Corrective welding method and seam processing device for corrective welding

Abstract

The present invention relates to a (first) method for corrective welding of a workpiece having at least one defective weld seam, comprising the following steps: - providing a workpiece having at least one defective weld seam; capturing image data of the workpiece using a camera device; - performing a first artificial intelligence based image analysis of the image data to identify the location of at least one weld seam; - performing a second artificial intelligence based image analysis of the image data to identify at least one defect class in the at least one weld seam; - checking whether corrective welding of the at least one weld seam is possible based on the identified seam location and the identified defect class of the at least one weld seam.

Description

[Technical Field]

[0001] The present invention relates to a method for corrective welding and a seam treatment device for corrective welding.

[0002] When welding workpieces, defects may occur in the weld seam, and these defects can be categorized into different defect classes. If defects occur in the weld seam, the workpiece can undergo corrective welding in a post-processing station. Currently, the location of the defective workpiece cannot be solved algorithmically for all defect classes, so corrective welding is performed manually by an operator. The operator also determines whether corrective welding is possible and, if so, which welding parameters and / or scanner geometry will be used for the corrective welding.

[0003] Due to their cognitive capabilities, humans are particularly well suited for "manual" or human evaluation of workpieces with respect to weld seam location and defect class. Nevertheless, human evaluation of workpieces requires training and experience, is subjective, prone to error, and is time- and resource-intensive. Therefore, "manual" evaluation is often unsuitable when reliable high throughput is required when analyzing large numbers of workpieces in real time. Automated, fast, and reliable computer-aided image recognition and analysis are desirable. [Background technology]

[0004] It is therefore an object of the present invention to provide a method for corrective welding of workpieces having at least one defective weld seam, which method is fast, reliable and / or can be at least partially automated.

[0005] Summary of the Invention The object of the present invention is achieved by a method having the features of claim 1. The method, specifically a computer-implemented method, for corrective welding of a workpiece having at least one defective weld seam comprises the following steps, specifically in the listed order: A defective workpiece having at least one defective weld seam is provided to a seam processing device. A camera device is used to capture image data of the workpiece, specifically the weld seam. Subsequently, a first artificial intelligence-based image analysis of the image data is performed to identify a seam position of the at least one weld seam. An absolute seam position or a relative seam position of the weld seam is thereby determined, the relative seam position referring to the seam processing device, specifically the camera device and / or the laser device. Additionally, a second artificial intelligence-based image analysis of the image data is performed to identify at least one defect class of the at least one weld seam. As a result, the defect class of the weld seam, i.e., the type of weld seam defect, is determined. Based on the identified seam position and / or the identified defect class of the at least one weld seam, it is determined whether corrective welding of the at least one weld seam is possible (correction suitability confirmation). It is possible that some weld defects or defect classes may not be suitable for repair welding. Suitability may depend, for example, on the capabilities and parameters of the seam processing device, particularly the laser device.

[0006] As a result, it is possible to automatically, quickly, and reliably determine whether a defective weld seam can be corrected in a practical sense. In this regard, the classification of the weld defect is deterministic and is not subject to variations due to the subjective evaluation of different operators. The subsequent determination of the welding parameters can still be performed manually.

[0007] If the check indicates that corrective welding is possible, particularly in a practical sense, it is advantageous if the welding parameters are determined based on the identified seam location and / or the identified defect class of the at least one weld seam. This makes post-processing even faster and more reliable and allows further steps of the method to be automated. In this regard, the determination of the welding parameters is based on the information, in particular the seam location and / or the defect class, determined to check whether corrective welding is possible.

[0008] The present description further includes a method for repair welding a workpiece having at least one defective weld seam, the method comprising, specifically in the listed order, the following steps: - providing a workpiece having at least one defective weld seam; capturing image data of the workpiece using a camera device; - performing a first artificial intelligence based image analysis of the image data to identify a seam location of at least one weld seam; - performing a second artificial intelligence based image analysis of the image data to identify at least one defect class in the at least one weld seam; - determining welding parameters based on the identified seam location and / or the identified defect class of the at least one weld seam.

[0009] Computer-assisted image analysis solutions include artificial intelligence (AI) methods based on machine learning (ML), specifically deep learning techniques that use artificial neural networks (ANNs). When the term "artificial intelligence" (AI) is used herein, it refers to, for example, machine learning (ML) methods, preferably deep learning. The currently most widely used variant for training AI is known as "supervised learning." This involves using training data to train an AI toward one or more specific goals or tasks. For training, an AI is thus presented with a combination of training images and corresponding results ("labels," "annotations," "ground truth") on which the AI ​​should be trained, i.e., example solutions for the tasks to be solved using images. This combination of training images and example solutions allows the AI ​​to learn a task set, verify and correct its results, and thereby undergo successful training.

