Flexographic printing machine control method and system based on machine vision

Abstract

The invention relates to the technical field of printing, in particular to a flexo printing machine control method and system based on machine vision. The method comprises the following steps: collecting surface image data of a printed matter in a motion state in real time, separating and identifying a preset printing feature region from the image data, and obtaining geometric feature parameters of the region relative to a reference position; generating a multi-axis cooperative position compensation instruction of the printing unit according to the deviation value of the geometrical characteristic parameter and a preset target characteristic; converting the position compensation instruction into an adjusting parameter of a servo driving system in real time, and dynamically adjusting the pressure of a printing roller of a printing unit and the displacement of a registration mechanism; and synchronously outputting a defect type judgment signal and correlating the adjustment parameter correction value based on the form change of the printing feature region in the updated image data. According to the method, the stability and the adjusting efficiency of the printing quality are improved, and the defects that a traditional method is lagged in response and inaccurate in adjustment are overcome.

Description

Technical Field

[0001] The present invention relates to the field of printing technology, and in particular to a machine vision-based flexographic printing press control method and system. Background Art

[0002] Flexographic printing, a highly efficient and environmentally friendly printing technology, is widely used in the packaging industry. However, the printing process is prone to defects such as misregistration and uneven ink layers due to mechanical vibration, material deformation, and multi-axis motion coupling. Traditional control methods rely primarily on offline detection and manual adjustment, making them difficult to adapt to the real-time control requirements of high-speed, dynamic printing scenarios. While existing machine vision technologies can detect surface defects in printed products, feature extraction accuracy is significantly affected by motion blur and texture interference, and they lack a coordinated compensation mechanism for multi-axis servo systems. Especially under high-speed printing conditions, material property variations and mechanical response delays lead to nonlinear error propagation, making existing PID-based control models unable to effectively decouple multi-degree-of-freedom error components. Furthermore, closed-loop optimization of the dynamic relationship between defect identification and control parameters has not yet been established, leading to control lag and overcompensation. Although some research has attempted to improve detection accuracy using frequency domain analysis or neural networks, bottlenecks such as high algorithm complexity and insufficient real-time performance remain in phase compensation, dynamic decoupling, and defect traceability. Therefore, there is an urgent need for an intelligent control method that integrates sub-pixel texture feature recognition, multi-axis collaborative compensation and defect feedback optimization to solve the problem of collaborative optimization of precision, efficiency and stability in high-speed flexographic printing. Summary of the Invention

[0003] The present invention overcomes the deficiencies of the prior art and provides a machine vision-based flexographic printing press control method and system.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: The first aspect of the present invention discloses a flexographic printing press control method based on machine vision, comprising the following steps: Collecting surface image data of a printed product in motion in real time, separating and identifying a preset printing feature area from the image data, and obtaining geometric feature parameters of the area relative to a reference position; generating a multi-axis coordinated position compensation instruction for a printing unit according to a deviation between the geometric feature parameter and a preset target feature; Convert the position compensation instruction into the adjustment parameter of the servo drive system in real time, and dynamically adjust the printing roller pressure and the registration mechanism displacement of the printing unit; Based on the morphological changes of the printing feature area in the updated image data, a defect type determination signal is synchronously output and an associated adjustment parameter correction value is associated.

[0005] Preferably, a preset printing feature area is separated and identified from the image data, and geometric feature parameters of the area relative to a reference position are obtained, specifically: Perform motion trajectory compensation filtering on the collected continuous frame surface images to generate a feature-enhanced image with de-motion blur; In the feature-enhanced image, the printed mark area with micro-texture features is separated as the texture fingerprint through the preset chroma key threshold and morphological gradient detection; Performing a local Fourier descriptor transform on the texture fingerprint to generate a rotation-invariant feature map including spatial frequency distribution characteristics; Performing a cross-correlation operation on the phase of the rotation-invariant feature map and the reference template to obtain a sub-pixel offset vector of the texture fingerprint center point relative to a preset reference grid; Based on the ratio of the offset vector components in the printing travel direction and the axial direction, the stretching distortion rate and torsion angle parameters of the printed feature area are calculated.

[0006] Preferably, a multi-axis coordinated position compensation instruction of the printing unit is generated according to the deviation between the geometric feature parameter and the preset target feature, specifically: The detected sub-pixel offset vector, stretch distortion rate and torsion angle parameters are input into the pre-trained error space mapping model and decomposed into three independent degrees of freedom: axial registration error component, radial pressure error component and circumferential torsion error component through nonlinear transformation. A dynamic decoupling matrix is ​​constructed using real-time updated material property coefficients to perform orthogonal decoupling operations on the three-degree-of-freedom error components. The nonlinear coupling strength factors between the degrees of freedom are simultaneously calculated to generate independent error corrections with coupling compensation. According to the real-time linear speed of the printing unit and the response delay parameters of the servo system, a phase advance correction factor is superimposed on the independent error correction value, and a dynamic reference instruction with time domain advance characteristics is generated through a pre-compensation algorithm; The dynamic reference instructions are input into the multi-axis collaborative optimizer based on the mechanical configuration constraints of the printing unit. A fuzzy weight allocation strategy is used to coordinate the displacement instruction weights of the printing roller pressure axis, axial registration axis and circumferential torsion compensation axis, and the optimal position compensation instruction sequence that meets the mechanical interference boundary conditions is output.

[0007] The material property coefficients include the elastic modulus of the substrate and the ink viscosity threshold.

[0008] Preferably, according to the real-time linear speed of the printing unit and the response delay parameter of the servo system, a phase advance correction factor is superimposed on the independent error correction amount, and a dynamic reference instruction with a time domain advance characteristic is generated through a pre-compensation algorithm, specifically: Based on the step response test data of the servo system, the response delay parameter is decomposed into a phase lag component and an amplitude attenuation component, and a discretized transfer function model positively correlated with the linear velocity is established. The independent error correction amount is converted into an amplitude-phase vector representation. When it is detected that the phase lag component exceeds a preset threshold, the vector is subjected to a complex domain phase compensation operation with a dynamically adjusted lead correction factor to generate a frequency domain compensation vector with a phase lead characteristic. Perform inverse transformation on the frequency domain compensation vector, extract its real part as the time domain amplitude envelope signal, and retain the phase advance compensation amount constructed by the imaginary part; When the real-time linear velocity change rate exceeds the set threshold, the gain coefficient is dynamically adjusted according to the product of the amplitude attenuation component and the phase advance compensation amount, and a time domain reference curve for anti-saturation limiting is generated through a sliding integrator; The time domain amplitude envelope signal and the phase advance compensation amount are fused in the hybrid domain to output a dynamic reference instruction that compensates for both amplitude attenuation and phase lag.

