Method for determining the quality of a laser beam weld

Abstract

Method for determining the quality of a laser beam weld joint (2) of at least two components (4) of a battery, in particular a lithium-ion traction battery of an electrically powered motor vehicle, - in which a micrograph (6) of the laser beam weld joint (2) is created, - in which the polished surface (6) is semantically segmented using a neural network (8), and - where a quality characteristic (10) of the laser beam weld joint (2) is determined based on the semantically segmented micrograph (6 seg ) is determined, - wherein, to determine the quality characteristic (10), a geometric size of a defect in the laser beam weld joint (2) is determined, wherein the geometric size is an area or extent of the defect, and - wherein the laser beam welding connection (2) is a connection between at least one cell contact (4b) of a battery cell and a busbar (4a), or wherein the laser beam welding connection (2) is a connection between a housing frame and a housing cover.

Description

[0001] The invention relates to a method for determining the quality of a laser beam welded joint of at least two components of a battery, which is in particular a lithium-ion traction battery of an electrically powered motor vehicle.

[0002] An electrically powered motor vehicle typically has a traction battery (high-voltage battery, HV battery) that supplies energy to an electric motor to drive the motor vehicle. An electrically powered motor vehicle is understood to mean, in particular, an electric vehicle that stores the energy required for propulsion solely in the traction battery (BEV, battery electric vehicle), an electric vehicle with a range extender (REEV, range extended electric vehicle), a hybrid vehicle (HEV, hybrid electric vehicle), a plug-in hybrid vehicle (PHEV, plug-in hybrid electric vehicle), and / or a fuel cell vehicle (FCEV, fuel cell electric vehicle), which temporarily stores the electrical energy generated by a fuel cell in the traction battery.

[0003] During the production of such a traction battery, components of the battery are joined together using a laser welding process, for example. The resulting laser beam welds are then examined, for example, as part of a quality control process. However, there are numerous evaluation criteria for the quality of laser beam welds. For example, the laser beam welds are examined with regard to their geometric characteristics, such as the extent of the laser beam weld seam, i.e., its penetration depth and band width, or with regard to a number of different defects, such as pores or cracks.

[0004] This multitude of evaluation criteria results in a comparatively complex and / or time-consuming evaluation. Typically, this evaluation is performed manually, which is relatively error-prone and therefore unreliable.

[0005] DE 10 2020 101 422 A1 discloses a method for the automated measurement and / or evaluation of weld nuggets on a component manufactured using the laser powder bed fusion process. For this purpose, digital image data of a micrograph of the component are provided, weld nuggets are identified by segmenting the image data using a neural network, and the weld nuggets are automatically evaluated using digital image processing to determine at least one weld nugget characteristic number.

[0006] DE 10 2019 212 881 A1 discloses a method for laser micro-welding two components. In this method, the two components are joined together in a joining zone using a laser beam by melting the material of the two components and subsequently solidifying the material. The two components consist at least substantially of copper. An optical analysis of the resulting microstructure is performed using a micrograph.

[0007] The invention is based on the object of providing a method for determining the quality of a laser beam weld between at least two components of a battery. In particular, this method should be capable of being carried out automatically and / or should be particularly reliable.

[0008] With regard to the method, this object is achieved according to the invention by the features of claim 1. Advantageous further developments and embodiments are the subject of the subclaims.

[0009] The method is used to determine the quality of a laser beam weld between at least two components of a battery. In other words, the method is intended and configured to determine the quality of the laser beam weld between two or more components of the battery. In particular, the battery is a lithium-ion traction battery intended for an electrically powered motor vehicle.

[0010] Each of the components is expediently formed, at least in the region of the laser beam weld, from copper, from a copper compound, for example a copper alloy, from aluminum or from an aluminum compound, for example an aluminum alloy. One of the components is a cell contact (also referred to as a "lead") of a battery cell, which cell contact is formed in particular from aluminum or copper. The further component joined to this in this case is a busbar (which is also referred to as a "busbar"), in particular made of copper or alternatively from an aluminum alloy. Alternatively, two or more such cell contacts are stacked one on top of the other and joined to the busbar by laser beam welding.Alternatively, one of the components is a housing frame and the other is a housing cover, with the housing frame and housing cover being made, for example, of aluminum or an aluminum alloy. For example, this housing is the battery housing or, alternatively, a (cell) module housing.

