Method for the mark-free tracking of components produced in moulding processes

Abstract

The invention relates to a method, in particular an at least partially computer-implemented method, for identifying components from a moulding process during manufacture of a product comprising the component (3), the method comprising the following steps: - providing (S1) reference image recordings of components (3) produced in a plurality of moulding devices (2), wherein the reference image recordings are each assigned a component identification specification, - assigning (S2) the reference image recordings to an associated group of components (3); - providing or selecting (S3) a candidate for each group of reference image recordings, wherein the candidate in each case corresponds to a comparison image recording; - providing a query image recording of a component (3) to be identified; - carrying out (S5) first pairwise image comparisons of the query image recording with the candidates for each group of reference image recordings in order to identify the group of reference image recordings as the one whose candidate has the greatest similarity to the query image recording; - carrying out (S6) second pairwise image comparisons of the query image recording with the reference image recordings of the identified group of reference image recordings in order to ascertain the most similar reference image recording and to determine the component identification specification assigned to the ascertained reference image recording.

Description

[0001] Description

[0002] title

[0003] Methods for marking-free tracking of components manufactured in molding processes

[0004] Subject matter of the invention

[0005] The invention relates to methods for tracking components manufactured in molding processes, such as casting, stamping, or other forming processes, so that the components can be uniquely identified. The invention further relates to methods for reducing the identification effort for such components.

[0006] Technical background

[0007] In product manufacturing, components are often produced in large quantities using casting, stamping, or similar molding processes. Particularly for defect tracking, it is necessary to be able to clearly trace components manufactured in this way throughout the product's manufacturing process. This means providing a unique identifier for each component, enabling the unambiguous identification of its manufacturing steps with regard to batch, production time, and other relevant factors.

[0008] To track components manufactured using molding processes, identification codes are often printed onto the components or marked during the molding process. In contrast, markless tracking methods eliminate the need for marking components and instead use the surface texture or structure of the component as a feature that allows for the unique identification of each component. Markless tracking of a large number of identically manufactured components utilizes a reference image of the component's surface, created immediately after its production. Characteristic structural features are captured and stored from this reference image. Each component is assigned an individual identifier, which is stored along with the characteristic structural features.

[0009] If, in a product state where such a component is installed or intended to be installed, it is necessary to precisely identify the component, the component is captured again using a camera, and structural features of the resulting so-called query image are determined. The structural features of the query image are then compared with the structural features of the registered reference image. If the compared images are highly similar, the component can be unambiguously identified.

[0010] This method makes it possible to track a component without marking, thus saving the effort of marking a component while still enabling a clear assignment of a component to a molding process carried out at a specific time.

[0011] The publication DE 10 2019 210 580 A1 describes a method for determining the position of an image area corresponding to a first image area in a second image.

[0012] The publication DE 10 2021 203 081 A1 describes a method for verifying the authenticity of an object, whereby the user, e.g., via smartphone (app), photographs an object, e.g., a banknote, and provides part of the serial number, and shortly afterwards receives confirmation from the app, if the object is authentic, as well as the rest of the serial number.

[0013] For components manufactured using plastic injection molding processes, the surface structures often exhibit a very high degree of similarity, meaning that the existing method for the unambiguous identification of the component in question may be unreliable. This can lead to a risk of misidentification, and in particular, components produced from the same mold could be confused.

[0014] The object of the present invention is to provide a method for identifying components manufactured using a molding process, which enables higher reliability and faster identification.

[0015] Disclosure of the invention

[0016] This problem is solved by the method for identifying components from a molding process during the manufacture of a product comprising the component, according to claim 1, as well as a corresponding device and a manufacturing system according to the dependent claims.

[0017] Further details are specified in the dependent claims.

[0018] According to a first aspect, a procedure for identifying components from a molding process during the manufacture of a product that includes the component is provided, with the following steps:

[0019] Providing reference images of components produced in multiple molding devices, wherein each reference image is assigned a component identification identifier that uniquely identifies the component in question.

