EP4399675A1 - Verfahren zum beurteilen einer oberfläche eines karosseriebauteils sowie verfahren zum trainieren eines künstlichen neuronalen netzes - Google Patents
Verfahren zum beurteilen einer oberfläche eines karosseriebauteils sowie verfahren zum trainieren eines künstlichen neuronalen netzesInfo
- Publication number
- EP4399675A1 EP4399675A1 EP22762073.9A EP22762073A EP4399675A1 EP 4399675 A1 EP4399675 A1 EP 4399675A1 EP 22762073 A EP22762073 A EP 22762073A EP 4399675 A1 EP4399675 A1 EP 4399675A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- variable
- polygon
- characterizing
- curvature
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/64—Analysis of geometric attributes of convexity or concavity
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30136—Metal
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Definitions
- the invention relates to a method for assessing a surface of a body part of a motor vehicle according to patent claim 1.
- the invention also relates to a method for training an artificial neural network according to patent claim 6.
- Such a method for assessing a surface of a body component of a motor vehicle and such a method for training an artificial neural network can be taken from the general prior art as known.
- the surface of the body component can be examined or checked, in particular manually.
- the surface can be checked for any surface defects.
- the surface defects can be used to assess the surface.
- the object of the invention is to create a method for assessing a surface of a body component of a motor vehicle and a method for training an artificial neural network, so that the amount of work and costs for producing the body component can be kept particularly low.
- this object is achieved by a method for assessing a surface of a body component of a motor vehicle having the features of patent claim 1 and by a method for training an artificial neural network having the features of patent claim 6 .
- Advantageous embodiments of the invention are the subject of the dependent patent claims and the description.
- a first aspect of the invention relates to a method for assessing a surface of a body component of a motor vehicle.
- the motor vehicle can be designed, for example, as a passenger car, commercial vehicle or truck.
- the body component can be understood in particular as a component of a body of the motor vehicle.
- the body can in particular be a act self-supporting body of the motor vehicle.
- the body component is preferably an outer skin component of the body or of the motor vehicle.
- the outer skin component can in particular be understood to mean that the body component is a component of an outside of the motor vehicle that delimits the motor vehicle to the outside.
- the body component can be formed from sheet metal, for example. As a result, the body component can be referred to in particular as a body panel.
- At least one virtual polygon mesh of the surface of the body component is created in the method.
- at least one virtual image of the surface of the body component is created or formed, with the virtual image being in the form of a polygon network.
- the virtual image can be understood in particular as a virtual model of the surface.
- the polygon network can be understood in particular as points connected to edges, with the polygon network comprising a multiplicity of polygons which can be designed, for example, as a triangle or as a square.
- the respective polygon comprises a plurality of points, referred to in particular as nodes, which are connected to one another via edges of the polygon.
- the surface of the body component is approximated by the virtual polygon mesh.
- the respective node can in particular be referred to as a “vertex”.
- the nodes form respective corners of the respective polygon.
- the polygon mesh is preferably three-dimensional.
- the polygon mesh can be two-dimensional.
- the polygon mesh is preferably an STL mesh.
- STL mesh the surface of the body component is described or approximated by triangular facets.
- Each triangle facet can be characterized by three corner points and an associated surface normal of the triangle.
- the STL network is in particular a polygon network which has a format referred to in particular as the STL format.
- At least one variable characterizing a curvature of the polygon mesh is determined at at least one node of the polygon mesh.
- at least one variable characterizing a curvature at at least one node of the polygon network is determined.
- the at least one variable characterizing the curvature of at least one partial area of the surface is determined as a function of the polygon network.
- the curvature characterizing variable can be calculated, for example.
- the variable characterizing the curvature can be understood in particular as a measure or a value by means of which the curvature of the polygon network, in particular at the respective node, can be characterized or described.
- the curvature of the polygon network in particular at the respective nodes, or the curvature of the surface of the body component can be inferred.
- the variable characterizing the curvature of the polygon mesh is preferably determined at or for all nodes of the polygon mesh.
- a mapping referred to in particular as a Weingarten mapping can be calculated.
- a matrix referred to in particular as a vineyard matrix or vineyard curvature matrix can be calculated for the respective node.
- Eigenvalues of the vineyard map or the vineyard matrix correspond to the principal curvatures of the polygon mesh, evaluated at the respective node.
- the main curvatures can be referred to as K1 and K2.
- the main curvature can generally be understood as a curvature of a flat curve, which results from a normal section.
- the normal section can in particular be a section of a surface with a plane determined by a surface normal vector of the surface and a tangential direction of the surface.
- the principal curvatures are then a minimum value and a maximum value of these curvatures.
- Directions of the eigenvectors can in particular be referred to as main directions of curvature.
- the main directions of curvature can be understood to mean, in particular, tangential directions.
- One of the main curves can in particular be referred to as the so-called first main curve.
- the other main curvature can in particular be referred to as the so-called second main curvature.
- the vineyard mapping can be calculated at at least one node, in particular at all nodes, of the polygon mesh. This can be done, for example, as follows:
- a normal vector is first calculated or approximated at at least one of the nodes, in particular at all nodes, of the polygon network.
- the normal vector corresponds to or approximates a normal of the polygon mesh, in particular an area of the polygon mesh, in the respective node.
- Two vectors which are orthonormal to one another and which are perpendicular to the respective normal vector are then calculated at at least one of the nodes, in particular at all nodes, of the polygon network.
- the vectors and the normal vector are used as a local, orthonormal coordinate system at the respective node.
- the vineyard image or the vineyard curvature matrix is then calculated in at least one of the nodes, in particular in all nodes, of the polygon network, expressed in the respective local coordinate system.
- the main curvatures and/or the main directions of curvature can be used as the variable characterizing the curvature.
- the main curvatures and/or the main directions of curvature can be determined as the variable characterizing the curvature at at least one, in particular at all, nodes of the polygon network.
- the Weinberg image can be calculated for at least one, in particular for all nodes of the polygon network, from which the main curvatures and/or the main directions of curvature can be calculated.
- the virtual polygon network is preferably stored in an electronic computing device after it has been created.
