Intelligent feeding and discharging method and system for die-cutting machine

Abstract

The application relates to the technical field of industrial automation control, and discloses an intelligent feeding and discharging method and system for a die-cutting machine. The method comprises the following steps: acquiring material thickness sampling signals, material tension sampling signals and environmental light intensity of the die-cuting machine, and preprocessing to obtain state fusion data; inputting regional images and the state fusion data into a detection network to obtain actual grabbing coordinates; calculating a coordinate deviation value of the actual grabbing coordinates and a preset reference position, feeding back a posture compensation amount obtained by calculation, and obtaining an adjusted grabbing instruction if the coordinate deviation value is greater than a preset deviation threshold value; comparing and processing a real-time beat signal to obtain a speed matching degree and a speed fluctuation rate; inputting a fuzzy rule base to infer a fuzzy adjustment amount and coordinated operation parameters; performing time sequence prediction processing on a material flow rate to obtain an interruption risk probability; and obtaining a channel allocation instruction according to a preset reserve logic. The method can realize dynamic self-adaptive coordination of feeding and discharging.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to an intelligent loading and unloading method and system for die-cutting machines. Background Technology

[0002] Currently, in modern manufacturing, automated loading and unloading technology for die-cutting machines is a crucial pillar for improving production efficiency and product quality. Especially in fields such as packaging and electronic component processing, efficient and precise loading and unloading directly affects the stability of the production line.

[0003] In existing technologies, basic robotic arms are typically used in conjunction with fixed feeding programs for material handling. However, existing loading and unloading methods often reveal significant shortcomings when dealing with complex production scenarios. Many solutions struggle to adapt to the dynamic changes in material conditions during production and cannot effectively balance the relationship between equipment operating rhythm and material supply speed. Specifically, due to the lack of a dynamic data acquisition and control mechanism, the system struggles to perceive and adjust the matching between material conditions and equipment operation in real time. The remaining thickness and tension of the material dynamically change during continuous operation, directly affecting the selection of gripping points and the stability of feeding. Failure to accurately grasp these changes can easily lead to gripping position deviations, resulting in the risk of material accumulation or feeding interruptions. Furthermore, such deviations can interfere with the coordination between die-cutting cycle time and feeding speed, making it difficult for the equipment to maintain stability during high-speed operation, disrupting the production rhythm, and potentially causing equipment damage or material waste.

[0004] Existing technologies have the problem of difficulty in achieving dynamic adaptive coordination in the loading and unloading of die-cutting machines. Summary of the Invention

[0005] This invention provides an intelligent loading and unloading method and system for die-cutting machines to solve the problem of dynamic adaptive coordination in the loading and unloading of die-cutting machines in the prior art.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an intelligent loading and unloading method for a die-cutting machine, comprising:

[0007] The material thickness sampling signal, material tension sampling signal, and ambient light intensity of the die-cutting machine are acquired and preprocessed to obtain state fusion data;

[0008] A region image of the die-cut material is acquired, and the region image and the state fusion data are input into a pre-trained detection network for feature inference to obtain the actual grasping coordinates.

[0009] Calculate the coordinate deviation between the actual grasping coordinates and the preset reference position. If the coordinate deviation is greater than the preset deviation threshold, perform feedback calculation on the coordinate deviation to obtain the attitude compensation amount. Perform control mapping processing based on the attitude compensation amount to obtain the adjusted grasping command.

[0010] The real-time cycle signal of the die-cutting equipment is acquired, and the simulated path is compared with the real-time cycle signal according to the adjusted grasping instruction to obtain the speed matching degree and speed fluctuation rate.

[0011] The speed matching degree and the speed fluctuation rate are input into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount. The fuzzy adjustment amount is then parsed to obtain coordinated operation parameters.

[0012] The material flow rate of the production line is obtained, and time-series prediction processing is performed based on the coordinated operating parameters and the material flow rate to obtain the probability of interruption risk.

[0013] Based on the interruption risk probability and the preset reserve logic, the main and auxiliary channels are allocated traffic to obtain the channel allocation instruction.

[0014] Secondly, the present invention provides an intelligent loading and unloading system for a die-cutting machine, comprising:

[0015] The signal processing module is used to acquire the material thickness sampling signal, material tension sampling signal and ambient light intensity of the die-cutting machine, and perform preprocessing to obtain state fusion data;

[0016] The coordinate deduction module is used to acquire the regional image of the die-cutting material, input the regional image and the state fusion data into the pre-trained detection network for feature deduction, and obtain the actual grasping coordinates;

[0017] The attitude correction module is used to calculate the coordinate deviation between the actual grasping coordinates and the preset reference position. When the coordinate deviation is greater than the preset deviation threshold, the module performs feedback calculation on the coordinate deviation to obtain the attitude compensation amount. Based on the attitude compensation amount, the module performs control mapping processing to obtain the adjusted grasping command.

[0018] The matching evaluation module is used to acquire the real-time cycle signal of the die-cutting equipment, and perform simulated path comparison processing based on the adjusted capture command and the real-time cycle signal to obtain the speed matching degree and speed fluctuation rate.

[0019] The coordination and scheduling module is used to input the speed matching degree and speed fluctuation rate into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount, and to analyze the fuzzy adjustment amount to obtain coordinated operation parameters.

[0020] The risk prediction module is used to obtain the material flow rate of the production line, and perform time-series prediction processing based on the coordinated operating parameters and material flow rate to obtain the probability of interruption risk.

[0021] The emergency response module is used to calculate the flow allocation between the main and auxiliary channels based on the probability of interruption risk and the preset reserve logic, and to obtain the channel allocation instruction.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention collects multi-source signals by combining an ambient light compensation sensor array and uses a deep convolutional neural network to interactively deduce the directional velocity component and heat flow path. This process realizes deep computation and cross-validation of physical sensing data and visual features, which can accurately remove background noise and sampling bias in dynamic production environments, thereby achieving high-precision positioning of material grasping coordinates. It significantly reduces positioning deviations caused by changes in illumination or fluctuations in material roll diameter, and improves the success rate and reliability of robotic gripping under complex working conditions.

[0024] (2) This invention introduces Bézier curve path simulation and fuzzy control algorithm to monitor the matching degree between the feeding speed and the operating cycle of the die-cutting equipment in real time, and dynamically optimizes the motor speed and tension regulating valve opening accordingly. This adaptive feedback scheduling mechanism changes the limitation of traditional fixed feeding programs in dealing with dynamic disturbances in the system, enabling the material conveying process to be aligned with parameters at the millisecond level according to the actual operating load of the equipment. It achieves deep synchronization between the feeding rhythm and the die-cutting frequency, effectively preventing the accumulation or pulling damage of materials in the die-cutting area, and ensuring the stable operation of the production line.

