] From the image information (image data) transmitted by the matrix camera to the data processing unit, error types such as inclusions, craters (dents), protrusions (bumps), contamination (dust, electrolyte residues), pseudo-edges, orange peel, pores, cracking, grinding marks, specks, surface defects, blistering, scratches, and wet marks are determined by appropriate data processing in the data processing unit, which is described in more detail below.

[0019] In one embodiment of the method, the classification of the at least one second image information section is carried out using a classifier with two states or a classifier with at least 3 states, wherein the classifier with at least 3 states allows, for example, the assignment of different error types, while a classifier with two states is used to assess the presence or absence of an error.

[0020] In one embodiment of the method, for each object, a multitude of line-by-line image information is generated by a camera, for example a matrix camera, from reflected light of a line illumination device on linear areas of the top surface (optionally including the upper tab surface of the first tab and / or the second tab) in a moving state of the object to be inspected, and transmitted to a data processing device, wherein the following steps are carried out by means of the data processing device: • Assembling the image information captured row by row into a second overall matrix comprising the image information of the top side of the object, • Determining the type of error of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, additionally based on the image information of the second overall matrix.

[0021] Determining the error type and / or severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on the image information of the composite second matrix, can in one embodiment also be carried out instead of the determination described above, based on the separate determination for each patch of the multitude of patches, whether the respective patch of the first image information section has one or more anomalies, using a correspondingly trained first NN algorithm, and in combination with the detection described above, whether an error is present in the at least one second image information section, and a corresponding classification of the respective second image information section using a correspondingly trained second NN algorithm.Alternatively, determining the error type and / or severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on the image information of the composite second overall matrix, can also be performed in an exemplary embodiment instead of the above-described detection of whether an error is present in the at least one second image information section and a corresponding classification of the respective second image information section using a suitably trained second NN algorithm, and in combination with the above-described determination based on the separate determination for each patch of the multitude of patches whether the respective patch of the first image information section has one or more anomalies, using a suitably trained first NN algorithm.

[0022] When assembling the image information of the object, captured line by line, the recorded "image lines" are combined to form a matrix image (second composite matrix). This second composite matrix image is then rectified so that the resulting image is aligned with the first composite matrix or one of the first n composite matrices. The second composite matrix is ​​used for the further analysis of the image data described below.

[0023] The linear lighting device illuminates a linear area of ​​the top of the housing (optionally including the upper tab surfaces of the first and second tabs). For example, the linear lighting device is formed by a luminaire with a plurality of LEDs arranged to illuminate a desired linear area. This can be a single LED array or, for a wider linear area, several adjacent LED arrays (e.g., 2 to 10 LED arrays). In one embodiment, the linear lighting device is configured to illuminate each point of the linear area with light at two different intensities (i.e., high intensity A and low intensity B).Accordingly, the light reflected from the linear area of ​​the top surface (possibly the upper tab surface of the first and / or second tab) is captured line by line using a modified switching rhythm in the form ABABAB... (i.e., the two different intensities A and B are switched alternately). The line-by-line acquisition of the image information (frequency and timing of acquisition) and the feed rate of the motion device are synchronized for this purpose. This illumination is also known as HDR Reflection Bright-Field illumination and, due to the special lighting technique, offers advantages in the detection of certain types of defects, such as contamination of the object with a transparent substance.The line-by-line acquisition of image information can be performed by a camera with a suitable field of view and resolution, for example, also by a matrix camera, which also performs matrix-by-matrix image acquisition. After acquisition by the matrix camera, the data (image information) is transmitted to the data processing unit and used for the inspection of the 3-dimensional object. Before analysis, as described above, the line-by-line acquired image information is combined to form a complete image (second overall matrix) of the respective top surface of the object.

[0024] It should be emphasized that in one embodiment, a single matrix camera captures the reflected light of the line-illuminated area of ​​the top of the housing of the object to be inspected in the moving state, line by line, in the form of image information, and the reflected light of the entire top of the housing (possibly including the upper tab surface) of an object to be inspected in a stationary state, illuminated obliquely from above, is captured matrix by matrix, and this captured image information (image data) is transmitted to the data processing device.Line-by-line acquisition represents a sub-area of ​​the matrix camera's field of view, resulting in one or more adjacent pixel rows (e.g., 16 to 128 pixel rows) containing image information. Matrix-by-matrix acquisition, on the other hand, results in a pixel matrix containing image information, which also represents a sub-area of ​​the field of view. In one embodiment, the image information can be determined within a predefined wavelength range. The field of view of the matrix camera is designed such that both matrix-by-matrix and pixel-by-pixel image information can be acquired using a single matrix camera, which is then assigned to the respective object by the data processing unit.Using a matrix camera for matrix-wise and line-wise image acquisition has the advantage that the resulting image data does not need to be calibrated with respect to the acquisition instrument. They all contain the same camera characteristics. Incorporating the line-wise acquired image information into the inspection process further improves quality assessment.

[0025] In one embodiment of the method, the following further steps are carried out using the data processing device with regard to the above analysis of the image information of the second overall matrix: • Segmentation of the second overall matrix ◯ into a third image information section comprising the image information of the upper surface section and / or ◯ into at least one fourth image information section, wherein each fourth image information section includes the image information of at least one predefined section of the lateral surface sections (where corner sections may be excluded) and / or at least one predefined corner section, • Decomposition of the third image information section into a multitude of individual patches, • where determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, based on • a separate determination for each patch of the multitude of patches as to whether the respective patch of the third image information section has one or more anomalies, using the first NN algorithm, whereby an error is detected if an anomaly is present, and / or ◯ a detection is performed to determine whether an error is present in at least one fourth image information section, and a corresponding classification of the respective fourth image information section is performed using the second NN algorithm.

[0026] The image information, assembled line by line and previously acquired line by line, can be examined for the presence of errors in the third image information section defined above, analogous to the matrix-acquired image information, using the first NN algorithm after decomposition into a multitude of individual patches, and / or in the fourth image information section using the second NN algorithm. This method is also applicable to the image information acquired line by line and offers the advantages mentioned above.

[0027] In one embodiment of the method, at least n (n ≥ 2) images (captures by the matrix camera) of the entire top surface (optionally including the upper tab surface) of matrix-wise captured image information are generated for each object by the sequential capture of reflected light. This light is generated from an illumination sequence of the entire top surface of the housing of the object to be inspected in its resting state (optionally including the upper tab surfaces of the first and second tabs of the object to be inspected) from n different directions obliquely above using the area illumination device. The n images are transmitted to the data processing device.Accordingly, in this embodiment, the area lighting device is configured to illuminate the entire top surface of the housing of the object to be inspected in its resting state (optionally including the upper tab surfaces of the first and second tabs of the object to be inspected) sequentially from at least n different oblique directions from above. The matrix camera is configured to capture the image information sequentially in a matrix-wise manner when illuminated from the at least n directions by the area lighting device. The data processing device is configured to receive and process the n matrix-wise captured image information when illuminated from the n directions by the area lighting device, and to assign this image information to the respective object.The image information, captured matrix-wise in the form of n images, is further processed as n first total matrices using the data processing device.

[0028] In one embodiment of the method according to the invention, the following further steps can be carried out using the data processing device: • Segmentation of the first n total matrices ◯ in n fifth image information sections comprising the image information of the upper surface section of each of the n first total matrices and / or ◯ in n sixth and optionally further image information sections of each of the n first total matrices, where each sixth and optionally further image information section contains the image information of at least one given ◯ Section of the lateral surface sections and / or at least one specified corner section includes, • Determining one maximum image and / or one absorption image and / or one topology image from the image information of the fifth image information section and / or the sixth image information section and / or any further image information sections, • Determining, analyzing, and characterizing errors in the maximum image and / or the absorption image and / or the topology image of the fifth image information section and / or the sixth image information section and / or, if applicable, the further image information sections and / or the third image information section and / or the at least one fourth image information section, • wherein the determination of a fault type of a detected fault and / or a severity of a detected fault and / or the determination of a quality indicator which allows an assessment of the quality of the object is based on the result of the analysis and / or characterization of the faults identified in each case.