[0010] For the purposes of the present invention, "defect class" is to be understood as different types of defects in the weld seam. Defects in the weld seam can be, for example: pores and bubbles, inclusions, cracks, discontinuities and defects, incomplete penetration or melting, overfilling or underfilling, distortion or deformation, discoloration and oxidation. Different welding strategies can be selected depending on the seam location and / or defect class of the weld seam defect. The welding strategy is determined, inter alia, by the welding parameters of the seam processing device, in particular the laser device.

[0011] An advantageous further development provides that the first artificial intelligence-based image analysis and / or the second artificial intelligence-based image analysis include a deep convolutional network. Convolutional neural networks (CNNs) are a class of deep feedforward neural networks used in image processing (e.g., image recognition), speech processing (e.g., speech recognition), and similar fields. A convolutional layer in a deep convolutional network applies a convolution operation to its input and passes the result to the next layer. In general, deep convolutional networks use relatively little preprocessing, which means that the network learns filters manually developed in conventional algorithms, making them more independent of existing knowledge and human effort in feature engineering.

[0012] The artificial intelligence for the first image analysis and the artificial intelligence for the second image analysis can be designed as a collaborative artificial intelligence.

[0013] To determine the seam location of the weld seam, image data, specifically one or more camera images, are preferably captured by a camera device. The AI ​​for component location detection and / or seam location detection uses a deep convolutional network to recognize the seam location of the weld seam. Here, a decision is made for each pixel in the image data regarding whether the pixel belongs to the weld seam. Depending on the workpiece, other classes (e.g., non-welded hairpin surfaces on hairpins) are considered during component location detection and forwarded as information to downstream correction suitability checks and / or parameter adjustments. Once the seam location is accurately determined, the workpiece, specifically at least one weld seam, can preferably be aligned with the laser device to ensure optimal correction welding.

[0014] The AI ​​for defect class detection, specifically the classifier, can be a deep convolutional network pre-trained with representative data for each defect class, as in the case of the AI ​​for component location detection and / or seam location detection. The trained network calculates a probability vector for each input, indicating how likely it is that the current input belongs to the respective defect class. Using this approach, new defect classes are taught through a new training process involving all the data.

[0015] Alternatively or additionally, the AI ​​can be trained for defect class detection using a few-shot / one-shot training approach. This involves prior, preferably one-time, training of the AI ​​to learn a specific metric to distinguish data. The metric can be viewed as a more abstract capability in terms of a deep convolutional network. This trained model preferably resides on a computing unit and requires only a few portions of the sample data of the defect class to perform classification. Furthermore, this approach can learn new defect classes during ongoing operation. Computationally intensive training of the network is preferably not required. By implementing the few-shot / one-shot training approach, it is also easy to introduce new defect classes and welding strategies directly onto the system.

[0016] To improve the classification, sensor data is preferably acquired using a sensor device, and the identification of the defect class is also performed based on the sensor data. The sensor device preferably performs optical coherence tomography and / or optical sectioning and / or triangulation and / or 3D reconstruction.

[0017] It is further advantageous if a seam shape and / or a center of mass of at least one weld seam is identified in the second artificial intelligence-based image analysis, and the welding parameters are determined based on the identified seam shape and / or the identified center of mass, which further increases the reliability of the method.

[0018] A further advantageous further development of the invention provides that information for determining whether corrective welding is possible and / or information regarding the welding parameters is transmitted to the laser device. It is advantageous if corrective welding is performed using the laser device with the determined welding parameters. The results of the classification, together with the parameters from the first AI, are transmitted to an algorithm that selects welding parameters based on the classification. In addition, the numerical parameters can be adapted to the respective welding situation based on sensor data, such as the shape and / or center of mass. The results of this algorithm are transmitted to the laser, and corrective welding is performed using the parameters.

[0019] Preferably, the camera device captures 2D image data. Preferably, a third artificial intelligence-based image analysis is performed in which 3D image data is generated from the 2D image data. The 2D and / or 3D image data are preferably used to identify seam locations and / or defect classes and / or welding parameters.

[0020] The object underlying the present invention is also achieved by a seam processing device having the features of claim 10. The seam processing device for corrective welding of a workpiece having at least one defective weld seam comprises a camera device for capturing image data of the workpiece, a computing unit for performing image analysis, and a laser device for corrective welding of the at least one weld seam, the computing unit being configured to: perform a first artificial intelligence-based image analysis of the image data to identify a seam location of the at least one weld seam; perform a second artificial intelligence-based image analysis of the image data to identify a defect class of the at least one weld seam; and check whether corrective welding of the at least one weld seam is possible based on the identified seam location and the defect class of the at least one weld seam.

