Method and system for tracking location on material strip

Abstract

The invention relates to a method for tracking a position (w1, w2, w3) on a material strip (120, 122, 124) during the machining of the material strip (120, 122, 124), comprising the following steps: acquiring first measurement data (300) of a first surface region (w1) of the material strip (120), machining the material strip (122) in a machining step, -determining modified first measurement data by applying at least one processing machine learning model (172) to the first measurement data (300), and identifying the first surface area (s1) on the processed material strip (124) using the modified first measurement data.

Description

Technical Field

[0001] The present invention relates to a method and system for tracking a material strip during processing using measurement data about a surface area. Background Technology

[0002] This system is known from existing technology.

[0003] For example, the article "Integration of Traceability Systems in Battery Production" (Günther Riexinger et al., Procedia CIRP, Vol. 93, 2020, pp. 125-130, ISSN 2212-8271, https: / / doi.org / 10.1016 / j.procir.2020.04.002. (https: / / www.sciencedirect.com / science / article / pii / S221282712030531X) discloses examples of traceability concepts focusing on identification technologies in battery production. The article introduces the developed traceability concepts and examples of their implementation. For instance, examples include methods for identifying objects or production components by correspondingly tracking inherent object characteristics or applied identification markers.

[0004] The drawback of the aforementioned prior art is that the known methods still have limitations in terms of the accuracy of identifying the identification location or mark on the product within the scope of the relevant production process. This is because the production process can affect the identification location or mark, and then the location or mark no longer precisely corresponds to the originally existing identification location or mark. Summary of the Invention

[0005] Therefore, the object of the present invention is to provide a method and / or system by means of which the accuracy of identification of product identification locations and / or markings can be improved during the production of the product.

[0006] The objective is achieved by a method having the features of claim 1.

[0007] This method is implemented and constructed for tracking the position on a material strip during its processing, and includes the following steps: - Obtain first measurement data of the material strip regarding the first surface region. -Process the material strip during the processing steps. - By applying at least one processing machine learning model to the first measurement data, the modified first measurement data is determined, and the modified first measurement data is used to identify the first surface region on the processed material strip.

[0008] Herein, and particularly within the scope of this specification, the abbreviation “ML” is understood to mean “machine learning” or “Machine Learning”. Machine learning methods, ML models, the term “ML model” and / or the term “processing ML model” are further described within the scope of this specification.

[0009] This method, for example, can be implemented and constructed as a computer-implemented method. Furthermore, all method steps of this method can be implemented and constructed as computer-implemented method steps. It can also be specified that a selection of method steps can be implemented and constructed as computer-implemented method steps or computer-aided method steps.

[0010] Furthermore, the method steps mentioned do not necessarily have to be performed sequentially or in a predetermined order. For example, they can be performed partially simultaneously, or in other orders. Additionally, it can be specified that the method steps are performed separately or that other method steps are included.

[0011] The identification of surface regions after processing is improved because the first measurement data modified by the processing ML model is used to identify the surface regions of the material strip after processing in the processing step.

[0012] The aforementioned similarity comparisons, or the corresponding identification processes, more frequently produce so-called "false positives." These "false positives" are surface areas of a processed material strip that are incorrectly identified as another surface area of ​​an unprocessed material strip. The cause of such "false positives" can be, for example, the ambiguity of the visual fingerprint used in this similarity comparison, due to incomplete images (regarding size, resolution, lighting conditions, noise), and the use of relatively low identification thresholds to achieve adequate identification of the surface areas.

[0013] If modified first measurement data is used within the range of identifying the first surface area on the material strip being processed, such false alarms can be reduced or sometimes even completely avoided. Through a correspondingly implemented, constructed, and / or trained processing ML model, it is possible to consider, or at least partially consider, various variations of the material strip, such as those caused by processing steps, and / or the acquisition devices, measurement methods, and / or measurement conditions that may be used within the identified range of the first surface area on the material strip being processed.

[0014] For this purpose, the processing ML model can be constructed, designed, and trained such that applying at least one processing ML model to the first measurement data simulates, or at least partially simulates, the effects of the aforementioned factors (e.g., variations in the material strip, such as those caused by processing steps, and / or the acquisition apparatus, measurement method, and / or measurement conditions that may be used within the identified range) on the first measurement data. This simplifies the corresponding similarity comparison between the first measurement data acquired before processing and, for example, measurement data acquired within the range of the first surface area on the identified material strip being processed (e.g., other acquired measurement data or second measurement data).

[0015] Using the modified first measurement data thus obtained, the identification of surface areas after processing is improved by also improving the identification accuracy of identification parts and / or markings of the product within the production product range.

[0016] Here, the material strip can be made of any possible material, which can be shaped, manufactured, and / or processed into a material strip. For example, it can be a material accordingly implemented and constructed to unfold from and / or wind onto a roll. Typically, it is a very long and relatively thin material strip. Such a material strip can be made of, for example, plastic, metal, fabric, paper, or similar materials.

[0017] Furthermore, it can be specified that the moving material strip moves, for example, along the transport direction, particularly linearly along the transport direction. Furthermore, it can be specified that the moving material strip moves at a transport speed. This transport speed can be, for example, a constant transport speed or a variable transport speed. For this purpose, it can be specified that, for example, at least one speed measuring device is used to measure the speed of the material strip. Furthermore, the speed of the material strip can also be determined using other parameters, such as the rotational speed of the unwinding or winding rollers used for the material strip, sometimes together with the corresponding radii of the unwinding and / or winding rollers. Furthermore, other measured values ​​can also be used, by means of which the speed of the moving material strip can be determined in one or another of the above-described ways, or by means of said other measured values.

[0018] Generally, within the scope of this specification, measurement data, i.e., first and / or second measurement data, can be any type of zero-dimensional, one-dimensional, and / or multi-dimensional data structure suitable for similarity comparison and / or re-identification, such as when identifying a first surface region on a processed material strip.

[0019] In a preferred design, the measurement data can be implemented and constructed as two-dimensional image data. Here, the image data can be acquired, or has been acquired, by, an optical acquisition device, such as a camera. The optical acquisition device can be implemented and constructed to acquire measurement data within different spectral ranges, such as from the far-infrared to the ultraviolet spectral range, or sub-ranges thereof.

[0020] In addition, the corresponding one-dimensional or multi-dimensional image data may also include temperature data acquired by a thermal imager, or electrical data (such as charge, current, or electric field or field strength) or magnetic data (magnetic field or field strength) acquired by the corresponding measuring device.

[0021] With the help of appropriate measuring equipment, one-dimensional or multi-dimensional, such as two-dimensional, chemical data or surface structure data (e.g., roughness, height profile or height map or similar data) can also be acquired as measurement data.

[0022] The corresponding surface region of the material strip can be any finite two-dimensional region of the material strip. In particular, the surface region can be a continuous region or comprise multiple discontinuous sub-regions. Furthermore, the surface region can be defined, for example, by a regular geometric shape. The surface region can be, for example, rectangular, square, circular, elliptical, or have a similar shape.

[0023] The first measurement data can be acquired, for example, by means of a first acquisition device.

[0024] Furthermore, identifying the first surface region on the processed material strip can also be done using second measurement data regarding the second surface region of the processed material strip. Here, the second measurement data can be acquired using a second acquisition device, or it can also be acquired using a first acquisition device, or it has already been acquired.

[0025] Here, within the range of acquiring first measurement data or other measurement data using the first acquisition device, the same acquisition or measurement parameters and / or transport speed can be used respectively. Alternatively, different acquisition or measurement parameters and / or transport speeds can also be used.

[0026] Therefore, in the same manner, within the range of identifying the first surface area, when using the second acquisition device, it is possible to use the same acquisition parameters and / or transport speed, or it is also possible to use different acquisition parameters and / or transport speed.

[0027] Furthermore, when identifying the first surface area on the material strip being processed, the same acquisition parameters and / or transport speed can be used as when acquiring the first measurement data using the first acquisition device.

[0028] Furthermore, the method can be implemented and configured such that a first time point of acquisition (hereinafter also referred to as the first acquisition time point) is recorded within the range of acquiring the first measurement data. Additionally, it can be specified that second measurement data is acquired within the range of identifying the first surface region, and then a corresponding second time point (hereinafter also referred to as the second acquisition time point) is recorded for each acquisition of the respective second measurement data.

[0029] Furthermore, when acquiring measurement data within the scope of this specification, other acquisition parameters can be acquired and / or stored separately. Such other acquisition parameters may include, for example, the acquisition time point, acquisition date, acquisition device setting parameters, lighting settings, temperature during acquisition, acquisition angle, acquisition spacing, and / or similar acquisition parameters.

[0030] Furthermore, it can be specified here that, in identifying the first surface region, in addition to using the modified first measurement data, other measurement parameters obtained within the range in which the first measurement data was acquired are also used. Such other measurement parameters can be implemented and constructed, for example, according to this specification.

[0031] Therefore, for example, the acquisition of measurement data can typically be performed using an acquisition device and corresponding acquisition parameters. These acquisition parameters may include, for example, the transport speed of the material belt (e.g., at the time point when the first measurement data is acquired), the first time point at which the measurement data is acquired, the type and / or model of the acquisition device, additional information about the acquired measurement data (e.g., image size, data type, physical units of the acquired measurement data, etc.), setup parameters of the acquisition device, setup parameters or acquisition arrangement of the acquisition device (e.g., the angle and spacing between the acquisition device and the transport track), lighting type, lighting arrangement, lighting parameters, and / or similar acquisition parameters. The acquisition arrangement may include, for example, data describing the arrangement of the acquisition device, such as, the spacing and angle between the acquisition device and the transport track, lighting type, lighting arrangement, lighting parameters, and / or similar arrangement parameters.

[0032] Therefore, for example, first measurement data can be acquired using the first acquisition device and first acquisition parameters according to this specification.

[0033] In the same manner, when acquiring second measurement data with the aid of the second acquisition device and when identifying the first surface region using the second acquisition device, the corresponding acquisition parameters can be acquired and / or stored respectively.

[0034] The corresponding acquisition parameters can be generally acquired and / or stored when acquiring all measurement data according to this specification.

[0035] As described above, to acquire the various measurement data mentioned in the method, one or more acquisition devices can be provided, such as cameras or other sensors implemented and configured accordingly, to acquire the corresponding measurement data. Here, the acquisition of one-dimensional or multi-dimensional data structures can be performed, for example, in a single measurement step, such as in a 2D camera, or through a corresponding scan, such as by means of a point sensor or a 1D sensor. In this way, by using a 1D camera oriented perpendicular to the direction of material strip movement (e.g., a so-called line scan camera) and acquiring images while the material strip moves under the line scan camera, it is possible to generate, for example, 2D data structures, i.e., optical images.

[0036] The acquisition device can be implemented and constructed as a camera, such as a 3D, 2D, or line scan camera, an RGB or grayscale camera, an infrared or ultraviolet camera, and / or a hyperspectral camera. Furthermore, the acquisition device can be implemented and constructed as a height or surface acquisition device.

[0037] Furthermore, the acquisition device according to this specification can be implemented and constructed as, for example, a camera (e.g., a hyperspectral line scan camera or an RGB camera), an optical line sensor, an optical sensor, a magnetic sensor, a motion sensor, a position sensor, a material sensor, a touch sensor, and / or any other type of sensor.

[0038] The measurement data described in this specification can be data or data structures (e.g., databases, datasets, 1D data structures or 1D data, 2D data structures or 2D data, 3D data structures or 3D data, or higher-dimensional data or data structures) output from and / or transmitted to external devices, such as control devices, edge devices, industrial PCs (IPCs) and / or other computer devices. Such measurement data can be, for example, optical data (e.g., 0D data, 1D data, 2D data (e.g., 2D images), 3D data (e.g., 3D images)), electrical properties or information (e.g., resistance, conductivity, ...), magnetic properties or information, volume measurements, area measurements, state data (e.g., pressure, temperature, ...), motion parameters (e.g., position, velocity, flow rate or velocity, ...), or other measurement data.

[0039] The acquisition of a time point can be performed, for example, by an acquisition device or another time measuring device, such as a corresponding clock, which is synchronized with the acquisition device accordingly. The mentioned time point can be acquired and / or stored, for example, as time or other time signals, such as a counter or similar data, or can be acquired and / or stored.