- the variable characterizing the curvature is preferably stored in the electronic computing device or in a second electronic computing device that is configured separately from the electronic computing device.
- at least one output variable characterizing a surface defect of the surface is determined at least indirectly, in particular directly, by means of an artificial neural network for assessing the surface.
- the variable characterizing the curvature is an input variable of the neural network, with the neural network determining the output variable, which is a variable characterizing the surface defect of the surface, as a function of the variable characterizing the curvature.
- the output variable characterizing the surface defect of the surface is an output variable of the neural network.
- the neural network can be referred to in particular as a neural network.
- the artificial neural network can be understood in particular as a software code which is stored on a computer-readable storage medium and represents one or more networked artificial neurons or can simulate their function.
- the software code can in particular contain a number of software code components which can have different functions, for example.
- an artificial neural network can implement a non-linear model or algorithm that maps an input to an output.
- the input can be an input feature vector or an input sequence.
- the output can include, for example, an output category for a classification task, one or more precessed values, or a precessed sequence.
- the entry or input can include the variable that characterizes the curvature.
- the output or output can include the variable that characterizes the surface defect.
- the neural network is preferably a trained neural network.
- the surface defect can be understood in particular as a deviation between an actual state of the surface of the body component and a target state of the surface of the body component.
- the surface defect can thus be a deviation between a target surface of the body component and an actual surface of the body component.
- the desired state or the desired surface can be understood in particular as a defined, desired state of the surface. This state can be provided during the manufacture of the surface of the body component.
- the actual state or the actual surface can be understood in particular as a real state of the surface of the body part, with the real State after the manufacture of the body component or the surface is present. The deviation or a difference between the target surface and the actual surface can thus result from the manufacture of the body component, in particular a manufacturing error or a manufacturing inaccuracy.
- the surface defect can be a dent, for example.
- the surface defect can be a waviness of the surface, for example. It can be provided that the deviation between the target surface and the actual surface is only a surface defect or is regarded as the surface defect when the deviation exceeds a predefined threshold value.
- the output variable characterizing the surface defect can be understood in particular as any measure or any value which characterizes or describes the surface defect.
- the surface defect can be inferred by means of the output variable characterizing the surface defect.
- the output variable characterizing the surface defect can assume or have at least two, in particular discrete, values.
- a first of the values can mean, for example, that the respective surface has the surface defect.
- the second value can mean, for example, that the respective surface does not have the surface defect and is therefore free of defects.
- the output variable that characterizes the surface defect can include a plurality of defect classes that differ from one another.
- the output variable that characterizes the surface defect can assume at least one of several different values, with the respective value characterizing the respective defect class of the surface defect.
- the values can in particular be referred to as error class values.
- the dent in the surface can be one of the defect classes.
- the wavy surface can be another of the defect classes.
- a first of the error class values is determined by means of the artificial neural network if the respective surface of the body component has a surface error of a first error class.
- the artificial neural network is used to determine a second error class value that differs from the first error class value if the respective surface of the body component has a surface error of the second error class.
- the output variable characterizing the surface defect can include at least one measure for characterizing a defect severity of the surface defect.
- the artificial neural network is used to determine a first intensity value of the output variable that characterizes the surface defect if the respective surface of the respective body component has a first surface defect, and the neural network is used to determine a second intensity value, which is greater than the first intensity value, of the output variable that characterizes the surface defect , if the respective surface of the respective body component has a second surface defect that is more severe than the first surface defect.
- the output variable that characterizes the surface defect has at least one measure for characterizing a position of the defect on the surface of the body component.
- the output variable that characterizes the surface defect includes coordinates for this, the coordinates describing the position or location of the surface defect on the surface of the respective body component.
- body components can usually be produced by means of forming.
- Forming tools are used for this.
- a surface defect occurs.
- the forming tool to produce test parts of the body component, referred to in particular as test impressions.
- test impressions can then be painted, in particular black, and checked manually, in particular by experts, for surface defects. In this case, development of the forming tool can usually already be completed.
- any changes or adjustments to the forming tool can be particularly time-consuming and particularly cost-intensive.
- the development of the forming tool it is usually only possible to identify insufficiently whether body components to be produced using the forming tool tend to have surface defects or whether differences between a target geometry and an actual geometry of the body component represent a relevant surface defect.
- This problem can, for example, in a Process step of the development of the forming tool exist, in which the forming tool is developed by means of simulation, in particular finite element simulation.
- the problem can alternatively or additionally exist during the incorporation of tool sets in tool construction of or for the forming tool.
- the method according to the invention can already be used to identify the surface defect or defects during the development of the forming tool.
- the surface defect is evaluated or evaluated using the neural network.
- the expert can be emulated by the neural network.
- necessary measures can be taken at an early stage during the development of the forming tool.
- the measures can be understood, for example, as measures by means of which the forming tool can be adapted or improved, so that the surface defects occurring during the production of the respective body component can be avoided or reduced.
- costs, in particular manufacturing costs and effort, in particular manufacturing effort, of the body component can be kept particularly low.
- the optical detection device is preferably a camera, in particular a stereo camera, for example a GOM stereo camera.
- the virtual polygon mesh is created as a function of at least one simulation result of an in particular structural-mechanical and/or thermomechanical simulation of at least one step of a simulated manufacturing process of the body component.
- the simulation is used to simulate at least one step of the manufacturing process or the entire manufacturing process of the body component.
- the simulation result is calculated.
- the simulation can be structural and/or thermo-mechanical.
- the simulation can also include other effects.
- the simulation result can be, for example, a virtual model or a virtual image of the surface of the body component, it being possible for the virtual image to be deformed or deformed in relation to the target surface of the body component, for example.
- This deformation is a consequence of the, in particular simulated, manufacturing process of the body component.
- the virtual polygon mesh of the surface is created from the simulation result or from the simulated image of the surface.
- a virtual or simulated surface of the body component can be assessed using the neural network.
- the assessment can take place, for example, at a particularly early stage of a development process for the body component or the forming tool.
- a respective smoothing of the respective polygon network is carried out before the determination of the respective variable characterizing the curvature.
- the variable that characterizes the curvature of the polygon mesh can thus be determined at the at least one node of the smoothed polygon mesh.
