Method for identifying anomalies, and thus also actual treatment positions, of a web and for classifying these anomalies, system and computer program product

Abstract

The invention relates to a method, wherein a web (4) is provided, incorporating a number of treatment features (8, 10) made in it by means of a cutting / grooving unit (6), in each case on the basis of a predefined treatment position (24) and a predefined type of cut or type of groove, wherein a sensor unit (12) is used for measuring a height profile (14, 16) of the web (4) on at least one side of the web (4) and transversely to a conveying direction (F) of the web (4), wherein a number of anomalies (20), and thus also actual treatment positions (18), are identified in the height profile (14, 16) by identifying as an anomaly (20) such sections of the height profile (14, 16) in which the height profile (14, 16) reaches a predefined threshold value (26), wherein a respective anomaly (20) is classified by assigning it to the type of cut or type of groove that the anomaly (20) most closely resembles. The invention also relates to a system (2) and to a computer program product.

Description

[0001] Description

[0002] Methods for identifying anomalies and thus also actual processing positions of a railway line, as well as for classifying these anomalies, system and computer program product

[0003] The invention relates to a method for identifying anomalies, in particular cuts and / or creases, and thus also actual processing positions of a web, in particular corrugated board, as well as for classifying these anomalies. The invention further relates to a system and a computer program.

[0004] The web is, for example, a corrugated board web, composed of several layers of paper. The machine is, for example, a corrugated board machine, which first produces the web and then processes it further. The web is typically processed by a cutting and creasing unit, which makes a number of cuts and / or creases. This serves two purposes: firstly, to divide the web into several separate sheets, and secondly, to provide each sheet with a number of folds, creases, or other features, allowing it to be transformed from a flat shape into a typically three-dimensional final form (e.g., a box).

[0005] For the most economical and environmentally friendly production of corrugated board, including sheets and panels, it is desirable to minimize waste during manufacturing and processing. Waste regularly occurs when cuts and / or creases are made incorrectly or not at all. The earlier and more reliably any defects in the cutting and / or creasing process are detected, the less waste is generated and the more targeted the necessary corrective action can be, such as replacing a tool or adjusting an operating parameter.

[0006] DE 10 2015 200 397 A1 describes a testing device for checking the quality of processing results of a processing device for substrate sheets, comprising: a light source for illuminating a processing section of a substrate sheet that is mechanically processed by the processing device using a processing tool to change its structure, an image sensor which is configured to capture the illuminated processing section in an image and to generate Istbi ID data of the processing section on this basis.Furthermore, an evaluation device is available which is set up to output an evaluation result for the quality of the machining results, wherein target data for the machined machining section are stored in the evaluation device and the evaluation device is set up to compare the actual image data of the machining section with the target data under predetermined evaluation criteria and to determine correction data for the machining parameters as part of the evaluation result based on a resulting comparison result.

[0007] DE 10 2019 105 217 A1 describes a folding machine for folding carton blanks comprising at least one control unit, means for conveying the folding carton blanks from an input to an output of the folding machine, a device for folding the folding carton blanks, and an alignment unit for folded folding carton blanks. The folding machine has a device for optically detecting creases and / or gaps in the folding carton blanks, and a device for optically detecting creases in the transport direction is arranged upstream of the means for folding the folding carton blanks, and / or a device for optically detecting gaps is arranged downstream of the means for folding the folding carton blanks and upstream of the alignment unit in the transport direction.US Patent 5,581,353A describes a device for measuring at least two properties of corrugated board, the device comprising: at least one pair of laser triangulation sensors, one of which is located on one side of the corrugated board and the other on the opposite side. The distance between each sensor and the surface of the corrugated board is measured and converted into at least two properties of the corrugated board.

[0008] Against this background, an object of the invention is to improve the verification of the processing of a web by a cutting and creasing unit. For this purpose, a corresponding method, a system, and a computer program product are to be specified.

[0009] The problem is solved according to the invention by a method with the features of claim 1, by a system with the features of claim 23, and by a computer program product with the features of claim 24. Advantageous embodiments, further developments, and variants are the subject of the dependent claims. The descriptions relating to the method also apply mutatis mutandis to the system and the computer program product, and vice versa. Advantageous embodiments for the system result from its configuration to execute at least parts of the method or one or more steps of the method. For this purpose, the system then includes a suitably designed control unit. Analogously, advantageous embodiments for the computer program product result from its inclusion of instructions which, when executed by a computer, cause the computer to execute at least parts of the method or one or more steps of the method.

[0010] The method is specifically designed to identify anomalies and thus the actual processing positions of a web, as well as to classify these anomalies. In particular, the method enables improved verification of operations performed by a cutting / creasing unit and, especially, improved verification of the processing positions of these operations, i.e., cutting and / or creasing positions. First, a web is provided into which a number of operations have been performed by a cutting / creasing unit, each based on a predefined processing position and a predefined cutting or creasing type (generally referred to as "operation type"). "A number of" here, and generally, means "one or more" or "at least one." The operations are mechanical modifications of the web and each consists of either a cut or a crease in the web.Each cut was produced according to a predefined cut type, and each crease according to a predefined crease type. The cutting and creasing unit is therefore used to introduce cuts and / or creases into the web. A cut represents a separation point where the web is severed. In contrast, a crease initially only represents an embossed area, in particular a predetermined fold line (or similar), in the web, which allows a finished blank or sheet to be folded along this predetermined fold line. Accordingly, cuts and creases differ primarily in that a cut completely severs the web, which is not the case, or only partially so, with a crease. The cuts and / or creases serve, in particular, to assemble the web into individual blanks or sheets and / or to create break, fold, or crease edges of the blanks or the subsequent sheets, e.g.to form each of them into a box, container, or similar.

[0011] A cutting and creasing unit is understood to be, in particular, a cutting and / or creasing unit, i.e., a unit designed either for cutting the web or for creasing the web, or both. Hereinafter, without limiting generality, it is assumed that this is a cutting and creasing unit with which the web is both cut and creased. The cutting and creasing unit is, in particular, part of a web processing system. Preferably, the system is a corrugated board plant for the production of corrugated board, especially sheets of corrugated board. The web is, in particular, made of paper and is preferably a corrugated board web composed of several layers of paper. Preferably, the cutting and creasing unit is part of the corrugated board plant, and the web is a corrugated board web that is produced and processed by the corrugated board plant.In a suitable configuration, the cutting / creasing unit is arranged downstream of a so-called double-facer of the system. Within the double-facer, several intermediate products for a corrugated board web are assembled to form the actual corrugated board web. The double-facer also marks the end of a so-called "wet end" of the system (and is part of this system), followed by a so-called "dry end," which is used to convert the corrugated board web into individual panels and subsequently into individual sheets. The cutting / creasing unit is specifically part of the dry end of the system. For the purposes of this description, it is assumed, without limitation of generality, that the web is a corrugated board web. While such a configuration is preferred, the explanations given here apply in principle to any web, especially paper, possibly even single-ply, and to any system that processes a web.

[0012] The web is processed at a number of actual processing positions (i.e., cutting and scoring or creasing positions) using the cutting and scoring unit. At each actual processing position, a characteristic anomaly is formed due to the processing process. By identifying these anomalies (as explained in detail below), the actual processing positions are automatically identified, i.e., recognized, and are then referred to as recognized processing positions. The processing positions are defined by order data for a job to process the web. Accordingly, the order data contains a number of predefined processing positions.The type of cut or groove to be performed at each specified processing position is also defined in the order data, meaning that each specified processing position is assigned a corresponding cut or groove type. The cutting / grooving unit is controlled according to the order data; that is, the cutting and / or creasing elements of the cutting / grooving unit are controlled accordingly to produce the specified cut or groove type at the respective processing positions. Each specified processing position is assigned exactly one cutting or creasing element, which is positioned appropriately in the transverse direction to engage the workpiece and create the corresponding cut or groove.Due to errors or inaccuracies, the cutting and creasing unit does not necessarily produce the machining operations at the specified machining positions, but rather at the actual machining positions. Furthermore, the required cut or creasing type may not be reproduced exactly, or an incorrect cut or creasing type may even be used. In other words, the cuts and creasing operations that are actually produced are not necessarily located precisely at the specified machining positions, and the execution of a cut or creasing operation may also be faulty, resulting in an incorrect cut or creasing type being produced at a particular machining position. Consequently, a deviation (i.e., a difference) between the specified and actual machining positions may occur. Therefore, it is advantageous to check the machining result.Any deviation that may occur is expediently quantified and used by a control unit as an error signal to control the cutting and grooving unit depending on the difference in such a way that this deviation is minimized (closed-loop approach), as described in more detail below.

[0013] Furthermore, an anomaly is not necessarily caused by machining with the cutting and creasing unit, but can also be generated by other machining operations or damage. These anomalies and their corresponding actual machining positions are also detected; however, the latter is not assigned to any predefined machining position, nor does the anomaly necessarily correspond to a known cutting or creasing type. Therefore, the inspection is advantageously designed in such a way that such anomalies are not incorrectly identified as predefined machining operations.

[0014] Advantageously, the anomalies are evaluated and characterized independently of each other, i.e., classified as a whole, so that erroneous conclusions (e.g., based on the assumption that all anomalies are cuts or grooves) are avoided as much as possible from the outset.

[0015] In the method described here, a sensor unit is used to measure the height profile of the web on at least one side, preferably on both sides, and perpendicular to a conveying direction (i.e., also the longitudinal direction) of the web. The sensor unit is preferably part of the system. Preferably, the height profile is measured inline within the system and downstream of the cutting / grooving unit. The web has a width perpendicular to the conveying direction, and the height profile preferably extends across the entire width. The height profile is also referred to as a transverse profile, since it is generally measured transversely and preferably—but not necessarily—perpendicular to the conveying direction. The height profile is also referred to as a surface contour, as it represents the contour of a surface of the web. As described, deviations from the specifications according to the order data can occur during processing.Processing with the cutting and creasing unit creates characteristic changes in the web's height profile. This allows the actual processing positions and the operations performed there to be identified and verified by measuring and evaluating the height profile. The height profile is measured, for example, with a distance sensor and then represents the distance between the web and the sensor as a function along a transverse direction perpendicular to the conveying direction. The height profile can be represented, in particular, as a vector with a multitude of measured values, each at a different position in the transverse direction. Each measured value and its corresponding position (i.e., position value) together form a data point; the length of the vector corresponds to the number of data points.

[0016] In the height profile, the machining operations performed by the cutting and grooving unit are recognizable as anomalies at actual machining positions, i.e., as deviations from, for example, a normal state of the path, i.e., a path without machining operations. Therefore, a number of anomalies, and thus also actual machining positions, are identified in the height profile by identifying those sections of the height profile as an anomaly where the height profile reaches, in particular exceeds or falls below, a predefined (first) threshold value. Each anomaly is thus a section of the height profile and therefore two-dimensional, i.e., it contains multiple data points (measured values ​​and position values). Each anomaly corresponds to an actual machining position, which contains the positions of the corresponding section.Only the positional values ​​of a given anomaly are considered, and therefore the data is one-dimensional, essentially a positional interval or range. By using a threshold, a definition of the normal state is unnecessary; rather, the threshold specifies what is considered normal / non-normal. A given section is, in particular, a contiguous set of data points. A given section is expediently not merely the area where the elevation profile actually reaches the threshold, but includes a specific allowance, especially on both sides. This allowance is expediently sized such that the anomaly, which regularly begins along the elevation profile even before the threshold is reached, is captured as completely as possible by the section.

[0017] Each anomaly is then classified by assigning it to the cutting and / or creasing type to which it most closely resembles. In other words, each anomaly, and thus its actual processing position, is compared to predefined cutting and / or creasing types at predefined processing positions. Based on this comparison, a similarity is calculated, indicating how similar the anomaly at its actual processing position is to a given cutting and / or creasing type at a predefined processing position. Various aspects can be considered in calculating this similarity, particularly the distance between the actual and predefined processing positions, as well as the shape of the anomaly and the cutting and / or creasing type.The system therefore does not merely check whether or to what extent a given machining operation corresponds to the cutting or creasing type specified in the order data, but rather identifies the actual cutting or creasing type. This check is initially independent of the order data. The classification thus differs from a simple target / actual comparison, particularly in that the anomaly is not compared to a single cutting or creasing type, but to several different ones. Furthermore, the use of a threshold value actively searches for anomalies across the entire height profile, rather than merely examining the height profile at the specified machining positions. This allows for the detection and advantageous classification of non-specified machining operations (e.g., damage).Overall, this improves the verification of the web processing and, in particular, enables more detailed quality control than a simple target / actual comparison and a check of the processing positions of cuts and grooves in only predefined sections. This advantageously allows for the detection of any incorrect cuts and grooves that occur outside the expected sections.