[0021] It is advantageous if the calculation unit is further configured to determine welding parameters based on the identified seam location and defect class of the at least one weld seam, and the laser device is configured to perform corrective welding of the at least one weld seam using the determined welding parameters. The calculation unit can be further configured to control the laser device.

[0022] It is also advantageous if the seam processing device further comprises a sensor device for generating sensor data, in particular based on optical coherence tomography and / or optical sectioning and / or triangulation and / or 3D reconstruction. The computing unit is further preferably adapted to identify the defect class based on the sensor data.

[0023] Preferably, the laser device comprises a laser source for generating laser light, a processing optical system for deflecting the laser light, and a camera device, the camera device being arranged coaxially with the laser light on the processing optical system. Both a scanner (PFO) and a fixed optical system can preferably be used as the processing optical system. With fixed optical systems, an axis system must take on the task of tracking the optical system.

[0024] The setup preferably corresponds to the standard setup of Trumpf's VisionLine on a processing optical system with a laser. Optionally, VisionLine 3D can be used with an OCT sensor if the 3D information provides important information for defect classification. Alternatively, the 3D information can be calculated from the 2D images using additional AI. Here, the VisionLine camera is aimed coaxially at the component from above. A PanelPC from VisionLine can be used as the computational unit, computing both the neural network and the algorithm for determining the welding strategy. In the initial expansion phase, this strategy selection can be performed using a lookup table method (Active Process Logic, APL) in VisionLine. The results are transmitted to the laser device via the VisionLine network connection.

[0025] Further details and advantageous embodiments of the invention can be found in the following description, on the basis of which embodiments of the invention are further explained and explained. [Brief explanation of the drawings]

[0026] [Figure 1] 1 shows a schematic sequence for training a deep convolutional network for a first image analysis and / or a second image analysis. [Figure 2] 1 shows a general sequence for training an AI for a first image analysis and / or a second image analysis using a few-shot / one-shot training approach. [Figure 3] 1 shows a schematic sequence of a method for corrective welding.

[0027] The first image analysis and the second image analysis according to Fig. 3 are used to identify the seam location and / or defect class of at least one defective weld seam. The image analysis is based on an AI that can be trained according to Fig. 1 and / or Fig. 2.

[0028] The AI ​​shown in FIG. 1 is based on a deep convolutional network. Training such an AI requires data preparation 10, which involves collecting labeled or specified examples, such as weld seams on workpieces and / or weld seams with weld defects of different defect classes. To make the model more robust, additional examples can be generated, for example, by rotating, mirroring, or cropping existing samples. The dataset can be further divided into a training set, a validation set, and a test set. This is followed by determining the model architecture 12. The type and number of layers the network should have are selected, including convolutional layers, pooling layers, activation functions, and fully connected layers. Training 14 then occurs by first passing an input dataset containing workpieces and / or weld seams through the network, with all calculations performed at each layer. The results of the calculations are compared with the labeled values, i.e., the actual weld seams and / or actual defect classes, to calculate the defects. The loss quantifies how well or poorly the model fulfills its task. Finally, defects are run backward through the network to adjust the weights in each layer. After training is complete, the network is validated and tuned16. The network is tested using a validation set to demonstrate performance. Depending on the error rate on the validation set, parameters such as the learning rate, number of layers, or type of activation function can be adjusted. Finally, the network is evaluated using the test set. Various performance metrics (accuracy, F1 score, etc.) are calculated to assess the quality of the network.

[0029] The AI ​​shown in FIG. 2 is trained using a few-shot / one-shot training approach. According to step 20, data is also prepared; a dataset with far fewer examples per class is required for few-shot learning and one-shot learning. The dataset is also divided into a training set, a validation set, and a test set. According to step 22, a base model is selected that is fine-tuned for a specific task, in this case, the identification of weld seam and / or defect classes. According to step 24, an architecture that is particularly suitable for identifying weld seam and / or defect classes is selected. Meta-learning, such as Siamese or triplet networks and MAML (Model Independent Meta-Learning), is suitable for this purpose. Training according to step 26 is performed using episodic training. In each training episode, a small selection of classes and examples is randomly chosen. The model is then simply trained on this small dataset. A loss function specialized for small datasets, such as triplet loss, can be used to train the network by encouraging the network to place similar objects closer together and dissimilar objects farther apart in the feature space. The network 28 is then validated again. Finally, testing is performed according to step 30, where the network is evaluated using unseen test sets or new classes. Performance metrics are particularly important for assessing the model's ability to generalize to new, unknown data.