[0025] (3) This invention constructs a flow uniformity index and uses an autoregressive moving average algorithm to perform time-series prediction, combined with buffer reserve monitoring and dynamic flow allocation logic for main and auxiliary channels. This scheme achieves a leap from passive response to forward-looking early warning in dealing with production interruption risks. The algorithm calculates the risk coverage window in advance and automatically activates auxiliary channels for flow compensation. This minimizes the risk of downtime due to material shortages, ensures production continuity under high-intensity continuous operation, and improves the overall capacity and intrinsic safety level of die-cutting processing. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the intelligent loading and unloading method for a die-cutting machine provided in the first embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the intelligent loading and unloading system for a die-cutting machine provided in the second embodiment of the present invention. Detailed Implementation

[0028] 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.

[0029] Reference Figure 1 The first embodiment of the present invention provides an intelligent loading and unloading method for a die-cutting machine, comprising the following steps:

[0030] S11: Acquire the material thickness sampling signal, material tension sampling signal, and ambient light intensity of the die-cutting machine, and perform preprocessing to obtain state fusion data;

[0031] S12, acquire the region image of the die-cutting material, input the region image and the state fusion data into the pre-trained detection network for feature inference, and obtain the actual grasping coordinates;

[0032] S13, calculate the coordinate deviation value between the actual grasping coordinates and the preset reference position. If the coordinate deviation value is greater than the preset deviation threshold, perform feedback calculation on the coordinate deviation value to obtain the attitude compensation amount. Perform control mapping processing based on the attitude compensation amount to obtain the adjusted grasping command.

[0033] S14, acquire the real-time cycle signal of the die-cutting equipment, and perform simulated path comparison processing with the real-time cycle signal according to the adjusted grabbing instruction to obtain the speed matching degree and speed fluctuation rate.

[0034] S15, input the speed matching degree and the speed fluctuation rate into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount, and perform parsing processing on the fuzzy adjustment amount to obtain coordinated operation parameters;

[0035] S16, obtain the material flow rate of the production line, and perform time-series prediction processing based on the coordinated operating parameters and the material flow rate to obtain the probability of interruption risk;

[0036] S17, calculate the flow allocation for the main and auxiliary channels based on the interruption risk probability and the preset reserve logic to obtain the channel allocation instruction.

[0037] In step S11, the material thickness sampling signal, material tension sampling signal, and ambient light intensity of the die-cutting machine are acquired and preprocessed to obtain state fusion data, including:

[0038] Acquire material thickness sampling signal, material tension sampling signal and ambient light intensity of the die-cutting machine, and calculate signal gain compensation value based on the ambient light intensity using a preset illumination model;

[0039] The material thickness sampling signal and the material tension sampling signal are offset-corrected using the signal gain compensation value to obtain calibrated thickness data and tension data.

[0040] Kalman filtering iterative calculations are performed on the calibrated thickness data and the tension data to obtain state fusion data.

[0041] In one implementation, this embodiment acquires a material thickness sampling signal containing optical reflection voltage characteristics using a laser displacement sensor deployed above the die-cutting machine's feed rack, and acquires a material tension sampling signal using a strain gauge pressure sensor installed at the bottom of the tension roller bearing seat. Simultaneously, a photosensitive sensor deployed directly above the material monitors ambient light intensity in real time, and a temperature sensor deployed near the pressure sensor acquires the real-time ambient temperature.

[0042] In this embodiment, the illuminance value of the ambient light intensity is extracted and substituted into a preset illumination model to perform a first-order linear polynomial calculation to obtain the signal gain compensation value. It should be noted that the preset illumination model is constructed as follows: under a calibration environment where the die-cutting machine is stopped and there is no physical load, multiple discrete gradient illuminance values ​​are simulated using a controlled light source as independent variables, and the static zero-point drift voltage of the laser displacement sensor is recorded as the dependent variable; a least squares method is used for linear regression fitting, and the slope of the fitted line is extracted as the illumination bias coefficient.

[0043] In one implementation, this embodiment uses the signal gain compensation value to perform offset correction processing on the material thickness sampling signal to obtain calibrated thickness data, and performs temperature drift correction on the material tension sampling signal to obtain calibrated tension data. Specifically, the voltage value of the material thickness sampling signal is subtracted from the signal gain compensation value to eliminate common-mode photoelectric interference caused by ambient light intensity fluctuations, resulting in a net thickness voltage. For the material tension sampling signal generated by physical strain, this embodiment retrieves the real-time ambient temperature collected by the temperature sensor and calculates a temperature compensation value based on preset thermal expansion coefficients and sensitivity coefficients. The temperature compensation value is then subtracted from the material tension sampling signal to perform temperature drift correction, resulting in a net tension voltage. Subsequently, the net thickness voltage and the net tension voltage are multiplied by their corresponding physical conversion constants. These physical conversion constants are directly extracted from the sensitivity parameters specified in the manufacturer's specifications of the laser displacement sensor and strain gauge pressure sensor, such as millimeters per volt and Newtons per volt, thereby outputting the calibrated thickness data and tension data with physical dimensions.

[0044] In one implementation, this embodiment performs Kalman filtering iterative calculations on the calibrated thickness data and the tension data to obtain state fusion data. This embodiment constructs a two-dimensional state observation vector from the two data streams, and calculates the optimal state estimate for the current moment through prediction in the time update phase and Kalman gain allocation in the measurement update phase. The measurement noise covariance matrix involved in the Kalman filtering is statistically calibrated using the background noise variance statistics of the acquisition device under no-load uniform speed operation for 24 consecutive hours; the process noise covariance matrix is ​​determined by fitting the residual of the servo motor's speed fluctuation using an offline system identification method; the initial state estimate calculated by the Kalman filtering iterative calculation is taken as a vector composed of the calibrated thickness data and tension data obtained from the first measurement, and the initial covariance matrix is ​​taken as the identity matrix multiplied by ten times the sensor measurement noise variance.

[0045] For example, the system acquires an ambient light intensity of 500 lux and calculates a light compensation value of 0.001 volts. Subtracting this value from the acquired thickness signal of 2.501 volts yields 2.500 volts. Simultaneously, the system detects a temperature drift of 0.002 volts in the tension signal due to an increase in ambient temperature. After temperature correction, the net tension voltage is obtained. The system retrieves the physical conversion constants from the sensor's manufacturer's specifications, quantizes the voltage into thickness data of 0.501 mm and tension data of 201.2 Newtons, and finally smooths the output using a Kalman filter.

[0046] In step S12, a region image of the die-cutting material is acquired, and the region image and the state fusion data are input into a pre-trained detection network for feature deduction to obtain the actual grasping coordinates.

