A corrugated carton high-speed offset printing positioning control system based on machine vision

Abstract

This invention belongs to the field of electronic digital data processing technology and discloses a high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision. The synchronous acquisition module processes encoder pulse signals to convert rotation pulses into micron-level displacement digital trigger signals. The fusion positioning module synchronously extracts four types of positioning features and uses sub-pixel edge detection algorithms, adaptive weight scheduling, and random sampling consistency abnormal data removal algorithms to complete damaged features and filter invalid data, thereby calculating high-precision cardboard actual pose data. The error calculation module completes full-dimensional global error calculation and decomposes it into independent compensation parameters for each monochrome group. The feedforward control module builds a master-slave synchronous control architecture, combining feedforward control and fuzzy PID closed-loop control to synchronize the preceding error of each slave color group and compensate it in advance. The prediction compensation module constructs a dynamic delay model, uses a Kalman filter algorithm to predict deviations in advance, and issues synchronous compensation commands.

Description

Technical Field

[0001] This invention belongs to the field of electronic digital data processing technology, specifically a high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision. Background Technology

[0002] Currently, the operating speed of high-end corrugated box offset printing equipment in the industry continues to increase, and high-speed production has become the mainstream of industry development. Traditional mechanical positioning methods can no longer meet the accuracy requirements under high-speed conditions. Positioning control systems based on machine vision are gradually becoming the mainstream technical solution in the industry. However, in actual industrial applications, existing technologies still have the following technical problems in terms of electronic digital data processing: Existing vision positioning systems mostly adopt a serial digital processing architecture for image acquisition, image processing, and command output. The entire process of data operation and transmission has significant delays. Under high-speed operation, this delay will directly cause deviations in the physical displacement of the cardboard, resulting in a serious mismatch between the control data and the actual position data of the cardboard, which in turn leads to quality problems such as printing ghosting and pattern misalignment. Simply optimizing the hardware acquisition rate cannot fundamentally solve the delay defect at the data processing level.

[0003] The existing solution only uses a single set of quasi-marker data to complete the positioning calculation. It does not perform adaptive filtering and weight scheduling of feature data for corrugated cardboard defects and working condition interference. When faced with problems such as cardboard warping, edge roughness, surface scratches, and ink splatter, the failure rate and deviation rate of feature data recognition increase significantly. Under high-speed working conditions, the data noise caused by equipment vibration further amplifies the recognition error. There is no effective digital filtering or abnormal data removal mechanism, and the fault tolerance rate is extremely low.

[0004] Existing multicolor offset printing positioning control only adopts an open-loop and semi-closed-loop data processing architecture, which can only achieve successive fine-tuning of errors between single-color groups. It has not built a full-link error digital model and cannot complete global error feedforward compensation. As the number of color groups increases, the error data continues to accumulate. Under high-speed conditions, the digital processing delay further amplifies the deviation, which cannot meet the high-end printing registration accuracy requirements. Summary of the Invention

[0005] The purpose of this invention is to provide a high-speed offset printing positioning control system for corrugated boxes based on machine vision, so as to solve one or more problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision, comprising the following modules: Preferably, the benchmark modeling module acquires full-width raw image data based on a high-speed linear array vision pre-scanning unit, and completes digital benchmark modeling and positioning feature data preprocessing before the corrugated cardboard enters the offset printing unit; The benchmark modeling module acquires grayscale images of the cardboard to be printed using a pre-scan line array camera. Based on digital image processing algorithms, it performs image noise filtering, edge feature extraction, positioning mark pre-identification, and texture interference noise reduction. It also performs targeted texture smoothing for the multi-flute structure of corrugated cardboard, removes surface rough edges and wrinkles from cardboard production and transportation, and constructs an initial positioning model that includes cardboard physical size parameters, preset positioning mark benchmark coordinates, and cardboard edge benchmark contours. Simultaneously, it completes defect data marking, interference data masking, and feature data standardization and normalization.

[0007] Preferably, the synchronous acquisition module receives the standardized digital reference output by the reference modeling module, adapting to the requirements of motion distortion and acquisition timing deviation under high-speed working conditions; the synchronous acquisition module acquires the pulse digital signal of the encoder of the main motion shaft of the offset printing machine in real time, performs electromagnetic interference filtering on the signal, and converts the rotational motion pulse data into micron-level equal displacement digital trigger signals bound to the physical displacement of the paperboard through digital frequency division algorithm and phase-locked loop digital processing algorithm; Through the above digital processing logic, the printing bit array vision acquisition unit is driven to complete the image data acquisition, ensuring that the acquired image is not stretched or distorted under high-speed conditions of 300m / min or above, the pixel coordinates correspond to the physical position of the paperboard, the timing deviation is eliminated, and the original digital image data is obtained.

