A cutting control method and system for an automatic cutting and foaming machine

Abstract

The present application relates to the technical field of automatic cutting equipment control, and discloses a cutting control method and system for an automatic cutting bubble press, the method comprising collecting resistance data and material surface signals by a sensor array, determining material hardness and toughness through sensor fusion and neural network after synchronous denoising processing, calculating depth adjustment increment in combination with cutter parameters when exceeding the limit, optimizing cutting speed in combination with thermal and cutting force parameters through synchronous simulation, forming a comprehensive scheme, feeding back evaluation accuracy, optimizing physical properties when not meeting the standard, and iteratively determining final parameters to realize stable cutting. The method can adapt to material characteristic fluctuations in real time, suppress thermal and cutting force abnormalities, improve cutting stability and accuracy, reduce waste rate and cutter wear, and ensure consistency and production efficiency of complex material processing.

Description

Technical Field

[0001] This invention relates to the field of automatic cutting equipment control technology, and in particular to a cutting control method and system for an automatic cutting bubble press machine. Background Technology

[0002] In industrial sectors such as paper processing and packaging manufacturing, automatic cutting and bubbling machines are core equipment for ensuring production efficiency and product quality. Their cutting precision directly affects material utilization and finished product quality. With the improvement of industrial automation, industrial control software is increasingly widely used in cutting equipment, providing basic control logic for equipment operation. However, challenges still exist when dealing with complex changes in material properties.

[0003] In existing technologies, the cutting control of automatic cutting and bubbling machines largely relies on preset parameters or simple feedback adjustments. The core logic of the industrial control software focuses on the execution of fixed processes, lacking the ability to perceive and dynamically adapt to the real-time characteristics of materials. Different batches and types of paper exhibit significant differences in physical properties such as hardness and toughness. Furthermore, resistance fluctuations may occur during processing due to uneven fiber distribution and changes in surface texture. Existing industrial control software cannot accurately capture these dynamic changes and can only drive the cutter according to fixed parameters.

[0004] Furthermore, the parameter configuration of industrial control software lacks a linkage mechanism with material properties. When the material hardness and toughness exceed the preset range, the tool depth and cutting speed cannot be adjusted in time, easily leading to abnormal cutting resistance and causing problems such as burrs, material tearing, and tool wear. This limitation is particularly prominent in multi-variety, small-batch production scenarios, which not only reduces production stability but also increases scrap rates and equipment maintenance costs, thus restricting the level of intelligence of cutting equipment.

[0005] In existing technologies, industrial control software has failed to achieve real-time linkage between material properties and cutting parameters, and lacks a precise identification and adaptive adjustment mechanism for fluctuations in material hardness, toughness, and resistance. This results in poor stability and insufficient precision in the cutting process, failing to meet the dual requirements of cutting quality and production efficiency in complex material processing scenarios. Summary of the Invention

[0006] This invention provides a cutting control method and system for an automatic cutting and bubbling machine, which solves the problems in the prior art where industrial control software fails to achieve real-time linkage between material properties and cutting parameters, and lacks a precise identification and adaptive adjustment mechanism for fluctuations in material hardness, toughness, and resistance, resulting in poor stability and insufficient precision in the cutting process. This invention achieves real-time perception and dynamic adaptation of material properties during paper processing, and improves the stability and cutting accuracy of the cutting process through the coordinated optimization of tool depth and cutting speed, reducing scrap rate and tool wear, and meeting the production needs of complex material processing scenarios.

[0007] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a cutting control method for an automatic cutting and bubbling machine, comprising: Collect paper processing resistance data and perform noise reduction processing to generate a resistance fluctuation sequence; Based on the resistance fluctuation sequence, the elastic modulus characterization value and energy dissipation ratio are obtained and input into a preset neural network model to output the current hardness value and current toughness value of the material. If the current hardness value or the current toughness value exceeds the preset safety threshold, the tool geometry parameters are obtained and the influence weight of each parameter on the cutting resistance is determined. The current resistance deviation is calculated, and an objective function is constructed and solved based on the current resistance deviation and the influence weight to obtain the tool depth adjustment increment. The current cutting speed is obtained, and simulation prediction is performed based on the adjustment increment. If the simulation result exceeds the preset processing stability boundary, the current cutting speed is corrected, a target cutting speed is generated and associated with the adjustment increment to form a comprehensive adjustment scheme. The integrated adjustment scheme is analyzed to generate timed tool control commands, which are then loaded into the servo control loop. The actual cutting resistance data stream is collected, the deviation from the theoretical resistance value is calculated, and a resistance deviation sequence is constructed. Contact state features are extracted based on the resistance deviation sequence, and cutting accuracy indicators are calculated based on the contact state features. If the cutting accuracy index is lower than the preset cutting accuracy qualified threshold, then the resistance change characteristics are extracted from the resistance deviation sequence. Based on the resistance change characteristics, the current physical properties are corrected in combination with the pre-acquired material thermal conductivity characteristics to generate optimized physical property indexes. Based on the optimized physical property index, the core features of the material are extracted and input into the preset cutting force prediction model to obtain the theoretical maximum cutting force. If the theoretical maximum cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization.

[0008] Secondly, the present invention provides a cutting control system for an automatic cutting and bubbling machine, comprising: The data acquisition and processing module is used to collect paper processing resistance data and perform noise reduction processing to generate a resistance fluctuation sequence. The material property determination module is used to obtain the elastic modulus characterization value and energy dissipation ratio based on the resistance fluctuation sequence, input them into a preset neural network model, and output the current hardness value and current toughness value of the material. The tool depth adjustment module is used to obtain tool geometry parameters and determine the influence weight of each parameter on cutting resistance if the current hardness value or the current toughness value exceeds a preset safety threshold, calculate the current resistance deviation, construct an objective function based on the current resistance deviation and the influence weight, and solve it to obtain the tool depth adjustment increment. The speed coordination adjustment module is used to obtain the current cutting speed, perform simulation prediction based on the adjustment increment, and if the simulation result exceeds the preset processing stability boundary, the current cutting speed is corrected, a target cutting speed is generated and associated with the adjustment increment to form a comprehensive adjustment scheme. The instruction execution feedback module is used to parse the comprehensive adjustment scheme to generate time-sequential tool control instructions, load them into the servo control loop, collect the actual cutting resistance data stream, calculate the deviation from the theoretical resistance value, and construct a resistance deviation sequence. The cutting accuracy evaluation module is used to extract contact state features based on the resistance deviation sequence and calculate the cutting accuracy index based on the contact state features. The physical property optimization module is used to extract resistance change features from the resistance deviation sequence if the cutting accuracy index is lower than the preset cutting accuracy qualification threshold, and to modify the current physical properties based on the resistance change features and the pre-acquired material thermal conductivity characteristics to generate optimized physical property index. The parameter iteration configuration module is used to extract the core features of the material based on the optimized physical property index and input them into the preset cutting force prediction model to obtain the theoretical maximum cutting force. If the theoretical maximum cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects resistance data and material surface signals through a sensor array, extracts parameters such as elastic modulus through a fusion algorithm, and determines the material hardness and toughness through a neural network, thereby realizing real-time perception of material properties, accurately adapting to different paper processing needs, avoiding cutting defects caused by material variations, and improving processing adaptability.

[0010] (2) This invention calculates the tool depth adjustment increment through cross-correlation analysis and regression model, and optimizes the cutting speed by combining thermal simulation to form a comprehensive adjustment scheme, so as to achieve coordinated adaptation of depth and speed, suppress cutting force fluctuation and thermal damage, ensure the stability of the cutting process, and reduce tool wear and scrap rate.

[0011] (3) The present invention evaluates the cutting accuracy through the resistance deviation sequence, optimizes the physical property index when the standard is not met, iteratively adjusts the tool parameters, and continuously corrects the deviation through closed-loop feedback to improve the stability of cutting accuracy and meet the quality requirements of high-requirement processing scenarios. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of a cutting control method for an automatic cutting and bubbling machine provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a cutting control system for an automatic cutting and bubbling machine provided in the second embodiment of the present invention. Detailed Implementation

[0013] 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.