[0047] Specifically, the region image and the state fusion data are input into a pre-trained detection network for feature deduction to obtain the actual grasping coordinates, including:

[0048] The Canny edge detection algorithm is used to process the image of the region to obtain hard edge features;

[0049] The image of the region is processed using the gray-level co-occurrence matrix algorithm to obtain texture features;

[0050] The hard edge features, texture features, and state fusion data are subjected to tensor concatenation to obtain a comprehensive feature vector.

[0051] The comprehensive feature vector is input into a pre-trained detection network for regression analysis to obtain the actual capture coordinates.

[0052] In one implementation, this embodiment acquires a regional image of the die-cutting material using an industrial camera deployed above the die-cutting station with its optical axis perpendicular to the strip plane. The resolution of the regional image is set to 1920×1080 pixels. This embodiment utilizes the Canny edge detection algorithm to perform dual-threshold gradient detection on the regional image, extracting a set of spatial abrupt change points representing the physical boundaries of the material to obtain hard edge features.

[0053] It is worth noting that the method for determining the high threshold in the Canny edge detection algorithm is as follows: The gradient intensity histogram of material images collected during historical production cycles is statistically analyzed, and the value corresponding to the gradient cumulative distribution function reaching 75% is extracted and determined as the high threshold; one-third of the high threshold is then determined as the low threshold. After obtaining the preliminary edge, this embodiment uses a 3×3 structuring element to perform morphological operations such as dilation and erosion. The 3×3 structuring element is determined based on the average pixel width of typical mechanical burrs on the edge of the die-cut material, obtained through microscopic visual statistics. Since the burr width is determined to be no more than 2 pixels, a 3×3 operator is selected to ensure complete coverage and filter out edge noise.

[0054] In one implementation, this embodiment uses a gray-level co-occurrence matrix algorithm to extract the contrast and uniformity values ​​of the region image and assembles them into texture features. Before performing tensor stitching, since the hard edge features are two-dimensional spatially distributed tensors, while the texture features and the state fusion data are one-dimensional scalars, this embodiment performs spatial dimension expansion processing. Using a tensor broadcasting mechanism, the one-dimensional scalar data is copied and filled in the spatial dimension to generate multiple scalar feature maps with the same height and width as the hard edge features. Subsequently, the hard edge features are stitched together with all the expanded scalar feature maps in the channel dimension to construct a comprehensive feature tensor with a three-dimensional structure. This processing method achieves deep alignment of physical state parameters and visual features at the low-level operator level while preserving the image spatial inductive bias.

[0055] In one implementation, this embodiment inputs the comprehensive feature tensor into a pre-trained detection network for regression inference, outputting the actual grasping coordinates. The pre-trained detection network uses ResNet-18 as the feature extraction backbone, and connects a regression output head consisting of three fully connected layers at its end. The regression output head maps the extracted high-dimensional features to horizontal and vertical axis values ​​in the image coordinate system, obtaining the actual grasping coordinates.

[0056] It is worth noting that the construction and training process of the pre-trained detection network involves collecting 5000 material sample images labeled with precise grasping center point locations as a training set; using the extracted comprehensive feature tensor as input, and manually calibrated optimal grasping coordinates as supervision labels; and employing mean squared error as the loss function for the regression task. In terms of training hyperparameter settings, the initial learning rate is set to 0.01, the batch size to 32, and backpropagation iterative updates are performed using the stochastic gradient descent algorithm with momentum. This embodiment employs a cosine annealing strategy to dynamically decay the learning rate and monitors the prediction bias on the independent validation set in real time during training until the mean squared error of the model on the validation set drops below 0.05 pixels, at which point the network is considered converged and the weight matrix is ​​fixed.

[0057] For example, the system acquires a region image and processes it to obtain hard edge features. The system extracts texture features with a contrast of 0.35 and a uniformity of 0.82 using a gray-level co-occurrence matrix algorithm. The system expands the texture feature scalar with state fusion data of 0.5005 mm thickness into a feature map through a broadcast mechanism, and concatenates it with the hard edge features to form a comprehensive feature tensor. The comprehensive feature tensor is input into a ResNet-18 model for feature inference, outputting the actual grasping coordinates in the image coordinate system as 320.5 pixels on the x-axis and 240.3 pixels on the y-axis. The prediction result shows good mean squared error on the validation set, meeting the accuracy requirements for dynamic high-speed grasping.

[0058] In step S13, the coordinate deviation between the actual grasping coordinates and the preset reference position is calculated. If the coordinate deviation is greater than the preset deviation threshold, the coordinate deviation is fed back to calculate the attitude compensation amount. Based on the attitude compensation amount, control mapping processing is performed to obtain the adjusted grasping command.

[0059] In one implementation, this embodiment retrieves a preset reference position from the system storage unit. The preset reference position is the three-dimensional physical coordinate point corresponding to the end effector of the gripping mechanism under standard feeding conditions of the die-cutting machine. Specifically, after the die-cutting machine is installed and debugged, the end effector of the gripping mechanism is moved to directly above the center of the material under standard feeding conditions, the three-dimensional physical coordinates of this point are recorded, and the arithmetic mean is taken after 10 repeated measurements and stored in the system storage unit as the preset reference position. This embodiment calculates the coordinate deviation between the actual gripping coordinates and the preset reference position. Before performing the subtraction operation, to eliminate the difference between pixel dimensions and physical length dimensions, this embodiment performs a projection mapping operation based on the pre-calibrated camera intrinsic parameter matrix and hand-eye calibration extrinsic parameter matrix to convert the actual gripping coordinates in the pixel coordinate system into physical space coordinates in the robot's base coordinate system. Subsequently, this embodiment performs a subtraction operation between the converted physical space coordinates and the preset reference position to obtain a three-dimensional deviation vector containing lateral deviation, longitudinal deviation, and angular deviation.

[0060] It should be noted that this embodiment determines whether the coordinate deviation value is greater than a preset deviation threshold. The preset deviation threshold is determined by obtaining the rated repeatability accuracy value of the robot end effector and the maximum measurement residual value of the sensor; calculating the algebraic sum of the squares of the rated repeatability accuracy value and the maximum measurement residual value, and calculating the arithmetic square root of the summation result to obtain the engineering tolerance boundary; multiplying the engineering tolerance boundary by a preset proportional adjustment coefficient to obtain the preset deviation threshold. The preset proportional adjustment coefficient is determined based on the Six Sigma quality management principle and the actual material extensibility tolerance of the die-cutting process. In this embodiment, the value is set to 1.2 to 1.5 to ensure that it covers the vast majority of normal distribution fluctuations in a statistical sense, while taking into account positioning accuracy and system anti-mistriking rate. Specifically, during the stable operation phase of the die-cutting machine, the distribution of coordinate deviation values ​​within a large number of normal production cycles is statistically analyzed, and their standard deviation is calculated. According to the Six Sigma principle, the engineering tolerance boundary usually covers ±6 times the standard deviation. Then, based on the extensibility experiment of a specific material, a maximum allowable deviation magnification factor that will not cause material deformation or damage is determined. The engineering tolerance boundary is combined with this maximum allowable magnification factor, and through experimental verification, a specific coefficient value is finally determined within the range of 1.2 to 1.5, so that more than 99.73% of normal fluctuations are allowed, while avoiding process damage.