[0011] According to the process, a micrograph, i.e., an image of a metallographic section, is first created. This involves first separating, in particular sawing or cutting, the joined components at the point to be examined. The separation surface, in particular the separation plane, expediently runs through the laser beam weld seam of the laser beam weld. The sample thus obtained is expediently cleaned and embedded in a mass. Subsequently, the sample is ground several times at the separation surface using increasingly finer grits of the sandpaper used for this purpose and finally polished.

[0012] An image of the metallographic section thus produced is conveniently captured using a microscope and a camera. The image is preferably captured in such a way that the joined components are sharply depicted, and at least in the area of ​​the components, the image is neither underexposed nor overexposed, i.e., it contains no burnt-out areas or completely black areas.

[0013] The micrograph is then semantically segmented using a neural network, particularly a deep neural network. Semantic segmentation divides the image into regions, each of which is assigned to an object, such as one of the components of the laser beam weld or a background, and assigns a class to each region. Semantic segmentation therefore divides the micrograph into segments (regions) and classifies each segment. This segmentation is performed automatically. In summary, the micrograph is evaluated using the neural network, resulting in a semantic segmentation of the micrograph. The semantically segmented micrograph is also referred to as a classification map or mask.

[0014] According to a practical embodiment, the neural network is pre-trained using a standardized dataset. This dataset contains, for example, a large number of images, with at least one region of the image assigned to an object being classified. For example, a dataset is used that includes a large number of object categories. The components of the laser beam weld can - but do not have to - be present in the dataset. If these are not present in the dataset, the neural network pre-trained using this dataset is still suitable for semantically segmenting the micrograph image based on what it has learned ("transfer learning").

[0015] A standardized (training) dataset is one that contains a fixed, unchangeable set of labeled images. In other words, a specific, consistent version of the dataset is used. For example, a defined version of a dataset from the "ImageNet" database is used for this purpose.

[0016] Alternatively or preferably additionally, the neural network is (further) trained using a data set comprising micrographs as a pattern, wherein each of the micrographs is assigned an image as an annotation (“label”) that is divided into regions, wherein each region corresponds to an object shown in the micrograph or a background, and wherein each region is assigned a corresponding (object) class.

[0017] According to a practical embodiment, the neural network is a so-called "convolutional neural network," in particular a "deep convolutional neural network." Such neural networks are particularly suitable for image processing / image analysis.

[0018] In a preferred embodiment, the neural network comprises an input layer and an output layer. The micrograph is sent to the input layer of the neural network. In particular, each neuron of the input layer receives a pixel from a channel of the micrograph. In particular, the micrograph comprises three channels: a red channel, a green channel, and a blue channel. If necessary, the micrograph is cropped accordingly beforehand. In particular, the semantically segmented micrograph is output at the output layer. This micrograph comprises, in particular, only one channel.

[0019] In addition, the neural network expediently has at least one convolutional layer and a deconvolutional layer downstream of it.

[0020] The so-called U-Net is used as a suitable neural network.

[0021] Following the semantic segmentation of the micrograph using the neural network, a quality characteristic of the laser beam weld is determined based on the semantic segmentation of the micrograph.

[0022] According to a suitable embodiment of the method, a geometric size, in particular an area or an extension, of the laser beam weld joint, in particular a feature thereof, is determined to determine the quality feature.

[0023] According to the invention, in order to determine the quality feature, a particular geometric size, namely an area or an extent, of a defect in the laser beam weld connection, in particular a defect in the laser beam weld seam, is determined.

[0024] For this purpose, the micrograph is assigned length information about a spatial extent, which is represented by a pixel in the recorded separation plane. For example, a scale is assigned to the micrograph.

[0025] Using the length information, a length or area, i.e. the geometric size, of the feature of the laser beam weld joint or a defect thereof to be examined can be determined for the regions determined by the semantic segmentation, in particular for those regions assigned to the components or a weld seam, for example by counting the pixels. To this end, a region is expediently first selected from the micrograph based on the segmentation, for example by appropriately masking it with the assigned segmented region, and within this region a length or area of ​​the feature is determined by counting pixels arranged next to one another, in other words neighboring pixels. This determination is expediently carried out automatically. For this purpose, known software libraries such as "OpenCV" or "Scikit-image" can be used, for example.