[0008] Preferably, the fusion positioning module receives the raw digital image data output by the synchronous acquisition module and builds the core positioning operation logic of the system; based on the synchronously acquired distortion-free digital images, a multi-feature adaptive weight scheduling fusion positioning model is built. The fusion positioning module simultaneously extracts four types of positioning features: registration marks, cardboard edge contours, corrugated peak textures, and printing reference corners. It then uses a sub-pixel edge detection algorithm to extract the coordinates of each feature. For scenarios involving ink smudging and slight cardboard deformation in high-speed printing, it prioritizes the extraction of effective features, completes the restoration of locally damaged features, and incorporates an adaptive weight allocation algorithm. This algorithm dynamically allocates computational weights in real time based on the completeness, clarity, and anti-interference capabilities of various feature data. Combined with a random sampling consistency algorithm, it eliminates abnormal positioning data points and finally obtains the actual pose data of the cardboard through data fitting.

[0009] Effective feature extraction priority is determined by the stability and recognizability of the features. Registration marks are given first priority, printing reference corners are given second priority, cardboard edge contours are given third priority, and corrugated peak textures are given fourth priority. When the proportion of a certain level feature that is damaged exceeds 50%, the weight allocation ratio of the next level complete feature is automatically increased by 0.1~0.2, and the total weight remains at 1. High-priority complete feature data is used first for pose calculation to ensure the effectiveness of the calculated data.

[0010] Preferably, the error calculation module receives the actual pose data of the cardboard output by the fusion positioning module, and, together with the previous reference digital model, builds a global calculation model for multi-degree-of-freedom overprinting error to complete the global error data calculation and color group specific compensation data decomposition. The error calculation module registers the cardboard reference pose data with the actual pose data, and calculates the global pose deviation data in all dimensions, such as lateral and longitudinal translation errors, rotation errors, and high-speed motion dynamic errors. Simultaneously, it verifies the rationality of the calculated errors and eliminates abnormal error values. Combining the physical arrangement parameters of each printing color group of the offset printing machine and the transmission link calculation parameters, the global error data is decomposed into independent compensation parameters for each color group. Pose deviation is the basic physical quantity of registration error, and registration error is the final manifestation of pose deviation in the printing process.

[0011] Preferably, the feedforward control module receives the independent compensation digital parameters of each monochrome group from the error calculation module and builds a multi-color group master-slave synchronous digital control architecture; with the calculated compensation data of each color group as the core, a real-time data interaction link for each color group is established to complete the timing pre-adjustment optimization. Using the first color group as the primary color group and the remaining color groups as synchronous slave color groups, each slave color group synchronizes the phase data of the primary color group, the error and compensation data of the preceding color group in real time. By combining feedforward digital control with fuzzy PID closed-loop digital control algorithm, the corresponding color group's roller phase and lateral displacement fine-tuning parameters are calculated in real time. The error of the preceding color group is compensated and synchronously corrected in the current and subsequent color groups in advance. The error accumulation channel is blocked from the data processing logic, and the closed-loop control response rate is optimized.

[0012] Preferably, the prediction compensation module receives real-time control data output by the feedforward control module, and combines it with the overall system runtime sequence to form a collaborative processing system with the synchronous acquisition module; The prediction and compensation module collects real-time delay data from image acquisition, algorithm processing, data transmission, and actuator response throughout the entire process. It dynamically corrects model parameters based on equipment runtime and ambient temperature changes to construct an adaptive iterative total system delay model. Combined with the Kalman filter prediction algorithm, it optimizes prediction logic for changing operating conditions using historical cardboard motion data, encoder real-time data, and error change trends. It predicts the cardboard's future position and deviation status in advance, generates compensation control commands, and sends them to the execution end.

[0013] For every hour the equipment runs, a micro-correction of 0.0001 seconds is made to the base delay value of the total system delay model to offset the mechanical response sluggishness caused by equipment operation. For every 1°C increase in ambient temperature from the baseline of 20°C, the delay parameters of the image acquisition and algorithm processing stages are corrected. For every 1°C decrease in ambient temperature, the delay parameters of the hydraulic / motor response stages of the actuator are corrected. After the model parameters are corrected, they are verified by actual operating data. If the delay prediction deviation still exceeds 0.001 seconds after the micro-correction, a new correction is immediately triggered, which overwrites all previous micro-correction results.

[0014] Preferably, the self-learning calibration module receives real-time operating data and global compensation and control results transmitted from each module of the system, continuously collects positioning error, overprinting accuracy, and digital parameters of the entire process of mechanism operation, constructs a system operation database, compares and iterates real-time calibration data with historical parameters, fits the system error drift law through machine learning algorithms, and automatically identifies systematic deviations such as lens distortion, phase shift, and transmission deviation; and completes fully automatic parameter calibration and dynamic updating of the benchmark model during non-production intervals of the equipment.

