A multi-process parameter adaptive intelligent control system for slitting machines
Through multi-dimensional perception and state estimation models, the adaptive control system of the slitting machine estimates the internal defect state of the composite material in real time, which solves the problem that external optimization exacerbates internal defects in existing technologies, realizes the synergistic optimization of internal and external quality, and improves the slitting yield and stability.
Patent Information
- Application Number
- CN202611361220.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-03
- Publication Date
- 2026-11-17
AI Technical Summary
The existing adaptive control system of slitting machine cannot effectively detect internal defects in composite materials, which may exacerbate internal defects when optimizing external quality, affecting the yield and production stability of high-end composite materials.
A multi-dimensional sensing module is used to collect electrical, acoustic and mechanical data. The internal defect state is estimated in real time through a state estimation model, and control instructions are generated by combining external quality indicators to adaptively adjust process parameters.
It enables real-time quantitative estimation of internal defects in composite materials and synergistic optimization of external quality, significantly improving the yield and production stability of high-end composite material slitting.
Smart Images

Figure CN123113353A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of slitting machine control, and more particularly to an adaptive intelligent control system for multi-process parameters of a slitting machine. Background Technology
[0002] Slitting machines are core equipment for cutting wide rolls of material into multiple narrow rolls, widely used in the processing of films, paper, composite materials, and other roll materials. During the slitting process, process parameters such as unwinding tension, slitting speed, winding tension, and pressure roller pressure have a decisive impact on the final product quality. Existing adaptive control methods mainly rely on sensors such as machine vision and laser rangefinders to monitor the external macroscopic geometric dimensions of the finished rolls online, including end-face flatness, width, and roll diameter, and then adjust the process parameters in real time through closed-loop control. This method is highly effective in ensuring the dimensional accuracy of products and has become a standard technical approach in the industry.
[0003] In existing technologies, taking sandwich-structured medical tape containing a cotton core layer as an example, the shearing force applied by the slitting tool may cause minute, invisible relative displacements between the middle cotton core layer and the upper and lower tape layers, known as interlayer slippage. This defect cannot be detected by external dimensional inspection at the time of slitting, but during subsequent storage and use, it gradually develops into serious quality problems such as edge delamination and warping, leading to product scrapping. Existing adaptive control systems lack a perception channel for internal structural defects, and their optimization decisions are entirely based on external surface quality indicators. Under certain operating conditions, when the system adjusts process parameters to improve end-face flatness, it may unintentionally increase interlayer shear stress, thereby exacerbating interlayer slippage and creating a paradoxical situation where external quality is acceptable but internal defects worsen. This severely restricts the yield rate and production stability of high-end composite material slitting. Summary of the Invention
[0004] To address the technical problems existing in the background art, this invention proposes an adaptive intelligent control system for multi-process parameters of a slitting machine, the specific solution of which is as follows: A multi-process parameter adaptive intelligent control system for a slitting machine is applied to control a slitting machine that includes at least an unwinding process, a slitting process, and a rewinding process, comprising: The multidimensional sensing module is used to collect multidimensional real-time working condition data of the slitting machine during the slitting process. The multidimensional real-time working condition data is data reflecting the internal state of the composite material in at least three different physical dimensions during the slitting process. The feature extraction module is used to preprocess the multidimensional real-time operating condition data to generate multidimensional feature data. The defect tracking module is used to use the multidimensional feature data as real-time observations and, using a pre-built state estimation model, to perform sequential importance sampling and resampling iterations to estimate the state of internal defects in composite materials in real time and obtain internal defect state estimates. The collaborative decision-making module is used to acquire external quality indicators that characterize the external quality of finished products, and generate control instructions with the goal of balancing the internal defect state estimate and the external quality indicators based on the real-time internal defect state estimate and the external quality indicators. The control module is used to execute the control instructions and adaptively adjust one or more process parameters of the slitting machine.
[0005] Furthermore, the multi-dimensional sensing module contains at least three different physical dimensions of multi-dimensional real-time operating condition data, including: Dielectric property data collected by an array of capacitive sensors deployed on both sides of the composite material's travel path; Stress wave data acquired by an acoustic emission sensor mounted on a guide roller near the downstream of the slitting blade assembly; And drive load response data acquired through the take-up motor servo driver.