[0009] The step response test data includes the response delay time, rise time, overshoot and steady-state establishment time of the servo system from command input to steady-state output establishment.

[0010] Preferably, the position compensation instruction is converted into an adjustment parameter of the servo drive system in real time to dynamically adjust the printing roller pressure and the registration mechanism displacement of the printing unit, specifically: receiving an optimal position compensation instruction sequence outputted by a multi-axis collaborative optimizer, wherein the optimal position compensation instruction sequence comprises three independent components: a printing roller pressure axis displacement instruction, an axial registration axis displacement instruction, and a circumferential torsion compensation axis displacement instruction; For each displacement command component, combined with the phase lag component and amplitude attenuation component pre-stored in the servo system, an initial parameterization conversion is performed through the dynamic gain coefficient mapping table to generate the corresponding axis-specific drive parameters; The axis-specific drive parameters and the phase advance compensation amount are hybrid-domain fused in the complex domain to generate a compensated drive vector with time-domain advance characteristics. Applying a sliding integrator to the compensated driving vector to perform anti-saturation limiting processing and output a smooth time-domain amplitude envelope signal; Based on the time domain amplitude envelope signal, the pulse frequency modulation duty cycle parameter and the axial displacement step parameter are calculated in real time through the servo drive interface module; The output torque of the printing roller pressure servo motor is dynamically adjusted by utilizing the pulse frequency modulation duty cycle parameter, and the linear displacement of the registration mechanism is precisely controlled by the axial displacement step parameter, thereby realizing coordinated dynamic adjustment of the printing roller pressure and the registration mechanism displacement.

[0011] Preferably, based on the morphological changes of the printed feature area in the updated image data, a defect type determination signal is synchronously output and an associated adjustment parameter correction value is associated, specifically: Multi-scale morphological gradient decomposition is performed on the updated rotation-invariant feature map collected in real time, and the edge sharpness attenuation and micro-texture diffusion of the printed feature area are extracted as morphological degradation feature vectors. Inputting the morphological degradation feature vector into a pre-trained defect classification convolution kernel group, and generating a three-dimensional probability distribution vector including the confidence of ink layer diffusion defect, dot loss defect and registration shift defect through a spatial attention weighting mechanism; Based on the maximum entropy increase rate direction of each component in the three-dimensional probability distribution vector and the time-domain derivative of the current phase advance compensation, the dynamic coupling coefficient between the entropy increase rate of the defect morphology and the servo adjustment response is calculated. The defect morphology entropy increase rate is input into the inverse transformation module of the dynamic decoupling matrix, and the defect traceability characteristic scalar corresponding to the axial registration error component, radial pressure error component and circumferential torsion error component is generated through orthogonal projection decomposition; According to the ratio between the defect traceability characteristic scalar and the preset safety threshold, the three-dimensional probability distribution vector is mapped into a discrete defect type judgment signal, and the defect traceability characteristic scalar is converted into a signed parameter deviation vector. A tensor dot product operation is performed on the parameter deviation vector and the position compensation instruction output by the multi-axis collaborative optimizer to generate a parameter correction association matrix that integrates the defect traceability characteristics. The row vectors of this matrix correspond to the defect type judgment signal encoding, and the column vectors are associated with the dynamic reference instruction correction value of the servo drive system.

[0012] The second aspect of the present invention discloses a flexographic printing press control system based on machine vision, the flexographic printing press control system includes a memory and a processor, the memory stores a flexographic printing press control method program, when the flexographic printing press control method program is executed by the processor, the steps of any one of the flexographic printing press control methods are implemented.

[0013] The present invention solves the technical defects existing in the background technology, and has the following beneficial effects: the present invention uses machine vision to collect and analyze the geometric parameters of the printing feature area in real time, combines the multi-axis collaborative control algorithm to generate dynamic compensation instructions, and realizes the precise and rapid adjustment of the printing unit; at the same time, closed-loop feedback is performed based on the compensated image data, and the defect judgment results are directly linked to the parameter correction, forming a complete control closed loop from detection to adjustment, thereby improving the stability of printing quality and the adjustment efficiency, and effectively overcoming the defects of traditional methods such as delayed response and inaccurate adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0015] Figure 1 This is a flow chart of the overall method of the flexographic printing machine control method; Figure 2 This is a partial flow chart of the control method of the flexographic printing machine; Figure 3 This is the system block diagram of the flexographic printing machine control system. DETAILED DESCRIPTION

[0016] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0018] like Figure 1 As shown, the first aspect of the present invention discloses a flexographic printing press control method based on machine vision, comprising the following steps: S102, collecting surface image data of a printed product in a moving state in real time, separating and identifying a preset printing feature area from the image data, and obtaining geometric feature parameters of the area relative to a reference position; S104, generating a multi-axis coordinated position compensation instruction for a printing unit according to a deviation between the geometric feature parameter and a preset target feature; S106, converting the position compensation instruction into adjustment parameters of the servo drive system in real time, and dynamically adjusting the printing roller pressure and registration mechanism displacement of the printing unit; S108 , based on the morphological changes of the printing feature area in the updated image data, synchronously output a defect type determination signal and associate it with an adjustment parameter correction value.

[0019] Among them, the updated image data refers to the surface image of the printed product that is re-collected after real-time position compensation adjustment, which contains the morphological change information of the printing feature area after dynamic adjustment, and is used for defect judgment and parameter closed-loop optimization.

[0020] It should be noted that the present invention uses machine vision to collect and analyze the geometric parameters of the printing feature area in real time, combines it with a multi-axis collaborative control algorithm to generate dynamic compensation instructions, and realizes accurate and rapid adjustment of the printing unit; at the same time, closed-loop feedback is performed based on the compensated image data, and the defect judgment results are directly linked to parameter correction, forming a complete control closed loop from detection to adjustment, thereby improving the stability of printing quality and adjustment efficiency, and effectively overcoming the defects of traditional methods such as delayed response and inaccurate adjustment.

