Computer-implemented method, computing device, and computer program product for processing volume scan data and analyzing performance of structural elements

Abstract

The present invention relates to a computer-implemented method, a computing device, and a computer program for processing computed tomography (CT) scan data and analyzing the performance of a structural element represented by the computed tomography data. A computer-implemented method for processing volume scan data includes obtaining (110) volume scan data of a structural element, the structural element including multiple instances of two or more types of components. The volume scan data is represented by a plurality of voxels. The method includes assigning (120) the plurality of voxels to one of two or more different component types; identifying (130) for each component type a voxel that is part of a surface of a component instance. The method includes extracting (140) a surface of the component instance using the voxels that are part of a surface of a component of the corresponding type. The method includes determining (170) information about the structural integrity of the structural element based on the extracted surface.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method, a computing device and a computer program product for processing volume scan data and analyzing properties of structural elements represented by the computed tomography data. Background Art

[0002] Analyzing failures that occur during the manufacture of components, such as automotive parts, is an area of research and development. Because many components are housed in closed enclosures, manual analysis of failures in such components may only be performed at the expense of destroying the enclosure, which may result in additional failures that did not exist before the enclosure was opened, and may not be suitable for spot checks performed during manufacturing. Therefore, in some cases, computed tomography (CT) scanning is used for failure analysis, or more generally, for on-site testing during manufacturing.

[0003] A CT scan typically includes a plurality of voxels (i.e., three-dimensional pixels). The brightness of a voxel indicates the absorption rate of the X-rays used for the CT scan. Different components within a housing typically exhibit different absorption rates, so they can be distinguished in a CT scan. Based on the CT scan, the internal components of a component can be analyzed. For example, U.S. patent applications US 2017 / 0292922 A1 and US 2011 / 0182495 A1, as well as Chinese patent application CN109658396A, illustrate the concept of using CT scans to analyze faults in components.

[0004] There may be a desire for an improved concept for analyzing volume scan data regarding faults in components.

[0005] This expectation is solved by the subject matter of the independent claims. Summary of the Invention

[0006] Embodiments of the present disclosure are based on the discovery that analysis of volume scan data, such as computed tomography (CT) scan data, functional magnetic resonance imaging (FMRI) scan data, or positron emission tomography (PET) scan data, can be improved by extracting surfaces of individual components of a structural element and analyzing the interrelationships of the surfaces to determine the structural integrity of the structural element. Thus, voxels of the volume scan data that are part of the bulk of the component (i.e., not at the surface) can be discarded from the analysis. The interrelationships of the surfaces can be analyzed to determine distances between surfaces, alignment of surfaces, etc. These structural integrity features can be important for analyzing potential failures in structural elements, such as batteries, where overlap of cathodes and anodes can be avoided.

[0007] Various embodiments of the present disclosure relate to a computer-implemented method for processing volume scan data. The method includes obtaining volume scan data of a structural element. The structural element includes multiple instances of two or more component types. The volume scan data is represented by a plurality of voxels. The method includes assigning the plurality of voxels to one of two or more different component types. The method includes identifying, for each component type, voxels that are part of a surface of a component instance (in this case, no instantiation occurs). The method includes extracting the surface of the component instance using the voxels that are part of the surface of the corresponding type of component. The method includes determining information about the structural integrity of the structural element based on the extracted surface. Generally speaking, a surface exhibiting irregularities may be detrimental to the structural integrity of the structural element and, therefore, may be analyzed to determine the structural integrity of the structural element.

[0008] For example, the structural element may be a battery. The two or more different component types may include a cathode and an anode of the battery. Generally speaking, in batteries, overlap of the cathode and anode can be avoided, and the surfaces of the cathode and anode can be analyzed to detect such potential overlap.

[0009] In general, a voxel can be part of a surface of a component instance or part of the bulk of the component instance. The method can include discarding voxels that are part of the bulk of the component instance. The bulk voxels can be discarded to reduce the processing power required to analyze the structural integrity of the structural element.

[0010] In various embodiments, the method includes segmenting component instances based on the extracted surfaces. Information regarding the structural integrity of the structural element can be determined based on the extracted surfaces of the component instances. For example, returning to the example of a battery as the structural element, each cathode layer and anode layer can be considered a separate instance of the corresponding component. The same principle applies to cylindrical batteries.

[0011] For example, information regarding structural integrity can be determined based on at least one of a distance between surfaces, an alignment of surfaces, a uniformity of distance along the extent of two surfaces, an intersection of surfaces in extracted surfaces, and a difference between the extracted surfaces and a three-dimensional model of the structural element. The distance and / or intersection of the surfaces can reveal structural faults within the structural element. Differences between the three-dimensional model (e.g., a CAD drawing) and the extracted surfaces can also indicate potential faults.