[0026] For example, the feature to be examined is a crater or crack in the weld seam. The geometric dimension to be determined is the width or depth of the crater or crack. Alternatively, the feature could be pores in the weld seam or at an interface between the weld seam and one of the battery components. The geometric dimension to be determined is the (total) area of ​​the pores.

[0027] Furthermore, the number of cracks, craters or pores can be used as a quality feature.

[0028] Alternatively, the feature is a gap between two components of the battery in the area adjacent to the laser beam weld seam of the laser beam welded joint. The extent (height) of the gap is then determined as a geometric parameter. Alternatively, a penetration width of the laser beam weld seam or a penetration depth of the laser beam weld seam, or alternatively, a penetration depth of the laser beam weld seam, and their spatial extent are determined as a geometric parameter.

[0029] The geometric size of the respective feature is expediently compared with a specified threshold value. If this threshold value is exceeded or fallen below, or if the presence of such a defect is determined, information about this is output to a user, for example. In summary, a quantitative or qualitative evaluation of the quality of the laser beam weld can be performed based on the determined size, particularly the geometric one.

[0030] According to a useful development of the method, a further laser beam welding operation (laser beam welding process), i.e., a laser beam welding operation following the determination of the quality characteristic, which is intended to produce a laser beam weld joint similar to the type of laser beam weld joint in the examined micrograph, is adjusted depending on the quality characteristic, in particular depending on the determined (geometric) size of the laser beam weld joint, which quality characteristic results from the examination of the laser beam weld joint based on the micrograph. This adjustment is expediently only made if the determined quality characteristic does not meet a predefined quality criterion. Thus, the further laser beam welding process is adjusted, i.e., changed, if the quality of the examined laser beam weld joint is insufficient.It is advisable to then carry out the further, appropriately adjusted laser beam welding process accordingly.

[0031] For example, if the penetration width or penetration depth falls below a certain threshold, an adjustment is made. For example, the exposure time, the laser intensity or a focus diameter of the laser beam used for laser beam welding or the area of ​​impact of the laser beam on the components is adjusted, i.e. increased or decreased. The same applies if a certain threshold is exceeded due to the width or depth of a crater or crack. Furthermore, for example, a welding speed and / or a cooling time is adjusted if the total area of ​​the pores exceeds a specified threshold. Furthermore, for example, the contact pressure of the components of the battery against one another for the laser beam welding process is increased if the height of a gap between these components exceeds a specified threshold.

[0032] The advantage of the neural network and the preferably automated determination of the quality characteristic is that it simplifies the quality assessment. This method enables a time-saving and reliable, thus error-free, evaluation of the micrograph.

[0033] In the following, exemplary embodiments of the invention are explained in more detail with reference to a drawing. In the drawings: Fig. 1 shows a flow chart of a process for determining the quality of a laser beam welded joint of at least two components of a battery using a micrograph, Fig. 2a a schematic cross-section of a laser beam welded connection between two cell contacts and a busbar, Fig. 2b a semantic segmentation of the micrograph, and Fig. 2c the micrograph with geometric dimensions of the laser beam weld joint for determining a quality feature.

[0034] Corresponding parts and sizes are always provided with the same reference symbols in all figures.

[0035] In the Fig. 1 shows a flowchart representing a method for determining the quality of a laser beam weld 2 of at least two components 4 of a battery.

[0036] In a first step (I), a micrograph 6 of the laser beam weld 2 is created. This involves irreversibly sawing through the joined components at the location to be examined, with the resulting parting line passing through the laser beam weld seam of the laser beam weld. The sample thus obtained is then cleaned and embedded in a mass. Subsequently, the sample is ground several times at the parting line using increasingly finer grits of the sandpaper used for this purpose and finally polished.

[0037] In the Fig. 2a shows an exemplary micrograph 6, wherein the laser beam weld shown therein is a connection between two cell contacts 4b, made of aluminum, of a battery cell of a lithium-ion battery and a busbar 4a, made of copper. Thus, the cell contact and the busbar form the components 4 of the laser beam weld 2. According to an alternative not further illustrated, the components 4 are a housing cover, made of aluminum, for example, and all housing frames, made of aluminum, for example. The following explanations apply analogously.