[0040] The second acquisition device may correspond to the first acquisition device, or it may be the first acquisition device itself. Furthermore, the second acquisition device may have the same measurement principle as the first acquisition device. Additionally, the second acquisition device may be implemented and constructed in a manner that is the same as, similar to, or comparable to the first acquisition device, or it may differ from it. Here, the second measurement device may be implemented and constructed such that the measurement data acquired by it is comparable to the measurement data acquired by the first acquisition device, making it possible to identify the first surface region according to method step d and to identify the second surface region according to method step e. For this purpose, it may also be necessary to perform appropriate adaptation, normalization, conversion, or similar processing steps on the data after acquiring the first and / or second measurement data, and before performing or being able to perform the aforementioned surface region identification.

[0041] Within the scope of acquiring measurement data using the second acquisition device, for example, the same measurement parameters as when acquiring measurement data using the first acquisition device can be used, or other, adapted, or comparable measurement parameters can be used.

[0042] The processing steps or sub-steps within the scope of this specification can also be implemented and configured to acquire second measurement data within the range of identifying a first surface region on the material strip being processed. Here, identifying the first surface region on the material strip being processed can be performed, for example, using modified first and second measurement data. Specifically, within the range of identifying the first surface region on the material strip being processed, a comparison between the modified first and second measurement data can be performed.

[0043] For example, it can be specified that first measurement data is acquired using a first acquisition device, first acquisition parameters, and / or a first acquisition arrangement. The aforementioned second measurement data can be acquired using a second acquisition device, second acquisition parameters, and / or a second acquisition arrangement. Here, the second acquisition device can be the same as or different from the first acquisition device. The second acquisition parameters can be the same as or different from the first acquisition parameters. Furthermore, the second acquisition arrangement can be the same as or different from the first acquisition arrangement.

[0044] Then, for example, it can be specified that the processing steps or processing sub-steps within the scope of this specification include acquiring second measurement data by means of a second acquisition device different from the first acquisition device and / or by using second acquisition parameters different from the first acquisition parameters and / or by using a second acquisition arrangement different from the first acquisition arrangement, or implemented and constructed for such acquisition.

[0045] The processing steps or sub-steps described in this specification can also be any type of method or processing step for processing material strips. This could include, for example, applying material to the material strip, heating, cooling, drying, pressing / calendering, winding, unwinding, and / or printing the material strip and / or similar processing steps. The processing steps according to this specification can also include multiple of the above-described processing steps as processing sub-steps.

[0046] For example, such material strips are used in the manufacture of batteries and accumulators to produce electrodes for said batteries and accumulators. Here, for example, the electrode material is applied onto a carrier foil, dried, and then pressed (“calendered”). The electrode strip can then be divided into individual material segments for subsequent production steps.

[0047] Since this electrode strip is used in subsequent processing steps to manufacture or to manufacture batteries and / or accumulators, it is helpful, or even necessary, to track the manufacturing process of each battery or accumulator by spatially resolving the electrode strip used in the manufacturing process through different processing steps. The goal is, for example, in the case of defective batteries and / or accumulators, to be able to subsequently trace the cause of the defect. For this purpose, it is advantageous to understand the area of ​​the electrode strip used in the respective battery or accumulator and to trace it across the multiple production steps. In this way, for example, the cause of the failure can be determined, which provides the feasibility of corresponding improvements to the manufacturing process. Furthermore, with the above information, it is also possible to identify other batteries or accumulators that also have defects or may have defects due to the tracking data.

[0048] Furthermore, the processing steps and / or processing sub-steps can be any type of acquisition of measurement data regarding the surface area of ​​the material strip—for example, using a suitable acquisition device. In the context of this specification, the term "spatial resolution" is understood as information that is at least partially associated with, or has, location or orientation information in at least one spatial dimension, or is associated with or can be associated with location or orientation information. In the present case, spatial resolution can be implemented and constructed, for example, such that the measurement data is at least partially associated with or has, location or orientation information in the longitudinal direction and / or transport direction of the material strip, or that the measurement data is associated with or can be associated with such location or orientation information.

[0049] Another example of processing material strips is printing newspapers and magazines, where color is applied to the paper strip in different processing steps, and additional coatings can be applied. In manufacturing roofing strips, for example, different sealing materials are applied to the paper or fabric strips, and various other coating materials are used for reinforcement and setting.

[0050] In these cases, spatially resolved tracking of the corresponding material strips, whether later or during production, helps in quality control or fault identification and can significantly improve such measures.

[0051] The methods according to this specification can be implemented and configured, for example, such that measurement data regarding a specific surface region characterizes, for example, the appearance, structure, surface relief and / or other surface properties, surface parameters and / or material properties of that surface region. Furthermore, processing steps for processing the surface region or material strip can be implemented and configured to, for example, alter the surface with respect to the aforementioned properties. Here, such alterations can be implemented and configured to, for example, change the surface properties such that comparing measurement data of the surface region before the processing step with measurement data obtained after the processing step, and re-identifying the surface region after performing the processing step, becomes difficult, may be difficult, or even infeasible or impossible.

[0052] Therefore, surface properties can change, for example, by winding a material strip, such as by transferring or being able to transfer the surface structure of an adjacent material strip segment on a winding roller to a lower material segment (the structure, for example, can be "imprinted"). For example, when coating a material strip by pressing it with a pressure roller or when drying the material strip, the surface properties and / or structure will change or can be changed.

[0053] Furthermore, within the scope of identifying the first surface area on the processed material strip, other measurement principles, acquisition parameters, settings, data formats, and / or different measurement arrangements can be used when acquiring measurement data. Even in this case, re-identifying the surface area based on the acquired measurement data becomes difficult or even infeasible due to such changes.

[0054] If the first measurement data is acquired using a first acquisition device, and further, within the range of identifying the first surface area of ​​the processed material strip, second measurement data is acquired, for example, using a second acquisition device, for example, of the surface area of ​​the processed material strip, then further variations or differences will occur, affecting the measurement data acquired before and after processing. For example, such variations or effects can occur when there are differences between the first and second acquisition devices, differences between the measurement parameters when acquiring the first and second measurement data, and / or differences in different measurement arrangements within the range of acquiring the first and second measurement data. Here, the acquisition device, measurement parameters, and / or measurement arrangement can be implemented and constructed, for example, according to this specification.

[0055] Therefore, for example, the machining ML model can be implemented and constructed such that applying the machining ML model to the first measurement data can simulate, partially simulate, the influence of machining steps on the measurement data acquired during the identification of the first surface region or material strip, or include such simulation. Furthermore, the machining ML model can, for example, be implemented and constructed such that applying the machining ML model to the first measurement data can simulate, partially simulate, the influence of using a changed acquisition device, changed measurement parameters, and / or changed measurement arrangement on the measurement data acquired within the range of identifying the first surface region, or include such simulation.

[0056] In an advantageous design, for example, it can be specified that the machining ML model is implemented and constructed such that applying the machining ML model to the first measurement data can simulate, partially simulate, the effect of machining steps on the first surface region of the material strip, and / or simulate the effect of using a changed acquisition device, changed measurement parameters, and / or changed measurement arrangement on the measurement data acquired within the range identifying the first surface region, or include such simulation.

[0057] Applying a processing ML model to specific measurement data, including specific simulations, within the scope of this specification means that the effects mentioned by the processing ML model are only or only a part of the overall effects of the processing ML model.

[0058] Therefore, for example, the statement "applying a machining ML model to specific measurement data can simulate or partially simulate changes in the measurement data caused by the influence of specific machining steps on the surface area associated with the measurement data, or includes such simulation" can mean that applying the machining ML model to the measurement data in the manner mentioned also simulates the influence of other machining processes, settings, parameters, acquisition devices, measurement arrangements, or similar factors on the measurement data. Thus, a machining ML model can be implemented and constructed, for example, such that applying the machining ML model to specific measurement data can simulate or partially simulate changes in the measurement data caused by the influence of machining steps on the surface area associated with the measurement data and by other acquisition devices, other measurement parameters, and / or other measurement arrangements.

[0059] The above-described design, which applies a machining ML model to the first measurement data, has the following advantages: In known machining steps, the machining ML model can be trained such that the effect of the machining steps on the measurement data obtained about the surface region can be simulated or partially simulated by applying the machining ML model to the initially recorded measurement data about the surface region. Here, the measurement data can be, for example, the first measurement data according to this specification. In this way, changes in the surface region due to the machining steps can be simulated.

[0060] Furthermore, the processing ML model can simulate, or include, the use of a second acquisition device different from the first acquisition device. The processing ML model can also simulate, or include, the use of measurement parameters and / or measurement arrangements that differ from those acquired using the first acquisition device.

[0061] Within the scope of this specification, the term "measuring arrangement" is understood, for example, as the spatial arrangement and / or installation of the measuring device. Such a spatial arrangement may include, for example, the distance between the acquiring device and the material strip, and the corresponding viewing angle. Furthermore, the measuring arrangement may also include, for example, the type and arrangement of lighting devices, which are used or can be used within the range of measurement data acquired by the acquiring device.

[0062] The process ML model simulation or partial simulation of specific processing steps and / or acquisition of measurement data using specific measurement parameters and / or measurement arrangements with the aid of specific acquisition devices, or including such simulation, means that the above simulation is at least a part of the effect of the process ML model on the measurement data input into the process ML model.

[0063] Here, the processing ML model includes, but is not limited to, being able to be implemented and constructed such that applying the processing ML model to measurement data acquired about the surface region of the material strip can simulate, or at least partially simulate, changes in measurement data caused by the influence of specific processing steps on the surface region and / or changes in measurement data caused by acquiring the surface region using a second acquisition device, or includes such simulation. Here, the simulation of acquiring measurement data using the second acquisition device includes the measurement parameters and / or measurement arrangement used. Here, the measurement data, the acquisition of the measurement data, the second acquisition device, the surface region, the processing steps and / or the influence of the processing steps on the measurement data can be implemented and constructed, for example, according to this specification.

[0064] By using modified measurement data generated with the aid of a processing ML model within the range of the identified surface region (e.g., the first surface region), it is possible to improve, or sometimes even achieve for the first time, the re-identification of the mentioned surface region by comparing the measurement data obtained from the processed material strip with the mentioned modified measurement data.

[0065] For example, a processing ML model can be generated for the winding of a strip of material or the sequential winding and unwinding of a strip of material, and, for example, simulates or at least partially simulates the transfer of the structure of a region of the strip of material to an adjacent region of the strip of material on the winding roll. Furthermore, a processing ML model can also be created, for example, for the pressing or calendering of a strip of material, and, hereby, for example, through appropriate training, simulates or at least partially simulates the effect of the surface structure of a particular calender roll on the strip of material. The corresponding processing ML model can, for example, be generated or can be generated for all processing steps according to this specification and / or also correspondingly for acquiring measurement data using specific acquisition devices, specific acquisition parameters, and / or specific acquisition arrangements.

[0066] When a single processing step is associated with a processing ML model, as described above, it is possible to specify that at least one processing ML model is implemented and constructed such that applying the processing ML model to measurement data acquired about a surface region prior to the execution of the processing step can simulate, at least partially simulate, the changes in the measurement data caused by the effect of the processing step on the mentioned surface region, or include such simulation. Then, by applying the modified measurement data generated through the processing ML model to the mentioned measurement data, the identification of the surface region after the execution of the processing step can be simplified, or sometimes even made possible for the first time.

[0067] Furthermore, it is possible to specify that a processing step comprises multiple processing sub-steps, but for this purpose, only one processing ML model is created for the entire processing step consisting of multiple sub-steps. In this case, it is also possible to specify that the processing ML model simulates, or at least partially simulates, the impact of the entire processing step on the corresponding measurement data. Here, each processing sub-step can be implemented and constructed as a processing step according to this specification.

[0068] In another scenario, where a particular processing step comprises multiple processing sub-steps, it can be specified that each processing ML model in at least one processing ML model is associated with a specific processing sub-step. Each processing ML model then simulates the effect of the particular processing sub-step on the corresponding measurement data. In a processing step comprised of processing sub-steps, the modified measurement data throughout the entire processing step can then be calculated as follows: Measurement data regarding the surface area of ​​a material strip not yet processed by the processing step is acquired. The processing ML model of the first processing sub-step is then applied to the measurement data. The processing ML models of subsequent processing sub-steps are then applied to the modified measurement data generated therein to produce further modified measurement data. In this way, the effect of the entire processing step on the measurement data acquired regarding the mentioned surface area is simulated by applying the respective associated processing ML models, in the order of the executed processing sub-steps. The measurement data obtained after applying the processing ML model of the last processing sub-step of the processing step then corresponds to the modified measurement data according to this specification.

