A parameter self-optimization method for intelligent packaging box hollowing process

Abstract

The present application belongs to the field of automation control technology, and relates to a parameter self-optimization method of intelligent packaging box hollowing process. The present application solves the problem that the existing packaging box hollowing process relies on static parameters or simple feedback adjustment, and is difficult to cope with the problems caused by material anisotropy and complex trajectory. The present application obtains the packaging box surface texture image and the hollowing design vector path, extracts the fiber arrangement and geometric curvature features to generate the processing resistance prediction field, calculates the energy compensation gain and generates the dynamic process control parameter flow. The present application collects the edge heat affected zone width and the notch flatness parameters in real time, generates the process deviation vector by comparing the standard quality threshold, and uses the PID control algorithm to close loop correct the dynamic process parameters to ensure the processing quality. Through accumulating the processing data to improve the process knowledge base, the quality consistency and self-adaptive control level of the hollowing operation of different batches of packaging materials are improved, and the production efficiency and product quality are enhanced.

Description

Technical Field

[0001] This invention belongs to the field of automation control technology and relates to a parameter self-optimization method for intelligent packaging box hollowing process. Background Technology

[0002] Cutout technology for packaging boxes is a key step in modern packaging manufacturing to enhance the aesthetic appeal and structural design flexibility of products. It typically relies on CNC laser die-cutting machines or mechanical cutters to make penetrating cuts into packaging materials such as cardboard along a pre-set vector path. During this process, the control system needs to adjust process parameters such as laser power, feed rate, or cutter pressure in real time to adapt to changes in material properties and the processing trajectory. Adaptive control technology, as a core branch of automation, aims to enable the system to automatically adjust its control strategy based on dynamic changes in the external environment and internal state, thereby maintaining the expected operational performance.

[0003] Existing packaging box cutout processing equipment mostly adopts an open-loop control scheme based on a static parameter library. Before processing, operators empirically set globally uniform operating parameters according to the material type and thickness. Some advanced equipment is equipped with a basic feedback adjustment module, which judges the current load state by monitoring the current fluctuation of the spindle motor or the reflected light intensity of the laser generator, and triggers a conventional proportional adjustment mechanism when the load exceeds a preset threshold to increase or decrease the current output power or reduce the operating speed.

[0004] This control mode, which relies on feedback from global static parameters or specific physical quantities, struggles to cope with the anisotropy of fiber distribution within packaging materials and the abrupt changes in nonlinear resistance caused by complex geometric trajectories. Due to the lack of feedforward prediction of the coupling effect between material microtexture and path curvature, the control system often exhibits lag in response when encountering high-density fiber clusters or sharp bends, easily leading to localized carbonization and scorching of cuts or rough edges. Furthermore, existing feedback mechanisms do not incorporate final quality indicators such as the width of the heat-affected zone and the micromorphology of the cut into the closed-loop control circuit, and lack self-learning capabilities based on historical processing data. This results in the equipment being unable to automatically evolve the optimal combination of control parameters when faced with multiple batches of processing tasks with material fluctuations. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a parameter self-optimization method for intelligent packaging box hollowing process is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: This invention provides a parameter self-optimization method for intelligent packaging box cutout process, including: S1, obtaining the surface texture image of the packaging box to be processed and the cutout design vector path.

[0007] S2. Extract fiber arrangement features from the surface texture image and geometric curvature features from the vector path of the hollow design, and fuse the fiber arrangement features and geometric curvature features to generate a processing resistance prediction field.

[0008] S3. Calculate the energy compensation gain at different coordinate points based on the processing resistance prediction field.

[0009] S4. Map the initial processing parameters of the actuator according to the energy compensation gain to generate a dynamic process control parameter stream.

[0010] S5. Real-time acquisition of parameters such as the width of the heat-affected zone at the edge of the work area and the flatness of the cut.

[0011] S6. Generate a process deviation vector by comparing the edge heat-affected zone width, cut flatness parameters and standard quality threshold.

[0012] S7. Use the process deviation vector to perform closed-loop correction of the dynamic process control parameter flow.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention generates a processing resistance prediction field by fusing the fiber arrangement features in the surface texture image with the geometric curvature features in the hollow design vector path, which changes the control logic of traditional equipment that only relies on fixed parameters for operation. This mechanism can predict the coupling effect of local physical properties of the material and trajectory complexity before the processing action occurs, and plan the energy compensation gain in advance, thereby realizing precise feedforward adjustment of control commands in complex texture areas or sharp turning paths.

[0014] (2) This invention constructs a basic control model that includes the rated response frequency and the maximum load limit, and converts the energy compensation gain into instantaneous control commands that match the hardware characteristics. This mapping process ensures that the generated dynamic process control parameter flow can meet the energy requirements of the processing resistance fluctuations, and is strictly controlled within the safety boundary of the actuator, avoiding equipment overload shutdown or high-frequency oscillation caused by parameter mutations.

[0015] (3) This invention introduces a multidimensional quality feedback dataset and a process deviation vector, and combines it with a proportional-integral-derivative control algorithm to realize the closed-loop correction of the dynamic process control parameter flow, and uses parameter evolution samples to update the calculated weights. This self-optimization mechanism enables the control system to adaptively adjust parameters according to real-time processing quality, and continuously improves the process knowledge base as processing tasks accumulate, thereby improving the quality consistency of different batches of packaging material hollowing operations and the system's adaptive control level. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

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

[0019] Please see Figure 1 The present invention provides a parameter self-optimization method for intelligent packaging box cutout process, including: S1, obtaining the surface texture image of the packaging box to be processed and the cutout design vector path.

[0020] In a specific embodiment of the present invention, obtaining the surface texture image and the cutout design vector path of the packaging box to be processed includes: using a visual sensing component to scan the surface of the packaging material to obtain the original reflectivity matrix, and performing grayscale gradient processing on the original reflectivity matrix to obtain the surface texture image.

[0021] The coordinate sequence is extracted from the CAD design document of the packaging box, and the coordinate sequence is continuously interpolated to obtain the vector path of the hollow design.

[0022] Align the surface texture image with the vector path of the cutout design in spatial coordinates to generate a processing base map with a textured background.

[0023] Specifically, the visual sensing component acquires photoelectric signals from the packaging material surface using a high-resolution linear scan sensor, converting the acquired light intensity information into a digital representation to form an original reflectivity matrix. Each element in the original reflectivity matrix represents the light reflection intensity at a specific coordinate point on the packaging material surface, reflecting the density of the fiber distribution. To extract the surface texture image, the original reflectivity matrix undergoes grayscale gradient processing, and the fiber edge features are enhanced by calculating the amplitude change rate between adjacent pixels. The generation of the surface texture image follows the formula:

[0024]

[0025] In the formula, Represents the surface texture image in coordinates grayscale gradient value at that location Represents the original reflectivity matrix in coordinates The reflectance value at that location. These represent the light reflectance values ​​of the positions directly to the right and left of the central pixel in the original reflectance matrix, respectively. These represent the light reflectance values ​​of the adjacent positions directly below and above the central pixel, respectively. This calculation process is based on measured data from 500 industrial cardboard samples, ensuring that the gradient operator accurately captures the contours of plant fibers with diameters between 0.02 mm and 0.05 mm.

[0026] While acquiring the surface texture image, the system reads the CAD design document of the packaging box and extracts the coordinate sequence defining the hollow shape. This coordinate sequence, composed of a series of discrete geometric vertices, cannot be directly used for high-precision laser processing control. Therefore, continuous interpolation is performed on the coordinate sequence, and a smooth curve equation is constructed between the discrete points using a cubic spline interpolation algorithm, thus obtaining the hollow design vector path. The mathematical expression for the hollow design vector path is as follows:

[0027]

[0028] In the formula, Indicates the vector path of the hollow design in the parameters The spatial position vector below, Represents the first position in the coordinate sequence The coordinate vector of each control point express order basis functions, This represents the total number of control points. Using this formula, a discrete coordinate sequence is transformed into a continuously differentiable vector trajectory.