[0038] In a second step II., the micrograph 6 is semantically segmented using a neural network 8. Here, the image is divided into areas (segments), with each area being assigned to an object, namely one of the components 4 of the laser beam weld or a background, and with each of the areas being assigned a class, here the classes “cell contact” S Z , “Busbar” S S , “Background” S H and “weld seam” S N The result of the semantic segmentation of the micrograph 6 of the Fig. 2a using the neural network 8 is in the Fig. 2b. The semantically segmented image is labeled 6 seg provided.

[0039] In a third step III., the semantic segmentation micrograph 6 is used to sega quality feature 10 of the laser beam weld 2 is determined. For this purpose, a region corresponding to the respective segment is first selected from the micrograph 6 based on the segmentation. Within this region, a length or area of ​​a feature of the laser beam weld in this region is automatically determined by counting pixels arranged next to one another. In this process, a length information item I associated with the micrograph is also generated. L which indicates the spatial extent, i.e. the length, of the area depicted by a pixel in the separation plane. For example, a scale is assigned to the micrograph 6, see also Fig. 2a.

[0040] In the Fig. 2c, region 16 is selected as an example, which represents the laser beam weld seam. A penetration depth dt of the laser beam weld seam into the busbar 4a, the penetration depth d bthe laser beam weld seam at the level of the interface between the two cell contacts 4b and the penetration depth d b the laser beam weld seam at the level of the interface between one of the cell contacts 4b and the busbar 4a. Furthermore, Fig. 2c, three of the pores P in the weld seam are provided with a reference symbol. According to this embodiment, the respective spatial extent of the penetration depth d is used as geometric parameters. t , and the penetration widths d b , or the total area of ​​all pores P in the weld by counting the corresponding pixels and using the length information I L Another parameter used is, for example, the number of pores P in the laser weld seam. These geometric parameters and the number of pores are used as a quality characteristic.

[0041] According to embodiments of the method not further illustrated, a width or a depth of a crack in the laser beam weld seam and / or an extent (height) of a gap between the components 4 in the area adjacent to the laser beam weld seam are additionally or alternatively determined as geometric variables.

[0042] In summary, to determine the quality feature 10, at least one size, in particular a geometric size of the laser beam weld 2 and / or a geometric size of a defect in the laser beam weld 2 is determined.

[0043] In a fourth step IV, a laser beam welding operation 14 (laser beam welding process 14) following the determination of the quality feature 10 is set depending on the quality feature 10, and thus depending on the determined geometric size of the laser beam weld, provided that the determined quality feature does not meet a predetermined quality criterion 12. In this case, a parameter of the laser beam welding process 14 is changed. Otherwise, the laser beam welding process 14 that was used to produce the laser beam weld 2 shown in the micrograph 6 is used.

[0044] For example, the welding speed is reduced and / or a cooling time is increased if the total area of ​​the pores P exceeds a specified threshold value as quality criterion 12. Furthermore, for example, the exposure time, the laser intensity or a focus diameter of the laser beam used for laser beam welding or an area of ​​influence of the laser beam on the components is changed if the penetration depth d t , or one of the penetration widths d b falls below a respective threshold as further quality criteria 12.

[0045] Additionally or alternatively, the contact pressure of the components of the battery against each other for the laser beam welding process is increased if the height of a gap between the components 4 exceeds a predetermined threshold value as a further quality criterion.

[0046] The neural network 8 is, for example, a convolutional neural network comprising an input layer, an output layer, at least one convolutional layer, and at least one deconvolutional layer. Preferably, a U-net is used as the neural network.

[0047] Suitably, the neural network comprises a plurality of convolutional layers, for example, 3, 5, 10, 25, 50, 75, 100, 125; 150, 200, 250, 300, 500, 750, or 1000 convolutional layers and / or 3, 5, 10, 25, 50, 75, 100, 125; 150, 200, 250, 300, 500, 750, or 1000 deconvolution layers, which are arranged downstream of the convolutional layer(s). For each convolution, an nxm convolution kernel is used, where n and m are natural numbers greater than 0, and preferably where n = m. For example, n and / or m is 4, 5, 6, 7, 8, 9, or 10, preferably 3. A stride in the convolution is preferably 1.

