Packaging die-cutting precision correction method based on edge contour tracking

Abstract

The application discloses a packaging die-cutting precision correction method based on edge contour tracking and relates to the technical field of die-cutting precision correction, which is used for solving the problem of precision correction failure caused by motion blur interference and die-cutting phase aliasing in the process of roll material feeding. An exposure period displacement vector is generated by collecting edge images, instantaneous speed of material feeding, camera exposure duration and die-cutting mechanism phase coding, a contour segment blur vector is constructed by point-by-point projection along the tangent direction sequence of the design standard contour data, and a deblurring observation contour is generated. The deblurring observation contour is compared with the design standard contour data to generate an original deviation residual. The in-phase and out-of-phase deviation components are separated according to the die-cutting instantaneous phase. The drift reference is estimated by an exponential weighted moving average, and whether to update is determined by Mahalanobis distance. The single-cycle correction instruction is generated by logical regression mapping for the out-of-phase deviation component and is output to the die-cutting mechanism, so that the die-cutting precision closed-loop correction is realized, and the correction stability and real-time response capability are improved.

Description

Technical Field

[0001] This invention relates to the field of die-cutting accuracy correction technology, and more specifically, to a packaging die-cutting accuracy correction method based on edge contour tracking. Background Technology

[0002] In the industrial production of packaging materials, roll die-cutting is a key process that precisely cuts continuously fed substrate into the required contour shape according to the design. The die-cutting mechanism periodically performs punching action through the cutter roller or die during the continuous feeding of the roll. To ensure product yield, the production system usually deploys industrial vision devices downstream of the die-cutting station to collect images of the edge of the die-cut roll and compare them with the pre-stored design standard contour data to detect die-cutting errors and feed back correction instructions to the die-cutting mechanism. The correction system tracks the drift based on historical deviation records and performs real-time evaluation and compensation for deviations in each punching cycle.

[0003] The existing technology has the following shortcomings: Currently, existing technologies for contour extraction of edge images after die-cutting of roll materials lack directional quantitative modeling of material feeding motion fuzz. This makes it impossible to effectively separate the directional stretching component introduced by motion fuzz from the actual die-cutting deviation, resulting in reduced accuracy of die-cutting precision correction. Furthermore, it is impossible to accurately distinguish between steady-state and transient deviation components, leading to the aliasing of steady-state drift and transient deviation in the correction command. Consequently, drift reference maintenance and single-cycle real-time homing cannot be executed independently, resulting in correction overshoot and drift estimation distortion. Therefore, a packaging die-cutting precision correction method based on edge contour tracking is proposed.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a packaging die-cutting accuracy correction method based on edge contour tracking. This method addresses the problems mentioned in the background art by employing contour segment fuzzy vector modeling based on tangent direction projection and iterative deconvolution defuzzification, phase bucketing deviation attribution based on fuzzy C-means clustering, and drift benchmark adaptive update and logistic regression mapping correction instruction generation based on Mahalanobis distance determination.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a packaging die-cutting accuracy correction method based on edge contour tracking, the specific operation process of which is as follows: Step S1: Acquire edge images of the rolled material after die-cutting, extract the observation contour curve from the edge images, and simultaneously read the instantaneous feeding speed of the feeding servo encoder, the camera exposure time, and the phase encoding of the die-cutting mechanism. Combine the instantaneous feeding speed and the camera exposure time to generate the displacement vector of the exposure period. Step S2: Retrieve the design standard contour data, sample along the design standard contour data to generate a tangent direction sequence, project the displacement vector of the exposure period along each sampling point of the tangent direction sequence to generate a contour segment blur vector, calculate the directional stretching amount of each segment based on the contour segment blur vector, and determine the deblurred observation contour by combining the observed contour curve. Step S3: Compare the defuzzified observation contour with the design standard contour data point by point to generate the original deviation residual. Identify the instantaneous phase of die cutting according to the phase encoding of the die cutting mechanism. Based on the instantaneous phase of die cutting, perform bucketing processing on the original deviation residual to obtain the in-phase deviation component and the out-of-phase deviation component. Step S4: Retrieve historical die-cutting residual timing records to generate a drift reference, compare the in-phase deviation component with the drift reference, select whether to update the drift reference based on the comparison result, generate a single-cycle correction command based on the out-of-phase deviation component and output it to the die-cutting mechanism.

[0007] In a preferred embodiment, in step S1, the edge image of the die-cut roll material is acquired, and the edge image of the die-cut roll material is extracted using the Canny edge detection algorithm to obtain the observation contour curve composed of interconnected edge pixels. The material feeding servo encoder synchronously reads the instantaneous material feeding speed, the camera exposure time synchronously reads the camera controller, and the phase code of the die-cutting mechanism synchronously reads the die-cutting servo encoder. An exposure time displacement vector is generated based on a combination of the instantaneous feed speed and the camera exposure duration. ,in, The displacement vector during the exposure period. This is a scalar value representing the instantaneous speed of the material feed. For camera exposure time, This is the unit vector for the material movement direction.