[0061] Specifically, the coordinate deviation value is fed back to calculate the attitude compensation amount, and control mapping processing is performed based on the attitude compensation amount to obtain the adjusted grasping command, including:

[0062] The PID control algorithm is used to perform feedback calculations on the coordinate deviation value to obtain the attitude compensation amount.

[0063] The actual grasping coordinates are transformed using the attitude compensation amount to obtain the updated control target coordinates;

[0064] The updated control target coordinates are processed by machine code mapping to obtain the adjusted capture instruction.

[0065] In one implementation, this embodiment utilizes a PID control algorithm to perform feedback calculations on the coordinate deviation value to obtain the attitude compensation amount. To ensure the physical rigor of the control dimension, this embodiment employs three independent parallel PID controllers. Preset proportional coefficients, integral coefficients, and derivative coefficients are obtained; the lateral deviation, longitudinal deviation, and angular deviation are respectively input to the corresponding lateral PID controller, longitudinal PID controller, and angular PID controller. Each controller calculates its corresponding dimensionless control output; each control output is multiplied by a preset dynamic mapping constant to convert it into an attitude compensation amount containing lateral displacement compensation, longitudinal displacement compensation, and angular rotation compensation. The preset dynamic mapping constant is a linear conversion ratio between the dimensionless control amount and the end-effector physical displacement / angle, calibrated offline by establishing the inverse kinematics Jacobian matrix of the grasping mechanism and combining the gear reduction ratio of each axis servo motor.

[0066] It is worth noting that the proportional coefficient, integral coefficient, and derivative coefficient involved in the PID controller are objective parameters determined by applying a step signal during the no-load commissioning phase of the equipment using the critical proportional gain method, measuring the critical oscillation period and gain of the system, and then fine-tuning them offline based on experimental feedback.

[0067] In one implementation, this embodiment uses the attitude compensation amount to perform coordinate transformation on the actual grasping coordinates to obtain updated control target coordinates. To meet the dimension alignment requirements of matrix operations, this embodiment first performs a dimension increase process, appending a constant '-' to the end of the two-dimensional vector of the actual grasping coordinates with physical dimensions, transforming it into a homogeneous coordinate vector $[x,y,1]^T$. This embodiment constructs a rotation matrix based on the angle rotation compensation amount in the attitude compensation amount and a translation vector based on the displacement compensation amount, forming a 3x3 homogeneous transformation matrix. :

[0068]

[0069] In the above formula, Represents the homogeneous transformation matrix; Indicates angular rotation compensation; Indicates the lateral displacement compensation; This represents the longitudinal displacement compensation. In this embodiment, the homogeneous transformation matrix and the homogenized actual grasping coordinates are multiplied together to calculate the corrected three-dimensional physical coordinates, which are then used as the updated control target coordinates.

[0070] In one implementation, this embodiment performs machine code mapping processing on the updated control target coordinates to obtain the adjusted grasping instruction. This embodiment extracts the pulse equivalent conversion constant by retrieving the underlying communication protocol manual of the robot controller; it then multiplies the coordinate values ​​of each axis in the updated control target coordinates by the corresponding pulse equivalent conversion constant to convert them into target pulse values ​​for each axis servo motor, and encapsulates this to generate the adjusted grasping instruction. It is worth noting that if the coordinate deviation value is less than or equal to a preset deviation threshold, the actual grasping coordinates are directly used as the adjusted grasping instruction without feedback calculation or attitude compensation.

[0071] For example, the system converts pixel coordinates into physical coordinates using a hand-eye calibration matrix. The system identifies a lateral deviation of 4.5 mm and a longitudinal deviation of 4.7 mm between the actual coordinates and the reference position. Since the deviation exceeds the threshold set based on the Six Sigma principle, the system invokes the PID controller and calculates the attitude compensation amount using the Jacobian matrix mapping relationship. The system uses a homogeneous transformation matrix to transform the homogeneous coordinates, calculates the updated control target coordinates, and finally converts them into machine code instructions containing the target pulse sequence for output.

[0072] In step S14, the real-time beat signal of the die-cutting equipment is acquired, and the simulated path is compared with the real-time beat signal according to the adjusted grasping instruction to obtain the speed matching degree and speed fluctuation rate.

[0073] The process of comparing the adjusted capture command with the real-time beat signal to obtain the speed matching degree and speed fluctuation rate includes:

[0074] The path corresponding to the adjusted capture command is smoothed by using a preset Bézier curve model to obtain a dynamic simulation trajectory.

[0075] The equipment dynamics parameters are obtained, and the theoretical feeding speed and the actual speed under obstruction are calculated by combining the real-time cycle signal, the dynamic simulation trajectory and the equipment dynamics parameters.

[0076] Calculate the percentage difference between the theoretical feeding speed and the actual obstructed speed to obtain the speed matching degree, and calculate the variance of the actual obstructed speed within a preset window to obtain the speed fluctuation rate.

[0077] In one implementation, this embodiment utilizes a preset Bézier curve model to perform smooth interpolation processing on the path corresponding to the adjusted grabbing command. The starting physical coordinates point in the adjusted grabbing command is then extracted. Update control target coordinates ;calculate and The linear Euclidean distance between them. The extraction robot is in... The instantaneous motion tangent direction vector at a given point, extending along this direction by a length equal to one-third of the Euclidean distance of the straight line, determines the first intermediate control point. The extraction robot arm is in The reverse tangent vector at the point is extended by the same length to determine the second intermediate control point. Finally, Substituting the values ​​into a cubic Bessel polynomial, spatial parameterization interpolation is performed to generate a three-dimensional continuous curve, i.e., the dynamic simulation trajectory.

[0078] It should be noted that this embodiment obtains the equipment dynamic parameters. These parameters include the end-load mass of the robotic arm, the bearing rotational friction coefficient, and the damping constant of the servo driver. The method for determining these dynamic parameters is as follows: offline system identification is performed on the feeding mechanism; pseudo-random sequence pulses are applied, and the angular displacement data fed back by the encoder is monitored; the transfer function is fitted using the least squares method, and the physical parameter values ​​are calibrated.

[0079] In one implementation, this embodiment combines the real-time beat signal, the dynamic simulation trajectory, and the equipment dynamic parameters for calculation. This embodiment analyzes the real-time beat signal to obtain the total pulse duration of the current die-cutting cycle. Dividing the total arc length of the dynamic simulation trajectory by this total pulse duration yields the ideal average speed across the entire path, which is determined as the theoretical feeding speed. Subsequently, this embodiment constructs a second-order dynamic equation including a nonlinear friction torque compensation term, using the theoretical feeding speed as the input excitation, and calculates the end-effector instantaneous velocity sequence after considering mechanical inertia and resistance losses, which is then determined as the actual obstructed speed.