[0006] Furthermore, in the multi-dimensional sensing module, the multi-dimensional real-time operating condition data is collected through clock synchronization via the PTP protocol.
[0007] Furthermore, in the feature extraction module, the preprocessing includes at least one of the following preprocessing operations: The dielectric characteristic data is digitally filtered to eliminate power frequency interference and high-frequency disturbance components are extracted to form dielectric characteristic data; When the signal amplitude of the stress wave data exceeds the preset signal amplitude threshold, the waveform is triggered for acquisition. The energy, peak amplitude, rise time, and arrival time difference of the signals received by the sensors at both ends of the waveform are calculated in real time to form stress wave characteristic data. The instantaneous torque command value, actual torque feedback value, and motor speed of the winding motor are collected synchronously. The following error between the actual torque and the command torque is calculated, and the energy spectral density change of the following error in a specific frequency band is analyzed to form drive load response characteristic data.
[0008] Furthermore, the internal defect is the slippage of interlayers within the composite material; In the defect tracking module, the internal defect state estimation is represented by a state vector, which consists of four elements: the position of the slip along the axial direction, the depth of the slip along the radial direction, the magnitude of the slip, and the rate of change of the magnitude of the slip. The state estimation model includes a state transition model for describing the evolution of the state vector over time.
[0009] Furthermore, in the defect tracking module, the pre-built state estimation model includes a feature observation model; the feature observation model is constructed in the following ways: On the slitting test bench, a high-speed microscopic camera is used to capture and quantize the slip state vector of the edge of the slid composite material as the ground value, and the multidimensional feature data is collected simultaneously to form a training dataset consisting of the slip state vector ground value and the corresponding multidimensional feature data. Based on the training dataset, a feature observation model is constructed using a deep neural network. The feature observation model is used to simulate the mapping relationship between internal defect states and multidimensional feature data. The feature observation model is used to predict the corresponding theoretical feature values based on the input state vector during real-time estimation.
[0010] Furthermore, in the defect tracking module, sequential importance sampling and resampling iterations are performed to estimate the state vector of internal defects in the composite material in real time, including: Initially, N particles are generated, each particle representing an assumption about the state vector; At each time step, the state transition model is used to predict the state of all particles from the previous time step to the current time step. For each particle, the theoretical feature value corresponding to its predicted state is calculated using the feature observation model. The theoretical feature value is compared with the multidimensional feature data to calculate the likelihood probability. The likelihood probability is used as the confidence weight of the particle. The optimal estimate of the state vector at the current moment is obtained by taking a weighted average of the state vectors represented by all particles. When the number of effective particles is lower than the preset effective particle number threshold, resampling is performed.
[0011] Furthermore, in the defect tracking module, after obtaining the optimal estimate of the state vector, the following steps are also included: Based on the optimal estimate of the state vector, an internal defect risk index is calculated. The internal defect risk index is positively correlated with the slip magnitude in the optimal estimate and also positively correlated with the absolute value of the slip magnitude change rate in the optimal estimate. The internal defect risk index is obtained by multiplying the slip magnitude and the absolute value of the slip magnitude change rate by a pre-set normalization coefficient and then summing them.
[0012] Furthermore, the collaborative decision-making module generates control instructions aimed at balancing the internal defect state estimate and the external quality indicators, including: Construct a multi-objective optimization function, which includes at least a first optimization term related to the internal defect state estimation and a second optimization term related to the external quality index; Based on the real-time internal defect state estimation and the external quality indicators, the multi-objective optimization function is solved to generate a control command that makes the multi-objective optimization function optimal.
[0013] Furthermore, in the execution control module, the control commands include the winding tension adjustment amount, the pressure roller pressure adjustment amount, and the slitting speed adjustment amount; Executing the control commands includes: adjusting the winding tension according to the winding tension adjustment amount, adjusting the pressure roller pressure according to the pressure roller pressure adjustment amount, and adjusting the slitting speed according to the slitting speed adjustment amount.
