Method for manufacturing a component, camera assembly

Abstract

The invention relates to a method for producing a component (11), wherein, in order to optically recognize the component (11), the following steps are carried out: a) creation of at least three, preferably four, photographic recordings (1) of the component (11) by means of a camera device (10), during which the component (11) is illuminated by at least one light source (12) and the direction of illumination (14) is varied for each photographic recording (1); b) reading out of two-dimensional and / or three-dimensional recognition features from the created photographic recordings (1) of the component (11) and entering of the recognition features into a database; c) re-execution of step a) in the further course of production of the component (11) and re-reading out of the two-dimensional and / or three-dimensional recognition features of the component (11); and d) assignment of the component (11) to the database entry to which the component belongs by comparison of the recognition features with the database. The invention also relates to a camera device assembly (15) for carrying out the method.

Description

Technical Field

[0001] This invention relates to a method for manufacturing a component, wherein the component is tracked during the manufacturing process. Furthermore, this invention relates to a camera device assembly. Background Technology

[0002] With the concept of Industry 4.0, industrial manufacturing processes are becoming increasingly digitalized. A particularly advantageous configuration lies in the personalization of similar components through post-processing and quality management. Many Industry 4.0 solutions require complete information on the location and movement of all workpieces. Particularly important is continuous, real-time awareness of where specific components or goods are located and what their respective work progresses. This presupposes the ability to individually identify goods or components. To meet this requirement, various tracking and identification schemes exist, such as assigning and reading individual serial numbers, integrating RFID chips into components, and also including non-invasive identification of component characteristics using imaging methods.

[0003] The latter approach typically proceeds as follows: Images of the component are captured at a first point in time, usually using a camera. These images are preprocessed using computer-aided image processing and recognition techniques and then registered, wholly or partially, in a database. Later in the manufacturing process, when component identification is required, the images are regenerated and preprocessed again using computer-aided image processing and recognition techniques. The images are then compared with database entries for consistency. Typically, a corresponding component is assigned to a database entry to which the corresponding images are completely or at least largely consistent.

[0004] However, this tracking method requires that the component not undergo significant optical changes between its registration in the database and its identification at a later point in time. Optical changes include, for example, changes in color, surface texture, and others. These optical changes also occur particularly during the heat treatment process. As a result, the consistency between the photographic record created for identification and its corresponding database entry decreases until the crucial identifying features used to distinguish it from other database entries no longer exist or can no longer be identified. Consequently, tracking suffers a significant loss in reliability. Summary of the Invention

[0005] Therefore, the objective of this invention is to ensure reliable tracking of components even when surface properties, for example, have changed due to heat treatment to the point that reliable tracking is no longer possible. To achieve this objective, a method having the features described in claim 1 is proposed. Advantageous improvements to the invention can be derived from the dependent claims. Furthermore, a camera device assembly for performing the method according to the invention is provided.

[0006] A method for manufacturing a component is proposed, wherein the following steps are performed in order to perform optical identification of the component: a) At least three, preferably four, video recordings of the component are created using a camera device, during which the component is illuminated by at least one light source and the direction of illumination changes with each video recording; b) Read out two-dimensional and / or three-dimensional identification features from the captured images created by the component, and enter the identification features into the database; c) In the further process of manufacturing the component (11), step a) is repeated, and the two-dimensional and / or three-dimensional identification features of the component are reread; and d) By comparing the identification features with the database, the component is assigned to the database entry to which the component belongs.

[0007] The proposed method enables component identification within a set of components of the same type, thereby facilitating tracking during manufacturing. Preferably, identification and tracking are based on two-dimensional and three-dimensional identification features. This allows for reliable identification even after different processing steps that alter the appearance of the components. Here, even when the three-dimensional properties of the component change, for example, due to the removal of small surface scratches through polishing and the loss of some three-dimensional identification features, two-dimensional features, such as surface shading, can still be reliably used. Conversely, three-dimensional identification features can still be reliably used when, for example, the characteristic color pattern on the surface changes due to heat treatment. Here, illumination from different sides produces regional brightness differences in the captured images. These differences contain three-dimensional features that can be derived using different methods.