[0015] The fully automatic calibration of the self-learning calibration module is triggered only during non-production intervals, specifically during periods when there is no paperboard printing, such as offset printing machine stoppages, roll changes, or brief standby. After triggering, the system first pauses the issuance of positioning control commands, and then calls historical and real-time deviation data from the system's operating database to correct systematic deviations such as lens distortion, phase shift, and transmission deviation one by one. After the parameter calibration is completed, the corrected parameters are substituted into the initial positioning model to complete the dynamic update of the benchmark model. The update process is fully automatic and requires no manual operation. The time taken for a single calibration and model update is controlled within 1 minute, without affecting the normal production rhythm of the equipment.

[0016] The system's operating database is categorized by data type and divided into three main categories: positioning error data, overprinting accuracy data, and mechanism operating parameter data. Each type of data is associated with tags such as acquisition time, printing speed, and ambient temperature. Data storage adopts a cyclic overwrite rule, retaining only the most recent 90 days of valid operating data, while automatically filtering out invalid data generated during the acquisition process due to equipment failure or human error, ensuring that all data in the database is real data under valid operating conditions.

[0017] The beneficial effects of this invention are as follows: 1. The synchronous acquisition module of this invention processes encoder pulse signals to convert rotation pulses into micron-level displacement digital trigger signals, ensuring that the acquired images are distortion-free and that pixels correspond to physical positions at high speeds; the fusion positioning module synchronously extracts four types of positioning features, and through sub-pixel edge detection algorithms, adaptive weight scheduling, and random sampling consistency abnormal data removal algorithms, it completes damaged features and filters invalid data to obtain high-precision cardboard actual pose data.

[0018] 2. The error calculation module of this invention completes the global error calculation in all dimensions and decomposes it into independent compensation parameters for each monochrome group to achieve error-oriented decomposition; the feedforward control module builds a master-slave synchronous control architecture, combining feedforward control and fuzzy PID closed-loop control to synchronize the preceding error of each slave color group and compensate it in advance; the prediction compensation module constructs a dynamic delay model, predicts the deviation in advance through the Kalman filter algorithm and issues synchronous compensation commands.

[0019] 3. The self-learning calibration module of this invention continuously collects the operating data of the entire system and builds a database. By fitting the error drift pattern through machine learning algorithms, it can automatically identify systematic deviations such as lens distortion and phase shift. It also completes fully automatic parameter calibration and dynamic updating of the benchmark model during non-production intervals of the equipment. At the same time, the benchmark modeling module completes the marking of cardboard defects and the shielding of interference data in advance to avoid interference from invalid information. Attached Figure Description

[0020] Figure 1 This is a flowchart of the high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision, as described in this invention. Figure 2 This is a flowchart of the high-precision positioning calculation module of the present invention. Detailed Implementation

[0021] 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.

[0022] like Figures 1 to 2 As shown, this embodiment of the invention provides a high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision, including the following modules: The benchmark modeling module collects full-width raw image data based on a high-speed linear array vision pre-scanning unit, and completes digital benchmark modeling and positioning feature data preprocessing before the corrugated cardboard enters the offset printing unit. The benchmark modeling module acquires grayscale images of the cardboard to be printed using a pre-scan line array camera. Based on digital image processing algorithms, it performs image noise filtering, edge feature extraction, positioning mark pre-identification, and texture interference noise reduction. Targeted texture smoothing is performed on the multi-flute structure of the corrugated cardboard to remove surface roughness and wrinkles caused by cardboard production and transportation. An initial positioning model is constructed, including the physical size parameters of the cardboard, the preset positioning mark reference coordinates, and the cardboard edge reference contour. Simultaneously, defect data marking, interference data masking, and feature data standardization and normalization are completed to avoid invalid data interference caused by surface defects and warped textures on the cardboard.

[0023] The high-speed linear array vision pre-scanning unit's acquisition resolution is adapted to the printing width requirements of corrugated cardboard. The acquisition frame rate is adjusted in conjunction with the offset printing press's feed speed, achieving a frame rate of no less than 2000 lines per second even at a high speed of 300m / min. The grayscale image acquisition has 256 gray levels. The cardboard defect marking criteria are surface roughness exceeding 0.5mm and wrinkle area exceeding 1cm². 2 The area is automatically marked as a defective region, and the feature data in this area is directly masked. Feature extraction is only performed in the effective defect-free area.

[0024] The feature data standardization and normalization process uses the physical dimensions of corrugated cardboard as a benchmark. It transforms the pixel coordinates of all positioning features into physical coordinates with the lower left corner of the cardboard as the origin, and unifies the coordinate unit to millimeters. At the same time, it maps the grayscale, outline size and other data of the features to the range of 0-1, eliminates the feature data scale deviation caused by the difference in the size of different batches of cardboard and the change in the light of the acquisition, and ensures that the positioning feature data of different cardboards have a unified calculation standard.