[0012] In some embodiments, extracting the surface of a component instance includes performing local plane fitting on voxels that are part of the surface of a component of the corresponding type. Local plane fitting can be used to generate a continuous (and smooth) surface from the voxels that are part of the surface of the corresponding component. For example, local plane fitting can be used as a plausibility check for the voxels that are part of the surface.

[0013] In various embodiments, a machine learning model suitable for image segmentation is used to assign voxels to one of two or more different component types. For example, a machine learning model based on a U-Net architecture can be used to assign voxels to one of two or more different component types. In general, various suitable machine learning frameworks exist that can be used to perform corresponding image segmentation.

[0014] In general, it may also be useful to determine the orientation of each surface (e.g., whether the surface is the top or bottom surface of a component instance). The method may include determining the orientation of the surface of the component instance and classifying the surface of the component instance (e.g., top or bottom) based on the orientation of the corresponding surface. Information regarding structural integrity can be determined based on the classification of each surface. For example, the classification can be used to apply a high-level rule set to determine structural integrity, such as, "the top surface of the anode may not intersect the bottom surface of the cathode."

[0015] For example, the orientation of a surface of a component instance can be determined based on the average positioning of the body of the corresponding component instance (e.g., without requiring instantiation at that point). For example, if the average positioning of the body is below the surface, the surface can be the top surface, and if the average positioning of the body is above the surface, the surface can be the bottom surface. In some embodiments, only the body of the instance associated with the surface can be considered. However, in some embodiments, the entire body of components of the same type can be considered.

[0016] In various embodiments, the method includes correlating the performance of the structural component with information regarding the structural integrity of the component. This correlation can then be used to determine an estimated performance of the structural element based on its volume scan data.

[0017] Various embodiments of the present disclosure relate to a computer-implemented method for determining a correlation between the structural integrity of a structural element and the performance of the structural element. The method includes obtaining information regarding the structural integrity of a plurality of structural elements, generating the information regarding the structural integrity using the method described above, obtaining information regarding the performance of the plurality of structural elements, and determining a correlation between the structural integrity of the plurality of structural elements and the performance of the plurality of structural elements. The correlation can be used to determine an estimated performance of the structural element based on volume scan data thereof.

[0018] Various embodiments of the present disclosure are directed to a computing device comprising an interface for exchanging information and processing circuitry configured to carry out a method for processing volume scan data.

[0019] Various embodiments of the present disclosure relate to a computing device comprising an interface for exchanging information and a processing circuit configured to perform a method for determining correlation. For example, the computing device may correspond to the computing device mentioned above.

[0020] Various embodiments of the present disclosure relate to a computer program having a program code for carrying out the method for processing volume scan data and / or the method for determining a correlation when the computer program is executed on a computer, a processor or a programmable hardware component. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Some further features or aspects will be described using the following non-limiting embodiments of apparatus or methods or computer programs or computer program products, by way of example only and with reference to the accompanying drawings, in which:

[0022] Figure 1a A flow chart illustrating an embodiment of a computer-implemented method for processing volume scan data;

[0023] Figure 1b A block diagram illustrating an embodiment of a computing device for processing volume scan data is shown;

[0024] Figure 2a A flow chart illustrating an embodiment of a computer-implemented method for determining relevance;

[0025] Figure 2b A block diagram illustrating an embodiment of a computing device for determining correlations is shown;

[0026] Figure 3 An overview of the overall workflow for analyzing volume scan data is shown;

[0027] Figure 4 An overview of the processing steps for calculating KPIs in an automated manner from computed tomography scans is shown;

[0028] Figure 5a The figure shows the fitting of the local plane;

[0029] Figure 5b illustrates determination of surface orientation; and

[0030] Figure 6 Illustration showing detection of overlap between bodies of two components. DETAILED DESCRIPTION

[0031] Various example embodiments will now be described more fully with reference to the accompanying drawings, in which some example embodiments are illustrated. In the drawings, the thickness of lines, layers, or regions may be exaggerated for clarity. Optional components may be illustrated using broken lines, dashed lines, or dot-dash lines.

[0032] Therefore, while the example embodiments are capable of various modifications and alternative forms, embodiments thereof are shown by way of example in the figures and will be described in detail herein. However, it should be understood that there is no intention to limit the example embodiments to the particular forms disclosed, but on the contrary, the example embodiments are intended to cover all modifications, equivalents, and alternatives falling within the scope of the invention. Throughout the description of the drawings, like numbers refer to like or similar elements.

[0033] As used herein, the term "or" refers to a non-exclusive or, unless otherwise indicated (e.g., "or otherwise" or "or instead"). Further, as used herein, unless otherwise indicated, words used to describe relationships between elements should be broadly interpreted to include direct relationships or the presence of intermediate elements. For example, when an element is referred to as being "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or there can be intermediate elements. In contrast, when an element is referred to as being "directly connected" or "directly coupled" to another element, there are no intermediate elements. Similarly, words such as "between," "adjacent," etc. should be interpreted in a similar manner.