[0029] As explained in more detail below, further acceleration of the inspection process can be achieved through additional error analysis and determination using methods that do not employ neural network algorithms. This is because, for example, the threshold analysis of the pixels in the maximum image and / or the absorption image and / or the topology image, as well as the generation of the maximum image and / or the absorption image and / or the topology image, can be performed very quickly. Both matrix-generated and row-generated image information can be used for this purpose.

[0030] In this embodiment of the method with n images of the matrix-acquired image information, in particular the image information of the fifth, sixth, or subsequent image information sections, the image information of a maximum image and / or a topology image and / or an absorption image is used as image information for further analysis. The matrix image, the topology image, and / or the absorption image is each generated from the n images of the respective image information section, acquired sequentially in a matrix-like manner. The maximum image represents the image information of the areas that are most easily accessible with respect to the respective lighting conditions and are therefore displayed most brightly. The topology image has the advantage of highlighting topological changes in the image, while the absorption image highlights errors that are based on the absorption of light (e.g.,(Soil on the surface), accentuated.

[0031] For example, the image information of the n acquisitions is generated pixel-identically, meaning that the image information of the at least two matrix-wise acquisitions of the entire top surface (optionally including the top tab surface of the first and second tabs) is generated from the same locations on the surface. Each of these matrix-wise acquisitions is called an image information matrix M, with at least two image information matrices Mk (k ≥ 2, k = 2 ... n) being acquired for each object. A pixel Pi of the acquired first image information matrix M1 thus corresponds to the same location on the surface of the top surface (optionally including the top tab surface) as the same pixel Pi of the acquired second (third, fourth, etc.) image information matrix Mk (M2, M3, M4, ... Mn). The acquired light intensity in pixel Pi is denoted as i(Pi).The detected light intensity of the first image information matrix M1 at pixel Pi is denoted as i1(Pi). Each image information matrix Mk contains, for example, the image information of the first image information section defined above.

[0032] The maximum image can be determined by calculating the maximum of the light intensities of all image information matrices Mk in the respective pixel Pi, i.e., Max(i1(Pi), i2(Pi)) for two determined image information matrices M1, M2 for two illuminations from two different directions, or Max(i1(Pi), i2(Pi), ... i n (Pi)), when n illuminations from n different directions are used. In one embodiment, n = 4. The maximum is calculated for each pixel Pi and, represented in the entire matrix (maximum matrix), yields the maximum image.

[0033] The topology image and the absorption image can be determined by first applying two differently parameterized low-pass filters (e.g., box filters) independently to each image information matrix Mk of each lighting situation and then subtracting them from each other: Fk=lowpass1(Mk)−lowpass2(Mk)

[0034] The parameters of the two low-pass filters, lowpass1 and lowpass2, differ, for example, such that the first parameter of the first low-pass filter, lowpass1, is smaller than the second parameter of the second low-pass filter, lowpass2. The light intensity assigned to each pixel Pi of the matrix Fk by this operation is denoted as fk(Pi) (k = 2 ... n). Subsequently, from the resulting matrices Fk, analogous to the maximum image above, a minimum matrix, MinM, and a maximum matrix, MaxM, are calculated by determining the minimum and maximum pixel-wise, respectively, across all matrices Fk. Each point Pi of the minimum matrix, MinM, is calculated as Min(f1(Pi), f2(Pi), ... fn(Pi)), and each point Pi of the maximum matrix, MaxM, is calculated as Max(f1(Pi), f2(Pi), ... fn(Pi)).Subsequently, a matrix H containing the values ​​h(Pi) is determined, which is calculated from the product – again determined pixel by pixel – of the minimum and maximum values ​​calculated at each point Pi, using a scaling factor a (for example, a = 64). This means that the value is obtained for each point Pi. h(Pi)=Min(f1(Pi),f2(Pi),…fn(Pi))∗Max(f1(Pi),f2(Pi),…fn(Pi))∗a

[0035] From this, a matrix Q with the values ​​q(Pi) is finally determined, where q(Pi)=sqrt(abs(h(Pi))), where abs(q(Pi)) is the absolute value of q(Pi) and sqrt() is the square root function. From this, the values ​​of the topology matrix T with the values ​​t(Pi) are obtained as follows: t(Pi)=q(Pi) if h(Pi)≤0 or t(Pi)=0 if h(Pi)>0.

[0036] Accordingly, the values ​​of the absorption matrix A with the values ​​a(Pi) are obtained as follows: a(Pi)=q(Pi) if h(Pi)>0 or a(Pi)=0 if h(Pi)≤0.

[0037] The topology matrix T calculated in this way with the values ​​t(Pi) is also called the topology image and the absorption matrix A with the values ​​a(Pi) is also called the absorption image.

[0038] If the matrix-wise acquisition of the light reflected upwards from the entire top surface (including the upper tab surface, if applicable) by the surface illumination device is performed four times under illumination from four different oblique directions from above, the directions are chosen, for example, such that the illumination comes from both opposite long sides and from both opposite short sides of the housing. Alternatively, the illumination can strike the top surface from the direction of the four corners of the housing. In all cases, it is advantageous if the images are acquired under illuminations that, with respect to the component lying in the plane of the upper surface section, cover as close as possible to a 360° angle (i.e.,When illuminated from four different directions, the illumination is provided from directions offset by 90° each time, or when illuminated from six different directions, the illumination is provided from directions offset by 60° each time, etc.).

[0039] In this embodiment, the maximum image and / or absorption image and / or topology image and / or the third image information section and / or at least one fourth image information section generated for the fifth image information section (e.g., an upper surface section) and / or the sixth image information section (e.g., four lateral surface sections) and, if applicable, a further image information section (e.g., four corner sections) are then used for further inspection. For example, pixel by pixel, using one or more threshold values ​​specified for the respective image (maximum image, absorption image, topology image) and / or for the third image information section and / or for the at least one fourth image information section, locations in the respective image / section are searched for where the respective value of the respective pixel exceeds or falls below the respective threshold value.It is assumed that if the respective threshold is exceeded or fallen below, an error has occurred. Subsequently, further properties of these errors are determined, such as their extent (in pixels) and a histogram of grayscale values ​​within the error area. From this, the respective error type and / or severity can be determined. Analysis values ​​and the assignment of the error type can be performed, for example, using corresponding tables stored in the data processing system.

[0040] In one embodiment, the data processing unit extracts four corner image information matrices from the segmented image information, or from the segmentation itself, which includes at least a second, fourth, sixth, or further image information section. Each matrix contains one corner of the housing, and its position is precisely known, for example, after the position correction described below. Alternatively, a different image information section of the object can be extracted. Optionally, a maximum matrix is ​​generated beforehand from several images.