[0021] Preferably, a preset printing feature area is separated and identified from the image data, and geometric feature parameters of the area relative to the reference position are obtained, such as Figure 2 As shown, specifically: S202, performing motion trajectory compensation filtering on the collected continuous frame surface images to generate a feature-enhanced image with motion blur removed; S204, in the feature-enhanced image, separating the printed mark area with microtexture features as a texture fingerprint through a preset chroma key threshold and morphological gradient detection; S206, performing a local Fourier descriptor transform on the texture fingerprint to generate a rotation-invariant feature map including spatial frequency distribution characteristics; S208, performing a cross-correlation operation on the phase of the rotation-invariant feature map and the reference template to obtain a sub-pixel offset vector of the texture fingerprint center point relative to a preset reference grid; The reference template refers to a pre-stored digitized reference image of a standard printing feature area, which contains texture features and geometric parameters under ideal conditions and is used to compare with the actual collected printing features to calculate the deviation.

[0022] S210 , based on the ratio of the offset vector components in the printing travel direction to the axial direction, calculating the stretching distortion rate and torsion angle parameters of the printed feature area.

[0023] It should be noted that the generated rotation-invariant feature map is subjected to phase cross-correlation in the frequency domain with the pre-stored reference template, and the optimal matching position of the two is determined by peak detection. Then, the cross-correlation result is precisely positioned at the sub-pixel level using the cubic spline interpolation algorithm to determine the offset vector of the texture fingerprint center point relative to the reference grid (including components in the X-axis and Y-direction directions). Finally, based on the ratio of the Y-direction to the X-direction offset, combined with the preset printing material characteristic parameters, the actual tensile distortion rate and torsion angle of the printed feature area are calculated using trigonometric function relationships.

[0024] In one embodiment of the present invention, for example, in a high-speed flexographic printing line (line speed 200 m / min), continuous frame images (resolution 5 μm / pixel) captured in real time by the printing unit are subjected to motion trajectory compensation filtering (exposure time 1 ms, compensation angle 15°) to generate a deblurred image. A texture fingerprint of the preset micro-character mark "FX-0.3" (size 0.3×0.3 mm) is isolated using a chroma key threshold (cyan channel C ≥ 85%) combined with morphological gradient detection (structuring element size 3×3). A local Fourier descriptor transform (window size 64×64 pixels) is performed on this fingerprint to generate a rotationally invariant feature map. This map is then subjected to phase cross-correlation with a reference template (grid accuracy 0.1 pixel) to measure the center point offset vector (X: -0.07 mm, Y: +0.12 mm). Based on the ratio of the Y (printing direction) to X (axial) offset of 1.71, a tensile distortion rate of 0.8% and a torsion angle of 0.6° are calculated.

[0025] In summary, the present invention achieves high-precision feature positioning and deformation parameter calculation at the sub-pixel level, which can improve the stability of printing feature recognition and the accuracy of geometric deviation detection, and provide a reliable data basis for subsequent multi-axis collaborative compensation of printing machines.

[0026] Preferably, a multi-axis coordinated position compensation instruction of the printing unit is generated according to the deviation between the geometric feature parameter and the preset target feature, specifically: The detected sub-pixel offset vector, stretch distortion rate and torsion angle parameters are input into the pre-trained error space mapping model and decomposed into three independent degrees of freedom: axial registration error component, radial pressure error component and circumferential torsion error component through nonlinear transformation. The error space mapping model is a nonlinear transformation model established through machine learning, linking print geometry deviations (such as offset vectors and deformation parameters) with multi-axis control system error components (registration, pressure, and torsion). The training process begins with historical data collected under typical operating conditions (e.g., line speeds of 100-250 m / min and various substrates, such as PE / PP film). This includes sub-pixel offset vectors, tensile distortion, and torsion angle parameters as inputs. High-precision sensors measure axial registration error, radial pressure error, and circumferential torsion error as target outputs. After normalization of 10,000 samples, a support vector regression (SVR) algorithm combined with a radial basis kernel function is used for nonlinear fitting. A grid search is used to optimize the penalty factor C = 1.2 and the kernel parameter γ = 0.05. Finally, the model is validated using mean squared error (MSE < 0.003) as the metric, and dynamic generalization testing is performed within a closed-loop control system to ensure an error decomposition accuracy exceeding 98%.

[0027] A dynamic decoupling matrix is ​​constructed using real-time updated material property coefficients to perform orthogonal decoupling operations on the three-degree-of-freedom error components. The nonlinear coupling strength factors between the degrees of freedom are simultaneously calculated to generate independent error corrections with coupling compensation. It should be noted that, first, the elastic modulus of the current substrate (such as 2.1GPa for PE film) and the ink viscosity threshold (such as 450cP) are obtained in real time as material property coefficients; secondly, a dynamic decoupling matrix is ​​constructed based on these coefficients - the elastic modulus is converted into a stiffness weight factor for radial pressure error, and the ink viscosity is mapped into a damping proportional coefficient for axial / circumferential error; then, the three error components of axial registration, radial pressure, and circumferential torsion are orthogonally decoupled: the cross-interference terms between the components are eliminated through matrix transformation (such as the influence of axial displacement on circumferential torsion); at the same time, the nonlinear coupling strength factor (such as 0.25) is determined to quantify the degree of mutual interference between the three degrees of freedom; finally, the inverse compensation value of the coupling strength is superimposed on the independent error correction amount (for example, the axial error correction amount -0.05mm is superimposed on the coupling compensation amount -0.012mm, and the output is -0.062mm), and the final correction instruction with dynamic compensation is generated to ensure that the control quantities of each axis are independent of each other and offset the system coupling effect.

[0028] According to the real-time linear speed of the printing unit and the response delay parameters of the servo system, a phase advance correction factor is superimposed on the independent error correction value, and a dynamic reference instruction with time domain advance characteristics is generated through a pre-compensation algorithm; The dynamic reference instructions are input into the multi-axis collaborative optimizer based on the mechanical configuration constraints of the printing unit. A fuzzy weight allocation strategy is used to coordinate the displacement instruction weights of the printing roller pressure axis, axial registration axis and circumferential torsion compensation axis, and the optimal position compensation instruction sequence that meets the mechanical interference boundary conditions is output.

[0029] The multi-axis collaborative optimizer coordinates and optimizes compensation commands for multiple axes (pressure, registration, and torsion) based on the mechanical constraints and kinematic models of the printing unit, ensuring coordinated movement of each axis without exceeding mechanical limits. The fuzzy weight allocation strategy dynamically adjusts the weight coefficients of each compensation axis' commands (e.g., 0.6 for the pressure axis and 0.3 for the registration axis) through fuzzy logic based on real-time operating conditions (e.g., error severity and axis response speed) to achieve optimal allocation of control resources.