[0034] The terms used herein are for the purpose of describing particular embodiments only and are not intended to limit the example embodiments. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should be further understood that the terms "comprises," "includes," "has," or "having," when used herein, specify the presence of stated features, integers, steps, operations, elements, or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof.

[0035] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It should be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless so explicitly defined herein.

[0036] Figure 1aA flow chart illustrating an embodiment of a computer-implemented method for processing volume scan data is shown. The method includes obtaining 110 volume scan data of a structural element. The structural element includes multiple instances of two or more types of components. The volume scan data is represented by a plurality of voxels. The method includes assigning 120 the plurality of voxels to one of two or more different component types. The method includes identifying 130, for each component type, voxels that are part of a surface of a component instance. The method includes extracting 140 surfaces of the component instances using the voxels that are part of a surface of a component of the corresponding type. The method includes determining 170 information regarding the structural integrity of the structural element based on the extracted surfaces.

[0037] Figure 1b A block diagram of an embodiment of a corresponding computing device 10 for processing volume scan data is shown. The computing device 10 includes an interface 12 for exchanging information (e.g., volume scan data and / or information about structural integrity), and a processor coupled to the interface 12 and configured to perform Figure 1a Generally speaking, the functionality of the computing device may be provided by the processing circuitry, for example in conjunction with the interface 12 (for exchanging data) and one or more storage devices (not shown) for storing data (e.g., volume scan data, information about structural integrity, and / or intermediate results).

[0038] The following description involves Figure 1a A computer-implemented method, and Figure 1b corresponding computing devices and computer programs.

[0039] Various aspects of the present disclosure relate to computer-implemented methods, computing devices, and computer programs for processing volumetric scan data (i.e., three-dimensional scan data, such as computed tomography (CT) scan data, functional magnetic resonance imaging (FMRI) scan data, or positron emission tomography scan data) of a structural element. As mentioned above, the proposed concepts can be used for fault analysis and / or spot checks, and in particular for determining correlations between the volumetric scan data of a structural element and the performance of the structural element. In this context, the structural element can be an object, such as a component or assembly for use in a machine, such as a component or assembly for use in a vehicle. In the context of the present disclosure, the present invention is discussed with respect to a battery. In other words, the structural element can be a battery. However, the same concepts also apply to other structural components, such as multi-layer sensors, or any structural element comprising a component having surfaces arranged in a predefined geometry. For example, the structural element can be an electric motor, a hydraulic damper, or a fuel cell, the former two of which have a cylindrical configuration.

[0040] The method includes obtaining 110 volume scan data of a structural element. For example, the volume scan data may be obtained via an interface or from a storage device. For example, the volume scan data may include output from a CT device (i.e., a CT scanner), output from an fMRI scanner, or output from a PET scanner. Generally speaking, the volume scan data may include scan data from a single scan of the structural element. However, in some cases, multiple scans may be performed.

[0041] A structural element includes multiple instances of two or more types of components. In the context of this disclosure, a distinction is made between "types" and "instances" of components. For example, a battery includes an anode and a cathode, which are types of battery components. In other words, the two or more different component types may include a cathode and an anode of a battery. Within each component, a structural element may include one or more instances. For example, a structural element may include multiple cathodes (i.e., multiple instances of a cathode) and multiple anodes (i.e., multiple instances of an anode).

[0042] Generally speaking, volume scan data is represented by a plurality of voxels. In the context of volume scan data, a voxel is an image point with coordinates in three dimensions (on a three-dimensional grid). In CT scan data, each voxel indicates the brightness of the scanned structural element at that three-dimensional coordinate. This brightness, in turn, indicates the absorption of the x-rays used to scan the structural element. In other words, each voxel is characterized by a brightness value corresponding to the absorption (of the x-rays).

[0043] The method includes assigning 120 a plurality of voxels to one of two or more different component types. By assigning the plurality of voxels to the different component types, a first step is to segment the components visible in the volume scan data. Generally speaking, a machine learning model suitable for image segmentation can be used to assign the voxels to one of the two or more different component types. In other words, the volume scan data or a derivative thereof can be provided at the input of the machine learning model, and the assignment between the voxels and the component type can be obtained at the output of the machine learning model. There are various frameworks suitable for image segmentation. Generally speaking, the machine learning model can be suitable for image segmentation in 2D or in 3D. One framework that has proven to be suitable is the so-called U-Net framework, which is primarily used for image segmentation (of 2D image data) in medical applications. Therefore, a machine learning model based on the U-Net architecture can be used to assign voxels to one of the two or more different component types. More information can be found in the article "U-Net: Convolutional Networks for Biomedical Image Segmentation" by Ronneberger et al. (2015). Alternatively, a machine learning model for segmenting volume image data, such as a model used in the biomedical field, can be used. Different types of machine learning models for segmentation may be suitable, for example, as long as they assign each voxel to one of multiple types / categories.