- noise can be suppressed in the polygon meshes or in the variable characterizing the curvature.
- the smoothing can in particular be referred to as filtering.
- the image can include disturbances referred to as image noise in particular.
- the disturbances cannot be related to the desired image content of the image, namely the image of the surface.
- the polygon mesh Due to the fact that the polygon mesh is created or has been created depending on the image, the polygon mesh can have the image noise or the disturbances.
- a position of at least one node of the polygon network can have no relation to the real surface.
- coordinates of the node can differ significantly from the respective coordinates of the surface of the body component.
- the polygon mesh can be smoothed.
- a position or coordinates of at least one node of the polygon network can be shifted, in particular in a respective normal direction, whereby the image noise can be reduced.
- An example smoothing method can be found in the literature: Fleishman, Shachar and Drori, Iddo and Cohen-Or, Dabniel (2003). Bilateral mesh denoising, ACM Transactions on Graphics, https://doi.org/10.1145/1201775.882368.
- Another embodiment provides that at least one two-dimensional geometric image of the polygon network is formed, with the output variable characterizing the surface defect being determined at least indirectly, in particular directly, by means of the artificial neural network as a function of the geometric image.
- the polygon network is transformed as a geometric image into a two-dimensional plane, with the geometric image being used as an input or as an input variable of the neural network.
- the geometric mapping can be understood in particular to mean that a respective point in the two-dimensional plane is assigned to each point of the polygon network.
- a respective node or point can be assigned to at least one or each node of the polygon network in the two-dimensional plane.
- a respective edge in the two-dimensional plane can be assigned to at least one or each edge of the polygon mesh.
- the polygon network thus enters the neural network indirectly as an input variable or as an input via the geometric mapping.
- the geometric mapping is a particularly well suited input variable of the neural network.
- memory requirements and learning complexity can be kept particularly low in the neural network.
- the geometric mapping can in particular be referred to as geometry mapping.
- the geometric mapping can resemble a two-dimensional matrix, for example.
- the geometric mapping is an authalic spherical parameterization.
- An exemplary geometric mapping or an exemplary method for performing such a geometric mapping can be taken from the literature: Sinha, Ayan and Bai, Jing and Ramani, Karthik (2016). Deep Learning 3D Shape Surfaces Using Geometry Images, European Conference on Computer Vision, pp 223-240, https://doi.org/10.1007/978-3-319-46466-4_14.
- a different from the geometric mapping and in particular a two-dimensional geometric image of the polygon network can be created by parameterizing a three-dimensional shape, in particular of the polygon network.
- the intermediate image can be mapped or scanned onto an octahedron.
- the octahedron can then be cut or cut off along its edges to create the geometric image.
- the authalic parameterization can be understood in particular as an area-preserving parameterization.
- the spherical authalic parameterization can be understood in particular as a combination of the spherical parameterization and the authalic parameterization.
- spatial distortions can be iteratively minimized in the polygon mesh used as the input variable of the authalic spherical parameterization and a bijective image can be created on a spherical surface.
- At least one pixel of the geometric image is assigned the respective variable characterizing the curvature
- the output variable characterizing the surface defect being at least indirectly determined by means of the artificial neural network as a function of the at least one pixel in particular directly, is determined.
- the respective pixel in the geometric mapping is assigned to at least one, in particular each, node of the polygon network, with the variable characterizing the curvature of the respective node or for the respective node being assigned to the respective pixel of the geometric mapping .
- the output variable characterizing the surface defect is determined by means of the artificial neural network as a function of the pixel and the variable associated with the respective pixel that characterizes the curvature.
- the output variable characterizing the surface defect is determined by means of the artificial neural network as a function of the geometric image, which includes the respective variable characterizing the curvature.
- variable characterizing the curvature is assigned to the respective pixel of the geometric image
- the geometric image or the pixel is encoded, in particular colored, with the respective variable characterizing the curvature.
- the variable that characterizes the curvature can be characterized or represented as a function of a color or a color intensity of the geometric image, in particular of the pixel.
- the variable that characterizes the curvature can, for example, include a number of sub-variables. Provision can be made for a respective two-dimensional geometric image of the polygon network to be formed for each of the partial variables, the respective partial variable characterizing the curvature being assigned to at least one pixel, in particular all pixels, of the respective geometric image. Alternatively, it can be provided that a, in particular exactly one, two-dimensional geometric image of the polygon network is formed, with at least one pixel, in particular all pixels, of the geometric image being assigned all the respective partial variables characterizing the curvature.
- a first of the part sizes can be the first principal curvature.
- a second of the part sizes can be the second principal curvature.
- the respective variable characterizing the curvature can, for example, include at least one extended curvature value.
- the extended curvature value can be, for example, a shape index of curvature and/or an intensity index of curvature.
- the shape index and/or the curvature index can be used as the variable that characterizes the curvature.
- a third of the part quantities can be the shape index.
- a fourth of the sub-variables can be the intensity index.
- the variable characterizing the curvature can include the first main curvature and/or the second main curvature and/or the shape index and/or the intensity index.
- shape index shape index
- intensity index curvedness intensity
- a shape, in particular a geometric shape, of the curvature can be characterized or described by means of the shape index.
- the shape index can be calculated depending on the principal curvatures (ki, k2).
- the curvature shape by exactly one parameter, namely the shape index, to be characterized.
- the shape index (s) can be calculated as follows:
- the shape index preferably assumes values between -1 and 1.
- the form index can be used to describe or characterize convex, concave and hyperbolic surfaces.
- convex and concave surfaces can each be located on different sides or areas (in particular with different signs) of the scale. For example, a value pair that differs from one another only with regard to its respective sign characterizes two surfaces designed to correspond to one another, for example a punch and a die.
- an intensity or a strength of the curvature can be characterized or described by means of the intensity index.
- the intensity index can be calculated as a function of the principal curvatures (ki, k2).
- the curvature intensity can be characterized by precisely one parameter, namely the curvature index.
- the index of curvature (c) can be calculated as follows:
- the intensity index corresponds to a magnitude of a reciprocal of a radius of the sphere.