[0029] In one specific embodiment of the present invention, in order to ensure that the generated hollow design vector path has a continuous second derivative throughout the entire field in subsequent feature extraction, the cubic spline interpolation algorithm sets... Order of basis functions ,and follow Recurrence formula:

[0030] when At that time, if ,but ,otherwise ;

[0031] when When, its mathematical recurrence relation is:

[0032] ;

[0033] In the formula, These are parameter nodes in a predefined set of node vectors. This recursive calculation eliminates the curvature discontinuity at the corners of the original CAD polygonal lines, transforming discrete coordinate sequences into continuous and differentiable vector trajectories. This provides stable and oscillatory mathematical input for subsequent high-precision spatial second derivative calculations.

[0034] It's important to note that the key prerequisite for achieving 'spatial coordinate alignment' lies in establishing benchmark anchor points between the physical and digital worlds. In practice, while scanning the surface texture of the packaging material, the vision sensing component simultaneously executes edge detection or Mark point recognition algorithms to extract the pixel coordinates of the four physical right-angle boundaries of the physical packaging cardboard or the pre-printed cross-shaped alignment Mark points. When the system reads the CAD design document, it simultaneously extracts the virtual coordinates of the corresponding design boundaries or digital Mark points. Subsequently, using these at least three pairs of paired physical-virtual coordinate points as a benchmark, the system employs the least squares method to solve for a 3x3 affine transformation matrix that includes translation, rotation, and scaling factors. This matrix not only eliminates linear distortion during the linear camera scanning process but also accurately projects and anchors the abstract vector paths from the CAD onto the surface texture image with real physical dimensions, ensuring a one-to-one correspondence between the subsequently predicted resistance points and the actual tool cutting points in physical space.

[0035] After completing the above data preparation, the surface texture image and the cutout design vector path are aligned in spatial coordinates. This process involves establishing a mapping relationship from the design coordinate system to the physical image coordinate system, and eliminating linear distortions generated by the visual sensing components during scanning through an affine transformation matrix. Finally, the cutout design vector path is superimposed on the surface texture image to generate a processing base map with a textured background. The coordinate mapping relationship of the processing base map with a textured background is as follows:

[0036]

[0037] In the formula, This represents the original physical coordinates in the vector path of the cutout design. This represents the pixel coordinates corresponding to the processed base map with a textured background. This represents a 3x3 spatial transformation matrix. (Transformation matrix) The parameter settings are determined based on the installation height of the visual sensing component and the focal length of the lens through alignment experiments with 20 sets of standard calibration plates.

[0038] For example, consider the perforation process of a 350g white cardboard packaging box from a certain brand. The visual sensing component uses an 8K resolution line scan camera to scan the surface of the packaging material under uniform strip light illumination, outputting a raw reflectance matrix of 4096 x 8192 dimensions. After grayscale gradient processing, the originally blurred fiber interweaving state in the raw reflectance matrix is ​​transformed into a surface texture image with obvious contrast, where the grayscale gradient values ​​of the fiber edges are distributed between 120 and 255. The system simultaneously reads the CAD design document and extracts the coordinate sequence containing 150 key vertices. Through continuous interpolation processing, these 150 vertices are expanded into a perforation design vector path containing 12,000 sampling points, with smooth and continuous changes in path curvature. A pre-calibrated spatial transformation matrix is ​​then used. The vector path of the hollow design is aligned to the surface texture image with an accuracy of 0.01 mm. The resulting processing base map with textured background clearly shows the specific location of the hollow lines cutting the fibers, providing accurate basic data for subsequent processing resistance prediction.

[0039] S2. Extract fiber arrangement features from the surface texture image and geometric curvature features from the vector path of the hollow design, and fuse the fiber arrangement features and geometric curvature features to generate a processing resistance prediction field.

[0040] In a specific embodiment of the present invention, fiber arrangement features and geometric curvature features in the vector path of the hollow design are extracted from the surface texture image, and the fiber arrangement features and geometric curvature features are fused to generate a processing resistance prediction field, including: performing a Fourier transform on the surface texture image to obtain a frequency domain distribution map, and identifying the dominant frequency direction and amplitude intensity in the frequency domain distribution map to obtain fiber arrangement features.

[0041] The local curvature is obtained by calculating the second derivative of the vector path of the hollow design at each sampling point, and the geometric curvature characteristics are obtained based on the rate of change of the local curvature with the path length.

[0042] A resistance correlation matrix is ​​established, and the fiber arrangement characteristics are used as row vectors and the geometric curvature characteristics are used as column vectors to perform Kronecker product operation to generate a processing resistance prediction field.

[0043] Specifically, the system transforms the surface texture image from the spatial domain to the frequency domain using mathematical transformations. It then utilizes Fourier transform, a technique that reveals the periodic structure of the image by analyzing the frequency components of the signal, to generate a frequency domain distribution map. In this map, regions of concentrated energy reflect the orientation and distribution density of the packaging material's fibers. By searching for energy extrema in the frequency domain distribution map, the direction with the highest energy distribution is identified as the dominant frequency direction, and the energy amplitude in this direction is extracted as the amplitude intensity, thus constituting the fiber arrangement characteristics. The extraction process of the fiber arrangement characteristics follows the formula below:

[0044]

[0045] In the formula, This represents the fiber arrangement feature vector. The frequency domain distribution plot is shown on the frequency coordinates. Complex values ​​at that location, This represents the frequency coordinate point with the highest energy value. The first component of this eigenvector represents the principal axis angle of the fiber, and the second component represents the concentration of fiber distribution. It is set based on the microstructure analysis of 300 sets of paperboards with different basis weights to ensure that the anisotropy of the material can be quantified.

[0046] In a specific embodiment of the present invention, the fiber arrangement formed in the papermaking process of the packaging paperboard exhibits physical anisotropy. The first time component, the principal axis angle, physically defines the orientation of the normal axis of the main fibers within the paper, enabling the system to perceive in real time whether the current processing trajectory is predicted to have low mechanical resistance by cutting along the fiber gaps or by cutting across the fibers, which is predicted to have high mechanical resistance. The second component, the concentration degree, effectively quantifies the severity of anisotropy by calculating the ratio of the energy at the extreme frequency point to the total energy in the frequency domain. If this ratio approaches the extreme value, it indicates that the fibers are arranged in a highly consistent unidirectional manner, and the system will predict a sudden and significant resistance step when the trajectory changes direction and reserve high compensation energy in advance. If the ratio is small, it indicates that the material is close to isotropy, and the system will reduce the sensitivity weight of the trajectory direction. Through the synergy of these two components, the previously invisible underlying mechanical properties of the material are accurately transformed into a controllable mathematical mapping, forming the basis for accurately calculating dynamic processing resistance.

[0047] When processing the vector path of the hollow design, the system samples at equal intervals along the path trajectory and calculates the second derivative of each sampling point with respect to the path length, thus obtaining the local curvature characterizing the degree of path bending. To further describe the severity of tool or laser beam deflection during processing, the system calculates the rate of change of local curvature with path length, defining it as the geometric curvature feature. The formula for calculating the geometric curvature feature is as follows:

[0048]

[0049] In the formula, Indicates path length Geometric curvature characteristics at the location, Represents the processing arc length variable The spatial differential operator for finding the first derivative directly quantifies the degree of abrupt change in local curvature as processing progresses. and These represent the lengths of the vector paths in the cutout design. The first and second derivatives. This feature reflects the dynamic geometric changes of the machining trajectory and is of great significance for predicting force fluctuations at corners.