[0044] The method comprises: identifying 130 for each component type voxels that are part of the surface of the component instance. Once the voxels are assigned to one of the component types, the surface voxels can be identified. In general, a voxel is part of the surface of the component instance, or is part of the body of the component instance. The identification of surface voxels can be done by comparing, for a given voxel, whether adjacent voxels belong to the same component type. If all adjacent voxels belong to the same component type, the voxel can be identified as a body voxel, if not, for example, if only five out of six adjacent voxels belong to the same component type, the voxel can be identified as a surface voxel. The method may comprise: discarding 135 voxels that are part of the body of the component instance. For example, in the following, only surface voxels and surfaces created by surface voxels may be considered.

[0045] The method comprises extracting 140 the surface of a component instance using voxels that are part of the surface of a component of the corresponding type. In other words, the surface of the component can be abstracted from individual pixels and transformed into a continuous plane for subsequent processing. In general, surface extraction can include one or more subtasks. For example, assigning voxels to different component types and identifying voxels that are part of the surface can be considered as subtasks of surface extraction. Additionally, extracting 140 the surface of a component instance can include performing 145 local plane fitting on the voxels that are part of the surface of the component of the corresponding type. In general, the local plane can be as close to the local surface as possible. This can be done by computing the local rotation tensor of all surface vectors and identifying the eigenvectors and the minimum eigenvalue using the plane normal. In combination Figure 5a An example of this process is given. Additionally, the orientation of the surface may be determined. In other words, the method may comprise determining 150 the orientation of the surface of the component instance and classifying 155 the surface of the component instance based on the orientation of the respective surface (relative to various coordinate systems, e.g. relative to the world coordinate system, or perpendicular to a radial direction), e.g. by classifying the surface into "top surface", "bottom surface" and "other surface / side surface". For example, the orientation of the surface may be determined by considering the positioning of one or more surface voxels relative to the body voxels. In other words, the orientation of the surface of the component instance is determined based on the average positioning of the body of the respective instance of the component. If the body voxels are mainly located below the surface, the surface may be the top surface, and if the body voxels are mainly located above the surface, the surface may be the bottom surface. If significant body parts are found below and above the respective voxels, the surface may be the side surface. Thus, based on the classification of the respective surfaces, information about the structural integrity may be determined, e.g. by determining the distance between the bottom surface of a first component and the top surface of a second adjacent component.

[0046] In various embodiments, the method further includes segmenting 160 component instances based on the extracted surfaces. For example, segmenting 160 component instances may include identifying consecutive instances of a component formed by multiple voxels of the same type. For example, the orientation of the surface may be used. For example, a component may be formed by all voxels of the same type that are directly adjacent to another voxel of the same type. Similarly, information regarding the structural integrity of the structural element may be determined based on the extracted surfaces of the component instances.

[0047] Once the surfaces of the component have been extracted, and optionally, instances of the component have been segmented, the resulting surfaces can be used to determine information regarding the structural integrity of the structural element. In other words, the method includes determining 170 information regarding the structural integrity of the structural element based on the extracted surfaces. Generally speaking, the information regarding the structural integrity can indicate the conformity of the structural element to an idealized version of the structural element, such as the conformity of the surface of the component. For example, an idealized version of the structural element (e.g., a three-dimensional model of the structural element) can be compared to the extracted surfaces to determine information regarding the structural identity. Thus, the information regarding the structural integrity can indicate deviations of the extracted surfaces from the idealized version of the structural element.

[0048] If no such model is available, or in addition to a model, the interrelationships of the surfaces can be analyzed to determine information about the structural integrity. For example, determining the information about the structural integrity can include determining the distance between the extracted surfaces. Thus, the information about the structural integrity can be determined based on at least one of the distance between the surfaces (of the extracted surfaces), the alignment of the surfaces (of the extracted surfaces), the uniformity of the distance (of the extracted surfaces) along the extent of the two surfaces, and the intersection of the extracted surfaces.