[0008] In a preferred embodiment, in step S2, the design standard contour data is a set of coordinate points for the theoretical cut shape after the roll material is die-cut, and the coordinate points are located in the physical coordinate system where the edge image of the roll material is located after die-cutting; Several sampling points are obtained by sampling along the design standard contour data with equal arc length. For each sampling point, several neighborhood points are taken on the design standard contour data with the corresponding sampling point as the center and the radius as the preset neighborhood scale. A covariance matrix is ​​constructed on the coordinate vector of the neighborhood points. Eigenvalue decomposition is performed on the covariance matrix, and the eigenvector corresponding to the largest eigenvalue is taken as the tangent direction unit vector of the corresponding sampling point. The unit vectors of the tangent directions of all sampling points are arranged in the sampling order to form a sequence of tangent directions. The displacement vector during the exposure period is multiplied by the unit vector of the tangent direction at the corresponding sampling point in the tangent direction sequence. The result of the dot product is then multiplied by the unit vector of the tangent direction to obtain the projection component, which is used as the blur vector of the contour segment.

[0009] In a preferred embodiment, in step S2, the blur vector of the contour segment is first processed into a dimensionless form according to the magnitude of the displacement vector during the exposure period, and then the dimensionless result is nonlinearly mapped using the Sigmoid function. Finally, it is multiplied by a preset upper limit value of stretching and a direction sign is added to obtain the directional stretching amount. Using the coordinates of the corresponding sampling points on the observed contour curve as the initial values ​​for iteration, and combining the directional stretching amount with the unit vector along the tangent direction, an iterative deconvolution with directional prior is performed to obtain the iteration result; The Euclidean distance between two consecutive iterations is compared with a preset convergence threshold: When the Euclidean distance between two adjacent iterations is less than the preset convergence threshold, the iteration is determined to be converged, and the current contour coordinates of the sampling points are arranged in the sampling order as the deblurred observation contour output. Conversely, if the iteration fails to converge, the next iteration will be performed to correct the failure.

[0010] In a preferred embodiment, in step S3, for each sampling point on the deblurred observation contour, a local neighborhood with the corresponding sampling point as the center and a radius of the preset neighborhood is taken on the design standard contour data, and the minimum Euclidean distance from the sampling point of the deblurred observation contour to all points in the local neighborhood is calculated. The minimum Euclidean distance between the corresponding point of the sampled point in the design standard contour data and all points in the same local neighborhood of the deblurred observation contour is calculated in reverse. The larger of the two minimum values ​​is taken as the original bias residual of the sampling point.

[0011] In a preferred embodiment, in step S3, the phase encoding of the die-cutting mechanism is converted into the instantaneous phase of die-cutting via linear mapping. The instantaneous phase of die-cutting is the phase angle of the die-cutting mechanism within its complete punching cycle at the moment when the edge image of the current roll material after die-cutting is acquired. The original deviation residuals are binned according to the instantaneous phase of the die-cutting using the fuzzy C-means clustering algorithm, and the membership degree of the original deviation residuals of each sampling point to each bin is obtained.

[0012] In a preferred embodiment, in step S3, the membership degree of the original deviation residual point is compared with a preset membership threshold, and the absolute value of the difference between the instantaneous phase of the die-cutting corresponding to the original deviation residual point and the phase center of the corresponding bucket is compared with a preset phase window. When the membership degree of the original deviation residual point to a certain bucket is higher than the preset membership threshold, and the absolute value of the difference between the instantaneous phase of the original deviation residual point and the phase center of the bucket is less than the preset phase window, the original deviation residual is assigned to the same phase deviation component. When the membership degree of the original deviation residual point to all buckets is lower than or equal to the preset membership threshold, or when the absolute value of the difference between the instantaneous phase of the original deviation residual point and the phase center of its respective bucket is greater than or equal to the preset phase window, the original deviation residual is classified into the out-of-phase deviation component.

[0013] In a preferred embodiment, in step S4, the historical die-cutting residual time series is recorded as an ordered sequence of the original deviation residual values ​​at historical moments. An exponentially weighted moving average algorithm is used to generate a drift baseline for each original deviation residual value; The difference between the in-phase deviation component and the drift reference is measured using the Mahalanobis distance to obtain the Mahalanobis distance between the in-phase deviation component and the drift reference. Compare the Mahalanobis distance with the preset convergence window: When the Mahalanobis distance is less than the preset convergence window, the residual value of the latest moment in the historical die-cutting residual time series record is replaced by the in-phase deviation component to update the drift reference. When the Mahalanobis distance is greater than the preset convergence window, the current drift reference remains unchanged.

[0014] In a preferred embodiment, in step S4, a single-cycle correction command is generated after the out-of-phase deviation components are dimensionless and mapped by logistic regression. The single-cycle correction command is output to the die-cutting mechanism, which then performs the corresponding blade roller phase adjustment or roll material lateral position adjustment at the next punching moment.

[0015] The technical effects and advantages of this invention are as follows: This invention generates an exposure period displacement vector by synchronously acquiring the instantaneous material feeding speed and camera exposure time. It then constructs a blurred vector of the contour segment by projecting it point by point along the tangent direction sequence of the design standard contour data and generates a deblurred observation contour through iterative deconvolution. This eliminates the interference of continuous material feeding motion blur on contour comparison. Based on the fuzzy C-means clustering of the instantaneous phase of die cutting, it separates the in-phase deviation component and the out-of-phase deviation component, eliminating phase aliasing in the correction command. It uses Mahalanobis distance to determine the shielding of abnormal mutations from the drift reference. Finally, it generates a single-cycle correction command for the out-of-phase deviation component through logistic regression mapping. This achieves high stability and real-time response capability of the die cutting accuracy correction system in continuous material feeding production. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of the packaging die-cutting accuracy correction method based on edge contour tracking according to the present invention.