[0030] The material property coefficients include the elastic modulus of the substrate and the ink viscosity threshold.

[0031] Preferably, according to the real-time linear speed of the printing unit and the response delay parameter of the servo system, a phase advance correction factor is superimposed on the independent error correction amount, and a dynamic reference instruction with a time domain advance characteristic is generated through a pre-compensation algorithm, specifically: Based on the step response test data of the servo system, the response delay parameter is decomposed into a phase lag component and an amplitude attenuation component, and a discretized transfer function model positively correlated with the linear velocity is established. The independent error correction amount is converted into an amplitude-phase vector representation. When it is detected that the phase lag component exceeds a preset threshold, the vector is subjected to a complex domain phase compensation operation with a dynamically adjusted lead correction factor to generate a frequency domain compensation vector with a phase lead characteristic. Perform inverse transformation on the frequency domain compensation vector, extract its real part as the time domain amplitude envelope signal, and retain the phase advance compensation amount constructed by the imaginary part; When the real-time linear velocity change rate exceeds the set threshold, the gain coefficient is dynamically adjusted according to the product of the amplitude attenuation component and the phase advance compensation amount, and a time domain reference curve for anti-saturation limiting is generated through a sliding integrator; The time domain amplitude envelope signal and the phase advance compensation amount are fused in the hybrid domain to output a dynamic reference instruction that compensates for both amplitude attenuation and phase lag.

[0032] The step response test data includes the response delay time, rise time, overshoot and steady-state establishment time of the servo system from command input to steady-state output establishment.

[0033] In one embodiment of the present invention, in a high-speed flexographic printing unit (line speed 200 m / min, substrate 0.2 mm thick PE film), based on the detected offset vector (X: -0.07 mm, Y: +0.12 mm), tensile distortion rate of 0.8%, and torsion angle of 0.6°: first, a pre-trained error space mapping model (300 sets of historical data) is used to decompose the axial registration error of -0.05 mm, radial pressure error of +120 N (ink viscosity threshold 450 cP), and circumferential torsion error of 0.4°; combined with the real-time updated PE film elastic modulus (2.1 GPa), a dynamic decoupling matrix (coupling strength factor 0.25) is constructed to generate independent correction values ​​with coupling compensation (axial -0.062 mm, radial +110 N, circumferential +0.32°); then, a transfer function is established based on the servo step response data (delay time 50 ms). Model, for example, when the linear speed suddenly changes to 220 m / min, the phase advance correction factor (compensation angle 20°) is superimposed on the independent correction amount to generate a frequency domain compensation vector, and the time domain amplitude envelope (radial force 105 N) is extracted through inverse transformation and combined with the sliding integrator to generate an anti-saturation reference curve; finally, the multi-axis collaborative optimizer (registration axis displacement limit ±0.1 mm, fuzzy weight: pressure axis 0.6 / registration axis 0.3 / torsion axis 0.1) outputs the optimal compensation instruction sequence: printing roller pressure axis -0.15 mm (corresponding to -108 N), axial registration axis -0.058 mm, circumferential torsion axis +0.28°, forming a closed-loop control link with the previous detection parameters.

[0034] In summary, in order to solve the technical problems of difficulty in compensating multi-axis error coupling in flexographic printing, untimely adjustment due to servo system response lag, and changes in material properties affecting control accuracy, the present invention achieves independent and accurate compensation of multi-degree-of-freedom errors through error space mapping and dynamic decoupling operations; effectively overcomes system response delays by combining phase advance correction and time domain pre-compensation algorithms; and finally, outputs optimal compensation instructions through multi-axis collaborative optimization, thereby improving the collaborative control accuracy and dynamic response speed of the multi-axis system under high-speed printing conditions and ensuring the stability of printing quality.

[0035] Preferably, the position compensation instruction is converted into an adjustment parameter of the servo drive system in real time to dynamically adjust the printing roller pressure and the registration mechanism displacement of the printing unit, specifically: receiving an optimal position compensation instruction sequence outputted by a multi-axis collaborative optimizer, wherein the optimal position compensation instruction sequence comprises three independent components: a printing roller pressure axis displacement instruction, an axial registration axis displacement instruction, and a circumferential torsion compensation axis displacement instruction; For each displacement command component, combined with the phase lag component and amplitude attenuation component pre-stored in the servo system, an initial parameterization conversion is performed through the dynamic gain coefficient mapping table to generate the corresponding axis-specific drive parameters; The dynamic gain coefficient mapping table refers to a gain parameter comparison table prepared in advance and dynamically adjusted according to the servo system characteristics and real-time working conditions, and is used to convert the displacement command into a drive signal that matches the response characteristics of each axis.

[0036] The axis-specific drive parameters and the phase advance compensation amount are hybrid-domain fused in the complex domain to generate a compensated drive vector with time-domain advance characteristics. Applying a sliding integrator to the compensated driving vector to perform anti-saturation limiting processing and output a smooth time-domain amplitude envelope signal; The sliding integrator is an integration module with a time window. It smoothes the command signal through sliding average to avoid sudden changes and suppress integral saturation. Anti-saturation limiting processing involves amplitude limiting and integral reset control of the drive signal to prevent the servo actuator from entering a saturated state due to overload commands.

[0037] Based on the time domain amplitude envelope signal, the pulse frequency modulation duty cycle parameter and the axial displacement step parameter are calculated in real time through the servo drive interface module; Among them, the servo drive interface module is a conversion unit that combines hardware and software, which converts the control signal into PWM duty cycle and step pulse instructions that can be executed by the servo motor.

[0038] The output torque of the printing roller pressure servo motor is dynamically adjusted by utilizing the pulse frequency modulation duty cycle parameter, and the linear displacement of the registration mechanism is precisely controlled by the axial displacement step parameter, thereby realizing coordinated dynamic adjustment of the printing roller pressure and the registration mechanism displacement.