[0014] Reference Figure 1 The first embodiment of the present invention provides a cutting control method for an automatic cutting and bubbling machine, comprising the following steps: S11: Collect paper processing resistance data and perform noise reduction processing to generate a resistance fluctuation sequence; S12, Based on the resistance fluctuation sequence, obtain the elastic modulus characterization value and energy dissipation ratio and input them into a preset neural network model, and output the current hardness value and current toughness value of the material. S13, if the current hardness value or the current toughness value exceeds the preset safety threshold, then obtain the tool geometry parameters and determine the influence weight of each parameter on the cutting resistance, calculate the current resistance deviation, construct an objective function based on the current resistance deviation and the influence weight, and solve it to obtain the tool depth adjustment increment; S14, obtain the current cutting speed, perform simulation prediction based on the adjustment increment, if the simulation result exceeds the preset processing stability boundary, then correct the current cutting speed, generate the target cutting speed and associate it with the adjustment increment to form a comprehensive adjustment scheme; S15, parse the comprehensive adjustment scheme to generate timed tool control commands, load them into the servo control loop, collect the actual cutting resistance data stream, calculate the deviation from the theoretical resistance value, and construct a resistance deviation sequence; S16, Extract contact state features based on the resistance deviation sequence, and calculate cutting accuracy index based on the contact state features; S17. If the cutting accuracy index is lower than the preset cutting accuracy qualified threshold, then the resistance change characteristics are extracted from the resistance deviation sequence. Based on the resistance change characteristics, the current physical properties are corrected in combination with the pre-acquired material thermal conductivity characteristics to generate optimized physical property index. S18. Based on the optimized physical property index, extract the core features of the material and input them into the preset cutting force prediction model to obtain the theoretical maximum cutting force. If the theoretical maximum cutting force is less than the instantaneous load, determine the final tool and cutting parameter configuration through adaptive control iterative optimization.

[0015] In step S11, the process of collecting paper processing resistance data and performing noise reduction to generate a resistance fluctuation sequence includes: The servo motor torque sensor in the control sensor array collects the friction torque data of the paper running, and the fiber texture signal on the material surface is collected by the high-speed optical sensor. Low-pass filtering is performed on the friction torque data and the fiber texture signal respectively to remove noise, and the time delay is calibrated by cross-correlation analysis to obtain a synchronous digital signal stream; Surface signal components are extracted from the synchronous digital signal stream, and the standard deviation is calculated to obtain roughness feature values. These roughness feature values ​​are then used as initial physical property indicators. Based on the roughness feature value, the filtering window is adaptively adjusted to filter the resistance data component in the synchronous digital signal stream, generating a resistance fluctuation sequence.

[0016] It should be noted that the servo motor torque sensor has a sampling frequency set to 100Hz, with a sampling range covering 0.1–5 N·m, which can adapt to the resistance monitoring needs of paper of different thicknesses and is used to acquire friction torque data during paper movement in real time. The high-speed optical sensor has a scanning resolution of 500 dpi and a sampling frame rate of 200 frames / second, which can clearly capture the texture features of the paper surface, such as fiber arrangement direction, micro-undulations, and coating uniformity, and outputs a digital signal with a grayscale value range of 0–255, which is used to accurately capture the fiber texture signal of the paper surface.

[0017] Next, when performing low-pass filtering on the friction torque data and fiber texture signal, the filter cutoff frequency was set to 10Hz. This frequency was determined based on the main distribution range of electromagnetic interference in the industrial environment, which can effectively filter out high-frequency interference components such as motor operation and circuit noise. After filtering, cross-correlation analysis was performed on the two signals to calculate the time offset corresponding to the peak value of the cross-correlation coefficient of the two signals. The original signal deviation of 3 to 8 milliseconds was calibrated to within ±0.1 milliseconds, completing the precise calibration of the time delay and ensuring that the resistance data and surface texture signal correspond precisely in the same time dimension, forming a synchronous digital signal stream.

[0018] After extracting the surface signal component from the synchronous digital signal stream, the data of this signal component is converted to grayscale to construct a discrete sequence. The standard deviation of this sequence is calculated to obtain the roughness characteristic value. The roughness characteristic values ​​of different types of paper have typical ranges. The roughness characteristic value of offset paper is usually between 0.014 and 0.021. Coated paper, due to its smooth surface coating, has a lower characteristic value, approximately 0.008 to 0.012. Corrugated base paper has a higher characteristic value, reaching 0.025 to 0.038. This characteristic value directly reflects the smoothness of the paper surface and the degree of fiber texture.

[0019] When adaptively adjusting the filtering window based on roughness feature values, a segmented adjustment strategy is adopted. This strategy is based on the strong correlation between paper surface roughness and the noise characteristics of the resistance signal. When the paper surface is smooth, i.e., the roughness feature value is less than 0.010, the noise in the resistance signal is mainly low-frequency interference. A wide, smooth window of 20 sampling points is used, which can filter out noise while preserving the main trend of resistance change. When the paper surface has slight texture undulations, i.e., the feature value is between 0.010 and 0.018, high-frequency noise and effective signal intertwine in the resistance signal. The window is adjusted to 10 sampling points to balance noise suppression and signal fidelity. When the paper surface is rough, i.e., the feature value is greater than 0.018, the surface texture will cause high-frequency fluctuation noise. If the window is too wide, it will misjudge the actual resistance abrupt change as noise. Therefore, the window is reduced to 5 sampling points and high-frequency suppression is strengthened to ensure accurate identification of the actual resistance changes caused by local fiber agglomeration, uneven coating, etc.

[0020] After this adaptive filtering process, the smoothness of the dynamic resistance variation value is significantly improved. During normal operation, the dynamic resistance usually fluctuates in the range of 0.25 to 0.38 N·m. When encountering local fiber agglomeration or uneven coating, the resistance will instantly rise to above 0.52 N·m. The resistance fluctuation sequence generated accordingly is updated every 50 milliseconds, recording the peak and valley values ​​and the slope of change in the most recent 1.5 seconds, providing accurate data support for subsequent material property analysis.

[0021] In step S12, the step of obtaining the elastic modulus characterization value and energy dissipation ratio based on the resistance fluctuation sequence and inputting them into a preset neural network model, and outputting the current material hardness value and current toughness value, includes: The resistance wave sequence is decomposed by multi-scale wavelet transform to separate signal components including high-frequency instantaneous impact components and low-frequency fiber deformation rate components. A multidimensional coupling matrix is ​​constructed with time windows as rows and the signal components as columns. The elastic modulus characterization value and plastic rheological threshold in the multidimensional coupling matrix are extracted by singular value decomposition. The material deformation equation is constructed using the elastic modulus characterization value and the plastic rheology threshold, and the energy dissipation ratio is calculated. The elastic modulus value and energy dissipation ratio are input into a preset radial basis function neural network model, which outputs the current hardness and toughness values ​​of the material.

[0022] It should be noted that when decomposing the resistance fluctuation sequence using multi-scale wavelet transform, the db4 wavelet basis function was selected, and the number of decomposition layers was set to 5. This parameter combination was verified by 100 sets of test data from different paper types, and it can optimally separate the signal components, ultimately obtaining two types of signal components: high-frequency transient impact components and low-frequency fiber deformation rate components. Among them, the high-frequency transient impact components are concentrated in the decomposition layers d1 to d3, mainly capturing sharp pulses caused by local abrupt changes on the paper surface, such as the instantaneous resistance jump caused by fiber agglomeration, coating peeling, or tiny impurities, with a typical amplitude range between 0.08 and 0.22 N·m; the low-frequency fiber deformation rate components are concentrated in the layers d6 to d8, reflecting the overall tensile deformation rate of the paper and demonstrating a continuous elastic deformation trend, with an amplitude stable in the range of 0.015 to 0.038 N·m / ms.

[0023] When constructing a multidimensional coupling matrix with time windows as rows and signal components as columns, the time window length is set to 100 sampling points. This length is determined based on the sampling frequency and the fluctuation period of paper cutting resistance. 100 sampling points correspond to a duration of 1 second, which can not only fully cover the fluctuation process of single cutting resistance, but also avoid signal lag or information redundancy caused by excessively long windows. The number of matrix columns includes 3 levels of high-frequency components and 3 levels of low-frequency components, forming a 100×6 coupling matrix, which comprehensively reflects the correlation between resistance dynamics and material response.

[0024] Then, when the coupling matrix is ​​reduced in dimensionality using singular value decomposition, the eigenvectors corresponding to the first three dominant singular values ​​are extracted to obtain the elastic modulus and plastic rheological threshold. These two parameters differ significantly among different types of paper. The elastic modulus of ordinary offset paper typically falls in the range of 18–24 MPa, and the plastic rheological threshold is around 0.42–0.51 N·m. High-gloss coated paper has a higher elastic modulus of 28–35 MPa and a lower plastic rheological threshold of 0.31–0.38 N·m, indicating that it is more prone to entering the irreversible deformation stage. Corrugated base paper has a lower elastic modulus of 12–16 MPa and a plastic rheological threshold of 0.55–0.63 N·m, exhibiting relatively stronger toughness.