[0017] Figure 2 This is a schematic diagram illustrating the steps of the packaging die-cutting accuracy correction method based on edge contour tracking according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention achieves closed-loop precision correction of packaging roll die-cutting accuracy by performing motion blur perception on the edge image of the roll material to restore the outline, combined with deviation component separation based on the die-cutting phase and adaptive drift benchmark maintenance, and outputting correction instructions in real time after mapping the transient die-cutting deviation through logistic regression.

[0020] Example 1: Please refer to Figures 1 to 2 The specific operation process of the packaging die-cutting accuracy correction method based on edge contour tracking is as follows: Step S1: Acquire edge images of the rolled material after die-cutting, extract the observation contour curve from the edge images, and simultaneously read the instantaneous feeding speed of the feeding servo encoder, the camera exposure time, and the phase encoding of the die-cutting mechanism. Combine the instantaneous feeding speed and the camera exposure time to generate the displacement vector of the exposure period. Step S2: Retrieve the design standard contour data, sample along the design standard contour data to generate a tangent direction sequence, project the displacement vector of the exposure period along each sampling point of the tangent direction sequence to generate a contour segment blur vector, calculate the directional stretching amount of each segment based on the contour segment blur vector, and determine the deblurred observation contour by combining the observed contour curve. Step S3: Compare the defuzzified observation contour with the design standard contour data point by point to generate the original deviation residual. Identify the instantaneous phase of die cutting according to the phase encoding of the die cutting mechanism. Based on the instantaneous phase of die cutting, perform bucketing processing on the original deviation residual to obtain the in-phase deviation component and the out-of-phase deviation component. Step S4: Retrieve historical die-cutting residual timing records to generate a drift reference, compare the in-phase deviation component with the drift reference, select whether to update the drift reference based on the comparison result, generate a single-cycle correction command based on the out-of-phase deviation component and output it to the die-cutting mechanism.

[0021] The specific steps are as follows: In step S1, synchronous data acquisition is performed on the continuous feeding of the roll material through the die-cutting station. An industrial camera set downstream of the die-cutting station acquires the edge image of the roll material after die-cutting. The gray value of each pixel in the acquired edge image of the roll material after die-cutting reflects the cumulative light intensity response of that position during the exposure period. The observation contour curve is a continuous closed or open edge line extracted from the edge image of the die-cut roll material, used to reflect the shape of the cut of the roll material after die-cutting under conditions containing motion blur. Specifically, the edge image after die-cutting of the roll material is processed using the Canny edge detection algorithm to extract the edges, resulting in an observation contour curve composed of interconnected edge pixels.

[0022] The Canny edge detection algorithm is an image processing algorithm based on three stages: Gaussian filtering for denoising, gradient magnitude for non-maximum suppression, and double thresholding for edge connectivity. It can maintain edge connectivity by relying on the double thresholding mechanism when motion blur causes edge intensity attenuation in the edge image after die-cutting of roll material.

[0023] While the industrial camera is performing exposure, the instantaneous feeding speed is read synchronously from the feeding servo encoder, the camera exposure time is read synchronously from the camera controller, and the phase code of the die-cutting mechanism is read synchronously from the die-cutting servo encoder.

[0024] It should be explained that the material feeding servo encoder is a rotary encoding device installed on the material feeding traction shaft, used to provide real-time feedback on the instantaneous feeding speed based on the rotation pulses of the material feeding traction shaft; the camera controller is an electrical device that controls the industrial camera to perform image acquisition, used to set and provide feedback on the camera exposure time; the die-cutting servo encoder is a rotary encoding device installed on the die-cutting mechanism's cutter roller or die drive shaft, used to provide feedback on the die-cutting mechanism's phase encoding based on the rotation angle of the drive shaft.

[0025] The exposure time displacement vector refers to the vector representation of the cumulative displacement of the roll material along the feeding direction during the start and end time of an exposure by an industrial camera in the physical coordinate system of the edge image after die-cutting of the roll material. It is used to reflect the motion blur direction and blur length that all edge pixels are subjected to during this exposure.

[0026] An exposure time displacement vector is generated based on a combination of the instantaneous feed speed and the camera exposure duration. ,in, The displacement vector during the exposure period. This is a scalar value representing the instantaneous speed of the material feed. For camera exposure time, The unit vector in the material feeding direction; The greater the instantaneous speed of the material feed, the greater the magnitude of the displacement vector during the exposure period, the greater the cumulative displacement of the roll material along the feeding direction during the exposure period, and the stronger the motion blur of the edge image after the roll material is die-cut; the longer the camera exposure time, the greater the magnitude of the displacement vector during the exposure period, the wider the single exposure integration window, and the more severe the motion blur.

[0027] It should be noted that the instantaneous speed of the material feed is measured in meters per second, and the exposure time of the camera is measured in seconds. After multiplying the two, the scalar amplitude of the displacement vector during the exposure period is measured in meters. This is consistent with the coordinate dimension of the edge image of the rolled material after die-cutting, which is mapped to the physical coordinate system after calibration by the industrial camera. Therefore, no additional dimension normalization is required.