- the artificial neural network is preferably a neural network designed as a convolutional neural network (CNN).
- CNN network can be referred to as a convolutional neural network.
- the neural network is preferably an artificial neural network designed as a region-based convolutional neural network (R-CNN).
- R-CNN region-based convolutional neural network
- the two-dimensional mapping is particularly well suited for use as an input variable in the CNN network or R-CNN network, in particular compared to the three-dimensional polygon network. The same applies to training the CNN network or R-CNN network. Memory requirements and learning complexity can be kept particularly low in the CNN network or R-CNN network.
- the R-CNN network is preferably designed as a Fast R-CNN network or as a Faster R-CNN network.
- a neural network is described, for example, in the following references:
- the output variable that characterizes the respective surface defect comprises at least one bounding box, which completely surrounds the respective at least one pixel in the respective geometric image, at which a respective surface defect has been determined.
- the output variable that characterizes the respective surface defect includes at least one piece of position information or location information that characterizes or describes a position of the bounding box in the geometric image.
- the bounding box is thus an indication of the location or position of the surface defect in the geometric image.
- the output variable that characterizes the surface defect can include coordinates that describe or form the respective bounding box.
- the bounding box can completely surround a number of pixels for which a respective surface defect has been determined.
- the output variable characterizing the respective surface defect is preferably assigned to the respective pixel of the respective geometric image.
- the output variable characterizing the respective surface defect is assigned to the respective node of the polygon network, with the respective node of the polygon network being the node to which the respective pixel is assigned. This allows the respective surface defects or the Position of the respective surface defect on the surface are shown particularly clearly on the respective polygon mesh.
- a second aspect of the invention relates to a method for training an artificial neural network.
- Advantages and advantageous configurations of the first aspect of the invention are to be regarded as advantages and advantageous configurations of the second aspect of the invention and vice versa.
- a respective virtual polygon mesh is created for at least one respective surface of a multiplicity of body components.
- a respective digital image of the respective surface of the body components is formed, with the digital image being in the form of the virtual polygon network.
- a first virtual polygon mesh is created for at least one respective surface of a first of the body components.
- a second virtual polygon mesh which is different from the first virtual polygon mesh, is created on at least one respective surface of a second one of the body parts, which is different from the first body part.
- the body components can be structurally identical or differ from one another in terms of their respective design, in particular their geometry.
- the respective surface or the body components are preferably real components.
- the respective surface of the respective body component is preferably arranged in a detection range of an optical detection device, with at least one image of the respective surface arranged in the detection range of the optical detection device being detected by means of the optical detection device and with the respective polygon network being created as a function of the detected image.
- the respective virtual polygon network is preferably stored in an electronic computing device, in particular a database of the electronic computing device.
- the artificial neural network is trained at least indirectly, in particular directly, using the variables characterizing the curvature.
- a learning process of the artificial neural network referred to in particular as learning, is carried out by means of the variables characterizing the curvature.
- At least one respective variable characterizing a respective surface defect of the respective surface of the respective body component is determined, in particular manually.
- the manual determination can preferably be carried out by experts.
- the respective surfaces can be examined and evaluated by the experts.
- the body components can be installed, i. H. are in the installation position in the motor vehicle, or are available as individual parts and can be stored in any way.
- the presence as an individual part can be understood in particular as meaning that the respective body part is not in its respective installed position in the motor vehicle.
- variable that characterizes the surface defect can be understood in particular as any measure or any value that characterizes or describes the surface defect.
- the surface defect can be inferred by means of the variable characterizing the surface defect.
- the variable characterizing the surface defect can assume or have at least two, in particular discrete, values.
- a first of the values can mean, for example, that the respective surface has the surface defect.
- the second value can mean, for example, that the respective surface does not have the surface defect and is therefore free of defects.
- the variable that characterizes the surface defect can include a number of defect classes that differ from one another.
- the variable that characterizes the surface defect can assume at least one of a plurality of mutually different values, with the respective value characterizing the respective defect class of the surface defect.
- the values can in particular be referred to as error class values.
- the dent in the surface can be one of the defect classes.
- the wavy surface can be another of the defect classes.
- a first of the error class values is determined, in particular manually, if the respective surface of the body component has a surface error of a first error class.
- a second error class value that differs from the first error class value is determined, in particular manually, if the respective surface of the body component has a surface error of the second error class.
- variable that characterizes the surface defect can include at least one measure for characterizing a defect severity of the surface defect. For example, a first intensity value of the variable characterizing the surface defect is determined, in particular manually, if the respective surface of the respective body component has a first surface defect, and a second intensity value, which is greater than the first intensity value, of the variable characterizing the surface defect is determined, in particular manually, if the respective surface of the respective body component has a second surface defect that is more severe than the first surface defect.
- variable that characterizes the surface defect has at least one measure for characterizing a position of the defect on the surface of the body component.
- the variable that characterizes the surface defect includes coordinates for this, the coordinates describing the position or location of the surface defect on the surface of the respective body component.
- the respective variable characterizing the respective surface defect of the surface of the respective body component is assigned to at least one respective polygon of the respective polygon network, in particular an area and/or a node and/or an edge of the polygon.
- the variable characterizing the respective surface defect is assigned to the respective polygon or polygons that determine the position of the respective surface defects on the respective polygon mesh, in particular as best as possible.
- the variable characterizing the respective surface defect of the surface of the respective body component is preferably stored in the electronic computing device, in particular in the database.
- the respective determined surface defect or a position of the respective surface defect can be marked on the, in particular real, respective surface of the body component.
- the respective marked surface defects can be detected when detecting the respective surface using the optical detection device.
- the respective detected surface defects or the respective detected positions of the respective surface defects can then be assigned to the respective virtual polygon mesh, in particular to the respective node and/or the respective area and/or the respective edge.
- the respective surface defects or the respective position of the respective surface defects on the respective polygon, in particular respective nodes and/or or the respective surface and/or respective edge of the respective polygon, of the polygon network is marked, in particular manually, and is thereby assigned to the respective polygon.
- At least one two-dimensional geometric image of the respective polygon network is formed, with at least one pixel, in particular several pixels, of the respective geometric image being assigned the variable characterizing the respective surface defect, and with the artificial neural network being at least indirectly, in particular is trained directly, by means of the geometric mapping comprising the respective surface defect.