[0008] Its basis is Lambert's law of reflection. Therefore, the surface brightness captured by the camera device... Depends on surface color Irradiation intensity and surface normal vector With the direction vector of illumination The scalar product, where the surface normal vector and the illumination direction vector are typically unit vectors: .

[0009] The scalar product is derived from the surface normal vector. With the direction vector of illumination Angle between Determine as follows: .

[0010] From this relationship, we can see that surface brightness... The surface normal vector is at its maximum when illuminated perpendicularly and decreases with increasing tangential illumination. Therefore, based on the surface brightness captured under different known illumination directions, three-dimensional properties, especially surface orientation, can be inferred. The possible orientation of the surface normal vector is defined by changing the illumination direction. In particular, the orientation of the surface normal vector can be uniquely determined in this way by superimposing multiple images from different illumination directions. Thus, important three-dimensional features of the component can be detected.

[0011] Furthermore, it is proposed that in step a), the at least one light source and / or the component be moved to change the illumination direction. The advantage of this preferred embodiment is that the illumination direction can be dynamically adjusted according to the component to be identified. For this purpose, the light source can be moved accordingly. However, this variant is relatively time-consuming. In contrast, the second variant changes the illumination direction by rotating the component. With this variant, the required imaging records can be created quickly and continuously, saving time. However, both variants utilize mechanical mechanisms that are prone to failure.

[0012] Therefore, it is proposed that in step a), multiple light sources be used to illuminate the component, the multiple light sources being arranged around the imaging axis of the imaging device, preferably at the same angular spacing, and being switched on sequentially. Compared with using a single light source or moving the component, this preferred embodiment retains the advantage of saving time, while eliminating the need for complex and potentially faulty mechanical mechanisms.

[0013] Further, in step a), the at least one light source is positioned relative to the imaging device such that the illumination direction forms an acute angle with the imaging axis of the imaging device, preferably 30° to 60°. This preferred embodiment ensures that the member is illuminated not parallel to the imaging axis. If the illumination direction is not parallel to the imaging axis, more pronounced shadows are cast on the three-dimensional topography of the member facing the imaging device, such as peaks and valleys, while illumination directions parallel to the imaging axis do not produce such shadows. Analyzing and evaluating the corresponding shadows from different illumination directions allows for the inference of the three-dimensional topography causing these shadows.

[0014] In an extended embodiment of the invention, it is proposed that in step a), the image recording is created using a telecentric imaging device and the component is illuminated by a telecentric light source, wherein three illumination directions among different illumination directions of the component are selected in a linearly independent manner. Through this preferred embodiment, photometric stereotechnology can be applied to the component, using this technology to perform photometric measurements of the three-dimensional properties of the component. Here, an associated surface normal vector is estimated for each pixel.

[0015] The analysis and evaluation of different shooting records are based on the aforementioned Lambert reflection principle; however, this analysis and evaluation is a tensor algebra problem and is performed pixel-by-pixel in practice. Here, the light intensity vector... This includes all light intensities measured at the examined pixels across different shooting records. The normalized illumination direction vector is recorded in the illumination direction matrix. In the context, the illumination direction matrix has dimension 1. and This refers to the number of images captured from different illumination directions. Two-dimensional surface properties, such as color and matte finish, are determined by an unknown scalar. This takes into account the so-called albedo. Light intensity vector Irradiation direction matrix and albedo The surface normal vector applicable at the pixel under consideration and to be determined They are linearly related in the following ways: .

[0016] Here, the tensor operation within the parentheses is a linear mapping. Solving this equation for the albedo... With surface normal vector The product of , where the length of the vector result corresponds to the albedo. The direction corresponds to the surface normal vector at the pixel under consideration. Applying this solution to each pixel yields the albedo and surface normal vector for each pixel accordingly.