[0018] The classification described here is also more reliable compared to a simple minimum or maximum identification function, which, by its very nature, always results in a cut or groove being detected, even when none of these have occurred. Furthermore, this method allows for validation, which advantageously identifies whether the correct groove profile (i.e., the correct groove cutter) was used for each groove, in accordance with the job specifications. Finally, the method described here also enables the inspection of the web for damage away from the cuts and grooves. Overall, this method makes it possible to detect missing or incorrectly positioned cuts and grooves, incorrectly inserted cutting and groove cutters, insufficiently executed cuts and grooves, and web damage.In response, a corresponding message is appropriately issued and / or an operator of the system is informed accordingly. Optionally, the system (and thus further processing of the track) is stopped or at least slowed down, thereby preventing the production of rejects.

[0019] The method performed and described here, especially the classification, also differs from the aforementioned DE 102015 200 397 A1, DE 10 2019 105 217 A1, and US 5,581,353 A. In this classification, a specific cutting or creasing type is assigned to a signal segment of the height profile. This identifies the actually produced cut or creasing not only as such, but also by type. Furthermore, the entire path, or at least extensive sections thereof, is checked, and not just those individual positions where cuts or creasing are expected. For cutting and / or creasing, the cutting / creasing unit has, in particular, one or more corresponding cutting elements and / or creasing elements (generally a tool), i.e., preferably several cutting elements and / or creasing elements, and not merely a combination of a single cutting element and a single creasing element.In both cases, the web undergoes mechanical processing; in the case of cutting, this is simply separation, and in the case of creasing, it is formed. A cutting unit typically has a first blade, which rotates around an axis transverse to the web during operation and is then inserted into the web to produce a cut. The blade can be, for example, a through blade for producing continuous cuts or a perforating blade for producing a perforation, i.e., a perforating cut. The cutting unit also has a counter roller, which is located on the opposite side of the web and interacts with the first blade (a second blade instead of the counter roller is also conceivable). Similarly, a creasing unit typically has a first roller with a profiled running surface, which rotates around an axis transverse to the web during operation and is then pressed against the web to produce a groove.Furthermore, the creasing body has a counter roller, typically with a second profiled running surface, which is located on the opposite side of the web and interacts with the first profiled running surface. The first and second running surfaces are not necessarily identically profiled. Different pairings of running surfaces are suitable for producing various creasing patterns, such as one-point creasing, three-point creasing, five-point creasing, dot-dot creasing, or asymmetric creasing. A particularly common creasing pattern is three-point creasing using a three-point creasing body. A so-called offset creasing is also possible by shifting the running surfaces of a creasing body transversely relative to each other. This shift is also referred to as the offset. Such offset creasing patterns can also be detected and verified using the method described here.A further adjustment is also regularly possible perpendicular to both the transverse and conveying directions. This sets a distance between the two running surfaces of a creasing body, the so-called creasing body spacing. This is expediently determined based on the thickness of the web (caliber). The creasing body spacing can also be identified and verified using the method described here.

[0020] The following section describes in detail the advantageous aspects and preferred configurations of the process. First, an overview is provided, followed by a more detailed discussion of the following aspects: measurement, preprocessing, position detection, validation of the processing positions, classification, and validation of the classification result.

[0021] Overview

[0022] As previously described, this method involves measuring at least one elevation profile of the railway line and then evaluating it by identifying anomalies (i.e., conspicuous sections of the elevation profile). For evaluation, the elevation profile is advantageously pre-processed, primarily using digital signal processing methods, and may be filtered in the process.

[0023] Subsequently, anomalies in the elevation profile, and thus the actual processing positions, are identified using position detection, which employs suitable statistical methods. Position detection is performed during normal operation of the system to ensure smooth processing of the track. Advantageously, the identified processing positions are then assigned to the predefined processing positions from the order data.

[0024] For further analysis, the aforementioned classification is used in the procedure described here. This classification categorizes the height profile section by section at the identified machining positions with regard to the machining operations performed (cuts and grooves), and in particular with regard to any damage, deviations, production defects, etc. In other words, the anomalies are classified. Various methods are suitable for the actual classification, i.e., the assignment of an anomaly to a specific cutting or grooving type (preferably also to a damage type, deviation type, production defect type, etc.), some of which are described below.

[0025] Advantageously, validation is also performed. This validation is either a validation of the processing positions or a validation of the classification result, preferably both, so that the validation is then two-part, which is assumed hereafter without limitation of generality. In the validation of the processing positions, a warning is issued, the system is stopped, or another suitable measure is initiated as soon as it is determined (e.g., already during position recognition) that the recognized processing positions do not match the specified processing positions from the order data within a given tolerance, or are directly identified as damage, deviations, production errors, or the like. The validation of the classification result based on the order data is carried out analogously to the validation of the processing positions, i.e.,The system checks whether the correct cutting or creasing type was applied and / or executed correctly, in particular whether the cuts and grooves were produced with the correct cutting or creasing tool. If, during this validation process, it is determined that the detected machining operations do not match the specified cutting and / or creasing types from the order data within a given tolerance, a corresponding message is issued, the system is stopped, or another suitable measure is initiated. Overall, the validation, in any configuration, thus checks and ensures a certain quality of the machining positions and the operations performed there, and initiates an appropriate response if necessary.

[0026] Preferably, the method is implemented by several modules, which are functionally separate from one another. A first module is preferably a data preprocessing module that performs the preprocessing and provides the measured elevation profile as a uniform data basis for the subsequent modules. A second module is preferably a detection module that performs position detection and recognizes anomalies in the elevation profile, then calculates the recognized (also detected or actual) processing positions and finally outputs these, preferably to a third module, which is an assignment module. The assignment module performs the assignment of the processing positions and assigns the recognized processing positions to the predefined processing positions according to the order data.A fourth module is preferably a classification module, which performs the classification and assigns a cut or groove type to each anomaly, or optionally even a damage type, deviation type, production defect type, or the like. A fifth module is preferably a validation module, which performs the validation and compares the detected machining positions with the specified machining positions and / or compares the detected cut and / or groove types with the specified cut and / or groove types. Optionally, one or more of the modules are present multiple times, in particular to evaluate cuts and grooves separately, especially to classify and / or validate them.

[0027] The various modules are preferably combined in a single control unit (integrated design), or alternatively, one or more modules are housed separately in one or more other control units (distributed design). One or more of the aforementioned control units, and thus also the modules they may contain, are advantageously part of the system described here.

[0028] measurement

[0029] The elevation profile is ideally measured using a sensor unit (also called a sensor module) with a number of sensors. A suitable sensor is, in particular, a distance sensor as mentioned above, e.g., a laser triangulation sensor, which can be moved transversely to the track (e.g., by means of a linear motor) and measures a distance in the direction of the track, i.e., the distance between the sensor and the track. A camera is also suitable as a sensor, especially one that captures an image of the track from which the elevation profile is then derived. The sensor outputs a measured value (distance) for each measurement position across the track, which varies depending on the surface characteristics of the track. As a result, the sensor outputs an elevation profile containing a series of measured values, particularly in the form of a vector. Each measured value is also assigned a position (measurement position) along the track, which, for example,The position is determined from the sensor's travel path. A pair consisting of a measured value and its corresponding position constitutes a data point. The sensor unit is preferably arranged directly downstream of the cutting / creasing unit, i.e., the web path from the cutting / creasing unit to the sensor unit is at most 10 m, preferably at most 1 m. Advantageously, the sensor unit has a pressure roller to press the web against a support as it passes through the sensor unit and during the measurement of the height profile, thus suppressing any curvature of the web as much as possible, at least during the measurement.

[0030] Preferably, the sensor unit comprises two sensors as described, one of which measures an upper height profile along the top side of the web, and the other similarly measures a lower height profile along the underside. This advantageously takes into account detailed information that is only found in one of the two height profiles. Both height profiles are processed in the same way, so for simplicity, only one height profile will be referred to below. In principle, it is sufficient to use only one height profile (i.e., on the top side or the underside; one-sided measurement), but the use of both height profiles (two-sided measurement) is preferred, especially for the classification of grooves, since the height profiles on the top and underside regularly differ, thus enabling more precise identification of the groove type and therefore classification.Furthermore, cuts and grooves can then be distinguished more clearly. However, the processing steps for both elevation profiles are expediently identical. In principle, the data points obtained from the measurement can be used directly for the subsequent steps and modules; however, it is advantageous to preprocess the data points, i.e., the measured elevation profile is preprocessed and is then a preprocessed elevation profile.

[0031] Pre-processing with the data pre-processing module

[0032] During preprocessing, the measured elevation profile is prepared specifically for subsequent position detection. For example, it's possible that the measurement range is exceeded during the elevation profile measurement, preventing a distance from being measured. This regularly occurs at a crosscut site, as the path is interrupted at the processing position. Consequently, if the measurement range is exceeded, a valid measurement value may not be generated; instead, a NaN (not a number) is returned, and a corresponding gap appears in the elevation profile at the associated measurement position. During preprocessing, these gaps (NaNs) in the elevation profile are expediently removed to obtain a complete profile. However, the gaps are not simply omitted but replaced with suitable values, for example, using linear interpolation based on the measurements of neighboring measurement positions.Alternatively, a gap can simply be replaced by a previously defined value, e.g., using a sample-and-hold function with the last valid measurement. Alternatively, a value of "0" (i.e., a vanishing gap) can be used for each gap. In any case, the number of data points should be preserved as much as possible. However, the aforementioned methods may have significant effects in the frequency domain, so it is advisable to additionally filter the height profile processed in this way with an anti-aliasing filter.

[0033] Furthermore, the data preprocessing module expediently crops the measured elevation profile to the width of the track, i.e., the zero point of the elevation profile is shifted to a side edge of the track. Alternatively or additionally, the elevation profile is also limited to the width of the track, which facilitates a subsequent comparison of the detected machining positions with the specified machining positions. The elevation profile is expediently mapped onto a grid, preferably an equidistant one, using the data preprocessing module. New data points with new (measured) values ​​and positions are generated from the measured values ​​and positions, distributed along an equidistant grid. This determines the distance between the data points of the elevation profile. The elevation profile is then, in particular, a vector in which each individual component corresponds to a value at a position along the elevation profile.If necessary, further data points are derived (e.g., interpolated) from the measured values ​​to obtain a specific grid and, in particular, a specific data point density (e.g., number of data points per millimeter). The data point density is especially relevant for the accuracy when viewing and processing the height profile in the frequency domain. Studies have shown that a data point density of at most 10 / mm is still sufficient; this corresponds to a distance of 0.1 mm between the individual positions of the height profile with an equidistant grid. Preferably, the data point density is predefined, and based on this, support points are defined as new positions at equidistant intervals. At each of these positions, a new value is then determined based on the measured values ​​and measurement positions, e.g.,If a measurement position and a reference point coincide, then this measurement position is used as a data point along with the corresponding measurement value; otherwise, the data point is interpolated from the nearest measurement values ​​and measurement positions. The resulting elevation profile is also referred to as a corrected elevation profile.

[0034] The height profile is ideally calibrated to compensate for deviations in the parallelism of the two sensors due to manufacturing tolerances or temperature fluctuations. For this purpose, a calibration vector is determined using a calibration measurement and then subtracted from the height profile. However, the specific implementation of such a calibration is of secondary importance here and will therefore not be described in detail. The processed, trimmed, cleaned, and / or calibrated height profile is then output as a pre-processed height profile for the subsequent steps and modules.