[0069] Within the scope of this specification, the term "simulation" will be understood both as a relatively good approximation or simulation, and as any simulation, even if only partially, approximately, or only fundamentally, of the impact of a processing step on the measurement data. Furthermore, the quality of such a simulation can be related to, or at least partially related to, or influenced by, the measurement data itself. Therefore, for individual measurements, applying the specific processing ML model according to this specification may result in a relatively good simulation of the processing step's impact on the measurement data, while for other measurements, applying the processing ML model may only yield a poor, or even insufficient or unusable, simulation.

[0070] The machining ML model can simulate, partially simulate the impact of machining steps on measurement data obtained about a specific surface area, or include such simulations. It can also be implemented and constructed such that the machining ML model, in addition to the said machining steps, simulates the impact of one or more other machining steps on the obtained measurement data.

[0071] Processing ML models, for example, can be implemented and constructed as ML models according to this specification.

[0072] The machining ML model, particularly the ML model according to this specification, can be implemented and constructed such that applying the machining ML model to measurement data acquired about the surface region of the material strip prior to machining through the machining steps can simulate, or at least partially simulate, the direct influence of the machining steps on the surface region and / or the changes in acquisition parameters and / or acquisition conditions on the measurement data. Here, the measurement data, the acquisition of the measurement data, the surface region, the machining steps, and / or the influence of the machining steps on the measurement data can be implemented and constructed, for example, according to this specification.

[0073] The processing ML model, particularly the ML model according to this specification, is implemented and constructed such that applying the processing ML model to first measurement data and possible other measurement data can simulate, or at least partially simulate, the changes in the first and possible other measurement data caused by the effects of the processing steps on the measurement data obtained with respect to the first and possible other associated surface areas.

[0074] In an advantageous embodiment of the invention, the processing ML model can be implemented and constructed, for example, as a neural network, particularly a deep neural network (DNN).

[0075] Here, machine learning methods are understood, for example, as automated (“machine”) methods that do not generate results through predefined rules, but rather (automatically) identify patterns from a number of examples by means of machine learning algorithms or learning methods, and then generate statements about the data to be analyzed based on those patterns.

[0076] This machine learning method can be implemented and constructed as supervised learning, semi-supervised learning, unsupervised learning, or reinforcement learning.

[0077] Examples of machine learning methods include regression algorithms (such as linear regression), the generation or optimization of decision trees (so-called "Decision Trees"), neural network learning methods, clustering methods (such as so-called "k-means-Clustering"), learning methods for generating support vector machines ("SVM"), learning methods for generating sequential decision models, or learning methods for generating Bayesian models or networks.

[0078] Therefore, an example of a machine learning method is "linear regression." Linear regression is a parametric method where the label is approximated by a weighted sum of all features. In a standard linear model variant, the mean squared error (MSE) is minimized during optimization. There are other linear model variants, which differ depending on the form of the error function. One variant is, for example, the Huber estimator, where, for example, a parameter ε is introduced to eliminate outliers in the input.

[0079] Another example of a machine learning approach is the k-nearest neighbor method. The principle of the k-nearest neighbor (k-NN) model is to determine the k closest neighbors for each input. This is a non-parametric method where the similarity criterion is a defined metric. This metric can be a norm or distance, which can be determined for all inputs. The neighborhood of the label is derived from the neighborhood or similarity of the inputs.

[0080] Decision trees are another example of machine learning (ML) models upon which machine learning methods are built. A decision tree (DT) is a hierarchical structure that enables nonparametric estimation. When processing data using a decision tree, the input is divided into local regions, the distances between which are defined by a specific metric. These local regions are the decision trees themselves.

[0081] A decision tree is a recursively partitioned sequence of decision nodes and terminal nodes or leaves. At each decision node, a discrete decision is made via a defined function (a so-called discriminant function), the result of which (yes or no) leads to subsequent nodes. If a leaf node is reached, the process ends and an output value is provided.

[0082] The result of applying such machine learning algorithms or learning methods to specific data, particularly in this specification, is referred to as a "Machine-Learning" model or ML model. Here, such an ML model is the digital storage or storable result produced by applying machine learning algorithms or learning methods to analytical data.

[0083] Here, the generation of ML models can be implemented and constructed such that new ML models are formed by applying machine learning methods, or existing ML models are modified or adapted by applying machine learning methods.

[0084] Examples of such ML models are the results of regression algorithms (e.g., linear regression), neural networks, decision trees, clustering methods (including, for example, the clusters or cluster categories obtained, definitions and / or parameters), support vector machines (SVM), sequential decision models, or Bayesian models or networks.

[0085] Furthermore, different types of ML models can be combined into a single ML model. This ensemble learning combines different ML models to achieve better reasoning. The combined ML models form what is called an ensemble. There are various methods for merging models; more precisely, they can be combined through voting, Bagging, or Boosting.

[0086] In addition, there exists so-called automated machine learning. Automated machine learning (AutoML) is a method in which, for a given task or dataset, the algorithm attempts to determine the optimal learning strategy from a specific number of machine learning methods or ML models. In AutoML, the algorithm seeks the optimal preprocessing steps and the optimal set of machine learning methods. AutoML can be combined with meta-learning. Meta-learning, also known as "learning to learn," is the process of a system observing how different machine learning methods perform on a variety of learning tasks and then learning from that experience (metadata) to learn new tasks much faster than typically can.

[0087] The AUTO-SKLEARN software library (https: / / www.automl.org / automl / auto-sklearn / ) provides a robust implementation of AutoML. The system can generate a set of up to 15 estimators. Furthermore, it supports up to 14 feature preprocessing methods and 4 dataset preprocessing methods.

[0088] Here, the neural network can be, for example, a so-called "deep neural network," "feedforward neural network," "recurrent neural network," "convolutional neural network," or "autoencoder neural network." Applying the corresponding machine learning method to the neural network is also commonly referred to as "training" the corresponding neural network.

[0089] Decision trees can be implemented and constructed, for example, as the so-called "Iterative Divider 3" (ID3), Classification or Regression Tree (CART), or the so-called "Random Forest".

[0090] Specifically within the scope of this specification, an ML model or "Machine-Learning" model is understood as the result of applying machine learning algorithms or learning methods to specific data. Here, an ML model is the digital storage or storable result produced by applying machine learning algorithms or learning methods to analytical data.

[0091] Here, the ML model can exist, be stored, or be storable, for example, in a form that allows the ML model to be applied to new data to produce results. Applying the ML model to data in this way is also called "inference". Furthermore, it can be specified that the ML model is also implemented and constructed in its current, stored, or storable form for training the ML model. However, it can also be specified that the ML model is no longer implemented and constructed in its current, stored, or storable form for training the ML model; in particular, in this case, the ML model can only be implemented and constructed for applying the data to produce results, i.e., the aforementioned "inference".

[0092] The data used to create and / or train the ML model according to the invention can be, for example, historical data of the first and / or second acquisition devices or include such data. Furthermore, the data used to create and / or train the ML model can include historical data of acquisition components with the same, similar, or comparable structure to the first and / or second acquisition components, or historical data of components of the same category or type as the first and / or second acquisition components.

[0093] When using an ML model with measurement data from other acquisition components, or when using an ML model with measurement data from other acquisition components, the creation and / or training of such an ML model can also be carried out with the help of historical data of the acquisition components, historical data of comparable acquisition components, or historical data of similar or comparable component types.

[0094] At least in the context of this specification, a neural network is understood to be, for example, an electronic device having a computer program, computer program product, or software—or the computer program, computer program product, or software itself or stored in a storage device—that comprises a network of so-called nodes, wherein typically each node is connected to multiple other nodes. Furthermore, in the context of this specification, a neural network is also understood, for example, as a computer program product or software stored in a storage device that generates such a network according to this specification when run on a computer. A node is also referred to, for example, as a neuron, unit, or cell. Here, each node has at least one input connection and one output connection. An input node of a neural network refers to a node capable of receiving signals (data, stimuli, patterns, or the like) from the external world. An output node of a neural network is understood to be a node capable of transmitting signals, data, or the like to the external world. A so-called "hidden node" is understood to be a neural network node that is neither an input node nor an output node.

[0095] Here, the neural network can be implemented and constructed, for example, as a so-called deep neural network (DNN). Such a deep neural network is a neural network in which network nodes are arranged in layers (wherein the layers themselves can be one-dimensional, two-dimensional, or higher-dimensional). Here, the deep neural network includes at least one so-called hidden layer, which consists only of nodes that are neither input nor output nodes. That is, the hidden layer is not directly connected to the input or output signals.

[0096] The term "deep learning" can be understood as, for example, a class of machine learning techniques that utilize many layers of nonlinear information processing for supervised or unsupervised feature extraction and transformation, as well as for pattern analysis and classification.

[0097] Neural networks can also have so-called autoencoder structures. Such autoencoder structures can be adapted to reduce the dimensionality of data, thereby enabling, for example, the identification of similarities and commonalities.

[0098] Neural networks can also be configured as so-called classification networks, which are particularly well-suited for classifying data into categories. Such classification networks can be used, for example, in conjunction with handwriting recognition.

[0099] Another possible structure for neural networks could be a design called a "deep-believe-network".

[0100] Neural networks can also, for example, combine multiple structures described above. Thus, for example, a neural network architecture can include an autoencoder structure to reduce the dimensionality of the input data, and this architecture can then be combined with other network structures to, for example, identify features and / or anomalies in the dimensionality-reduced data, or classify the dimensionality-reduced data.

[0101] The values ​​describing each node and its connections include other values ​​describing a specific neural network, which can be stored in a value set describing the neural network. Such a value set then represents, for example, a design scheme for the neural network. If such a value set is stored after the neural network has been trained, then, for example, the design scheme of the trained neural network is stored. Therefore, it is feasible, for example, to train the neural network in a first computer system using corresponding training data, then store the corresponding value set associated with the neural network, and transmit it to a second system as the design scheme of the trained neural network.

[0102] Furthermore, software or computer programs that implement neural networks can also be used as design schemes for neural networks. Furthermore, variations of such computer programs designed for inference and training of the neural network are possible design schemes for the neural network, as are variations of such computer programs that only implement, construct, and / or optimize inference for the neural network.

[0103] Neural networks are typically trained by inputting data into them and analyzing the corresponding output data from the network using various known learning methods to determine the parameter values ​​of individual nodes or their connections. In this way, neural networks can be trained in a known manner using known data, patterns, stimuli, or signals, so that such trained networks can then be used, for example, to analyze other data.

[0104] Generally, training a neural network is understood as follows: the data used to train the neural network is processed in the neural network by one or more training algorithms to calculate or change the so-called bias values ​​("Bias"), weight values ​​("weights"), and / or transfer functions ("TransferFunctions") of the individual nodes of the neural network or the connections between any two nodes in the neural network.

[0105] To train a neural network, for example, according to this specification, one of the so-called "supervised learning" methods can be used. Here, by training with the aid of appropriate training data, the network is trained to have results or abilities associated with said data. Furthermore, to train a neural network, so-called unsupervised learning methods can also be used. This algorithm generates a model for a given input set, which describes the input and makes predictions therefrom. Here, for example, there are clustering methods, which can divide the data into different categories when the data can be distinguished from each other by feature patterns. When training neural networks, it is also possible to combine supervised and unsupervised learning methods, for example, when one part of the data has trainable properties or capabilities, while another part of the data does not.

[0106] Furthermore, it is also possible to apply so-called reinforcement learning methods, including but not limited to those used for training neural networks.

[0107] For example, training requiring high computing power can be performed on a high-performance system, while further work or data analysis using the trained neural network can be performed entirely on a lower-performance system. Such further work and / or data analysis using the trained neural network can, for example, be performed on edge devices and / or control devices, programmable logic controllers or modular programmable logic controllers, or other appropriate devices according to this specification.

[0108] Monitoring of machine learning and / or machine learning systems works in two main phases: training and inference.

[0109] Inference is the process of generating ML model results by inputting new data into the artifact / model to produce outcomes, where the new data is typically not used for training and / or setting up the ML model. For example, machine learning inference is the ability of a machine learning system to make predictions from new data. There are three key components required for machine learning or monitoring inference: a data source, a machine learning or monitoring system for processing the data, and the data target.

[0110] Training involves the process of creating a model using machine learning algorithms. Training includes using deep learning frameworks (such as TensorFlow) and training datasets. IoT data provides a source of training data that data scientists and engineers can use to train machine learning models for a wide range of applications, from fault detection to consumer intelligence.