[0025] The synchronous acquisition module receives the standardized digital benchmark output by the benchmark modeling module, adapting to the requirements of motion distortion and acquisition timing deviation under high-speed working conditions; the synchronous acquisition module acquires the pulse digital signal of the encoder of the main motion shaft of the offset printing machine in real time, performs electromagnetic interference filtering on the signal, and converts the rotational motion pulse data into micron-level equal displacement digital trigger signals bound to the physical displacement of the paperboard through digital frequency division algorithm and phase-locked loop digital processing algorithm. Through the above digital processing logic, the printing bit array vision acquisition unit is driven to complete the image data acquisition, ensuring that the acquired image is not stretched or distorted under high-speed conditions of 300m / min or above, the pixel coordinates correspond to the physical position of the paperboard, the timing deviation is eliminated, and the original digital image data is obtained.

[0026] The trigger interval of the micron-level displacement digital trigger signal is set according to the paperboard printing precision requirements, with a minimum trigger interval of 1 micron. The trigger signal is linked in real time with the rotation angle of the offset printing machine's main motion axis. Electromagnetic interference filtering adopts a combination of digital filtering and hardware filtering. First, high-frequency electromagnetic noise is filtered out by hardware filtering circuit, and then pulse interference points in the signal are eliminated by digital algorithm to ensure that the effective recognition rate of the encoder pulse signal reaches 100%. The acquisition action and response delay of the printing bit linear array vision acquisition unit are controlled within 0.001 seconds to ensure accurate acquisition timing. The synchronous acquisition module and the printed line array vision acquisition unit adopt a timed linkage calibration method. The equipment automatically performs a calibration once before starting production each day; during production, a quick calibration is performed every 4 hours of continuous operation. Quick calibration: only calibrates the synchronization timing of the trigger signal and vision acquisition, taking ≤30 seconds, with a matching deviation allowable value of ≤0.02mm; Regular calibration: full-process calibration, covering trigger signal synchronization, pixel-physical coordinate matching, and lens distortion correction, taking ≤3 minutes, with a matching deviation allowable value of ≤0.01mm. During calibration, the synchronous acquisition module outputs a standard trigger signal, the vision acquisition unit receives the signal and acquires an image of the standard calibration board. The system compares the pixel coordinates of the acquired image with the physical coordinates of the calibration board. If the matching deviation exceeds the preset value (±0.01mm), the acquisition parameters of the vision acquisition unit and the synchronization timing of the trigger signal are automatically adjusted. After adjustment, a secondary verification is immediately performed. If the secondary verification still fails, an equipment alarm is triggered, and production is suspended to ensure that the linkage accuracy of the two always meets the positioning requirements of high-speed printing.

[0027] The fusion positioning module receives the raw digital image data output by the synchronous acquisition module and builds the core positioning operation logic of the system; based on the synchronously acquired distortion-free digital images, it builds a multi-feature adaptive weight scheduling fusion positioning model to achieve sub-pixel level high-precision pose data calculation. The fusion positioning module simultaneously extracts four types of positioning features: registration marks, cardboard edge contours, corrugated peak textures, and printing reference corners. It then uses a sub-pixel edge detection algorithm to extract the coordinates of each feature. For scenarios involving ink smudging and slight cardboard deformation in high-speed printing, it prioritizes the extraction of effective features, completes the restoration of locally damaged features, and incorporates an adaptive weight allocation algorithm. This algorithm dynamically allocates computational weights in real time based on the completeness, clarity, and anti-interference capabilities of various feature data, automatically filtering out damaged, occluded, and noisy invalid feature data. Combined with a random sampling consistency algorithm, it eliminates abnormal positioning data points and finally obtains the actual pose data of the cardboard through data fitting. This approach can withstand multiple interferences such as equipment vibration and cardboard defects.

[0028] The input to the multi-feature adaptive weight scheduling fusion localization model is sub-pixel coordinate data of four types of localization features: registration mark pixel coordinates (x1, y1), cardboard edge contour pixel coordinate set {(x1, y1)}, and so on. 2n y 2n )}, Corrugated peak texture pixel coordinate set {(x 3n y 3n The model output is the actual pose data of the cardboard (lateral displacement X, longitudinal displacement Y, rotation angle θ). The core parameters of adaptive weight allocation are feature integrity coefficient K1 (0-1), feature clarity coefficient K2 (0-1), and anti-interference coefficient K3 (0-1). The single feature weight W = K1×0.4 + K2×0.3 + K3×0.3. The initial weights of the four types of features are all 0.25. The weights are dynamically normalized in real time to a sum of 1. When the weight of a damaged feature is lower than 0.1, it is automatically masked and filled by the other features. Subpixel edge detection employs a step-by-step detection method for four types of positioning features. Corner subpixel edge detection algorithm is used for registration marks and printing reference corners, while line subpixel edge detection algorithm is used for cardboard edge contours and corrugated peak textures. During the detection process, the gray-level gradient of the feature edges is continuously sampled to ensure the accuracy of feature coordinate extraction. When applying the random sampling consistency algorithm, a basic sample set is first randomly extracted from the feature coordinate data for pose fitting. Then, all feature coordinates are compared with the fitting results. Coordinate points with deviations exceeding the preset range are identified as outliers and removed. This operation is repeated until the confidence level of the fitting result reaches 99% or higher before the final cardboard pose data fitting is performed.