[0017] Furthermore, it is proposed that, before reading out the identification features in step a) and / or step c), a texture image and a gradient image be created based on the captured image. Here, the texture image contains two-dimensional characteristics of the component, corresponding to the reflectivity of the surface; the gradient image contains three-dimensional characteristics derived from the spatial orientation of the surface. Through this preferred embodiment, the two-dimensional and three-dimensional characteristics of the component can be used as images to apply other known image processing methods. With these image processing methods, in particular, methods for highlighting or isolating features deemed relevant and suitable as identification features can be applied. Simultaneously, unimportant or interfering features can be filtered out. In the case of the preferred application of photometric stereo, as described above, the texture image is generated by assigning a value of its associated albedo to each pixel. The result is a grayscale image, such as an 8-bit grayscale image. The gradient image contains information about the curvature of the surface at the corresponding pixel. This gradient image is created by first considering the surface normal vector determined pixel-by-pixel using photometric stereo as a vector field. This vector field is available for various mathematical operations. Preferably, the average curvature of this vector field can be determined. Furthermore, this vector field can be convolved with the derivative of a Gaussian function, thereby enhancing the morphology of three-dimensional features, such as scratches or bumps. The result is a gradient image, which graphically encodes the morphology of these three-dimensional features.

[0018] Furthermore, it is proposed that in steps a) and / or c), the identifying features are read from the texture image and the gradient image by applying a feature extraction algorithm, preferably according to Förstner's algorithm. Not all of the two-dimensional features contained in the texture image and the three-dimensional features contained in the gradient image can be meaningfully used as identifying features for identifying components. With this preferred embodiment, those two-dimensional and three-dimensional features that can be used as identifying features are extracted. These two-dimensional and three-dimensional features, as a set of points or pixels, can be used for further computation. Several different feature extraction algorithms are known, and these algorithms can be considered. These algorithms generally have a common feature: they isolate points that are significantly different from their neighboring regions. In the case of Förstner's algorithm, the non-uniformity and isotropy of each point are determined; these are purely mathematical properties. When both the non-uniformity and isotropy of a point exceed a predefined threshold, the point is classified as "significant." "How to accurately select these thresholds" depends on the requirements of the various application areas of this invention and must be determined experimentally or iteratively.

[0019] In an extended embodiment of the invention, it is proposed that, in steps a) and / or c), the identifying features are read from the texture image and the gradient image by edge detection, preferably by applying the Sobel operator. This preferred embodiment adds additional relevant points to the previously extracted identifying features. This is particularly advantageous when surface properties, such as color, have changed significantly due to steps in the manufacturing process, such as heat treatment. Different algorithms for edge detection are known and can be used. When applying the Sobel operator to a determined edge detection filter, the first derivative of the corresponding pixel brightness value is determined by convolution, while simultaneously smoothing orthogonal to the derivative direction. The result is a strong highlighting of image regions with the most prominent edges. Relevant points are extracted by applying a threshold function. The parameters of this threshold function depend on the requirements in various application areas of the invention and must be determined experimentally or iteratively.

[0020] Furthermore, it is proposed that in step d), the identification features read in steps a) and / or c) are compared with the database by point-by-point determination of correspondences, preferably by applying a resampling algorithm, such as Random Sample Consensus. Preferably, the number of consistent identification features is determined for each database entry. During the identification of components during the manufacturing process, the components often change to such an extent that they no longer match the database entry to which they belong in all identification features. With this preferred embodiment, complete consistency is not necessary. Instead, assignment is based on the entry in the database that has the most consistent features with the component to be identified. Since the component has been explicitly recorded in the database in step a), it is ensured from the outset that a corresponding database entry exists.

[0021] Furthermore, a camera device assembly is proposed. According to the invention, the camera device assembly includes at least one camera device having an imaging axis and at least one light source having an illumination direction, wherein the imaging axis forms an acute angle with the illumination direction, and this angle is preferably 30° to 60°. Using the proposed camera device assembly, photographic records of the component can be created during the manufacturing process of the component with different illumination directions. By utilizing the effect of Lambertian reflection through different illumination directions, the three-dimensional properties of the component can be inferred. By additionally utilizing various three-dimensional identification features detectable in this manner, in addition to the two-dimensional identification features obtainable using conventional image recognition methods, reliable tracking of the component can be achieved during the manufacturing process. This can also be reliably achieved when two-dimensional properties, such as surface color, change due to manufacturing steps, such as heat treatment and similar steps.

[0022] Furthermore, it is proposed that the imaging device has a telecentric objective lens, and the at least one light source is telecentric. By using a telecentric imaging device and a telecentric light source, the necessary prerequisites for applying the aforementioned photometric stereoscopic technique are met, provided the illumination direction is correctly selected. This technique allows for the measurement of the three-dimensional properties of components with high accuracy. Consequently, the quantity and quality of extractable three-dimensional identification features are improved, which enhances the reliability of component tracking.