[0049] In some embodiments, this information regarding structural integrity can be used to determine a correlation between the information regarding structural integrity and the performance of the corresponding structural element. For example, in addition to performing a CT scan-based analysis of the structural elements, each structural element can be tested. The performance determined through testing can be correlated with the structural identity information determined based on the volume scan data. In other words, the method can include correlating 180 the performance of the structural assembly with the information regarding the structural integrity of the assembly. Thus, the method can include obtaining information regarding the performance of the structural element. For example, the performance information can relate to the structural performance of the structural element, such as its flexibility, or to the functionality of the structural element, such as whether the structural element is functional or the performance of a key performance indicator (such as an electrical characteristic variable) of the structural element. Both the determination of the information regarding structural integrity and the correlation of the information regarding structural integrity with the performance of the structural element can be repeated using additional volume scan data sets, the resulting information regarding structural integrity, and the corresponding performance to generate correlations based on multiple structural elements (of the same model). This correlation can be applied based on the information regarding the structural integrity of each element, and by determining the information regarding structural integrity from the volume scan data representing the structural element, key performance indicators of the structural element can be determined. For example, statistical methods can be used to determine the correlation. Alternatively, supervised training can be used to train the machine learning model, where information about the structural integrity is used as training input and corresponding information about the performance is used as the desired output of the machine learning model training. Figure 2a and Figure 2b , the same correlation determination is described separately from the determination of information about structural integrity.

[0050] The interface 12 may correspond to one or more inputs and / or outputs for receiving and / or transmitting information, which may be in the form of digital (bit) values according to a specified code within a module, between modules, or between modules of different entities. For example, the interface 12 may include interface circuitry configured to receive and / or transmit information.

[0051] In an embodiment, the processing circuit 14 may be implemented using one or more processing units, one or more processing devices, or any other means for performing processing (such as a processor, a computer, or a programmable hardware component), which may operate in conjunction with correspondingly adapted software. In other words, the described functions of the processing circuit 14 may also be implemented in software, which is then executed on one or more programmable hardware components. Such hardware components may include general-purpose processors, digital signal processors (DSPs), microcontrollers, and the like.

[0052] In at least some embodiments, computing device 10 may include one or more storage devices, for example, at least one element of the group of computer-readable storage media, such as magnetic or optical storage media, for example, a hard drive, flash memory, a floppy disk, a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electronically erasable programmable read-only memory (EEPROM), or a network storage device.

[0053] Further details and aspects of computer-implemented methods, computing devices, and computer programs for processing volume scan data may be found in conjunction with the proposed concepts or one or more examples described above or below (e.g., Figures 2a to 6 The computer-implemented method, computing device and computer program for processing volume scan data may include one or more additional optional features corresponding to one or more aspects of the proposed concept or one or more examples described above or below.

[0054] Figure 2a A flow chart illustrating an embodiment of a computer-implemented method for determining a correlation between the structural integrity of a structural element and a performance of the structural element is shown. The method includes obtaining 210 information about the structural integrity of a plurality of structural elements. Figure 1a Methods, Figure 1b The method includes obtaining 220 information about the performance of a plurality of structural elements. The method includes determining 230 a correlation between the structural integrity of the plurality of structural elements and the performance of the plurality of structural elements.

[0055] Figure 2b A block diagram of an embodiment of a corresponding computing device 20 for determining correlation is shown. The computing device 20 comprises an interface 22 for exchanging information, such as information about structural integrity and / or information about performance; and a processor coupled to the interface 24 and configured to perform Figure 2a In general, the functionality of the computing device may be provided by the processing circuitry, for example, in combination with an interface 22 (for exchanging data) and one or more storage devices (not shown) for storing data (e.g., information about structural integrity, information about performance, dependencies, and / or intermediate results).

[0056] As combined Figure 1a and 1b As mentioned, the determination of the correlation can be decoupled from the generation of information about the structural integrity. Figure 1aThe method generates information about the structural identity, or obtains 210 information about the structural integrity of a plurality of structural elements by obtaining the information from another computing device. The method includes obtaining 220 information about the performance of the plurality of structural elements, for example, from a device that has performed structural or functional testing of the plurality of structural devices or from a database. The method includes determining 230 a correlation between the structural integrity of the plurality of structural elements and the performance of the plurality of structural elements, for example, similar to that already combined Figure 1a Determination of the introduced correlation. Thus, the correlation can be suitable for determining key performance indicators of a structural element based on information about the structural integrity of the corresponding element.

[0057] The interface 22 may correspond to one or more inputs and / or outputs for receiving and / or transmitting information, which may be in the form of digital (bit) values according to a specified code within a module, between modules, or between modules of different entities. For example, the interface 22 may include interface circuitry configured to receive and / or transmit information.

[0058] In an embodiment, the processing circuit 24 may be implemented using one or more processing units, one or more processing devices, or any other means for performing processing, such as a processor, a computer, or a programmable hardware component operable with corresponding adapted software. In other words, the described functions of the processing circuit 24 may also be implemented in software, which is then executed on one or more programmable hardware components. Such hardware components may include a general-purpose processor, a digital signal processor (DSP), a microcontroller, and the like.

[0059] In at least some embodiments, the computing device 20 may include one or more storage devices, for example, at least one element of the group of computer-readable storage media, such as magnetic or optical storage media, for example, a hard drive, a flash memory, a floppy disk, a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electronically erasable programmable read-only memory (EEPROM), or a network storage device.