[0111] Reasoning involves the process of making predictions using trained machine learning algorithms. IoT data can be used as input to trained machine learning models and enable predictions that control decision-making logic on devices, at edge gateways, or other locations within the IoT system.

[0112] Trained or trained neural networks are inherently complex software. However, the overhead required to train a neural network is far greater than the overhead required to use the trained network later. Therefore, it is generally desirable to remove the "burden" of training when it is desired to use the trained neural network later.

[0113] Although this is a completely new area in computer science, there are two main approaches to taking this massive neural network and modifying it for speed and improved latency in applications.

[0114] The first approach observes the parts of the neural network that are not activated after training. These parts are unnecessary and can be "pruned." The second approach explores the possibility of merging multiple layers of the neural network into a single computational step.

[0115] Here, the modified first measurement data is used to identify the first surface region on the processed material strip, for example, to detect the similarity between the modified first measurement data and the data acquired by the second acquisition device.

[0116] For this purpose, there are numerous technically mature similarity methods for identifying similarities in 2D data structures such as images. Examples of such similarity methods are so-called fingerprinting methods, such as so-called “template matching” or “keypoint matching.” An overview of some of these methods can be found, for example, in the article “Integration of Traceability Systems in Battery Production” mentioned above (Günther Riexinger et al., Procedia CIRP, Vol. 93, 2020, pp. 125-130, ISSN 2212-8271, https: / / doi.org / 10.1016 / j.procir.2020.04.002. (https: / / www.sciencedirect.com / science / article / pii / S221282712030531X).

[0117] Therefore, for example, identifying a first surface region on a processed material strip can be implemented and configured such that a second acquisition device is used to acquire different second measurement data for different second surface regions of the processed material strip, and each of the second measurement data is compared with modified first measurement data. This comparison can be performed, for example, by means of a similarity method as described above.

[0118] Therefore, for example, identifying a first surface region on the material strip being processed can be implemented and configured such that the second surface region in the second surface region that has the highest similarity to the modified first measurement data is identified and / or distinguished as the first surface region on the material strip being processed.

[0119] In an alternative design, to identify the first surface area on the processed material strip, special markings on the material strip can be used, such as printed characters and / or patterns, edge markings, ID information (e.g., QR codes or barcodes), or similar identifying features. The first surface area can then be identified using a comparison method that matches the special markings.

[0120] Within the scope of this specification, a surface region is generally understood as a specific area on the surface of a movable or mobile material strip. Here, each surface region is associated with and / or may be associated with a physically fixed position on the moving material strip. Here, according to this specification, the names of the first and / or second surface regions are respectively associated with the corresponding first and / or second measurement data. As mentioned above, for example, a second surface region on a processed material strip can also completely or partially correspond to a first surface region on an unprocessed material strip, be a part of such a surface region, overlap with or extend beyond said surface region—or a combination of the above.

[0121] In this way, for example, it is possible to identify the position on the material strip defined by the first surface area by re-identifying the first surface area of ​​the unprocessed material strip on the processed material strip, and then track the material strip within the range of the processing steps.

[0122] In an advantageous design of this method, it is possible to specify that a first surface region of the material strip is associated or may be associated with positional coordinates on the material strip, wherein the positional coordinates are implemented and constructed such that they describe the orientation of the first surface region at least in the longitudinal direction and / or transport direction of the material strip.

[0123] In this way, according to the method of this specification, the processing of the material strip is spatially resolved by recognizing the surface area again after different processing steps within the processing range by the method, thereby enabling the processing of the material strip to be tracked specifically at the position coordinates defined by the surface area.

[0124] Correspondingly, the method explained above can also be extended to track more than one surface area. Thus, here, when the method explained in this specification is extended to acquiring measurement data of more than one surface area of ​​an unprocessed material strip and then subsequently re-identifying said surface area on a processed material strip, the currently mentioned method enables the re-identification and tracking of various precisely defined positions on a moving or movable material strip resulting from the processing steps.

[0125] Here, the method can be implemented and configured such that, in order to identify a specific surface region within the more than one surface region on the material strip being processed, modified first measurement data are generated according to this specification for each of the first measurement data acquired before material strip processing that relates to the corresponding surface region. Then, for example, by means of a similarity or fingerprinting method according to this specification, the measurement data is compared with second measurement data acquired within the range for identifying the specific surface region on the material strip being processed. If these second measurement data are identified as similar to the modified first measurement data relating to the specific surface region within the aforementioned surface region, then the surface region is defined as being identified herein.

[0126] In an advantageous design, it can be specified that when comparing the second measurement data obtained from the material strip to be processed with the modified first measurement data, only the first measurement data from the modified first measurement data that has not yet identified any surface area on the material strip to be processed is used.

[0127] In a further improvement, it is possible to specify that the identification of the first surface region and / or other first surface regions is performed using only the first measurement data of the modified first measurement data that has not yet been identified as belonging to the surface region on the material strip being processed.

[0128] The design improves the efficiency of the method and / or reduces costs because it reduces the number of comparisons to be performed.

[0129] In extensions of the described method, it is possible to extend this method to multiple processing steps, or even across multiple processing steps. Here, it is possible to specify that, after each processing step, the surface areas known before the processing step are re-identified or detected using the method according to this specification. It is also possible to specify that, only after some processing steps, the previously known surface areas are re-identified or detected using the method according to this specification.

[0130] In this way, an improved and / or simplified method is provided for position tracking or tracing of one or more processing steps or the effects of processing steps on a moving and / or movable or extended strip of material.

[0131] In this way, it is possible to track or trace an extended strip of material with relatively high spatial resolution by means of processing steps within the processing range.

[0132] In another advantageous approach, it is also possible to combine the above-described method design scheme, the extension of the method to many surface areas, and the further extension to many processing steps to achieve high spatial resolution tracking of moving or movable material strips across the entire processing chain.

[0133] In an advantageous design, it can be specified that at least one machining ML model is implemented and constructed such that at least one machining ML model is applied to a first measurement data simulation or a partial simulation of changes in the first measurement data caused by the influence of machining steps on the first surface region, or includes such a simulation.

[0134] Here, the processing of the ML model, the first measurement data of the processing steps, and the effect of the simulated or partially simulated processing steps on the measurement data can be implemented and constructed according to this specification.

[0135] Furthermore, it can be specified that at least one processing ML model is implemented and constructed as a processing ML model, which has been trained using the following method steps: - Obtain training data from the surface area of ​​the training material strip. - The training material strip or its surface area is processed through processing steps. - Obtain the modified training data for the surface region. - Train a processing ML model using the training data and the modified training data. - Store the trained processing ML model.

[0136] Here, it can be specified that the processing ML model is implemented and constructed as exactly one processing ML model.

[0137] In an advantageous design of the method for training a processing ML model according to this specification, it is possible to specify that individual identification marks on the training material strip are assigned to surface regions on the training material strip, and training data and modified training data are obtained from said surface regions.

[0138] Such identification marks can be implemented and constructed such that, even after the training material strip has been processed according to the processing steps of this specification, the identification marks can still be reliably or relatively reliably identified or detected by the corresponding acquisition and / or identification devices.

[0139] Such identification markings can be implemented and constructed as graphic symbols, QR codes, barcodes, ID information, color markings, RFID tags, magnetic materials, or similar markings, or include such markings.

[0140] Furthermore, it is possible to specify that relative orientation information about such identification marks is associated with surface areas.

[0141] In this way, the association between the training data and the modified training data can be performed more accurately, thereby improving the training process.

[0142] Here, training data can be acquired, for example, by means of a first acquisition device using first measurement parameters and a first measurement arrangement.

[0143] The modified training data can be acquired, for example, by using the first acquisition device with the first measurement parameters and the first measurement setup.

[0144] Furthermore, the modified training data can also be acquired using the first acquisition device as well as using other measurement parameters and / or other measurement arrangements.

[0145] The modified training data can also be acquired using a second acquisition device, a second measurement parameter, and / or a second measurement arrangement.

[0146] Here, the training data can be implemented and constructed as measurement data according to this specification. The acquisition of the training data can be implemented and constructed as measurement data acquisition according to this specification.

[0147] The modified training data can also be implemented and constructed as measurement data according to this specification. The acquisition of the modified training data can also be implemented and constructed as measurement data acquisition according to this specification.

[0148] Here, the training material strip can be implemented and constructed as a material strip according to this specification. The training material strip can also be any material strip within the scope of the corresponding manufacturing process. The term "training material strip" used in the context of describing the training process of a ML model is for clarity only within the scope of this specification.

[0149] Here, the machining ML model used for the machining step is applied to the measurement data obtained about the surface area of ​​the material strip before the machining step is applied, to simulate or at least partially simulate the change in measurement data caused by the effect of the machining step on the surface area, and can be implemented and constructed such that the machining ML model is trained by at least, but not limited to, the methods described above and / or similar methods.

[0150] Consider, for example, a machining ML model for a specific machining step according to this specification, and measurement data about the surface area of ​​the material strip before machining the material strip through the machining step. Applying the machining ML model to the acquired measurement data to simulate, or at least partially simulate, changes in the measurement data due to the effect of the machining step on the surface area, or including such simulation, can be implemented and constructed such that the machining ML model is trained at least by the methods described above and / or similar methods according to this specification.

[0151] The ML model can be implemented and constructed as a neural network, such as a deep neural network. In this case, the neural network can be trained or has been trained, for example, using a supervised learning method according to the present specification and the training method according to the present specification.

[0152] In the above-described case, for example, the training method according to this specification can be implemented and constructed such that training data is input into the ML model or neural network to be trained, and then the neural network to be trained outputs result data. The ML model or neural network is then trained by calculating the deviation between the modified training data and the result data, and the deviation is used to train the ML model or neural network using methods known per se for supervised learning (such as those briefly outlined elsewhere in this specification).

[0153] Furthermore, it can be specified that the processing step includes multiple processing sub-steps, and at least one processing ML model in the processing ML model is associated with each processing sub-step of the processing sub-step, wherein each processing ML model in the processing ML model is trained using the following method steps: - Obtain training data from the surface area of ​​the training material strip. -Process the training material strip or the surface area of ​​the training material strip through corresponding processing sub-steps. - Obtain the modified training data for the surface region. - Train a processing ML model associated with the processing sub-steps using the training data and the modified training data. - Store the trained processing ML model associated with the processing sub-steps.

[0154] Here, each of the processing sub-steps can be implemented and constructed as a processing step according to this specification.

[0155] The training of each processing ML model associated with each processing sub-step is then carried out by means of the training method according to this specification.

[0156] The design scheme of this method is implemented by extending the method for tracking the surface area of ​​a material strip to multiple processing steps of the material strip and / or acquisition steps of the surface area by means of multiple acquisition devices and / or acquisition parameters.

[0157] In an advantageous design, it is possible to specify the creation of multiple processing ML models for different processing steps or processing procedures according to this specification. These processing ML models can then be stored, for example, in a processing ML model database. Thus, for very specific processing and measurement data acquisition regarding the material strip, a suitable processing ML model can be selected from these processing ML models, and spatially resolved tracking can be initiated using the method according to this specification.

[0158] Here, the training material strip can be implemented and constructed as a material strip according to this specification. Also within the scope of the corresponding manufacturing process, the training material strip can be any material strip. The term "training material strip" used in the context of describing the training process of a fabricated ML model is for clarity only within the scope of this specification.

[0159] Here, the machining ML model used for the machining step is applied to simulate or at least partially simulate the changes in measurement data caused by the influence of the machining step on the surface area of ​​the material strip before the application of the machining step. It can be implemented and constructed such that the machining ML model is trained by the above method and / or similar methods.

[0160] Consider, for example, a machining ML model for a specific machining step, and measurement data about the surface area of ​​the material strip before machining the material strip through the machining step. Here, applying the machining ML model to the acquired measurement data to simulate, or at least partially simulate, the changes in measurement data caused by the effect of the machining step on the surface area can be implemented and constructed such that the machining ML model is trained by the methods described above and / or similar methods according to this specification.

[0161] Processing ML models can, for example, be implemented and constructed as neural networks, such as deep neural networks. In this case, the neural network can be trained or has been trained, for example, using the supervised learning method according to this specification or the training method described herein.

[0162] In the above context, the training method according to this specification can be implemented and constructed such that training data is input into the ML model or neural network to be trained, and then the neural network to be trained outputs result data. The ML model or neural network is then trained by calculating the deviation between the modified training data and the result data, and the deviation is used to train the ML model or neural network using methods known per se for supervised learning (such as those briefly outlined elsewhere in this specification).