[0025] When constructing the material deformation equation using the elastic modulus and the plastic rheological threshold, a linear elastic-plastic constitutive model is employed. This equation characterizes the relationship between stress and strain. The strain energy density is derived by analyzing the change in resistance during cutting, and then elastic strain energy and plastic strain energy are separated. Elastic strain energy is calculated from the elastic modulus and elastic strain, while plastic strain energy is determined by the plastic rheological threshold and plastic strain. The energy dissipation ratio is the ratio of plastic strain energy to total strain energy, directly quantifying the proportion of energy involved in irreversible deformation of paper under cutting action. For example, when the total strain energy generated by cutting resistance is 100 mJ, of which 18 mJ is plastic strain energy, the energy dissipation ratio is 0.18, indicating that 18% of the energy is converted into irreversible plastic deformation, and the remaining 82% is elastic strain energy, which can be recovered after unloading.

[0026] When the energy dissipation ratio is below 0.18, the paper is mainly in the elastic stage, suitable for maintaining the original cutting speed; when the ratio is between 0.18 and 0.32, the paper enters the elastoplastic stage, and resistance changes need to be monitored; when the ratio exceeds 0.32, it indicates that the material has undergone significant plastic flow, and there is a risk of breakage or tearing ahead. This classification standard was established by conducting cutting experiments on 10 common types of paper, collecting energy dissipation data at different deformation stages, and combining the results with statistical analysis of cut quality inspection. The experiment found that when the ratio is below 0.18, 99% of the samples have no traces of plastic deformation and the edges are smooth; when the ratio is between 0.18 and 0.32, about 75% of the samples have slight plastic deformation, but no risk of tearing, and resistance fluctuations need to be dynamically monitored; when the ratio exceeds 0.32, 88% of the samples show obvious plastic flow, and 30% of the samples are accompanied by edge tearing or delamination, therefore this is used as the risk threshold.

[0027] The radial basis function neural network model uses two features in its input layer: the elastic modulus and the energy dissipation ratio. The hidden layer has 40 nodes, and the activation function is a Gaussian kernel. The output layer displays the current hardness and toughness values. During training and optimization, the training dataset was collected from multiple centers, covering 10 common types of paper. For each paper type, 20 sets of data were collected, totaling 200 samples. Each set of samples included the elastic modulus, energy dissipation ratio, and corresponding measured values ​​of actual hardness and toughness. The training process consisted of three stages: the first stage initialized the model parameters using stochastic gradient descent, iterating 500 times; the second stage optimized the Gaussian kernel parameters and regularization coefficients through cross-validation, using 5-fold cross-validation, dividing the dataset into 5 groups, alternating between 4 groups as the training set and 1 group as the validation set, ultimately selecting the parameter combination with the smallest validation set error; the third stage fine-tuned the model, specifically correcting the prediction bias for each paper type to ensure that the prediction error for different paper types was controlled within ±3%. The model outputs hardness values ​​in Shore D and toughness values ​​in J / m². Standard double-sided coated paper has a hardness value of approximately 62–71 Shore D and a toughness value of 0.84–0.92 J / m². Paper with uneven coating may have a hardness value that drops sharply to 54–59 Shore D, and a corresponding decrease in toughness value to 0.61–0.73 J / m². Corrugated base paper has a hardness value of 58–65 Shore D and a toughness value that can reach 1.05–1.18 J / m². It should be further explained that the actual hardness and toughness measured values ​​in the training dataset of this radial basis function neural network model were obtained as follows: First, for each type of paper, according to GB / T 2411-2008 "Determination of indentation hardness (Shore hardness) of plastics and hard rubber using a durometer", 10 test points were evenly selected on the paper surface using a Shore D durometer, and the average value was taken as the measured hardness value of the sample. Second, according to GB / T 1040.1-2018 "Determination of tensile properties of plastics - Part 1: General rules", a universal testing machine was used to conduct a tensile test on the paper sample at a tensile speed of 10 mm / min until fracture. The stress-strain curve was recorded, and the fracture energy was calculated by the area under the integral curve. Then, the value was divided by the cross-sectional area of ​​the sample to obtain the measured toughness value in J / m². All measured values ​​of samples were completed in a constant temperature and humidity laboratory environment (temperature 23±2℃, relative humidity 50±5%) to ensure data consistency and repeatability.

[0028] In step S13, if the current hardness value or the current toughness value exceeds a preset safety threshold, the tool geometry parameters are obtained and the influence weight of each parameter on the cutting resistance is determined. The current resistance deviation is calculated, and an objective function is constructed and solved based on the current resistance deviation and the influence weights to obtain the tool depth adjustment increment, including: If the current hardness value exceeds the preset hardness safety threshold, or the current toughness value exceeds the preset toughness safety threshold, then extract the tool geometry parameters including the cutting edge angle, rake angle size, and cutting edge radius. A cross-correlation analysis of the resistance fluctuation sequence and the tool geometry parameters is performed using an adaptive length sliding window to generate a time-varying correlation feature vector. The feature vector is input into a preset support vector regression model, and a sensitivity coefficient matrix is ​​output. The influence weight of each parameter on the cutting resistance is quantified based on the sensitivity coefficient matrix. Based on the resistance fluctuation sequence and the preset theoretical resistance, the resistance deviation is calculated. Combined with the influence weight, an adaptive weighted objective function is constructed, and the adjustment increment of the tool depth is obtained by solving it.

[0029] It should be noted that the preset hardness safety threshold and toughness safety threshold are determined based on the processing adaptation range of common paper. The hardness safety threshold is set to 55-75 Shore D, and the toughness safety threshold is set to 0.6-1.2 J / m². When the current hardness value is lower than 55 Shore D or higher than 75 Shore D, or the current toughness value is lower than 0.6 J / m² or higher than 1.2 J / m², it is determined that the safety threshold has been exceeded, and the tool depth adjustment process is triggered.

[0030] The extracted tool geometry parameters specifically include the cutting edge angle, rake angle, cutting edge radius, and cutting edge width. The cutting edge angle is set to three selectable specifications: 18°, 22°, and 26°. The rake angle is adjustable from 5° to 15°. The cutting edge radius is 0.01 to 0.05 mm, and the cutting edge width is 0.02 to 0.08 mm. These parameters are pre-configured according to the processing requirements of different paper types and are used as fixed input parameters in the analysis during the adjustment process.

[0031] It is worth noting that when performing cross-correlation analysis on the resistance fluctuation sequence and tool geometry parameters using an adaptive-length sliding window, the window length is dynamically adjusted according to the sampling frequency. The window length is set to 80 sampling points at a sampling frequency of 100Hz and 120 sampling points at a sampling frequency of 200Hz, ensuring that the window can capture the complete resistance fluctuation cycle. The cross-correlation analysis generates a time-varying correlation feature vector by calculating the correlation coefficient between the two sets of data point by point. This vector has 10 dimensions and includes five peak correlation intensities and five corresponding hysteresis shifts at different lag times. For example, for a tool with a cutting edge angle of 18°, when the paper experiences a local abrupt change in hardness, the cross-correlation peak appears at lag 3–5 sampling points, with a peak intensity of 0.76–0.89, indicating that the resistance response is highly sensitive to geometry.

[0032] The support vector regression model uses a radial basis function (RBF) as its kernel function, with a penalty coefficient C set to 10 and a gamma parameter set to 0.1. The RBF was chosen because it effectively handles nonlinear feature mappings and adapts to the complex relationship between resistance fluctuations and tool parameters. The penalty coefficient C was determined through 5-fold cross-validation from four candidate values: 5, 10, 15, and 20. This value balances the model's fitting and generalization abilities, avoiding overfitting or underfitting. The gamma parameter was also optimized through cross-validation. This value determines the distribution density of samples in the high-dimensional feature space, ensuring the model remains sensitive to the feature differences of different paper types while avoiding excessive response to noise. Testing showed that this parameter combination achieved a prediction accuracy of over 94%. The sensitivity coefficient matrix output by the model is 4×2 dimensional, with rows corresponding to the four tool geometry parameters and columns corresponding to the direction of hardness and toughness changes. The element values ​​range from 0 to 1; larger values ​​indicate a more significant impact of the parameter on cutting resistance. For example, when processing high-gloss coated paper, the sensitivity coefficient of the blade corner radius is as high as 0.42, the sensitivity of the front angle is 0.31, and the sensitivity of the blade width is relatively low at only 0.14; when processing corrugated base paper, the sensitivity coefficient of the blade angle is 0.45, which is the main influencing factor.