[0060] Further details and aspects of computer-implemented methods, computing devices, and computer programs for determining relevance may be found in conjunction with the proposed concepts or one or more examples described above or below (e.g., Figures 1a to 1b 、 Figures 3 to 6 ) is mentioned. The computer-implemented method, computing device and computer program for determining relevance may include one or more additional optional features corresponding to one or more aspects of the proposed concept or one or more examples described above or below.

[0061] Various embodiments of the present disclosure relate to robust instance segmentation for automatically extracting key performance indicators from computed tomography scans of structural elements such as batteries using computer vision.

[0062] The production of batteries (e.g., for electric vehicles) is becoming increasingly important. However, in many battery production plants, the relationship between production quality and product performance may not be known. Various embodiments of the present disclosure use a computed tomography system to quantify product performance. While the general concept of using CT scanning for quality control in production systems is known, these known methods may use different techniques than the method outlined below. The described method is generally applicable to any manufacturing process where non-destructive analysis is required. The proposed method assumes that the product has certain symmetry (e.g., layered or rounded). However, the method can also be applied to structural elements other than batteries.

[0063] The concept described is part of an overall workflow that attempts to correlate the performance of manufactured parts with measurements collected during production. Using the insights gained from analyzing how performance changes due to production measurements, overall production can be optimized. Figure 3 An overview of the overall workflow for analyzing volume scan data is shown. The overall workflow may include production 310 of a structural element, generation of a CT scan 320 of the structural element, determination of the performance 330 of the structural element, and determination of automatic KPIs (key performance indicators). The described concepts may be used in the block "Automatic KPIs" 340.

[0064] The motivational application is a battery pack, but the described features can also be applied to other manufacturing processes. In the following, an analysis process is described that is improved or optimized to extract information about production quality from computed tomography scans of components (i.e., structural elements) such as battery packs during production.

[0065] The basis for the analysis is a computed tomography scan, ie a volume image in which each voxel is characterized by a brightness value corresponding to the absorption of x-rays.

[0066] Figure 4 An overview of the processing steps for automatically calculating KPIs from a computed tomography scan 410 is shown. The analysis is broken down into the following steps. Although the term "step" is used below, the steps can be performed in a different order and one or more steps can be omitted.

[0067] 1. Component segmentation by type. For a battery, the relevant components are the battery cathode 420 and anode 425. This step assigns each voxel to a component type. Note that at this step, all voxels can be independent and there may be no information about which voxels form the entire object (e.g., the anode sheet). Segmentation can be performed using a neural network-based image segmentation model. In an exemplary implementation, a so-called U-Net architecture is used, which was originally developed for image segmentation in the medical field.

[0068] 2. Surface extraction and classification 430. Instead of analyzing voxels directly, only surface information can be used to improve robustness. In this step, the surface of each class can be extracted and each voxel can be assigned to a surface type, for example, the top or bottom of the anode.

[0069] 3. Instance segmentation 440. In this step, it can be assumed that all connected voxels of a single type are part of the same object, eg, the top layer of a single anode.

[0070] 4. Automatic KPI calculation 450. Based on the detected objects, key performance indicators that quantify production quality can be calculated (i.e., structural integrity can be determined). In the case of batteries, this could be the alignment of the anode sheets, or a comparison with the CAD file on which the production is based.

[0071] In this disclosure, the focus is on step 2. The second step can be further divided into the following sub-steps:

[0072] 2a: Surface extraction. Each voxel can be classified as a bulk voxel and a surface voxel. Any voxel whose total number of direct neighboring voxels (6 in 3D) are of the same type can be classified as a bulk voxel, and all other voxels can be classified as a surface voxel. For example, an anode voxel with 4 direct neighboring voxels that are also anodes can be classified as a surface voxel.

[0073] 2b: Fit a local plane for each surface voxel. Around each surface voxel, a local plane can be fitted that tries to fit as close to the local surface as possible. This can be done by computing the local rotation tensor (for all surface voxels)

[0074]

[0075] And use the plane normal to identify the eigenvector with the minimum eigenvalue

[0076] Gv=λ min v

[0077] However, this vector is not oriented, as it can point in any direction. The squared magnitude of the eigenvector projected onto the principal axis can be denoted as the local plane parameter. The local plane parameter can be defined as:

[0078]

[0079] See also Figure 5a , which illustrates the fitting of a local plane. Figure 5a Surface 510, ellipsoid of inertia 520, and principal axes 530 are shown.

[0080] 2c: Orientation of the plane. The correct sign of the normal can be calculated by computing the average positioning of the body around the surface voxels. For example, if most of the body is below the surface voxels, then the local surface is the top of the object. For example, the following formula can be used to calculate the average body positioning:

[0081]

[0082] See also Figure 5b , where the determination of the surface orientation is shown. Figure 5b Shown are a surface 540, a body 550, a neighborhood 560, a reference point 570 (ie, the voxel being processed), and a vector 580 toward the local centroid.