[0050] To integrate the influence of material properties and processing trajectory, the system establishes a resistance correlation matrix, using fiber arrangement characteristics as row vectors and geometric curvature characteristics as column vectors. By performing the Kronecker product on these two vectors—that is, multiplying each element of the row vector with each column vector and arranging the results—a processing resistance prediction field is generated. The formula for generating the processing resistance prediction field is as follows:

[0051]

[0052] In the formula, This represents the prediction field matrix for processing resistance. This represents the Kronecker product operator. This represents the drag conversion factor. (Drag conversion factor) It is a dimensionless correction constant used to convert dimensionless geometric and frequency characteristics into numerical values ​​that characterize processing resistance. Its value is calibrated to 0.85 based on 150 sets of laser cutting resistance test experiments, ensuring a linear correlation between the calculation results and the actual physical resistance.

[0053] For example, consider processing corrugated cardboard with a distinct longitudinal fiber arrangement. After performing a Fourier transform on the surface texture image, the system identifies the dominant frequency direction as 90 degrees and the amplitude intensity as 0.75 in the frequency domain distribution map, forming a fiber arrangement feature vector. When processing a circular arc cutout design vector path with a radius of 5 mm, the local curvature at the sampling point remains constant. However, when the path enters the junction of the straight line and the arc, the rate of change of local curvature with path length, i.e., the geometric curvature feature, surges to 1.25. The system uses the fiber arrangement feature vector as a row vector and performs a Kronecker product operation with the column vector formed by the geometric curvature feature, multiplying it by a resistance conversion coefficient of 0.85. Finally, a machining resistance prediction field is generated in the corner coordinate region of the machining base map. In this prediction field, the predicted resistance value at a specific coordinate point is higher than that in the straight line segment region, accurately predicting the peak machining resistance caused by the superposition of fiber blockage and trajectory deflection when the tool or laser passes through that position.

[0054] S3. Calculate the energy compensation gain at different coordinate points based on the processing resistance prediction field.

[0055] In a specific embodiment of the present invention, the energy compensation gain at different coordinate points is calculated based on the processing resistance prediction field, including: retrieving the resistance value at each coordinate point in the processing resistance prediction field, and calculating the degree of deviation between the resistance value and the average resistance benchmark to obtain the resistance fluctuation coefficient.

[0056] The energy compensation gain is obtained by calculating the correction amount for laser power or tool pressure based on the energy adjustment function matched with the resistance fluctuation coefficient.

[0057] The energy compensation gain is smoothed by filtering to eliminate oscillations caused by abrupt changes in gain between adjacent coordinate points, thus generating a smooth gain sequence.

[0058] Specifically, the system retrieves the resistance values ​​at each coordinate point in the processing resistance prediction field. To quantify the differences in energy demand at different locations, a unified reference benchmark needs to be established. The system statistically analyzes the stress conditions of a large number of similar packaging materials at standard processing speeds, sets an average resistance benchmark, and calculates the deviation between the resistance values ​​in the processing resistance prediction field and the average resistance benchmark, thereby obtaining the resistance fluctuation coefficient. The calculation process for the resistance fluctuation coefficient follows the formula below:

[0059]

[0060] In the formula, Indicates coordinates Resistance fluctuation coefficient at the location, The processing resistance prediction field is represented in coordinates. The resistance value at that location, This represents the average resistance benchmark. The value of the average resistance benchmark is determined based on the average value measured by 400 sets of industrial sensors under standard cutting conditions, ensuring that the resistance fluctuation coefficient can accurately reflect the combined effects of local material strength and machining trajectory complexity.

[0061] In one specific embodiment of the present invention, in order to obtain the average resistance reference The calibration and data collection steps specifically include: establishing a standard cutting environment with a constant temperature of 25±2℃ and a relative humidity of 50%±5%; selecting 10 standard substrates including white cardboard, kraft paper, and corrugated paper, and combining them with 8 curvature radius ranges. to Based on the typical cutting vector path, 400 independent orthogonal test models (10×8×5) were constructed under 5 different rated speed configurations. The system drove the actuator equipped with a beam power attenuation meter or a high-frequency piezoelectric force gauge to sequentially cut and read the 400 models; the acquired raw sensor signals were processed to remove the interference from start-stop oscillations. In the data segment, the remaining 80% of the steady-state output signal is denoised by passing it through a low-pass filter with a cutoff frequency of 500Hz. Finally, the time-domain integral average of these 400 sets of effective steady-state signals is calculated, and the arithmetic mean of the output is fixed as the average resistance benchmark of the system (e.g., 1.20 Newtons). The measured acquisition process provides an objective and universal physical calibration reference for predicting system fluctuations.

[0062] After obtaining the resistance fluctuation coefficient, the system matches the energy adjustment function to the resistance fluctuation coefficient and calculates the correction amount for laser power or tool pressure, thereby obtaining the energy compensation gain. The energy adjustment function is a mathematical model that maps physical resistance deviation to the increment of actuator energy output, aiming to offset changes in machining resistance by dynamically adjusting the energy input. The formula for calculating the energy compensation gain is as follows:

[0063]

[0064] In the formula, Indicates energy compensation gain. Represents the energy mapping sensitivity coefficient. The sign function is used to determine the positive and negative directions of the gain. Energy mapping sensitivity coefficient. The setting is based on 250 sets of laser power response experiments. By observing the changes in cut quality under different power increments, the value range was determined to be between 0.15 and 0.35, so as to ensure that the energy compensation can overcome the resistance without causing the material to overheat and burn.

[0065] Considering the potential for drastic jumps in the spatial distribution of the predicted processing resistance field within a very small range, directly applying the energy compensation gain to the actuator could cause high-frequency oscillations in the laser output power or servo motor pressure, affecting mechanical lifespan and processing smoothness. Therefore, the system performs smoothing filtering on the energy compensation gain, employing a method that mitigates numerical abrupt changes by calculating a weighted average of local regions, eliminating oscillations caused by gain jumps between adjacent coordinate points, and ultimately generating a smoothed gain sequence. The formula for generating the smoothed gain sequence is as follows:

[0066]

[0067] In the formula, Represents the first in the smoothed gain sequence The value of each sampling point Indicates the first Within the neighborhood centered at point , the first Energy compensation gain at each point, Indicates the radius of the filtering window. This represents the smoothing factor. The settings are based on the rated response frequency of the actuator and are determined through 100 sets of dynamic response tests to ensure that the generated smooth gain sequence conforms to the physical adjustment characteristics of the hardware in the time domain.

[0068] It is particularly important to note that, in order to accurately map the actuator's 'rated response frequency' in the time domain to the machining base map in the spatial domain, the system introduces 'real-time feed rate' as a bridge for spatiotemporal transformation. Specifically, the system multiplies the reciprocal of the actuator's shortest response time period (rated response frequency) by the current real-time feed rate to calculate the physical spatial distance the tool travels on the material surface within this response limit. This distance is then divided by the resolution of the spatial coordinate points to obtain the 'limited step-by-step pixels' in the spatial domain. The 'radius of the filtering window' in the aforementioned smoothing filtering process is strictly limited by this limit of step-by-step pixels. By introducing feed rate to achieve dynamic decoupling and transformation between the spatiotemporal domains, it ensures that the smooth gain gradient in the spatial coordinates will never exceed the actuator's physical work limit on the time axis, completely closing the loop of the underlying logic of feedforward prediction.