[0014] Compared with the prior art, the present invention can achieve at least the following beneficial effects: 1. This invention collects real-time operating condition data from at least three different physical dimensions, uses a state estimation model to quantitatively estimate the interlayer slip defects inside composite materials in real time, and incorporates the internal defect state estimation and external quality indicators into the control decision. With the goal of balancing internal and external quality, it generates control instructions, fundamentally solving the technical problem that the lack of an internal defect perception channel in existing technologies may exacerbate internal defects when optimizing external quality. This invention achieves comprehensive and coordinated optimization of the internal and external quality of products, and significantly improves the yield and production stability of high-end composite material slitting.
[0015] 2. This invention collects real-time operating data from at least three different physical dimensions, simultaneously capturing the internal state information of composite materials from multiple dimensions. This fills the perception blind spot of existing technologies, which can only monitor external macroscopic geometric dimensions, enabling the slitting machine to penetrate the surface and perceive internal structural defects. By using multi-dimensional feature data as real-time observations and performing sequential importance sampling and resampling iterations using a state estimation model, real-time quantitative estimation of the location, magnitude, and rate of change of interlayer slip defects is achieved, providing internal defect state information that cannot be obtained by existing technologies for control decisions. By incorporating the internal defect state estimation and external quality indicators into the generation process of control instructions, optimization decisions are made with the goal of balancing internal and external quality. This avoids the paradox of unintentionally exacerbating internal defects in order to improve external quality, as seen in existing technologies. It achieves comprehensive and coordinated optimization of internal and external product quality, significantly improving the yield and production stability of high-end composite material slitting. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1This is a system principle block diagram of the present invention. Detailed Implementation
[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0018] Please refer to Figure 1 A multi-process parameter adaptive intelligent control system for slitting machines is used to control slitting machines that include at least unwinding, slitting, and rewinding processes. The system includes: a multi-dimensional perception module, a feature extraction module, a defect tracking module, a collaborative decision-making module, and an execution control module. The multidimensional sensing module is used to collect multidimensional real-time working condition data of the slitting machine during the slitting process. The multidimensional real-time working condition data is data reflecting the internal state of the composite material in at least three different physical dimensions during the slitting process.
[0019] It should be noted that multidimensional real-time operating condition data refers to real-time measurement data that can reflect the changes in the internal structural state of composite materials during the slitting process from different physical dimensions. Existing adaptive control methods for slitting machines mostly rely on sensors such as machine vision and laser rangefinders to monitor the external macroscopic geometric dimensions of the finished rolls, but cannot detect hidden defects inside the material. This invention collects data from at least three different physical dimensions, capturing information about the internal state of the material from electrical, acoustic, and mechanical dimensions, respectively, providing a multidimensional observational basis for subsequent estimation of internal defect states.
[0020] In an optional embodiment, the multidimensional sensing module contains at least three different physical dimensions of multidimensional real-time operating condition data, including: Dielectric property data collected by an array of capacitive sensors deployed on both sides of the composite material's travel path; Stress wave data acquired by an acoustic emission sensor mounted on a guide roller near the downstream of the slitting blade assembly; And drive load response data acquired through the take-up motor servo driver.
[0021] It should be noted that dielectric property data is used to reflect minute changes in interlayer distance and dielectric constant within the composite material. A capacitance sensor array is arranged along the transverse direction of the slitting machine, covering the edge region of the finished roll, and collects the capacitance values of each electrode pair at a predetermined sampling frequency. When interlayer slippage occurs within the composite material, the local change in interlayer distance causes a disturbance in the capacitance value. By extracting this disturbance component, the location and extent of the slippage can be indirectly characterized.
[0022] It should be noted that stress wave data is used to capture high-frequency stress wave signals generated by the propagation of microcracks or interlaminar friction within the material. The acoustic emission sensor is installed on the guide roller bearing housing near the downstream of the slitting blade assembly. When slippage occurs within the material, interlaminar friction and microcrack propagation release stress waves. Upon receiving the signal, the sensor triggers waveform acquisition. By analyzing the waveform's energy, amplitude, rise time, and arrival time difference, the location and intensity information of the slippage source can be obtained.
[0023] It should be noted that the drive load response data is used to reflect subtle fluctuations in motor load caused by changes in the internal state of the material during the winding process. When delamination slippage occurs within the composite material, the material's mechanical properties change. This change is transmitted to the winding motor via the drive chain, manifesting as a specific frequency band energy change in torque following error. By analyzing the frequency domain characteristics of this following error, an indirect characterization of internal defects can be obtained from a mechanical perspective.