[0020] Assigning the reference images to a respective group of components;

[0021] Providing or selecting one candidate for each group of reference images, with each candidate corresponding to a comparison image;

[0022] Providing a query image of a component to be identified; performing initial pairwise image comparisons of the query image with the candidates for each group of reference images to identify the group of reference images whose candidate is most similar to the query image; performing second pairwise image comparisons of the query image with the reference images of the identified group of reference images to determine the most similar reference image and to determine the component identification code associated with the identified reference image.

[0023] Furthermore, it may be provided that the assignment of the reference images to a respective group of components is carried out using pairwise image comparisons, which evaluate a similarity between each pair of reference images, whereby the reference images are assigned to a respective group using a clustering procedure, the similarities of which to each other exceed a predetermined similarity threshold.

[0024] The components are typically manufactured in high volumes using various molding equipment, particularly injection molding, deep drawing, or other molding processes. This results in characteristic surface properties that are unique to each component.

[0025] The different surface textures and patterns generally arise from varying shrinkage processes during cooling, resulting in unique fine or microstructures. The non-homogeneity of the molten granules also contributes to the different optical properties of the surface. Furthermore, the uneven distribution of fiber, color pigment, or solid content creates distinct patterns that uniquely identify each component. This results in a random texture or pattern that, like a fingerprint, uniquely identifies the corresponding component.

[0026] This utilizes the above method for the markless identification of components manufactured using a molding process. The method involves a two-step process: first, using initial pairwise image comparisons, it determines which group of components manufactured in the same molding unit, or in which specific molding unit, the component was manufactured; and then, using second pairwise image comparisons, the component is uniquely identified. This results in fewer pairwise image comparisons overall, saving computational effort, time, and energy. For this purpose, a reference image of a component surface is acquired after the component is manufactured using the molding process.From the reference image, characteristic structural features can be captured and stored, and an individual molding device identification number and a component identification number can be assigned to the component, which are stored together with the structural features.

[0027] Immediately after the impression process, reference images of the components are created and analyzed through pairwise image comparisons to identify similar reference images and assign them to a group. This group corresponds to components manufactured in the same impression system. It is not necessary to determine which impression system produced the component in question, as this information is known immediately after the impression process. Each reference image of the component is assigned an impression system identification code corresponding to its group. This identification code then simply identifies the group of components. The assignment of reference images to groups can be achieved, for example, using a clustering method.Pairwise image comparisons can be based on a known method for comparing images, each yielding a similarity value that indicates the degree of agreement between the images. For example, the similarity value can be determined by the height of the global maximum in a matching matrix according to the teaching of DE 10 2019 210 580 A1, preferably after applying a smoothing filter. Overall, the clustering method serves to assign components manufactured in the same molding device to one another.

[0028] Subsequently, one candidate from each group is selected from the reference images or determined in some other way.

[0029] The candidate can be selected from the reference images using random selection. This means that a reference image can be chosen arbitrarily from the respective group.

[0030] Furthermore, the reference image that yields the highest or lowest similarity score when compared to all other reference images within the group under consideration can be selected as a candidate. This means that the candidate selected in this way is a good representative for the group for the initial comparison because its similarity to every element in this group corresponds at least to this similarity value.

[0031] Furthermore, a modified reference image can be generated from the reference images through preprocessing to best represent the group of reference images. This preprocessing is specifically designed to reduce the similarity measures within the respective group for non-identical components. Alternatively, a modified reference image can be generated from several reference images of a group through combination or image processing to serve as a substitute image, thus providing a candidate for the group that best represents it.

[0032] For recognition and later identification of the component, it is captured again using a camera, and a query image is generated. Structural features are determined from this query image. These features are then compared to the structural features of the candidate component in the reference image. If the compared images are highly similar, the component can be uniquely identified.

[0033] The image acquisitions can correspond to an RGB image, an infrared image, a polarization image (i.e., an image captured with polarized light and / or a polarizing filter), an image captured with cross-polarization, a depth image (structure image) or albedo image (reflection-free image) generated using shape-from-shading, a depth image generated with a light field camera, a grayscale image, a monochromatic image, and the like. Suitable illumination sources include white light, colored light, narrowband light, monochromatic light, or polarized light, as well as combinations thereof. The light can be incident from above, at an angle, or at a very shallow angle, for example, to highlight characteristic irregularities. Multiple images acquired under different illuminations can also be processed and combined, for example, to create a depth image.Such a depth image reproduces the surface structure, which can be characteristic, especially of the molding device that formed the component.