[0028] In step S2, starting from the pre-stored design standard contour data in the upper system, the observed contour curve is subjected to directional stretching estimation and deblurring processing. The design standard contour data is an ordered set of coordinate points that are pre-stored by the upper system according to the geometric design of the die-cutting die before the start of the roll die-cutting operation. Each coordinate point is located in the physical coordinate system of the edge image after the roll die-cutting, and is used as a geometric reference for directional stretching estimation and a benchmark for comparing the deviation after deblurring.

[0029] The tangent direction sequence is an ordered set of local tangent directions at several sampling points obtained by sampling along the standard design contour data at equal arc lengths. It is used to reflect the geometric orientation of the design contour at different locations. Specifically, several sampling points are obtained by sampling along the design standard contour data with equal arc length. For each sampling point, several neighboring points are taken on the design standard contour data with the sampling point as the center and the radius as the preset neighborhood scale. A covariance matrix is ​​constructed from the coordinate vectors of the neighboring points. Eigenvalue decomposition is performed on the covariance matrix, and the eigenvector corresponding to the largest eigenvalue is taken as the tangent direction unit vector of the sampling point. The tangent direction unit vectors of all sampling points are arranged in the sampling order to form a tangent direction sequence.

[0030] The preset neighborhood scale is set as follows: during the calibration stage before the roll die-cutting operation begins, principal component analysis algorithm is performed on the design standard contour data according to different neighborhood scales, the mean square angle between the unit vector of the tangent direction and the analytical tangent direction is statistically calculated, and the neighborhood scale corresponding to the minimum mean square angle is taken as the preset neighborhood scale.

[0031] The contour blur vector is the projection component of the displacement vector during the exposure period onto the unit vector in the tangent direction of the sampling point at each sampling point. It is used to reflect the direction and length of the directional stretching caused by the cumulative displacement of the roll material during the exposure period at that sampling point.

[0032] Specifically, the displacement vector during the exposure period is multiplied by the unit vector of the tangent direction at the corresponding sampling point in the tangent direction sequence. The result of the dot product is then multiplied by the unit vector of the tangent direction to obtain the projection component, which serves as the contour segment blur vector for that sampling point. ,in, Let i be the blur vector of the contour segment at the i-th sampling point. The displacement vector during the exposure period. Let be the unit vector of the tangent direction at the i-th sampling point.

[0033] The closer the angle between the tangential unit vector and the feed direction is to 90 degrees, the closer the dot product of the exposure period displacement vector and the tangential unit vector is to zero, the smaller the magnitude of the blur vector of the contour segment, and the weaker the directional stretching caused by the exposure period displacement at that sampling point. The closer the tangential unit vector is to being parallel to the feed direction, the larger the absolute value of the dot product, the larger the magnitude of the blur vector of the contour segment, and the more significant the directional stretching at that sampling point.

[0034] It should be added that the dimension of the displacement vector during the exposure period is meters, the unit vector in the tangent direction is a dimensionless vector, the dimension of the dot product result is meters, and the dimension of the blurred vector of the contour segment after multiplying with the unit vector in the tangent direction is still meters, which is consistent with the coordinate dimension of the design standard contour data.

[0035] The directional stretching amount is a scalar band direction value obtained by dimensionless and nonlinear mapping of the fuzzy vector of the contour segment, used to characterize the magnitude and direction of the actual stretching displacement to be deducted at the sampling point.

[0036] The blurred vector of the contour segment is first dimensionless according to the magnitude of the displacement vector during the exposure period. Then, the dimensionless result is nonlinearly mapped using the Sigmoid function. Finally, it is multiplied by a preset upper limit of stretching and a direction sign is added to obtain the directional stretching amount. Dimensionless is defined as: ,in, Let i be the directional stretching amount at the i-th sampling point. Let be the Euclidean modulus of the blur vector of the contour segment. To be according to The magnitude of the blurred vector of the contour segment after dimensionless processing. Let x be the magnitude of the displacement vector during the exposure period. To preset the nonlinear steepness coefficient, To preset the upper limit of the stretching amount, Let be the unit vector of the tangent direction at the i-th sampling point. It is a symbolic function.

[0037] The Sigmoid function is used in this scenario to maintain mapping stability when the magnitude of the blur vectors of different segments of the curve contour is large. It avoids excessive stretching estimation that exceeds the real physical dimensions after linear mapping of the maximum blur quantity, thereby suppressing the overcompensation in the subsequent deblurring process. Preset nonlinear steepness coefficient The setup method is as follows: During the calibration phase, a set of calibration samples containing known fuzzy values ​​and actual stretching values ​​are collected. The calibration samples are then fitted using a least-squares method that minimizes the squared error between the predicted and actual stretching values ​​to determine the values. Values.

[0038] Preset stretching limit The setting method is as follows: statistically analyze the distribution of the actual tensile amount in the same set of calibration samples, and take the upper quartile of the actual tensile amount as... .

[0039] The defuzzy observation profile is an ordered set of coordinate points obtained by subtracting the directional stretching amount at each sampling point from the observation profile curve. It is used to approximately restore the cut shape of the roll material after die-cutting without being affected by motion blur.

[0040] Specifically, the coordinates of the corresponding sampling points on the observed contour curve are used as the initial values ​​for iteration. Combined with the directional stretching amount, an iterative deconvolution with directional prior is performed along the tangent direction using a unit vector. The iterative formula is as follows: The iteration termination condition is: ,in, The contour coordinates of the i-th sampling point after the k-th iteration. The initial coordinates of the observed contour curve at the corresponding sampling points are used. Let i be the directional stretching amount at the i-th sampling point. Let be the unit vector of the tangent direction at the i-th sampling point. To preset the iteration step size, To preset the convergence threshold, The distance is Euclidean.