[0024] In the multi-dimensional sensing module, the multi-dimensional real-time operating condition data is collected by clock synchronization via the PTP protocol.
[0025] It should be noted that clock synchronization via the PTP (Precise Time Protocol) is to ensure precise alignment of data collected by different sensors on the time axis. Since dielectric property data, stress wave data, and drive load response data are subsequently input into the state estimation model as observation vectors at the same moment, the time synchronization accuracy of the sensor data directly affects the accuracy of the state estimation. The PTP protocol can achieve microsecond-level clock synchronization accuracy, meeting the time consistency requirements of multi-sensor data fusion.
[0026] The feature extraction module is used to preprocess the multidimensional real-time operating condition data to generate multidimensional feature data.
[0027] It should be noted that the multidimensional real-time operating condition data collected by the multidimensional sensing module is the raw signal, which contains irrelevant components such as environmental noise and power frequency interference, and cannot be directly used for accurate estimation of internal defect states. The purpose of preprocessing is to extract feature components that are sensitive to internal defects and have clear physical meaning from the raw signal, forming multidimensional feature data to improve the accuracy and robustness of subsequent state estimation.
[0028] In an optional embodiment, the preprocessing in the feature extraction module includes at least one of the following preprocessing operations: The dielectric characteristic data is digitally filtered to eliminate power frequency interference and high-frequency disturbance components are extracted to form dielectric characteristic data; When the signal amplitude of the stress wave data exceeds the preset signal amplitude threshold, the waveform is triggered for acquisition. The energy, peak amplitude, rise time, and arrival time difference of the signals received by the sensors at both ends of the waveform are calculated in real time to form stress wave characteristic data. The instantaneous torque command value, actual torque feedback value, and motor speed of the winding motor are collected synchronously. The following error between the actual torque and the command torque is calculated, and the energy spectral density change of the following error in a specific frequency band is analyzed to form drive load response characteristic data.
[0029] It should be noted that digital filtering of dielectric property data to eliminate power frequency interference is necessary because electromagnetic interference at 50Hz and its harmonics exists in industrial environments. This interference superimposed on the signal acquired by the capacitance sensor affects the accurate measurement of minute changes in capacitance. After filtering out power frequency interference with a notch filter, the high-frequency disturbance component of the capacitance value is extracted. This high-frequency disturbance component is more sensitive to instantaneous changes in interlayer distance. When interlayer slippage occurs within the composite material, minute changes in interlayer distance cause rapid fluctuations in capacitance; this fluctuation characteristic is the core component of the dielectric property data.
[0030] It should be noted that stress wave data acquisition employs a threshold-triggered method, meaning waveform acquisition only occurs when the signal amplitude output by the acoustic emission sensor exceeds a preset threshold. This threshold is typically set to be a certain multiple higher than the background noise level, such as three times the root mean square value of the background noise. Using a threshold-triggered method effectively avoids the massive data storage and processing burden of continuous acquisition, recording only meaningful acoustic emission events. The real-time calculated waveform energy reflects the intensity of the stress wave event, the peak amplitude characterizes the severity of local damage, and the rise time is related to the characteristics of the acoustic emission source. The arrival time difference of the signals received by the two sensors, utilizing the known propagation speed of the acoustic emission signal in the material, allows for the location of the slip source. A larger arrival time difference indicates a greater distance difference between the slip source and the two sensors.
[0031] It should be noted that the simultaneous acquisition of the instantaneous torque command value and actual torque feedback value of the winding motor during the preprocessing of the drive load response data is for calculating the following error between the two. Ideally, the actual torque should closely follow the changes in the command value; however, when delamination slippage occurs within the composite material, the mechanical properties of the material change instantaneously, causing the actual torque to fail to respond promptly to changes in the command value, thus increasing the following error. Energy spectral density analysis is performed on this following error in a specific frequency band (e.g., 5-50Hz) because this frequency band corresponds precisely to the characteristic frequency range of material stress fluctuations during the slitting process, effectively reflecting load response anomalies caused by internal defects. The change in energy spectral density is the core component of the drive load response characteristic data, providing a quantitative basis for identifying internal defects from a mechanical perspective.