[0030] The AI ​​trained according to Fig. 1 and / or Fig. 2 can be used in a method for corrective welding according to Fig. 3. For this purpose, a workpiece having at least one defective weld seam is provided according to step 100. A camera device is used to capture image data of the workpiece according to 102. Subsequently, a first artificial intelligence-based image analysis, specifically trained according to the method according to Fig. 1 and / or Fig. 2, is performed according to step 104 to identify a seam location of the at least one weld seam from the image data. Additionally, according to step 106, a second artificial intelligence-based image analysis, specifically trained according to the method according to Fig. 1 and / or Fig. 2, is performed to identify the image data, specifically at least one defect class of the at least one weld seam from the same image data.

[0031] Depending on the seam location and / or defect class, step 108 checks whether corrective welding of the at least one weld seam is possible. If a positive assessment is made, welding parameters are further determined based on the identified seam location and / or the identified defect class of the at least one weld seam, according to step 110. The seam location and welding parameters are then fed to a laser device, which performs corrective welding of the defective weld seam using the welding parameters, according to step 112. Finally, the workpiece or weld seam can be run through the method again to check whether the weld defect has been corrected and / or whether an existing weld defect can be corrected again.

Claims

1. 1. A method for repair welding of a workpiece having at least one defective weld seam, comprising the steps of: - providing said workpiece having at least one defective weld seam; - capturing image data of said workpiece using a camera device; - performing a first artificial intelligence based image analysis of said image data to identify a seam location of at least one weld seam; - performing a second artificial intelligence based image analysis of said image data to identify at least one defect class in said at least one weld seam; - checking whether corrective welding of the at least one weld seam is possible based on the identified seam location and the identified defect class of the at least one weld seam.

2. If corrective welding of the at least one weld seam is possible, the method further comprises the steps of: The method of claim 1, further comprising the step of: determining welding parameters based on the identified seam location and / or the identified defect class of the at least one weld seam.

3. 3. The method of claim 1 or 2, wherein the first artificial intelligence based image analysis and / or the second artificial intelligence based image analysis comprises a deep convolutional network.

4. 4. The method of claim 1, wherein training data of individual defect classes is provided to the artificial intelligence of the second image analysis for training purposes, and / or the training data is provided to the artificial intelligence of the second image analysis in a few-shot / one-shot training approach to teach specific metrics for distinguishing image data.

5. The method according to any one of claims 1 to 4, wherein sensor data is acquired by at least one sensor device, and the identification of defect classes is also performed on the basis of said sensor data.

6. The method of claim 5 , wherein optical coherence tomography and / or light sectioning and / or triangulation and / or 3D reconstruction are performed by the sensor device.

7. 7. The method according to claim 1, wherein a seam shape and / or a center of mass of the at least one weld seam is identified in a second image analysis, and welding parameters are determined based on the identified seam shape and / or the identified center of mass.

8. 8. The method according to claim 1, wherein information for checking whether a correction welding is possible and / or information on welding parameters is transmitted to a laser device, and the correction welding is performed by the laser device using the welding parameters.

9. 9. The method of claim 1, wherein the camera device captures 2D image data, and a third artificial intelligence-based image analysis is performed to generate 3D image data from the 2D image data.

10. 1. A seam treatment device for repair welding of a workpiece having at least one defective weld seam, comprising: a camera device for capturing image data of said workpiece; a calculation unit for carrying out image analysis; a laser device for carrying out correction welding of at least one weld seam, The computing unit: a first artificial intelligence based image analysis of said image data to identify a seam location of said at least one weld seam; - a second artificial intelligence based image analysis of said image data to identify a defect class of said at least one weld seam; - checking whether corrective welding of the at least one weld seam is possible based on the identified seam location and the defect class of the at least one weld seam.

11. The computing unit: - determining welding parameters based on the identified seam location and defect class of the at least one weld seam; The seam processing device of claim 10 , wherein the laser device is configured to perform the modified welding of the at least one weld seam using the determined welding parameters.

12. The seam processing device of claim 10 or 11, further comprising a sensor device for generating sensor data, in particular based on optical coherence tomography and / or optical section and / or triangulation and / or 3D reconstruction, and wherein the calculation unit further identifies the defect class based on the sensor data.

13. the laser device a laser light source for generating laser light; a processing optical system for deflecting said laser light; A seam treatment device according to any one of claims 10 to 12, comprising a camera device, which is arranged on the treatment optical system coaxially with the laser light.

Images