[0041] The reason for choosing the iterative deconvolution method with directional prior is that it uses the contour state of the previous round as the geometric constraint of the next round in each iteration. Compared with the method of directly subtracting the observed contour curve by directional stretching at one time, it can gradually approximate the real cut shape. It is suitable for the characteristics of this scenario where the fuzzy vector of the contour segment changes with the sampling point position and each segment needs to converge with different time lengths.

[0042] The Euclidean distance between two consecutive iterations is compared with the preset convergence threshold. Comparison: When the Euclidean distance between two adjacent iterations is less than the preset convergence threshold, the iteration is determined to be converged, and the current contour coordinates are arranged in the sampling order as the deblurred observation contour output to step S3. If the Euclidean distance between two consecutive iterations is greater than or equal to a preset convergence threshold, the iteration is deemed not to have converged, and the iteration formula is applied. Continue with the next round of iteration and correction.

[0043] Through the above process of directional estimation of the design standard contour data and iterative deconvolution of the observed contour curve along the directional stretching amount, the directional stretching component carried in the edge image of the die-cut roll containing motion blur is separated from the real die-cutting deviation component, providing a morphological benchmark that can be directly compared with the design standard contour data for deviation attribution in subsequent steps. It should be noted that the preset iteration step size With preset convergence threshold The setup method is as follows: during the calibration phase, a set of calibration samples containing both true contours and motion-blurred contours are collected, and a series of candidate... and Combined iterative deconvolution is performed, and the number of iterations required for convergence to the true contour is statistically analyzed. The value that minimizes the number of iterations and ensures that the mean square error between the final contour and the true contour falls within a preset acceptable range is selected. and The combination is considered the preferred value.

[0044] In step S3, deviation measurement and phase attribution are performed on the deblurred observation profile and the design standard profile data; the original deviation residual is the symmetrical maximum value of the local nearest neighbor distance between the deblurred observation profile and the design standard profile data at each sampling point; it is used to reflect the local deviation of the deblurred observation profile relative to the design standard profile data at that sampling point.

[0045] For each sampling point on the deblurred observation contour, take a local neighborhood on the design standard contour data with the sampling point as the center and the radius as the preset neighborhood radius, and calculate the minimum Euclidean distance from the sampling point of the deblurred observation contour to all points in the local neighborhood. Simultaneously, the minimum Euclidean distance from the corresponding sampling point in the design standard contour data at that location to all points within the same local neighborhood of the deblurred observed contour is calculated in reverse. The larger of the two minimum values ​​is taken as the original deviation residual for that sampling point, and its calculation formula is as follows: ,in, The original bias residual for the i-th sampling point is... To determine the coordinates of the unblurred observation contour at the i-th sampling point, To design the coordinates of the j-th point in the local neighborhood centered on the i-th sampling point, To design the coordinates of the i-th sampling point on the standard contour data, The coordinates of the j-th point in the neighborhood are used to deblur the observed contour.

[0046] The preset neighborhood radius is set as follows: during the calibration phase, a set of calibration samples containing known deviation points and no deviation points are collected. The separability of the original deviation residuals with respect to known deviation points and no deviation points is statistically analyzed under different neighborhood radii. The neighborhood radius corresponding to the highest separability is taken as the preset neighborhood radius.

[0047] The instantaneous phase of die-cutting is the phase angle of the die-cutting mechanism within its complete punching cycle at the moment when the edge image of the current roll material is acquired after die-cutting. It is used to classify the original deviation residual according to the die-cutting phase at the time of occurrence.

[0048] The phase encoding of the die-cutting mechanism is converted into the instantaneous phase of die-cutting via linear mapping: ,in, For the instantaneous phase of die cutting, The readings are the phase codes for the die-cutting mechanism. This is the full-scale value of the die-cutting servo encoder.

[0049] It should be noted that the dimensions of the phase encoder reading of the die-cutting mechanism and the full-scale value of the die-cutting servo encoder are both encoder counts. The ratio between the two is dimensionless. After multiplying by twice pi, the dimension of the instantaneous phase of the die-cutting mechanism is in radians, which is consistent with the dimension of the phase center in the objective function of fuzzy C-means clustering.

[0050] The original deviation residuals are binned according to the instantaneous phase of the die-cutting using the fuzzy C-means clustering algorithm, and the membership degree of the original deviation residuals of each sampling point to each bin is obtained. Fuzzy C-means clustering is a clustering algorithm that allows a single data point to belong to multiple clusters with consecutive membership degrees. In this example, it is used to bin the original bias residuals in phase space.

[0051] The objective function for fuzzy C-means clustering is constructed as follows: The phase similarity function is: ,in, The objective function for fuzzy C-means clustering is... Let be the membership degree of the i-th original deviation residual point to the j-th bucket. To preset the fuzzy index, The original bias residual for the i-th sampling point is... Let j be the phase center of the j-th bucket. Let i be the instantaneous phase of the die-cutting corresponding to the i-th sampling point. The phase similarity function uses the cosine square operation to transform the phase difference into a similarity between zero and one; this is achieved through iterative minimization. The membership degree of each original deviation residual point to each bin is obtained by solving the problem. Phase center with each sub-bucket .