[0035] Filter

[0036] The detection module is designed to identify any anomalies (contiguous subsets of data points) in the processed elevation profile and thereby recognize the actual processing positions (contiguous subsets of positions). It is initially assumed that the elevation profile, for a path without cuts or grooves (i.e., in its normal state), can be described, at least to a good approximation, as a constant function. Therefore, an anomaly is conveniently defined as any deviation of the elevation profile from this constant function. In this case, it was found that the anomalies are also detectable in the frequency domain, specifically in the spectrogram of the elevation profile. The spectrogram is a two-dimensional spectral representation and contains the frequency-dependent amplitude of the elevation profile as a function of the position of the respective data point (especially along the grid).The spectrogram is generated from the elevation profile, for example, using a short-time Fourier transform (STFT), where the time dimension is replaced by position (i.e., a spatial dimension), so that the frequencies are then inverse positions / positions. The achievable frequency range depends on the data point density described above (Nyquist-Shannon sampling theorem). With ten data points per millimeter, a frequency range of 0 to 5 mm is possible. -1 representable.

[0037] Investigations have shown that continuous lines (parallel to the frequency axis) appear in the spectrogram at the exact same positions as actual cuts. This means that a cut can be identified across the entire frequency spectrum and that by filtering out a suitable sub-range from the frequency spectrum (i.e., by selecting a specific frequency range) containing only the frequencies of a cut, the influence of this cut on the measured height profile can be isolated, thus enabling improved identification of the cut. Accordingly, in an advantageous embodiment for better identification of cuts in particular, a corresponding frequency range is isolated using a bandpass filter. The bandpass filter then has a passband that corresponds to the frequency range. A suitable frequency range extends, for example, from 1.5 mm. -1up to, in particular, 5 mm -1 This is because, in such a frequency range, only the effect of the cuts is visible in the spectrogram, and all other influences on the height profile lie completely or at least predominantly below this frequency range. Accordingly, the upper limit is not strictly necessary; rather, it is primarily determined by the data point density. Similarly, anomalies in the frequency range of 0 mm are also visible at the processing positions of grooves in the spectrogram. -1 up to approximately 0.1 mm -1 Recognizable, so that in an advantageous embodiment, a corresponding frequency range is specifically reserved using a bandpass filter for better identification of grooves. A suitable frequency range (i.e., passband of the bandpass filter) extends, for example, from a frequency > 0 mm. -1 (e.g. 0.001 mm) -1 ), to suppress the DC component in the height profile at 0 mm -1, down to 0.1 mm -1 A combination of both of the aforementioned bandpass filters is also useful. Depending on the application, frequency ranges other than those mentioned above may also be advantageous.

[0038] In general, to better identify one or more anomalies, it is advantageous to use a bandpass filter to isolate a frequency range containing the anomalies. Accordingly, in a suitable embodiment, the height profile is filtered with a bandpass filter before the anomalies are identified within it. The bandpass filter isolates a frequency range selected to specifically preserve those parts of the height profile that are generated either by cuts or grooves. The isolated frequency range (i.e., passband) is achieved by a suitably selected bandwidth of the bandpass filter with appropriate cutoff frequencies.The selected frequency range is chosen appropriately depending on the application, possibly differently than described above, and thus also represents an adjustable parameter of the detection module, the identification of anomalies can be optimized by adjusting this parameter. In an advantageous embodiment, the preprocessing is further improved by one or more additional sensors for vibration measurement on the track. A corresponding sensor measures vibration of the track as it moves through the system. Furthermore, the vibration frequency is determined, and this vibration frequency is then expediently filtered out from the height profile, e.g., by means of a suitable frequency filter or by selecting the passband of the aforementioned bandpass filter such that the vibration frequency is suppressed.

[0039] The filtering described above with one or more bandpass filters is optionally performed within the detection module and then as part of the position detection.

[0040] Position detection using the detection module

[0041] The (first) threshold is suitably determined using a statistical method and, in a preferred embodiment, is a standard deviation of the elevation profile, e.g., 20. The threshold depends particularly on the position and forms an envelope around the elevation profile. If the elevation profile breaks through this envelope, i.e., reaches the threshold, then an anomaly exists at that position and is identified as such. By multiplying the standard deviation by a factor, the threshold can be scaled as needed, and this factor thus represents a parameter for optimization. The threshold, and especially its determination, particularly benefits from the filtering of the elevation profile using a bandpass filter described above, since this limits the elevation profile to the specific anomaly (cut or groove) being sought and facilitates its identification using statistical methods.The positions of the identified anomalies in the elevation profile are considered candidates for recognized processing positions, particularly for subsequent assignment to a predefined processing position using the mapping module. To detect sections, anomalies are expediently identified separately in both elevation profiles (upper and lower). An anomaly at an (actual) processing position is then recognized as a section if an anomaly has been identified at that processing position in both the upper and lower elevation profiles. Alternatively or additionally, an anomaly at an (actual) processing position is expediently recognized as a section if a gap (NaN) exists in either the upper or lower elevation profile at that processing position. This also includes the case where a gap exists in both elevation profiles simultaneously.Suitablely, these three criteria are linked together (calculated) so that the presence of one of the criteria is sufficient to identify a cut.

[0042] To detect anomalies, especially grooves, a differential height profile is conveniently calculated from the lower and upper height profiles, for example, by subtracting them. In other words, a differential height profile is created by identifying a number of anomalies in a profile where the height reaches a predefined threshold. This profile is defined as an anomaly in sections where the height reaches a predefined threshold. It is assumed that an anomaly, specifically a groove, reduces the thickness of the web at a specific (actual) processing position. The differential height profile is then conveniently filtered with a bandpass filter, as described above in the preprocessing section, making it particularly robust against measurement disturbances.If the difference height profile reaches the (first) threshold value at a machining position, an anomaly, especially a grooving, is detected at that machining position.

[0043] Using the two aforementioned steps for detecting cuts on the one hand and grooves on the other, the identified anomalies are conveniently divided into two groups before classification: a first group containing all anomalies identified as cuts, and a second group containing all anomalies identified as grooves. The anomalies are thus pre-sorted. These two groups of anomalies are then processed separately, i.e., classified and / or validated. In principle, the same classification module can be used for both groups, or a separate, specialized classification module can be used for each group.

[0044] Assignment of processing positions using the assignment module

[0045] As mentioned above, each anomaly is assigned an actual processing position, which is automatically recognized upon identification of the anomaly and is then also referred to as the recognized processing position. During the assignment of processing positions, the actual processing positions are assigned to the predefined processing positions, and these are conveniently compared. In a suitable implementation, the assignment module generates a correlation table for this purpose, which contains the similarity of each actual processing position to each predefined processing position. The similarity arises in particular from the deviation of the actual processing position from the (assigned) predefined processing position; this deviation is therefore appropriately used as a measure of the similarity between the actual and the predefined processing positions.For example, the correlation table is a simple distance table containing the distances of each actual processing position to each predefined processing position. These specific distances represent a measure of the deviation and thus also of the similarity of an actual processing position to a predefined processing position. Each actual processing position is then assigned the predefined processing position that is most similar to it, i.e., the one with the greatest similarity, for example, the one with the smallest distance. Advantageously, this also avoids double assignments, i.e., assigning multiple actual processing positions to only one predefined processing position or vice versa. Instead, each actual processing position is assigned to at most one predefined processing position, and vice versa.The described comparison also advantageously identifies any excess processing positions, i.e., those processing positions that have been identified but cannot be assigned to any predefined processing position in the order data. This is particularly the case when the number of predefined processing positions is less than the number of identified processing positions. Conversely, missing predefined processing positions are also conveniently identified if the number of predefined processing positions is greater than the number of identified processing positions. In this way, excess or missing processing positions are identified because not all identified / predefined processing positions can be assigned.

[0046] It is also advantageous to have a design in which a threshold for similarity is defined, which must be reached at a minimum to perform an assignment. If the distance is used as a measure of similarity, the threshold is accordingly a maximum distance (e.g., 5 mm). This avoids forced assignments and improves the detection of redundant and missing machining positions.

[0047] In a suitable implementation, during validation, an anomaly is marked as an expected anomaly if a detected machining position could be assigned to a predefined machining position; otherwise, the anomaly is marked as an unexpected anomaly. Specifically in the latter case, a notification is appropriately issued indicating that unexpected machining operations, damage, cuts, and / or grooves have been detected. Conversely, during validation, a predefined machining position is appropriately marked as a found machining position if it could be assigned to a detected machining position; otherwise, the machining position is marked as not found.During validation, this is expediently applied analogously to the cut or groove type, so that a given cut or groove type is marked as not found if it was not found using the classification.

[0048] Classification (characterization) using the classification module

[0049] The identified processing positions, which, as described, can optionally be assigned to predefined processing positions and preferably validated in the process, are now classified using the classification module and thereby assigned to one of several cutting or creasing types. In general terms, the classification process evaluates each anomaly, particularly with regard to its shape and / or severity, and assigns it to a cutting or creasing type based on this evaluation. Classification is an important component of quality assurance for the finished web (e.g., sheets or panels made of corrugated board) and, in particular, enables advantageous order traceability. For example, in the event of a complaint, the predefined cutting and / or creasing types from the order data are compared with the classified anomalies, thereby identifying a possible reason for the complaint.The same applies analogously to the identified and specified processing positions. The classification described here makes it advantageously possible to compare and verify the actual production result with the order data, i.e., to validate it.

[0050] Since several different cutting and / or creasing tools are typically used, correspondingly different cuts and / or grooves, and therefore different path anomalies, are produced. Each cutting and creasing tool has its own characteristics, which lead to a corresponding characteristic (shape and / or severity) of the anomaly it produces. This characteristic, in particular, is the subject of the classification described here. A cut, compared to a groove, regularly has a significantly simpler characteristic. Since a cut is essentially a separation of the path at a specific machining position, it is initially sufficient to simply search for such a separation (e.g., by means of gaps in the measured values). Validation with the job data is therefore generally possible for cuts even without classification.It is also advantageously possible to distinguish between cuts and grooves without classification. Accordingly, in an advantageous embodiment, cuts are first distinguished from grooves as described, e.g., based on gaps in the height profile, and then only those anomalies are classified at whose machining positions grooves were detected.

[0051] The following discussion focuses primarily on the classification of grooves. However, the same principles apply to cuts, as it is generally advantageous to consider the overall characteristics of any anomaly in order to draw conclusions, such as whether a particular cut was executed correctly or to check whether the cutting tool is still functioning properly or needs to be replaced.

[0052] In addition to cutting and creasing types, it is also advantageous to define other types that are recognized during classification. It is particularly useful to define a zero type, which describes a section of the web without either cutting or creasing. This zero type makes it especially easy to identify when no processing has taken place at a given processing position, meaning that any creasing or cutting is missing there. Specifically, during the validation of processing positions, the system checks the height profile at points where processing is specified according to the order data, but no anomaly has been detected, to determine whether the zero type is present at that position. Alternatively or additionally, anomalies at any excess processing positions are classified by comparison with other types and then, for example, recognized as damage.The determination of other types for classification, especially the null type, is carried out in particular analogous to the possibilities for determining cutting and grooving types described below, i.e., for example, by means of modeling, measurement or machine learning.

[0053] In a suitable embodiment, a given anomaly is compared for classification purposes with a representative of a given cut or groove type, and the similarity (i.e., correspondence) between the anomaly and the cut or groove type is calculated using a distance norm. Analogous to the anomaly, the respective representative is a set of data points, in particular a vector of the same length as the anomaly, and represents a typical or ideal form of the corresponding cut or groove type. In other words, analogous to the anomaly, the representative is a set of data points (values ​​as a function of position). A difference is calculated from the data points of the anomaly and the representative at a given position (i.e., component-wise in the case of a vector), and these differences are then summed. For example, the difference is calculated according to d = a - v, and the similarity is then calculated according to... where a and v are the anomaly and the representative respectively, and each has 2-N+1 data points (more precisely: values).

[0054] In another suitable embodiment, the representative is not merely a one-dimensional vector as described, but has one or more additional dimensions besides its position, in particular the groove body spacing. Such a representative is then essentially a set of several one-dimensional representatives parameterized by one or more parameters, such as the groove body spacing. The explanations regarding the one-dimensional representative apply analogously to the comparison with the anomaly.

[0055] The existence of representatives is primarily important for the classifications described below, which are based on modeling and measured values. For classification using a machine learning system, relevant information is incorporated through training the machine, which can also be done using representatives as described.