- the respective polygon network is transformed as a geometric image into a two-dimensional plane, with the respective variable characterizing the respective surface defect being assigned to the at least one pixel of the respective geometric image.
- the neural network is trained by means of the geometric images and the variables that characterize the surface defects and are assigned to the pixels. As a result, the neural network can be trained particularly advantageously, as a result of which it can make particularly precise predictions.
- At least one pixel of the geometric image is assigned the variable characterizing the respective curvature and the artificial neural network is trained at least indirectly, in particular directly, by means of the geometric image, which includes the variable characterizing the curvature.
- at least one two-dimensional geometric image of the respective polygon network is formed, with at least one pixel of the geometric image being assigned the variable that characterizes the respective curvature, and with the artificial neural network being curvature characterizing size includes, is trained.
- the bounding box surrounding the respective pixel in particular a plurality of pixels, is formed in the respective geometric image, the variable characterizing the respective surface defect being assigned to the respective pixel.
- the neural network is trained using the respective geometric mapping that includes the bounding box.
- the neural network can be trained with position or location information of the respective surface defect.
- the position or attitude information can be predicted by means of the neural network.
- the respective pixel which is enclosed by the bounding frame, is preferably a respective pixel at which the surface of the respective body component has a surface defect.
- the variable characterizing the respective surface defect assigned to the respective pixel has a value which corresponds to an existing surface defect.
- Bounding box specifically coordinates of the bounding box, in which electronic computing device, in particular the database, stored.
- the surface defects marked in the respective polygon network are surrounded by the bounding box in the geometric image, the coordinates of which are stored in the electronic computing device, in particular in the database.
- a respective virtual desired polygon mesh is created for at least one respective surface of a respective virtual desired geometry of the respective body component.
- the respective virtual desired polygon network which approximates the virtual desired geometry, is created from the respective virtual desired geometry.
- the target geometry can be, for example, a CAD geometry of the surface of the respective body component.
- the target geometry has no manufacturing-related surface defects and is therefore free of the manufacturing-related surface defects.
- the target geometry is thus a desired ideal geometry of the surface of the respective body component.
- the target polygon mesh is preferably an STL polygon mesh.
- the respective target polygon mesh can be stored in the electronic computing device, in particular in the database.
- At least the variable characterizing a respective curvature at at least one node of the respective target polygon mesh is determined.
- the at least one variable characterizing the curvature of the target polygon mesh is determined at the at least one node of the target polygon mesh.
- At least one respective two-dimensional geometric image of the respective target polygon network is preferably formed, with the artificial neural network being trained at least indirectly, in particular directly, by means of the respective geometric image.
- the respective target polygon network is transformed as a geometric image into a two-dimensional plane, with the neural network being trained by means of the respective geometric image.
- the neural network can learn the ideal geometry of the target polygon network, as a result of which precise predictions with regard to the surface defects can be made possible by means of the neural network.
- FIG. 2 shows a schematic perspective view of a body component on which a method according to the invention can be carried out
- Fig. 3 is a schematic perspective view of a virtual polygon mesh of
- Fig. 4 is a schematic perspective view of a virtual polygon mesh of
- Fig. 5 is a schematic representation of a two-dimensional geometric
- Fig. 6 is a schematic process diagram of a process for
- FIG. 1 shows a schematic method diagram of a method for training 1 an artificial neural network 2.
- the artificial neural network 2 is preferably a region-based convolutional neural network (R-CNN).
- FIG. 2 shows a body component 3 of a motor vehicle in a schematic perspective view.
- the body component 3 is designed as a side frame of the motor vehicle or a body of the motor vehicle.
- the body component 3 is an outer skin component of the body of the motor vehicle.
- the body component 3 shown in FIG. 2 is an exemplary body component, since the method is carried out for a large number of such body components 3 .
- an assessment 4 is first carried out for the body components 3 .
- Rating 4 can in particular be referred to as an assessment.
- at least one respective surface 5 of the respective body component 3 is examined, in particular manually, for surface defects 6 in the track.
- the assessment 4 or the examination is preferably carried out by experts.
- the respective surface defect 6 can be understood in particular as a deviation between a target state of the respective surface 5 and an actual state of the respective surface 5 .
- the respective surface defect 6 can be a dent or a waviness of the surface 5, for example.
- the respective body component 3 is preferably a body component 3 that is painted, in particular black.
- variable 7 characterizing the respective surface defect 6 of the respective surface 5 of the respective body component 3 is determined, in particular manually.
- size 7 can be referred to as a defect size.
- the variable 7 characterizing the respective surface defect 6 can, for example, comprise a respective parameter which describes whether the respective surface 5 has a surface defect 6 or whether the respective surface 5 is free of the surface defect 6 .
- the variable 7 characterizing the respective surface defect 6 can include at least one parameter which characterizes a respective defect class of the respective surface defect 6 .
- the respective defect class can be the dent or the waviness of the respective surface 5, for example.
- variable 7 characterizing the respective surface defect 6 can comprise at least one parameter which characterizes a defect severity of the respective surface defect 6, referred to in particular as defect intensity.
- the variable 7 characterizing the respective surface defect 6 can comprise at least one parameter which characterizes or describes a respective location or respective position of the respective surface defect 6 on the respective surface 5 .
- At least one respective image 9 of the respective surface 5 arranged in a detection region 10 of the optical detection device 8 is detected by means of an optical detection device 8 .
- a respective virtual polygon mesh 11 of the respective surface 5 of the respective body component 3 is created as a function of the respective recorded image 9 .
- the body components 3 on which the evaluation 4 was carried out are digitized as a polygon network 11 by means of the optical detection device 8 .
- Fig. 3 shows a schematic perspective view of the polygon network 11 of the surface 5 of the body component 3.
- the respective polygon network 11 comprises a plurality of polygons 12.
- Fig. 4 shows a schematic perspective view of the polygon network 11, wherein in a partial view A respective polygons 12 are illustrated.