[0052] The preset fuzzy index m is set as follows: during the calibration stage, a set of original deviation residual samples containing known phase assignments are collected, and fuzzy C-means clustering is performed on a series of candidate m values. The value that makes the clustering result have the highest consistency rate with the known phase assignment is taken as m.

[0053] The membership degree of the original deviation residual point is compared with the preset membership threshold, and the absolute value of the difference between the instantaneous phase of the die-cutting corresponding to the original deviation residual point and the phase center of the corresponding bucket is compared with the preset phase window. When the membership degree of the original deviation residual point to a certain bucket is higher than the preset membership threshold, and the absolute value of the difference between the instantaneous phase of the die-cutting corresponding to the original deviation residual point and the phase center of the bucket is less than the preset phase window, the original deviation residual is assigned to the same phase deviation component. Conversely, when the membership degree of the original deviation residual point to all buckets is lower than or equal to the preset membership threshold, or when the absolute value of the difference between the instantaneous phase of the original deviation residual point and the phase center of the bucket to which it belongs is greater than or equal to the preset phase window, the original deviation residual is classified into the out-of-phase deviation component.

[0054] In-phase deviation component refers to the original deviation residual subset that is statistically strongly correlated with the current die-cutting instant phase, and is used to reflect the steady-state deviation that the die-cutting mechanism repeatedly exhibits at a fixed phase. The out-of-phase deviation component refers to the subset of original deviation residuals that is statistically unrelated or weakly correlated with the current die-cutting instantaneous phase. It is used to reflect the transient deviation of the die-cutting mechanism from the steady state in the current single cycle.

[0055] It should be noted that the preset membership threshold is set as follows: during the calibration stage, for a set of original deviation residual samples with known phase affiliation, the membership threshold corresponding to the maximum separation degree between the same-phase deviation component and the opposite-phase deviation component in the membership degree dimension is taken as the preset membership threshold; the preset phase window is set as follows: the total phase length of the die-cutting mechanism in a single cycle is taken as a multiple of the minimum resolvable phase determined by the resolution of the die-cutting servo encoder.

[0056] In step S4, the calibration database is a structured data set established by the die-cutting accuracy calibration system during the roll die-cutting operation. It is used to store the original deviation residuals and their phase assignment information in chronological order. In this example, it is used to provide historical die-cutting residual time-series records and corresponding covariance matrices.

[0057] Historical die-cutting residual time series records are ordered sequences of raw deviation residual values ​​representing several historical moments, provided by the calibration database in chronological order, and are used to make benchmark estimates of steady-state drift during long-term production. The drift reference is a reference value estimated based on historical die-cutting residual time series records, which characterizes the center position of the steady-state deviation of the die-cutting mechanism in the current production period. It is used to measure the difference with the in-phase deviation component and to determine whether to update the reference value.

[0058] The drift baseline was estimated using an exponentially weighted moving average algorithm on the historical die-cut residual time series records. ,in, Let k be the drift reference after the kth update. The original bias residual value at time k is the historical die-cutting residual time series record. To pre-set the forgetting factor, The drift reference is the one updated after the (k-1)th time; the initial drift reference. Take the average residual value within the initial time period of the historical die-cut residual time series record.

[0059] Preset forgetting factor The value range is limited to between zero and one. The closer it is to one, the more sensitive the drift baseline is to recent residuals and the shorter its memory of historical residuals; The closer to zero, the stronger the drift baseline's dependence on historical residuals and the smoother its response to recent residuals.

[0060] Preset forgetting factor The setup method is as follows: During the calibration phase, a set of historical die-cutting residual time-series records containing known drift evolution curves are collected as calibration samples, and a series of candidate... Estimate the drift baseline and calculate the root mean square (RMS) tracking error between it and the known drift evolution curve. Take the value that minimizes this RMS as the baseline. .

[0061] The difference between the in-phase deviation component and the drift reference is measured using Mahalanobis distance: ,in, The Mahalanobis distance between the in-phase deviation component and the drift reference. For in-phase deviation components, As the current drift reference, The covariance matrix of the historical die-cut residual time series records. It is the inverse of the covariance matrix. This is a vector transpose operation.

[0062] The reason for choosing Mahalanobis distance is that it standardizes the difference between the in-phase deviation component and the drift reference by using the inverse of the covariance matrix. It can simultaneously consider the variance differences and correlations between different directions of the historical die-cutting residual time series records in different directions (including the material feeding direction and the transverse direction of the material). It is suitable for the situation in which the in-phase deviation component fluctuates at different scales in different directions. Compared with directly using Euclidean distance, it can avoid misjudgment caused by equal weighting of different directions.

[0063] Compare the Mahalanobis distance with the preset convergence window: When the Mahalanobis distance is less than the preset convergence window, the difference between the in-phase deviation component and the drift reference is determined to be within the normal drift range. The in-phase deviation component replaces the latest residual value in the historical die-cutting residual time series record, and the update formula is applied. Calculate the drift reference for the next time step and write it back to the correction database; When the Mahalanobis distance is greater than the preset convergence window, it is determined that the in-phase deviation component has an abnormal change relative to the drift reference. The current drift reference is kept unchanged, and an abnormality mark is triggered for the current period for manual review. When the Mahalanobis distance equals the preset convergence window, it is processed according to the normal drift range, and the drift reference is updated.