[0069] In a specific embodiment of the present invention, the smoothing factor The determination process and mechanism of the calibration are not based on preset constants, but rather on deduction from the actual physical time constant of the actuator and establishment through closed-loop testing. Specifically, the system first captures the rated response bandwidth of the hardware server or laser power supply and converts it into the device's limit command throughput slope based on the first-order inertial characteristics; then, in offline control calibration, a set of 100 step fluctuation test sets covering the full range of resistance jumps from 10% to 100% is constructed, and the smoothing factor is adjusted through the test program. Perform a step-by-step decreasing scan, simultaneously monitoring the overshoot of the actuator's output waveform. When real-time recording indicates a certain... When the value triggers a high-frequency rebound oscillation exceeding the 2% safety tolerance, the system automatically extracts the safety boundary parameters before the oscillation occurs. Based on this, it integrates the extreme value envelopes of 100 operating conditions to finally determine the global smoothing factor calibration value of the system. This closed-loop calibration process with rigid index constraints mathematically filters out high-frequency steep components that exceed the limits, strictly ensuring that the rate of change of the generated smooth gain sequence in the time domain completely converges within the true undamped throughput envelope of the actuator, thereby eliminating mechanical resonance and power howling phenomena induced by abrupt changes in control compensation.

[0070] For example, consider processing a special packaging paper with uneven fiber clusters on its surface. The system retrieves a resistance value of 1.85 Newtons from the processing resistance prediction field for a high-density fiber region, while the preset average resistance benchmark is 1.20 Newtons. The resistance fluctuation coefficient at this location is calculated to be 0.54. Subsequently, the system calls the energy adjustment function, and with the energy mapping sensitivity coefficient set to 0.25, calculates the energy compensation gain at this coordinate point to be 0.108. Because the resistance change at the edge of this high-density region is relatively steep, the original energy compensation gain exhibits a step-like distribution in space. The system then applies a smoothing filter, setting the filter window radius to 5 sampling points and the smoothing factor to 1.2, transforming the originally sharp gain jump into a numerical curve with a gentle slope, generating a smooth gain sequence. This sequence ensures that the laser power can smoothly increase from the base power to the compensation power when passing through the fiber clusters, avoiding carbonization of the cut edge caused by a sudden surge in power.

[0071] S4. Map the initial processing parameters of the actuator according to the energy compensation gain to generate a dynamic process control parameter stream.

[0072] In a specific embodiment of the present invention, the initial processing parameters of the actuator are mapped according to the energy compensation gain to generate a dynamic process control parameter stream, including: obtaining the rated response frequency and maximum load limit of the actuator, and constructing a basic control model in combination with the initial processing parameters.

[0073] The energy compensation gain is injected into the basic control model, and the laser duty cycle or servo motor speed at each time step is calculated to obtain the instantaneous control command.

[0074] The instantaneous control commands are arranged according to the processing sequence to form a dynamic process control parameter flow that is synchronized with the vector path of the hollowing design.

[0075] Specifically, the system reads the hardware parameter configuration manual of the actuator, extracts the highest frequency at which the device can stably respond to control signals as the rated response frequency, and obtains the maximum power or torque threshold allowed for output within the safe operating range as the maximum load limit. The system extracts pre-set basic operating values ​​as initial processing parameters and combines them with the rated response frequency to construct a basic control model describing the dynamic response characteristics of the device. The basic control model simulates the delay process from receiving a signal to achieving the output target using an exponential approximation function. The mathematical expression of the basic control model is as follows:

[0076]

[0077] In the formula, Indicates the basic control model in time The output state under the following conditions Indicates the initial processing parameters. This indicates the rated response frequency. The value of the rated response frequency is determined based on 150 no-load step response tests of the device to ensure that the model can accurately reflect the physical hysteresis characteristics of the hardware.

[0078] In one specific embodiment of the present invention, the determination process of the rated response frequency and the mechanism for ensuring physical hysteresis characteristics are not simply based on the ideal theoretical value on the equipment's nameplate. Instead, a rigorous closed-loop testing process is established to obtain the true electrical control characteristics of the hardware. Specifically, under a pure no-load state without mechanical cutting resistance, the system inputs 150 sets of instantaneous step commands spanning different basic output ranges to the laser generator or servo motor through the signal excitation module, and simultaneously uses a high-frequency oscilloscope acquisition card to capture the light intensity or speed rise curves at the output end. The test program analyzes these 150 sets of transient waveforms, accurately extracts the real time constant required for the output to reach 63.2% of the steady-state peak value, and then applies the frequency band conversion rule of a first-order inertial system to calculate the limit bandwidth limit of each set. Finally, the lower envelope boundary extreme value covering 150 measured samples is extracted as the current rated response frequency of the system. This calibration mechanism, based on pure physical measurements, precisely quantifies the 'disconnect between command and action' phenomenon caused by the charging and discharging delay of capacitors at the hardware level or the electrical inertia of the motor rotor into anti-hysteresis time constraints in mathematical equations. This ensures that the basic control model has high hardware fidelity, so that each time a dynamic instantaneous control command is injected in the time domain, its actual start-up timing can match the feedforward prediction of the processing position in terms of time parameters.

[0079] The system converts the energy compensation gain in spatial coordinates into a time series and injects it into the basic control model. During this process, the system calculates the theoretical output value after superimposing the gain based on the current time step and uses maximum load limits to safely truncate this theoretical output value to prevent equipment overload. Through this mapping process, the system calculates the laser duty cycle or servo motor speed at each time step, using it as the instantaneous control command to directly drive the hardware. The calculation formula for the instantaneous control command is as follows:

[0080]

[0081] In the formula, Indicates time The instantaneous control command issued Indicates time The corresponding energy compensation gain, Represents the gain conversion factor. Indicates the maximum load limit. Gain conversion factor. It is used to convert the dimensionless energy compensation gain into the same physical dimension as the initial processing parameters. Its value is calibrated to 45.5 based on 200 material cutting penetration experiments to ensure that the gain amplitude is precisely matched with the actual required laser duty cycle or servo motor speed increment.

[0082] In a specific embodiment of the present invention, the gain conversion coefficient The conversion logic and its precise matching mechanism essentially build a mathematical bridge spanning the abstract algorithm domain and the physical electronic control domain. This coefficient is not calculated based on empirical assumptions but rather established using a rigorous linear regression fitting model. Specifically, during the calibration phase, the system designed 200 gradient-based material cutting penetration experiments. When facing various typical high-resistance material clusters, the system, on the one hand, recorded the dimensionless energy compensation gain theoretical value of 0.15 calculated by the feedforward prediction model; on the other hand, it used a binary approximation exhaustive method to perform physical trial cuts at the execution end, accurately detecting the absolute increment of physical parameters actually required for compensation by the actuator when the cut reaches the critical mass state of 'just completely cut through without overheating and carbonization at the edge'. By capturing these 200 sets of corresponding data covering low, medium, and high resistance bands, the system used the least squares method to plot a scatter plot with 'dimensional theoretical gain' as the x-axis and 'actual physical parameter increment' as the y-axis, interpolating the optimal fitting line. The slope of this line was extracted, solidified, and given dimensions, becoming the gain conversion coefficient, calibrated to 45.5. This fitting operation, based on the accumulation of massive physical cutting destructive tests, eliminates the mapping gap between pure digital prediction and hardware power consumption, ensuring that every bit of abstract gain number issued by the system can be equivalently converted into the real machine physical energy dissipated for cutting the current material.

[0083] After acquiring the instantaneous control commands for all time steps, the system linearly arranges these commands along the time axis according to the machining sequence. Since the machining sequence strictly corresponds to the movement trajectory of the tool or laser beam on the material surface, the arranged command sequence can form a dynamic process control parameter flow that is completely synchronized in space and time with the hollowing design vector path. This dynamic process control parameter flow is then sent to the underlying controller as the final execution file, guiding the equipment to complete complex machining operations with varying parameters.