[0032] The defect tracking module is used to perform sequential importance sampling and resampling iterations using the multidimensional feature data as real-time observations and a pre-built state estimation model to estimate the state of internal defects in the composite material in real time and obtain an internal defect state estimate.
[0033] It should be noted that the generated multidimensional feature data is used as the basis for observation to infer the internal defect state that cannot be directly measured in real time. Since interlayer slip occurs inside the composite material, its position, magnitude, and rate of change cannot be directly measured by external sensors. Therefore, a state estimation method is needed, using multidimensional feature data as indirect observations to deduce the most probable internal defect state. Sequential importance sampling and resampling iteration are the core schemes of the particle filter algorithm. Its basic idea is to approximate the posterior probability distribution of the true state in the state space using a large number of weighted random samples, which is particularly suitable for the nonlinear, non-Gaussian state estimation problem faced in this invention.
[0034] In an optional embodiment, the internal defect is interlayer slippage within the composite material; In the defect tracking module, the internal defect state estimation is represented by a state vector, which consists of four elements: the position of the slip along the axial direction, the depth of the slip along the radial direction, the magnitude of the slip, and the rate of change of the magnitude of the slip.
[0035] It should be noted that the state vector is: The state estimation model includes a state transition model to describe the evolution of the state vector over time. The four elements of the state vector provide a complete description of the interlayer slip defect from both spatial location and severity perspectives. The axial location x represents the position of the slip along the transverse direction of the slitting machine; the radial depth y represents the interlayer interface along the thickness direction of the composite material where the slip occurs; the slip magnitude s represents the severity of the slip (e.g., the slip area); and the rate of change v of the slip magnitude represents whether the slip is accelerating or stabilizing. These four elements together constitute a complete quantitative description of the interlayer slip defect state, providing a comprehensive information basis for subsequent control decisions.
[0036] The state estimation model includes a state transition model for describing the evolution of the state vector over time.
[0037] It should be noted that the state transition model is constructed as follows: Where A is the state transition matrix and B is the noise driving matrix. This represents Gaussian process noise. The state transition model describes the evolution of the state vector over time. In actual cutting processes, the precise physical evolution model of interlayer slip is extremely complex, involving multiple disciplines such as nonlinear contact mechanics, material plastic deformation, and tribology, making it difficult to establish an accurate analytical model. Therefore, a simplified second-order dynamic model is adopted. The evolution of the state is described by the state transition matrix A, which reflects the inertial characteristics of the slip state, and the noise driving matrix B and Gaussian process noise. This is used to describe the uncertainty of the model. This model acknowledges the unavailability of an exact physical model and instead adopts an engineering-realizable and robust stochastic process model. The parameters of matrices A and B can be initially calibrated through offline simulation or with a small amount of experimental data.
[0038] In an optional embodiment, the pre-built state estimation model in the defect tracking module includes a feature observation model; the feature observation model is constructed in the following ways: On the slitting test bench, a high-speed microscopic camera is used to capture and quantize the slip state vector of the edge of the slid composite material as the ground value, and the multidimensional feature data is collected simultaneously to form a training dataset consisting of the slip state vector ground value and the corresponding multidimensional feature data. Based on the training dataset, a feature observation model is constructed using a deep neural network. The feature observation model is used to simulate the mapping relationship between internal defect states and multidimensional feature data. The feature observation model is used to predict the corresponding theoretical feature values based on the input state vector during real-time estimation.
[0039] It should be noted that the feature observation model is used to establish the mapping relationship between the internal defect state and multi-dimensional feature data. Since this mapping relationship involves the coupling of multiple physical fields such as electricity, acoustics, and mechanics, it is difficult to establish directly through analytical methods. Therefore, this invention adopts a data-driven approach to construct the model. On the slitting experimental platform, a high-speed microscopic camera is used to directly photograph and quantify the true state of edge slip from the side of the material being slid, serving as the ground truth label for the training data. Simultaneously, all sensor data are collected as input features for the training data, forming a complete training dataset. The deep neural network trained based on this dataset can learn the complex mapping relationship between the state vector and multi-dimensional feature data. Thus, in actual production processes, it can quickly predict the corresponding theoretical sensor feature values based on a hypothetical defect state, providing a basis for particle weight updates.