[0023] Further improvements are proposed: to change the illumination direction, the at least one light source is movable, preferably mobile, and / or multiple light sources are arranged around the imaging axis of the imaging device. The advantage of the first variation is that the illumination direction can be dynamically adjusted according to the component to be identified. For this purpose, the light source is moved accordingly. However, this variation is time-consuming. The second variation differs in that the light source can be fixedly installed. With this variation, the required recordings can be created quickly and continuously, saving time. Furthermore, complex and potentially faulty mechanical mechanisms are not required. Attached Figure Description

[0024] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings. The drawings are as follows: Figure 1 A schematic diagram of an exemplary manufacturing process; Figure 2 : A schematic diagram of the optimal process for reading out the identified features; Figure 3 A schematic diagram illustrating the conversion of four images into texture and gradient images; Figure 4 A simplified side view of a preferred embodiment of a camera device assembly having multiple light sources according to the present invention; Figure 5 A top view of a preferred embodiment of a camera device assembly having multiple light sources according to the present invention. Detailed Implementation

[0025] Figure 1 An exemplary manufacturing process for component 11 is schematically illustrated. This manufacturing process is carried out at multiple workstations. Component 11 is produced in the first manufacturing step S0. In the subsequent method step S1, four photographic records 1 of component 11 are created using a camera device 10. Here, component 11 is illuminated each time by a different light source 12 from a different illumination direction 14. The photographic records 1 are merged and processed using photometric stereo (see also...). Figure 3This is converted into texture image 2 and gradient image 3. Here, texture image 2 holds the two-dimensional properties of component 11, and gradient image 3 holds the three-dimensional properties. Two-dimensional and three-dimensional identification features are read from texture image 2 and gradient image 3, and these identification features are registered in a database. In another method step S2, component 11 is further processed. For example, the surface of component 11 is heat-treated. In a subsequent step S3, component 11 is re-imaged by camera device 10 under different illumination directions 14, and the identification features of the component are read (see also...). Figure 2 These identification features are compared with the database, and component 11 is assigned to the database entry to which it belongs. Component 11 is thus identified. In another method step S4, component 11 is then subjected to further post-processing. Here, the tracking of component 11 according to method step S3 can be inserted at any time. In the final method step S5, the manufacturing process is completed.

[0026] Figure 2 A preferred process for tracking component 11 is shown. In the first step E1, the component is photographed by camera device 10. Here, the component is illuminated from one side by light source 12. After creating image record 1, the light source is turned off. In the second step E2, the second light source 12 illuminates component 11 from a new direction. Camera device 10 creates another image record 1 of component 11, and the second light source 12 is turned off. In the subsequent step E3, the third light source 12 illuminates component 11 from yet another new direction. Camera device 10 creates another image record 1 of component 11, and the third light source 12 is turned off. In the subsequent step E4, the fourth light source 12 illuminates component 11 from a direction deviating from all previous directions. Camera device 10 creates a fourth image record 1 of component 11, and the fourth light source 12 is turned off. In the subsequent method step E5, the four created image records 1 are converted into texture image 2 and gradient image 3 using photometric stereo. Here, texture image 2 contains the two-dimensional surface characteristics of component 11, such as surface color. The three-dimensional properties of component 11 are encoded in gradient image 3. In the subsequent step E6, the identifying features of component 11 are extracted from texture image 2 and gradient image 3. This is preferably achieved using a feature extraction algorithm, such as the one based on Förster's algorithm. In the subsequent method step E7, the identifying features extracted in E6 are compared with all database entries for consistency. Component 11 is assigned to the database entry with the most consistency.

[0027] Figure 3An illustrative example is shown where four photographic records 1 of component 11 under four different illumination directions 14 are converted into texture image 2 and gradient image 3 using a photometric stereo model. Here, texture image 2 displays the two-dimensional surface properties of component 11 in grayscale. This specifically relates to surface reflectivity, such as surface color, gloss or matte finish, and similar properties. The gradient image 3 encodes the three-dimensional properties of component 11. This includes scratches, protrusions, edges, and comparable three-dimensional morphology. The gradient image 3 is created in the photometric stereo model by calculating a surface normal vector for each pixel based on the photographic record 1, and calculating the surface curvature based on the resulting vector field, encoding this surface curvature into pixel values.