[0029] For locally damaged positioning features, a neighborhood feature completion method is used for calculation. If the registration mark or printing reference corner is partially occluded or ink is smudged, the effective feature texture and coordinate trend within 5mm around the damaged feature are extracted and interpolated for completion. If the cardboard edge contour or corrugated peak texture is locally broken, continuous fitting is performed based on the contour direction and peak spacing rules of adjacent areas. The completed feature data must be checked for grayscale consistency. Only after passing the check can it be included in the effective dataset of pose calculation. The grayscale consistency check is based on the grayscale difference between the completed area and the surrounding effective area. If the grayscale difference is ≤20, the check is considered passed; otherwise, the completion is repeated.

[0030] The error calculation module receives the actual pose data of the cardboard output by the fusion positioning module, and, together with the previous reference digital model, builds a global calculation model for multi-degree-of-freedom overprinting error, and completes the global error data calculation and color group-specific compensation data decomposition. The error calculation module registers the cardboard reference pose data with the actual pose data, and calculates the global pose deviation data in all dimensions, including lateral and longitudinal translation errors, rotation errors, and high-speed motion dynamic errors. Simultaneously, it verifies the rationality of the calculated errors and eliminates abnormal error values. Combining the physical arrangement parameters of each printing color group of the offset printing machine and the transmission link calculation parameters, the global error data is decomposed into independent compensation parameters for each color group, realizing accurate decomposition and directional classification of error data, and avoiding the spread and transmission of single error data.

[0031] The verification of the rationality of multi-degree-of-freedom overprinting error is based on the overprinting accuracy standard of the corrugated cardboard offset printing industry. When the horizontal and vertical translation error exceeds ±0.1mm, the rotation error exceeds ±0.05°, and the high-speed motion dynamic error exceeds ±0.08mm, it is judged as an abnormal error value and is directly rejected. When the global error is decomposed into independent compensation parameters for each monochrome group, it is allocated according to the transmission distance, transmission ratio and printing sequence difference between each monochrome group and the main color group. The color group close to the main color group is allocated compensation parameters according to the actual error value at a ratio of 1:1, and the color group far from the main color group is adapted and adjusted according to the error attenuation coefficient of the transmission link. The cardboard reference pose data and the actual pose data are matched using feature point registration. Preset positioning marks and printing reference corners are selected as core registration feature points. The coordinates of the actual collected feature points are matched one by one with the coordinates of the feature points in the reference model. During the registration process, the origin of the reference model is used as a fixed reference point to eliminate the registration error caused by coordinate system offset. After the registration is completed, the average value of the matching deviation of the feature points is calculated to ensure that the basic data for global error calculation is free from registration deviation interference.

[0032] The transmission link error attenuation coefficient is adapted to the actual transmission distance between the color group and the main color group. For each additional color group's transmission distance, the attenuation coefficient is finely adjusted by a fixed ratio. The adjustment of the attenuation coefficient is based on the accuracy detection data of the actual transmission link. If the actual operating deviation of a certain color group's transmission link is less than the preset value, the attenuation coefficient remains at the baseline value. If the actual operating deviation of a certain color group's transmission link is greater than the preset value (±0.05mm), the attenuation coefficient is increased by 1.2 times the actual deviation exceeding the value, with a maximum increase of no more than 0.5, to ensure that the compensation parameters are accurately matched with the actual transmission error.

[0033] The feedforward control module receives the independent compensation digital parameters of each monochrome group from the error calculation module; with the calculated compensation data of each monochrome group as the core, a multi-color group master-slave synchronous digital control architecture is built, and a real-time data interaction link for each monochrome group is established. In order to address the control lag problem caused by the spacing between monochrome groups and the transmission gap in high-speed printing, the timing pre-adjustment optimization is completed. With the first color group as the reference master color group, the remaining color groups are synchronous slave color groups. Each slave color group synchronizes the phase data of the master color group, the error and compensation data of the preceding color group in real time. By combining feedforward digital control with fuzzy PID closed-loop digital control algorithm, the corresponding color group's roller phase and lateral displacement fine-tuning parameters are calculated in real time. The error of the preceding color group is compensated and synchronously corrected in the current and subsequent color groups in advance. The error accumulation channel is blocked from the data processing logic, and the closed-loop control response rate is optimized. Optimize the control effect of multi-color group step-by-step printing deviation, while strictly controlling the delay of data calculation and instruction issuance to adapt to the real-time requirements of high-speed printing.