[0163] The above objective is also achieved by a method for training a machining ML model, wherein the machining ML model is implemented and constructed such that the machining ML model is applied to the simulation of measurement data about a surface region of a material strip, or to the partial simulation of changes in measurement data caused by the effect of machining steps on a first surface region, or includes such simulation, the method comprising the following method steps: - Obtain training data from the surface area of ​​the training material strip. - The training material strip or its surface area is processed through processing steps. - Obtain the modified training data for the surface region. - Train a processing ML model using the training data and the modified training data. - Store the trained processing ML model.

[0164] Here, the training data can be implemented and constructed as measurement data according to this specification. The acquisition of the training data can be implemented and constructed as acquisition of measurement data according to this specification.

[0165] The modified training data can also be implemented and constructed as measurement data according to this specification. The acquisition of the modified training data can also be implemented and constructed as measurement data acquisition according to this specification.

[0166] Here, the machining ML model, measurement data, material strips, surface areas, simulation or partial simulation of changes in measurement data caused by the effects of machining steps on this surface area, and training of the machining ML model can be implemented and constructed according to this specification.

[0167] The above objective is also achieved by a computer-readable storage medium comprising a processing ML model, wherein the processing ML model has been trained according to the method of this specification, and wherein the processing ML model is implemented and constructed for use as one of at least one processing ML model in a method for tracking the position on a material strip according to this specification.

[0168] Here, the processing ML model can be implemented and constructed, for example, to cause the modified first measurement data to be determined by applying the processing ML model to the first measurement data on a computer with the first measurement data.

[0169] Herein, the first measurement data, the computer, the determination of the modified first measurement data, the processing ML model, and the application of the processed ML model to the first measurement data can be implemented and constructed according to this specification.

[0170] The above objective is also achieved by a system for tracking the position on a material strip during processing, wherein the system is implemented and configured to perform a method for tracking the position on a material strip according to this specification, comprising: a first acquisition device for acquiring first measurement data of a first surface region of the material strip; a second acquisition device for acquiring second measurement data regarding a surface region of the processed material strip; and a computing device implemented and configured to apply at least one processing ML model to the first measurement data to generate modified first measurement data, and further for identifying the first surface region on the processed material strip using the modified first measurement data and the second measurement data.

[0171] Here, the computing device also includes, for example, implementations and configurations for comparing acquired measurement data about the surface region of the processed material strip with modified first measurement data, and for using the modified first measurement data to identify a first surface region on the processed material strip according to this specification.

[0172] Herein, the moving or movable material strip, the material strip processing and acquisition device, the surface area of ​​the material strip, the first and second measurement data, the modified measurement data, the processing ML model, and the identification of the first surface area can be implemented and constructed according to this specification.

[0173] As explained within the scope of this specification, identifying a first surface region, for example, on a processed material strip, can be done by comparing second measurement data obtained about a surface region of the processed material strip with modified first measurement data about the first surface region—for example, via a similarity method or a fingerprint method. As described above, the surface region of the processed material strip whose associated second measurement data is most similar to the modified first measurement data is thus identified as the first surface region.

[0174] Here, the moving or movable material belt can, for example, move along the transport direction at a transport speed and / or move along a transport speed curve. For example, the material belt can also move along the direction of movement.

[0175] The system described herein achieves, for example, spatially resolved tracking of the effects of processing on the material strip within the processing area. Here, an objective is to generate, configure, and / or calibrate a moving coordinate system for the material strip to a certain extent by re-identifying surface regions within the processing area, so as to monitor, for example, the effects of processing on the material strip spatially. Furthermore, it is also possible to monitor and / or analyze other regions of the material strip through interpolation between different considered surface regions.

[0176] As described above, by implementing and constructing the system mentioned above to perform the methods according to this specification, it is possible to improve or increase the accuracy of identification of surface areas, identification locations and / or markings within the material strip processing range.

[0177] The computing device can be implemented and configured as, for example, a computer, an edge device, a PLC, a virtual PLC, one or more modules of a PLC or virtual PLC, a control device, a cloud, an automation server, a cloud-based control application, or a similar computing device, or includes such components. The computing device can also be implemented and configured as a system consisting of multiple of the above-described components, which are capable of or have been communicatively coupled.

[0178] Furthermore, the computing device includes a processing ML model. Identification of a first surface region on the processed material strip using modified first and second measurement data is also performed by the computing device.

[0179] Furthermore, it can be specified that the processing ML model is stored in a sub-device of the computing device, such as a module for control devices, a separate edge device, the cloud, or another computing device according to this specification. Furthermore, it can be specified that the generation of modified first measurement data according to this specification is also performed in said sub-device.

[0180] It can also be specified here that the training of such processing ML model is performed either in one of the aforementioned devices or in another computing device according to this specification. In particular, it can be specified that the training of the processing ML model according to this specification is performed in a computing device with relatively high computing performance, such as a computer, workstation, server, cloud, edge device, or comparable computing device.

[0181] Furthermore, it can be specified that the identification of the first surface region by the method according to this specification is performed in the aforementioned sub-device of the computing device, or also in another sub-device of the computing device, such as a module of the control device, a separate edge device, the cloud, or another computing device according to this specification.

[0182] Here, the computing device can, for example, have a storage area for storing measurement data, both pre-processing and post-processing measurement data. Furthermore, the storage area, or another storage area, can also be configured and designed to store acquisition time points, processing parameters, and transport speeds. Additionally, the computing device can implement and construct methods for comparing the measurement data. This comparison can be performed using comparison methods, fingerprinting methods, or recognition methods known in the prior art. If the measurement data is implemented and constructed as image data, then, for example, image comparison methods or image recognition methods known in the prior art can be used.

[0183] A control device can be any type of computer or computer system implemented and configured to control equipment, machines, facilities, instruments, components, or mechanisms. A control device can also be a computer, computer system, or so-called cloud, on which control software or control software applications, such as control applications, are implemented, instantiated, or installed. Such control applications implemented in a computer or cloud can, for example, be implemented and configured as one or more applications with programmable logic controller functionality.

[0184] The control device can also be implemented and configured as a so-called edge device, which, for example, can include an application for controlling the equipment or facility. For example, such an application can be implemented and configured as an application with programmable logic controller (PLC) functionality. Here, the edge device can, for example, connect to another control device of the equipment or facility, or connect directly to the equipment or facility to be controlled. Furthermore, the edge device can be implemented and configured such that it is additionally connected to a data network or cloud, or implemented and configured for connecting to a corresponding data network or cloud.

[0185] Control devices can also be implemented and constructed as so-called programmable logic controllers (SPS) (or PLCs). Furthermore, control devices can also be implemented and constructed as so-called modular programmable logic controllers (modular SPSs).

[0186] Here, the control device can include a control module or a central module, which is implemented and configured to run a control program. For example, the control module can include functions defined by the standard IEC 61131.

[0187] Here, the control module can also be implemented and configured as a software application, for example, to run a control program in real time to control the machined parts. Here, the software application can include, for example, the functions defined in standards IEC 61131 and / or IEC 61499.

[0188] The control module can also be implemented and constructed as a separate mechanical module or component, which is implemented and constructed for real-time operation of the control program. Such a mechanical module or component can, for example, include functions defined by standards IEC 61131 and / or IEC 61499. For example, the control module can be implemented and constructed as a programmable logic controller itself, or, for example, as a central module of a modular programmable logic controller. Here, the control module can, for example, include the functionality of input-output components, or it may not include the functionality of input-output components.

[0189] A programmable logic controller (SPS) is a component that is programmed and used to regulate or control a system or machine. Specific functions, such as sequential control, can be implemented in an SPS, enabling the control of input and output signals of a process or machine. Programmable logic controllers are defined, for example, in standards IEC 61131 and / or IEC 61499.

[0190] To connect a programmable logic controller (PLC) to a system or machine, both actuators, typically connected to the PLC's output, and sensors are used. Status indicators are also used. In principle, the sensor is located at the SPS input, through which the PLC obtains information about what is happening in the system or machine. Suitable sensors include, for example: photoelectric switches, limit switches, pushbuttons, incremental encoders, level sensors, and temperature sensors. Suitable actuators include, for example: contactors for activating motors, solenoid valves for compressed air or hydraulic pressure, drive control modules, motors, and drivers.

[0191] SPS can be implemented in various ways and methods. That is, it can be implemented as a standalone electronic device, software simulation, a so-called "virtual PLC" or "soft-PLC", a PC plug-in card, etc. Modular solutions are also common, in which SPS consists of multiple plug-in modules. Such modules can be, for example, a central control module, an input-output module, a communication module, a frequency converter module, an application module, or a comparable module.

[0192] A virtual PLC, or so-called Soft-PLC, is understood as a programmable logic controller implemented as a software application and capable of running or currently running on computer equipment, industrial PCs or other PCs, computing devices, or, for example, edge devices. In this context, it is also possible to modularize the virtual PLC or Soft-PLC. Here, the various functions of the programmable logic controller or PLC are designed as individual software modules, which are connected or interoperable via so-called middleware. Such modules can be, for example, a central control software module (e.g., the module includes at least, but is not limited to, features predefined by standard IEC 61131), an Ethernet communication module for connecting to a fieldbus, a specific organization or device, OPC-UA or a comparable communication standard, a web server module, an HMI module (HMI: Human Machine Interface), and / or an application module according to this specification.

[0193] Here, a modular programmable logic controller can be implemented and configured such that it can have multiple modules, typically including, in addition to a so-called central module (also called a control central module or CPU), which is implemented and configured to run control programs, such as to control components, machines, or facilities (or parts thereof), one or more expansion modules. Such expansion modules can be implemented and configured, for example, as current / voltage supply devices, or as input and / or output of signals, or as functional modules or application modules to perform specific tasks (e.g., counters, converters, data processing using artificial intelligence methods, including, for example, neural networks or other ML models). In the present case, for example, it can be specified that processing ML models is implemented in such modules for the programmable logic controller.

[0194] For example, functional modules or application modules can also be implemented and constructed as AI modules for performing actions using artificial intelligence methods. Such functional modules can, for example, include neural networks or ML models according to this specification, or another ML model according to this specification.

[0195] Edge devices or edge mechanisms can, for example, include applications for controlling devices or facilities. For instance, such applications can be implemented and configured as applications with programmable logic controller (PLC) functionality. Here, the edge device can, for example, connect to another control device of the device or facility, or connect directly to the device or facility to be controlled. Furthermore, the edge device can be implemented and configured such that it is additionally connected to a data network or cloud, or implemented and configured for connecting to a corresponding data network or cloud.

[0196] Edge devices can also implement and configure additional functions related to the control of, for example, machines, equipment, or components (or parts thereof). Such functions could include, for example: - Collect data and transmit it to the cloud and / or perform appropriate preprocessing, compression and / or analysis on such data; - For example, data can be analyzed using AI methods, such as neural networks or corresponding ML models. To this end, edge devices can, for example, include ML models; - Managing or executing the training of neural networks or ML models. Here, the training itself can be performed at least partially on the edge device itself, but may also include, but is not limited to, the cloud. If the training is performed in the cloud, the edge device can, for example, implement and construct methods for downloading the trained neural network or ML model and subsequently using it.

[0197] The present invention will now be illustrated by way of example, using the tracking (so-called “Tracking” or “Tracing” of production steps or processes) in electrode manufacturing within the scope of battery production.

[0198] In the manufacture of battery cells, the manufacture of battery electrodes is an important step. The electrode manufacturing consists of a number of separate, sequential steps in which large rolls of material, known as “foil rolls” (e.g., aluminum or copper strips up to 4 kilometers long), are processed.

[0199] In the coating step, the roll containing the foil is unrolled, coated with an electrode material, also known as a "slurry," and dried. The resulting coated foil, also called an electrode strip, is then wound into a new roll. The resulting roll is then passed to a calendering step, in which the coated foil is pressed between the rolls. Here, each step can consist of a sequence of unrolling, processing, and winding the roll.

[0200] In the subsequent slitting process, the electrode strip is cut into multiple narrower strips along its length. Optionally, additional cutting can also be performed between the cladding and calendering processes.

[0201] The resulting "sub" rolls of coated, calendered, and slit electrode strips are conveyed to the cell assembly station, where the rolls are cut laterally into plates (e.g., 20 cm in length), stacked, and placed in containers to protect them from environmental impact. Each battery cell consists of multiple electrode plates. While it is easy to identify and trace a single cell as an independent element (e.g., with the aid of printed markings, such as barcodes), it is difficult to identify the origin of the electrode plates across different processing steps on a continuous electrode strip.