[0041] The corner image information matrix of each corner (i.e., each corner segment) can be analyzed separately, for example, using a convolutional neural network (CNN) algorithm as a second NN algorithm, which includes a binary classifier (a classifier with two states, namely "intact corner" and "defective corner"). The corner image information matrix, or alternatively the image information matrix of the other image information segment, can be defined as comparatively small (e.g., 128 x 128 pixels) and includes only the area of ​​the corner of the housing. This CNN model was specifically designed for this classification task with few classes and for matrices with few pixels. This CNN model has a compact structure, which is, for example, more compact than conventional deep architectures. It consists, for example, of three strands with different convolutional sizes that are subsequently merged.This structure reduces the number of parameters to be trained, meaning the model learns fewer features. Therefore, it is ideally suited for binary or other low-dimensional classification. For training and evaluating the CNN model, a large dataset containing matrices of corner structures and their corresponding expected defects is used. This dataset is compiled for each object under investigation and annotated by engineers. It includes images of the corners of the respective objects, divided into two classes: "defective" and "intact." The images were carefully selected and annotated to ensure they covered a wide range of defects and variations in the battery corners. The model was trained using this dataset, with the images divided into training sets and validation sets.For example, a ratio of 80% for the training data and 20% for the validation data can be used. Additionally, five-fold cross-validation can be performed to ensure that all images are present in both the training and validation data. The model consists of several convolutional layers, pooling layers, and fully connected layers, which enable it to extract key features from the images and detect the subtle differences between defective and intact corners. The convolutional layers are used to train the weights of the convolutional operations, which are then used to identify image features. These features are then aggregated using the pooling layers. Finally, the weights of the fully connected layers are iteratively trained to determine a probability for the corresponding class based on the features.Additionally, prior to analysis with the NN algorithm, all images / sections were cut out in the same size as the corner image information matrices to be analyzed (128x128 pixels) and aligned in such a way that they were brought into the same orientation to enable a consistent view.

[0042] In another embodiment, the data processing device extracts at least one predefined area of ​​the lateral surface sections (e.g., terrace sections of a pouch cell) from the segmented image information, which includes at least one second image information section and / or at least one fourth image information section and / or a sixth or further image information section, or from the segmentation itself. Each predefined area is analyzed using an object detection network based on the mask RCNN model as a further second NN algorithm.The algorithm detects a wide variety of defect types (for example, six different defect types such as protrusion / nose / bulge, indentation, crease, scratch, contamination, and particle) and adds a corresponding bounding box to the data of the relevant second image information section in the defect area. To train the Mask-RCNN model as an algorithm, data representing a corresponding second image information section is used, with the relevant defects annotated and bounded within these image information matrices. The architecture of the Mask-RCNN model was carefully selected. This model represents an evolution of the Faster R-CNN model and is capable of generating bounding boxes and masks for defects in the specified areas.Rare error types are artificially inserted into corresponding image information sections of the predefined areas for model training. The performance of each model is evaluated using a separate validation dataset. For example, a validation dataset with a ratio of 80% for training data and 20% for validation data can be used. This ensures that the model efficiently and accurately detects and correctly classifies errors.

[0043] In another embodiment, the data processing device extracts at least one predefined area of ​​the upper surface section from the segmented image information, which includes at least one first image information section, or through the segmentation itself. In one embodiment, this area comprises the entire upper surface section. Furthermore, the resolution in the predefined area is reduced to a predefined value (e.g., from 5120 × 2216 to 841 × 265 pixels) to accelerate the process. The image is divided into several small patches (sub-areas). Subsequently, a pre-trained CNN algorithm, Wide ResNet-50, is used as the first NN algorithm to examine each patch to determine whether one or more predefined features are present.In "Wide ResNet-50," the network layers are made "wider" by increasing the number of channels in the convolutional layers. This CNN is capable of recognizing complex patterns and textures. Furthermore, it has been observed that such wider CNNs often generalize better, meaning they can process new, unknown data more effectively. The method, also known as PaDiM (Patch Distribution Modeling Framework for Anomaly Detection and Localization), is an algorithm for anomaly detection and localization. This approach is particularly well-suited for industrial defect detection, where the goal is to identify irregularities or deviations from the norm in visual data. PaDiM models the distribution of features in an image. Subsequently, the features extracted by the CNN are collected for each patch.For each patch, the Mahalanobis distance between the patch's features and a normal distribution derived from the training data is calculated. This step determines how 'abnormal' or unusual each patch is compared to normal training data. The calculated Mahalanobis distance serves as an anomaly score, with a higher value indicating a greater deviation from normality. Based on the anomaly score, a threshold is established. Patches with a score exceeding this threshold are considered abnormal. Anomalies are localized by marking the positions of the patches classified as abnormal within the image information section, thus enabling the localization of the anomalies within that specific image information section.

[0044] The anomaly score is calculated separately for each patch by determining the Mahalanobis distance of its features from the expected normal distribution, represented by the mean and covariance matrix from the training data. A large Mahalanobis distance indicates that the patch's features deviate significantly from the normal distribution, suggesting a potential anomaly. Mathematically, the Mahalanobis distance D of a point x to a distribution with mean µ and covariance matrix Σ is calculated as follows: D(x)=(x−μ)T∑−1(x−μ) • D(x) is the Mahalanobis distance for the point • x is the vector of observed values. • µ is the mean vector, based on the training data set. • Σ is the covariance matrix of the training data. • Σ -1 is the inverse of the covariance matrix. • T denotes the transposition of the vector.

[0045] The anomaly score of each patch provides an assessment of the severity of the error present in that patch.

[0046] Applying the second NN algorithm (CNN algorithm, Mask-RCNN algorithm) to the corresponding image information sections also yields detected errors, an error type, and, for each detected error (including defective corners), a severity level for the respective error (or defective corner). Furthermore, as explained above, errors in the respective image / section are also detected, analyzed, and characterized by error detection using a maximum image and / or an absorption image and / or a topology image and / or an image information section based on row-by-row determined values. Here, an error type is also assigned, and the severity level of the respective error is determined. The respective severity level can, for example, be assessed based on the size of the respective error, with a small error receiving a severity level from a small class (e.g.,A minor defect (e.g., class 1) is assigned a severity level corresponding to a higher class (e.g., class 4), while a more extensive defect is assigned a higher severity level. Alternatively or additionally, further parameters of the defect can be used to assess its severity, such as the location of the defect on the object.

[0047] The magnitude of the error can be determined, for example, by the data processing unit after taking into account the perspective and / or the optical distortion of the matrix camera. Similarly, a dimension of the object can be determined (e.g., the edge length of the housing).

[0048] After the above information regarding the detected errors and their severity has been collected, the data processing system performs an overall assessment for each object, for example, using a quality indicator. All detected errors and their severity are taken into account, with different errors being weighted differently depending on their severity. From this, for example, by applying a corresponding table provided in the data processing system, it is determined whether the object meets the specified quality requirements or not, i.e., whether the calculated quality indicator is less than, greater than, or equal to a quality indicator threshold.In both cases, the determined data (error, severity of the error, size of the error, location / place of the error) can be output to an interface of the data processing device and made available for further processing of the object.

[0049] For image acquisition, in one embodiment, the matrix camera can be configured (e.g., controlled by the data processing unit) such that the image information is captured line by line and matrix by matrix in a single acquisition sequence (a temporal sequence of images taken by the matrix camera across its entire field of view). This can be synchronized with corresponding lighting control (i.e., the line lighting and / or area lighting). In one embodiment, the image information to be captured line by line for a first object to be inspected (in motion) can be captured at least partially simultaneously with the image information to be captured matrix by a second object to be inspected, different from the first (at rest).Such a design of the recording sequence can shorten the overall time required for the quality assessment of the object. This means that the matrix camera is set up in such a way that at least one of its recordings of a capture sequence is captured in the same recording. • the line-by-line captured image information from the linear area of ​​the top (possibly including the top tab surface of the first tab and / or the second tab) reflected light from the line lighting device in the moving state of a first object and • The matrix-acquired image information includes upward-reflected light from the area illumination device of the entire top surface (optionally including the top tab surface of the first and second tabs) in the resting state of a second object, and light reflected from the lateral surface sections of the second object via the first deflecting mirrors into the matrix camera, the second object being different from the first object (i.e., in the same acquisition). Acquisition and illumination sequences may, for example, include a large number of line-wise acquisitions of the light reflected from the linear area of ​​the top surface (optionally including the top tab surface of the first and / or second tab) (e.g., between 50 and 120 line-wise acquisitions) and, optionally partially in the same acquisition, some (between 5 and 20) matrix-wise acquisitions of the entire top surface.Alternatively, the image information to be captured line by line and the image information to be captured matrix by matrix from two different objects can be acquired sequentially by the matrix camera in the acquisition sequence. In this case, to save time, only sections of the entire pixel matrix of the matrix camera can be read out, e.g., the corresponding section of the line-by-line acquisition and the corresponding section of the matrix-by-matrix acquisition.