[0034] The inputs to the fuzzy PID closed-loop digital control algorithm in offset printing color group control are the pose deviation values ​​(ΔX, ΔY, Δθ) between the current color group and the main color group, and the rate of change of the deviation. The outputs are the phase fine-tuning amount Δφ and the lateral displacement fine-tuning amount ΔS of the color group roller. The fuzzification rule uses a triangular membership function to divide the deviation / deviation change rate into 7 fuzzy sets: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The core PID tuning parameter is the proportional coefficient Kp. Integral coefficient Ki Differential coefficients Kd The parameters are dynamically adjusted according to the offset printing speed. Under high-speed conditions above 300m / min, Kp is 4.0-5.0, Ki is 0.05-0.1, and Kd is 0.3-0.5. For every 50m / min increase in offset printing speed, Kp increases by 0.5, Ki decreases by 0.01, and Kd increases by 0.05, with the maximum not exceeding the above upper limit.

[0035] The multi-color group master-slave synchronous digital data interaction link uses industrial Ethernet to build a dedicated communication link. The master color group acts as the data sender, sending phase reference data and its own error compensation data to each slave color group in real time. Each slave color group only feeds back its own error calculation results and compensation execution status to the master color group, reducing the amount of data transmitted in the link. The data interaction transmission frequency is linked to the offset printing machine's operating speed. Under high-speed conditions above 300m / min, the transmission frequency is no less than 500 times / second to ensure the data synchronization of the master and slave color groups. The link communication delay is controlled within 0.003 seconds, with no data mistransmission or omission. The timing pre-adjustment optimization calculates the time it takes for the paperboard to be transferred to each slave color group based on the physical distance between each slave color group and the master color group and the operating speed of the offset printing machine. The phase data and error compensation data of the master color group are sent to the corresponding slave color group in advance, so that the compensation calculation and execution actions of the slave color group start in advance. The pre-adjustment time difference is perfectly matched with the paperboard transfer time. At the same time, a fixed compensation advance amount is added in the timing pre-adjustment to eliminate the instruction execution lag problem caused by the transmission gap.

[0036] The feedforward digital control and fuzzy PID closed-loop control adopt a superimposed execution logic of feedforward followed by closed-loop control. First, based on the error data of the preceding color group, the feedforward control outputs basic compensation parameters in advance to complete coarse adjustment compensation. Then, the fuzzy PID closed-loop control collects the actual pose deviation of the current color group in real time after the feedforward coarse adjustment and performs fine adjustment on the basic compensation parameters of the feedforward control to complete fine adjustment compensation. The coarse and fine adjustment actions are executed continuously without interval. The feedforward control provides parameter reference for the closed-loop control, and the closed-loop control corrects the deviation of the feedforward control, achieving a dual compensation effect.

[0037] The prediction compensation module receives real-time control data output by the feedforward control module, and combines it with the overall system runtime sequence to form a collaborative processing system with the synchronous acquisition module. This system manages the digital processing delay of the entire process under high-speed conditions, ensuring accurate matching between the positioning data and the cardboard position. The prediction and compensation module collects real-time delay data from image acquisition, algorithm processing, data transmission, and actuator response throughout the entire process. It dynamically corrects model parameters based on equipment runtime and ambient temperature changes to construct an adaptive iterative total system delay model. Combined with the Kalman filter prediction algorithm, it optimizes prediction logic for variable operating conditions such as cardboard acceleration and deceleration by using historical cardboard motion data, encoder real-time data, and error change trends. It predicts the cardboard's future position and deviation state in advance, generates compensation control commands in advance, and sends them to the execution end, ensuring that the compensation action is completely synchronized with the cardboard's arrival time.

[0038] The input state variables for the Kalman filter prediction algorithm in offset printing paperboard pose prediction are the real-time pose (X, Y, θ) of the paperboard, the running speed v, and the acceleration a. The measured variables are the encoder pulse data from the synchronous acquisition module and the pose data acquired by vision. The output is the predicted pose (X', Y', θ') and prediction deviation ΔE of the paperboard at a future time t (t=0.01-0.05s, dynamically adjusted according to system delay). The initial filter core parameters are set as process noise covariance Q=diag([0.01, 0.01, 0.001, 0.05, 0.1]) and measurement noise covariance R=diag([0.02, 0.02, 0.002]). Q and R are dynamically corrected according to the ambient temperature (20-40℃). For every 5℃ increase in temperature, the velocity and acceleration dimension parameters of Q are increased by 10%. The predictive compensation module and the synchronous acquisition module adopt a real-time bidirectional data interaction mode. The synchronous acquisition module transmits the real-time pulse data of the encoder and the timing data of the visual acquisition image to the predictive compensation module 1,000 times per second. The predictive compensation module synchronously feeds back the real-time system delay correction data to the synchronous acquisition module to adjust the timing of the trigger signal. The collaborative data interaction between the two adopts a high-speed serial port transmission method, and the data transmission packet loss rate is controlled within 0.01%, ensuring the real-time performance and accuracy of the data interaction.