[0033] It is worth noting that the true values ​​of the "sensitivity coefficient matrix" required for training this support vector regression model were obtained through orthogonal cutting experiments. The specific experimental design is as follows: four factors were selected: cutting edge angle (18°, 22°, 26°), rake angle (5°, 10°, 15°), cutting edge radius (0.01mm, 0.03mm, 0.05mm), and cutting edge width (0.02mm, 0.05mm, 0.08mm). Each factor had three levels, calculated according to L9(3)... 4 Nine cutting experiments were arranged using an orthogonal array. Each experiment was repeated five times on the same paper (standard double-sided offset paper as an example). The resistance fluctuation sequence during the cutting process was recorded using a high-precision force gauge. The rate of change of resistance relative to the baseline parameters (cutting edge angle 22°, rake angle 10°, fillet radius 0.03mm, cutting edge width 0.05mm) was calculated for each parameter combination. The partial derivative of the rate of change of resistance with respect to each parameter was used to obtain the sensitivity coefficient for that parameter combination. The sensitivity coefficients of all nine experiments constituted the training output labels, and the corresponding input feature vectors (time-varying correlation feature vectors obtained from cross-correlation analysis) were extracted from the resistance fluctuation sequence. Through this supervised learning process, a support vector regression model that can predict the sensitivity coefficients under any combination of tool parameters was trained.

[0034] It should be noted that in the sensitivity coefficient matrix output by the support vector regression model, the values ​​of each element directly correspond to the weights of the tool geometry parameters on cutting resistance. For example, the sensitivity coefficient for the cutting edge radius is 0.42, indicating that for every 1 mm change in this parameter, the cutting resistance changes by 42%; the sensitivity coefficient for the rake angle is 0.31, indicating that for every 1° change in the rake angle, the cutting resistance changes by 31%. This quantification method clarifies the degree of influence of each parameter on cutting resistance, providing a precise weighting basis for constructing the objective function. For instance, when processing high-gloss coated paper, the sensitivity coefficient for the cutting edge radius reaches the highest level of 0.42, the rake angle sensitivity is 0.31, and the cutting edge width sensitivity is relatively low at only 0.14; when processing corrugated base paper, the sensitivity coefficient for the cutting edge angle is 0.45, making it the main influencing factor.

[0035] The current resistance deviation is the absolute value of the difference between the actual cutting resistance and the theoretical resistance in the resistance fluctuation sequence, reflecting the mechanical equilibrium state of the current cutting process. The adaptive weighted objective function uses the current resistance deviation, the extent of hardness exceeding the limit, and the proportion of toughness reduction as weighting terms. The weights of each term are dynamically adjusted based on the sensitivity coefficient matrix. For example, when hardness exceeding the limit is the main cause, the sensitivity coefficients of each tool parameter corresponding to the direction of hardness change in the sensitivity coefficient matrix are extracted, and weights are allocated according to the coefficient proportions, so that tool parameters with a more significant impact on hardness occupy a higher weight in the objective function. Similarly, when toughness reduction or resistance deviation is the main problem, the weight allocation is adjusted accordingly to ensure that the objective function focuses on the core contradictions. The core optimization objective of the objective function is to minimize the extent of hardness exceeding the limit and the proportion of toughness reduction while controlling the cutting resistance deviation ≤ 0.5N, achieving a balanced optimization of multiple objectives.

[0036] The objective function is solved using the gradient descent method. The specific solution process is as follows: First, initialize the cutting depth correction to 0 and calculate the initial objective function value; then, with a step size of 0.0001 mm, iteratively update the correction along the gradient descent direction of the objective function, recalculating the objective function value after each iteration; when the difference between the objective function values ​​of two adjacent iterations is less than 1 × 10⁻⁶, the objective function is considered solved. -6When the change in correction amount is less than 0.0001mm, the convergence condition is met, and iteration stops. The convergence accuracy is set to 0.001mm, which is determined based on the mechanical control accuracy of the automatic cutting and bubbling machine and can meet the cutting accuracy requirement of ±0.1mm. The final optimal cutting depth correction range is controlled between -0.18 and +0.09mm. The negative correction range is -0.18 to 0mm, which is suitable for scenarios where the material hardness is high, the toughness is low, or the resistance is too high. By reducing the cutting depth, the cutting resistance is reduced, and the material tearing is avoided. The positive correction range is 0 to +0.09mm, which is suitable for scenarios where the material hardness is low, the toughness is high, and the resistance is too low. By increasing the cutting depth, the cutting is ensured to be thorough and the incomplete cutting is avoided. This range has been verified by 100 sets of tests on different paper types and can effectively avoid new instability caused by over-adjustment. By superimposing the optimal cutting depth correction amount with the current feed position data, the tool depth adjustment increment can be obtained, realizing precise dynamic adjustment of the tool depth.

[0037] In step S14, the current cutting speed is obtained, and simulation prediction is performed based on the adjustment increment. If the simulation result exceeds the preset processing stability boundary, the current cutting speed is corrected, a target cutting speed is generated and associated with the adjustment increment to form a comprehensive adjustment scheme, including: The current cutting speed data is collected via a servo motor encoder; The change in material removal rate is calculated based on the tool depth adjustment increment. The change in material removal rate is then input into a preset cutting thermo-mechanical coupling simulation unit to obtain the predicted value of the tool tip instantaneous temperature and the amplitude of the cutting force fluctuation. A preset processing stability boundary is defined, which is based on the heat resistance characteristics of paper and the cutting limit of the tool, and includes a temperature boundary and a cutting force fluctuation boundary. If the instantaneous temperature prediction exceeds the temperature boundary or the cutting force fluctuation amplitude exceeds the cutting force fluctuation boundary, a genetic algorithm is run to solve for the rotation speed compensation factor. The current cutting speed is then corrected using the rotation speed compensation factor to obtain the target cutting speed, which is then correlated with the adjustment increment to form a comprehensive adjustment scheme.

[0038] It should be noted that the current cutting speed data is collected by a servo motor encoder with the encoder resolution set to 1000 lines / revolution. After conversion by the motor transmission ratio, the detection accuracy of the cutting speed can reach ±0.1mm / min. The range of the collected current cutting speed data is 1000~3000mm / min, covering the conventional processing speed range of the automatic cutting and bubbling machine.

[0039] When calculating the change in material removal rate based on the incremental adjustment of tool depth, the paper width is considered in the calculation. The paper width is set according to the actual processing specifications, ranging from 50 to 1000 mm. The change in material removal rate directly reflects the reduction in cutting volume. The change in material removal rate is input into a preset cutting thermo-mechanical coupling simulation unit. This unit is built based on a finite element analysis model, and the simulation mesh uses an unstructured tetrahedral mesh. The mesh size of the core cutting region of the tool tip is set to 0.01 mm, and the mesh size of the outer region of the tool tip is gradually increased in a 1.2-fold gradient to ensure the simulation accuracy of the core region while taking into account the simulation efficiency. The simulation time step is 0.001 seconds, which is determined based on the dynamic response characteristics of the contact between the tool tip and the paper during the cutting process, and can accurately capture the instantaneous fluctuations of cutting force and rapid changes in temperature.

[0040] The simulation model embeds the thermophysical properties and constitutive relations of the paper. The thermophysical properties include thermal conductivity, specific heat capacity, and coefficient of thermal expansion. Parameters for different paper types are precisely configured as needed. The constitutive relations relate the cutting force, cutting speed, and the plastic deformation of the material. This model can accurately simulate the temperature distribution and cutting force changes in the blade tip region, reproducing the thermo-mechanical coupling state during actual cutting. After inputting the change in material removal rate into the simulation unit, the predicted instantaneous temperature at the blade tip and the amplitude of cutting force fluctuation are analyzed and obtained as the core indicators of the simulation results. For example, for high-gloss coated paper, when the removal rate decreases by 9.2%, the simulation shows that the highest instantaneous temperature in the blade tip region decreases from 128℃ to 109℃, and the amplitude of cutting force fluctuation decreases from a peak of 48N to 31N; for corrugated base paper, when the removal rate decreases by 10%, the temperature decreases from 95℃ to 82℃, and the amplitude of cutting force fluctuation decreases from 65N to 42N.

[0041] The preset processing stability boundaries are based on the heat resistance characteristics of paper and the cutting limit calibration of the tool. The temperature boundary is set at 115℃, which is the critical temperature for thermal damage of common paper. Exceeding this temperature will lead to defects such as carbonization and adhesion of the paper edges. The cutting force fluctuation boundary is set at 35N. This value is determined based on the fatigue strength of the tool and the tear limit of the paper. Exceeding this boundary will increase the risk of tool wear and paper tearing.