[0040] In an optional embodiment, the defect tracking module performs sequential importance sampling and resampling iterations to estimate the state vector of internal defects in the composite material in real time, including: Initially, N particles are generated, each particle representing an assumption about the state vector; At each time step, the state transition model is used to predict the state of all particles from the previous time step to the current time step. For each particle, the theoretical feature value corresponding to its predicted state is calculated using the feature observation model. The theoretical feature value is compared with the multidimensional feature data to calculate the likelihood probability. The likelihood probability is used as the confidence weight of the particle. The optimal estimate of the state vector at the current moment is obtained by taking a weighted average of the state vectors represented by all particles. When the number of effective particles is lower than the preset effective particle number threshold, resampling is performed.
[0041] It should be noted that the initially generated N particles are uniformly or randomly distributed in the possible state space, and each particle carries a complete hypothesis about the state vector. At each time step, the state of all particles is first predicted using a state transition model, allowing it to evolve over time. Then, a feature observation model is used to calculate the theoretical eigenvalue corresponding to the predicted state for each particle. Next, the theoretical eigenvalue is compared with the actual collected multidimensional feature data. The closer the two are, the more credible the state hypothesis represented by the particle is, and the higher its confidence weight is. Finally, a weighted average of the states of all particles is taken to obtain the optimal estimate of the true state at the current time. When the weight of most particles becomes extremely low, and only a few particles have high weights, it indicates that the number of effective samples in the particle set is insufficient. At this time, a resampling operation is performed, copying high-weight particles and discarding low-weight particles, so that the particle set is refocused on the most likely state region, ensuring the accuracy and stability of subsequent estimations.
[0042] It should be noted that the preset effective particle count threshold is usually set to one-half to two-thirds of the total number of particles N. For example, when N is 1000, the effective particle count threshold can be set to 500 to 600. When the effective particle count is lower than this threshold, it indicates that the particle set has severely degraded, with a few high-weight particles dominating the state estimation results. The particle set's ability to cover other possible regions in the state space has decreased significantly, and resampling must be performed to restore particle diversity.
[0043] It should be noted that the specific value of the effective particle count threshold is determined based on the following criteria: if high real-time performance of state estimation is required, the threshold can be appropriately lowered to reduce the frequency of resampling operations and lower computational overhead; if high accuracy of state estimation is required, the threshold can be appropriately increased to recover particle diversity more promptly and ensure estimation accuracy. This threshold can be optimized and calibrated through simulation experiments or statistical analysis of actual operating data.
[0044] In an optional embodiment, after obtaining the optimal estimate of the state vector, the defect tracking module further includes: Based on the optimal estimate of the state vector, an internal defect risk index is calculated. The internal defect risk index is positively correlated with the slip magnitude in the optimal estimate and also positively correlated with the absolute value of the slip magnitude change rate in the optimal estimate. The internal defect risk index is obtained by multiplying the slip magnitude and the absolute value of the slip magnitude change rate by a pre-set normalization coefficient and then summing them.
[0045] It should be noted that the internal defect risk index R is calculated using the following formula: ,in and These represent the slip magnitude and rate of change in the optimal estimation of the state vector, respectively. and The normalization coefficient is a pre-defined value. The internal defect risk index is a comprehensive quantification of the optimal estimate of the state vector, aiming to provide an intuitive and comparable numerical indicator for subsequent control decisions. The risk index is obtained by weighted summation of two parts: the magnitude of slip and the absolute value of the rate of change of the slip magnitude. The larger the slip magnitude, the more severe the interlayer slip that has already occurred; the larger the absolute value of the rate of change of the slip magnitude, the more rapidly the slip is accelerating or deteriorating. Combining the two, it can reflect both the current severity of the defect and its development trend, giving the risk index a forward-looking early warning capability. Normalization coefficient and This is used to adjust the slip magnitude and rate of change to the same order of magnitude; its specific value can be set according to the quality control requirements of the actual production process. For example, if more sensitivity is needed to the development trend of defects, the value can be appropriately increased. The value of can be increased appropriately if more attention is paid to the absolute severity of the defect. The value of .