[0039] In one embodiment of the present invention, in a high-speed flexographic printing unit (line speed 200 m / min), the compensation instruction sequence output by the multi-axis collaborative optimizer of the previous example is received (pressure axis -0.15 mm, registration axis -0.058 mm, torsion axis +0.28°): combined with the pre-stored servo characteristics (phase lag 30 ms / amplitude attenuation 15%), the initial drive parameters (pressure axis -0.18 V, registration axis -0.046 V) are generated through the dynamic gain coefficient mapping table (pressure axis gain 1.2, registration axis 0.8); the initial parameters are compared with the previous example. The phase advance compensation amount (compensation angle 18°) is fused in the complex domain to generate a compensated drive vector (real part -0.175V, imaginary part +0.032j). A sliding integrator (window width 50ms, amplitude limit ±5V) outputs a smooth time-domain signal (pressure axis amplitude envelope -0.172V). The servo interface module converts the envelope signal into a 62% PWM duty cycle for the pressure servo motor (corresponding to a torque of -108N·m) and a registration step size of -0.057mm. This synchronously drives the printing roller to decompress and the registration mechanism to retract, achieving coordinated dynamic adjustment.

[0040] In summary, to address the technical issues of insufficient compensation accuracy caused by servo drive system response lag and asynchronous multi-axis adjustment in traditional flexographic printing presses, the present invention achieves high-precision command conversion by dynamically fusing position compensation commands with servo system characteristic parameters and employing complex-domain lead compensation and anti-saturation processing. Finally, through pulse frequency modulation and step-by-step calculation, pressure and displacement adjustment are synchronously controlled, improving the dynamic response speed and coordinated control accuracy of the printing roller pressure and register mechanism displacement, ensuring that the printing unit executes compensation actions quickly and stably.

[0041] Preferably, based on the morphological changes of the printed feature area in the updated image data, a defect type determination signal is synchronously output and an associated adjustment parameter correction value is associated, specifically: Multi-scale morphological gradient decomposition is performed on the updated rotation-invariant feature map collected in real time, and the edge sharpness attenuation and micro-texture diffusion of the printed feature area are extracted as morphological degradation feature vectors. Inputting the morphological degradation feature vector into a pre-trained defect classification convolution kernel group, and generating a three-dimensional probability distribution vector including the confidence of ink layer diffusion defect, dot loss defect and registration shift defect through a spatial attention weighting mechanism; The pre-trained defect classification convolutional kernel set refers to a set of specialized convolutional filters pre-trained on a large number of printing defect samples (e.g., 100,000 annotated images containing ink layer spread, dot dropout, and registration shift). The training process involves constructing a neural network containing multi-scale convolutional kernels, such as 3×3, 5×5, and 7×7, and performing end-to-end training on the defect dataset using transfer learning (using a ResNet50 backbone network). A spatial attention mechanism (SE module) automatically learns feature weights for different regions (e.g., a dot region weight coefficient of 0.8). Parameters are optimized using a cross-entropy loss function, ultimately retaining a set of convolutional kernels (e.g., 32 kernels) that effectively extract features such as ink layer non-uniformity and dot deformation. These parameters are fixed during deployment to enable real-time classification of morphological degradation feature vectors (inference speed ≤ 5ms / frame). This kernel set is rotationally invariant and scale-adaptive, making it directly embeddable in real-time control systems.

[0042] Based on the maximum entropy increase rate direction of each component in the three-dimensional probability distribution vector and the time-domain derivative of the current phase advance compensation, the dynamic coupling coefficient between the entropy increase rate of the defect morphology and the servo adjustment response is calculated. The specific calculation steps are as follows: First, determine the direction of the component with the fastest probability change in the three-dimensional probability distribution vector (for example, the ink layer diffusion confidence level changes from 0.72 to 0.85, with an entropy increase rate of 0.12 / s). Then, obtain the current rate of change of the phase advance compensation (for example, a time-domain derivative of 0.05 mm / s² indicates the trend of compensation strength change). Next, multiply the entropy increase rate by the derivative (0.12 / s × 0.05 mm / s²), and then multiply by a preset adjustment strength factor (for example, an empirical coefficient of 0.85), to finally generate the dynamic coupling coefficient (0.85 in this example). This coefficient quantifies the degree of match between the defect deterioration rate and the servo control response rate. A value greater than 0.8 indicates a need for increased compensation, while a value less than 0.3 indicates an over-response.

[0043] The defect morphology entropy increase rate is a quantitative indicator of the time-dependent deterioration of morphological degradation characteristics (such as edge sharpness and texture structure) in a printed feature area during the dynamic printing process, reflecting the rate of defect spread or aggravation. The specific acquisition method involves performing multi-scale morphological analysis on the real-time rotationally invariant feature map to extract feature vectors representing the attenuation of edge sharpness and the diffusion of microtexture. The rate of change of these feature vectors is calculated through time series tracking (e.g., five consecutive image frames). The rate of increase in disorder is quantified using information entropy theory (e.g., an entropy increment of 0.12 / s per unit time), and finally normalized to an entropy increase rate in the range of 0-1. A higher value for this parameter indicates accelerated defect deterioration and requires prioritization.

[0044] The defect morphology entropy increase rate is input into the inverse transformation module of the dynamic decoupling matrix, and the defect traceability characteristic scalar corresponding to the axial registration error component, radial pressure error component and circumferential torsion error component is generated through orthogonal projection decomposition; According to the ratio between the defect traceability characteristic scalar and the preset safety threshold, the three-dimensional probability distribution vector is mapped into a discrete defect type judgment signal, and the defect traceability characteristic scalar is converted into a signed parameter deviation vector. A tensor dot product operation is performed on the parameter deviation vector and the position compensation instruction output by the multi-axis collaborative optimizer to generate a parameter correction association matrix that integrates the defect traceability characteristics. The row vectors of this matrix correspond to the defect type judgment signal encoding, and the column vectors are associated with the dynamic reference instruction correction value of the servo drive system.