[0202] Therefore, one of the topics considered in this specification is traceability of electrode strips across all production steps at the plate level. For example, consider a scenario where a single cell is identified as defective during final testing. Understanding which segments of the electrode strip comprised the single cell layer is from is crucial for comprehensive root cause analysis and processing optimization. However, in today's battery production and / or gigabit plants, corresponding tracking and / or traceability are still performed at the slurry batch and roll ID level, allowing only very coarse inferences about specific single cells.

[0203] For this type of traceability, it is possible to use markings (e.g., barcodes), printing / laser-engraving the markings onto the foil or applying them to the foil. However, the following problems may arise here: - Marking can impair product quality, (2) - It is difficult to find a suitable location for the marking because most of the uncoated edges of the foil are removed in subsequent processing steps, and the location of the pre-reserved notch cannot be known in advance. (3) - In addition, the cut serves as a flow collection point, and surface contamination can impair the primary function of the cut.

[0204] Such traceability includes, but is not limited to, the use of so-called "visual fingerprinting" to identify and compare images of electrode strips.

[0205] The limitations of this known scheme are as follows: it assumes that images from the same strip region from two independent processing steps are so similar that they produce the same fingerprint. Otherwise, the accuracy of the comparison deteriorates (missed matches and false alarms).

[0206] For example, the winding and unwinding of the rolls thus alters the optical appearance of the surface. This can particularly affect the beginning sections of the strip, where the roll radius is small, the curvature is high, and the layers on the rolls are subjected to pressure from the layers above. For instance, defects or distinctive structures on the surface can create indentations on adjacent layers of the rolls, which subsequently degrade visual legibility.

[0207] The method described in this specification can also be used with visual fingerprints or other fingerprints (in which case the acquisition device is configured as a camera, and the measurement data is therefore the acquired image). Here, for example, the so-called "keypoint scheme" and / or the so-called "template matching scheme" can be followed. In the keypoint scheme, specific keypoints (e.g., specific gray-level non-uniformity) are first detected. Then, the keypoints are described by feature vectors, which are ultimately used to find matching keypoints in the image pair. Existing methods, such as ORB (Oriented BRIEF), keypoint detectors and descriptor extractors, or ASLFeat, can be used for keypoint-based image matching. Alternatively, some schemes exist for template matching, such as those based on the Fast Fourier Transform (FFT) in OpenCV. A less common but known scheme is to adapt only the phase portion of the FFT (ignoring amplitude), which can be advantageous because it focuses only on the structural similarity between images and reduces the impact of brightness variations.

[0208] An example of visual fingerprinting based on template matching is... Figure 2 It is shown in the figure and described in detail below.

[0209] The method according to this specification can, for example, have one or more of the following improvements: - Fusion of image-based fingerprint recognition and material strip physical model - Use a probabilistic model of reasonable consistency (based on distance) to reduce false positive consistency. -Probabilistic modeling using the uncertainty of distance, - Faster search by considering monotonicity constraints.

[0210] Visual fingerprinting, whether using template matching or keypoint descriptors, more frequently produces so-called "false positives." These "false positives" are scan windows of the processed material strip that incorrectly match the error window w of the unprocessed material strip (see...). Figure 2 False alarms can be caused by the ambiguity of visual fingerprints, such as due to incomplete images (regarding size, resolution, lighting conditions, and noise), and the necessity of using a relatively low recognition threshold to achieve full recognition of surface areas.

[0211] False alarms can be filtered, for example, by associating the longitudinal position (in meters) on the strip with each window. This position can be calculated from the timestamp of the associated image and a meter counter, the latter also derived from the time-integrated velocity of the moving electrode strip. The longitudinal spacing between windows is then calculated based on the position.

[0212] An unreliable match is one whose distance does not correspond to the expected value of other matches. The decision of which matches are reasonable can be based on a weighted majority decision, where the weights are derived from the confidence of the matches (e.g., based on correlation coefficients). Alternatively, this decision can be made iteratively, for example, using Random Sample Consensus (RANSAC), a known technique in computer vision. In another version of this design, the distance can be represented as a probability density function, which explicitly describes the uncertainty of the value. For example, the timestamp of a camera image can have an uncertainty of 50 ms. At a line speed of 80 m / min, this means an uncertainty of nearly 7 cm for the location. Calculating the distance between two locations additionally increases the uncertainty. In particular, if the probability density functions of two locations are described as normal distributions with variances σ_1^2 and σ_2^2, then the distance between them has a variance σ_1^2 + σ_2^2, which means: the individual variances are added together. In the case of a typical electrode plate length of, for example, 20 cm, the uncertainty may be large enough to become relevant in the physical model.

[0213] Another (independent) advantage of fusing visual fingerprints with physical strip models is that the monotonicity on the strip is constrained, and the resulting sequence of images can be used to accelerate alignment. Specifically, this leverages the fact that the position of the window in the strip monotonically increases. Therefore, instead of comparing each scan window s with all windows w, only a subset of windows w must be considered. For all subsequent consistency, windows w that have already reached a stable consensus on matching scan windows s can be excluded from the window set.

[0214] A special case in this method is strip breakage, i.e., accidental strip breakage, such as due to improper tension control. When repairing a strip breakage, a specific length of the coating is removed, and the new end is manually spliced. The length of the removed section is usually not recorded, which can affect the spacing between windows. However, if a strip breakage occurs, it is usually recorded / documented at least at the roll level (e.g., for quality control reasons), and its location can usually be roughly estimated from control data. Therefore, strip-related optimizations can be temporarily disabled in the section surrounding the strip breakage.

[0215] This specification specifically relates to improvements to fingerprinting methods known from the prior art, such as improvements in image consistency recognition across production steps by predicting processing effects (e.g., winding, calendering) via AI-based image transformation models (image-to-image transformation).

[0216] The motivation lies in the fact that the surface of the electrode strip can be altered by methods and / or processing steps, such as through winding and / or unwinding. Of course, the surface can also be significantly altered partially through calendering.

[0217] These changes can impair consistency and accuracy, or even make comparison impossible. For example, for calenders, it is feasible to circumvent the problem by recording images before calendering, but this would require an additional optical inspection system. However, effects such as those caused by winding and / or unwinding cannot be avoided.

[0218] The basic idea is to train an ML model, such as a deep neural network, according to this specification to predict the effects on the surface. Here, the training set can consist of image pairs from the same area, with and without corresponding effects, and a vector x with additional contextual information (e.g., the location of the segment counter on the strip, product type). An example of creating such training data to model winding and / or unwinding effects could be: acquiring an image recorded before winding and unwinding (e.g., in coating processes, directly after drying and before winding onto a roll) and acquiring another image of the same area after unwinding. Another example of modeling calendering effects could be: recording an image before calendering and acquiring another image of the same area after calendering.

[0219] When creating such a training set, using markers (e.g., stickers) can help achieve high accuracy in pixel-level correspondence between images acquired before and after corresponding method steps. Further improving the accuracy of the training set can be achieved by considering pixel-level image offsets to increase the correlation between images acquired before and after corresponding method steps.

[0220] Based on such a training set, it is possible to train a corresponding ML model, such as a deep neural network (DNN), to perform transformations between images with and without processing or processing steps. Here, for example, known schemes such as pix2pix can also be used.

[0221] Then, this trained ML model can, for example, as follows: Figure 3 Let's take an example. An image w of the material strip before the processing step is fed to the model to predict a transformed image w', which predicts how the original image w will look after the considered processing step (e.g., after calendering). Then, w' is used instead of w in the above comparison. The overall scheme is... Figure 3 As shown in the image.

[0222] As used within the scope of this specification, the term "image" generally refers to a 2D image produced by a linear or area scan camera that can generate grayscale or RGB images. In principle, an "image" can also be a hyperspectral camera image, or an image generated by virtual measurements, such as 3D tomography.

[0223] Other advantageous design options are derived in the dependent claims. Attached Figure Description

[0224] The invention will now be further described by way of example with reference to the accompanying drawings.

[0225] Figure 1 An example system is shown for tracking the position on a moving or movable material strip during pressing. Figure 2 Showing according to Figure 1 Exemplary strip segments and exemplary scanning areas of the material strip before and after pressing; Figure 3 An exemplary method flow is shown for comparing images before and after compression using ML transformation of image data before compression. Detailed Implementation

[0226] Figure 1 A processing station 140 is shown for pressing material strips 120, 122, and 124. Here, material strips 120, 122, and 124 are unwound from the unwinding roll 126 and move at a constant speed v through the pressing station 140 to the winding roll 128, where they are rewound after pressing. In the pressing station 140, the material strips are guided between the upper pressure roll 142 and the lower pressure roll 144. Here, the pressure rolls 142 and 144 and the material strips 120, 122, and 124 are arranged, implemented, and configured such that the material strips are pressed or compressed. This pressing step is present, for example, in the production of electrodes for batteries or accumulator cells, such as after the electrode material is applied to a corresponding carrier layer and dried. In the context of battery electrode production, such a processing step is referred to, for example, as "calendering."

[0227] Figure 1 The entire strip of material from the unwinding roll 126 to the winding roll 128 is not shown. Instead, the same strip of material is shown three times: first as a first image 120 of strip 120 immediately after being unwound from the unwinding roll 126; second as a second image 122 of strip 122 during the pressing of strips 120, 122, and 124; and third as a third image 124 of strip 124 shortly before winding strips 120, 122, and 124. Here, for ease of explanation, corresponding reference numerals are used both for the corresponding images of the strip of material in the figures and for the corresponding strip of material itself.

[0228] also, Figure 1 The tracking system 110 is shown for tracking the material strip during the processing of the material strip—for example, for spatially resolving the effect of pressing the material strips 120, 122, 124 by the pressing station 140.

[0229] Here, the tracking system 110 includes a first camera 150 and a first clock 152 (which can also be implemented and constructed as a timer, clock, or counter, for example), wherein the first camera 150 and the first clock 152 are implemented and constructed such that the clock 152 records the acquisition time point 300 of each image 300 recorded by means of the first camera 150, and then stores the time point together with metadata 300, 310 stored for the image 300 and / or also stores it as metadata 300, 310 associated with the image 300.

[0230] The tracking system 110 also includes a second camera 160 and a second clock 162 (which can also be implemented and configured as a timer, clock, or counter), wherein the second camera 160 and the second clock 162 are implemented and configured such that the clock 162 acquires the acquisition time point 400 of each image 400 recorded by means of the second camera 160, and then stores the time point together with metadata 400, 410 stored for the image 400 and / or also stores it as metadata 400, 410 associated with the image 400.

[0231] Within the scope of this specification, and particularly in the description of the accompanying drawings, it is assumed that the acquired images 300, 400 are acquired, stored, and / or transmitted together with their respective associated acquisition times 300, 400. For example, this is not necessarily mentioned separately each time.

[0232] Furthermore, the tracking system 110 includes a computer 170, to which the first camera 150 transmits images 300 recorded by it along with corresponding acquisition timestamps 300. The second camera 160 also transmits images 400 acquired by it along with corresponding acquisition timestamps 400 to the computer 170.

[0233] Another component of the tracking system 110 is a velocity sensor 130, used to acquire the velocities of the material strips 120, 122, and 124 in a time-resolved manner. The velocity data is then transmitted to a computer 170, where it is stored as time-resolved velocity data or a corresponding velocity graph. The computer then implements and constructs, for example, methods for processing and / or evaluating the velocity data, such as smoothing, averaging, normalization, drift correction, etc. To simplify... Figures 1 to 3 The method flow shown will be explained in the following sections. Figures 1 to 3 In the accompanying drawings and the corresponding method description, it is assumed that the material belt speed v of material belts 120, 122, and 124 is constant.

[0234] Computer 170 is implemented and configured to perform a so-called fingerprinting method, for example, by installing and / or running appropriate software thereon. Here, the fingerprinting system is implemented and configured such that, for example, an image of a specific surface area is acquired before the material strips 120, 122, 124 are processed, and then the surface area is re-identified by comparing an image recorded after the processing of the material strips 120, 122, 124 with a previously recorded image using fingerprinting, identification, or authentication methods known in the art.