[0050] During image acquisition, the matrix camera is stationary (i.e., it does not move, nor do any parts of it), and the dimensions of the matrix camera's field of view are designed such that both the image information to be acquired line by line and the image information to be acquired matrix by line are contained within the same field of view. During line-by-line acquisition, the object being inspected is in motion; that is, it continues to move while the image information is being acquired. In contrast, during matrix-by-line acquisition, the object being inspected is stationary at a predetermined position for a predetermined period (i.e., in a state of rest), thus enabling precise determination of the matrix-acquired image information.In addition to the object to be inspected, at least two first deflection mirrors are provided, which are also captured by the field of view of the matrix camera and which provide further image information from the lateral surface sections of the top of the object to be inspected. This information is captured together (simultaneously, i.e., in the same image) with the matrix-wise acquisition of the object to be inspected. The image information acquired line by line and the image information acquired matrix-wise, including the image information transmitted via the first deflection mirrors, are assigned to the respective inspected object and included in the determination of the presence of a defect of at least one defect type and / or a quality indicator.

[0051] The arrangement and inclination of the first deflecting mirrors are designed such that the matrix camera receives the light reflected from as large an area as possible of the respective lateral surface sections of the top surface. In one embodiment of the device, at least two, and in particular four, first deflecting mirrors are provided, with each first deflecting mirror being arranged next to one side of the housing when the object is at rest. With four first deflecting mirrors, the reflected light from the lateral surface sections of all sides of the housing can be captured. Each first deflecting mirror is designed, for example, such that its length (largest dimension, dimension parallel to the respective side next to which the first deflecting mirror is arranged) corresponds at least to the length of the respective side of the housing.Furthermore, in one embodiment, each first deflecting mirror is arranged horizontally at a distance of at least 30 mm from the respective side of the housing. In another embodiment, the width of each first deflecting mirror (dimension perpendicular to the respective side next to which the first deflecting mirror is arranged) is at least 20 mm. The tilt angle of the first deflecting mirror is, for example, at least 30° to the horizontal direction. Moreover, for the accuracy of the inspection, it is advantageous if the deflecting mirrors achieve very good optical imaging quality in order to avoid distortions in the image of the matrix camera.

[0052] The illuminated linear area can extend over the entire length of the top surface, including any protruding first and second tabs. The length of the top surface is defined as the dimension of the housing in the direction of its greatest extent. In this embodiment, image information for the entire top surface can be obtained using the illuminated linear area as the entire object is moved past the line illumination device.

[0053] The matrix camera can be calibrated such that the data processing unit can take perspective and / or optical distortion into account from the matrix-acquired image information. For such calibration, the method described, for example, in the article "Digital camera self-calibration," CS Fraser, ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997), pages 149-159, is used. In one embodiment of the device, the data processing unit is configured to determine at least one dimension of the object and / or at least one magnitude of a detected error after taking perspective and optical distortion into account. For this purpose, a lookup table is determined beforehand based on the calibration, which allows the conversion of a pixel count into a unit of length or area.The look-up table is stored, for example, in a storage unit of the data processing facility.

[0054] In one embodiment, position correction can also be performed using the data processing unit based on calibration and by utilizing fixed points (for example, the corners of the housing). Here, the coordinates of the four corners of the housing are determined by software-based "probing" in the horizontal and vertical directions. This probing involves examining the respective rows and columns of the image information matrix for a jump in intensity (a significant increase or decrease in intensity from one pixel to the next). This is advantageous for comparing the captured image information of the matrix with corresponding target values ​​to determine errors or to establish a quality indicator, since the object cannot always be positioned exactly in the same location when stationary.In one embodiment, position correction allows the location (position) of each detected defect to be determined, particularly on the top surface (possibly including the upper tab surface). Based on this location information, a marking device downstream of the inspection apparatus can, for example, mark the defect by applying (e.g., spraying) a water-soluble paint by circling it on the object's surface. Alternatively or additionally, knowing the defect location can facilitate the control of a defect removal device.

[0055] In one embodiment, the data processing unit can incorporate image information from at least one additional camera into the inspection process. This camera is mounted, for example, below the matrix camera on the frame. It views the top surface of the object from above, such as the upper tab surface of a tab, and optionally an adjacent section. This additional camera can, for example, generate higher-resolution images of the specified sections of the object. The image information acquired from the corresponding viewing areas is transmitted to the data processing unit, which then generates further information regarding minor defects in these sections.

[0056] The inspection procedure for the object can be implemented as a computer-implemented procedure based on the captured image information, i.e., as a procedure carried out with the data processing unit (computer). The procedure can also include the control of the line illumination unit and / or the area illumination unit and / or the matrix camera such that a predefined capture and / or illumination sequence is implemented. For this purpose, the data processing unit and the line illumination unit and / or the area illumination unit are connected to each other either wired or wirelessly. The matrix camera is also connected to the data processing unit either wirelessly or via a wired connection, also for the purpose of transmitting the image information captured by the matrix camera to the data processing unit.

[0057] The data processing unit for processing image information and determining whether an error of at least one type is present, and / or determining which quality indicator can be assigned to the object, comprises a processor that represents a functional module which interprets and executes instructions / commands from algorithms and includes an instruction control unit, an arithmetic unit, and a logic unit. The processor can include at least one microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a discrete logic circuit, and any combination of these components. The data processing unit can also include a storage unit, an input module (e.g., keyboard or touchpad), a power supply module (e.g., a power supply unit), and a power supply module.The data processing unit comprises a battery and a display module (e.g., a display). The data processing unit can be a physical hardware resource, such as a smartphone, desktop computer, server, notebook, cluster / warehouse scale computer, embedded system, or similar, or a virtualized computer resource. Furthermore, the data processing unit can include a transceiver for exchanging data / image information with a display. The data processing unit also has an interface for exchanging data with the line lighting device, the area lighting device, the matrix camera, and / or a control unit for the motion device.

[0058] As already explained above, the procedure described above can be realized, for example, as a computer program or computer-implemented procedure comprising instructions which, when executed, cause a processor of the data processing unit to perform the steps of the procedure described above, wherein the computer program includes a combination of the steps and data definitions described above which enable the computer hardware to perform computational or control functions, and / or which represents a syntactic unit which conforms to the rules of a particular programming language and which consists of declarations and statements or instructions required for the functions, tasks, or problem solutions described above.

[0059] Furthermore, a computer program product is disclosed that includes instructions which, when executed by the processor of the data processing unit, cause the device to perform the steps of one or all of the procedures defined above. Correspondingly, a computer-readable data carrier that stores such a computer program product is disclosed. The computer program product may be a software routine.