[0045] In one embodiment of the present invention, in a high-speed flexographic printing production line (working speed 200m / min), for the rotation-invariant feature map of the printing feature area collected in real time (size 512×512 pixels), multi-scale morphological gradient decomposition (scale factor σ=1.2, 2.0, 3.0) is first used to extract the edge sharpness attenuation (0.35) and micro-texture diffusion (0.28) as the morphological degradation feature vector; the vector is input into the pre-trained defect classification convolution kernel group (including three convolution kernels of 3×3, 5×5, and 7×7), and a three-dimensional probability distribution vector [such as ink layer diffusion 0.72, dot loss 0.15, and registration offset 0.13] is generated through a spatial attention weighting mechanism (weight coefficient 0.6); based on the maximum entropy increase rate direction (ink layer diffusion direction, entropy increase rate 0.12 / s) combined with the current phase advance compensation amount The time-domain derivative (0.05 mm / s²) is used to calculate a dynamic coupling coefficient of 0.85. The entropy increase rate of the defect morphology is input into the inverse transformation module of the dynamic decoupling matrix, decomposing the axial alignment error component of 0.08 mm, the radial pressure error component of 90 N, and the circumferential torsion error component of 0.25°. When these scalar values ​​exceed the preset safety thresholds (axial 0.05 mm, radial 80 N, circumferential 0.2°), an "ink layer diffusion defect" judgment signal and a signed parameter deviation vector [-0.03 mm, -10 N, -0.05°] are generated. Finally, a tensor dot product operation is performed on this vector with the position compensation instruction output by the previous multi-axis collaborative optimizer to generate a parameter correction association matrix (correction values: axial alignment -0.035 mm, radial pressure -8 N, circumferential torsion -0.03°), realizing closed-loop feedback control from defect characteristics to adjustment parameters.

[0046] In summary, the present invention achieves high-precision classification of defects such as ink layer diffusion and dot loss through multi-scale morphological analysis and spatial attention weighted mechanism. Combined with dynamic coupling coefficient calculation and defect tracing decomposition, a real-time correlation between defect morphological changes and servo adjustment parameters is established. Finally, through the parameter correction association matrix, the defect judgment results are directly mapped to the correction values ​​of each axis control instruction, forming a closed-loop control from defect identification to parameter adjustment, thereby improving the accuracy and automation of defect correction.

[0047] In this embodiment, the control method may further include the following steps: At the quantum transmitter of the distributed servo node, a polarization-orbital angular momentum doubly entangled photon pair is prepared to form a composite entangled channel across the group, where the local instruction register of each node is bound to a pair of entangled photons; Among them, distributed servo nodes refer to independent control units with quantum communication capabilities deployed in multiple printing groups. Each node establishes quantum connections with other nodes through entangled photon pairs to achieve collaborative control across groups.

[0048] The axial component of the position compensation instruction sequence is mapped into the polarization angle modulation parameter, and the circumferential component is converted into the orbital angular momentum topological charge phase. The two degrees of freedom are orthogonally encoded through the Pauli-X gate and the Hadamard gate to generate a quantum fingerprint state carrying multi-dimensional instructions. The multi-node Bell-based measurement device is triggered within the synchronization time window, causing the local and correlated entangled photons to undergo quantum interference in the polarization beam splitter and spiral phase plate, and the probability of the joint collapse of the two degrees of freedom is obtained through coincidence counting detection. Among them, the multi-node Bell-based measurement device is a synchronous detection system based on the principles of quantum optics. It performs joint Bell-state measurement on entangled photons of multiple nodes through polarization beam splitters and spiral phase plates to achieve instantaneous synchronization and verification of cross-unit instructions.

[0049] It should be noted that the specific steps for obtaining the joint collapse probability of two degrees of freedom through coincidence counting detection are as follows: First, the entangled photons, after interference through a polarization beam splitter and a spiral phase plate, are introduced into an array of single-photon detectors. Then, a time-to-digital converter (TDC) records the timestamps of each detector's response. When two correlated detectors trigger simultaneously within a preset time window (e.g., 5ns), it is recorded as a valid coincidence event. Next, the coincidence counts for different Bell basis combinations are counted, and the proportion of each combination in the total coincidence events is calculated. Finally, a joint probability integration is performed on the coincidence proportion of the polarization degree of freedom and the spiral phase distribution of the orbital angular momentum degree of freedom. After normalization, the joint collapse probability matrix of two degrees of freedom is output for subsequent quantum fingerprint feature extraction. This process is implemented in real time using a field-programmable gate array (FPGA), with latency controlled to the microsecond level.

[0050] According to the quantum fingerprint characteristics of the joint collapse probability, the polarization angle and phase are decoupled using the quantum Fourier inverse transform, and the original instruction components are restored after correction by the Pauli-Z gate to generate a synchronous instruction stream aligned in time and space; The fidelity between the decoupling instruction and the original quantum fingerprint state is calculated. When the fidelity is less than a preset threshold (the threshold is 0.99), the polarization modulator bias voltage and the orbital angular momentum generator pitch are dynamically adjusted according to the fidelity deviation until a compensation instruction without timing jitter is output to each servo drive interface.

[0051] It should be noted that the specific calculation steps for the fidelity of the decoupled instruction and the original quantum fingerprint state are as follows: First, the decoupled and restored instruction parameters (including polarization angle and topological charge phase) are re-encoded into quantum states and inner-producted with the original quantum fingerprint state in Hilbert space. Then, the density matrices of the two are reconstructed through quantum state tomography, and their trace distance is calculated as the fidelity benchmark value. Next, a projection operator based on the measurement basis is introduced to perform amplitude-phase decomposition on the two-degree-of-freedom joint collapse probability matrix to obtain the transmission characteristic correction factor of the actual quantum channel. Finally, the benchmark value is convolved with the correction factor and normalized to the range of 0-1 to output the final fidelity value F. When the F value falls below the threshold, the system automatically records the degree-of-freedom component (polarization or phase) with the largest deviation as the priority adjustment object for subsequent dynamic tuning.

[0052] For example, in a high-speed multi-unit flexographic printing system (e.g., 8 printing units working together at a line speed of 300 m / min), for the multi-axis compensation instructions output in the previous example (axial registration -0.05 mm, circumferential twist +0.3°): polarization-orbital angular momentum dual entangled photon pairs are generated between distributed nodes through periodically polarized lithium niobate waveguides (wavelength 1550 nm), and a pair of entangled photons is bound to the FPGA instruction register of each node (entanglement efficiency ≥ 85%); the axial compensation amount -0.05 mm is mapped to a polarization angle θ = π / 8 (corresponding to a voltage of 3.2 V), and the circumferential twist angle of +0.3° is converted to a topological charge phase φ = 0.6π, which is modulated by a Pauli-X gate (pulse width 5 ns) and a Hadamard gate (fidelity 99.2%) to generate a quantum fingerprint state (encoding dimension 4 ×4); a Bell Base measurement was triggered within a 50μs synchronization window. After the entangled photons interfered through a polarization beam splitter (extinction ratio 30dB) and a spiral phase plate (topological charge l=1), a joint collapse probability of 0.91 was measured. θ=π / 8±0.01 and φ=0.6π±0.02 were decoupled via inverse quantum Fourier transform (QFT point number 64). After correction by a Pauli-Z gate, the instructions (axial -0.049mm, circumferential +0.298°) were restored with a fidelity of F=0.987. When F<0.99, the polarization modulator bias voltage was adjusted to 3.25V (in 0.01V steps) and the spiral phase plate pitch to λ / 8 (originally λ / 10). Finally, the synchronization instructions (axial -0.050mm, circumferential +0.300°) were output to the servo drive with a timing jitter of ≤100ns.