[0034] For example, a component can be manufactured in 20 different plastic injection molds, all designed to produce the same type of component. The components are transported or temporarily stored in bulk, e.g., in a batch of 1,000 components. With 1,000 components per batch, each of the 20 injection molds has produced an average of 50 components. In the two-stage identification process, 20 comparisons are initially required per component query to identify the mold in which the component was manufactured. Subsequently, an average of 50 comparisons of images are needed to uniquely identify the component. This reduces the number of comparisons to an average of 70, compared to 1,000 comparisons in the single-stage process.Furthermore, the models for identifying the assigned impression device and the model for identifying and distinguishing different components produced in a specific impression device can be significantly less complex, so that the evaluations of the query image recordings can be carried out considerably faster.

[0035] If necessary, measures are required to better highlight texture differences when capturing images for comparison. For example, the principle of cross-polarization can be used, employing polarized illumination and a polarizing filter in front of the camera, oriented perpendicular to the polarization direction of the illumination. This eliminates specular reflections (which retain their polarization direction) and allows only scattered light (which loses its polarization) to pass through. Unwanted reflections, such as those on plastic surfaces, are thus very effectively suppressed. Furthermore, narrowband or monochromatic light and / or a bandpass filter can be used in front of the camera. Additionally, dyeing and high-quality spectral imaging can selectively reveal specific textures. By using oblique illumination for the images, surface irregularities of the component can be better highlighted.

[0036] If a batch of already registered components is available, a query image is first created for the component to be identified. The query image is acquired under the same or comparable conditions and acquisition parameters as those used for the reference image. Subsequently, the query image can be evaluated by comparing it to the selected candidates from the reference images to identify the group of reference images to which the query image is most similar.

[0037] Based on the result of the group determination, a second pairwise image comparison between the query image and the reference images assigned to the determined group can finally identify the reference image that shows the greatest similarity to the query image.

[0038] Furthermore, after the production of at least one additional component, a new reference image can be created and each of these assigned to a group, e.g. using pairwise image comparisons and a clustering method.

[0039] After further processing of a component in the product, the associated reference image can be removed from its assigned group of reference images. Removing unnecessary reference images prevents the number of reference images from growing unnecessarily and avoids unnecessary comparisons. Furthermore, the influence of the reference image on subsequent calculations in which it was used should be eliminated (e.g., by including the negatively weighted reference image in the averaging). A device for carrying out the above procedure is also provided.

[0040] Brief description of the drawings

[0041] The embodiments are explained in more detail below with reference to the accompanying drawings. These show:

[0042] Figure 1 shows a schematic representation of a production line with a molding system having multiple molding devices for the production of components used to manufacture a product; and

[0043] Figure 2 shows a schematic representation of a production line of an alternative embodiment; and

[0044] Figure 3 is a flowchart illustrating a process for identifying a component from a lot of unsorted components.

[0045] Description of embodiments

[0046] Figure 1 schematically shows a manufacturing system 1 for producing a product with components.

[0047] The manufacturing system 1 comprises a number of multiple molding devices 2, each of which may include an injection mold, a deep casting mold or the like, to produce identical components 3.

[0048] The components 3 are further processed into products 5 in a manufacturing facility 4.

[0049] To identify the individual components 3, and in particular to track components 3 for quality control purposes, a component identification number for the component 3 used in the product 5 must be assigned to it. This component identification number is linked to the component's production batch, the production time, the molding device 2, the material used, and other manufacturing parameters in the molding device 2.

[0050] Due to their size, the components 3 are often transported as bulk material or unsorted in a container 8 to the next production step of the production facility 4. In this case, the component identification information is no longer present or possible to be assigned to the respective components 3. To manufacture product 5, a random component 3 is taken from the quantity of components 3 in container 8 and used for the production of product 5.

[0051] A reference camera 6 is provided to take a reference image after the manufacturing of component 3. Furthermore, a query camera 7 is provided to capture a reference image after the component 3 has been removed from the container 8 and before it is installed in the product 5, in order to obtain the component identification information.