[0048] For example, pooling is performed after one convolution, after two convolutions, after three convolutions, or after four convolutions. In particular, average pooling or preferably max pooling is performed, preferably using nxn pooling with a stride of n. For example, here n is 3, 4, 5, or preferably 2. For example, the number of channels ("feature maps") is increased in each pooling step, for example, multiplied by n, preferably doubled. Preferably, the number of deconvolutions is equal to the number of pooling operations.

[0049] For example, following each convolution or convolutional layer, a ReLU activation function Φ ReLU (Rectified Linear Unit activation function) is applied, which is defined by ϕReLU(a)=max{a,0}.

[0050] For example, after each deconvolution or after each deconvolution layer, the deconvolution result is combined, e.g., by addition or concatenation, with a (possibly trimmed) result of one of the previous convolutions. For example, the first convolution uses the result of the convolution performed immediately before the last pooling operation. If necessary, the subsequent convolution uses the result of the convolution performed immediately before the second-to-last pooling operation. The subsequent convolution uses the result of the convolution performed immediately before the third-to-last pooling operation, and so on.

[0051] For example, one or two further convolutions are carried out following such a link and before each further unfolding (in an analogous manner to the convolutions described above), whereby the ReLU activation function Φ ReLU is applied. In particular, the number of channels is reduced, for example halved, with each unfolding.

[0052] For example, a 1×1 convolution is applied at the end.

[0053] The neural network 8, for example, is pre-trained using a standardized data set, whereby what has been learned from this data set is applied to the semantic segmentation of the micrograph (“transfer learning”).

[0054] Alternatively, or preferably in addition, the neural network 8 is or is trained using a data set that includes, for example, a plurality of training pairs with the micrograph 6 as input variable and with the semantic segmentation of the micrograph as the result. Such a training pair is, for example, composed of the images of the Fig. 2a and Fig. 2b formed.

[0055] The invention is not limited to the exemplary embodiments described above. Rather, other variants of the invention can also be derived therefrom by those skilled in the art without departing from the subject matter of the invention. In particular, all individual features described in connection with the exemplary embodiments can also be combined with one another in other ways without departing from the subject matter of the invention. List of reference symbols 2 laser beam welding connection 4 Components of the battery 4a Busbar 4b Cell contact 6 Micrograph 6 seg segmented micrograph 8 neural network 10 Quality features 12 Quality criteria 14 further laser beam welding processes 16 selected area d b penetration width d t penetration depth I L Length information P Pore S Z Class “Cell Contacting” S S Class “Busbar” S H Class “Background” S N Class “Weld Seam” I. Creating a micrograph II. Semantic segmentation of the micrograph III. Determination of a quality characteristic IV. Setting up a laser beam welding process

Claims

[1] Method for determining the quality of a laser beam weld (2) of at least two components (4) of a battery, in particular a lithium-ion traction battery of an electrically driven motor vehicle, - in which a micrograph (6) of the laser beam weld (2) is created, - in which the micrograph (6) is semantically segmented using a neural network (8), and - in which a quality feature (10) of the laser beam weld (2) is determined on the basis of the semantically segmented micrograph (6 seg ) is determined, - wherein, to determine the quality feature (10), a geometric size of a defect in the laser beam weld (2) is determined, wherein the geometric size is an area or an extension of the defect, and - wherein the laser beam welded connection (2) is a connection between at least one cell contact (4b) of a battery cell and a busbar (4a), or wherein the laser beam welded connection (2) is a connection between a housing frame and a housing cover. [2] Method according to claim 1, characterized by that in order to determine the quality feature (10) a, in particular geometric, size of the laser beam weld joint (2) is determined. [3] Method according to claim 1 or 2, characterized by that a further laser beam welding process (14) is set depending on the determined quality feature (10), provided that the determined quality feature (10) does not meet a predetermined quality criterion (12). [4] Method according to one of claims 1 to 3, characterized by that the neural network (8) is a convolutional neural network. [5] Method according to claim 4, characterized bythat the neural network (8) comprises an input layer, an output layer, at least one convolutional layer and at least one deconvolutional layer. [6] Method according to one of claims 1 to 5, characterized by that the neural network (8) is pre-trained using a standardized data set.

Images