- the respective variable 7 characterizing the respective surface defect 6 of the respective surface 5 of the respective body component 3 is assigned to a respective polygon 12 of the respective polygon network 11 .
- the variable 7 can be assigned, for example, to at least one respective area 13 and/or at least one respective node 14 and/or at least one respective edge 15 of the respective polygon 12 .
- the assignment of the respective variable 7 characterizing the respective surface defect 6 can be carried out in two variants, for example.
- the respective polygon 12 or a respective region 16, which comprises a plurality of the respective polygons 12, is marked in the respective polygon network 11.
- the area 16 corresponds to a respective area 17 or approximates the respective area 17 in which the respective surface defect 6 on the real component has been identified.
- the area 16 is a virtual image of the area 17.
- the marking can be carried out manually.
- the respective area 17, which includes the respective surface defect 6, is marked on the body component 3 and thus on the real component.
- the image 9 captured by means of the optical capture device 8 also includes the marked area 17.
- the respective virtual polygon network 11 is created, depending on the image 9, which includes in particular the area 17, the respective polygon 12 or the respective virtual area 16 , In particular automatically, are marked when creating the respective polygon mesh 11 .
- the respective polygon mesh 11 and the respective variable 7 are preferably stored in an electronic computing device, in particular a database.
- the polygon meshes 11 are preferably smoothed, that is to say the polygon meshes 11 can be smoothed.
- the smoothed polygon meshes 11 can be stored in the electronic computing device, in particular the database. It is possible that further method steps, in particular all further method steps, can be carried out using the smoothed polygon meshes 11 or the unsmoothed polygon meshes 11 .
- the variable 18 characterizing the respective curvature can include, for example, main curvatures k1, k2 and/or main directions of curvature and/or a shape index s and/or an intensity index c of the respective node 14 or at the respective node 14.
- variable 18 that characterizes the curvature can include a plurality of partial variables 18a-d, which in particular can be referred to as partial information.
- a first of the partial sizes 18a can be a first of the main curvatures K1.
- a second of the part sizes 18b can be the second main curvature K2.
- a third of the part sizes 18c can be the shape index s.
- the fourth partial quantity 18d can be the intensity index c.
- the main curvatures k1, k2 and the main directions of curvature can be calculated, for example, by means of a mathematical mapping referred to in particular as a Weingarten mapping as a function of the respective polygon network 11, in particular the nodes 14.
- the form index s and the intensity index c can be calculated, for example, as a function of the main curvatures k1, k2 and/or the main directions of curvature.
- the respective variable 18 of the polygon meshes 11, which is referred to in particular as curvature information and characterizes the curvature, is preferably stored in the electronic computing device, in particular the database.
- At least one two-dimensional geometric image 19 of the respective polygon network 11 is formed.
- 5 shows the two-dimensional geometric image 19 in a schematic representation.
- the respective geometric image 19 comprises a multiplicity of pixels 20, with each of the pixels 20 being assigned at least one node 14 of the respective polygon network 11.
- the respective geometric image 19 can have at least one partial area 21 to which no polygon 12 or node 14 of the respective polygon network 11 is assigned.
- the partial area 21 is an image of an image area of the respective image 9 , the image area, if the image 9 were to be displayed, not showing the respective surface 5 of the body component 3 .
- the partial area 21 can thus, for example, characterize or correspond to a recess in the body component 3 .
- the sub-area 21 differs from the remaining areas of the geometric image 19 in that the sub-area 21 does not have any networks or does not depict the polygon network 11 .
- Several of the partial areas 21 are shown in FIG. 5 . These correspond to the respective door areas or window areas of the motor vehicle or depict the respective door areas or window areas.
- the geometric image 19 is preferably formed from the respective polygon mesh 11 by means of authenic spherical parameterization.
- the variable 18 characterizing the respective curvature is assigned to at least one of the pixels 20 of the respective geometric image 19 .
- the geometric image 19, in particular the pixels 20, can be colored depending on the variable characterizing the curvature.
- exactly one of the geometric images 19 can be formed for each of the polygon meshes 11, with the at least one pixel 20 of exactly one geometric image 19 being assigned several, in particular all, of the partial variables 18a-d of the variable 18 characterizing the respective curvature.
- a plurality of geometric images 19 can be formed for each of the polygon meshes 11, with each of the plurality of geometric images 19 being assigned precisely one of the partial variables 18a-d at at least one of the pixels 20.
- the geometric images 19 are preferably stored in the electronic computing device, in particular in the database.
- the artificial neural network 2 can be trained using the respective geometric image 19 which includes the respective variable 7 characterizing the respective surface defect 6 .
- At least one bounding box 22 is created, which completely surrounds the respective at least one pixel 20 to which the variable 7 characterizing the respective surface defect 6 is assigned.
- the artificial neural network 2 is trained using the respective bounding frame 22 or using the respective geometric image 19 which includes the respective bounding frame 22 .
- a respective virtual desired polygon mesh 23 of at least one respective surface 24 of a respective virtual desired geometry 25 of the respective body component 3 is created.
- the virtual target geometry 25 is preferably a CAD geometry of the respective surface 5 or of the respective body component 3.
- the target geometry 25 preferably does not include any production-related surface defects 6.
- the virtual target polygon mesh 23 can be stored in the electronic computing device, in particular in the database.
- variable 18 characterizing the respective curvature is preferably determined on the respective virtual target polygon meshes 23 . It is therefore preferably provided that the at least one respective variable 18 characterizing the curvature of the respective target polygon mesh 23 is determined at at least one node 14 of the respective target polygon mesh 23 .
- the geometric image 26 differs in particular from the geometric image 19 in that the geometric image 26 does not include the manufacturing-related surface defects 6 or the bounding box 22 .
- the geometric image 26 is free from the manufacturing-related surface defects 6 and the bounding box 22.
- At least one pixel of the geometric image 26 is assigned the variable 18 characterizing the respective curvature of the target polygon network 23, the artificial neural network 2 using the geometric image 26, which includes the variable 18 characterizing the curvature of the virtual target polygon network 23, is trained.
- the respective geometric image 26 can be stored in the electronic computing device.
- Data stored in the electronic computing device, in particular in the database, are preferably used for training 1 of the artificial neural network 2 .