[0042] If the instantaneous temperature prediction exceeds the temperature boundary or the cutting force fluctuation amplitude exceeds the cutting force fluctuation boundary, a genetic algorithm is run to solve for the speed compensation factor. The genetic algorithm has a population size of 50, 80 generations, a crossover probability of 0.8, and a mutation probability of 0.05. The fitness function comprehensively considers the temperature over-limit penalty, the force fluctuation suppression objective, and the machining efficiency constraint. The temperature over-limit penalty is gradient-weighted according to the difference in temperature exceeding the boundary; the greater the temperature exceeds the boundary, the larger the penalty value. The force fluctuation suppression objective aims to optimize the cutting force fluctuation amplitude to approach the set boundary. The machining efficiency constraint limits the lower limit of the speed compensation factor to avoid a significant drop in production efficiency due to excessively low cutting speed. After normalization, the three indicators are weighted and superimposed. The weight of the temperature over-limit penalty is 0.45, the weight of the force fluctuation suppression objective is 0.4, and the weight of the machining efficiency constraint is 0.15. This multi-objective weighted fitness function achieves an accurate solution for the speed compensation factor. The final converged speed compensation factor ranges from 0.84 to 0.91.

[0043] Finally, the current cutting speed is corrected using the rotation speed compensation factor to obtain the target cutting speed. This target cutting speed is then correlated with the tool depth adjustment increment to form a comprehensive adjustment scheme that includes depth and speed parameters, providing a complete basis for subsequent command generation.

[0044] In step S15, the process of parsing the comprehensive adjustment scheme to generate timed tool control commands, loading them into the servo control loop, acquiring actual cutting resistance data streams, calculating the deviation from theoretical resistance values, and constructing a resistance deviation sequence includes: The target cutting speed and the adjustment increment in the comprehensive adjustment scheme are analyzed, and control commands are generated through multi-axis interpolation. The control command is loaded into the servo control loop, the servo motor load current signal is collected, and the actual cutting resistance data stream is calculated. Calculate the difference between the actual cutting resistance data stream and the theoretical resistance point by point; Arrange the differences in chronological order to construct a resistance deviation sequence that includes the peak value, duration, and standard deviation of the deviation.

[0045] It should be noted that after analyzing the target cutting speed and tool depth adjustment increments in the comprehensive adjustment scheme, time-sequential control commands are generated through multi-axis interpolation. The interpolation algorithm uses linear interpolation, and the command cycle is dynamically set according to the equipment's control precision. In conventional processing scenarios, the command cycle is 5 milliseconds, suitable for scenarios with moderate precision requirements such as ordinary paper cutting and packaging paper processing. This cycle can meet the ±0.1 mm cutting precision requirement while ensuring processing efficiency. In high-precision processing scenarios, the command cycle is adjusted to 2 milliseconds, targeting processing scenarios with extremely high requirements for cut smoothness, such as art paper and specialty coated paper. By shortening the command cycle, the response speed of tool movement is improved, increasing the cutting precision to within ±0.05 mm, ensuring burr-free and delamination-free cuts. The control commands include X-axis speed commands and Z-axis position commands. The X-axis corresponds to the feed direction, and the Z-axis corresponds to the tool lifting direction. The Z-axis position command is updated in real time according to the tool depth adjustment increment, while the X-axis speed command remains constant according to the target cutting speed.

[0046] After the control commands are loaded into the servo control loop, the load current signal of the servo motor is acquired through a current sensor. The acquisition frequency is set to 1000 Hz, which is determined based on the dynamic response characteristics of the servo motor. This frequency can completely capture the instantaneous changes in the motor load and avoid the distortion of resistance data caused by too low a sampling frequency. The current signal range is set to 0 to 5 amps, covering the normal operating current range of the servo motor of the automatic cutting and compression machine. When the current exceeds this range, the system will trigger overload protection to prevent motor damage or tool jamming. The current signal is calculated into cutting torque based on the motor's torque constant, and then the actual cutting resistance data stream is obtained by converting it through the tool radius. The final output cutting resistance data stream ranges from 0 to 50 Newtons, with an accuracy of ±0.1 Newtons.

[0047] The theoretical resistance values ​​are pre-calibrated based on parameters such as material density and fiber strength. The core calibration process involves first selecting a standard sample corresponding to the paper type and measuring its basic physical parameters using a material testing machine, including actual density, longitudinal and transverse fiber strength, and elongation at break. Then, combined with the cutting equipment's tool geometry parameters, including the cutting edge angle, rake angle, and blade radius, the theoretical cutting resistance benchmark value for the paper at different cutting depths and speeds is calculated using cutting mechanics theory. Finally, multiple sets of actual cutting verification experiments are conducted to correct the theoretical values. The calculated theoretical resistance is compared with the actual resistance collected during the actual cutting process, and the theoretical value set is iteratively calibrated according to the deviation value to ensure that the matched degree between the calibrated theoretical resistance value and the actual processing scenario is above 95%. The theoretical resistance benchmark values ​​differ for different types of paper. The theoretical resistance benchmark for offset paper is 31.5 Newtons, for coated paper it is 28.3 Newtons, and for corrugated base paper it is 35.7 Newtons. The calibration process involves measuring the paper's breaking strength using a tensile testing machine and calculating the resistance based on the tool's cutting angle and depth to ensure consistency between the theoretical value and the actual processing scenario. When calculating the difference between the actual cutting resistance data stream and the theoretical resistance value set point by point, the difference is calculated one by one in chronological order. The range of the difference is -5 to +5 Newtons. A positive value indicates that the actual resistance is greater than the theoretical value, and a negative value indicates that the actual resistance is less than the theoretical value.

[0048] When constructing a resistance deviation sequence by arranging the differences in chronological order, the sequence length is the sampling data within the most recent second. In a standard machining scenario, this includes 200 data points, corresponding to a 5-millisecond command cycle; in a high-precision machining scenario, it includes 500 data points, corresponding to a 2-millisecond command cycle. This sequence not only records the raw difference data but also extracts key statistical features, including the deviation peak value, duration, and standard deviation. The deviation peak value is the largest absolute value in the sequence, with a normal range of ±3 Newtons; the duration is the longest consecutive positive or negative deviation, with a normal range of no more than 0.5 seconds; and the standard deviation is the dispersion of the differences, with a normal range of no more than 1 Newton. These features provide core basis for subsequent cutting accuracy evaluation.

[0049] In step S16, the step of extracting contact state features based on the resistance deviation sequence and calculating cutting accuracy indicators based on the contact state features includes: The frequency of peak deviation, duration of continuous positive deviation, and standard deviation of deviation are extracted from the resistance deviation sequence as contact state features. The contact state features are mapped to a preset processing quality correlation matrix to obtain the mapping result; The comprehensive influence coefficient is calculated based on the mapping result, and then converted into a cutting accuracy index according to a preset conversion rule.

[0050] It should be noted that the contact state characteristics extracted from the resistance deviation sequence specifically include three items: the frequency of deviation peak occurrence, the duration of continuous positive deviation, and the standard deviation of deviation. The frequency of deviation peak occurrence refers to the number of times the deviation peak exceeds ±3N per unit time; under normal processing conditions, this frequency should be ≤2 times / second. The duration of continuous positive deviation refers to the longest time interval between consecutive positive deviation values ​​exceeding 1N; the normal range is ≤0.5 seconds. The standard deviation reflects the dispersion of resistance deviation; the normal range is ≤1N, and a larger standard deviation indicates a more unstable cutting process. This judgment standard is based on statistical analysis of 1200 sets of actual processing historical data for different paper types, including eight mainstream processed paper types such as offset paper, coated paper, and corrugated base paper. Each set of data simultaneously recorded contact state characteristics and final cut quality. Statistical analysis determined that 98% of the qualified processing samples had contact state characteristics falling within the above normal range, which was used as the basis for judging the stability of the cutting process.

[0051] The pre-defined processing quality correlation matrix is ​​3×3 in dimension. Rows correspond to three contact state characteristics, and columns correspond to three main quality defects: edge burrs, thermal damage, and delamination / tearing. Matrix elements are the correlation coefficients between features and defects, ranging from 0 to 1. Higher values ​​indicate a more significant impact of the feature on the defect. The correlation coefficients are calculated using Pearson correlation coefficient analysis combined with processing experimental data. First, based on 1000 sets of processing experimental data, the correlation coefficients between each contact state characteristic and the quantitative values ​​of various quality defects are calculated. Then, low-correlation terms are eliminated through significance testing. The final retained and calibrated coefficients are the matrix elements, ensuring the accuracy of the correlation relationships. For example, the correlation coefficient between the frequency of peak deviation and edge burrs is 0.85, the correlation coefficient between the duration of continuous positive deviation and thermal damage is 0.92, and the correlation coefficient between the standard deviation of deviation and delamination / tearing is 0.88.