[0060] The above problem is also solved by a device for inspecting 3-dimensional objects, for example pouch cells, wherein each object has a substantially cushion-shaped or cuboid housing with a top and a bottom, wherein the top of the housing is composed of at least one upper surface section and a plurality of lateral surface sections which are oblique, parallel or perpendicular to the at least one upper surface section or which represent corner sections. wherein the underside of the housing is composed of at least one lower surface section and a plurality of lateral surface sections which run obliquely, parallel or perpendicular to the at least one lower surface section or which represent corner sections, with a matrix camera which generates matrix-wise captured image information for each object from light reflected from the top surface of an area lighting device in a resting state of the object to be inspected and transmits it to a data processing device, wherein the matrix-wise captured image information includes light reflected from the lateral surface sections, wherein the data processing device is set up such that the matrix-wise captured image information is further processed as a first overall matrix and that it performs the following steps: • Segmentation of the first overall matrix ◯ in a first image information section encompassing the image information of the upper surface section and ◯ into at least one second image information section, wherein every second image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the first image information section into a multitude of individual patches, • Determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on ◯ a separate determination for each patch of the multitude of patches as to whether the respective patch of the first image information section has one or more anomalies, using a correspondingly trained first NN algorithm, whereby an error is detected if an anomaly is present, and ◯ a detection of whether an error is present in at least one second image information section, and a corresponding classification of the respective second image information section using a correspondingly trained, second NN algorithm that is different from the first NN algorithm.

[0061] In one embodiment of the above device, a camera, for example a matrix camera, is provided which is configured to generate a plurality of line-by-line captured image information from reflected light of a line lighting device on linear areas of the top surface in a moving state of the object to be inspected and to transmit it to a data processing device. the data processing facility is set up to carry out the following steps: • Assembling the image information captured row by row into a second overall matrix comprising the image information of the top side of the object, • Determining the type of error of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, additionally based on the image information of the second overall matrix.

[0062] In one embodiment of the device, the data processing unit is set up to perform the following steps: • Segmentation of the second overall matrix ◯ into a third image information section comprising the image information of the upper surface section and / or ◯ into at least one fourth image information section, wherein each fourth image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the third image information section into a multitude of individual patches, • where determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, based on • a separate determination for each patch of the multitude of patches as to whether the respective patch of the third image information section has one or more anomalies, using the first NN algorithm, whereby an error is detected if an anomaly is present, and / or ◯ a detection is performed to determine whether an error is present in at least one fourth image information section, and a corresponding classification of the respective fourth image information section is performed using the second NN algorithm.

[0063] In one embodiment of the device, the data processing unit is configured to determine at least one dimension of the object and / or at least one magnitude of a detected error, taking into account the perspective and / or the optical distortion of the matrix camera.

[0064] For the above embodiments of the device, reference is made to the above explanations of the method and the device. Further embodiments and their advantages are presented here, which are also considered disclosed for the corresponding device.

[0065] Further advantages, features, and applications of the invention are described below with reference to exemplary embodiments and the figures. All features described and / or illustrated constitute the subject matter of the present invention, even independently of their compilation in the claims and their cross-references.

[0066] They show schematically: Fig. 1 An embodiment of a device used for the method according to the invention in a first perspective view from the side, Fig. 2 the device according to Fig. 1 in a second perspective view from the side, Fig. 3 the device according to Fig. 1 in a side view showing the marginal rays of lighting devices and the field of view of the matrix camera, Fig. 4 the device according to Fig. 1 in a front view with the marginal rays of a lighting device and a central ray of the field of view of the matrix camera, Fig. 5 the device according to Fig. 1 in a front view showing the marginal rays of lighting devices and the marginal rays of the matrix camera's field of view, Fig. 6 the device according to Fig. 1. The field of view of the matrix camera is shown in a top view. Fig. 7 an embodiment of a system consisting of two devices arranged one after the other according to Fig. 1 in a perspective view from the side, Fig. 8. A first example of the design of an object (pouch cell) in a perspective view from the side, Fig. 9 a second example of the design of an object (pouch cell) in a perspective view from the side and Fig. 10 an embodiment of a method according to the invention as a flowchart.

[0067] The in the Fig. The inspection device shown in Figures 1 to 6 is used for inspecting objects, e.g. in the form of pouch cells.

[0068] Two examples of pouch cells 11, 111 are in the Fig. 8 and Fig. 9 shown.

[0069] Pouch cell 11 (see Fig. 8) has a substantially cushion-shaped housing 12. The housing 12 includes a first terminal tab (Tab) 14 projecting from one short side and a second terminal tab (Tab) 15 projecting from the opposite second short side. The cushion-shaped housing 12 has a top surface with a substantially horizontal upper surface section 13. The first Tab 14 has an upper Tab surface 17 and the second Tab 15 has a corresponding upper Tab surface 18. In Fig. The corresponding lower tab surfaces of tabs 14 and 15 are not visible in Figure 8. The housing 12 is essentially cuboid in shape. On its short sides, the housing 12 has side surfaces 21 and 22, and on its long sides, side surfaces 23 and 24. The side surfaces 21, 22, 23, and 24 run approximately perpendicular to the horizontal upper surface section 13. The horizontal upper surface section 13, the upper tab surfaces 17 and 18 of tabs 14 and 15, and the side surfaces 21, 22, 23, and 24 together form the top surface of the housing 12. The underside is similarly designed and has a horizontal lower surface section, the lower tab surfaces of tabs 14 and 15, and the side surfaces 21, 22, 23, and 24. In this embodiment, the side surfaces 21, 22, 23, 24 belong to both the top and the bottom, since they can also be detected during the inspection of the top or the bottom, respectively. Fig. In section 8, the length of the housing 12 is denoted by L (excluding tabs 14 and 15) and the width by B (see dashed double arrow lines). Due to the simple design of the pouch cell 11, it is used to explain the operation of the inspection device 1 (see Fig. 1 to 6). Accordingly, the inspection device 1 can also be used for other types of objects, especially in the form of pouch cells.

[0070] Fig. Figure 9 shows a second example of a pouch cell 111, which has a substantially cushion-shaped housing 112 with a horizontal upper surface section 113. The housing 112 also includes a first terminal tab 114 and a second terminal tab 115, which are arranged side by side and project from a single short side. The first tab 114 has an upper tab surface 117 and the second tab 115 has an upper tab surface 118.

[0071] The pouch cell 111 further has side edges 121, 122, 123, 124 on the housing 112 around the horizontal upper surface section 113, which run obliquely to the horizontal upper surface section 113 and transition into it with a curve. Corners 125 are formed at the transition of one side edge to the adjacent side edge 121, 122, 123, 124. The housing 112 also has terrace sections 126, 127, 128, 129, which each adjoin the side edges 121, 122, 123, 124 and run essentially parallel to the horizontal upper surface section 113. The top of the housing 112 is formed by the horizontal upper surface section 113, the side edge 121, 122, 123, 124, the corners 125 and the terrace sections 126, 127, 128, 129.

[0072] The in the Fig. The device 1 shown in Figures 1 to 6 for inspecting the pouch cell 11 has a frame 3 on which a base plate 5 is arranged, which has a first through opening 7 and a second through opening 8 (compare Figure 1 to 6). Fig. 1, Fig. 2 and Fig. 6) possesses. From a large number of pouch cells to be inspected, the following are found in the Fig. Figures 1 to 6 show two pouch cells 11 and 31 to be inspected, which are guided past the inspection device 1 below the base plate 15, as illustrated by arrows 11a and 31a. Pouch cell 31 is another pouch cell with a structure as shown in Fig. 8 shown.

[0073] Above the base plate 5, a matrix camera 40 is mounted on the frame 3, looking down through openings 7, 8 at the pouch cells 11, 31. The pouch cells 11, 31 are arranged such that the top of the housing 12 is always facing upwards, allowing the matrix camera 40 to view the horizontal upper surface section 13 from above. The matrix camera 40 also captures the upper tab surface 17 of the first tab 14 and the upper tab surface 18 of the second tab 15. The field of view 42 of the matrix camera 40 is such that (see in particular Fig. 6) that it extends over both openings 7, 8 in such a way that both the pouch cell 11, 31 arranged under the respective opening 7, 8 in its entire length and width (viewed from above) and the deflecting mirrors 65, 67 arranged next to the pouch cell 11 are also captured.