[0053] It should be noted that in practical applications, to address the technical challenges of asynchronous compensation instructions and accumulated timing jitter caused by communication delays between distributed printing units, this embodiment achieves ultra-high-speed parallel transmission of multi-axis instructions through a two-dimensional quantum entanglement channel and quantum fingerprint state encoding. Quantum interference and fidelity feedback mechanisms ensure instantaneous, precise synchronization and anti-interference capabilities of cross-unit instructions, completely eliminating the timing deviations of traditional bus communications. Ultimately, this achieves absolute spatiotemporal consistency in the coordinated control of multiple printing units, improving the overall registration accuracy and dynamic stability of large-scale printing systems.

[0054] In this embodiment, the control method may further include the following steps: Projecting near-infrared-visible multispectral bands (wavelength range 450-1550nm) onto the printed feature area, collecting a sequence of sub-surface scattering interference images, and generating a raw interference atlas containing the refractive index gradient of the ink-substrate interface; The original interference pattern set is subjected to time-space phase unwrapping processing and the ink penetration depth distribution is inverted through finite element analysis to obtain the ink penetration profile gradient, curing phase transition stress field and dynamic viscoelastic response spectrum, generating a three-dimensional material property tensor. It should be noted that the collected multispectral interference image sequence is first subjected to time-space phase unwrapping processing to eliminate phase jumps and reconstruct a continuous phase distribution map; then, a finite element model is established based on the substrate material parameters (such as fiber orientation and porosity), and the phase distribution data is input as boundary conditions. The three-dimensional distribution of the ink penetration depth in the substrate is inverted through iterative calculation; then, the gradient change of the penetration profile is extracted through differential operation, and the phase change stress field distribution is calculated in combination with the temperature-time curve during the curing process; at the same time, based on the time domain response characteristics of interference signals of different wavelengths, the dynamic viscoelastic response spectrum is analyzed and obtained; finally, these parameters (penetration gradient, stress field, response spectrum) are integrated into a three-dimensional material property tensor containing spatial distribution and mechanical properties, where each dimensional data corresponds to the penetration depth (XY plane distribution), curing stress (Z-direction gradient) and viscoelasticity (frequency response) characteristics.

[0055] Perform eigenvalue decomposition on the three-dimensional material property tensor and extract the main penetration direction vector corresponding to the maximum eigenvalue (representing the maximum penetration trend of ink along the substrate fibers); The main penetration direction vector is projected onto the printing pressure action plane (XY plane), and the cosine value of the angle between the projection vector and the radial pressure axis direction is calculated. Combined with the mean variance of the local elastic modulus dimension, the dynamic compensation coefficient of the elastic modulus is generated by orthogonal polynomial fitting; The dynamic compensation coefficient of the elastic modulus is input into the elastic modulus weight channel of the dynamic decoupling matrix to replace the original static parameter. At the same time, the damping proportional coefficient threshold is adjusted according to the curing shrinkage dimensional data, and the corrected real-time decoupling matrix is ​​output for subsequent error component decoupling operations.

[0056] It should be noted that to address the technical issues of insufficient compensation accuracy caused by the inability to perceive ink penetration status and changes in material properties in real time in traditional printing control, this embodiment achieves three-dimensional dynamic monitoring of sub-surface properties such as ink penetration depth and curing stress through multispectral interferometric imaging and finite element inversion. Combining main penetration direction analysis with dynamic compensation of elastic modulus, the real-time material properties are integrated into the decoupling control algorithm, enabling the system to adapt to changes in the ink-substrate interface. Ultimately, by dynamically updating the decoupling matrix parameters, the accuracy of pressure compensation and material adaptability under high-speed printing conditions are improved, effectively suppressing ink layer defects caused by uneven penetration.

[0057] like Figure 3 As shown, the second aspect of the present invention discloses a flexographic printing press control system 8 based on machine vision, wherein the flexographic printing press control system includes a memory 60 and a processor 80, wherein the memory 60 stores a flexographic printing press control method program, and when the flexographic printing press control method program is executed by the processor 80, the steps of any one of the flexographic printing press control methods are implemented.

[0058] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0059] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0060] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0061] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0062] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods of the various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0063] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A flexographic printing press control method based on machine vision, characterized in that: The following steps are involved: Collecting surface image data of a printed product in motion in real time, separating and identifying a preset printing feature area from the image data, and obtaining geometric feature parameters of the area relative to a reference position; generating a multi-axis coordinated position compensation instruction for a printing unit according to a deviation between the geometric feature parameter and a preset target feature; Convert the position compensation instruction into the adjustment parameter of the servo drive system in real time, and dynamically adjust the printing roller pressure and the registration mechanism displacement of the printing unit; Based on the morphological changes of the printing feature area in the updated image data, a defect type determination signal is synchronously output and an associated adjustment parameter correction value is associated.

2. The method for controlling a flexographic printing press based on machine vision according to claim 1, characterized in that: Separate and identify a preset printing feature area from the image data, and obtain geometric feature parameters of the area relative to a reference position, specifically: Perform motion trajectory compensation filtering on the collected continuous frame surface images to generate a feature-enhanced image with de-motion blur; In the feature-enhanced image, the printed mark area with micro-texture features is separated as the texture fingerprint through the preset chroma key threshold and morphological gradient detection; Performing a local Fourier descriptor transform on the texture fingerprint to generate a rotation-invariant feature map including spatial frequency distribution characteristics; Performing a cross-correlation operation on the phase of the rotation-invariant feature map and the reference template to obtain a sub-pixel offset vector of the texture fingerprint center point relative to a preset reference grid; Based on the ratio of the offset vector components in the printing travel direction and the axial direction, the stretching distortion rate and torsion angle parameters of the printed feature area are calculated.