[0039] The self-learning calibration module receives real-time operating data and global compensation and control results transmitted from various modules of the system, maintaining the long-term stable operation of the entire visual positioning control system and avoiding accuracy data drift. This module continuously collects positioning error, overprinting accuracy, and digital parameters of the entire mechanism operation process, constructs a system operation database, compares and iterates real-time calibration data with historical parameters, fits the system error drift pattern through machine learning algorithms, and automatically identifies systematic deviations such as lens distortion, phase shift, and transmission deviation. During non-production intervals, it completes fully automatic parameter calibration and dynamic updates of the benchmark model without manual intervention or shutdown, gradually optimizing the self-calibration accuracy and ensuring long-term stable operation of the equipment.

[0040] The self-learning calibration module uses a gradient boosting tree machine learning algorithm to fit the systematic error drift pattern. The algorithm's input feature set includes the device running time t, ambient temperature T, positioning error ΔE, overprinting accuracy value P, and transmission mechanism speed n. The output is the systematic deviation correction amount (ΔX correction, ΔY correction, Δθ correction) and the baseline model update coefficient K. The baseline model update coefficient K is the correction coefficient for the feature coordinates in the positioning baseline model. The value of K ranges from 0.98 to 1.02. The feature coordinates of the positioning baseline model = original baseline coordinates × K. When K > 1.02 or K < 0.98, it is determined that the model drift is too large, and the baseline model is immediately remodeled.

[0041] The triggering condition for model iteration and update is that the average positioning error of 100 consecutive cardboard sheets exceeds ±0.05mm, or the cumulative running time of the equipment exceeds 8 hours. The iteration step size parameter is 0.01. Each iteration adds the latest 500 sets of running data to the training set and removes historical redundant data that exceeds 30 days. After the systematic bias of the model identification is corrected, the deviation value must be controlled within ±0.02mm to complete the calibration. After the compensation control commands generated by each module are sent to the execution terminal, the execution terminal executes the actions according to the priority of the commands. The position deviation compensation command has a higher priority than the roller phase fine adjustment command. The emergency compensation command under high-speed conditions takes precedence over the regular compensation command. The action response accuracy of the actuator must match the offset printing positioning accuracy requirements. The roller phase fine adjustment response accuracy is not less than 0.001°, the lateral displacement adjustment response accuracy is not less than 0.01mm, and the action start delay of the actuator after receiving the compensation command is controlled within 0.002 seconds to ensure that the compensation action is accurately matched with the actual position of the cardboard.

[0042] A tiered adaptive adjustment rule is set for different offset printing speeds. Under normal conditions where the offset printing speed is below 300m / min, the data acquisition, transmission, and processing frequency of each module is appropriately reduced to decrease system resource consumption. Under high-speed conditions where the offset printing speed reaches 300m / min and above, the system automatically starts high-speed mode, synchronously increasing the trigger frequency of the synchronous acquisition module, the processing speed of the fusion positioning module, and the data interaction frequency between modules. At the same time, the prediction time of the prediction compensation module is adjusted in real time according to the speed. The faster the printing speed, the more dynamically the prediction time is adjusted according to the ratio of the total system delay value to the instantaneous speed of the paperboard. Under high-speed conditions of 300m / min and above, the prediction time is controlled within 0.01~0.03s to ensure that the compensation command is accurately synchronized with the paperboard arrival time, and to ensure that the working rhythm of each module is fully matched with the high-speed operation of the offset printing machine, maintaining positioning and registration accuracy throughout the process.

[0043] A dedicated response mechanism is implemented for common abnormal conditions in high-speed offset printing, such as cardboard acceleration / deceleration, changes in roll tension, and slight equipment vibration. During cardboard acceleration / deceleration, the prediction and compensation module adjusts the prediction time and compensation lead in real time, while the synchronous acquisition module increases the output frequency of the trigger signal. When changes in roll tension cause slight stretching or shrinkage of the cardboard, the fusion positioning module automatically identifies subtle changes in cardboard dimensions and corrects the pose calculation benchmark. When the equipment vibrates slightly, the fusion positioning module raises the threshold for rejecting abnormal data. Slight equipment vibration is defined as encoder pulse signal fluctuation amplitude ≤ 0.005 mm / ms. The abnormal data rejection threshold is increased from the conventional ±0.03 mm to ±0.05 mm. Simultaneously, the error calculation module temporarily stores five consecutive sets of pose data and averages them to calculate the error, avoiding positioning deviations caused by single vibrations. All responses to abnormal conditions are automatically triggered by the system without manual intervention, maintaining stable positioning accuracy.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. 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 process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision, characterized in that, include; The benchmark modeling module is used to perform digital preprocessing on the full-width image of the corrugated cardboard to be printed, filter out image noise and external interference data, complete benchmark parameter calibration, construct a standardized initial positioning model, and output cardboard positioning benchmark data. The synchronous acquisition module uses the encoder pulse signal of the offset printing equipment to achieve synchronous displacement triggering acquisition, performs special anti-interference processing on the original acquisition signal, and obtains a distortion-free original acquisition image. The fusion positioning module extracts multiple positioning features synchronously based on the distortion-free original image, and completes the multi-feature fusion pose calculation through multi-algorithm collaborative operation. It can resist multiple external interferences, including equipment operation vibration and cardboard defects, and finally outputs the real-time actual pose data of the cardboard. The error calculation module connects the reference positioning data and the actual pose data of the paperboard to complete the global registration error calculation. It decomposes the overall registration error into independent compensation parameters for each printing color group, realizing error classification and directional decomposition. The feedforward control module builds a multi-color group master-slave synchronous control architecture based on the independent compensation parameters of each color group. It adopts a composite logic of feedforward control and closed-loop control to realize synchronous pre-correction of multi-color group overprinting error and block the cumulative transmission of error from the control link. The prediction and compensation module collects system delay data throughout the entire process, constructs an adaptive iterative system total delay model, and combines a prediction algorithm to predict the cardboard pose deviation in advance, generating and issuing compensation commands in advance to ensure that the control commands match the real-time position of the cardboard under high-speed conditions. The self-learning calibration module collects full-process operation data of the entire system, fits the system error drift law, automatically identifies various systematic deviations in equipment operation, and completes fully automatic operation parameter calibration and dynamic update of the positioning reference model during non-production intervals.