[0046] It should be noted that the normalization coefficient and This is used to adjust the magnitude and rate of change of slip to a comparable numerical scale. Its specific value can be set according to the quality control strategy of the actual production process. For example, if the production process is more sensitive to sudden defects and a rapid response from the system to the slip development trend is desired, the value can be appropriately increased. The value of is adjusted to give the rate of change a greater weight in the risk index; if the production process is more concerned about cumulative defects, the value can be appropriately increased. The value of is chosen to give greater weight to the absolute magnitude of the slip in the risk index. This is achieved by adjusting... and The relative proportions allow the risk index calculation strategy to be flexibly adapted to different product quality control needs.
[0047] It should be noted that the normalization coefficient and The value of is in the range of 0 to 1, and satisfies . . Used to adjust the weighting of the slip magnitude on the risk index. Used to adjust the contribution weight of the rate of change of the slip magnitude to the risk index. When When the value is large, the risk index focuses more on reflecting the current severity of the slippage; when... When the value is larger, the risk index focuses more on reflecting the development trend of slippage, so as to enhance the early warning capability of the system. and The specific value can be set comprehensively based on factors such as the interlayer bonding strength of the material being cut, the tolerance of the production process to defects, and the cutting speed, and can be optimized through offline experiments or statistical feedback from production data.
[0048] The collaborative decision-making module is used to acquire external quality indicators that characterize the external quality of the finished product, and to generate control instructions with the goal of balancing the internal defect state estimate and the external quality indicators based on the real-time internal defect state estimate and the external quality indicators.
[0049] It should be noted that this invention unifies the consideration of quality indicators from both internal and external dimensions to generate control commands that take both into account. In existing adaptive control methods for slitting machines, control decisions are based solely on external macroscopic quality indicators. When external quality indicators are normal but internal slippage has occurred, the system cannot detect and intervene, and may even exacerbate internal slippage by unilaterally pursuing external quality. This invention introduces internal defect state estimation as one of the inputs to the control decision, giving the system a "global perspective" that considers both internal and external quality simultaneously.
[0050] In an optional embodiment, the collaborative decision-making module generates control instructions aimed at balancing the internal defect state estimate and the external quality indicators, including: Construct a multi-objective optimization function, which includes at least a first optimization term related to the internal defect state estimation and a second optimization term related to the external quality index; Based on the real-time internal defect state estimation and the external quality indicators, the multi-objective optimization function is solved to generate a control command that makes the multi-objective optimization function optimal.
[0051] It should be noted that external quality indicators refer to quantitative indicators reflecting deviations in the appearance quality of finished products, which are collected and calculated online by an external visual inspection system. Examples include the deviation of the flatness of the finished roll end face from a set value, and the tolerance of the finished product width from the nominal width, all detected by a machine vision system. These indicators are commonly used quality control criteria in existing technologies. Integrating them with internal defect state estimation achieves synergistic optimization of internal and external quality.
[0052] It should be noted that the multi-objective optimization function unifies internal and external quality objectives into a solvable mathematical expression. The first optimization term is related to the estimation of internal defect states, and its physical meaning is to minimize the risk of internal defects; the second optimization term is related to external quality indicators, and its physical meaning is to minimize external quality deviations. By solving this multi-objective optimization function, an optimal balance point can be found between internal quality protection and external quality control, generating a set of control instructions that optimize overall quality. This mechanism avoids the dilemma of "paying for one thing at the expense of another" in traditional methods and achieves true multi-objective intelligent control.
[0053] It should be noted that solving the multi-objective optimization function is a real-time process. Each control cycle recalculates the optimal control command based on the latest internal defect state estimate and external quality indicators. When the internal defect risk index is low, the optimization result tends to prioritize improving external quality indicators to enhance product appearance and production efficiency. As the internal defect risk index begins to rise, the optimization result gradually shifts towards protecting internal quality, even if this may sacrifice the perfection of external quality or production efficiency to some extent. This dynamic balancing mechanism enables the control system to adaptively adjust its optimization focus according to changes in actual operating conditions, embodying the core concept of "intelligent control."
[0054] The control module is used to execute the control instructions and adaptively adjust one or more process parameters of the slitting machine.