[0084] For example, consider the processing operation of a certain type of CNC laser die-cutting machine. The system obtains the rated response frequency of the laser generator from the equipment firmware, which is 5000 Hz, and the maximum load limit is set to a duty cycle of 95. The system reads the set initial processing parameters, which are a duty cycle of 60, and constructs a basic control model based on the rated response frequency. When processing a certain high-density fiber area, the system injects the energy compensation gain of 0.15 corresponding to that time point into the basic control model. Combined with a gain conversion coefficient calibrated to 45.5, the system calculates the theoretically required increase in duty cycle to be 6.825. Adding the basic output to the increment yields 66.825, which does not exceed the maximum load limit of 95. Therefore, the system uses 66.825 as the laser duty cycle at that time step and generates instantaneous control commands. The system arranges the 12,000 instantaneous control commands generated throughout the entire processing cycle according to the processing sequence, forming a dynamic process control parameter flow that corresponds one-to-one with the cutout design vector path containing 12,000 sampling points. This control flow allows the laser die-cutting machine to instantly increase the laser duty cycle when encountering high-resistance areas, successfully cutting tough fibers, while maintaining the basic duty cycle in flat, low-resistance areas, avoiding energy waste and excessive material ablation.

[0085] S5. Real-time acquisition of parameters such as the width of the heat-affected zone at the edge of the work area and the flatness of the cut.

[0086] In a specific embodiment of the present invention, the edge heat-affected zone width and cut flatness parameters of the working area are collected in real time, including: using an infrared thermal imaging component to monitor the temperature gradient distribution of the cut edge and extracting the width of the area with a temperature higher than the material carbonization threshold to obtain the edge heat-affected zone width.

[0087] The microscopic undulation data of the cut surface are measured by a laser displacement sensor, and the standard deviation of the microscopic undulation data is calculated to obtain the cut surface smoothness parameter.

[0088] The width of the edge heat-affected zone and the cut flatness parameters are timestamped to generate a multidimensional quality feedback dataset.

[0089] Specifically, the actuator receives dynamic process control parameters and drives the laser head or cutter to perform the cutout operation along the vector path of the cutout design. During the operation, the infrared thermal imaging component closely follows the processing point. The infrared thermal imaging component adopts a coaxial optical path design with the laser head or is mounted in the normal direction via a paraaxial follower of the robotic arm to ensure that its field of view center is always aligned with the current cutting point and maintains a fixed angle with the feed direction, thereby monitoring the temperature gradient distribution at the edge of the cutout in real time. The system sets a material carbonization threshold, which represents the critical temperature at which irreversible thermal damage occurs to the packaging material. The width of the edge heat-affected zone is obtained by extracting the width of the area where the temperature is higher than the material carbonization threshold. The calculation of the edge heat-affected zone width follows the following formula:

[0090]

[0091] In the formula, Indicates time Width of the lower edge heat-affected zone This indicates the distance from the center line of the incision is... Real-time temperature at the location, Indicates the carbonization threshold of the material. This is a step indicator function. It takes the value 1 when the temperature difference is greater than 0, and 0 otherwise. This represents the lower limit of integration, which is the theoretical edge position of the cut (usually determined by the kerf or the radius of the light spot). This represents the integration upper limit, i.e., the maximum effective outer boundary of the field of view monitored by the infrared thermal imaging component at this moment. Material carbonization threshold. Based on temperature rise experiments on 300 sets of packaging boxes of different materials, the extent of heat damage can be accurately defined.

[0092] It should be noted that, considering the physical interference avoidance at the processing site, the infrared thermal imaging component is mounted with a specific tilt angle for off-axis tracking. To eliminate the trapezoidal perspective distortion in thermal imaging caused by oblique shooting, the system must perform dynamic perspective inverse mapping correction before extracting the width of the edge heat-affected zone. Specifically, the system pre-extracts the intrinsic parameter matrix of the infrared lens and the extrinsic parameter matrix of the camera, including the tilt angle, using a calibration checkerboard grid. During real-time acquisition, after acquiring the set of pixels exceeding the thermal tolerance, the system performs inverse perspective projection transformation using a homography matrix, forcibly projecting and reconstructing the two-dimensional infrared pixel surface from the tilted viewpoint into a top-view orthophoto map of the thermal distribution that is absolutely parallel to the surface of the cut material. Only by performing pixel width statistics in the orthophoto matrix after this projection mapping and multiplying by the absolute spatial resolution can the absolute span value of the physical edge heat-affected zone with an accuracy of up to 0.01 mm be accurately and error-free, thus eliminating the systematic ranging error caused by the optical installation angle.

[0093] In a specific embodiment of the present invention, the material carbonization threshold The establishment of the basis and the specific extraction mechanism of the edge heat-affected zone width aim to define the boundary of the cut defect by accurately quantifying the physical phenomenon of heat penetration. Specifically, the system is based on 300 sets of simultaneous comparative experiments of infrared gradient heating and visual discoloration on white cardboard and corrugated wood pulp substrates commonly used in the packaging industry. It accurately captures the thermal runaway critical point of irreversible yellowing or charring caused by the breakage of plant fiber molecular chains, and solidifies it as the carbonization threshold of typical materials. The carbonization threshold of conventional 350g white cardboard is strictly anchored to 280 degrees Celsius. In the actual calculation process of the extracted area width, the system uses a high-resolution infrared thermal imaging component that is dynamically installed to capture a one-dimensional temperature gradient pixel array perpendicular to the current cut feed direction in real time, and then retrieves... The step binarization algorithm filters out and captures continuous 'thermal exceedance' pixels in the array whose physical temperature remains positive after subtracting the 280℃ carbonization threshold. Finally, the system automatically extracts the farthest coordinates of these exceedance pixels and directly multiplies the total pixel span between them and the bottom line of the physical inner wall of the cut by the pre-calibrated infrared lens spatial resolution. This accurately transforms the thermal diffusion blur halo, which is difficult to define instantly with the naked eye, into a digitally explicit absolute span value of the edge thermally affected area width. This provides zero-latency real target data for the system to subsequently perform closed-loop containment of thermal damage.

[0094] Simultaneously, the system uses a laser displacement sensor to perform high-frequency sampling on the processed cut surface, acquiring microscopic undulation data of the cut surface. This microscopic undulation data reflects the roughness of the processed edge. The system quantifies the quality of the cut by calculating the standard deviation of the microscopic undulation data, thereby obtaining the cut smoothness parameter. The formula for calculating the cut smoothness parameter is as follows:

[0095]

[0096] In the formula, Indicates time The following cut smoothness parameters, Indicates the first The micro-undulation height value of each sampling point This represents the average height of micro-fluctuations within the sampling window. This indicates the total number of sampling points. This parameter is set based on actual measurements of the cut cross-section using 200 sets of industrial sensors; a smaller standard deviation indicates a smoother cut.

[0097] To achieve a precise correlation between processing quality and control parameters, the system timestamps the width of the edge heat-affected zone and the cut smoothness parameters. These timestamps are used to correlate quality data with instantaneous control commands in the dynamic process control parameter stream via a synchronized clock signal, generating a multidimensional quality feedback dataset. The structure of the multidimensional quality feedback dataset is as follows:

[0098]

[0099] In the formula, This represents a multidimensional quality feedback dataset. This indicates the synchronization timestamp corresponding to the processing position. In this way, the system can trace the actual processing effect of each cutout design vector path under specific process parameters.