[0083] 2D: Surface types are classified based on plane orientation, voxel location, and voxel component type. Each surface voxel can then be classified by thresholding the local plane parameters and the average body location. For example, the top of the anode piece would be p z >0.5 and n z > 0. This approach can be further extended to non-hierarchical geometries. For example, for cylindrical geometries, by using the radial vector as the projection target for the local plane parameter, each voxel can be classified as being orthogonal to or perpendicular to a perfectly aligned cylinder. For example, a structural element can be a structural element with a cylindrical configuration, such as a cylindrical battery, an electric motor, or a hydraulic damper.

[0084] The described method is applicable to, but not limited to, the analysis of volumetric scan data of batteries and explicitly takes into account the layered geometry of the battery pack. Consequently, additional robustness can be observed even for imperfect CT scans. Particularly at low resolutions, other instance segmentation methods can lead to undesirable artifacts due to overlapping slices. Here, any such cases are explicitly prohibited, and the corresponding voxels can be filtered out. Figure 6 Illustration showing detection of overlap between bodies of two components. Figure 6 Shown are a bottom surface 610 of one component, a top surface 620 of the other component, and an overlap 630 between the bodies of the two components. Figure 6Can be used as an illustration of how detecting objects using surfaces instead of volumes increases robustness. While bodies can have overlap, surfaces never overlap by construction.

[0085] As already mentioned, in embodiments, the corresponding methods can be implemented as a computer program or code that can be executed on corresponding hardware. Therefore, another embodiment is a computer program having a program code that, when executed on a computer, a processor, or a programmable hardware component, is configured to perform at least one of the methods described above. Another embodiment is a computer-readable storage medium storing instructions that, when executed by a computer, a processor, or a programmable hardware component, cause the computer to perform one of the methods described herein.

[0086] Those skilled in the art will readily appreciate that the step of various said methods can be implemented by programmed computers, for example, can determine or calculate the location of time slots.Herein, some embodiments are also intended to cover program storage devices, for example, digital data storage medium, it is machine or computer readable, and the machine executable or computer executable program of instruction are encoded, wherein said instruction is implemented some or all steps of method described herein.Program storage devices can be for example digital memory, magnetic storage medium such as disk and tape, hard disk drive or optically readable digital data storage medium.Embodiment is also intended to cover the computer programmed to implement the described steps of method described herein, or is programmed to implement (field) programmable logic array ((F) PLA) or (field) programmable gate array ((F) PGA) of the described steps of method described below.

[0087] The description and drawings illustrate only the principles of the present invention. It will be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the present invention and are included within the spirit and scope of the present invention. Furthermore, all examples cited herein are primarily and explicitly intended to be used for teaching purposes only, to help the reader understand the principles of the present invention and the concepts contributed by the inventor(s) to advance the art, and are to be interpreted as not being limited to such specifically cited examples and conditions. In addition, all statements of the principles, aspects, and embodiments of the present invention, as well as their specific examples, are intended to encompass their equivalents. When provided by a processor, the functionality may be provided by a single dedicated processor, by a single shared processor, or by multiple individual processors, some of which may be shared. In addition, the explicit use of the term "processor" or "controller" should not be interpreted as referring exclusively to hardware capable of executing software, but may implicitly include, but not limited to, digital signal processor (DSP) hardware, network processors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), read-only memories (ROMs), random access memories (RAMs), and non-volatile storage devices for storing software. Other hardware (conventional or custom) may also be included. Their functions may be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood by the context.

[0088] Those skilled in the art will appreciate that any block diagrams herein represent conceptual diagrams of illustrative circuitry embodying the principles of the invention. Similarly, it will be appreciated that any flow charts, flow diagrams, state transition diagrams, pseudocode, and the like represent and thus various processes executed by a computer or processor, whether or not such a computer or processor is explicitly shown, can be substantially represented in a computer-readable medium.

[0089] Furthermore, the following claims are hereby incorporated into the detailed description, where each claim can stand on its own as a separate embodiment. While each claim can stand on its own as a separate embodiment, it should be noted that—while a dependent claim may refer to a specific combination with one or more other claims in a claim—other embodiments may include combinations of a dependent claim with the subject matter of each other dependent claim. Unless stated otherwise, such combinations are herein proposed. Furthermore, it is intended that features of a claim be included in any other independent claim, even if that claim is not directly dependent on that independent claim.

[0090] It should further be noted that the methods disclosed in the specification or claims may be implemented by an apparatus having means for carrying out each of the respective steps of these methods.

