System and method for defect monitoring of a laser welding process

Abstract

The invention relates to a system for defect monitoring of a laser welding process carried out on a component processed by a laser processing system, comprising an image recording device for generating two-dimensional image data of the component and an evaluation unit for the image data, characterized in that the image recording device is configured to record the two-dimensional image data of the component, and the evaluation unit is configured to evaluate the two-dimensional image data of the component. The evaluation unit is arranged in such a way that the evaluation unit calculates an associated height value for two-dimensional image data generated by the image capturing device by means of a convolutional neural network and generates a height profile of the component.

Description

Technical Field

[0001] During laser welding, monitoring the quality of the weld seam is crucial to ensure high-quality components. Therefore, the weld seam needs to be post-processed after production, in which its quality is inferred by evaluating the image data of the weld seam. In addition, it is also possible to evaluate whether a stable process can be performed before welding based on the component configuration (Bauteilkonfiguration). This is also determined by evaluating the image data. In both applications, the height information of the component provides added value and can meet additional requirements for component configuration and quality assurance. Background Art

[0002] An example in which weld monitoring is very important is the manufacture of electric motors and generators, whose stators are wound using hairpin technology, in which adjacent hairpin ends are laser welded according to the circuit diagram to establish the electrical contact of the pins. Defective welds between the hairpin ends are reworked in a second step (if possible), because otherwise the entire stator would be unusable.

[0003] The quality of the weld seam is influenced by a number of parameters, such as material quality, beam deviation and environmental factors. Defects may occur, for example, due to excessive material discharge or insufficient penetration of the laser beam into the material of the component. When welding hairpins, defects can be detected, in particular, by excessively large or small weld beads, spatter or height misalignment of the hairpins relative to one another before the welding process. Therefore, in addition to the two-dimensional recording of the surface of the component, the height profile of the component is also important for finding and characterizing defects.

[0004] Visual quality monitoring of weld seams is usually carried out by 3D scanning or by optical coherence tomography (OCT) scanners on laser processing facilities, or by grayscale photographs obtained with the aid of cameras, which generate 2D grayscale images of the component. 3D scanning can provide very precise 3D images of the component, but requires long recording times and is therefore difficult to integrate into ongoing production processes. In the case of 2D grayscale images, the height structure of the weld seam can only be analyzed very imprecisely and / or only when multiple cameras are used.

[0005] According to DE 10 2018 129 425 A1, it is known to convert image data and height data of a workpiece into information about the processing quality by means of a transfer function, wherein the transfer function is formed by a deep convolutional network that can be adapted to changing situations, such as different workpieces, by transfer learning. The image data and height data are acquired by an OCT scanner, a stereo camera system and / or a triangulation system. Therefore, such known systems for defect monitoring are relatively complex and require a long evaluation time.

[0006] DE 10 2010 017 316 A1 describes a welding system that is equipped with a plurality of cameras that can record a component from a plurality of different points and generate a three-dimensional image of the weld bead. The height and width of the weld bead can be calculated from the image. Summary of the invention

[0007] The object of the present invention is to enable reliable and rapidly executable defect monitoring of laser welding processes.

[0008] This object is achieved by a system for defect monitoring of a laser welding process, which is carried out on a component processed by a laser processing facility, the system having an image recording device for generating two-dimensional image data of the component to be processed or processed and an evaluation unit for the image data, characterized in that the evaluation unit is arranged so that it calculates the corresponding height values ​​for the two-dimensional image data generated by the image recording device with the help of a convolutional neural network and generates a height profile of the component.

[0009] Using a convolutional neural network, preferably a deep convolutional neural network, a height profile of a component can be generated from a single two-dimensional image capture. The calculated height value is close to the actual height value of the component, wherein the deviation between the calculated height value and the actual height value can be kept at a very low level by a well-trained neural network. The calculated three-dimensional image is a component height map, from which not only the position of defects in the weld can be determined, but also the type of defect. Generating the height map requires very little computing time and can therefore be integrated into the production process of the component. For this purpose, there is no need for a time-consuming three-dimensional scan of the component. Such a scan is only required during the training of the network.

[0010] Preferably, the image recording device can generate two-dimensional image data for pixels of a pixel matrix of an image of the component recorded thereby, and the evaluation unit can calculate an associated height value for each pixel therein. The generated height profile thus has the same resolution as the two-dimensional image recording. The processing of the two-dimensional image data by the neural network thus does not lead to a loss in accuracy in the determination of defects.