[0100] For example, consider the laser engraving process of a high-end gift packaging box. The actuator processes the material according to dynamic process control parameters, and the infrared thermal imaging component detects that the highest temperature at the cut edge reaches 450 degrees Celsius. Based on a preset material carbonization threshold of 280 degrees Celsius, the system identifies the lateral span of the area with a temperature higher than this threshold as 0.12 mm, thus obtaining a heat-affected zone width of 0.12 mm. Simultaneously, the laser displacement sensor measures the cut surface at 20-micrometer intervals, acquiring a set of micro-undulation data. The calculated standard deviation is 15 micrometers, indicating a cut smoothness parameter of 15 micrometers. The system binds these two quality indicators to the current timestamp of 20.5 seconds and stores them in a multi-dimensional quality feedback dataset. This dataset records the processing effect after laser power compensation at 20.5 seconds, providing real-time and objective data support for subsequent comparison with standard quality thresholds and generation of process deviation vectors.

[0101] S6. Generate a process deviation vector by comparing the edge heat-affected zone width, cut flatness parameters and standard quality threshold.

[0102] In a specific embodiment of the present invention, a process deviation vector is generated by comparing the edge heat-affected zone width, cut flatness parameters and standard quality thresholds, including: obtaining the minimum heat damage boundary and maximum roughness tolerance value required by the packaging process as the standard quality threshold.

[0103] The thermal damage component is obtained by calculating the difference between the width of the edge heat-affected zone and the minimum thermal damage boundary, and the morphological distortion component is obtained by calculating the ratio of the cut smoothness parameter to the maximum allowable roughness value.

[0104] By combining the thermal damage component and the morphological distortion component, a process deviation vector is constructed to characterize the difference between the current processing state and the ideal state.

[0105] Specifically, the system extracts the real-time monitored edge heat-affected zone width and cut smoothness parameters from the multi-dimensional quality feedback dataset. To assess whether the current processing quality meets the standards, the system retrieves the minimum thermal damage boundary and maximum roughness tolerance value required for packaging processes from a pre-set process database, using these two indicators as standard quality thresholds. The minimum thermal damage boundary represents the limit width at which carbonization and discoloration are allowed at the material edge, and the maximum roughness tolerance value represents the highest acceptable standard for micro-undulations in the cut surface. The system calculates the difference between the edge heat-affected zone width and the minimum thermal damage boundary to obtain the thermal damage component. The formula for calculating the thermal damage component is as follows:

[0106]

[0107] In the formula, Indicates time The thermal damage component below, Indicates time Width of the lower edge heat-affected zone This indicates the minimum thermal damage boundary. The value of the minimum thermal damage boundary was determined to be 0.05 mm based on 150 visual acceptance tests of the packaging box, ensuring that carbonization marks on the processed edges are not visible to the naked eye under normal lighting.

[0108] In a specific embodiment of the present invention, the calibration mechanism of the standard quality threshold and the definition of its typical value are not arbitrarily given empirical estimates, but are established by strictly relying on the cross-verification of industrial quality inspection standards for high-end packaging products and ergonomic experiments. Specifically, for the 'minimum thermal damage boundary', the researchers conducted a double-blind visual sensitivity acceptance test on 150 groups of specimens with gradient thermal damage under standard industrial light source and conventional operating viewing distance conditions. They accurately captured the physical span of the carbonization and discoloration marks at the cut edge at the critical point that is imperceptible to the human eye, thus strictly anchoring the typical upper limit of the absolute hard benchmark of this thermal damage to 0.05 mm. Simultaneously, for the 'maximum roughness tolerance', the system combined the microscopic scanning of a high-precision profilometer with 100 simulated end-face tactile evaluation experiments of consumers opening boxes and sliding their fingers. This scientifically revealed the extreme value of the maximum surface profile standard deviation fluctuation when human skin touches the cut edge, causing a tactile jump from 'smooth and unobstructed' to 'scratching and rough feeling', thus solidifying the typical upper limit of the morphological distortion benchmark to 20 micrometers. This highly quantified and extracted subjective surface decoration visual specifications and physical tactile quality inspection requirements into a calibration conversion model with precise physical absolute values ​​of 0.05 mm and 20 micrometers. This model provides an objective, rigorous, and insurmountable digital benchmark for the subsequent construction of a comprehensive process deviation vector pool in the system, ensuring that the system has an absolutely reliable basis for issuing commands when triggering closed-loop corrections.

[0109] Simultaneously, the system calculates the ratio of the cut smoothness parameter to the maximum allowable roughness value, thereby obtaining the topographic distortion component. The calculation formula for the topographic distortion component is as follows:

[0110]

[0111] In the formula, Indicates time The shape distortion component below, Indicates time The following cut smoothness parameters, This indicates the maximum allowable roughness value. The maximum allowable roughness value is determined to be 20 micrometers based on 100 cut tactile evaluation experiments, ensuring that the edges of the packaging box are free of burrs when touched.

[0112] After obtaining the thermal damage component and the morphological distortion component, the system combines the thermal damage component and the morphological distortion component to construct a process deviation vector characterizing the difference between the current processing state and the ideal state. The mathematical expression of the process deviation vector is as follows:

[0113]

[0114] In the formula, Indicates time The process deviation vector is shown below. This vector comprehensively reflects the thermal damage caused by excessive laser energy input and the roughness of the cut caused by mechanical vibration or insufficient energy, providing multi-dimensional error feedback signals for subsequent closed-loop control.

[0115] For example, consider the cut-out process of corrugated cardboard with a thickness of 2 mm. The system reads from the multi-dimensional quality feedback dataset that the edge heat-affected zone width at a specific timestamp is 0.08 mm and the cut flatness parameter is 25 micrometers. The system retrieves standard quality thresholds, where the minimum thermal damage boundary is set to 0.05 mm and the maximum roughness tolerance is set to 20 micrometers. The system calculates the difference between the edge heat-affected zone width of 0.08 mm and the minimum thermal damage boundary of 0.05 mm, obtaining a thermal damage component of 0.03 mm. Simultaneously, the system calculates the ratio of the cut flatness parameter of 25 micrometers to the maximum roughness tolerance of 20 micrometers, obtaining a shape distortion component of 1.25. The system combines the thermal damage component of 0.03 mm and the shape distortion component of 1.25 to construct a process deviation vector containing these two values. This process deviation vector indicates that the current processing state deviates from the ideal state in both the thermal damage and shape distortion dimensions, directly triggering the subsequent closed-loop correction mechanism.

[0116] S7. Use the process deviation vector to perform closed-loop correction of the dynamic process control parameter flow.

[0117] In a specific embodiment of the present invention, the process deviation vector is used to perform closed-loop correction of the dynamic process control parameter flow, including: inputting the process deviation vector into the proportional-integral-derivative control algorithm, and calculating the feedback correction value for the dynamic process control parameter flow.

[0118] Determine whether the feedback correction value exceeds the safety threshold of the actuator. If the feedback correction value does not exceed the safety threshold of the actuator, then add the feedback correction value to the dynamic process control parameter stream to obtain the updated control stream.

[0119] If the feedback correction value exceeds the safety threshold of the actuator, the feed rate parameter in the dynamic process control parameter stream is reduced, the feedback correction value is recalculated, and an update is performed.

[0120] Specifically, the system inputs the process deviation vector into the proportional-integral-derivative (PID) control algorithm. This algorithm calculates the adjustment amount by integrating the current deviation, historical deviation accumulation, and deviation change trend, thereby achieving dynamic intervention in the processing. Since the process deviation vector contains thermal damage and morphological distortion components with different dimensions, the system performs normalized weighted fusion of these two components within the algorithm to calculate the feedback correction value for the dynamic process control parameter flow.