[0052] The contact state features are mapped to a preset processing quality correlation matrix to obtain the mapping result, which is the quantified value of the correlation strength between each contact state feature and the corresponding quality defect. When calculating the comprehensive influence coefficient based on the mapping result, a weighted summation method is used. The weights are set according to the severity of the quality defect: edge burrs have a weight of 0.3, thermal damage has a weight of 0.4, and delamination / tears have a weight of 0.3. The specific implementation process is as follows: First, the three contact state features are normalized to eliminate the influence of different dimensions, mapping the actual feature values ​​to a numerical range of 0 to 1. The more the feature value deviates from the normal range, the closer the normalization result is to 1. Then, the normalized feature value is multiplied one by one with the corresponding correlation coefficient in the processing quality correlation matrix to obtain the influence value of each feature on various types of quality defects. Subsequently, according to the preset weight of the quality defects, the influence values ​​corresponding to various types of defects are weighted and summed to obtain the comprehensive influence value of a single type of defect. Finally, the comprehensive influence values ​​of the three types of defects are added together to obtain the overall comprehensive influence coefficient. The comprehensive influence coefficient ranges from 0 to 1. The larger the coefficient, the more significant the negative impact of resistance deviation on processing quality. Based on the comprehensive influence coefficient, a new cutting accuracy index is obtained, with an index range of 0 to 100 points. The closer the comprehensive influence coefficient is to 0, the closer the cutting accuracy index is to 100 points. The higher the score, the better the cutting accuracy.

[0053] In step S17, if the cutting accuracy index is lower than a preset cutting accuracy qualification threshold, then resistance change characteristics are extracted from the resistance deviation sequence. Based on the resistance change characteristics, the current physical properties are corrected in conjunction with the pre-acquired material thermal conductivity characteristics to generate optimized physical property indices, including: Real-time resistance change features, including peak amplitude, fluctuation frequency, and rate of change, are extracted from the resistance deviation sequence to generate a time-domain synchronization sequence. Based on the influence weights of each feature in the resistance variation characteristics on cutting stability, the sequence weights are calculated, and a weighted fusion matrix is ​​constructed. The weighted fusion matrix is ​​subjected to feature analysis, and the cutting heat distribution data is calculated in combination with the material's thermal conductivity characteristics to generate correction values ​​for physical properties; wherein the physical properties include thermal conductivity coefficient and friction coefficient. The correction amount is loaded into the preset base material model to update the physical property parameters related to cutting stability and generate optimized physical property indices.

[0054] It should be noted that the preset cutting accuracy qualification threshold is set at 85 points. This threshold was obtained through statistical analysis of 500 sets of qualified processing samples and 300 sets of unqualified processing samples. 85 points effectively distinguishes between qualified and unqualified products. When the index is not lower than 85 points, 98% of the samples meet industry quality standards, with cut smoothness and non-damage meeting the standards. When the new cutting accuracy index is not lower than 85 points, it is judged as meeting the accuracy standard, and the current processing parameters are maintained. When the new cutting accuracy index is lower than 85 points, it is judged as not meeting the accuracy standard, triggering the subsequent physical property optimization process. For example, when the deviation peak frequency is 3 times / second, the duration of continuous positive deviation is 0.8 seconds, and the standard deviation of deviation is 1.5N, the comprehensive influence coefficient is high, the new cutting accuracy index is 79 points, which is lower than the qualification threshold, and the system initiates the physical property optimization process. When all characteristics are within the normal range, the comprehensive influence coefficient is low, the new cutting accuracy index is 92 points, and it is judged as meeting the accuracy standard, without needing to enter the subsequent optimization and adjustment process.

[0055] It should be noted that real-time resistance change characteristics are extracted from the resistance deviation sequence. These characteristics specifically include three items: peak amplitude, fluctuation frequency, and rate of change. A time-domain synchronization sequence is generated based on these characteristics. The peak amplitude is the peak-to-peak value of the resistance fluctuation, ranging from 0 to 10N; the fluctuation frequency is the number of times the resistance peak occurs per unit time, ranging from 0 to 5Hz; and the rate of change is the average rate at which the resistance rises from a trough to a peak, ranging from 0 to 20N / second. When generating the time-domain synchronization sequence based on these characteristics, the extracted peak amplitude, fluctuation frequency, and rate of change are aligned in time to form three parallel, continuous sequences with the same time index. This sequence is the time-domain synchronization sequence, which fully preserves the dynamic process of resistance change.

[0056] When calculating the sequence weights based on the influence of each feature in the resistance variation characteristics on cutting stability, the analytic hierarchy process (AHP) is used to determine the weight allocation. The peak amplitude weight is 0.4, the fluctuation frequency weight is 0.3, and the rate of change weight is 0.3. This weight allocation has been verified by multiple sets of machining experiments and can accurately reflect the degree of influence of different features on the cutting process. When constructing the weighted fusion matrix, the matrix dimension is 2000×3. The rows correspond to each data point in the time-domain synchronization sequence, and the columns correspond to the weight coefficients of the three features. The matrix elements are the product of the data point value and the corresponding weight. This matrix is ​​used to achieve weighted integration of multiple features.

[0057] When performing feature analysis on the weighted fusion matrix, principal component analysis was used to extract the first two principal components, with a cumulative variance contribution rate exceeding 95%. This was then combined with pre-acquired material thermal conductivity characteristics to calculate the cutting heat distribution data. The material thermal conductivity characteristics were set according to paper type: 0.12 W / (m·K) for offset paper, 0.15 W / (m·K) for coated paper, and 0.10 W / (m·K) for corrugated base paper. The cutting heat distribution data was presented as a two-dimensional heat map, with coordinates representing the spatial location of the cutting area and numerical values ​​representing temperature values ​​ranging from 25 to 120℃. This clearly shows the heat concentration area near the cutting tip, providing a targeted basis for physical property correction.

[0058] It is worth noting that the physical property corrections include corrections for thermal conductivity and friction coefficient, with a range of ±10% to ±30%, determined based on the uniformity of cutting heat distribution and peak temperature. When the peak temperature exceeds 90℃ and the heat distribution is concentrated, the thermal conductivity correction is set to +20% to +30%, and the friction coefficient correction is set to -10% to -20%. When the heat distribution is uniform but the peak temperature is slightly high, the thermal conductivity correction is set to +10% to +20%, and the friction coefficient correction is set to -5% to -10%. When the heat distribution is uniform and the peak temperature is normal, the correction is controlled within ±5% to avoid over-correction affecting cutting stability.

[0059] Then, when the correction is applied to the base material model, which contains the basic physical properties of paper, such as initial thermal conductivity, initial friction coefficient, density, and elastic modulus, the corresponding thermal conductivity and friction coefficient parameters in the model are directly updated during the application process, generating optimized physical property indices. The optimized physical property indices can more accurately reflect the material characteristics under actual cutting conditions. For example, after correction, the thermal conductivity of double-sided coated paper is adjusted from 0.12 W / (m·K) to 0.144 W / (m·K), and the friction coefficient is adjusted from 0.32 to 0.288, providing accurate input parameters for the subsequent cutting force prediction model and ensuring the effectiveness of iterative optimization.

[0060] In step S18, the core features of the material are extracted based on the optimized physical property indicators and input into a preset cutting force prediction model to obtain the theoretical maximum bearing cutting force. If the theoretical maximum bearing cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization, including: The material hardness value and coefficient of thermal expansion are extracted from the optimized physical property indicators to obtain the core characteristics of the material; The core characteristics of the material are input into a preset cutting force prediction model, and the theoretical maximum cutting force that the material can withstand is output. If the theoretical maximum cutting force is less than the real-time instantaneous load data, then a corrected feed rate is generated through adaptive control. A cutting force fluctuation function is constructed, and iterative optimization calculations are performed with the goal of minimizing the cutting force fluctuation until it converges to a preset accuracy threshold. The optimal cutting depth is derived, and the final tool depth and cutting speed configuration are determined in combination with the modified feed rate.