[0056] Classification with modeling

[0057] In an initial classification approach, representatives are determined using a physical model that simulates the generation of grooves (and / or cuts) based on the geometry of a respective groove body (or cutting element) of the cutting / grooving unit. First, a reference model is created for each groove type based on the corresponding groove body, containing all relevant properties of the groove. This reference model then serves as the representative for classification using the classification module. Each anomaly is then compared with various representatives and classified as the groove type whose representative most closely resembles the anomaly. In a suitable implementation, the representative is generated from the design data of the groove body, for example, by directly using its transverse contour as the representative.Since the grooving body typically does not transfer its contour identically to the path, but only approximately, it is advisable to consider a corresponding transfer function when defining the representative or to determine the representative from the design data or the contour of the grooving body using a finite element method, which then takes into account additional parameters regarding the process of generating a groove with the grooving body.

[0058] The modeling, i.e., the generation of representatives, is advantageously carried out using a machine learning system. This will be described in more detail below.

[0059] Classification based on measured values

[0060] In a second classification approach, the representative of each creasing (or cutting) type is determined based on previous measurements. This is achieved by recording numerous height profiles for each creasing (or cutting) type and then determining the representative for that type, particularly using statistical methods. This approach has the advantage that the representative is more closely aligned with the actual processing result and less so with the creasing body, as previously described. This potentially increases the similarity and thus the reliability of the result. In addition to the creasing type and the contour of the creasing body, the parameters of creasing body spacing and / or web thickness are also considered. These parameters are generally known and, for example, part of the order data. Therefore, they are not subject to any uncertainty.In a suitable implementation, the representative is simply determined as the mean of the elevation profiles, i.e., the mean as a function of the position along the elevation profile. Advantageously, any outliers are eliminated beforehand, for example, by ignoring elevation profiles that deviate from the mean by a certain (second) threshold, e.g., 2°, and then the mean of the remaining elevation profiles is calculated again and used as the representative.

[0061] In a suitable further development, the measured values ​​from different positions are weighted differently when calculating similarity, for example, with a weighting vector g, which contains a weight for each position. The weighting vector is multiplied element-wise by the difference vector before the distance norm is calculated. Advantageously, positions with lower dispersion are weighted more heavily. The dispersion as a function of position is determined from the measured elevation profiles. The dispersion is, for example, simply the width of an envelope at a given position of all elevation profiles, or the variance (i.e., the statistical quantity "variance"). In a particularly advantageous embodiment, the weight vector is expressed using an exponential function as g = e~ b determined where b indicates the width of the envelope at the different positions.

[0062] Classification based on measured values ​​also enables the continuous development of the representatives for training purposes. In a suitable implementation, the anomalies are each stored as a template and tagged by recording the corresponding cutting and / or creasing type. This creates known mappings of anomalies to cutting and / or creasing types, which are particularly suitable as representatives, for creating representatives, or as training data for a machine learning system (see below). Tagging is performed, for example, via user input. The tagged templates therefore contain correct mappings of anomalies to cutting and / or creasing types and are then advantageously used, as described above, for the continuous updating of each representative. Classification with a machine learning system.

[0063] In a third classification approach, a learning machine is used that receives an anomaly (i.e., at least one or exactly one anomaly) as an input parameter and outputs the corresponding cut or groove type as an output parameter. The learning machine is then part of the classification module. The learning machine is preferably a neural network with a large number of neurons, which will be assumed hereafter without loss of generality. In the previously described classification using modeling or based on measured values, the anomaly in the core is compared using a difference calculation and multiple representatives. Based on this comparison, the anomaly is assigned to a corresponding cut or groove type.When classifying anomalies with a machine learning system, the result is also the assignment of the anomaly to a corresponding cut or groove type. However, there is no direct comparison with representatives; rather, the relevant information is encoded within the machine learning system. For this purpose, the machine learning system is suitably trained, specifically through supervised learning. This means the machine learning system is trained with known assignments of anomalies to cut and / or groove types. The representatives already mentioned are also fundamentally suitable for this training.

[0064] The input parameter for the learning machine is a specific anomaly. The input parameters form an input vector; preferably, the anomaly is represented as a vector (which then contains only the values, but not the positions, of the anomaly's data points) and used directly as the input vector. When classifying using a learning machine, it is advantageous to use both height profiles simultaneously as input parameters. This means not only passing an anomaly in the height profile on one side of the track to the learning machine, but also the corresponding section of the height profile (i.e., at the same processing positions) on the other side of the track, i.e., the opposite anomaly. Two such opposite anomalies form an anomaly pair. This doubles the number of input parameters and the length of the input vector accordingly. This is particularly advantageous for classifying grooves.

[0065] The cutting and / or creasing types are also referred to as target classes in the context of the machine learning system. These target classes are made accessible to the machine through a suitable one-hot encoding (or equivalently, one-cold encoding). This means, in particular, that the cutting and / or creasing types, typically named with a string (e.g., "three-point creasing," "one-point creasing," "dot-dot creasing," "no creasing (i.e., null type)"), are each represented by a vector. Depending on the cutting and / or creasing type, this vector contains a high bit at a single position and otherwise only low bits. The number of bits, i.e., the length of the vector, corresponds to the number of cutting and / or creasing types. Optionally, the aforementioned null type and / or other types may also exist. The output parameters of the neural network form an output vector of the same length.

[0066] The activation function for the individual neurons of the neural network is preferably the so-called ReLU function (ReLU = rectified linear unit function), which is defined as relu(y) := max(y, 0). Since this ReLU function returns only a zero value over a certain range of values, individual neurons responsible for recognizing a different cut or groove type in the input vector are effectively deactivated. The classification result is therefore more precise. Furthermore, the ReLU function is advantageously differentiable (except at the origin).

[0067] The neural network generally comprises an input layer for receiving the input vector and an output layer for outputting the output vector. The input and output layers are connected via a number of hidden layers. In a suitable embodiment, the neural network has several, preferably three, hidden layers with a decreasing number of neurons towards the output layer, preferably 300 in the first, 200 in the second, and 100 in the third hidden layer. Such a configuration of the hidden layers has proven particularly advantageous. For training the neural network, a gradient descent method is chosen, for example, and a statistical measure, preferably cross-entropy, is used as the cost function for this.

[0068] In a first advantageous embodiment, the learning machine is trained with correctly identified anomalies (e.g., based on the templates and / or representatives mentioned above) and is then accordingly a supervised training machine. The length of the output vector corresponds, in particular, to the number of cut and / or groove types plus any additional types, so that the output vector indicates how similar the anomaly (or pair of anomalies, if both sides are input simultaneously) passed as an input vector is to a respective cut and / or groove type or other type.

[0069] In a second advantageous embodiment, the learning machine calculates the aforementioned representative of each creasing type (or cutting type), which is then used for classification as described above by comparing a given anomaly with the representative. The representative is also referred to as the comparison signal. The learning machine thus generates a representative for each cutting and / or creasing type and is used for modeling. Advantageously, the learning machine is pre-trained accordingly. In a suitable embodiment, the learning machine described above is essentially used in reverse. For training, the learning machine receives as an input vector, in particular a creasing type or cutting type, e.g., appropriately labeled templates, especially from previous measurements, as well as optionally further parameters such as corrugated board thickness (caliber) or creasing body spacing.The representative then serves as the output vector for the learning machine. The actual classification of an anomaly is not performed by this learning machine, but rather, as described above, and possibly by a second learning machine. In other words, the learning machine calculates a representative, which is then compared separately with a measured height profile to determine its similarity to the representative and thus also the similarity of the groove type (or cut type) to the anomaly. In an advantageous embodiment, the learning machine calculates the representative as a function of position on the one hand and additionally as a function of the groove body spacing on the other.

[0070] Regardless of the specific classification method used, the described optional detection of the zero type already makes it clear that, apart from cuts and grooves, all features of the track can be identified and classified, especially damage to the track. For this purpose, only appropriate representatives or marked measurements (i.e., correctly identified sections of the elevation profile) are used, and the classification is carried out as described. The same applies to the validation of the classification result described below.

[0071] Validation using the validation module

[0072] The previously described assignment and / or classification are advantageously validated using the validation module. This means that the respective assignment of anomalies at the detected machining positions to a cutting and / or creasing type is compared with the order data, specifically with the cuts and / or creasing specified at the given machining positions, and is also evaluated. The validation is based primarily on the assignment of detected machining positions to specified machining positions, which was carried out using the assignment module mentioned above, and on the classification of the anomalies, i.e., their assignment to specific cutting and / or creasing types.

[0073] The validation module appropriately calculates the distances between each detected processing position and the assigned predefined processing position. It then determines an overall measurement (i.e., a single measure for the total deviation) from all these distances, such as the maximum, the sum, the mean, or a similar value. This overall measurement is then compared to a further (third) threshold value. If the overall measurement does not reach the threshold value, the assignment is considered valid (i.e., validated); otherwise, it is considered invalid. Similarly, the order is considered correctly executed or faulty with respect to the processing position. The validation module then outputs a corresponding message. Preferably, it outputs for each individual predefined processing position whether it is valid or not, i.e., whether it was executed correctly or not.

[0074] Alternatively or additionally, the validation module calculates the similarity of each detected cutting and / or creasing type to the cutting and / or creasing type required (i.e., specified) in the order data for that processing position. This similarity is then compared, preferably separately for each specified processing position, with a further (fourth) threshold value. In this process, a measurement is determined from each distance (i.e., one measurement for each distance and thus for each pair of specified and actual processing position), for example, by simply using the distance itself as the measurement. If the similarity, i.e., the measurement, reaches the threshold value, the classification is considered valid; otherwise, it is considered invalid. Similarly, the order is considered correctly executed or faulty with regard to the cutting and / or creasing type. The validation module then issues a corresponding message.In principle, it is also advantageous to derive a single overall measure (see above) from the several similarities, e.g. the mean value, and to compare only this with the threshold value in order to validate the entire task in a simple way.

[0075] In case of discrepancies between the detected processing positions and / or the detected cutting and / or grooving types and the order data, a notification is expediently issued to an operator via an output element, e.g., as part of a human-machine interface (HMI), or forwarded to a control room of the plant. Preferably, such information about discrepancies, along with suitable key figures (timestamp, value, etc.), is stored in a database for further analysis or preprocessors (machine learning, training for a learning machine, predictive and prescriptive analytics).

[0076] Control (closed-loop approach): The actual and the specified machining positions, especially in conjunction with their similarity to each other, are also suitable for determining actual values ​​for a control system. Within this system, the cutting and creasing unit, and more generally any part of the system, is controlled based on a comparison with specified target values, specifically to optimize the similarities, i.e., to better align the actual machining positions with the specified machining positions. In a suitable configuration, a control unit generates a control signal based on the actual and specified machining positions. Using this control signal, the control unit controls the cutting and creasing unit in such a way that the similarities between the actual machining position and the specified machining position are optimized, and in particular, maximized.The control unit (e.g., higher-level control system module) is preferably part of the plant's control unit, but can also be designed separately.

[0077] The control unit determines a number of actual values ​​based on the actual and / or predefined machining positions, which are then compared to a corresponding number of target values. Based on this comparison, the control unit generates an error signal, which ultimately measures how well the actual and predefined machining positions correspond. The control unit then uses this error signal to generate the control signal for the cutting and creasing unit. This control signal then influences a corresponding manipulated variable to align the actual values ​​with the target values, thereby optimizing their similarity. In this way, a closed-loop control system is implemented, which automatically corrects and minimizes any deviations, particularly in real time.This is based on the understanding that measuring and interpreting the height profile (and thus the surface quality), as well as the correct execution of grooves, cuts, and other machining patterns, are crucial for automating and optimizing the system and the entire production process, especially in conjunction with a control system. Firstly, the control unit advantageously performs targeted error correction (optimization of similarities) in real time when a deviation is detected. Secondly, wear and tear or incorrect use of tools are advantageously detected automatically, which can ultimately even result in a production stoppage to prevent (further) scrap.

[0078] The control process is advantageously implemented as quickly as possible, ideally in real time, to minimize waste. The control unit therefore does not simply adjust from job to job or roll to roll, but advantageously continuously, i.e., even within a single job, and potentially even from panel to panel (which are not yet separated on the cutting and creasing machine).

[0079] The following describes two suitable methods for generating a control signal based on the actual and predefined machining positions. The key difference between the two methods lies in the selection of actual and target values.