[0080] It is worth noting that this embodiment calculates the speed matching degree. The absolute value of the deviation between the theoretical feeding speed and the actual obstructed speed is calculated at each calculation step. The absolute value of the deviation is divided by the theoretical feeding speed to obtain the deviation ratio. This deviation ratio is then subtracted from a constant to determine the speed matching degree. This indicator quantifies the real-time tracking performance of the feeding action to the main frequency of the equipment.

[0081] In one implementation, this embodiment calculates the variance of the obstructed actual speed within a preset window. This embodiment uses the time span corresponding to a single stamping cycle of the die-cutting machine as the preset window. The obstructed actual speed sequence within this window is extracted, and its second-order dispersion relative to the window mean is calculated, outputting the speed fluctuation rate.

[0082] For example, when the system detects a deviation exceeding the 3.0 pixel threshold, it calls the PID parameters P=0.8, I=0.2, D=0.1 for decoupling feedback, outputs the lateral displacement compensation, and performs a homogeneous transformation to obtain the updated coordinates. The system generates a smooth Bezier trajectory based on the starting and updated points using the 1 / 3 distance method, and calculates the theoretical speed to be 100.0 mm / s using a real-time cycle of 5.0 seconds. Substituting the friction coefficient into the dynamic equation, the resisted speed is calculated to be 95.0 mm / s, resulting in a speed matching degree of 95% and a speed fluctuation rate of 0.3.

[0083] In step S15, the speed matching degree and the speed fluctuation rate are input into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount. The fuzzy adjustment amount is then parsed to obtain coordinated operation parameters, including:

[0084] The speed matching degree and the speed fluctuation rate are matched with a preset fuzzy rule base, and inference calculations are performed to obtain the fuzzy adjustment amount;

[0085] The fuzzy adjustment amount is defuzzified and calculated using the center of gravity method to obtain the speed adjustment amount and the valve opening adjustment amount.

[0086] The current state of the equipment is updated based on the speed adjustment and the valve opening adjustment to obtain coordinated operating parameters.

[0087] In one implementation, this embodiment extracts the previously calculated speed matching degree and speed fluctuation rate, and transforms them into a fuzzy input vector using a preset membership function. The preset membership function is a Gaussian membership function, which divides the input universe of discourse into three fuzzy subsets: low, medium, and high. The center value and standard deviation parameters of the Gaussian membership function are obtained by constructing a neural fuzzy inference network and iteratively calibrating using a backpropagation algorithm.

[0088] In this embodiment, the fuzzy input vector is input into a preset fuzzy rule base, and the Mamdani fuzzy inference algorithm is used to perform logical deduction to obtain the fuzzy adjustment amount. It is worth noting that the preset fuzzy rule base contains twenty-five if-then-form association mapping rules. For example, the fuzzy rule base includes the following specific control logic: if the speed matching degree is low and the speed fluctuation rate is high, then the output speed adjustment amount is positive and the valve opening adjustment amount is positive; if the speed matching degree is high and the speed fluctuation rate is low, then the output speed adjustment amount is zero and the valve opening adjustment amount is maintained. The construction process of the preset fuzzy rule base is as follows: extract the debugging record sequence from the historical operation database of the die-cutting machine under stable system operation and no material accumulation conditions; use the mean-shift clustering algorithm to divide the feature space of the debugging record sequence, extract the cluster centers to generate the above rules.

[0089] In one implementation, this embodiment uses the centroid method to defuzzify the fuzzy adjustment amount to obtain the speed adjustment amount and the valve opening adjustment amount. This embodiment extracts the membership envelope curve of the fuzzy adjustment amount in the physical output universe of discourse, and calculates the corresponding geometric centroid abscissa using the following formula:

[0090]

[0091] In the above formula, This represents the precise value obtained after defuzzification, namely the speed adjustment amount or the valve opening adjustment amount; This indicates that the values ​​of variables in the universe of discourse should be output. The envelope function represents the membership degree of the fuzzy adjustment quantity in the output universe.

[0092] It is important to note that this embodiment achieves a closed loop in physical transmission by combining the output speed adjustment and valve opening adjustment. Adjusting the opening of the tension regulating valve aims to change the tension level of the material during the conveying process, thereby changing the physical slip coefficient between the material and the conveying roller. By coordinating the feed motor speed and the physical slip coefficient, speed fluctuations caused by local tensile deformation or slippage of the material can be eliminated, thus achieving precise matching between the feeding rhythm and the main frequency of the equipment.

[0093] In one implementation, this embodiment updates the current state of the equipment based on the speed adjustment amount and the valve opening adjustment amount to obtain coordinated operating parameters. This embodiment reads the current reference speed value of the feed motor and the current reference opening value of the tension regulating valve via the bus; adds the speed adjustment amount to the current speed reference value, and adds the valve opening adjustment amount to the current opening reference value; then encapsulates both according to the underlying driver message protocol format to generate the coordinated operating parameters.

[0094] For example, the system extracts the current speed matching degree as 92.5% (low) and the speed fluctuation rate as 0.8% (high). The system matches the fuzzy rule base and performs inference operations, and the envelope synthesis yields the fuzzy adjustment amount. After performing the center of gravity method integral calculation, the calculated speed adjustment amount is +30.0 rpm, and the valve opening adjustment amount is +5.0%. The system reads the current speed of 1500.0 rpm and the opening of 60.0%, performs an addition operation, and encapsulates it to generate coordinated operating parameters containing 1530.0 rpm and 65.0%.

[0095] In step S16, the material flow rate of the production line is obtained, and time-series prediction processing is performed based on the coordinated operating parameters and the material flow rate to obtain the probability of interruption risk.

[0096] The process of performing time-series prediction based on the coordinated operating parameters and the material flow rate to obtain the probability of interruption risk includes:

[0097] The theoretical conveying rate is calculated based on the coordinated operating parameters, and the discrete deviation between the theoretical conveying rate and the material flow rate is calculated to obtain the flow uniformity index.

[0098] A time series is constructed by extracting multiple consecutive flow uniformity indices, and the probability of interruption risk is obtained by calculating the time series using an autoregressive moving average algorithm.

[0099] In one implementation, this embodiment uses a photoelectric rotary encoder deployed on the conveyor roller shaft of the die-cutting machine to collect angular velocity signals, which are then multiplied by the physical radius of the conveyor roller to calculate the material flow rate. This embodiment extracts the feed motor speed from the coordinated operating parameters, multiplies it by the gearbox transmission ratio and the physical circumference of the drive wheel, and performs a product operation using the physical slip coefficient determined by the opening degree of the tension regulating valve to calculate the theoretical conveying rate.