[0235] Therefore, for example, images 300 of the material strip 120 before it is pressed by the processing station 140, along with corresponding acquisition time points 300, are recorded at specific time intervals, and then transmitted to the computer 170 along with the acquisition time points 300. Then, after the material strip 122 has been processed by the processing station 140, the second camera 160 again records images 400 of the material strip 124, again along with corresponding acquisition time points 400, which are then also transmitted to the computer 170.

[0236] Then, computer 170 compares the image 400 captured by the second camera with the image 300 recorded by the first camera, and attempts to identify, using a fingerprint method (e.g., a method known from the prior art), an image from the image 400 recorded by the second camera 160 that matches the image 300 recorded by the first camera 150. In this way, it is possible to identify and / or compare the same surface areas or locations on the material strips 120, 122, 124 before and after processing, thereby at least partially tracking the effect of processing on the respective surface areas.

[0237] To perform a fingerprinting method for identifying surface regions w1 after processing material strips 120, 122, and 124 in processing station 140, computer 170 includes a neural network 172. Here, neural network 170 is implemented, constructed, and trained such that it simulates the effect of the processed material strip 122 in processing station 140 on images w1, w2, and w3 of the unprocessed material strip 120 acquired by first camera 150.

[0238] For this purpose, neural network 172 has been trained with image pairs consisting of images of the surface region recorded by a first camera 150 before pressing and images of the same surface region recorded by a second camera 160 after pressing. Here, neural network 172 is implemented and constructed as a deep neural network (DNN) as known in the prior art. The training of neural network 172 is performed according to supervised learning methods known in the prior art for such deep neural networks (DNNs). However, other neural networks or ML models according to the prior art or this specification can also be trained and used for this purpose.

[0239] Therefore, the fingerprinting method is performed as follows: images 300 of surface regions w1, w2, and w3 recorded by the first camera 150 are input into a neural network 172. Then, the neural network 172 generates a modified image 330, which is subsequently used for comparison with an image of the material strip 124 after it has been pressed in the processing station 140 and recorded by the second camera 160.

[0240] The fingerprint method's design improves the recognition probability of the surface regions of material strips 120, 122, and 124 processed at the processing station by ensuring that the changes in the surfaces of material strips 120, 122, and 124 caused by processing are at least partially simulated and / or included in the image 300 processed by the neural network 172 from the first camera 150. This image 330, generated in this manner, may then be more similar to the image 400 recorded by the second camera for each corresponding surface region, thus enabling easier detection by the corresponding fingerprint method.

[0241] The following text combines Figure 3 The design scheme of the fingerprint method will be further elaborated here.

[0242] Furthermore, by recording an image 300 of the material strip 120 before processing using the first camera 150, the length scales of the material strips 120, 122, and 124 can be defined. Here, for example, this can be done by establishing the zero point of this material strip length scale as the surface region w1 associated with the first image w1 (the image recorded by the first camera 150 at a first time point immediately after the material strip 120 is unwound from the unwinding roller 126). From the known transport speed of the strip and the corresponding acquisition time points of the images w1 and w2 recorded by the first camera 150, the spacing between the surface regions w1 and w2 associated with the images w1 and w2 can be determined respectively. In this way, spatial length coordinates on the material strip can be assigned to each corresponding surface region w1 and w2—thus defining length scales, particularly virtual length scales, on the material strips 120, 122, and 124.

[0243] In the same manner, after identifying the first image s1 recorded by the second camera 160, the surface region s1 associated with the image s1 is again the zero point defined above on the material strip—in this way, the zero point is "found again" or "re-identified" to some extent. By re-identifying the images w1 and w2 recorded by the first camera 150 in the images s1-s9 recorded by the second camera 160, other surface regions s5 and s9 are correspondingly re-identified. Then, based on the known transport speed of the material strips 120, 122, and 124 and the corresponding recording time point 400 of the second image 400, the mentioned (virtual) length scale can be reproduced on the material strips 120, 122, and 124 after processing at the processing station 140.

[0244] In this way, for example, it is possible to assign corresponding length coordinates on material strips 120, 122, 124 to each surface region w1, w2, w3, s1, s5, s9 respectively associated with the recordings of the first camera 150 and / or the second camera 160. At least among other things, it is possible to immediately track the effects of processing on material strips 120, 122, 124 by processing station 140 with spatial resolution.

[0245] The length coordinates associated with specific surface areas w1, w2, w3, s1, s5, s9 can also remain associated with these surface areas throughout the entire preceding and subsequent processing chain, until one or more final products are manufactured. Therefore, for example, when a particular product manufactured using a material strip experiences a failure or defect, it is possible to trace which length coordinates or length coordinate ranges of the material strip were used in the product, and then infer the possible causes of the failure or defect based on stored processing data and images.

[0246] In an advantageous design, the known spatial distance between the first surface region w1 and the subsequent second surface region w2, as described above, can be used when successfully identifying the first surface region w1 and then re-identifying the second subsequent surface region w2. This can be implemented and configured, for example, such that after re-identifying the first image w1 or surface region w1 in images s1-s9 recorded by the second camera 160, in order to identify the second surface region w2, an image s5 of the processed material strip 124 is preferentially selected, said image being a certain time interval from the image w1 of the first surface region w1 recorded by the second camera 160, said time interval being derived from the aforementioned known spacing between the associated surface regions of the material strip and the transport speed.

[0247] For example, if when comparing images s1-s9 of the second surface region w2 recorded by the second camera 160 with the second image w2 recorded by the first camera, multiple images s4, s5, and s6 that are somewhat similar to the second image w2 recorded by the first camera 150 are obtained, then within the scope of the method, the following image in image s5 is selected, wherein the distance between the surface region s5 to which the image belongs and the identified first surface region s1 is closest to the expected distance.

[0248] Here, for example, measurement uncertainties in time and conveyor belt speed measurements can also be taken into account. Methodologically, this can be considered such that, in the above case, a spacing probability distribution is determined from the known spacing between the first surface region w1 and the second surface region w2, as well as the measurement uncertainties in image acquisition and conveyor belt speed measurement. This probability distribution can, for example, be implemented and constructed as a Gaussian distribution or a similar probability distribution, or approximated by such a probability distribution. Then, for example, the probability distribution can be compared with similarity values, obtained by comparing images s1-s9 recorded by the second camera 160 with a second image w2 of the second surface region w2 recorded by the first camera 150, according to fingerprint methods known from the prior art, to select the image s5 corresponding to the second surface region w2 from the second camera 160.

[0249] Here, the reference numerals w1, w2, and s1-s9 introduced above are... Figure 2 The description is detailed below.

[0250] Figure 2 The upper area shows the... Figure 1 The material strip segment 120 after being unfolded by the unfolding roller 126 is shown.

[0251] Here, the reference numerals for regions w1, w2, and w3 of the material strip segment 120 before processing respectively denote the corresponding surface regions w1, w2, and w3, and also indicate, for example, the composition of the surface regions w1, w2, and w3. Figure 1 Images w1, w2, and w3 acquired by the first camera 150 are shown in the figure. The reference numerals t1, t2, and t3 indicate the corresponding acquisition time points t1, t2, and t3 of the images w1, w2, and w3, for example, acquired by the first camera 150.

[0252] also, Figure 2 The same material strip segment 124 as the upper segment is shown in the middle and lower regions, respectively, but this time it is after processing the material strip segment 122, or just after winding. Figure 1 Before the winding roller 128 shown.

[0253] exist Figure 2 In the middle, the arrow 'v' symbolizes... Figure 2The material strip segments 120 and 124 shown move at a speed v in the direction of the arrow. Therefore, at the camera relative to... Figure 2 When stationary, material strip segments 120 and 124 move through the camera in the direction of the arrows shown.

[0254] Here, the reference numerals for regions s1, s2, s3, s4, s5, s6, s7, s8, and s9 (also abbreviated as s1-s9 in this specification) of the processed material strip segment 124 respectively represent the corresponding surface regions s1, s2, s3, s4, s5, s6, s7, s8, and s9 (abbreviated as s1-s9), and also indicate the composition of the surface regions s1, s2, s3, s4, s5, s6, s7, s8, and s9 (s1-s9). Figure 1 The images s1, s2, s3, s4, s5, s6, s7, s8, and s9 (s1-s9) acquired by the second camera 160 are shown. The reference numerals T1, T2, T3, T4, T5, T6, T7, T8, and T9 (T1-T9) indicate the corresponding acquisition timestamps T1, T2, T3, T4, T5, T6, T7, T8, and T9 (T1-T9) of the images s1, s2, s3, s4, s5, s6, s7, s8, and s9 (s1-s9) by the second camera 160.

[0255] Figure 2 The upper area shows Figure 1 The material strip 120 shown is in the region immediately after being unwound from the unwinding roller 130 and before processing in the processing station 140. Here, a first surface region w1 is shown on the left side of the material strip 120, a second surface region w2 is shown in the middle, and a third surface region w3 is shown on the right side.

[0256] exist Figure 2 In this process, characteristic shading lines are associated with corresponding surface regions w1, w2, w3 of the unprocessed material strip 120. The shading lines characterize both the specific appearance of the corresponding surface regions w1, w2, w3 and the images w1, w2, w3 obtained therefrom.

[0257] Figure 2 At the bottom is a length scale, which shows the length coordinates of the regions 120 and 124 of the material strip along the material strip. According to this length scale, Figure 2 As shown: along the material strip segments 120 and 124, the first surface region w1 has a starting coordinate x1, the second surface region w2 has a starting coordinate x3, and the third surface region w3 has a length coordinate x5.

[0258] Now, before the material strip segment 120 is processed in the processing station 140, at time point t1, by Figure 1The first camera 150 shown acquires an image w1 of the first surface region w1. At time t2, an image w2 of the second surface region w2 is acquired in the same manner. In the same manner, at time t3, the first camera 150 acquires an image w3 of the third surface region w3 from the third surface region w3.

[0259] The distance between the first surface region w1 and the second surface region w2 can be calculated, for example, from the difference between the acquisition time t2 of the image w2 of the second surface region w2 and the acquisition time t1 of the image w1 of the first surface region w1, and the transport speed v of the material strip segment 120. In this way, the starting coordinate x3 of the second surface region w2 can be calculated from the known position coordinate x1 of the first surface region w1. Similarly, the starting coordinate x5 of the third surface region w3 can then be determined from the acquisition time t3 of the image w3 of the third surface region w3, the acquisition time t2 of the image w2 of the previous surface region w2, and the transport speed v of the material strip segment 120.

[0260] To simplify the following description, it is assumed that the material strip moves at a constant speed v throughout the entire scope of the method described in this specification.

[0261] From this, it can be deduced that, based on the above explanation, the image acquisition process before the material strip 120 passes through the processing station 140 is known: after acquiring the image w1 of the first surface region w1, a time t2-t1 is elapsed before acquiring the image w2 of the second surface region w2. Therefore, the distance x3-x1 between the second surface region w2 and the first surface region w1 is also known. Furthermore, it is also known that after acquiring the image w2 of the second surface region w2, a time t3-t2 is elapsed before acquiring the image w3 of the third surface region w3. Here, the distance x5-x3 between the third surface region w3 and the second surface region w2 is also known.

[0262] according to Figure 2 The processed material strip segment 124 shown in the middle region is described below as a first variant of a tracking method for identifying or re-identifying the surface regions w1, w2, w3 as explained above on the material strip segment 124 processed in the processing station 140.

[0263] Here, by the relevant Figure 1 The second camera 160 mentioned above records images of the associated surface regions s1-s9 on the material strip segment 124 being processed at corresponding recording time points T1-T9.

[0264] Typically, images s1-s9 of the processed material strip segment 124 are then compared with images of the unprocessed material strip segment 120 converted by the neural network 172 using a fingerprinting method according to existing technology. The second camera 160 records images in a different image format than the first camera 150. Therefore, in Figure 2 In the view, the images s1-s9 recorded by the second camera have a different image format than the images w1, w2, w3 recorded by the first camera 150.

[0265] The results of these comparisons are Figure 2 The processed material strip 124 is symbolically represented by the corresponding shaded lines of the corresponding images s1-s9. Therefore, for example, the consistent shaded lines of the first s1 and the seventh surface region s7 of the processed material strip 124, or the consistent shaded lines of the first s1 and the seventh image s7 of said surface regions s1 and s7, indicate that these two images s1 and s7 are similar to the image converted by the neural network 172 of the first surface region w1 of the unprocessed material strip 120. The same applies to the similarity between the fifth s5 and the eighth image s8 of the processed material strip 124 and the image converted by the neural network 172 of the second surface region w2 of the unprocessed material strip 120; and to the similarity between the third s3 and the ninth image s9 of the processed material strip 124 and the image converted by the neural network 172 of the third surface region w3 of the unprocessed material strip 120. This applies to the processed material strip 124. Figure 2 In the surface regions s2, s4, and s6 without shaded lines, no similarity was found to the image converted by neural network 172 for comparison with the unprocessed material strip 120.