[0080] In a first suitable variant, a number of actual values ​​are determined from the actual machining positions, corresponding to the distances between these positions. In other words, the actual machining positions lie next to each other in the transverse direction and are spaced apart in pairs. The number of these spacings is, by definition, exactly one less than the number of actual machining positions. These spacings are also referred to as relative distances, as they each specify the distance between two machining positions that are consecutive in the transverse direction. Each distance is thus defined relative to one of the cutting and / or creasing bodies, which marks one end of this distance. Equivalently, the distances can also be defined relative to one or more other reference points, e.g.,relative to the drive or operator side, or to the center of the path, or multiple distances relative to a single cutting or creasing element. For these distances, corresponding target values ​​are now specified; these are suitably calculated analogously to determining the actual values ​​from the specified machining positions, namely, for example, as the difference between two specified machining positions.

[0081] In a second suitable variant, a number of actual values ​​are determined from the actual machining positions and the predefined machining positions. Each of these values ​​corresponds to a distance between one of the actual machining positions and its corresponding predefined machining position. Thus, exactly as many distances are determined as there are pairs of actual and assigned predefined machining positions. These distances then directly indicate how well the actual machining positions correspond to the predefined machining positions. These distances are therefore also referred to as absolute distances. Essentially, the absolute distances are positional differences between predefined and actual machining positions and are therefore particularly suitable as input parameters for the control system, i.e., as actual values. These distances are, in particular, a measure of the similarity between an actual and a predefined machining position.The target values ​​are then, in particular, specified distances; typically, the target values ​​are 0, but alternatively, a target value other than 0 is specified to allow for a corresponding tolerance, e.g., 1 mm, which is defined, for example, by the order data.

[0082] In principle, relative distances can also be calculated from absolute distances, and vice versa. Both of the methods presented above ultimately lead to the same result: the absolute distances are minimized, and the relative distances are set to specific, predetermined values. The two methods can also be combined.

[0083] Regardless of the chosen variant, each actual value is compared with a predefined target value, and the control signal is then generated based on this comparison result (e.g., the difference). The control signal is specifically designed to move one or more cutting and / or creasing elements of the cutting / creasing unit transversely to the conveying direction. In a suitable embodiment, all cutting and / or creasing elements are mounted on a common axis and move along this axis. However, an embodiment in which the cutting and / or creasing elements are mounted on several axes, particularly parallel ones, is also possible and suitable, e.g., on both sides of the web and / or one behind the other in the conveying direction.Every anomaly, and thus every detected actual machining position, is generated by a cutting or creasing element, unless it is an anomaly caused by another factor, such as a defect. If the cutting or creasing element is incorrectly positioned in the transverse direction, a corresponding discrepancy arises between the actual and the predefined machining position. The absolute and relative distances described above are therefore suitable, directly or indirectly, as error signals for controlling the position of the respective cutting or creasing element in the transverse direction. Accordingly, the cutting and / or creasing elements of the cutting / creasing unit are adjusted in the transverse direction depending on the relative and / or absolute distances. Each relative or absolute distance is inherently assigned to a predefined machining position and thus also to a specific cutting or creasing element.The position of this cutting or creasing element is now controlled depending on the distance, i.e., generally the respective actual value, in order to minimize the distance and, in particular, to keep it at least below a predetermined tolerance. For this purpose, a suitable direction of movement (in the transverse direction, i.e., towards the drive side or the operator side) and distance (also called path, likewise in the transverse direction) are determined for each cutting or creasing element, depending on the actual value compared to the target value. The control signal is then generated, which moves the cutting or creasing element by the specified distance in the direction of movement. In one embodiment, the cutting or creasing element is moved while it is engaged with the path. In another embodiment, the cutting or creasing element is moved when it is not engaged with the path, i.e.,It may even happen that the machine first moves out of the path, then moves laterally, and finally returns to the path. For the described control of the cutting and / or creasing body's position in the transverse direction, depending on the actual values, four possible error patterns and their consequences for the control system are described below. In the first error pattern, a single cutting or creasing body is incorrectly positioned, resulting in a corresponding absolute distance. The two adjacent processing positions, or neighbors, are correct. Due to the absolute distance, the relative distance to one neighbor is therefore greater than specified, and the relative distance to the other neighbor is less than specified. The position of the cutting or creasing body is corrected accordingly; the adjacent cutting and / or creasing bodies remain unaffected.In a second error pattern, only a relative distance is incorrect; that is, starting from one side (operating side or drive side), several consecutive cutting and / or creasing tools are incorrectly positioned and are corrected accordingly. While the relative distances between these cutting and / or creasing tools may be correct, the relative distance to the next, correctly positioned cutting or creasing tool is incorrect. Therefore, all incorrectly positioned cutting and / or creasing tools are corrected together. In a third error pattern, several relative distances are incorrect, as described in the second error pattern; that is, several cutting and creasing tools that are not adjacent to each other are incorrectly positioned. These are then corrected accordingly. In a fourth error pattern, several adjacent relative distances are incorrect.Several adjacent cutting and / or creasing bodies are incorrectly positioned. The cutting and / or creasing bodies are then moved transversely as needed, parallel to each other (with a constant distance), towards each other, or away from each other. This repositioning is expediently performed from the center of the path towards its lateral edges (on both the drive and operator sides).

[0084] Fundamental to the control system is, in particular, the most comprehensive possible measurement and determination of actual values ​​based on these measurements, as well as a comparison of the actual values ​​with the target values. This information, i.e., the actual and target values, is interpreted accordingly by the control unit; that is, the control unit generates a suitable control signal to ultimately align the actual values ​​with the target values. Ideally, all information generated from measurement, data processing, and validation is used by the control unit to control the system, specifically the cutting and creasing unit.

[0085] In a suitable embodiment, the actual and specified machining positions are transmitted to the control unit in real time via a suitable interface using a suitable protocol. The control unit then determines the actual values ​​and compares them with the target values, which are also transmitted to the control unit or generated by the unit itself. Each target value is preferably also assigned a tolerance (possible for both variants). If the tolerances are met, production (operation of the system) continues; however, any deviations are expediently logged and stored in a suitable database (data storage system) for later use. Furthermore, a notification is expediently issued simultaneously to relevant points on the system (e.g., at the operator console or control panel).This also gives the machine operator the opportunity to intervene manually and make corrections if necessary. However, if the control unit detects a deviation outside the tolerances, it makes a corresponding correction to the ongoing production so that it is not otherwise affected. In other words, the control unit controls the machine in such a way that the deviation is reduced to below the tolerance. If the deviation exceeds a further critical value, which is particularly greater than the tolerance, production is stopped. For example, an absolute distance of up to 1 mm between the specified and actual machining position, or a relative distance deviation of 1 mm from the target value, is still tolerated; then the tolerance is 1 mm. Above this, active intervention occurs, and countermeasures are taken as described. However, with a deviation of, for example, 2 mm, i.e., twice the tolerance, the machine is stopped.All information and values ​​generated for and as part of the control system are preferably logged in a database (e.g., the aforementioned database) and / or made available to the operating personnel at the system. The database, in particular, is executed locally (on-premise) and / or cloud-based. In all cases, in addition to the measured and sensor values, one or more of the following pieces of information are also appropriately stored in computer-aided (mathematical / AI) models: date, time, machine or system component or part identification, and other statistically and / or control-relevant information, especially for further, subsequent processing.

[0086] Preferably, the control unit includes an AI component (AI = artificial intelligence) that allows customer-specific production characteristics or peculiarities to be taken into account using (pre-)trained models based on empirical data, and the operation, i.e., the production, to be adapted or even optimized accordingly. This information is advantageously also incorporated into the control system as needed.

[0087] Analogous to the described control based on distances, a control based on the groove depth (depth of the groove) at the machining positions is also advantageous, either alternatively or additionally, especially for grooving. At an actual machining position, the groove has an actual depth. For a given machining position, a target depth is specified. For each pair of actual and target machining positions, the corresponding actual depth can then be compared with the target depth, analogous to the distance control, and their difference determined. Based on this difference, the control unit then controls the cutting / grooving unit to adjust the actual depth to the target depth.In a suitable embodiment, the actual groove depth is measured for at least one actual machining position, and the cutting / grooving machine is controlled based on the difference between the actual depth and a target depth in order to minimize this difference. The actual depth is taken directly from the height profile, for example, as the difference between a maximum and a minimum of the height profile. Alternatively or additionally, the actual depth is measured using pattern matching; that is, during classification, different patterns with varying groove depths are stored for the same groove type, so that the classification directly provides the groove depth. Alternatively or additionally, the groove depth is measured using a confidence level during classification, i.e., determined by ascertaining how similar the groove is to the most similar groove type.The target depth corresponds to the grooving depth specified for the given machining position, which is assigned to the actual machining position. This enables grooving depth detection and automatic control for maintaining the grooving depth. The respective grooving die is automatically advanced into and out of the path as needed, thus regulating the grooving depth to the specified target depth. The grooving depth typically differs for different grooving types; this relationship is documented in a corresponding table.

[0088] Apart from the described, special designs in which the distances are used to control the cutting and / or grooving elements of the cutting / grooving unit, it is also generally advantageous to transmit the deviations between actual and specified machining positions to the control unit in order to then generate a suitable control signal for a part of the system with which these deviations can be influenced.

[0089] Other aspects

[0090] Based on the anomalies, in conjunction with the specifications according to the predefined machining positions, the wear condition of the cutting and / or creasing elements of the cutting / grooving unit can advantageously be determined. In a suitable embodiment, the wear condition of a cutting or creasing element of the cutting / grooving unit is detected based on at least one anomaly—namely, the cutting or creasing element that produced the anomaly. A predefined machining position, and thus also a cut or creasing type with a corresponding depth and shape, is assigned to the anomaly. If the depth and / or shape of the anomaly do not exactly correspond to the depth and shape specified, then the corresponding cutting or creasing element may be worn and requires maintenance.Furthermore, the method described here is inherently robust against incorrect assignment / classification due to its specific classification and assignment. However, a less than optimal similarity regularly indicates wear of the cutting or grooving element and is evaluated accordingly. For example, a cutting element (blade) wears down over time (becomes blunt, for instance), so that the path may no longer be completely cut. A cut is specified for the given machining position, but only an incomplete cut with limited depth is detected. The similarity is still sufficient for a reliable assignment of this actual machining position to the specified machining position, but at the same time, the similarity is not maximum, which suggests wear of the cutting element.Alternatively or additionally, a given cutting or creasing type is stored or trained multiple times for classification, namely for several different wear states of the corresponding cutting or creasing tool, so that the classification of the anomaly automatically also determines the wear state of the cutting or creasing tool. As soon as the wear state reaches a predefined limit, a corresponding message is expediently issued, or maintenance, preferably automatic, such as resharpening or replacement, of the cutting or creasing tool is carried out.

[0091] The height profile measurement described here allows not only the identification of anomalies but also the measurement of the web's actual caliber. Caliber generally refers to the web's thickness, i.e., its dimension perpendicular to both the conveying and transverse directions, disregarding any processing positions. The caliber may be influenced during processing within the system, for example, by web compression or depending on a wrap angle that varies during operation. Therefore, measuring the actual caliber is generally advantageous. Accordingly, in a suitable configuration, the actual web caliber is determined based on the height profile. The actual caliber is expediently compared with a target caliber as part of a caliber control system. Depending on any difference between the actual and target calibers, a section of the system is then appropriately controlled to reduce this difference.In particular, this control is a caliber-influencing setting, e.g., a contact pressure or a wrap angle. The details of the caliber control are of secondary importance here; what is more important is that the measurement of the height profile has an additional use, namely, in addition to identifying anomalies, also to determine the actual caliber. In principle, it is conceivable to use only the upper or the lower height profile to determine the actual caliber, e.g., within the framework of a relative determination of the actual caliber. However, a preferred embodiment is one in which both the upper and the lower height profiles are used to determine the actual caliber, e.g., simply as the difference between the two height profiles, whereby any machining positions, i.e., anomalies, are disregarded.

[0092] Preferably, the system includes an input element, e.g., also as part of the HMI, via which the output instructions, measured height profiles, and the results of position detection, assignment, classification, and / or validation can be labelled by an operator with additional information, e.g., key figures or identifiers, particularly "just-in-time" (also referred to as "labeling"), in order to supplement or make available relevant information on production, process, and product quality, and specifically to generate the templates mentioned above. This information is preferably stored in a database and used for further evaluations and learning processes, e.g., the machine learning training described above.