[0100] In one implementation, this embodiment calculates the discrete deviation between the theoretical conveying rate and the material flow rate to obtain a flow uniformity index. The absolute value of the theoretical conveying rate minus the material flow rate is calculated; this absolute value is divided by the theoretical conveying rate to obtain an instantaneous deviation ratio; and a constant is used to subtract this instantaneous deviation ratio to determine the flow uniformity index.

[0101] It is worth noting that this embodiment extracts multiple flow uniformity indices generated continuously within a preset time window to construct a time series. The preset time window is determined by extracting the median of the sample set of severe fluctuation durations before material interruption shutdown in historical operation records, and multiplying it by a preset safety redundancy constant. The preset safety redundancy constant is determined based on the ratio of the maximum peak to the average value of historical material interruption fluctuations, and is set to 1.5, thus eliminating the error of blindly applying experience across disciplines.

[0102] In one implementation, this embodiment uses the Autoregressive Moving Average (ARIMA) algorithm to calculate the probability of interruption risk on the time series. This embodiment performs a unit root test on the time series to determine the difference order, and uses the Akaike Information Criterion (AIC) to establish the number of autoregressive and moving average terms in the model; subsequently, it performs forward multi-step prediction operations, extracting the lowest scalar value in the predicted sequence as the extreme fluctuation point.

[0103] In this embodiment, the extreme fluctuation points are substituted into a pre-constructed probability mapping function for calculation, and the interruption risk probability is output. The interruption risk probability is obtained by calculating 1 plus the natural constant to the power of an exponent, and taking the reciprocal of the result. The exponent is the inverse of the result of multiplying the independent variable by the stretching scale parameter and adding the center offset parameter. It should be noted that the center offset parameter and the stretching scale parameter of this probability mapping function are fitted by constructing a log-likelihood function that assumes the true interruption labels follow a Bernoulli distribution. This embodiment uses the following log-likelihood function formula for target optimization:

[0104]

[0105] In the above formula, Represents the log-likelihood function; This represents the total number of historical samples; Indicates the first The true interruption label of each historical sample is a value of one when an interruption occurs and a value of zero when no interruption occurs. Indicates the first The extreme fluctuation points of each sample; This represents the predicted probability calculated by the probability mapping function, which internally includes the center offset parameter to be solved. With tensile dimensional parameters This embodiment utilizes the gradient descent optimization algorithm to iteratively solve for the parameter combination that maximizes the probability of historical observation data by maximizing the log-likelihood function, thus achieving objective fitting.

[0106] For example, the system calculates a theoretical conveying rate of 5.16 m / s, and the measured material flow rate is 5.00 m / s. The calculated dispersion ratio is 3.1%, and the flow uniformity index is 96.9%. The system extracts a time series constructed over the past 10 minutes and inputs it into an ARIMA model with an optimal order. It predicts an extreme fluctuation point of 85.0% within the next 5 minutes. Substituting this extreme fluctuation point into the probability mapping function optimized by maximum likelihood estimation, the final interruption risk probability is calculated to be 3.2%.

[0107] In step S17, the main and auxiliary channels are allocated traffic based on the interruption risk probability and the preset reserve logic to obtain a channel allocation instruction, including:

[0108] Obtain the material reserves in the buffer zone and the production demand rate of the downstream cycle;

[0109] The available time can be calculated based on the material reserves and the production demand rate.

[0110] When the probability of interruption risk exceeds a preset probability threshold and the sustainable time is less than or equal to a preset risk coverage window, the target total flow is determined according to the production demand rate.

[0111] Based on the target total flow, the flow is proportionally divided between the main and auxiliary channels to obtain the channel allocation instruction.

[0112] In one implementation, this embodiment uses a weighing sensor deployed at the bottom of the spare buffer silo to read the analog electrical signal of the weight in real time, and obtains the material reserve of the buffer after analog-to-digital conversion; at the same time, by reading the production schedule table of the programmable logic controller of the downstream die-cutting station, the scalar value of the material consumption per unit time is extracted and determined as the production demand rate of the downstream cycle.

[0113] In this embodiment, the material reserve is divided by the production demand rate, and an arithmetic division operation is performed to calculate the time required to maintain normal operation without the replenishment of new materials, which is then determined as the sustainable time.

[0114] It should be noted that this embodiment compares the interruption risk probability with a preset probability threshold, and also performs a logical judgment on the sustainable time and the preset risk coverage window. The preset probability threshold is determined by statistically analyzing the risk probability distribution of actual downtime accidents in historical production data and extracting the inflection point value of the probability density function as the preset probability threshold. The preset risk coverage window is determined by measuring the maximum time required for the auxiliary feeding channel to go from a cold start state to outputting a stable material flow, adding the constant communication delay of the physical network, and defining this total time as the preset risk coverage window.

[0115] When the interruption risk probability exceeds the preset probability threshold and the sustainable time is less than or equal to the preset risk coverage window, it is determined that the current buffer capacity is insufficient to withstand the impending interruption. In this embodiment, the current production demand rate is extracted and multiplied by a preset safety compensation coefficient. An arithmetic multiplication operation is then performed to calculate the required total material supply rate, which is determined as the target total flow rate. The preset safety compensation coefficient is obtained by extracting the highest flow loss ratio of the equipment under maximum resistance conditions in its history. To avoid excessive transport leading to material accumulation, this value is set to 1.15. It should be noted that if the interruption risk probability does not exceed the preset probability threshold, or the sustainable time is greater than the preset risk coverage window, the current channel allocation instruction remains unchanged, and the proportional splitting of the main and auxiliary channels is not initiated.

[0116] The process of proportionally dividing the target total traffic between the primary and secondary channels to obtain channel allocation instructions includes:

[0117] Obtain the real-time load rate of the auxiliary feeding channel, and use a PID control algorithm to calculate the required flow rate of the auxiliary channel based on the real-time load rate;

[0118] The difference between the target total flow and the auxiliary channel demand flow is calculated to obtain the main channel demand flow.

[0119] The required flow rates of the auxiliary channel and the main channel are converted into frequency conversion control signals to obtain channel allocation instructions.

[0120] In one implementation, this embodiment collects the actual operating current through the current transformer of the auxiliary feeding channel drive motor, divides the actual operating current by the rated full-load current, and calculates the real-time load rate. This embodiment uses a PID control algorithm to calculate the required flow rate of the auxiliary channel based on the real-time load rate. A pre-set safe upper limit for the target load rate of the auxiliary channel is extracted; the real-time load rate is subtracted from the safe upper limit to obtain a load difference percentage; the load difference percentage is input into the PID control algorithm, which sequentially performs proportional amplification, historical deviation integral accumulation, and deviation change rate differentiation operations to output a dimensionless flow rate adjustment coefficient. Subsequently, the rated maximum flow rate reference value of the auxiliary feeding channel motor is extracted, multiplied by the flow rate adjustment coefficient, and converted into a flow rate value with physical dimensions, which is determined as the required flow rate of the auxiliary channel.