[0052] The reference camera 6 and the query camera 7 are connected to a manufacturing control unit 9, which controls the manufacturing process and assigns the component identification information of the component 3 used to the manufactured product 5. The manufacturing control unit 9 further assigns the component identification information to the component based on the reference images and the query images.

[0053] Figure 2 shows an alternative that differs from Figure 1 in that each impression device 2 is assigned a separate reference camera 6. This means that for each reference image acquisition, it is known in advance from which impression device 2 the component 3 captured by the reference camera 6 originates. Therefore, this information does not need to be determined from the image. This avoids assignment errors and reduces computational effort.

[0054] Furthermore, the fixed assignment of a reference camera 6 to the respective impression device 2 makes it easy to ensure that the sequence of reference images from this reference camera 6 corresponds to the sequence of components produced by this impression device 2, or that the production time stamp can be recorded. This results in advantages in terms of complete traceability of the produced components.

[0055] Figure 3 shows a flowchart illustrating a procedure for determining the component identification information for the component 3 that has been removed, is to be identified or is to be inserted into the product 5.

[0056] The procedure is carried out in conjunction with cameras 6, 7 in the manufacturing control unit 9, which is connected to the cameras.

[0057] First, in step S1, reference images of the manufactured components 3 are captured after production in the respective impression device 2. Ideally, at least one manufactured component 3 of the selected quantity should be available for each impression device 2 used in production.

[0058] In step S2, the similarities between the reference images are determined pairwise, and all components are thus divided into groups. Each group represents a common impression fixture from which all components of the identified group originate. This is done by evaluating the reference image and determining similarity measures. To assign a component to a group, a threshold comparison of the similarity measure with a predefined threshold value can be performed. If the similarity measure exceeds a predefined threshold value, it is assumed that the pair of reference images and assigned components originates from the same impression fixture 2. To increase the reliability of the method, it is advisable to compare each component with every other component and evaluate the result in a known manner using a similarity matrix of the similarity measures.

[0059] Step S2 is simplified if the manufacturing system 1 is set up as shown in Figure 2, because then the assignment of a component to a group is already known. Each impression device 2 is permanently assigned a reference camera 6. The information about which impression device 2 the produced component 3 originates from is therefore known and is assigned to the corresponding reference image.

[0060] In another embodiment, each impression device 2 can leave an individual distinguishing mark (stamp) on the produced component 3, for example, a sequential number for numbering the impression devices 2 or a corresponding code. This distinguishing mark is positioned and designed so that it is included in the reference image and easily evaluated. Subsequently, this distinguishing mark is evaluated, and thus the membership in the respective group is easily determined.

[0061] In step S3, a candidate reference image is selected from each group for the initial comparison. Several alternatives exist for this selection. A reference image can be chosen arbitrarily, for example, the first reference image from the respective group. Alternatively, the reference image that yields the highest minimum similarity score when compared with all other reference images within the group under consideration can be selected. This means that the candidate for the initial comparison is a good representative because its similarity to every element in this group is at least equal to this similarity score.

[0062] Alternatively, a candidate image for the group's reference images can be generated from several reference images by combining them. This candidate image serves as a comparison image, thus providing a candidate that best represents the group. In the simplest case, this can be achieved by ordinary or continuous averaging of the group's reference images. If the reference images are misaligned (translational, rotational, affine, or projective distortion), it is advantageous to compensate for or correct the misalignment before averaging and to first align the images uniformly. Alternatively, several candidates per group can be selected that exhibit below-average similarity to each other with respect to all reference images in the group.Depending on the group, the candidate for the initial comparison can always be the same, or it can be changed regularly or as needed. A change can be particularly useful once the component assigned to the reference image is no longer available because it has already left manufacturing system 1.

[0063] If a reference image is removed from the respective group, it should also be removed from the averaging of the reference images, if necessary.

[0064] In step S4, component 3 is removed from container 8 and a query image capture is created.

[0065] In step S5, the query image is compared pairwise with all previously selected candidates from the groups. The component 3 associated with the query image is then assigned to the group of components 3 where the comparison with the respective candidate shows the highest similarity. This determines which group the component under consideration, to which the query image belongs, is assigned, thus limiting the number of further image comparisons accordingly.