[0074] Furthermore, a linear lighting device 51 is provided on the frame 3, which illuminates a linear area 31b of the top of the housing and the upper tab surfaces of the tabs. As shown in Fig. As shown in Figure 4, the light reflected from the top of the housing (including the upper tab surface) passes through the viewing beam 41 into the matrix camera 40 and is detected there. The matrix camera 40 thus detects the illuminated linear area 31b of the pouch cell 31 line by line, whereby during detection the pouch cell 31 is positioned transversely to the length of the opening 8 (see arrow 31A in Figure 4). Fig. 6) is in motion, i.e., is in a state of motion. A large number of images are thus generated using the matrix camera 40. Each image includes a photograph of the illuminated linear area of ​​the top of the housing (including the upper tab surfaces) of the moving pouch cell 31. The pouch cell is moved in motion by means of a motion device described in more detail below and brought to a resting state for a predetermined period, and then moved out of the inspection device from this resting state.

[0075] In addition, four surface lighting devices 52, 53, 55, 56 are provided on the frame. As can be seen from the Fig. 2, Fig. 3, Fig. 5 and Fig. As shown in Figure 6 and the marginal rays 52a and 52b or 53a, 53b of the area illumination devices 52, 53 are shown, the area illumination devices 52, 53 illuminate the top of the housing 12 (including the upper tab surface 17, 18 of the tabs 14, 15) of the pouch cell 11 over the entire length L + tab length from obliquely above, so that in particular the side opposite the respective area illumination device 52, 53 with respect to the width of the pouch cell 11 (compare in particular Figure 6) Fig. 5) recorded, which is located below the opening 7 in the base plate 5. As shown Fig. As can be seen from Figure 3, the surface illumination devices 55, 56 illuminate the top of the housing 12 (including the upper tab surfaces 17, 18 of tabs 14, 15) via the mirrors 61, 62. The light emitted by the surface illumination device 55 falls via the mirror 62 essentially onto the opposite first end of the pouch cell 11, and the light emitted by the surface illumination device 56 via the mirror 61 falls essentially onto the second end of the pouch cell 11 opposite the first end (in the longitudinal direction). This can be observed by the marginal rays 55a, 55b, 56a and 56b. The surface lighting devices 52, 53, 55, 56 also partially illuminate the side surfaces 21, 22, 23, 24, so that reflections from these side surfaces and from the horizontally extending upper surface section 13 of the housing 12 are detected by the matrix camera 40.The light reflected from the side surfaces 21, 22, 23, 24 of the surface lighting devices 52, 53, 55, 56 is detected in particular by the deflecting mirrors 65, 67, which are provided next to the pouch cell 11 arranged below the opening 7 in such a way that the side surfaces 23, 24 on the long sides of the housing 12 are viewed by means of the deflecting mirrors 65, while the deflecting mirrors 67 serve to detect the side surfaces 21, 22 on the short side of the housing 12 of the pouch cell 11.

[0076] The illumination by means of the area lighting devices 52, 53, 55, 56 is now carried out such that these are switched on sequentially, so that the pouch cell 11 is illuminated from an oblique angle above, while the other three area lighting devices are switched off. For example, the illumination is first provided by area lighting device 52, then by area lighting device 55, then by area lighting device 53, and finally by area lighting device 56. The matrix camera 40 detects the reflected light in each of the four illumination states, whereby the pouch cell 11 is in its resting state, i.e., in the same predetermined position below the opening 7 in the base plate 5, in all four illumination states.Accordingly, four images of the entire top surface of the housing 12 (including the upper tab surface 17, 18) are generated using the matrix camera 40, which capture these areas four times matrix-wise, namely once each with the area illumination device 52, area illumination device 55, area illumination device 53 and area illumination device 56 switched on, with the pouch cell 11 being in the same position each time.

[0077] During the 4 matrix-wise acquisition of the entire top surface of the housing 12 (including the upper tab surface 17, 18), the side surfaces 21, 22, 23, 24 are also acquired via the deflecting mirrors 65, 67, because the light reflected from these side surfaces 21, 22, 23, 24 reaches the matrix camera 40 via the deflecting mirrors 65, 67, since the viewing area 42 of the matrix camera 40 covers these areas, as shown in Fig. 6 shown, one.

[0078] The numerous row-wise and matrix-wise acquisitions of the pouch cell 11, 31 by the matrix camera 40 are transmitted to the data processing unit (computer) 70 after their acquisition (compare Fig. 1) transmitted. The data processing unit 70 receives the image information from the line-by-line and matrix-by-matrix acquisition of the respective pouch cell 11, 31.

[0079] The movement and resting states of pouch cells 11 and 31 are detected by a motion detection unit of the inspection device. For example, by monitoring the movement of a predefined marker on pouch cell 11 or 31, such as a barcode, or by receiving corresponding signals from the motion detection unit, motion information is generated. This information includes, in particular, details of the current movement state of each pouch cell 11 or 31. For instance, the motion detection unit can transmit information to the motion detection unit indicating that a pouch cell is ready for inspection (start signal). From this point on, the motion detection unit can continuously record the movement information of the motion detection unit and thus of the respective pouch cell (e.g., the movement speed of the motion detection unit).Alternatively or additionally, the motion detection unit receives a signal when the line-by-line acquisition of the respective pouch cell is complete. Subsequently, the motion detection unit, or a marker on the pouch cell, generates a signal and transmits it to the motion detection unit when the respective pouch cell is in its predefined resting position and remains motionless. After the matrix-by-matrix acquisition of the respective pouch cell is complete, the motion detection unit generates another signal indicating that the respective pouch cell can be removed from the inspection device. These signals also represent important motion information required for processing the image data.

[0080] This image and motion information is further processed and evaluated by the data processing unit 70. As detailed below, this process assesses the object for the presence of at least one type of defect and / or determines a quality indicator that allows for an evaluation of the quality of pouch cells 11 and 31. The image information acquired at different times from the row-by-row and matrix-by-matrix acquisition of the respective pouch cells 11 and 31, and their motion and rest states, are assigned to them. This can be done, for example, using the motion information transmitted by the motion detection unit regarding the motion and rest states of the respective pouch cells 11 and 31.In this process, the image information of the generated recordings is corrected with regard to image cropping, mirror distortion, line-by-line capture (so-called line scan correction) and the position associated with the respective state of movement or rest.

[0081] The image information from the line-by-line acquisition of the respective pouch cell 11, 31 can be generated, for example, using bright-field illumination (Reflection-Bright-Field (RBF) illumination).

[0082] For example, as explained in more detail below, the data processing unit 70 generates a quality statement (quality indicator) from the number, error type and size / dimension of detected errors.

[0083] Additional cameras 45 and 46 are arranged on the frame 3, mounted below the matrix camera. They view the top surface of the pouch cell 11 from above (see marginal rays 45a, 45b, 46a, 46b) such that they observe an upper tab surface 17 of the first tab 14 and an upper tab surface 18 of the second tab 15, as well as, if applicable, an adjacent section of the upper surface of the top surface 13. The additional cameras 45 and 46 generate higher-resolution images in the specified sections of the pouch cell 11. The image information obtained from the corresponding viewing areas is transmitted to the data processing unit 70, which generates further information regarding minor defects.