[0266] Therefore, in the first step, using a fingerprinting method according to the prior art, images s1-s9 of the processed material strip 124 are now compared with images of the first surface region w1 of the unprocessed material strip 120 converted by the neural network 172. Here, it is confirmed that the image of the first surface region w1 of the unprocessed material strip 120 converted by the neural network 172 is similar to images s1 acquired at time point T1 and s7 acquired at time point T7 of the processed material strip 124. Furthermore, it is known that the first surface region w1 on the unprocessed material strip 120 and... Figure 2 The spacing of the previous surface regions is not shown. By comparing the acquired time points T1 and T7 with the known spacing (considering the transport speed v), only the surface region s1 of the processed material strip 124 acquired through the first image s1 can correspond to the first surface region w1 of the unprocessed material strip 120. From the corresponding acquisition time point T1, Figure 2Based on the acquisition time of the previous image (not shown) and the transport speed v of the material strips 120, 122, and 124, it is now possible to determine the position coordinate x2 of the surface region s1 on the processed material strip 124, which corresponds to the first surface region w1 on the unprocessed material strip 120.

[0267] In the second step, images s1-s9 of the material strip 124 to be processed are now compared with image w2 of the second surface region w2 of the unprocessed material strip 120. Here, it is confirmed that image w2 of the second surface region w2 of the unprocessed material strip 120 is similar to images s5 acquired at time point T5 and s8 acquired at time point T8 on the material strip 124 to be processed. Furthermore, it is known that the distance between the second surface region w2 on the unprocessed material strip 120 and the previous first surface region w1 is significant. By comparing the acquisition times T5 and T7 with the known distance, taking into account the transport speed v, only the surface region s5 of the material strip 124 acquired through the fifth image s5 corresponds to the second surface region w2 of the unprocessed material strip 120. From the acquisition time point T5, the acquisition time point T1 of image s1, and the transport speed v of material belts 120, 122, and 124, it is now possible to determine the position coordinate x4 of the surface region s5 on the processed material belt 124 that corresponds to the second surface region w2 on the unprocessed material belt 120.

[0268] Similarly, it is then determined that the surface region s9 of the processed material strip 124, obtained through the ninth image s9, corresponds to the third surface region w3 of the unprocessed material strip 120. In the same manner, the position coordinate x6 of the surface region s9 on the processed material strip 124 corresponding to the third surface region w3 on the unprocessed material strip 120 can also be determined.

[0269] according to Figure 2 The pressed material strip segment 124 shown below is another variation of the above-described tracking method described below, which is used to identify or re-identify the surface regions w1, w2, w3 described above on the pressed material strip segment 124 in processing station 140.

[0270] Unlike the methods described above, here the images s1-s9 of the material strip 124 to be processed are compared only with the images of the unprocessed material strip 120 that have not yet been identified by the previous comparison steps, which are converted by the neural network 172.

[0271] This is Figure 2 As shown in the figure, such that with Figure 2 Compared to the example shown in the middle, in Figure 2Some of the surface regions s1-s9 shown in the lower part no longer have shaded lines. This is because the images of the identified surface regions w1, w2, w3 of the unprocessed material strip 120, corresponding to the shaded lines and converted by the neural network 172, are no longer used for comparison with the images s1-s9 of the processed material strip 124.

[0272] Therefore, for example, in order to examine the second to ninth images s2-s9 of the processed material strip 124, only the images of the second and third surface regions w2 and w3 of the unprocessed material strip 120, converted by the neural network 172, are used. The reason for this is that the first surface region w1 of the unprocessed material strip 120 has already been identified or identified on the processed material strip, and therefore the corresponding structure is unlikely to appear again in the subsequent images s2-s9 of the processed material strip.

[0273] Furthermore, in order to examine the sixth to ninth images s6-s9 of the processed material strip 124, for example, only the image converted by the neural network 172 of the third surface region w3 of the unprocessed material strip 120 is used. This is because the first and second surface regions w1 and w2 of the unprocessed material strip 120 have indeed been identified or identified, and the corresponding structures can no longer appear in the subsequent images s6-s9.

[0274] exist Figure 2 In this context, it is possible to identify, for example, the seventh and eighth images s7 and s8. Figure 2 The lower region no longer has a shaded line, while the corresponding images s7 and s8 are in... Figure 2 The central region still contains a shaded line. This is because the first and second surface regions w1 and w2 of the unprocessed material strip were already identified when examining the seventh and eighth images s7 and s8. Therefore, to examine surface regions s6-s9 later, only the image of the third surface region w3 of the unprocessed material strip 120, converted by the neural network 172, is used. Furthermore, when examining, for example, the processed material strip 124, the aforementioned seventh and eighth images s7 and s8, no further similarity was detected.

[0275] Figure 3 An example of a method flow for processing an ML model 172, designed as a neural network 172 according to this specification, is shown.

[0276] Here, the combined Figure 1 The neural network 172 shown, implemented in computer 170, is used to consider, approximate, or simulate—or at least partially simulate or approximate—the effects of processed material strips 120, 122, 124 on images w1, w2, w3 recorded on surface regions w1, w2, w3 of unprocessed material strips 120.

[0277] As described above, for this purpose, the neural network 172 has been trained using training methods known from the prior art for supervised learning with the aid of image pairs, the images being images of a specific surface region recorded by a first camera 150 before pressing and images of the same surface region recorded by a second camera 160 after pressing. Furthermore, the neural network 172 is implemented and constructed as a "deep neural network" (DNN).

[0278] In the aforementioned advantageous design, the neural network 172, and for example, each processing ML model 172 according to this specification, can additionally be implemented and constructed such that the aforementioned image pairs are created using different processing conditions, i.e., processing parameters, measurement positions, material strip speeds, temperatures, and / or product characteristics. Furthermore, the aforementioned processing conditions, or portions thereof, can also be used as, or have been used as, part of the training data. Here, the neural network 172, or each processing ML model 172 according to this specification, can also be implemented and constructed such that, in addition to the corresponding image data, the processing conditions according to this specification regarding the material strip processing steps are also used as input data to the neural network 172 or the processing ML model 172.

[0279] exist Figure 3 The left-hand area shows a method flow for image acquisition of the unprocessed material strip 120, such as combining... Figure 1 and / or Figure 2 The following explanation is provided. Here, in the first method step 300, one or more images w are acquired from the unprocessed material strip 120, along with the acquisition time points of the corresponding images w. In parallel with this, and / or after or before, in another method step 310, processing data is determined, such as processing data regarding processing steps, camera position, surface area position or similar position, material strip speed, and / or product identification.

[0280] Using the aforementioned neural network 172, in the next method step 320, image transformation is performed on the image w acquired in the first method step 300 using the neural network 172. Here, the image data of the acquired image w, and at least optionally the aforementioned processing data, and in some cases, the acquired time points, are input into the neural network 172 as input data.

[0281] As output data of neural network 172, a transformed image w' is then obtained, which simulates the effect of the processing steps on the image w obtained from the unprocessed material belt 120. These transformed images w', along with their respective acquisition time points and, if necessary, the aforementioned processing conditions, are stored in a data storage device 340, such as computer 170, another computer, or a corresponding storage device 340.

[0282] Figure 3 The right-hand area is shown in accordance with Figure 1 Another data acquisition step after processing material belt 124 at processing station 140. Here, we again rely on... Figure 1 The second camera 160 shown acquires images s and their corresponding acquisition timestamps from the processed material strip 124 in method step 400. Furthermore, in method step 410, corresponding processing data processed by the processing station is acquired and / or stored again. The images and acquisition timestamps can then be stored in a corresponding storage device or stored therein (not in...). Figure 3 (As shown in the image).

[0283] exist Figure 3 In another method step 460 shown below, the converted image w' from the unprocessed material strip 120, stored in data memory 340, is then compared with the image s obtained from the processed material strip 124. This comparison can be performed, for example, by a fingerprint recognition method according to this specification or a similar known identification method.

[0284] Furthermore, in method step 450, a physical model of the material strip is used to identify the image w' recorded and converted from the unprocessed material strip 120 using the image s obtained from the processed material strip 124. Here, for example, the known positions and spacing of the surface regions obtained on the unprocessed material strip 120 are used to better identify the surface regions on the processed material strip 124. The method can also be implemented and constructed, for example, according to this specification.

[0285] Therefore, the result of the matching or comparison according to method step 460, taking into account physical model 450, is the positioning of the surface area recorded on the unprocessed material strip 120 on the processed material strip 124. This positioning can also be implemented and constructed according to this specification.

Claims

1. A method for tracking positions (w1, w2, w3) on material strips (120, 122, 124) during processing, the method comprising the steps of: - Obtain first measurement data (300) of the material strip (120) with respect to the first surface region (w1). -The material strip (122) is processed in the processing step. - By applying at least one processing machine learning model (172) to the first measurement data (300), the modified first measurement data is determined, and the modified first measurement data is used to identify the first surface region (s1) on the processed material strip (124).

2. The method according to claim 1, characterized in that, The at least one processing machine learning model (172) is implemented and configured such that the at least one processing machine learning model (172) is used to simulate or partially simulate the changes in the first measurement data (300) caused by the processing steps on the first surface region (w1) of the first measurement data (300), or the at least one processing machine learning model includes the simulation.

3. The method according to claim 1 or 2, characterized in that, The at least one processing machine learning model (172) is implemented and constructed as a processing machine learning model (172), which has been trained using the following method steps: Obtain training data (300) on surface region (w1) from training material band (120, 122, 124). The training material strip (120, 122, 124) or the surface region (w1) of the training material strip (120, 122, 124) is processed through the processing steps. Obtain modified training data (400) regarding the surface region (s1). The processing machine learning model (172) is trained using the training data (300) and the modified training data (400). Store the trained processing machine learning model (172).

4. The method according to claim 1 or 2, characterized in that, The processing step includes multiple processing sub-steps, each of which has an associated processing machine learning model of the at least one processing machine learning model (172), wherein each of the processing machine learning models has been trained using the following method steps: Obtain training data (300) on surface region (w1) from training material band (120, 122, 124). The training material strips (120, 122, 124) or the surface region (w1) of the training material strips (120, 122, 124) are processed through the corresponding processing sub-steps. Obtain modified training data (400) regarding the surface region (s1). The training data (300) and the modified training data (400) are used to train the processing machine learning model associated with the processing sub-step. Store the machine learning model for the processing that has been trained and is associated with the processing sub-step.

5. A method for training a processing machine learning model (172), wherein, The processing machine learning model (172) is implemented and constructed such that it simulates or partially simulates the changes in the measurement data (300) related to the surface region (w1) of the material strip (120, 122, 124) caused by the processing steps on the first surface region (w1), or the processing machine learning model includes the simulation, characterized by comprising the following method steps: Obtain training data (300) on surface regions (w1, w2, w3) from the training material band (120, 122, 124). The training material strip (120, 122, 124) or the surface region (w1, w2, w3) of the training material strip is processed through the processing steps. Obtain modified training data (400) for the surface regions (w1, w2, w3). The processing machine learning model (172) is trained using the training data (300) and the modified training data (400). Store the trained processing machine learning model (172).

6. A computer-readable storage medium, wherein, The computer-readable storage medium includes a processed machine learning model (172). The processing machine learning model (172) has been trained using the method according to claim 5, and The processing machine learning model (172) is implemented and constructed for use as a processing machine learning model (172) in the method according to any one of claims 1 to 4.

7. A system (110) for tracking positions (w1, w2, w3) on material strips (120, 122, 124) during processing of the material strips (120, 122, 124), wherein, The system (110) is implemented and constructed for performing the method according to any one of claims 1 to 5, the system comprising: A first acquisition device (150) is used to acquire first measurement data (300) of the material strip with respect to a first surface region (w1, w2, w3). A second acquiring device (160) is used to acquire second measurement data (400) related to the surface areas (s1-s9) of the material strip (124) being processed, and A computing device, implemented and configured to apply the at least one processing machine learning model (172) to the first measurement data (300) to generate modified first measurement data (330), and additionally Used to identify the first surface region (s1) on the processed material strip (124) using the modified first measurement data (330) and the second measurement data (400).

Images