[0121]

[0122]

[0123] In the formula, Indicates time The feedback correction value below, This represents the dimensionless composite deviation scalar constructed internally by the algorithm. , , These represent the proportional coefficient, integral coefficient, and differential coefficient, respectively. and These represent the weights for thermal damage and morphology distortion, respectively. Indicates the thermal damage component. Indicates the minimum thermal damage boundary. This represents the shape distortion component. The weighting coefficients are set based on 300 process optimization experiments with different materials, typically... Set to 0.6. The coefficients were set to 0.4 to prioritize the suppression of irreversible material ablation. The values ​​of the proportional, integral, and derivative coefficients were determined based on 100 closed-loop response tests to ensure that the system can quickly and stably eliminate process deviations.

[0124] After calculating the feedback correction value, the system determines whether it exceeds the actuator's safety threshold. The actuator's safety threshold represents the maximum parameter fluctuation the equipment can withstand within a single command cycle. It is set based on 50 motor and laser overload tests, designed to prevent drastic adjustments from triggering hardware protective shutdowns. If the absolute value of the feedback correction value does not exceed the actuator's safety threshold, the system directly adds the feedback correction value to the dynamic process control parameter stream, generating an updated control flow. The formula for calculating the updated control flow is as follows:

[0125]

[0126] In the formula, Indicates time The updated control flow This represents the instantaneous control command in the original dynamic process control parameter stream. Subtraction is used here because a positive deviation indicates excessive energy input or mechanical force, requiring a corresponding reduction in control parameters.

[0127] In a specific embodiment of the present invention, the typical value setting of the safety threshold of the actuator and its anti-shutdown protection logic are intended to impose a strict physical limit lock on the purely mathematical closed-loop intervention. The typical application value of this threshold in the system is strictly calibrated to 5.0, and the maximum absolute increment of the duty cycle allowed to be crossed in a single cycle is 5.0%. The definition of this specific value is based on 50 destructive limit measurements performed by the system in offline state: the test program forcibly injects aggressive instructions with a step-by-step increase in the jump amplitude into the device, and simultaneously detects the surge state of the hardware bus current. The actual test found that when the instantaneous jump amplitude of the instruction approaches the 6.2% boundary, it will instantly break through the tolerance limit of the servo motor or laser power supply and directly trigger the relay overcurrent power-off and crash. The system then accurately captures this physical collapse point and introduces a preset steady-state safety redundancy coefficient for lower limit calculation, and finally rigorously refines and solidifies the limit red line of 5.0. By assigning this specific benchmark, even if the PID algorithm aggressively calculates a feedback correction value higher than this benchmark for extremely severe process errors, this safety threshold mechanism can instantly execute hard cut-off and interception, forcing the system to abandon the burst power compensation and switch to the safe 'reduced speed replacement increment' strategy. This effectively prevents component breakdown and full-line protective shutdown accidents induced by transient and drastic adjustments at the underlying algorithm logic level.

[0128] If the absolute value of the feedback correction exceeds the safety threshold of the actuator, it indicates that simply adjusting the laser power or tool pressure is insufficient to compensate for the deterioration in machining quality within a safe range. In this case, the system reduces the feed rate parameter in the dynamic process control parameter stream. The feed rate parameter represents the speed at which the actuator moves across the material surface. By reducing the feed rate parameter, the system increases the machining time per unit path, thereby altering the spatial distribution of energy density. The formula for calculating the speed reduction is as follows:

[0129]

[0130] In the formula, This indicates the reduced feed rate parameter. This represents the original feed rate parameter. This represents the speed attenuation coefficient. The value of the speed attenuation coefficient is calibrated to 0.15 based on 80 variable acceleration cutting experiments. After reducing the feed rate parameter, the system recalculates the feedback correction value and executes the update process until the machining state returns to the ideal range.

[0131] For example, consider the corrugated cardboard cutout process. The system receives a process deviation vector at timestamp 20.5 seconds, where the thermal damage component is 0.03 mm and the shape distortion component is 1.25. The system inputs this process deviation vector into the proportional-integral-derivative (PID) control algorithm. Combining the minimum thermal damage boundary of 0.05 mm and the set weighting coefficients of 0.6 and 0.4, the algorithm calculates a comprehensive deviation scalar of 0.46. After comprehensive calculations of the proportional, integral, and derivative terms, the system calculates a feedback correction value of 8.5 for the dynamic process control parameter flow. The system retrieves the safety threshold of the actuator, which is set to 5.0. Since the feedback correction value of 8.5 exceeds the safety threshold of 5.0, the system determines that directly reducing the duty cycle will cause control oscillation. Therefore, the system reduces the feed rate parameter, currently set at 100 mm / s, to 85 mm / s. After the speed reduction, the local heating and stress state of the material changes, and the system recalculates a new feedback correction value of 3.2. The value did not exceed the safety threshold of the actuator, so the system superimposed it on the dynamic process control parameter stream, lowering the laser duty cycle command from 66.825 to 63.625, resulting in an updated control stream. This closed-loop correction process enabled subsequent hollowing operations to quickly deviate from the trend of escalating thermal damage, restoring a smooth cut quality.

[0132] In a specific embodiment of the present invention, after the step of using the process deviation vector to perform closed-loop correction of the dynamic process control parameter flow, the method further includes: statistically analyzing the parameter change rate of the dynamic process control parameter flow before and after correction, and establishing parameter evolution sample pairs in conjunction with the processing resistance prediction field.

[0133] By using the calculated weights of parameter evolution samples to update the energy compensation gain, a self-optimized process knowledge base is generated.

[0134] In the next processing task, the process knowledge base is called to pre-compensate the initial processing parameters.

[0135] Specifically, upon completion of a single processing task or during processing, the system's backend statistically analyzes the rate of change of dynamic process control parameters before and after correction. This rate of change reflects the absolute deviation between the theoretical prediction model and the actual physical processing state. The system calculates the relative deviation between the original instantaneous control commands and the updated control flow after closed-loop correction at the same time sequence. The calculation of the rate of change follows the formula below:

[0136]

[0137] In the formula, Indicates time The rate of change of the parameters under the following conditions This represents the instantaneous control commands in the original dynamic process control parameter stream. This represents the updated control flow after closed-loop correction.

[0138] Since each time step in the processing can be precisely aligned with the spatial trajectory, the system maps the calculated rate of change of parameters back to the coordinate system of the processing base map and retrieves the predicted field value of the processing resistance at the corresponding coordinate point. The system binds the predicted resistance at the same point with the actual rate of change of parameters, establishing parameter evolution sample pairs. The set expression of the parameter evolution sample pairs is as follows:

[0139]

[0140] In the formula, Indicates inclusion The set of parameter evolution sample pairs for each valid sampling point Representing coordinates Predicted processing resistance at the location, This represents the rate of change of the parameter at the corresponding time point.

[0141] Based on the constructed parameter evolution sample pairs, the system updates the calculation weights of the energy compensation gain using either gradient descent or a moving weighted average algorithm. Since the initially set energy compensation gain is often based on empirical constants, this step aims to adaptively adjust the feedforward calculation weights according to the actual physical response of the material batch. The evolutionary update formula for the calculation weights is as follows:

[0142]

[0143] In the formula, This represents the updated calculated weights. This represents the historical calculation weight of the previous period (initial value is 1.0). This represents the knowledge learning rate parameter. This represents the average resistance benchmark. Learning rate parameter. The setting is based on 500 batches of material cutting evolution experiments, and is usually fixed at... to To balance the system's learning speed with its noise resistance stability, the updated computational weights will be categorized and stored, generating a self-optimized process knowledge base.