[0093] The information mentioned is also suitable for being passed on to one or more higher-level systems in order to be used outside of the operation of the plant and the processing of the railway, e.g. in production planning or the generation or adaptation of order data.

[0094] The aforementioned database is implemented, for example, as a cloud solution and thus separately from the system, or as part of it and therefore as an on-premise solution. The process itself, and in particular the computer program, are expediently executed on the system itself. Alternatively or additionally, the process itself, and in particular the computer program, are implemented as a cloud solution to which the system is connected.

[0095] The computer program product according to the invention comprises instructions which, when executed by a computer, cause the computer to execute, in particular, the method described above, either partially or completely. In a suitable embodiment, when executed by a computer, the instructions cause the computer to identify a number of anomalies in a measured elevation profile of a railway by identifying those sections of the elevation profile as anomalies in which the elevation profile reaches a predetermined threshold value, and then classifying each anomaly by assigning it to a cut and / or groove type (i.e., one of several cut and / or groove types) to which the anomaly is most similar.

[0096] Exemplary embodiments of the invention are explained in more detail below with reference to a drawing. Each drawing schematically shows:

[0097] Fig. 1 a method,

[0098] Fig. 2 a process diagram,

[0099] Fig. 3 a system,

[0100] Fig. 4 shows a section of two height profiles,

[0101] Fig. 5 shows a grooved body in a cross-sectional view,

[0102] Fig. 6 shows two views of a sensor unit of the system from Fig. 3.

[0103] Fig. 7 shows an elevation profile and its spectrogram.

[0104] Figs. 8 to 10 each show a comparison of two height profiles, each with one representative example; Fig. 11 shows a multitude of height profiles for the same groove type.

[0105] Fig. 12 a learning machine,

[0106] Fig. 13 a representative,

[0107] Fig. 14 shows the system from Fig. 3 in a different representation.

[0108] Fig. 1 shows an embodiment of a method according to the invention. Fig. 2 then shows a process diagram with further details of the method. Fig. 3 shows an exemplary system 2 in which the method is implemented. In a first step S1, a web 4 is provided into which a number of operations 8, 10 have been made by means of a cutting / creasing unit 6 of the system 2, each according to a predetermined cutting type or crease type. Here, and also generally, "a number of" means "one or more" or "at least one." The operations 8, 10 are mechanical operations on the web 4 and, in this case, each is either a cutting 8 or a crease 10 in the web 4. Each cutting 8 was produced according to a predetermined cutting type and each crease 10 according to a predetermined crease type.The cuts 8 and scoring 10 are used to prepare the web 4 into individual panels or sheets and to form break, fold or crease edges of the panels or the later sheets, e.g. to form them into a box or similar.

[0109] In the illustrated embodiment, the cutting and creasing unit 6 both cuts and creases the web 4. The system 2 is a corrugated board production line for manufacturing corrugated board, specifically sheets. The web 4 is made of paper and is a corrugated board web composed of several layers of paper. As can be seen in Fig. 3, a height profile 14, 16 of the web 4 is measured inline downstream of the cutting and creasing unit 6 by means of a sensor unit 12; more precisely, two height profiles 14, 16 are measured: an upper height profile 14 on the top side of the web 4 and a lower height profile 16 on the underside of the web 4. The web 4 is processed by the cutting and creasing unit 6 at a number of processing positions 18 (i.e., cutting and / or creasing positions). An anomaly 20 is formed at each actual processing position 18 due to the system's design.These actual processing positions 18 are detected together with the anomalies 20 and are then also referred to as detected processing positions 18. The processing positions are defined by order data 22 for an order for plant 2. Accordingly, the order data 22 then contains predefined processing positions 24. The cutting or creasing type to be performed at each processing position 24 is also defined in the order data 22, so that a predefined cutting or creasing type is assigned to each predefined processing position 24 in the order data 22. The cutting / creasing unit 6 is controlled according to the order data 22.However, due to errors or inaccuracies, the cutting and grooving unit 6 does not necessarily produce the machining operations 8, 10 in such a way that the specified machining positions 24 correspond to the actual machining positions 18, and the required cutting or grooving type may not be reproduced exactly, or even an incorrect cutting or grooving type may be used.

[0110] In the method described here, in a second step S2, the sensor unit 12 measures a height profile 14, 16 on both sides of the track 4 and transversely to a conveying direction F of the track 4. An example of such height profiles 14, 16 is shown in Fig. 4, in which two anomalies 20 are visible at each of two actual processing positions 18, i.e., a total of two pairs of anomalies, one of which belongs to a cut 8 and the other to a grooving 10, more precisely a three-point grooving. Further height profiles 14, 16 are also shown in Figs. 7 to 10.

[0111] As can be seen in Fig. 4, the machining operations 8, 10 performed by the cutting and grooving unit 6 are identifiable as anomalies 20 in the respective height profiles 14, 16, i.e., as deviations from a normal state of the path. In a third step S3, the anomalies 20 are identified in the height profiles 14, 16 by identifying those sections of the height profiles 14, 16 as an anomaly 20 in which the respective height profile 14, 16 reaches a predetermined (first) threshold value 26, i.e., depending on the perspective, exceeds or falls below it. As indicated by the dashed frames in Fig. 4, the section is not merely the area in which the height profile 14, 16 actually reaches the threshold value 26, but includes a specific allowance on both sides.

[0112] In Fig. 4, reaching the threshold value 26 is particularly clear for the groove 10. The cut 8, however, is not necessarily recognized as shown, but can also be identified in other ways, as will be described below.

[0113] In a fourth step S4, each anomaly 20 is classified by assigning it to the cutting and / or creasing type to which it most closely resembles. Thus, the process does not merely check whether or to what extent a given processing operation 8, 10 corresponds to the cutting or creasing type according to the order data 22, but rather verifies which cutting or creasing type is actually present (possibly with a confidence value). This classification therefore differs from a simple target / actual comparison in that the anomaly 20 is not compared with just a single cutting or creasing type, but with several different ones. Furthermore, by using the threshold value 26, the entire height profile 14, 16 is actively searched for, rather than just the height profile 14, 16 at the specified processing positions 24.

[0114] For cutting and / or grooving, the cutting / grooving unit 6 has one or more corresponding cutting elements 27 and / or grooving elements 28. In both cases, the path 4 is mechanically processed; in the case of cutting, this is merely a separation, and in the case of grooving, a deformation. A cutting element 27 typically has at least one blade, which is inserted into the path 4 to produce a cut 8. Similarly, a grooving element 28 regularly has two profiled running surfaces 30, 32, which are arranged on opposite sides of the path 4 and interact. An example of a grooving element 28 for three-point grooving is shown in Fig. 5. By displacing the running surfaces 30, 32 of a grooving element 28 in the transverse direction Q, so-called offset grooving is also possible. This displacement is also referred to as the offset.A further shift is regularly also possible perpendicular to both the transverse direction Q and the conveying direction F in order to adjust the so-called groove body spacing 34.

[0115] Various aspects and configurations of the procedure are described in detail below, with particular reference to Fig. 2. To evaluate an elevation profile 14, 16 and to detect anomalies 20 therein, the elevation profile 14, 16 is first processed and filtered using preprocessing 36, which employs digital signal processing methods. Subsequently, processing positions 18 detected in the elevation profile 14, 16 are identified using position detection 38, which employs statistical methods; thus, the anomalies 20 are also identified here. Finally, the detected processing positions 18 are assigned 40 to the predefined processing positions 24 from the order data 22. The position detection 38 and the assignment 40 are shown in Fig.2. The process is carried out separately, once for cuts 8 and once for grooves 10; the anomalies 20 are divided into two corresponding groups by the preprocessing 36.

[0116] For further analysis, the aforementioned classification 42 is performed in the method described here. This classification categorizes the height profile 14, 16 section by section at the identified machining positions 18 with respect to the machining operations 8, 10 performed; that is, the anomalies 20 are classified. In the embodiment shown in Fig. 2, only the grooves 10 are classified; the cuts 8 are not classified, but this is also possible as an option. The cuts 8, or more precisely their machining positions 18, are validated here only by a validation 44. Optionally, the result of the classification 42 can also be validated; however, this is not explicitly shown in Fig. 2, but is possible analogously to the aforementioned validation 44.During validation 44 of the machining positions 18, a warning is issued, system 2 is stopped, or another suitable measure is initiated as soon as it is determined that the identified machining positions 18 do not match the specified machining positions 24 from the order data 22 within a given tolerance. The validation of the result of classification 42 based on the order data 22 is carried out analogously to the validation 44 of the machining positions 18; that is, it is checked whether the correct cutting or creasing type was applied and / or executed correctly, and whether the cuts 8 and grooves 10 were produced with the correct cutting or creasing body 28.As soon as it is determined during this validation that the detected processing operations 8, 10 do not match the specified cutting and / or grooving types from the order data 22 within a given tolerance, a corresponding message is issued, the system 2 is stopped or another suitable measure is initiated.

[0117] Measurement (first step S1)

[0118] The height profile 14, 16 is measured, for example, with a sensor unit 12 as shown in Fig. 6. In Fig. 6, the sensor unit 12 is shown on the left, viewed in the conveying direction F, and on the right, viewed in the transverse direction Q, perpendicular to the conveying direction F. The sensor unit 12 has two sensors 46, in this case distance sensors, specifically laser triangulation sensors, which are movable transversely to the track 4 and measure a distance in the direction of the track 4 (i.e., towards it), i.e., the distance between sensor 46 and track 4. For each measurement position transverse to the track 4, the sensor 46 outputs a measured value (distance), which varies depending on the surface characteristics of the track 4. As a result, each sensor 46 outputs a height profile 14, 16 containing a series of measured values. Each measured value is also assigned a position P (measurement position) along the track 4. A measured value, together with its corresponding position P, forms a data point.The sensor unit 12 is arranged directly downstream of the cutting / grooving unit 6. Furthermore, the sensor unit 12 shown here has a pressure roller 48 to press the web 4 against a support 50 as it passes through the sensor unit 12 and when measuring the height profile 14, 16, and to suppress any curvature of the web 4 as much as possible, at least during the measurement. Pre-processing (second step S2).

[0119] During preprocessing 36, the measured height profile 14, 16 is appropriately prepared for the subsequent position detection 38. For example, it is possible that the measurement of the height profile 14, 16 exceeds the measurement range, preventing a distance from being measured. This regularly occurs at a section 8, as the path 4 is interrupted at its processing position 18. Accordingly, if the measurement range is exceeded, a valid measurement value may not be generated; instead, a NaN (not a number) is returned as the measurement value, and a corresponding gap is found at the associated position P in the height profile 14, 16. During preprocessing 36, such gaps in the height profile 14, 16 are specifically removed in a block 52 to obtain a complete height profile 14, 16. However, the gaps are not simply omitted but replaced with suitable values.Furthermore, in block 52, the elevation profile 14, 16 is mapped onto an equidistant grid. New data points with new (measured) values ​​and new positions P are generated from the measured values ​​and measurement positions, and these are distributed along an equidistant grid.

[0120] The preprocessing 36 in Fig. 2 also includes a further block 54 in which the gaps and their positions P are identified.

[0121] The anomalies 20 are also recognizable in the frequency domain, specifically in the spectrogram of the height profile 14, 16. An exemplary spectrogram is shown in the lower part of Fig. 7, and the corresponding height profile 14, 16 is shown in the upper part. In Fig. 7, the data point density of the height profile 14, 16 is, by way of example, ten data points per millimeter, resulting in a spectrogram in a frequency range of 0 to 5 mm. -1(in the vertical direction). As can be clearly seen in Fig. 7, continuous lines appear in the spectrogram at the exact same positions P as the actually executed cuts 8 (parallel to the vertical frequency axis). This means that a cut 8 is identifiable across the entire frequency spectrum and that by filtering out a suitable sub-range from the frequency spectrum (i.e., by selecting a specific frequency range 56) containing only the frequencies of a cut 8, the influence of this cut 8 on the measured height profile 14, 16 can be isolated and thus enable the identification of the cut 8. In one possible embodiment, a corresponding frequency range 56 is therefore isolated using a bandpass filter B. In Fig. 7, a suitable frequency range 56 extends from 1.5 mm. -1 up to, in particular, 5 mm -1, since in this frequency range 56 only the effect of the cuts 8 is visible in the spectrogram, and all other influences on the height profile 14, 16 lie completely or at least predominantly below this frequency range 56. Accordingly, the upper limit is not strictly necessary; rather, it is determined by the data point density. Similarly, at the processing positions 18 of the grooves 10, anomalies in the frequency range 58 of 0 mm are also visible in the spectrogram. -1 up to approximately 0.1 mm -1 It is apparent that, in one possible embodiment, a corresponding frequency range 58 is isolated using a bandpass filter B. In Fig. 7, a suitable frequency range 58 extends from 0.001 mm -1 (to suppress the DC component in the height profile 14, 16 at 0 mm) -1 ) to 0.1 mm -1 .