[0064] The preset convergence window is set as follows: during the calibration phase, a set of historical die-cutting residual time series records containing both normal drift and abnormal mutation are collected as calibration samples. A series of candidate window values ​​are tested, and the upper limit of the window corresponding to the normal drift sample falling within the window and the abnormal mutation sample falling outside the window is taken as the preset convergence window.

[0065] The single-cycle correction command is a directional scalar value generated for the current die-cutting cycle and output to the die-cutting mechanism to adjust the phase of the cutter roller or the lateral position of the roll material. It is used to correct the transient deviation reflected by the out-of-phase deviation component at the next punching instant.

[0066] After dimensionless transformation and logistic regression mapping of the out-of-phase bias components, a single-cycle correction command is generated, and its calculation formula is as follows: Dimensionless is defined as: ,in, This is a single-cycle correction command. Let Euclidean mode be the Euclidean magnitude of the out-of-phase deviation component. The maximum magnitude of the displacement vector during the exposure period of the calibration phase. To preset the mapping steepness, For the preset bias, To preset the upper limit of the calibration command, This is the unit vector in the material feeding direction. It is a symbolic function; Preset mapping steepness With preset bias The setup method is as follows: during the calibration phase, a set of calibration samples containing the correspondence between known out-of-phase bias components and the optimal correction command are collected, and the maximum likelihood estimation method is used to fit the data. and Preset upper limit of calibration command The setting method is as follows: take the maximum allowable adjustment displacement of the die-cutting mechanism within a single cycle as... Its value is determined by the smaller of the maximum steady-state response displacement of the die-cutting mechanism servo system and the allowable displacement adjustment margin of the die under a single-cycle punching action.

[0067] Logistic regression mapping is used to smoothly map the dimensionless modulus of the out-of-phase deviation components to a bounded interval of zero to one. The modulus is then amplified to the maximum allowable adjustment displacement range of the die-cutting mechanism by a preset upper limit value of the correction command. This is suitable for scenarios in which the single correction amplitude needs to be limited to the steady-state response range of the servo system during the accuracy correction process, and to avoid overshoot oscillation of the die-cutting mechanism caused by large deviations after linear mapping.

[0068] It should be noted that the dimension of the out-of-phase deviation component is meters. After performing dimensionless processing based on the maximum modulus of the displacement vector during the exposure period of the calibration stage, the value range of the dimensionless result is limited to zero to one. Therefore, the input of the logistic regression function has a clear range meaning. After multiplying the output of the logistic regression function by the preset upper limit value of the correction command, the single-cycle correction command regains the dimension of displacement (meters), which can be directly used as the adjustment amount for the phase or lateral position of the die-cutting mechanism's roller.

[0069] The single-cycle correction command is output to the die-cutting mechanism, which then performs the corresponding blade roller phase adjustment or roll material lateral position adjustment at the next punching moment, thereby completing the die-cutting accuracy correction for the current cycle.

[0070] It should be added that the die-cutting mechanism is an execution device installed at the die-cutting station. It performs punching action on the continuously feeding roll material through the periodic rotation of the cutter roller or die drive shaft, cutting the roll material into the required contour shape according to the design.

[0071] By estimating the exponentially weighted moving average of the historical die-cutting residual time series records, determining the Mahalanobis distance between the in-phase deviation component and the drift reference, and performing logistic regression mapping on the out-of-phase deviation component, the benchmark maintenance of steady-state drift and the real-time correction of transient deviation are respectively implemented in the update of the drift reference and the output of the single-cycle correction command, thus realizing the packaging die-cutting accuracy correction based on edge contour tracking.

[0072] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0073] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0075] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0076] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A packaging die-cutting accuracy correction method based on edge contour tracking, characterized in that: Includes the following steps: Step S1: Acquire edge images of the rolled material after die-cutting, extract the observation contour curve from the edge images, and simultaneously read the instantaneous feeding speed of the feeding servo encoder, the camera exposure time, and the phase encoding of the die-cutting mechanism. Combine the instantaneous feeding speed and the camera exposure time to generate the displacement vector of the exposure period. Step S2: Retrieve the design standard contour data, sample along the design standard contour data to generate a tangent direction sequence, project the displacement vector of the exposure period along each sampling point of the tangent direction sequence to generate a contour segment blur vector, calculate the directional stretching amount of each segment based on the contour segment blur vector, and determine the deblurred observation contour by combining the observed contour curve. Step S3: Compare the defuzzified observation contour with the design standard contour data point by point to generate the original deviation residual. Identify the instantaneous phase of die cutting according to the phase encoding of the die cutting mechanism. Based on the instantaneous phase of die cutting, perform bucketing processing on the original deviation residual to obtain the in-phase deviation component and the out-of-phase deviation component. Step S4: Retrieve historical die-cutting residual timing records to generate a drift reference, compare the in-phase deviation component with the drift reference, select whether to update the drift reference based on the comparison result, generate a single-cycle correction command based on the out-of-phase deviation component and output it to the die-cutting mechanism.

2. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 1, characterized in that: In step S1, the edge image of the die-cut roll material is acquired, and the edge is extracted using the Canny edge detection algorithm to obtain the observation contour curve composed of interconnected edge pixels. The material feeding servo encoder synchronously reads the instantaneous material feeding speed, the camera exposure time synchronously reads the camera controller, and the phase code of the die-cutting mechanism synchronously reads the die-cutting servo encoder. An exposure time displacement vector is generated based on a combination of the instantaneous feed speed and the camera exposure duration. ,in, The displacement vector during the exposure period. This is a scalar value representing the instantaneous speed of the material feed. For camera exposure time, This is the unit vector for the material movement direction.

3. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 2, characterized in that: In step S2, the design standard contour data is the set of coordinate points of the theoretical cut shape after the roll material is die-cut, and the coordinate points are located in the physical coordinate system of the edge image after the roll material is die-cut; Several sampling points are obtained by sampling along the design standard contour data with equal arc length. For each sampling point, several neighborhood points are taken on the design standard contour data with the corresponding sampling point as the center and the radius as the preset neighborhood scale. A covariance matrix is ​​constructed on the coordinate vector of the neighborhood points. Eigenvalue decomposition is performed on the covariance matrix, and the eigenvector corresponding to the largest eigenvalue is taken as the tangent direction unit vector of the corresponding sampling point. The unit vectors of the tangent directions of all sampling points are arranged in the sampling order to form a sequence of tangent directions. The displacement vector during the exposure period is multiplied by the unit vector of the tangent direction at the corresponding sampling point in the tangent direction sequence. The result of the dot product is then multiplied by the unit vector of the tangent direction to obtain the projection component, which is used as the blur vector of the contour segment.

4. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 3, characterized in that: In step S2, the blur vector of the contour segment is first processed into a dimensionless form according to the magnitude of the displacement vector during the exposure period. Then, the dimensionless form is nonlinearly mapped using the Sigmoid function. Finally, it is multiplied by a preset upper limit of stretching and a direction sign is added to obtain the directional stretching amount. Using the coordinates of the corresponding sampling points on the observed contour curve as the initial values ​​for iteration, and combining the directional stretching amount with the unit vector along the tangent direction, an iterative deconvolution with directional prior is performed to obtain the iteration result; The Euclidean distance between two consecutive iterations is compared with a preset convergence threshold: When the Euclidean distance between two adjacent iterations is less than the preset convergence threshold, the iteration is determined to be converged, and the current contour coordinates of the sampling points are arranged in the sampling order as the deblurred observation contour output. Conversely, if the iteration fails to converge, the next iteration will be performed to correct the failure.

5. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 4, characterized in that: In step S3, for each sampling point on the deblurred observation contour, a local neighborhood with the corresponding sampling point as the center and a radius of the preset neighborhood is taken from the design standard contour data, and the minimum Euclidean distance from the sampling point of the deblurred observation contour to all points in the local neighborhood is calculated. The minimum Euclidean distance between the corresponding point of the sampled point in the design standard contour data and all points in the same local neighborhood of the deblurred observation contour is calculated in reverse. The larger of the two minimum values ​​is taken as the original bias residual of the sampling point.

6. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 1, characterized in that: In step S3, the phase encoding of the die-cutting mechanism is converted into the instantaneous phase of die-cutting through linear mapping. The instantaneous phase of die-cutting is the phase angle of the die-cutting mechanism in its complete punching cycle at the moment when the edge image of the current roll material is acquired after die-cutting. The original deviation residuals are binned according to the instantaneous phase of the die-cutting using the fuzzy C-means clustering algorithm, and the membership degree of the original deviation residuals of each sampling point to each bin is obtained.

7. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 6, characterized in that: In step S3, the membership degree of the original deviation residual point is compared with a preset membership threshold, and the absolute value of the difference between the instantaneous phase of the die-cutting corresponding to the original deviation residual point and the phase center of the corresponding bucket is compared with a preset phase window. When the membership degree of the original deviation residual point to a certain bucket is higher than the preset membership threshold, and the absolute value of the difference between the instantaneous phase of the original deviation residual point and the phase center of the bucket is less than the preset phase window, the original deviation residual is assigned to the same phase deviation component. When the membership degree of the original deviation residual point to all buckets is lower than or equal to the preset membership threshold, or when the absolute value of the difference between the instantaneous phase of the original deviation residual point and the phase center of its respective bucket is greater than or equal to the preset phase window, the original deviation residual is classified into the out-of-phase deviation component.

8. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 1, characterized in that: In step S4, the historical die-cutting residual time series is recorded as an ordered sequence of the original deviation residual values ​​at historical moments. An exponentially weighted moving average algorithm is used to generate a drift baseline for each original deviation residual value; The difference between the in-phase deviation component and the drift reference is measured using the Mahalanobis distance to obtain the Mahalanobis distance between the in-phase deviation component and the drift reference. Compare the Mahalanobis distance with the preset convergence window: When the Mahalanobis distance is less than the preset convergence window, the residual value of the latest moment in the historical die-cutting residual time series record is replaced by the in-phase deviation component to update the drift reference. When the Mahalanobis distance is greater than the preset convergence window, the current drift reference remains unchanged.

9. The packaging die-cutting accuracy correction method based on edge contour tracking according to claim 1, characterized in that: In step S4, the out-of-phase deviation components are dimensionless and mapped by logistic regression to generate a single-cycle correction command. The single-cycle correction command is output to the die-cutting mechanism, which then performs the corresponding blade roller phase adjustment or roll material lateral position adjustment at the next punching moment.

Images