[0144] In a specific embodiment of the present invention, the underlying mathematical logic of 'updating the calculated weights using gradient descent' is as follows: The system first establishes a loss function aimed at reducing the error between the theoretical prediction of processing parameters and the actual requirements. The loss function is defined as the rate of change of parameters, and the squared error of the relative deviation between the actual compensation parameters and the feedforward prediction parameters: During the processing, the actual deviation of the dynamic process control parameter flow is mainly due to the original calculated weights. This is due to inaccuracies. The system calculates the partial derivative of the loss function with respect to the weights (i.e., the gradient), and uses the learning rate parameter... Perform backpropagation updates. The specific gradient descent update iteration rule is as follows: By introducing the actual rate of change of parameters. As a penalty term guiding gradient descent, the system can drive the computation of weights. Through multiple batches of processing, the energy mapping physical background of the current material is continuously approached, completely upgrading the parameter update from 'blind trial and error' to 'directional convergence'.

[0145] In the next processing task of packaging materials of the same or similar specifications, before executing step S3, the system first calls the process knowledge base to match and retrieve the latest calculated weights, and pre-compensates the energy compensation stage of the initial processing parameters. At this time, the new energy compensation gain formula evolves into: Through this cross-cycle experience inheritance and iteration, the system can transform the posterior experience of the previous feedback correction into the prior prediction of the current stage, significantly reducing the triggering frequency and adjustment burden of PID closed-loop control.

[0146] For example, consider the batch cutting operation of a certain batch of thick cardboard. When processing the first packaging box, the laser duty cycle generated by the predictive model feedforward is 66.8. However, due to the abnormal glue content of this batch of cardboard, the cut is slightly rough. After the closed-loop correction module intervenes, the actual duty cycle is increased to 72.1 to successfully cut through the material. The system background statistics show the rate of change of active parameters in this high-resistance region. Approximately The system predicts the resistance value for this area. With rate of change Extract the data and establish hundreds of parameter evolution sample pairs.

[0147] Subsequently, the system calls the weight evolution update formula, with the learning rate... Set as In this case, the energy compensation calculation weight for processing this material will be adjusted. From the initial Updated to It is then stored in the process knowledge base under the category labeled "thick recycled paperboard - batch A".

[0148] When the next packaged box from the same batch enters the processing station on the production line, the system directly calls the process knowledge base and uses the updated calculation weights. Pre-compensation is performed. When encountering similar high-resistance regions, feedforward control directly generates a duty cycle command sufficient to cut through the material (approximately...). The entire cutting process achieved the standard quality threshold in one go, with excellent performance in both the edge heat-affected zone and cut smoothness. This avoided false alarms from the infrared thermal imager and displacement sensor, and improved the consistency of continuous production processes and the adaptive iteration capability of the equipment.

[0149] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A parameter self-optimization method for intelligent packaging box cutout process, characterized in that, Includes the following steps: S1. Obtain the surface texture image and cutout design vector path of the packaging box to be processed; S2. Extract fiber arrangement features from the surface texture image and geometric curvature features from the vector path of the hollow design. Fuse the fiber arrangement features and geometric curvature features to generate a processing resistance prediction field, including: performing Fourier transform on the surface texture image to obtain a frequency domain distribution map, and identifying the main frequency direction and amplitude intensity in the frequency domain distribution map to obtain fiber arrangement features. The local curvature is obtained by calculating the second derivative of the vector path of the hollow design at each sampling point, and the geometric curvature characteristics are obtained based on the rate of change of the local curvature with the path length. A resistance correlation matrix is ​​established, and the fiber arrangement characteristics are used as row vectors and the geometric curvature characteristics are used as column vectors to perform Kronecker product operation to generate a processing resistance prediction field. S3. Calculate the energy compensation gain at different coordinate points based on the processing resistance prediction field, including: retrieving the resistance values ​​at each coordinate point in the processing resistance prediction field, and calculating the degree of deviation between the resistance values ​​and the average resistance benchmark to obtain the resistance fluctuation coefficient. Based on the energy adjustment function matched with the resistance fluctuation coefficient, the correction amount for laser power or tool pressure is calculated to obtain the energy compensation gain. The energy compensation gain is smoothed by filtering to eliminate oscillations caused by sudden changes in gain between adjacent coordinate points, thus generating a smooth gain sequence. S4. Map the initial processing parameters of the actuator according to the energy compensation gain to generate a dynamic process control parameter stream; S5. Real-time acquisition of parameters such as the width of the heat-affected zone at the edge of the work area and the flatness of the cut; S6. Generate a process deviation vector by comparing the edge heat-affected zone width, cut flatness parameters and standard quality threshold. S7. Use the process deviation vector to perform closed-loop correction of the dynamic process control parameter flow.

2. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, The process of obtaining the surface texture image and cutout design vector path of the packaging box to be processed includes: The original reflectance matrix is ​​obtained by scanning the surface of the packaging material using a visual sensing component. The surface texture image is then obtained by performing grayscale gradient processing on the original reflectance matrix. The coordinate sequence is extracted from the CAD design document of the packaging box, and the coordinate sequence is continuously interpolated to obtain the vector path of the hollow design. Align the surface texture image with the vector path of the cutout design in spatial coordinates to generate a processing base map with a textured background.

3. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, The process of mapping the initial processing parameters of the actuator based on the energy compensation gain to generate a dynamic process control parameter stream includes: Obtain the rated response frequency and maximum load limit of the actuator, and construct a basic control model based on the initial processing parameters; Inject the energy compensation gain into the basic control model, calculate the laser duty cycle or servo motor speed at each time step to obtain the instantaneous control command; The instantaneous control commands are arranged according to the processing sequence to form a dynamic process control parameter flow that is synchronized with the vector path of the hollowing design.

4. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, The parameters for the width of the heat-affected zone and the flatness of the cut in the real-time acquisition work area include: The temperature gradient distribution at the edge of the cutout is monitored using an infrared thermal imaging component, and the width of the area with a temperature higher than the carbonization threshold of the material is extracted to obtain the width of the edge heat-affected zone. The micro-undulation data of the cut surface are measured by a laser displacement sensor, and the standard deviation of the micro-undulation data is calculated to obtain the cut surface smoothness parameter. The width of the edge heat-affected zone and the cut flatness parameters are timestamped to generate a multidimensional quality feedback dataset.

5. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, The process deviation vector is generated by comparing the edge heat-affected zone width, cut smoothness parameters, and standard quality thresholds, including: The minimum thermal damage boundary and the maximum roughness tolerance required by the packaging process are used as standard quality thresholds; The thermal damage component is obtained by calculating the difference between the width of the edge heat-affected zone and the minimum thermal damage boundary, and the morphology distortion component is obtained by calculating the ratio of the cut smoothness parameter to the maximum allowable roughness value. By combining the thermal damage component and the morphological distortion component, a process deviation vector is constructed to characterize the difference between the current processing state and the ideal state.

6. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, The method of using process deviation vectors to perform closed-loop correction of dynamic process control parameter flow includes: The process deviation vector is input into the proportional-integral-derivative control algorithm to calculate the feedback correction value for the dynamic process control parameter flow; Determine whether the feedback correction value exceeds the safety threshold of the actuator. If the feedback correction value does not exceed the safety threshold of the actuator, then add the feedback correction value to the dynamic process control parameter stream to obtain the updated control stream. If the feedback correction value exceeds the safety threshold of the actuator, the feed rate parameter in the dynamic process control parameter stream is reduced, the feedback correction value is recalculated, and an update is performed.

7. The parameter self-optimization method for an intelligent packaging box hollowing process according to claim 1, characterized in that, After the step of using the process deviation vector to perform closed-loop correction of the dynamic process control parameter flow, the following steps are also included: The rate of change of statistical dynamic process control parameters before and after correction is used to establish parameter evolution sample pairs in combination with the processing resistance prediction field; By using the calculated weights of parameter evolution samples to update the energy compensation gain, a self-optimized process knowledge base is generated. In the next processing task, the process knowledge base is called to pre-compensate the initial processing parameters.

Images