[0122] Position detection (third step S3)

[0123] The threshold 26 is, in this case, a standard deviation of the respective elevation profile 14, 16. The standard deviation forms a threshold 26, which depends on the position P and effectively forms an envelope (see Fig. 4) around the elevation profile 14, 16. If the elevation profile 14, 16 breaks through this envelope, i.e., reaches the threshold 26, then an anomaly 20 exists at this position P and is identified as such. This identification of an anomaly 20 by reaching the threshold 26 is shown in block 60 in Fig. 2. The threshold 26, and especially its determination, benefits from the filtering of the elevation profile 14, 16 by means of a bandpass filter described above, since this limits the elevation profile 14, 16 to the respective anomaly 20 being sought (sections 8 or grooves 16) and facilitates its identification using statistical methods.The positions P of the identified anomalies 20 in the elevation profiles 14 and 16 are also considered candidates for recognized processing positions 18, in order to be subsequently assigned to a predefined processing position 24 using assignment 40. For the detection of sections 8, anomalies 20 are identified separately in both elevation profiles 14 and 16 in block 60. An anomaly 20 is then recognized as a section 8 at a processing position 18 if an anomaly 20 has been identified at that processing position 18 in both the upper and lower elevation profiles 14 and 16. Additionally, in block 62 (detection based on a gap), an anomaly 20 at a processing position 18 is recognized as a section 8 if a gap exists at that processing position 18 in either the upper or lower elevation profile 14 and 16. For this purpose, the positions P identified therein, including gaps, are transferred from block 54 to block 62.The aforementioned criteria are linked together, so that the presence of one of the criteria is sufficient to identify a cut 8.

[0124] To detect grooves 10, a difference height profile is calculated from the lower and upper height profiles 14, 16 and then filtered with a bandpass filter B as described above. If the difference height profile reaches the threshold value 26 at a processing position 18 during identification in block 60, a groove 10 is detected at this processing position 18.

[0125] Using the two aforementioned steps for recognizing cuts 8 on the one hand and grooves 10 on the other, the identified anomalies 20 are divided into two groups before classification 42: a first group containing all anomalies 20 that were identified as cuts 8 (left in Fig. 2), and a second group containing all anomalies 20 that were identified as grooves 10 (right in Fig. 2). The anomalies 20 are thus pre-sorted. In Fig. 2, only the group of anomalies 20 identified as grooves 10 is then classified in order to assign a specific groove type to each of these anomalies 20.

[0126] Assignment of processing positions

[0127] Each anomaly 20 is assigned an actual processing position 18, which is identified by the anomaly 20 and is then also referred to as the identified processing position 18. Within the framework of the assignment 40, the actual processing positions 18 are assigned to the predefined processing positions 24 and compared. For this purpose, a correlation table is generated, for example, which contains the similarity of each actual processing position 18 to each predefined processing position 24. For instance, the correlation table is a simple distance table, which contains the distances of each actual processing position 18 to each predefined processing position 24. Each actual processing position 18 is then assigned the predefined processing position 24 that is most similar to it, i.e., that exhibits the greatest similarity to it.to which it has the smallest distance. This avoids duplicate assignments, so that each actual processing position 18 is assigned to at most one predefined processing position 24, and vice versa.

[0128] The comparison also identifies any excess processing positions, i.e., processing positions 18 that have been detected but cannot be assigned to any predefined processing position 24 in the order data 22. These can also be, for example, anomalies 20 that were not generated by the cutting / grooving unit 6. Conversely, missing predefined processing positions 24 are detected analogously. For this purpose, block 64 is specifically used in Fig. 2, which generates feedback for the assignment 40 in order to adjust one or more parameters (e.g., threshold 26, limit value (see below), search area in the height profile 14, 16) for identifying the anomalies 20, if necessary. In this case, a limit value for similarity is also defined for the assignment 40, which must be reached at a minimum to perform an assignment.If distance is used as a measure of similarity, the limit value is accordingly a maximum distance (e.g.

[0129] 5 mm). This avoids a forced assignment.

[0130] Classification (fourth step S4)

[0131] The identified processing positions 18, which were optionally also assigned to predefined processing positions 24 as described, are now classified and thereby each assigned to one of several cutting or creasing types. Since several different creasing bodies 28 are typically used, correspondingly different creasing patterns 10 and thus, depending on the job, also different anomalies 20 are generated (analogous for cuts 8). Each creasing body 28 has its own characteristic, which leads to a corresponding characteristic of the anomaly 20 generated by it. A cut 8 regularly has a significantly simpler characteristic compared to a creasing pattern 10. Since a cut 8 is essentially a separation of the web 4 at a specific processing position 18, it is initially sufficient to only search for such a separation (e.g., by means of gaps in the measured values). Validation with the job data 22 is—as shown in Fig.2. Recognizable – for cuts 8, therefore, classification is generally possible even without classification. It is also generally possible to distinguish 42 cuts 8 from grooves 10 without classification. Accordingly, in the embodiment shown here, cuts 8 are first distinguished from grooves 10, and then only those anomalies 20 are classified at whose machining positions 18 grooves 10 were detected. Therefore, the classification 42 of grooves 10 is primarily described below. However, the explanations apply analogously to cuts 10.

[0132] For classification 42, in one possible embodiment, a given anomaly 20 is compared with a representative 66 of a given groove type, and the similarity (i.e., correspondence) between the anomaly 20 and the groove type is calculated using a distance standard. This is shown by way of example in Figures 8, 9, and 10, which each show in their upper part a comparison of two height profiles 14, 16, each with three different representatives 66 (i.e., a total of six representatives 66), whereby two representatives 66 always form a pair and are assigned to a specific groove type. The height profiles 14, 16 are identical in all three figures 8, 9, and 10. The groove types shown are: dot-flat grooving in Fig. 8, dot-dot grooving in Fig. 9, three-point grooving in Fig. 10. In the lower part, the result pairs of the comparisons are shown as two bars, namely three comparisons each of the upper and lower height profile 14, 16 with a corresponding representative 66.A difference was calculated from the data points of each anomaly 20 and its respective representative 66 at a given position P (i.e., component-wise for a vector), and these differences were then summed for each of the two height profiles 14, 16. The two sums, one for the upper and one for the lower height profile 14, 16, are then represented as bars. Comparing the three pairs of bars, it becomes clear that the lowest overall sum (sum of both sums of the differences) is obtained for the three-point profile in Fig. 10, indicating the greatest similarity. Both anomalies 20 (i.e., the pair of anomalies) are therefore assigned the groove type three-point grooving.

[0133] Classification with modeling

[0134] In one possible embodiment for classification 42, the representatives 66 are determined using a physical model that simulates the generation of grooves 10 based on the geometry of a respective groove body 28. First, a reference model is created for each groove type based on the associated groove body 28, containing all relevant properties of the corresponding groove 10. The reference model then serves as a representative 66 for classification 42. A given anomaly 20 is then compared with various representatives 66 (see Figs. 8, 9, and 10) and classified as the groove type whose representative 66 is most similar to the anomaly 20. In one possible embodiment, the representative 66 is generated from the design data of the groove body 28, for example, by directly using its contour in the transverse direction Q, as shown in Fig. 5, as the representative 66.

[0135] Classification based on measured values

[0136] In another possible embodiment of the classification 42, the representative 66 of a respective groove type is determined based on previous measurements, e.g., by recording a large number of height profiles 14, 16 for a given groove type, as shown in Fig. 11, and then determining the representative 66 for this groove type from these, for example, simply as the mean of the height profiles 14, 16. Optionally, any outliers are eliminated beforehand, e.g., by ignoring height profiles 14, 16 that deviate from the mean by a certain threshold, and then calculating the mean of the remaining height profiles 14, 16 again and using it as the representative 66. Also optionally, the measurements from different positions P are weighted differently when calculating the similarity before the distance norm is determined as described above.In particular, positions P with lower dispersion are weighted more heavily; in Fig. 11, these are primarily the positions P where the upper elevation profile 14 reaches a minimum and the lower elevation profile 16 reaches a maximum (two maxima in Fig. 11).

[0137] Classification using a learning engine

[0138] In another possible embodiment for classification 42, a learning machine 68 is used, which receives an anomaly 20 as an input parameter and then outputs the corresponding groove type as an output parameter. An exemplary embodiment for the learning machine 68 is shown in Fig. 12, in which the learning machine 68 is a neural network with a plurality of neurons 70. The learning machine 68 is trained in this case by means of supervised training.

[0139] The input parameter for the learning machine 68 is a specific anomaly 20, i.e., a segment of the elevation profile 14, 16. The input parameters form an input vector. In this case, both elevation profiles 14, 16 are used simultaneously; that is, not only is an anomaly 20 in the elevation profile 14, 16 used on one side of track 4, but also the opposite anomaly 20. This doubles the number of input parameters and the length of the input vector accordingly. The cut and / or groove types are made accessible to the learning machine 68 via one-hot coding. The output parameters of the neural network then form an output vector with a length corresponding to the number of groove types. The ReLU function, for example, is used as the activation function for the individual neurons of the neural network.The neural network has an input layer 72 for receiving the input vector and an output layer 74 for outputting the output vector. The input layer 72 and the output layer 74 are connected via a number of hidden layers 76, e.g., with a decreasing number of neurons 70 towards the output layer 74. The learning machine 68 is trained, for example, with correctly identified anomalies 20 and is then accordingly a supervised training learning machine 68. It is also possible to configure the learning machine 68 to calculate the aforementioned representative 66 of each groove type, which is then used for classification 40 as described above by comparing each anomaly 20 with the representative 66. The learning machine 68 thus generates a representative 66 for each groove type and is used for modeling. In Fig.Figure 13 shows a representative 66 calculated in this way, which is a function of the position P on the one hand and additionally a function of the groove body spacing 34.

[0140] Validation

[0141] The validation 44 is based on the assignment 40 of detected to predefined processing positions 18, 24, which was carried out as further described, on the one hand (shown in Fig. 2 for the cutting types) or on the other hand on the classification 42 of the anomalies 20, i.e. their assignment to certain cutting and / or grooving types (not explicitly shown in Fig. 2).

[0142] In the validation 44 shown in Fig. 2, the distances between each detected processing position 18 and the assigned predefined processing position 24 are calculated, and then an overall dimension is determined from all distances, e.g., the maximum or the mean value or something similar, and this overall dimension is compared with a threshold value. If the overall dimension does not reach the threshold value, the assignment 40 is considered valid (i.e., validated); otherwise, it is considered invalid, and analogously, the order with regard to processing positions 18 and 24 is considered correctly executed or defective.

[0143] Alternatively or additionally, the similarity of each detected cutting and / or creasing type is calculated analogously to the cutting and / or creasing type required (i.e., specified) in the order data 22 at this processing position 18, and then the similarity for each specified processing position 24 is compared separately with a threshold value. If the similarity reaches the threshold value, the classification 42 is considered valid; otherwise, it is considered invalid, and analogously, the order is considered correctly executed or defective with regard to the cutting and / or creasing type.

[0144] In case of discrepancies between the detected processing positions 24 and / or the detected cutting and creasing types and the order data 22, a notification is issued to an operator via an output element, e.g., as part of a human-machine interface 78, and forwarded to a control station of system 2. System 2 shown here also features an input element (not explicitly shown) as part of the human-machine interface 78, through which an operator can add further information, e.g., key figures or identifiers, to the issued notifications, the measured height profiles 14, 16, the results of the position detection 36, the assignment 40, the classification 42, and / or the validation 44, particularly "just-in-time," in order to supplement or make available relevant information on production, process, and product quality, and specifically to generate the templates mentioned above.This information is stored in a database 80 and used for further evaluations and learning processes.

[0145] Annex 2 also includes a control unit 82, which is designed to execute the procedure as described.