[0011] The evaluation unit can generate a three-dimensional representation for at least a portion of the component surface that is relevant for defect detection. This can result in the number of pixels in the two-dimensional representation of the component being higher than the number of pixels in the three-dimensional representation. However, the resolution in the two-dimensional representation and the three-dimensional representation can be the same.

[0012] In order to obtain two-dimensional image data, the image recording device can have at least one camera device oriented coaxially with the laser processing head of the laser processing device. The camera device is preferably a grayscale camera device. However, color recording of the component is of course also possible. It is also possible to use multiple cameras, which record the component from different positions. The use of multiple cameras is particularly beneficial for the learning of convolutional neural networks and can replace the scanner in this case.

[0013] However, the system can also have a 3D scanner, in particular an OCT scanner, which generates a height profile of the component and transmits the height profile to the evaluation unit. The camera images and the height profile are used as training data for the neural network. The scans are only necessary for learning the network and are not responsible for defect monitoring. Their high accuracy can be used to train the neural network very effectively.

[0014] In addition, in production When a change occurs, a scan of the component is generated, with the aid of which the evaluation unit can retrain the convolutional neural network for calculating the associated height values ​​for the two-dimensional image data of the component. This enables the neural network to adapt to new situations by transfer learning. New production situations can be, for example, a new position of a component or a change to a new component type.

[0015] The component height profile generated by the evaluation unit can be compared with the stored three-dimensional defect representation. This allows the detected defects to be directly classified and thus the component to be evaluated accordingly.

[0016] The convolutional neural network can preferably be a modified UNet. This network has been widely recognized in image processing and especially semantic segmentation.

[0017] The invention also includes a laser processing installation having a laser processing head for producing a weld seam on a component, characterized in that it has a system according to one of claims 1 to 8 .

[0018] Furthermore, the present invention relates to a method for defect monitoring of a laser welding process, the laser welding process being performed on a component processed by a laser processing facility, the method comprising the following steps:

[0019] - Acquire two-dimensional image data of the surface of the component to be processed or processed,

[0020] - Calculate the associated height values ​​for the two-dimensional image data using a convolutional neural network,

[0021] - Generate height profiles of components,

[0022] In this case, the neural network has been previously trained with the image data and the associated height profiles for the calculation of the height values.

[0023] If further refinement of the results is required, the neural network can be trained at any time. It is also possible to adapt the convolutional neural network to changed production situations, such as different components, different component positions or changed production parameters, by transferring the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The application scenarios of the system and its results are described in detail below.

[0025] The accompanying drawings show:

[0026] Figure 1 a to Figure 1 g shows different illustrations of the hairpin after the laser welding process;

[0027] Figure 2a to Figure 2c shows a grayscale image, a 3D scan, and a 3D reconstruction of a first welded hairpin using the system of the present invention;

[0028] Figure 3a to Figure 3c A grayscale image, a 3D scan, and a 3D reconstruction performed using the system of the present invention are shown of a second welded hairpin. DETAILED DESCRIPTION

[0029] exist Figure 1 The hairpin 1 shown in a has two copper rod ends, which must be welded to each other to establish electrical contact. The hairpin 1 is part of a winding (not shown here) of the stator of an electric motor. The hairpins are often welded by laser, and the inspection of the weld quality is particularly important here, because all hairpins of the stator must be welded to each other so that electrical contact is reliably established.

[0030] exist Figure 1 In b is shown a good welding of the copper bars 10, 11 of the hairpin 1. The weld seam 12 has a smooth, slightly domed surface.

[0031] Figure 1 c to Figure 1g shows various defects that may occur when welding the hairpin 1. Figure 1 In c, hairpin 1 is not in the focus of the laser during welding. Figure 1 The weld shown in d is produced when the laser power is too low. Figure 1 In e, hairpin 1 is welded under excessive power. Figure 1 In f, the copper bars 10 and 11 are offset relative to each other, while in Figure 1 In g, the insulation of copper rods 10 and 11 is not stripped.

[0032] 2 and 3 illustrate how the weld seam 12 of the hairpin 1 can be three-dimensionally displayed and evaluated using the system and method of the present invention.