[0066] In step S6, the query image is compared with all reference images from the group identified in step S5. Component 3 is then identified as the component whose reference image shows the highest pairwise similarity to the query image. This completes the identification process for the query image.

[0067] In step S1, a new component is typically added to the manufacturing system for each production cycle, along with a new reference image. Simultaneously, on average, component 3 leaves the manufacturing system with the finished product 5. Therefore, the associated reference image is no longer needed and can be deleted from the relevant group.

[0068] In step S2, if this assignment is not already known (see embodiment of Figure 2), it is determined to which of the existing groups the new reference image best fits in terms of similarity and is assigned to this group accordingly.

[0069] The main advantage of the method described so far is the saving of computational effort or a saving of time, because on average fewer image comparisons need to be carried out in the two-stage comparison according to steps S4 and S5 than in the one-stage complete pairwise image comparison.

[0070] Furthermore, it may be possible to preprocess the images, resulting in the similarity measures within each group becoming smaller for non-identical components or larger for identical components. Various measures can achieve this, for example, by further processing the reference images of a group.

[0071] This further processing aims to reduce the similar components, i.e., those features of the image that the group has in common, and / or to increase the dissimilar components. The similar components can be eliminated, for example, by subtracting the substitute image described in step S3 pixel by pixel from the respective reference image, or by weighting the subtraction for all reference images within the respective group. The resulting modified reference images can then be used for comparison in step S4.

[0072] Additionally or alternatively, the query image captures can be pre-processed in a suitable manner to obtain modified query image captures that can be used to determine the similarity measures in the second pairwise image comparison.

Claims

Claims 1. Method, in particular a method that is at least partially computer-implemented, for identifying components from a molding process in the manufacture of a product comprising the component (3), comprising the following steps: Providing (S1) reference images of components (3) produced in several molding devices (2), wherein each reference image is assigned a component identification identifier, Assigning (S2) the reference images to a respective group of components (3); Providing or selecting (S3) one candidate for each group of reference images, the candidate corresponding to one comparison image; Providing a query image of a component to be identified (3); Performing (S5) initial pairwise image comparisons of the query image with the candidates for each group of reference images in order to identify the group of reference images whose candidate is most similar to the query image; Perform (S6) second pairwise image comparisons of the query image acquisition with the reference images of the identified group of reference images in order to determine the most similar reference image acquisition and to determine the component identification information assigned to the identified reference image acquisition.

2. The method of claim 1, wherein the assignment of the reference images to a respective group of components (3) is carried out using pairwise image comparisons that evaluate a similarity between each pair of reference images, wherein the reference images are assigned to a respective group using a clustering method. will be those whose similarities to each other exceed a predetermined similarity threshold.

3. Method according to claim 1 or 2, wherein the selection of the candidate from the reference image recordings is carried out by random selection.

4. Method according to claim 1 or 2, wherein the candidate reference image is selected which, when compared with all other reference images within the group under consideration, results in the greatest or greatest minimum similarity measure.

5. Method according to claim 1 or 2, wherein a modified reference image is generated as a candidate from the reference images by preprocessing, which represents the group of reference images in the best possible way, wherein the preprocessing is in particular designed such that similarity measures of a similarity between the reference images within the respective group and the modified reference image are minimized.

6. Method according to any one of claims 1 to 5, wherein the image acquisitions correspond to an RGB image, an infrared image, a polarization image, an image with polarized light and / or with a polarization filter, an image acquired with cross-polarization, a depth image generated with shape from shading or an albedo image, a depth image generated with a light field camera, a grayscale image or a monochromatic image.

7. Method according to one of claims 1 to 6, wherein after manufacturing at least one further component (3) a new reference image is created and, if the assigned group is known, this is directly assigned to the group or, using pairwise image comparisons and a clustering method, the newly created reference images are each assigned to a group.

8. Method according to any one of claims 1 to 7, wherein after further processing of a component (3) in the product the associated reference- The image is removed from the assigned group of reference images.

9. Device for carrying out one of the methods according to one of the claims 1 to 8.

10. Computer program product comprising instructions which, when the program is executed by at least one data processing device, cause it to perform the steps of the method according to any one of claims 1 to 8.

11. Machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause it to perform the steps of the method according to any one of claims 1 to 8.

Images