[0146] With reference to Fig. 14, an embodiment for controlling the cutting and creasing unit 6 by means of a control unit 84 is described below. Fig. 14 shows the arrangement from Fig. 3 in a simplified representation. Horizontal dashed lines have been added to clarify the actual machining positions 18, namely cuts 8 produced by the cutting elements 27 and grooves 10 produced by the creasing elements 28. Dashed lines have also been added to clarify the predefined machining positions 24, which are specified, for example, according to order data.Depending on the actual machining positions 18 and the predefined machining positions 24, the control unit 84 generates a control signal C, which controls the cutting and grooving unit 6 in such a way as to optimize, or in this case maximize, the similarity between the actual machining position 18 and the predefined machining position 24. Based on the actual and / or the predefined machining positions 18 and 24, the control unit 84 first determines a number of actual values, which are then compared with a corresponding number of target values. Based on this comparison of the actual values ​​with the target values, the control unit 84 generates an error signal, which ultimately measures how well the actual and the predefined machining positions 18 and 24 correspond.

[0147] The following describes, with reference to Fig. 14, a first and a second variant for generating a control signal C depending on the actual machining positions 18 and the specified machining positions 24. The two variants differ primarily in the selection of the actual and target values.

[0148] In the first variant, a number of actual values ​​are determined from the actual machining positions 18, corresponding to the distances A1 between these positions. Due to the principle, the number of these distances A1 (here seven) is exactly one less than the number of actual machining positions 18 (here eight). These distances A1 are also referred to as relative distances A1, as they each specify the distance A1 between two machining positions 18 that are consecutive in the transverse direction Q. Each distance A1 is thus defined relative to at least one cutting and / or creasing body 27, 28, e.g., as in Fig. 14 between two adjacent cutting and / or creasing bodies 27, 28, each marking one end of this distance A1. The distances A1 can, in principle, also be defined relative to one or more other reference points, with equivalent results.For these distances A1, corresponding target values ​​are now also specified; these are calculated analogously to the determination of the actual values ​​from the specified processing positions 24, namely, for example, as the difference between two specified processing positions 24.

[0149] In the second variant, a number of actual values ​​are determined from the actual processing positions 18 and the specified processing positions 24. Each of these values ​​corresponds to a distance A2 between one of the actual processing positions 17 and its assigned specified processing position 24. Therefore, exactly as many distances A2 are determined as there are pairs of actual and assigned specified processing positions 18 and 24 (here, eight pairs). The distances A2 then directly indicate how well the actual processing positions 18 and 24 correspond to the specified positions. These distances A2 are therefore also referred to as absolute distances.

[0150] Regardless of the selected variant, each actual value is compared with a predetermined target value, and the control signal C is then generated based on this comparison. The control signal C is designed such that it moves one or more of the cutting and / or creasing bodies 27, 28 of the cutting / creasing unit 6 transversely to the conveying direction F. In Fig. 14, this is illustrated by a vertical double arrow on one of the creasing bodies 28. In the embodiment shown here, all cutting and / or creasing bodies 27, 28 are mounted on a common axis and are also moved along this axis. If a cutting or creasing body 27, 28 is incorrectly positioned in the transverse direction Q, a corresponding distance A2 results between the actual and the predetermined machining position 18, 24.The absolute and relative distances A1, A2 as described are thus suitable, directly or indirectly, as error signals for controlling the position of the respective cutting or creasing body 27, 28 in the transverse direction Q. Accordingly, the cutting and / or creasing bodies 27, 28 are moved in the transverse direction Q depending on the relative and / or absolute distances A1, A2.

[0151] Reference symbol list

[0152] 2 Annex

[0153] 4 lane

[0154] 6 Cutting and grooving unit

[0155] 8. Processing, cutting

[0156] 10. Machining, grooving

[0157] 12 sensor units

[0158] 14 upper elevation profile

[0159] 16 lower elevation profile

[0160] 18 actual / detected processing positions

[0161] 20 anomaly

[0162] 22 Order data

[0163] 24 predefined processing positions

[0164] 26 (first) threshold

[0165] 27 cutting bodies

[0166] 28 grooved bodies

[0167] 30 profiled running surface

[0168] 32 profiled running surface

[0169] 34 Groove body spacing

[0170] 36 Preprocessing

[0171] 38 Position detection

[0172] 40 Assignment (of processing positions)

[0173] 42 Classification

[0174] 44 Validation

[0175] 46 Sensor

[0176] 48 Pressure roller

[0177] 50th edition

[0178] 52 Block (of preprocessing)

[0179] 54 more block (of preprocessing)

[0180] 56 Frequency range (for cuts)

[0181] 58 Frequency range (for grooves)

[0182] 60 Block (Anomaly identification by reaching the threshold) 62 Block (Detection based on a gap)

[0183] 64 blocks

[0184] 66 Representative

[0185] 68 Learning machine

[0186] 70 neurons

[0187] 72 input layers

[0188] 74 output layer

[0189] 76 hidden layers

[0190] 78 Human-Machine Interface

[0191] 80 database

[0192] 82 Control unit

[0193] 84 Control unit

[0194] A1 Distance (between (adjacent) actual processing positions)

[0195] A2 Distance (between an actual processing position and the predefined processing position assigned to it)

[0196] B bandpass filter

[0197] C control signal

[0198] F Conveyor direction

[0199] P Position

[0200] Q transverse direction

[0201] 51 First step (measurement)

[0202] 52 second step (preprocessing)

[0203] 53 third step

[0204] 54 fourth step

Claims

Claims 1. Procedure, - wherein a web (4) is provided into which a number of operations (8, 10) have been performed by means of a cutting and creasing unit (6), each based on a predetermined machining position (24) and a predetermined cutting type or creasing type, - wherein a height profile (14, 16) of the track (4) is measured with a sensor unit (12) on at least one side of the track (4) and perpendicular to a conveying direction (F) of the track (4), - wherein a number of anomalies (20) and thus also actual processing positions (18) are identified in the elevation profile (14, 16) by identifying such sections of the elevation profile (14, 16) as an anomaly (20) in which the elevation profile (14, 16) reaches a predetermined threshold (26), - wherein each anomaly (20) is classified by means of a classification (42) by assigning the anomaly (20) to the cut type or groove type to which the anomaly (20) is most similar.

2. Method according to claim 1, wherein for classification (42) a learning machine (68) is used which receives an anomaly (20) as an input parameter and then outputs the corresponding cut or groove type as an output parameter.

3. Method according to claim 1, wherein a learning machine (68) calculates a respective representative (66) of a groove type or cut type, which is then used for classification by comparing a respective anomaly (20) with the representative (66).

4. Method according to claim 2 or 3, wherein the learning machine (68) is a neural network with a plurality of neurons (70), wherein the ReLu function is used as the activation function for the individual neurons (70), wherein the neural network has an input layer (72) and an output layer (74), wherein the neural network has several hidden layers (76), with a decreasing number of neurons (70) towards the output layer (74).

5. Method according to claim 1, wherein for the classification (42) a respective anomaly (20) is compared with a representative (66) of a respective cut or groove type and the similarity of the anomaly (20) and the cut or groove type is calculated using a distance standard, wherein the representative (66) of a respective cut or groove type is determined on the basis of previous measured values ​​by recording a plurality of height profiles (14, 16) for a respective cut or groove type and then determining the representative (66) for this cut or groove type from these.

6. Method according to claim 1, wherein for the classification (42) a respective anomaly (20) is compared with a representative (66) of a respective cut or groove type and the similarity of the anomaly (20) and the cut or groove type is calculated using a distance norm, wherein the representatives (66) are determined using a physical model which simulates the generation of grooves (10) and / or cuts (8) based on a geometry of a respective groove body (28) or cutting body of the cutting groove unit (6).

7. Method according to any one of claims 1 to 6, wherein the similarity of a recognized cutting and / or creasing type is calculated with the specified cutting and / or creasing type and then the similarity for each specified processing position (24) is compared separately with a further threshold, whereby the classification (42) is considered valid if the similarity reaches the threshold, and otherwise not valid.

8. Method according to any one of claims 1 to 7, wherein each actual processing position (18) is assigned the predetermined processing position (24) which is most similar to the actual processing position (18).

9. Method according to claim 8, wherein a control unit (84) generates a control signal (C) depending on the actual machining positions (18) and the predetermined machining positions (24) and thereby controls the cutting and grooving unit (6) in such a way that the similarities of the actual machining position (18) to the predetermined machining position (24) are optimized, in particular maximized.

10. Method according to claim 9, wherein a number of actual values ​​are determined from the actual machining positions (18) which correspond to distances (A1 ) between the actual machining positions (18), wherein each actual value is compared with a predetermined target value and the control signal (C) is generated on the basis of which.

11. Method according to claim 9, wherein a number of actual values ​​are determined from the actual processing positions (18) and the predetermined processing positions (24), each corresponding to a distance (A2) between one of the actual processing positions (18) and the predetermined processing position (24) assigned to it, wherein each actual value is compared with a predetermined (order data) target value and the control signal is generated on the basis of this.

12. Method according to claims 9 to 11, wherein, to optimize the similarity, one or more cutting and / or grooving bodies (28) of the cutting / grooving unit (6) are moved transversely to the conveying direction (F).

13. Method according to any one of claims 8 to 12, wherein a limit value for similarity is specified which must at least be reached in order to perform an assignment (40) in order to avoid a forced assignment.

14. Method according to one of claims 8 to 13, wherein distances between a respective actual machining position (18) and the assigned predetermined machining position (24) are calculated, wherein a total dimension is determined from all distances and this total dimension is compared with a further threshold value, wherein the assignment (40) is considered valid if the total dimension does not reach this further threshold value, and otherwise as not valid.

15. Method according to any one of claims 1 to 14, wherein the height profile (14, 16) is filtered with a bandpass filter (B) before the anomalies (20) are identified therein, wherein the bandpass filter (B) exposes a frequency range (56, 58) which is selected such that specific components of the height profile (14, 16) remain, which are produced either by cutting (8) or by grooving (10).

16. Method according to any one of claims 1 to 15, wherein for at least one actual machining position (18) a groove depth is measured as actual depth, wherein the cutting and creasing machine (6) is controlled depending on a difference between the actual depth and a target depth in order to minimize the difference.

17. Method according to any one of claims 1 to 16, wherein the threshold (26) is a standard deviation of the height profile (14, 16).

18. Method according to any one of claims 1 to 17, wherein anomalies (20) are identified separately in an upper height profile (14) on a top side of the web (4) and in a lower height profile (16) on a bottom side of the web (4), wherein an anomaly (20) at a processing position (18) is recognized as a cut (8) if an anomaly (20) has been identified at this processing position (18) in both the upper and the lower height profile (14, 16), and / or wherein an anomaly (20) at a processing position (18) is recognized as a cut (8) if there is a gap in the upper or the lower height profile (14, 16) at this processing position (18).

19. Method according to any one of claims 1 to 18, wherein a difference height profile is calculated from an upper height profile (14) on a top side of the web (4) and a lower height profile (16) on a bottom side of the web (4), wherein, when the difference height profile reaches the threshold value (26) at a processing position (18), an anomaly (20) is detected at that processing position (18).

20. Method according to any one of claims 1 to 18, wherein a wear condition of a cutting or grooving body (28) of the cutting / grooving unit (6) is detected on the basis of at least one anomaly (20).

21. Method according to any one of claims 1 to 20, where the actual caliber of the track (4) is determined based on the elevation profile (14, 16).

22. Method according to any one of claims 1 to 20, wherein the cutting and creasing unit (6) and the sensor unit (12) are part of a corrugated board machine and wherein the web (4) is a corrugated board web which is produced and processed with the corrugated board machine, wherein the height profile (14, 16) is measured inline and downstream of the cutting and creasing unit (6) in the corrugated board machine.

23. System (2) comprising a control unit (82) configured to perform a method according to any one of claims 1 to 21.

24. Computer program product which contains instructions which, when executed by a computer, cause it to, - to identify a number of anomalies (20) and thus also actual processing positions (18) in a measured elevation profile (14, 16) of a track (4) by identifying such sections of the elevation profile (14, 16) as an anomaly (20) in which the elevation profile (14, 16) reaches a predetermined threshold (26), - to classify each anomaly (20) by means of a classification (42) by assigning the anomaly (20) to a cut and Z or groove type to which the anomaly (20) is most similar.

Images