[0061] It should be noted that the core material characteristic parameters extracted from the optimized physical property indicators are the material hardness value and the coefficient of thermal expansion. The material hardness value is the Shore D hardness after physical property correction, and the coefficient of thermal expansion is set according to the paper type. The coefficient of thermal expansion of offset paper is 12 × 10⁻⁶. -6 / K, coated paper is 10×10 -6 / K, corrugated base paper is 15×10 -6 / K. When constructing the multidimensional feature vector, the hardness value is normalized and directly combined with the coefficient of thermal expansion to form a two-dimensional feature vector, which provides input for the cutting force prediction model.

[0062] The cutting force prediction model is constructed using a backpropagation (BP) neural network. The input layer consists of two-dimensional feature vectors, with two hidden layers, each containing 16 nodes. The ReLU activation function is used, and the output layer represents the theoretical maximum cutting force. The model was trained on 300 sets of samples covering various combinations of hardness and thermal expansion coefficients. The prediction error was controlled within ±5%, and the theoretical maximum cutting force ranged from 20 to 60 N, consistent with the actual cutting capacity of paper. For example, the optimized model has a hardness of 65 Shore D and a thermal expansion coefficient of 12 × 10⁻⁶. -6 / K double-sided offset paper has a theoretical maximum cutting force of 48N; its hardness value is 58 Shore D, and its coefficient of thermal expansion is 15×10⁻⁶. -6 / K corrugated base paper has a theoretical maximum cutting force of 55N.

[0063] Instantaneous load data is the peak value of cutting resistance monitored in real time during the current cutting process. It is obtained by extracting the maximum value of the resistance data stream within the most recent 0.5 seconds, with a monitoring frequency of 10Hz to ensure timely capture of load changes. When the theoretical maximum withstand cutting force is less than the instantaneous load data, it indicates that the current cutting force is approaching or exceeding the material's tolerance limit, posing a risk of machining defects. In this case, adaptive control generates a corrected feed rate. The corrected feed rate is dynamically adjusted based on the ratio between the theoretical maximum withstand cutting force and the instantaneous load data, ensuring that the corrected feed rate keeps the cutting force within the material's tolerance range.

[0064] When constructing the cutting force fluctuation function, the function uses the cutting depth as the independent variable and the cutting force fluctuation amplitude as the dependent variable, and is obtained by fitting the output results of the cutting force prediction model. Iterative optimization calculations are performed with the goal of minimizing cutting force fluctuation, with an iteration step size set at 0.01 mm and a convergence accuracy set at a cutting force fluctuation amplitude ≤ 5 N. During the iteration process, an initial cutting depth is first set, i.e., the current actual cutting depth, and the cutting force fluctuation amplitude at the current depth is simulated using the cutting force prediction model. If the fluctuation amplitude does not meet the standard, the cutting depth is adjusted in the direction of decreasing cutting force fluctuation, and the simulation is repeated. This process is repeated until the fluctuation amplitude meets the convergence accuracy requirement, and the optimal cutting depth is derived. For example, with an initial cutting depth of 1.25 mm, an instantaneous load of 52 N, and a theoretical maximum cutting force of 48 N, the feed rate is first corrected from 1800 mm / min to 1728 mm / min using a proportional relationship, and then the cutting depth is iteratively adjusted. After 5 iterations, the cutting depth is adjusted to 1.12 mm, at which point the cutting force fluctuation amplitude drops to 4.2 N, satisfying the convergence condition. By combining the modified feed rate and the optimal depth of cut, the final tool and cutting parameter configuration is determined to achieve a stable cutting process.

[0065] In summary, this invention discloses a cutting control method for an automatic cutting and bubbling machine, comprising collecting and processing paper resistance and surface signals, determining material hardness and toughness through sensor fusion and neural network, adjusting tool depth and cutting speed in a coordinated manner if limits are exceeded, evaluating cutting accuracy based on feedback, and optimizing physical properties and iteratively calibrating parameters if standards are not met. This invention achieves stable and controllable cutting processes in complex material processing through multi-source data fusion to perceive material characteristics, multi-dimensional parameter collaborative optimization, and a closed-loop feedback calibration mechanism, thereby improving cutting accuracy and production consistency, and reducing scrap rate and tool wear.

[0066] Reference Figure 2 The second embodiment of the present invention provides a cutting control system for an automatic cutting and bubbling machine, comprising: The data acquisition and processing module is used to collect paper processing resistance data and material surface signals through a sensor array, and generate real-time resistance fluctuation sequence and initial physical property indicators after noise reduction and synchronization. The material property determination module is used to perform multi-scale decomposition of the signal through sensor fusion based on the real-time resistance fluctuation sequence and the initial physical property index, construct a multi-dimensional coupling matrix, extract the elastic modulus and plastic rheological threshold feature parameters in the multi-dimensional coupling matrix, and input them into the neural network model after feature integration processing to determine the current hardness value and current toughness value of the material. The tool depth adjustment module is used to preset the hardness safety threshold and the toughness safety threshold. If the current hardness value or the current toughness value exceeds the safety threshold, the real-time resistance fluctuation sequence and tool geometric parameters are extracted, and the tool depth adjustment increment is obtained through cross-correlation analysis and regression calculation. The speed coordination adjustment module is used to obtain the current cutting speed through the servo motor encoder, calculate the material removal rate change in combination with the tool depth adjustment increment, simulate and analyze thermal and cutting force parameters, set the processing stability boundary, and generate a speed compensation factor to correct the cutting speed if the thermal or cutting force parameters exceed the processing stability boundary. This, together with the tool depth adjustment increment, forms a comprehensive adjustment scheme. The instruction execution feedback module is used to parse the comprehensive adjustment scheme to generate time-sequential tool control instructions, load them into the servo control loop, collect the actual cutting resistance data stream, calculate the deviation from the theoretical resistance value, and construct a resistance deviation sequence. The cutting accuracy assessment module is used to extract the contact state characteristics of the resistance deviation sequence, map them to the processing quality correlation matrix to assess the impact, determine new cutting accuracy indicators, and set a cutting accuracy qualification threshold. The physical property optimization module is used to construct a weighted fusion matrix based on the real-time resistance change features extracted from the resistance deviation sequence if the new cutting accuracy index is lower than the cutting accuracy qualified threshold, analyze the physical property correction amount, and update and generate the optimized physical property index. The parameter iteration configuration module is used to extract the core features of the material based on the optimized physical property index, input the cutting force prediction model to obtain the theoretical maximum cutting force, and if the theoretical maximum cutting force is less than the instantaneous load, determine the final tool and cutting parameter configuration through adaptive control iteration optimization to achieve stable cutting.

[0067] It should be noted that the cutting control system for an automatic cutting and blotting machine provided in this embodiment of the invention is used to execute all the process steps of the cutting control method for an automatic cutting and blotting 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.

[0068] 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.

[0069] 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. A cutting control method for an automatic cutting and bubbling machine, characterized in that, include: Collect paper processing resistance data and perform noise reduction processing to generate a resistance fluctuation sequence; Based on the resistance fluctuation sequence, the elastic modulus characterization value and energy dissipation ratio are obtained and input into a preset neural network model to output the current hardness value and current toughness value of the material. If the current hardness value or the current toughness value exceeds the preset safety threshold, the tool geometry parameters are obtained and the influence weight of each parameter on the cutting resistance is determined. The current resistance deviation is calculated, and an objective function is constructed and solved based on the current resistance deviation and the influence weight to obtain the tool depth adjustment increment. The current cutting speed is obtained, and simulation prediction is performed based on the adjustment increment. If the simulation result exceeds the preset processing stability boundary, the current cutting speed is corrected, a target cutting speed is generated and associated with the adjustment increment to form a comprehensive adjustment scheme. The integrated adjustment scheme is analyzed to generate timed tool control commands, which are then loaded into the servo control loop. The actual cutting resistance data stream is collected, the deviation from the theoretical resistance value is calculated, and a resistance deviation sequence is constructed. Contact state features are extracted based on the resistance deviation sequence, and cutting accuracy indicators are calculated based on the contact state features. If the cutting accuracy index is lower than the preset cutting accuracy qualified threshold, then the resistance change characteristics are extracted from the resistance deviation sequence. Based on the resistance change characteristics, the current physical properties are corrected in combination with the pre-acquired material thermal conductivity characteristics to generate optimized physical property indexes. Based on the optimized physical property index, the core features of the material are extracted and input into the preset cutting force prediction model to obtain the theoretical maximum cutting force. If the theoretical maximum cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization.

2. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The process of collecting paper processing resistance data and performing noise reduction to generate a resistance fluctuation sequence includes: The servo motor torque sensor in the control sensor array collects the friction torque data of the paper running, and the fiber texture signal on the material surface is collected by the high-speed optical sensor. Low-pass filtering is performed on the friction torque data and the fiber texture signal respectively to remove noise, and the time delay is calibrated by cross-correlation analysis to obtain a synchronous digital signal stream; Surface signal components are extracted from the synchronous digital signal stream, and the standard deviation is calculated to obtain roughness feature values. These roughness feature values ​​are then used as initial physical property indicators. Based on the roughness feature value, the filtering window is adaptively adjusted to filter the resistance data component in the synchronous digital signal stream, generating a resistance fluctuation sequence.

3. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The process of obtaining the elastic modulus characterization value and energy dissipation ratio based on the resistance fluctuation sequence and inputting them into a preset neural network model, outputting the current hardness value and current toughness value of the material, includes: The resistance wave sequence is decomposed by multi-scale wavelet transform to separate signal components including high-frequency instantaneous impact components and low-frequency fiber deformation rate components. A multidimensional coupling matrix is ​​constructed with time windows as rows and the signal components as columns. The elastic modulus characterization value and plastic rheological threshold in the multidimensional coupling matrix are extracted by singular value decomposition. The material deformation equation is constructed using the elastic modulus characterization value and the plastic rheology threshold, and the energy dissipation ratio is calculated. The elastic modulus value and energy dissipation ratio are input into a preset radial basis function neural network model, which outputs the current hardness and toughness values ​​of the material.

4. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, If the current hardness value or the current toughness value exceeds a preset safety threshold, then the tool geometry parameters are obtained and the influence weight of each parameter on the cutting resistance is determined. The current resistance deviation is calculated, and an objective function is constructed and solved based on the current resistance deviation and the influence weights to obtain the tool depth adjustment increment, including: If the current hardness value exceeds the preset hardness safety threshold, or the current toughness value exceeds the preset toughness safety threshold, then extract the tool geometry parameters including the cutting edge angle, rake angle size, and cutting edge radius. A cross-correlation analysis of the resistance fluctuation sequence and the tool geometry parameters is performed using an adaptive length sliding window to generate a time-varying correlation feature vector. The feature vector is input into a preset support vector regression model, and a sensitivity coefficient matrix is ​​output. The influence weight of each parameter on the cutting resistance is quantified based on the sensitivity coefficient matrix. Based on the resistance fluctuation sequence and the preset theoretical resistance, the resistance deviation is calculated. Combined with the influence weight, an adaptive weighted objective function is constructed, and the adjustment increment of the tool depth is obtained by solving it.

5. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The process involves obtaining the current cutting speed, performing simulation prediction based on the adjustment increment, and if the simulation result exceeds a preset processing stability boundary, then correcting the current cutting speed, generating a target cutting speed, and associating it with the adjustment increment to form a comprehensive adjustment scheme, including: The current cutting speed data is collected via a servo motor encoder; The change in material removal rate is calculated based on the tool depth adjustment increment. The change in material removal rate is then input into a preset cutting thermo-mechanical coupling simulation unit to obtain the predicted value of the tool tip instantaneous temperature and the amplitude of the cutting force fluctuation. A preset processing stability boundary is defined, which is based on the heat resistance characteristics of paper and the cutting limit of the tool, and includes a temperature boundary and a cutting force fluctuation boundary. If the instantaneous temperature prediction exceeds the temperature boundary or the cutting force fluctuation amplitude exceeds the cutting force fluctuation boundary, a genetic algorithm is run to solve for the rotation speed compensation factor. The current cutting speed is then corrected using the rotation speed compensation factor to obtain the target cutting speed, which is then correlated with the adjustment increment to form a comprehensive adjustment scheme.

6. The cutting control method for an automatic cutting and bubbling machine according to claim 4, characterized in that, The process involves analyzing the comprehensive adjustment scheme to generate time-sequential tool control commands, loading them into the servo control loop, acquiring actual cutting resistance data streams, calculating the deviation from theoretical resistance values, and constructing a resistance deviation sequence, including: The target cutting speed and the adjustment increment in the comprehensive adjustment scheme are analyzed, and control commands are generated through multi-axis interpolation. The control command is loaded into the servo control loop, the servo motor load current signal is collected, and the actual cutting resistance data stream is calculated. Calculate the difference between the actual cutting resistance data stream and the theoretical resistance point by point; Arrange the differences in chronological order to construct a resistance deviation sequence that includes the peak value, duration, and standard deviation of the deviation.

7. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The step of extracting contact state features based on the resistance deviation sequence and calculating cutting accuracy indicators based on the contact state features includes: The frequency of peak deviation, duration of continuous positive deviation, and standard deviation of deviation are extracted from the resistance deviation sequence as contact state features. The contact state features are mapped to a preset processing quality correlation matrix to obtain the mapping result; The comprehensive influence coefficient is calculated based on the mapping result, and then converted into a cutting accuracy index according to a preset conversion rule.

8. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The process of extracting drag change characteristics from the drag deviation sequence, and then modifying the current physical properties based on these drag change characteristics and pre-acquired material thermal conductivity characteristics to generate optimized physical property indices includes: Real-time resistance change features, including peak amplitude, fluctuation frequency, and rate of change, are extracted from the resistance deviation sequence to generate a time-domain synchronization sequence. Based on the influence weights of each feature in the resistance variation characteristics on cutting stability, the sequence weights are calculated, and a weighted fusion matrix is ​​constructed. The weighted fusion matrix is ​​subjected to feature analysis, and the cutting heat distribution data is calculated in combination with the material's thermal conductivity characteristics to generate correction values ​​for physical properties; wherein the physical properties include thermal conductivity coefficient and friction coefficient. The correction amount is loaded into the preset base material model to update the physical property parameters related to cutting stability and generate optimized physical property indices.

9. The cutting control method for an automatic cutting and bubbling machine according to claim 1, characterized in that, The process involves extracting core material features based on the optimized physical property indicators and inputting them into a preset cutting force prediction model to obtain the theoretical maximum withstand cutting force. If the theoretical maximum withstand cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization, including: The material hardness value and coefficient of thermal expansion are extracted from the optimized physical property indicators to obtain the core characteristics of the material; The core characteristics of the material are input into a preset cutting force prediction model, and the theoretical maximum cutting force that the material can withstand is output. If the theoretical maximum cutting force is less than the real-time instantaneous load data, then a corrected feed rate is generated through adaptive control. A cutting force fluctuation function is constructed, and iterative optimization calculations are performed with the goal of minimizing the cutting force fluctuation until it converges to a preset accuracy threshold. The optimal cutting depth is derived, and the final tool depth and cutting speed configuration are determined in combination with the modified feed rate.

10. A cutting control system for an automatic cutting and bubbling machine, characterized in that, include: The data acquisition and processing module is used to collect paper processing resistance data and perform noise reduction processing to generate a resistance fluctuation sequence. The material property determination module is used to obtain the elastic modulus characterization value and energy dissipation ratio based on the resistance fluctuation sequence, input them into a preset neural network model, and output the current hardness value and current toughness value of the material. The tool depth adjustment module is used to obtain tool geometry parameters and determine the influence weight of each parameter on cutting resistance if the current hardness value or the current toughness value exceeds a preset safety threshold, calculate the current resistance deviation, construct an objective function based on the current resistance deviation and the influence weight, and solve it to obtain the tool depth adjustment increment. The speed coordination adjustment module is used to obtain the current cutting speed, perform simulation prediction based on the adjustment increment, and if the simulation result exceeds the preset processing stability boundary, the current cutting speed is corrected, a target cutting speed is generated and associated with the adjustment increment to form a comprehensive adjustment scheme. The instruction execution feedback module is used to parse the comprehensive adjustment scheme to generate time-sequential tool control instructions, load them into the servo control loop, collect the actual cutting resistance data stream, calculate the deviation from the theoretical resistance value, and construct a resistance deviation sequence. The cutting accuracy evaluation module is used to extract contact state features based on the resistance deviation sequence and calculate the cutting accuracy index based on the contact state features. The physical property optimization module is used to extract resistance change features from the resistance deviation sequence if the cutting accuracy index is lower than the preset cutting accuracy qualification threshold, and to modify the current physical properties based on the resistance change features and the pre-acquired material thermal conductivity characteristics to generate optimized physical property index. The parameter iteration configuration module is used to extract the core features of the material based on the optimized physical property index and input them into the preset cutting force prediction model to obtain the theoretical maximum cutting force. If the theoretical maximum cutting force is less than the instantaneous load, the final tool and cutting parameter configuration is determined through adaptive control iterative optimization.

Images