- the data include the geometric image 19, in particular the variable 18 of the polygon mesh 11 characterizing the curvature and/or the variable 7 characterizing the respective surface defect 6 and/or the respective bounding box 22, and/or the geometric image 26, in particular the respective the Variable 18 characterizing the curvature of the target polygon mesh 23.
- the first principal curvature k1 and/or the second principal curvature k2 and/or the shape index s and/or the intensity index c can be used as the variable 18 characterizing the curvature.
- the aim of training 1 is for the trained neural network 2 to be able to detect the respective surface defects 6, in particular already during a development of the Body component 3 and / or the forming tool to predict or to recognize.
- neural network 2 can be used to localize and/or classify respective surface defects 6 in polygon networks, in particular in target polygon network 23 .
- the manufacturing effort and manufacturing costs of the body components 3 can be kept particularly low, in particular compared to manual localization or classification of the surface defects 6.
- Fig. 6 shows a schematic method diagram of a method for assessing 27 the respective surface 5 of the body component 3 of the motor vehicle using the trained artificial neural network 2.
- the optical detection device 8 uses the optical detection device 8 to create 5 of the body component 3 is detected.
- the virtual polygon mesh 11 of the surface 5 is created as a function of the captured image 9 .
- the polygon network 11 can thus be generated by digitizing the body component 3 using the detection device 8 .
- the virtual polygon mesh 11 can be created as a function of at least one simulation result 28 of an in particular structural-mechanical and/or thermomechanical simulation 29 of at least one step of a simulated manufacturing process 30 of the body component 3 .
- the simulation 29 is preferably a finite element simulation (FEM), in particular a three-dimensional one.
- FEM finite element simulation
- the simulation 29 is carried out with a target geometry 25 of the body part 3 .
- the virtual polygon network 11 is preferably stored in the electronic computing device, in particular in the database.
- variable 18 characterizing the curvature for the virtual polygon mesh 11 is determined. Provision is thus made for the at least one variable 18 characterizing the curvature of the polygon mesh 11 to be determined at at least one node 14 of the virtual polygon mesh 11 .
- Variable 18 of polygon mesh 11 that characterizes the curvature is preferably stored in the electronic computing device, in particular in the database.
- At least one two-dimensional geometric image 31 of the polygon network 11 is formed, with depending on the geometric image 31 using the artificial neural network 2 at least one output variable 32 characterizing a surface defect 6 of the surface 5 is determined.
- Fig. 7 shows a schematic illustration of the neural network 2 and a prediction of the output variable 32 as a function of the geometric image 31. It is preferably provided that at least one pixel 20 of the geometric image 31 has the respective curvature of the polygon network 11 characterizing size 18 is assigned.
- the geometric image 31 differs in particular from the geometric image 19 in that the geometric image 31 does not include the surface defects 6 or the bounding box 22 .
- the geometric mapping 31 is free from the surface defects 6 and the bounding box 22.
- the geometric mapping 31 can be used as an input variable of the artificial neural network 2.
- exactly one geometric image 31 can be formed for the polygon mesh 11, with the at least one pixel 20 of the precisely one geometric image 31 being assigned several, in particular all, of the partial variables 18a-d of the variable 18 characterizing the respective curvature.
- a plurality of geometric images 31 can be formed for the polygon mesh 11, with each of the plurality of geometric images 31 being assigned to at least one of the pixels 20 exactly one of the part sizes 18a-d.
- the first partial variable 18a in particular the first principal curvature k1
- a second of the geometric images 31 can be assigned the second partial variable 18b, in particular the second principal curvature k2, on at least one of the pixels 20.
- a third of the geometric images 31 can be assigned to at least one of the pixels 20 the third partial variable 18c, in particular the shape index s.
- the fourth partial variable 18d in particular the intensity index c, can be assigned to a fourth of the geometric images 31 on at least one of the pixels 20 . This is illustrated in FIG.
- the artificial neural network 2 is used to generate the output variable 32 that characterizes the surface defect 6 is determined.
- the at least one output variable 32 characterizing the surface defect 6 of the surface 5 for assessing 27 the surface 5 is determined at least indirectly as a function of the variable 18 characterizing the curvature by means of the artificial neural network 2 .
- digital or digitized body parts 3, in particular from tool incorporation, can be assessed or evaluated with regard to surface defects 6 by means of the trained neural network 2.
- any surface defects 6 can be identified, in particular localized or classified, for example in a particularly early development phase of the body component 3 .
- the manufacturing effort and manufacturing costs of the body components 3 can thus be kept particularly low, in particular compared to manual localization or classification of the surface defects 6.
- the output variable 32 is preferably a vector referred to in particular as a result vector.
- the initial variable 32 can be identical to the variable 7 characterizing the surface defect 6, or the initial variable 32 and the variable 7 can be at least partially different from one another.
- the result vector preferably includes a plurality of components 33.
- at least one of the components 33 includes coordinates of the bounding box 22.
- the position of the surface defect 6, which has been predicted by means of the artificial neural network 2 can thus be localized using the output variable 32.
- at least one of the components includes the classification of the respective surface defect 6.