[0033] Figure 2a A two-dimensional grayscale image of a good weld 12 of a hairpin 1 is shown (see Figure 1 b). Figure 2c The weld 12 is shown by using a convolutional neural network Figure 2a A 3D illustration of the reconstruction of a grayscale image in . Convolutional neural networks are also called Convolutional Neuronal Netwerk. For this purpose, convolutional neural networks are Figure 2a For each pixel in the grayscale image, an approximation of the actual height value of the component is calculated. Figure 2c In the diagram in , the brighter the color, the larger the height value, as indicated by the scale next to Figure 2. For comparison, in Figure 2b A three-dimensional OCT scan of the same hairpin is shown in FIG. Figure 2c The computational reconstruction in does not differ much from the OCT scan and shows that the quality of the weld seam 12 can be assessed just as reliably with the system and method according to the invention as with the much more complex OCT scan.

[0034] Figure 3a A grayscale image of a defective weld is shown. Figure 3c In Figure 3a The reconstructed 3D image is obtained by Figure 2c The brighter color indicates that the weld seam is too high. From this and the still clearly discernible outlines of the two copper bars 10, 11 of the hairpin 1, it can be concluded that the hairpin in FIG. 3 was welded with too low a laser power, which is consistent with the Figure 1 The card issuance in c is similar. Here, too, Figure 3b A 3D OCT scan of the same hairpin is shown in FIG. 1 for comparison, which further demonstrates that: Figure 3c The height values ​​obtained by the convolutional neural network have been calculated very accurately.

[0035] Used to calculate the card issuance Figure 2c and Figure 3cThe convolutional neural network for 3D reconstruction shown in has been trained with the aid of 3D OCT scans of hairpins. The neural network is part of an evaluation unit (not shown) in which comparison images of defectively welded hairpins can also be stored. In this way, the evaluation unit can also Figure 2c and Figure 3c The 3D reconstruction of the hairpin is used to directly perform the evaluation and classification of the weld.

[0036] It is understood that the system and method according to the present invention can also be applied to welds of other components other than hairpins. In addition, the convolutional neural network can be modified so that other laser processing of components (such as laser cutting) can also be evaluated by three-dimensional reconstruction of the component from the grayscale image.

Claims

1. A system for defect monitoring of a laser welding process (12), the laser welding process being carried out on a component (1) to be processed or already processed by a laser processing facility, the system comprising an image recording device for generating two-dimensional image data of the component (1) and an evaluation unit for the image data, characterized in that: The evaluation unit is configured such that it calculates associated height values ​​for the two-dimensional image data generated by the image recording device using a convolutional neural network and generates a height profile of the component (1).

2. The system according to claim 1, characterized in that The image recording device generates two-dimensional image data for pixels of a pixel matrix of an image of the component (1) recorded by the image recording device, and the evaluation unit calculates an associated height value for each of the pixels.

3. The system according to claim 1 or 2, characterized in that: The image recording device has at least one camera device which is aligned coaxially to a laser processing head of the laser processing system.

4. The system according to claim 3, characterized in that The at least one camera is a grayscale camera.

5. System according to one of the preceding claims, characterized in that The system has a scanner, in particular an OCT scanner, which generates a scan of the component (1) and transmits the scan to the evaluation unit, which uses the scan to train the convolutional neural network for calculating associated height values ​​for two-dimensional image data.

6. The system according to claim 5, characterized in that In the event of a change in the production situation, the system generates a scan of the component (1), with the aid of which the evaluation unit retrains a convolutional neural network for calculating associated height values ​​for the two-dimensional image data of the component (1).

7. System according to one of the preceding claims, characterized in that The evaluation unit generates a three-dimensional representation of at least a section of the component surface.

8. System according to one of the preceding claims, characterized in that The evaluation unit performs a comparison of the generated three-dimensional representation of the component (1) with a three-dimensional defect representation stored in the evaluation unit.

9. System according to one of the preceding claims, characterized in that The convolutional neural network is a modified UNet.

10. A laser processing installation having a laser processing head for producing a weld seam on a component (1), characterized in that: The laser processing installation has a system according to one of the preceding claims.

11. A method for defect monitoring of a laser welding process, the laser welding process being performed on a component (1) to be processed or already processed by a laser processing facility, the method comprising the following steps: - Acquire two-dimensional image data of the surface of the processed component, - calculating the associated height values ​​for the two-dimensional image data using a convolutional neural network, - generating a height profile of the component (1), In this case, the neural network has already been trained beforehand using image data of the component (1) and the associated height profiles.

Images