[0084] Fig. Figure 7 shows a system according to the invention for inspecting a pouch cell. The upper side of the pouch cell (e.g., pouch cell 11) is first inspected using the inspection device 1. The pouch cell (e.g., pouch cell 11) then moves to the turning device 180, which has a rotating mechanism and grippers with suction cups. This device turns the pouch cell (e.g., pouch cell 11) so that the underside is now facing up. The pouch cell (e.g., pouch cell 11) is then inspected using an inspection device 101, which is identical in construction to the inspection device 1. Specifically, its underside (which is now facing up) is inspected. The processing of the image data obtained by the inspection devices 1 and 101 with regard to the pouch cell (e.g.,The inspection of the pouch cell (e.g., pouch cell 11) and the determination, generated from this image information, of whether one or more defects of at least one defect type are present, and / or the determination of a quality indicator that allows an assessment of the object's quality, is carried out by means of the data processing unit 170, which is connected to a display 172 for showing the inspection results. The pouch cell (e.g., pouch cell 11) is transported from the first inspection device 1 to the turning device 180 and to the second inspection device 101 by means of a motion device, which includes, for example, slides that are movable on a linear unit. The respective pouch cell is attached to a slide by means of suction cups.After the matrix camera 40 and, if applicable, the other cameras 45, 46 have completed capturing the image information, and in particular after the matrix-wise acquisition of the respective pouch cell has been completed, a corresponding signal is generated by the inspection device, which is transmitted to the control unit of the motion device. The motion device is then controlled so that it conveys the respective pouch cell out of the respective inspection device 1, 101 and, if applicable, transports it to the turning device 180, in order to be turned over and then transported to the second inspection device 101.

[0085] The evaluation of the captured image information to determine the presence of an error and / or to calculate a quality indicator can be carried out, for example, as follows. The procedure is based on the flowchart described in Fig. Figure 10 illustrates this.

[0086] The starting point for the evaluation of the captured image information is the four matrix-wise captured image information 200 of the top of the object generated by illumination from different lighting directions, as well as the row-wise captured image information 201 of the top of the object using the example of the pouch cell 101.

[0087] As described above, in step 202, the perspective and / or optical distortion of the matrix camera is first compensated for in each of the four matrix-acquired image data sets. Subsequently, in step 204 (position correction), the actual position of the object as it was optically captured by the matrix camera is corrected, if necessary, i.e., rotated and / or shifted, so that the image data occupies a predetermined position of the object within the field of view of the matrix camera. In parallel, the data processing unit 70 corrects the line-by-line acquired image data in step 203 for inconsistencies in the reconstruction of the image of the top of the object from the line-by-line acquired image data, as described above. Finally, this data is also corrected with regard to its position in step 204, as described above.The combined and corrected image information, captured line by line, forms the second overall matrix.

[0088] For the four matrix-wise acquired, compensated, and corrected image information sets, the determination of a maximum image, an absorption image, and / or a topology image follows in step 206 of the object's top surface. The calculation from the four matrices containing image information is described in detail above. The maximum image is also referred to as the first composite matrix.

[0089] Now, in step 210, segmentation is performed. As explained in detail above, layout recipes are used to extract desired image information sections from the respective (corrected) image information matrix of the top surface, determined by matrix-wise or row-wise acquisition. For example, a first image information section is created in the form of the upper surface section (image information from direct capture by the camera / matrix camera) 113, a second image information section is created from the capture areas via the mirrors on the shorter side in the form of the two corner sections 125, and a further second image information section is created in the form of the four terrace sections (image information from direct capture by the camera / matrix camera) 126, 127, 128, 129.Segmentation is performed with regard to both the corrected and composited line-by-line image information and the matrix-by-matrix image data, as well as the maximum image and / or the absorption image and / or the topology image. The image information segments obtained through segmentation are then processed in parallel and finally used for an overall evaluation of the object.

[0090] The result of the segmentation in step 230 is, for example, an image information section of the upper surface in each of the maximum image, the absorption image, the topology image, and the second composite matrix. In step 232, the data processing unit 70 checks, for each of these image information sections and with respect to each pixel, whether it exceeds a predefined threshold. Such a threshold could be 240 for the image information section of the maximum image, 220 for the image information section of the topology image, 203 for the image information section of the absorption image, and 120 for the image information section of the second composite matrix. If the image information value of the respective pixel is at or above the respective threshold, then an error is detected.In step 234, further properties of the detected defect are determined, such as its size (by analyzing whether a defect was also detected in neighboring pixels) in pixels, a histogram of the image information values ​​in the area of ​​the respective defect, and / or the shape and orientation of the defect. Based on the properties of the detected defect and, if applicable, other defects found in the image information section, the defect type is then determined in step 236, whereby different images / matrices relating to the same location of the image information section can be used for this purpose.For example, the absence of a defect at the location in the absorption image, a length-to-width ratio greater than 5 in the absorption image, and a mean value of an image information histogram along the defect in the topology image greater than 200 can all be used to conclude that the defect type is "scratch." In step 238, the severity of the detected scratch defect is then determined. This determination can be based, for example, on assigning the size of the scratch to severity classes. Here, if the defect is less than 2 pixels, the scratch defect can be assigned severity level zero; if the defect is greater than or equal to 2 pixels and less than 4 pixels, it can be assigned severity level 1; if the defect is greater than or equal to 4 pixels and less than 6 pixels, it can be assigned severity level 2, and so on.

[0091] The result of the segmentation in step 240, for example, is image information segments in the form of four corner segments (e.g., 128 × 128 pixels) from the maximum image determined in step 206. These corner segments are derived, for example, from the image information generated by the mirrors 67 on the short side of the pouch. As described above, the CNN algorithm with a binary classifier described above is applied to each of these image information segments in step 242. As a result, the attribute "defective corner" (step 244) or "intact corner" (step 246) is determined for each corner image information segment and assigned to the respective corner in step 248.

[0092] The result of the segmentation in step 250, for example, are image information sections in the form of terrace segments (corresponding to terraces 126, 127, 128, 129) from the maximum image determined in step 206, where the image information was generated by direct acquisition from above using the matrix camera 40. The mask RCNN algorithm is then applied to these terrace image information sections (see above, step 252, for a more detailed description). As a result, errors of various types are detected in the terrace image information sections and assigned a bounding box (step 254). The severity of each error is then determined for these terrace image information sections, based, for example, on the identified error type, the size of the bounding box, the shape of the bounding box, etc. (step 256).

[0093] The result of the segmentation in step 260 is, for example, image information sections in the form of the upper surface section 113, such as in the maximum image and in the second overall matrix (from the row-wise acquisition). These image information sections are also subdivided into patches in step 262, whereby the resolution of each section can optionally be reduced to a predefined value (see example above). Subsequently, in step 264, the image information sections are analyzed as described above using the pre-trained CNN algorithm "Wide ResNet-50," and anomalies are detected in some patches, if necessary. In step 266, the Mahalanobis distance to the normal distribution is determined for each patch in which an anomaly was detected, and for each detected anomaly. From this, the severity of the anomaly, and thus the respective error, is determined in step 268.

[0094] The severity of the error is expressed in the form of predefined classes in all the above cases.

[0095] Subsequently, in step 270, the data processing unit 70 evaluates the overall quality of pouch cell 111 based on all defects identified in the four analysis strands, their respective defect types, and their respective severity levels. It is assessed whether pouch cell 111 meets the specified quality requirements overall. In step 280, the result of the overall evaluation is displayed at an interface of the data processing unit, optionally along with a list of the identified defects and their characteristics. For example, a pouch cell with two "Dent" defects of severity class 5 might be deemed sufficient to meet the quality requirements. Conversely, a single "Dent" defect of severity class 7 might be classified as not meeting the quality requirements.

[0096] The above procedure can also be performed analogously for the underside of the pouch cell.

[0097] As shown above, the method according to the invention allows for a simple and quick inspection of a three-dimensional object, e.g. a pouch cell, taking into account in particular the different properties of the sections of the object during the analysis.