[0028] Figure 4 A side view is shown of a preferred embodiment of a camera device assembly 15 for creating a photographic record 1 according to the present invention. The camera device 10 is positioned vertically above the member 11 to be photographed. Four light sources 12, preferably telecentric projectors, are arranged to illuminate the member 11. Here, the light sources 12 are arranged such that the illumination direction 14 forms an angle of 30° to 60° with the image axis 13. .

[0029] Figure 5 Showing with Figure 4 The image shows a top view of the preferred camera assembly 15 arranged accordingly for creating the recording 1. The camera assembly 10 is located in the center. Four light sources 12 surround the camera assembly.

[0030] To create each recording 1, a light source 12 is turned on to illuminate the component 11 to be photographed. The light source 12 is then turned off, and the component 11 is then illuminated by another light source 12 and photographed again by the camera device 10. This process is performed four times, each time using a different light source 12.

Claims

1. A method for manufacturing a component (11), wherein, In order to perform optical identification of the component (11), the following steps are performed: a) At least three, preferably four, shooting records (1) of the component (11) are created by means of a camera device (10), during which the component (11) is illuminated by at least one light source (12) and the illumination direction (14) changes with each shooting record (1); b) Read out two-dimensional and / or three-dimensional identification features from the shooting record (1) created by the component (11) and enter the identification features into the database; c) Perform step a) in the further process of manufacturing the component (11) and reread the two-dimensional and / or three-dimensional identification features of the component (11); as well as d) By comparing the identification features with the database, the component (11) is assigned to the database entry to which the component belongs.

2. The method according to claim 1, characterized in that, In step a), in order to change the irradiation direction (14), the at least one light source (12) and / or the component (11) are moved.

3. The method according to claim 1 or 2, characterized in that, In step a), a plurality of light sources (12) are used to illuminate the component (11), the plurality of light sources being arranged around the imaging axis (13) of the imaging device (10), preferably arranged at the same angular spacing, and being switched on one after another.

4. The method according to any one of the preceding claims, characterized in that, In step a), the at least one light source (12) is positioned relative to the imaging device (10) such that the illumination direction (14) forms an acute angle with the imaging axis (13) of the imaging device (10). ), where the angle ( The preferred angle is 30° to 60°.

5. The method according to any one of the preceding claims, characterized in that, In step a), the shooting record (1) is created by means of a telecentric camera device (10) and the component (11) is illuminated by a telecentric light source (12), wherein three of the different illumination directions (14) of the component (11) are selected in a linearly independent manner.

6. The method according to any one of the preceding claims, characterized in that, Before reading the identification features in step a) and / or step c), a texture image (2) and a gradient image (3) are created based on the captured record (1).

7. The method according to claim 6, characterized in that, In step a) and / or step c), the identification features are read from the texture image (2) and the gradient image (3) by applying a feature extraction algorithm, preferably according to Förstner's algorithm.

8. The method according to claim 6 or 7, characterized in that, In step a) and / or step c), the identification features are read from the texture image (2) and the gradient image (3) by edge detection, preferably by applying the Sobel operator.

9. The method according to any one of the preceding claims, characterized in that, In step d), the corresponding relationships are obtained point by point, preferably by applying a resampling algorithm, such as a random sampling consensus algorithm, and the identification features read in step a) and / or step c) are compared with the database. Preferably, for each database entry, the number of consistent identification features is determined.

10. A camera device assembly (15) for performing the method according to any one of the preceding claims, characterized in that, The imaging device assembly includes at least one imaging device (10) having an imaging axis (13) and at least one light source (12) having an illumination direction (14), wherein the imaging axis (13) and the illumination direction (14) form an acute angle ( ). ), and that angle ( The preferred angle is 30° to 60°.

11. The camera device assembly (15) according to claim 10, characterized in that, The camera device (10) has a telecentric objective lens, and the at least one light source (12) is telecentric.

12. The camera device assembly (15) according to claim 10 or 11, characterized in that, In order to change the illumination direction (14), the at least one light source (12) is movable, preferably movable, and / or a plurality of light sources (12) are arranged around the imaging axis (13) of the imaging device (10).

Images