[0091] Reference Number List

[0092] 10 Computing Devices

[0093] 12 interfaces

[0094] 14 Processing circuit

[0095] 20 Computer equipment

[0096] 22 interfaces

[0097] 24 Processing circuit

[0098] 110 Obtain volume scan data

[0099] 120 Assign voxels to component types

[0100] 130 Identify surface voxels

[0101] 135 Discard the main voxels

[0102] 140 Extract Surface

[0103] 145 Perform local plane fitting

[0104] 150 Determine the orientation of the surface

[0105] 155 Classifying Surfaces

[0106] 160 Example of Splitting Components

[0107] 170 Determining Structural Integrity

[0108] 180 Correlating Performance with Structural Integrity

[0109] 210 Obtaining information about structural integrity

[0110] 220 Get information about performance

[0111] 230 Determine relevance

[0112] 310 Production

[0113] 320 CT scan

[0114] 330 Performance

[0115] 340 Automatic KPI

[0116] 410 CT scan

[0117] 420 Cathode and anode separation: cathode

[0118] 425 Cathode and Anode Separation: Anode

[0119] 430 Surface Extraction and Classification

[0120] 440 Instance Segmentation

[0121] 450 KPI Calculation

[0122] 510 Surface

[0123] 520 Inertial Ellipsoid

[0124] 530 spindle

[0125] 540 Surface

[0126] 550 main body

[0127] 560 Neighborhood

[0128] 570 reference point

[0129] 580 Arrow indicating local centroid

[0130] 610 bottom surface

[0131] 620 top surface

[0132] 630 Subject overlap.

Claims

1. A computer-implemented method for processing volume scan data, the method comprising: obtaining (110) volume scan data of a structural element, the structural element comprising a plurality of instances of two or more types of components, wherein the volume scan data is represented by a plurality of voxels; assigning (120) the plurality of voxels to one of two or more different component types; identifying (130) for each component type a voxel that is part of a surface of an instance of the component; extracting (140) a surface of an instance of a component using voxels that are part of a surface of a component of the corresponding type; and determining (170) information about the structural integrity of the structural element based on the extracted surface; wherein the method further comprises: determining (150) an orientation of a surface of an instance of the component, and classifying (155) the surface of the instance of the component based on the orientation of the respective surface, wherein information regarding structural integrity is determined based on the classification of the respective surface; wherein the orientation of the surface of the instance of the component is determined based on the following operations: Determine the normal of each voxel by fitting a local plane to each surface voxel, and The sign of the normal is determined based on the average positioning of the bodies of the instances of the corresponding component.

2. The method according to claim 1, wherein The structural element is a battery, wherein the two or more different component types comprise a cathode and an anode of the battery.

3. The method according to any one of claims 1 or 2, wherein The voxel is part of a surface of an instance of a component or part of a body of an instance of a component, and the method includes discarding (135) the voxel that is part of the body of the instance of a component.

4. The method according to any one of claims 1 or 2, comprising: Instances of the component are segmented (160) based on the extracted surfaces, wherein information about structural integrity of the structural element is determined based on the extracted surfaces of the instances of the component.

5. The method according to any one of claims 1 or 2, wherein Information about the structural integrity is determined based on at least one of a distance between the surfaces, an alignment of the surfaces, a uniformity of the distance along an extent of the two surfaces, an intersection of the surfaces in the extracted surfaces, and a difference between the extracted surfaces and a three-dimensional model of the structural element.

6. The method according to any one of claims 1 or 2, wherein Extracting (140) a surface of an instance of a component includes performing (145) local plane fitting on voxels that are part of the surface of a component of the corresponding type.

7. The method according to any one of claims 1 or 2, wherein The voxels are assigned to one of the two or more different component types using a machine learning model suitable for image segmentation.

8. The method according to claim 6, wherein: The voxels are assigned to one of the two or more different component types using a machine learning model based on a U-Net architecture.

9. The method according to any one of claims 1 or 2, comprising: The performance of the structural element is correlated with information regarding the structural integrity of the assembly (180).

10. A method for determining a correlation between the structural integrity of a structural element and a performance of the structural element, the method comprising: obtaining (210) information about the structural integrity of a plurality of structural elements, the information about the structural integrity being generated using the method according to any one of claims 1 to 9; obtaining (220) information about properties of a plurality of structural elements; as well as A correlation between the structural integrity of the plurality of structural elements and the performance of the plurality of structural elements is determined (230).

11. A computing device (10), comprising: Interface for exchanging information (12); as well as A processing circuit (14) configured to carry out the method of any one of claims 1 to 9.

12. A computing device (20), comprising: an interface (22) for exchanging information; as well as A processing circuit (24) configured to carry out the method of claim 10.

13. A computer program product having a program code for carrying out the method according to claim 1 or the method according to claim 10 when the program code is executed on a computer, a processor or a programmable hardware component.

Images