3. The method for controlling a flexographic printing press based on machine vision according to claim 1, characterized in that: According to the deviation between the geometric feature parameters and the preset target features, a multi-axis coordinated position compensation instruction of the printing unit is generated, specifically: The detected sub-pixel offset vector, stretch distortion rate and torsion angle parameters are input into the pre-trained error space mapping model and decomposed into three independent degrees of freedom: axial registration error component, radial pressure error component and circumferential torsion error component through nonlinear transformation. A dynamic decoupling matrix is ​​constructed using real-time updated material property coefficients to perform orthogonal decoupling operations on the three-degree-of-freedom error components. The nonlinear coupling strength factors between the degrees of freedom are simultaneously calculated to generate independent error corrections with coupling compensation. According to the real-time linear speed of the printing unit and the response delay parameters of the servo system, a phase advance correction factor is superimposed on the independent error correction value, and a dynamic reference instruction with time domain advance characteristics is generated through a pre-compensation algorithm; The dynamic reference instructions are input into the multi-axis collaborative optimizer based on the mechanical configuration constraints of the printing unit. A fuzzy weight allocation strategy is used to coordinate the displacement instruction weights of the printing roller pressure axis, axial registration axis and circumferential torsion compensation axis, and the optimal position compensation instruction sequence that meets the mechanical interference boundary conditions is output.

4. The method for controlling a flexographic printing press based on machine vision according to claim 3, characterized in that: According to the real-time linear speed of the printing unit and the response delay parameters of the servo system, a phase advance correction factor is superimposed on the independent error correction value, and a dynamic reference instruction with time domain advance characteristics is generated through the pre-compensation algorithm. Specifically: Based on the step response test data of the servo system, the response delay parameter is decomposed into a phase lag component and an amplitude attenuation component, and a discretized transfer function model positively correlated with the linear velocity is established. The independent error correction amount is converted into an amplitude-phase vector representation. When it is detected that the phase lag component exceeds a preset threshold, the vector is subjected to a complex domain phase compensation operation with a dynamically adjusted lead correction factor to generate a frequency domain compensation vector with a phase lead characteristic. Perform inverse transformation on the frequency domain compensation vector, extract its real part as the time domain amplitude envelope signal, and retain the phase advance compensation amount constructed by the imaginary part; When the real-time linear velocity change rate exceeds the set threshold, the gain coefficient is dynamically adjusted according to the product of the amplitude attenuation component and the phase advance compensation amount, and a time domain reference curve for anti-saturation limiting is generated through a sliding integrator; The time domain amplitude envelope signal and the phase advance compensation amount are fused in the hybrid domain to output a dynamic reference instruction that compensates for both amplitude attenuation and phase lag.

5. The method for controlling a flexographic printing press based on machine vision according to claim 1, characterized in that: The position compensation instruction is converted into the adjustment parameters of the servo drive system in real time, and the printing roller pressure and the registration mechanism displacement of the printing unit are dynamically adjusted. Specifically: receiving an optimal position compensation instruction sequence outputted by a multi-axis collaborative optimizer, wherein the optimal position compensation instruction sequence comprises three independent components: a printing roller pressure axis displacement instruction, an axial registration axis displacement instruction, and a circumferential torsion compensation axis displacement instruction; For each displacement command component, combined with the phase lag component and amplitude attenuation component pre-stored in the servo system, an initial parameterization conversion is performed through the dynamic gain coefficient mapping table to generate the corresponding axis-specific drive parameters; The axis-specific drive parameters and the phase advance compensation amount are hybrid-domain fused in the complex domain to generate a compensated drive vector with time-domain advance characteristics. Applying a sliding integrator to the compensated driving vector to perform anti-saturation limiting processing and output a smooth time-domain amplitude envelope signal; Based on the time domain amplitude envelope signal, the pulse frequency modulation duty cycle parameter and the axial displacement step parameter are calculated in real time through the servo drive interface module; The output torque of the printing roller pressure servo motor is dynamically adjusted by utilizing the pulse frequency modulation duty cycle parameter, and the linear displacement of the registration mechanism is precisely controlled by the axial displacement step parameter, thereby realizing coordinated dynamic adjustment of the printing roller pressure and the registration mechanism displacement.

6. The method for controlling a flexographic printing press based on machine vision according to claim 1, characterized in that: Based on the morphological changes of the printed feature area in the updated image data, a defect type determination signal is synchronously output and an associated adjustment parameter correction value is associated, specifically: Multi-scale morphological gradient decomposition is performed on the updated rotation-invariant feature map collected in real time, and the edge sharpness attenuation and micro-texture diffusion of the printed feature area are extracted as morphological degradation feature vectors. Inputting the morphological degradation feature vector into a pre-trained defect classification convolution kernel group, and generating a three-dimensional probability distribution vector including the confidence of ink layer diffusion defect, dot loss defect and registration shift defect through a spatial attention weighting mechanism; Based on the maximum entropy increase rate direction of each component in the three-dimensional probability distribution vector and the time-domain derivative of the current phase advance compensation, the dynamic coupling coefficient between the entropy increase rate of the defect morphology and the servo adjustment response is calculated. The defect morphology entropy increase rate is input into the inverse transformation module of the dynamic decoupling matrix, and the defect traceability characteristic scalar corresponding to the axial registration error component, radial pressure error component and circumferential torsion error component is generated through orthogonal projection decomposition; According to the ratio between the defect traceability characteristic scalar and the preset safety threshold, the three-dimensional probability distribution vector is mapped into a discrete defect type judgment signal, and the defect traceability characteristic scalar is converted into a signed parameter deviation vector. A tensor dot product operation is performed on the parameter deviation vector and the position compensation instruction output by the multi-axis collaborative optimizer to generate a parameter correction association matrix that integrates the defect traceability characteristics. The row vectors of this matrix correspond to the defect type judgment signal encoding, and the column vectors are associated with the dynamic reference instruction correction value of the servo drive system.

7. The method for controlling a flexographic printing press based on machine vision according to claim 3, wherein: The material property coefficients include the elastic modulus of the substrate and the ink viscosity threshold.

8. The method for controlling a flexographic printing press based on machine vision according to claim 4, wherein: The step response test data includes the response delay time, rise time, overshoot and steady-state establishment time of the servo system from command input to steady-state output establishment.

9. A flexographic printing press control system based on machine vision, characterized in that: The flexographic printing press control system includes a memory and a processor, wherein a flexographic printing press control method program is stored in the memory. When the flexographic printing press control method program is executed by the processor, the steps of the flexographic printing press control method as described in any one of claims 1 to 8 are implemented.

Images