Claims

[1] Method for inspecting 3-dimensional objects, for example pouch cells (11, 31, 111), wherein each object has a substantially cushion-shaped or cuboid housing with a top and a bottom, wherein the upper surface of the housing (12, 112) is composed of at least one upper surface section (13, 113) and a plurality of lateral surface sections (17, 18, 21, 22, 23, 24, 117, 118, 121, 122, 123, 124, 125, 126, 127, 128, 129) which are inclined, parallel or perpendicular to the at least one upper surface section (13, 113) or which form corner sections (125), wherein, for each object, matrix-wise image information is generated by means of a matrix camera from light reflected from the top surface of an area lighting device in a resting state of the object to be inspected and transmitted to a data processing device (70), wherein the matrix-wise acquired image information includes light that was reflected from the lateral surface sections, wherein the matrix-wise acquired image information is further processed as a first overall matrix by means of the data processing device, wherein the following steps are further carried out by means of the data processing device (70): • Segmentation of the first overall matrix ◯ in a first image information section encompassing the image information of the upper surface section and o in at least one second image information section, wherein every second image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the first image information section into a multitude of individual patches, • Determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on o a separate determination for each patch of the multitude of patches as to whether the respective patch of the first image information section has one or more anomalies, using a correspondingly trained first NN algorithm, whereby an error is detected if an anomaly is present, and o a detection of whether an error is present in at least one second image information section, and a corresponding classification of the respective second image information section using a correspondingly trained, second NN algorithm that is different from the first NN algorithm. [2] Method according to claim 1, characterized by , that the classification of the at least one second image information section is carried out using a classifier with two states or a classifier with at least 3 states, whereby the classifier with at least 3 states allows, for example, the assignment of different error types. [3] Method according to any one of the preceding claims, characterized by, that for each object, a multitude of line-by-line image information is generated from reflected light of a line lighting device on linear areas of the top surface in a moving state of the object to be inspected, and transmitted to a data processing device (70), wherein the following steps are further carried out by means of the data processing device (70): • Assembling the image information captured row by row into a second overall matrix comprising the image information of the top side of the object, • Determining the type of error of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, additionally based on the image information of the second overall matrix. [4] Method according to claim 3, characterized by, that the following steps are carried out using the data processing equipment (70): • Segmentation of the second overall matrix ◯ into a third image information section comprising the image information of the upper surface section and / or o in at least one fourth image information section, wherein each fourth image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the third image information section into a multitude of individual patches, • where determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, based on o a separate determination for each patch of the multitude of patches as to whether the respective patch of the third image information section has one or more anomalies, using the first NN algorithm, whereby an error is detected if an anomaly is present, and / or The process involves detecting whether an error is present in at least one fourth image information section, and classifying the respective fourth image information section accordingly using the second NN algorithm. [5] Method according to any one of the preceding claims, characterized by, that for each object at least n (n ≥ 2) images of the entire top surface of matrix-wise acquired image information are acquired by a temporally successive acquisition of reflected light and transmitted to the data processing device (70), wherein the matrix-wise acquired image information is further processed as n first total matrices by means of the data processing device, wherein the reflected light is generated by a temporally separated illumination of the entire top surface of the housing of the object in the rest state of the object to be inspected from n different directions from obliquely above. [6] Method according to claim 5, characterized by , that the following steps are carried out using the data processing equipment (70): • Segmentation of the first n total matrices ◯ in n fifth image information sections comprising the image information of the upper surface section of each of the n first total matrices and / or o in n sixth and, if necessary, further image information sections of each of the n first total matrices, wherein each sixth and, if necessary, further image information section contains the image information of at least one given section of the lateral surface sections and / or at least one given corner section, • Determining one maximum image and / or one absorption image and / or one topology image from the image information of the fifth image information section and / or the sixth image information section and / or any further image information sections, • Determining, analyzing, and characterizing errors in the maximum image and / or the absorption image and / or the topology image of the fifth image information section and / or the sixth image information section and / or, if applicable, the further image information sections, • wherein the determination of a fault type of a detected fault and / or a severity of a detected fault and / or the determination of a quality indicator which allows an assessment of the quality of the object is based on the result of the analysis and / or characterization of the faults identified in each case. [7] Method according to any one of the preceding claims, characterized by , that to detect an anomaly in a patch, a Patch Distribution Modeling Framework for Anomaly Detection is applied, whereby the degree of anomaly is determined using the Mahalanobis distance with respect to a normal distribution expected in the respective patch. [8] Method according to any one of the preceding claims, characterized by , that the location of the anomaly in the patch is determined and used to locate a possible error in / on the object. [9] Method according to any one of the preceding claims, characterized by , that the data processing device determines at least one dimension of the object and / or at least one magnitude of a detected error, taking into account the perspective and / or the optical distortion of the matrix camera. [10] Method according to any one of the preceding claims, characterized by , that a position correction is carried out using the data processing device via predefined fixed points of the object. [11] Method according to any one of the preceding claims, characterized by, that before the first image information section and / or the third image information section is decomposed into a multitude of individual patches, the resolution of the first image information section is reduced. [12] Device (1, 101) for inspecting 3-dimensional objects, for example pouch cells (11, 31, 111), wherein each object has a substantially cushion-shaped or cuboid housing with a top and a bottom, wherein the upper surface of the housing (12, 112) is composed of at least one upper surface section (13, 113) and a plurality of lateral surface sections (17, 18, 21, 22, 23, 24, 117, 118, 121, 122, 123, 124, 125, 126, 127, 128, 129) which are inclined, parallel or perpendicular to the at least one upper surface section (13, 113) or which form corner sections (125), with a matrix camera which generates matrix-wise image information for each object from light reflected from the top surface of an area lighting device in a rest state of the object to be inspected and transmits it to a data processing device (70), wherein the matrix-wise image information includes light reflected from the lateral surface sections, wherein the data processing device (70) is set up such that the image information acquired matrix-wise is further processed as a first overall matrix and that it performs the following steps: • Segmentation of the first overall matrix o in a first image information section encompassing the image information of the upper surface section and o in at least one second image information section, wherein every second image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the first image information section into a multitude of individual patches, • Determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the object's quality, based on o a separate determination for each patch of the multitude of patches as to whether the respective patch of the first image information section has one or more anomalies, using a correspondingly trained first NN algorithm, whereby an error is detected if an anomaly is present, and o a detection of whether an error is present in at least one second image information section, and a corresponding classification of the respective second image information section using a correspondingly trained, second NN algorithm that is different from the first NN algorithm. [13] Device according to claim 12, wherein a camera, for example a matrix camera, is provided which is configured to generate a plurality of line-by-line captured image information from reflected light of a line lighting device on line-shaped areas of the top surface in a moving state of the object to be inspected and to transmit it to a data processing device (70), wherein the data processing device is configured to perform the following steps: • Assembling the image information captured row by row into a second overall matrix comprising the image information of the top side of the object, • Determining the type of error of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, additionally based on the image information of the second overall matrix. [14] Device according to claim 13, wherein the data processing device is configured to perform the following steps: • Segmentation of the second overall matrix ◯ into a third image information section comprising the image information of the upper surface section and / or o in at least one fourth image information section, wherein each fourth image information section includes the image information of at least one predefined section of the lateral surface sections and / or at least one predefined corner section, • Decomposition of the third image information section into a multitude of individual patches, • where determining the type of a detected error and / or the severity of a detected error and / or determining a quality indicator that allows an assessment of the quality of the object, based on o a separate determination for each patch of the multitude of patches as to whether the respective patch of the third image information section has one or more anomalies, using the first NN algorithm, whereby an error is detected if an anomaly is present, and / or The process involves detecting whether an error is present in at least one fourth image information section, and classifying the respective fourth image information section accordingly using the second NN algorithm. [15] Device according to one of claims 12 to 14, wherein the data processing device is configured to determine at least one dimension of the object and / or at least one magnitude of a determined error after taking into account the perspective and / or the optical distortion of the matrix camera.

Images