[0121] It should be noted that when the PID control algorithm is used to calculate the auxiliary channel demand flow based on the real-time load rate, the proportional coefficient is 2.0, the integral coefficient is 0.5, and the derivative coefficient is 5.1. These coefficients are obtained by applying a step signal under the no-load condition of the auxiliary channel using the critical proportional method, and then measuring the critical oscillation period and gain of the system before offline tuning.

[0122] It is worth noting that the target load rate safety upper limit is determined by directly extracting the percentage of the maximum continuous operation load rate without overheating as indicated in the drive motor's manufacturer's specifications. In this embodiment, the target total flow rate is subtracted from the auxiliary channel's required flow rate, and an arithmetic subtraction operation is performed to obtain the remaining flow rate that needs to be borne by the main channel, which is then determined as the main channel's required flow rate.

[0123] In one implementation, this embodiment converts the required flow rates of the auxiliary channel and the main channel into frequency converter control signals to obtain a channel allocation command. The frequency-flow mapping curves of the main channel conveyor motor and the auxiliary channel conveyor motor are retrieved; the required flow rates of the main channel and the auxiliary channel are substituted into their respective mapping curves to calculate their corresponding target operating frequency values; the target operating frequency values ​​are encoded into a data frame containing the device address and frequency code according to the industrial Ethernet communication protocol, and the channel allocation command is output. The mapping curves are determined by offline measurement of material discharge data points corresponding to different frequency pulses and by fitting a polynomial using the least squares method.

[0124] For example, the system detects an interruption risk probability of 8.3% through the risk assessment module, which exceeds the preset probability threshold of 5.0%. The system obtains the material reserve in the buffer zone as 32.0 kg through the weighing sensor and reads the downstream production demand rate as 3.2 kg / s. The system calculates the sustainable time to be 10.0 seconds. Since 10.0 seconds is less than the preset risk coverage window of 15.0 seconds, i.e., the time required for the auxiliary channel to complete the speed-up, the splitting mechanism is triggered. The system multiplies the production demand rate of 3.2 kg / s by the safety compensation coefficient of 1.15, calculating the target total flow rate to be 3.68 kg / s.

[0125] Subsequently, the system collects auxiliary channel motor data, obtaining the current real-time load rate of 72.5%. This is then subtracted from the target load rate safety limit of 85.0%, yielding a load difference of 12.5%. This difference is input into the PID control algorithm, which outputs a dimensionless flow regulation coefficient of 0.4. The system extracts the rated maximum flow rate baseline value of 4.0 kg / s for the auxiliary channel, multiplies it by 0.4, and obtains a required flow rate of 1.60 kg / s for the auxiliary channel. Finally, the system subtracts 1.60 kg / s from the target total flow rate of 3.68 kg / s to obtain a required flow rate of 2.08 kg / s for the main channel. Utilizing the inverter frequency-flow mapping curve, the system converts 2.08 and 1.60 into data frames containing specific frequency parameters, which are then issued as channel allocation commands, ensuring a stable and precise supply of flow to the die-cutting station from both channels.

[0126] In summary, this invention discloses an intelligent loading and unloading method and system for die-cutting machines. This invention deeply integrates opto-electro-mechanical multi-source sensor data with material visual features, utilizing a spatial dimension broadcasting mechanism and residual networks for accurate regression deduction of grasping coordinates; it outputs attitude compensation quantities based on hand-eye calibration and decoupled PID algorithms and performs rigorous homogeneous coordinate transformation; it generates dynamic trajectories by combining Bessel smoothing interpolation and equipment dynamics models, and introduces fuzzy inference rules to coordinate the feeding speed and physical slip coefficient in real time to match the die-cutting cycle; it uses an autoregressive moving average algorithm and a probability mapping function optimized by maximum likelihood estimation to proactively assess interruption risks and accordingly performs dynamic proportional segmentation of the main and auxiliary feeding channels. This method and system completely establish a data closed loop from underlying physical dimensions, visual perception, attitude correction to flexible conveying, effectively overcoming the limitations of traditional fixed programs in handling dynamic system disturbances. It achieves a leap from passive response to forward-looking early warning of production interruption risks, comprehensively improving the positioning accuracy, system robustness, and intrinsic safety level of die-cutting processing under complex working conditions.

[0127] Reference Figure 2 The second embodiment of the present invention provides an intelligent loading and unloading system for a die-cutting machine, comprising:

[0128] The signal processing module is used to acquire the material thickness sampling signal, material tension sampling signal and ambient light intensity of the die-cutting machine, and perform preprocessing to obtain state fusion data;

[0129] The coordinate deduction module is used to acquire the regional image of the die-cutting material, input the regional image and the state fusion data into the pre-trained detection network for feature deduction, and obtain the actual grasping coordinates;

[0130] The attitude correction module is used to calculate the coordinate deviation between the actual grasping coordinates and the preset reference position. When the coordinate deviation is greater than the preset deviation threshold, the module performs feedback calculation on the coordinate deviation to obtain the attitude compensation amount. Based on the attitude compensation amount, the module performs control mapping processing to obtain the adjusted grasping command.

[0131] The matching evaluation module is used to acquire the real-time cycle signal of the die-cutting equipment, and perform simulated path comparison processing based on the adjusted capture command and the real-time cycle signal to obtain the speed matching degree and speed fluctuation rate.

[0132] The coordination and scheduling module is used to input the speed matching degree and speed fluctuation rate into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount, and to analyze the fuzzy adjustment amount to obtain coordinated operation parameters.

[0133] The risk prediction module is used to obtain the material flow rate of the production line, and perform time-series prediction processing based on the coordinated operating parameters and material flow rate to obtain the probability of interruption risk.

[0134] The emergency response module is used to calculate the flow allocation between the main and auxiliary channels based on the probability of interruption risk and the preset reserve logic, and to obtain the channel allocation instruction.

[0135] It should be noted that the intelligent loading and unloading system for a die-cutting machine provided in this embodiment of the invention is used to execute all the process steps of the intelligent loading and unloading method for a die-cutting machine in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0136] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0137] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An intelligent loading and unloading method for a die-cutting machine, characterized in that, include: The material thickness sampling signal, material tension sampling signal, and ambient light intensity of the die-cutting machine are acquired and preprocessed to obtain state fusion data; A region image of the die-cut material is acquired, and the region image and the state fusion data are input into a pre-trained detection network for feature inference to obtain the actual grasping coordinates. Calculate the coordinate deviation between the actual grasping coordinates and the preset reference position. If the coordinate deviation is greater than the preset deviation threshold, perform feedback calculation on the coordinate deviation to obtain the attitude compensation amount. Perform control mapping processing based on the attitude compensation amount to obtain the adjusted grasping command. The real-time cycle signal of the die-cutting equipment is acquired, and the simulated path is compared with the real-time cycle signal according to the adjusted grasping instruction to obtain the speed matching degree and speed fluctuation rate. The speed matching degree and the speed fluctuation rate are input into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount. The fuzzy adjustment amount is then parsed to obtain coordinated operation parameters. The material flow rate of the production line is obtained, and time-series prediction processing is performed based on the coordinated operating parameters and the material flow rate to obtain the probability of interruption risk. Based on the interruption risk probability and the preset reserve logic, the main and auxiliary channels are allocated traffic to obtain the channel allocation instruction.

2. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The process involves acquiring material thickness sampling signals, material tension sampling signals, and ambient light intensity from the die-cutting machine, and preprocessing them to obtain state fusion data, including: Acquire material thickness sampling signal, material tension sampling signal and ambient light intensity of the die-cutting machine, and calculate signal gain compensation value based on the ambient light intensity using a preset lighting model; The material thickness sampling signal and the material tension sampling signal are offset-corrected using the signal gain compensation value to obtain calibrated thickness data and tension data. Kalman filtering iterative calculations are performed on the calibrated thickness data and the tension data to obtain state fusion data.

3. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The step of inputting the region image and the state fusion data into a pre-trained detection network for feature inference to obtain the actual grasping coordinates includes: The Canny edge detection algorithm is used to process the image of the region to obtain hard edge features; The image of the region is processed using the gray-level co-occurrence matrix algorithm to obtain texture features; The hard edge features, texture features, and state fusion data are subjected to tensor concatenation to obtain a comprehensive feature vector. The comprehensive feature vector is input into a pre-trained detection network for regression analysis to obtain the actual capture coordinates.

4. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The process of feeding back the coordinate deviation value to obtain the attitude compensation amount, and then performing control mapping processing based on the attitude compensation amount to obtain the adjusted capture command includes: The PID control algorithm is used to perform feedback calculations on the coordinate deviation value to obtain the attitude compensation amount. The actual grasping coordinates are transformed using the attitude compensation amount to obtain the updated control target coordinates; The updated control target coordinates are processed by machine code mapping to obtain the adjusted capture instruction.

5. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The step of performing a simulated path comparison process based on the adjusted capture command and the real-time beat signal to obtain the speed matching degree and speed fluctuation rate includes: The path corresponding to the adjusted capture command is smoothed by using a preset Bézier curve model to obtain a dynamic simulation trajectory. The equipment dynamics parameters are obtained, and the theoretical feeding speed and the actual speed under obstruction are calculated by combining the real-time cycle signal, the dynamic simulation trajectory and the equipment dynamics parameters. Calculate the percentage difference between the theoretical feeding speed and the actual obstructed speed to obtain the speed matching degree, and calculate the variance of the actual obstructed speed within a preset window to obtain the speed fluctuation rate.

6. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The process involves inputting the speed matching degree and the speed fluctuation rate into a preset fuzzy rule base for inference calculations to obtain a fuzzy adjustment amount, and then parsing the fuzzy adjustment amount to obtain coordinated operation parameters, including: The speed matching degree and the speed fluctuation rate are matched with a preset fuzzy rule base, and inference calculations are performed to obtain the fuzzy adjustment amount; The fuzzy adjustment amount is defuzzified and calculated using the center of gravity method to obtain the speed adjustment amount and the valve opening adjustment amount. The current state of the equipment is updated based on the speed adjustment and the valve opening adjustment to obtain coordinated operating parameters.

7. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The step of performing time-series prediction processing based on the coordinated operating parameters and the material flow rate to obtain the interruption risk probability includes: The theoretical conveying rate is calculated based on the coordinated operating parameters, and the discrete deviation between the theoretical conveying rate and the material flow rate is calculated to obtain the flow uniformity index. A time series is constructed by extracting multiple consecutive flow uniformity indices, and the probability of interruption risk is obtained by calculating the time series using an autoregressive moving average algorithm.

8. The intelligent loading and unloading method for a die-cutting machine according to claim 1, characterized in that, The step of calculating the traffic allocation for the primary and secondary channels based on the interruption risk probability and preset reserve logic to obtain channel allocation instructions includes: Obtain the material reserves in the buffer zone and the production demand rate of the downstream cycle; The available time can be calculated based on the material reserves and the production demand rate. When the probability of interruption risk exceeds a preset probability threshold and the sustainable time is less than or equal to a preset risk coverage window, the target total flow is determined according to the production demand rate. Based on the target total flow, the flow is proportionally divided between the main and auxiliary channels to obtain the channel allocation instruction.

9. The intelligent loading and unloading method for a die-cutting machine according to claim 8, characterized in that, The step of proportionally dividing the total traffic between the primary and secondary channels based on the target total traffic to obtain channel allocation instructions includes: Obtain the real-time load rate of the auxiliary feeding channel, and use a PID control algorithm to calculate the required flow rate of the auxiliary channel based on the real-time load rate; The difference between the target total flow and the auxiliary channel demand flow is calculated to obtain the main channel demand flow. The required flow rates of the auxiliary channel and the main channel are converted into frequency conversion control signals to obtain channel allocation instructions.

10. An intelligent loading and unloading system for a die-cutting machine, characterized in that, include: The signal processing module is used to acquire the material thickness sampling signal, material tension sampling signal and ambient light intensity of the die-cutting machine, and perform preprocessing to obtain state fusion data; The coordinate deduction module is used to acquire the regional image of the die-cutting material, input the regional image and the state fusion data into the pre-trained detection network for feature deduction, and obtain the actual grasping coordinates; The attitude correction module is used to calculate the coordinate deviation between the actual grasping coordinates and the preset reference position. When the coordinate deviation is greater than the preset deviation threshold, the module performs feedback calculation on the coordinate deviation to obtain the attitude compensation amount. Based on the attitude compensation amount, the module performs control mapping processing to obtain the adjusted grasping command. The matching evaluation module is used to acquire the real-time cycle signal of the die-cutting equipment, and perform simulated path comparison processing based on the adjusted capture command and the real-time cycle signal to obtain the speed matching degree and speed fluctuation rate. The coordination and scheduling module is used to input the speed matching degree and speed fluctuation rate into a preset fuzzy rule base for inference calculation to obtain fuzzy adjustment amount, and to analyze the fuzzy adjustment amount to obtain coordinated operation parameters. The risk prediction module is used to obtain the material flow rate of the production line, and perform time-series prediction processing based on the coordinated operating parameters and material flow rate to obtain the probability of interruption risk. The emergency response module is used to calculate the flow allocation between the main and auxiliary channels based on the probability of interruption risk and the preset reserve logic, and to obtain the channel allocation instructions.

Images