- the class of the surface defect 6 referred to as the defect type and/or the defect severity of the surface defect 6, referred to in particular as the defect intensity, can be predicted by means of the artificial neural network 2.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102021122939.1A DE102021122939B4 (de) | 2021-09-06 | 2021-09-06 | Verfahren zum Beurteilen einer Oberfläche eines Karosseriebauteils sowie Verfahren zum Trainieren eines künstlichen neuronalen Netzes |
| PCT/EP2022/072442 WO2023030842A1 (de) | 2021-09-06 | 2022-08-10 | Verfahren zum beurteilen einer oberfläche eines karosseriebauteils sowie verfahren zum trainieren eines künstlichen neuronalen netzes |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4399675A1 true EP4399675A1 (de) | 2024-07-17 |
Family
ID=83151600
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP22762073.9A Pending EP4399675A1 (de) | 2021-09-06 | 2022-08-10 | Verfahren zum beurteilen einer oberfläche eines karosseriebauteils sowie verfahren zum trainieren eines künstlichen neuronalen netzes |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US20240311991A1 (de) |
| EP (1) | EP4399675A1 (de) |
| CN (1) | CN117501308A (de) |
| DE (1) | DE102021122939B4 (de) |
| WO (1) | WO2023030842A1 (de) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20240067420A (ko) * | 2022-11-09 | 2024-05-17 | 현대자동차주식회사 | 사이드 아우터 추출 시스템 및 사이드 아우터 추출 방법 |
| CN117952954B (zh) * | 2024-02-26 | 2024-12-27 | 河南许继仪表有限公司 | 应用bp神经网络的配电箱体数据解析系统 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| BR102016028266A2 (pt) | 2016-12-01 | 2018-06-19 | Autaza Tecnologia Ltda - Epp | Método e sistema para a inspeção automática de qualidade de materiais |
| US10346969B1 (en) | 2018-01-02 | 2019-07-09 | Amazon Technologies, Inc. | Detecting surface flaws using computer vision |
| DE102018207411A1 (de) | 2018-05-14 | 2019-11-14 | Audi Ag | Verfahren zur Ermittlung von Messinformationen in einem optischen Koordinatenmessgerät |
| MX2021007733A (es) | 2018-12-25 | 2021-08-05 | Jfe Steel Corp | Metodo de generacion de modelo aprendido, modelo aprendido, metodo de inspeccion de defectos de superficie, metodo de fabricacion de acero, metodo de determinacion de pasa/no pasa, metodo de determinacion de grado, programa de determinacion de defectos de superficie, programa de determinacion de pasa/no pasa, sistema de determinacion y equipo de fabricacion de acero. |
-
2021
- 2021-09-06 DE DE102021122939.1A patent/DE102021122939B4/de active Active
-
2022
- 2022-08-10 EP EP22762073.9A patent/EP4399675A1/de active Pending
- 2022-08-10 US US18/572,876 patent/US20240311991A1/en active Pending
- 2022-08-10 WO PCT/EP2022/072442 patent/WO2023030842A1/de not_active Ceased
- 2022-08-10 CN CN202280043005.5A patent/CN117501308A/zh active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| US20240311991A1 (en) | 2024-09-19 |
| DE102021122939A1 (de) | 2023-03-09 |
| WO2023030842A1 (de) | 2023-03-09 |
| DE102021122939B4 (de) | 2023-06-01 |
| CN117501308A (zh) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102018129425B4 (de) | System zur Erkennung eines Bearbeitungsfehlers für ein Laserbearbeitungssystem zur Bearbeitung eines Werkstücks, Laserbearbeitungssystem zur Bearbeitung eines Werkstücks mittels eines Laserstrahls umfassend dasselbe und Verfahren zur Erkennung eines Bearbeitungsfehlers eines Laserbearbeitungssystems zur Bearbeitung eines Werkstücks | |
| EP1173749B1 (de) | Bildbearbeitung zur vorbereitung einer texturanalyse | |
| DE69838581T2 (de) | Verfahren zur determinierung von fehlern im form einer willkürlichen krummen oberflache | |
| EP2753897B1 (de) | Verfahren und vorrichtung zum erkennen von abweichungen einer oberfläche eines objekts | |
| EP2284795A2 (de) | Quantitative Analyse, Visualisierung und Bewegungskorrektur in dynamischen Prozessen | |
| DE102016100101A1 (de) | System zum maschinellen sehen und analytisches verfahren zur segmentierung planarer oberflächen | |
| DE102021122939B4 (de) | Verfahren zum Beurteilen einer Oberfläche eines Karosseriebauteils sowie Verfahren zum Trainieren eines künstlichen neuronalen Netzes | |
| DE112023000151T5 (de) | Inspektionsunterstützungssystem, inspektionsunterstützungsverfahren und inspektionsunterstützungsprogramm | |
| DE102019120696B4 (de) | Vorrichtung und Verfahren zur Reifenprüfung | |
| WO2018206225A1 (de) | Computerimplementiertes verfahren zur vermessung eines objekts aus einer digitalen darstellung des objekts | |
| EP1882232B1 (de) | Verfahren und vorrichtung zum bestimmen von materialgrenzen eines prüfobjekts | |
| DE102020208474B4 (de) | Messung der Empfindlichkeit von Klassifikatoren anhand zusammenwirkender Störungen | |
| DE102009056467A1 (de) | Verfahren zur Bestimmung von Oberflächen in Voxeldaten | |
| DE112019007961T5 (de) | Modellgenerator und aufnahmeroboter | |
| DE10023377A1 (de) | Verfahren zur Erhöhung der Leistungsfähigkeit einer Computereinrichtung bei Finite-Elemente-Simulationen und eine solche Computereinrichtung | |
| DE102023209685A1 (de) | Verfahren zum Bewerten eines Trainingsdatensatzes für ein Maschinenlernmodell | |
| EP4202850A1 (de) | Verfahren zur erfassung eines schadens an einem fahrzeug | |
| EP1435065A1 (de) | Automatische ermittlung von geometrischen modellen für optische teilerkennungen | |
| EP3685349B1 (de) | Verfahren zur erkennung einer geometrie eines teilbereichs eines objekts | |
| EP4121950A1 (de) | Verfahren zum analysieren eines bauteils, verfahren zum trainieren eines systems, vorrichtung, computerprogramm und computerlesbares speichermedium | |
| WO2024132592A1 (de) | Verfahren und vorrichtung zur erzeugung von anomaliekarten von fertigungsobjekten | |
| EP4660574A1 (de) | Verfahren zur generierung synthetischer daten von objekten, vorzugsweise von produkten | |
| DE102019118924A1 (de) | Verfahren und Vorrichtung zur Generierung von Trainingsdaten für ein maschinelles Lernverfahren zum Bestimmen von Merkmalen von Strukturauffälligkeiten eines räumlich ausgedehnten Objekts | |
| DE102019208864A1 (de) | Erkennungssystem, Arbeitsverfahren und Trainingsverfahren | |
| DE102020212005B4 (de) | Verfahren zur Bestimmung der für einen Bildklassifikator entscheidungsrelevanten Bildanteile |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20240312 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20251022 |