2. The high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 1, characterized in that, The benchmark modeling module incorporates a high-speed linear array vision pre-scanning unit, which acquires grayscale images of the cardboard to be printed through a pre-scanning linear array camera. Based on digital image processing algorithms, it performs noise filtering, edge feature extraction, and positioning mark pre-identification. Targeted texture smoothing is applied to the special multi-flute structure of corrugated cardboard to remove surface roughness and wrinkles that cause interference data during cardboard production and transfer. Finally, an initial positioning model is constructed, including the physical size parameters of the cardboard, the preset positioning mark reference coordinates, and the cardboard edge reference contour. Simultaneously, interference data shielding and feature data standardization and normalization are completed.

3. The high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 2, characterized in that, The synchronous acquisition module acquires the pulse digital signal of the encoder of the main motion shaft of the offset printing machine in real time. First, it performs electromagnetic interference filtering on the signal, and then converts the equipment rotation motion pulse data into micron-level displacement digital trigger signals that are bound to the physical displacement of the paperboard through digital frequency division and phase-locked loop digital processing algorithms. The operation of the printing bit array vision acquisition unit is driven by a trigger signal, ensuring that the acquired image is not stretched or distorted under high-speed conditions, and that the pixel coordinates correspond one-to-one with the physical position of the paperboard.

4. The high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 3, characterized in that, The fusion positioning module incorporates a multi-feature adaptive weight scheduling fusion positioning model, which simultaneously extracts four core positioning features: registration marks, cardboard edge contours, corrugated peak textures, and printing reference corners. It extracts various feature coordinates through a sub-pixel edge detection algorithm. The feature extraction logic is optimized for common interference scenarios in high-speed printing. Combined with a dynamic weight allocation algorithm and a random sampling consistency abnormal data removal algorithm, it completes multi-feature data fitting and outputs the actual pose data of the cardboard.

5. A high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 4, characterized in that, The error calculation module constructs a global calculation model for multi-degree-of-freedom overprinting errors, registers the cardboard reference pose data with the actual pose data, calculates the global pose deviation in all dimensions, including lateral and longitudinal translation errors, rotation errors, and high-speed motion dynamic errors, and simultaneously completes error rationality verification and eliminates abnormal error values; then, combined with the physical arrangement parameters of each printing color group and the transmission link parameters of the offset printing machine, the global error is decomposed into independent compensation parameters for each color group.

6. A high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 5, characterized in that, The feedforward control module uses the first color group as the reference master color group and the other color groups as synchronous slave color groups to build a real-time data interaction link for multiple color groups. It adopts feedforward digital control combined with fuzzy PID closed-loop digital control algorithm to fine-tune the roller phase and lateral displacement parameters of the corresponding color group in real time, and compensates and corrects the error of the preceding color group in advance, thus blocking the error accumulation channel from the control logic.

7. A high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 6, characterized in that, The prediction compensation module and the synchronous acquisition module form a collaborative processing system to collect real-time delay data of the entire process, including image acquisition, algorithm processing, data transmission, and actuator response, and to construct an adaptive iterative total system delay model. Combined with the Kalman filter prediction algorithm, the cardboard pose deviation is predicted in advance based on the real-time operating data of the equipment, and compensation control commands are generated and sent to the execution terminal in advance.

8. A high-speed offset printing positioning control system for corrugated cardboard boxes based on machine vision according to claim 7, characterized in that, The self-learning calibration module builds a system operation database, continuously collects positioning errors, overprinting accuracy, and digital parameters of the entire process of mechanism operation, fits the system error drift pattern through machine learning algorithms, automatically identifies various systematic operational deviations of the equipment, and completes fully automatic parameter calibration and dynamic iterative updates of the positioning reference model during non-production intervals of the equipment.

Images