[0055] It should be noted that the execution and control module is the execution link in the entire adaptive intelligent control closed loop. It transforms the control commands generated by the collaborative decision-making module into the actual actions of each actuator of the slitting machine, completing the real-time adjustment of process parameters. Through the closed-loop iteration from the multi-dimensional perception module to the execution and control module, the system continuously collects operating condition data, estimates the internal defect status, generates control commands, and executes adjustments, forming a complete adaptive control loop of "perception-estimation-decision-execution".
[0056] In an optional embodiment, the control module includes control commands such as winding tension adjustment, pressure roller pressure adjustment, and slitting speed adjustment. Executing the control commands includes: adjusting the winding tension according to the winding tension adjustment amount, adjusting the pressure roller pressure according to the pressure roller pressure adjustment amount, and adjusting the slitting speed according to the slitting speed adjustment amount.
[0057] It should be noted that winding tension, pressure roller pressure, and slitting speed are key process parameters affecting the slitting quality of composite materials, and they are closely related to the generation and development of interlayer slippage. Winding tension determines the longitudinal tensile force on the material during winding; excessive tension increases interlayer shear stress, exacerbating the risk of interlayer slippage. Pressure roller pressure determines the radial pressure acting on the surface of the finished roll; excessive pressure increases the interlayer bonding force, while insufficient pressure may lead to loosening of the end faces. Slitting speed determines the shearing rate of the material by the cutter; excessive speed may cause instantaneous excessive shearing force, triggering localized interlayer slippage. The control commands simultaneously include adjustment values for these three parameters, enabling the system to coordinate adjustments from multiple dimensions according to actual working conditions, achieving comprehensive optimization of both internal and external quality.
[0058] It should be noted that adaptive adjustment means that the control system does not rely on preset fixed parameter values, but rather dynamically calculates and executes the parameter adjustment amount most suitable for the current working conditions in each control cycle based on real-time working condition data and state estimation results. This adjustment method enables the slitting machine to maintain the optimal parameter combination under complex working conditions such as changes in material properties, fluctuations in environmental conditions, and tool wear, ensuring the stability and consistency of product quality.
[0059] In summary, this invention solves the technical paradox of existing technologies that may exacerbate internal damage when optimizing appearance quality due to a lack of internal defect perception capabilities, and significantly improves the yield and production stability of high-end composite material slitting.
[0060] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.
[0061] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.
[0062] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0063] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.
[0064] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.
[0065] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A slitting machine multi-process parameter adaptive intelligent control system, applied to control a slitting machine comprising at least a unwinding process, a slitting process and a winding process, characterized in that, include: The multidimensional sensing module is used to collect multidimensional real-time working condition data of the slitting machine during the slitting process. The multidimensional real-time working condition data is data reflecting the internal state of the composite material in at least three different physical dimensions during the slitting process. The feature extraction module is used to preprocess the multidimensional real-time operating condition data to generate multidimensional feature data. The defect tracking module is used to use the multidimensional feature data as real-time observations and, using a pre-built state estimation model, to perform sequential importance sampling and resampling iterations to estimate the state of internal defects in composite materials in real time and obtain internal defect state estimates. The collaborative decision-making module is used to acquire external quality indicators that characterize the external quality of finished products, and generate control instructions with the goal of balancing the internal defect state estimate and the external quality indicators based on the real-time internal defect state estimate and the external quality indicators. The control module is used to execute the control instructions and adaptively adjust one or more process parameters of the slitting machine.
2. The slitting machine multi-process parameter self-adaptive intelligent regulation and control system of claim 1, wherein: The multi-dimensional sensing module contains at least three different physical dimensions of real-time operating condition data, including: Dielectric property data collected by an array of capacitive sensors deployed on both sides of the composite material's travel path; Stress wave data acquired by an acoustic emission sensor mounted on a guide roller near the downstream of the slitting blade assembly; And drive load response data acquired through the take-up motor servo driver.
3. The slitting machine multi-process parameter self-adaptive intelligent regulation and control system of claim 2, wherein: In the multi-dimensional sensing module, the multi-dimensional real-time operating condition data is collected by clock synchronization via the PTP protocol.
4. The slitting machine multi-process parameter self-adaptive intelligent regulation and control system of claim 2, wherein: In the feature extraction module, the preprocessing includes at least one of the following preprocessing operations: The dielectric characteristic data is digitally filtered to eliminate power frequency interference and high-frequency disturbance components are extracted to form dielectric characteristic data; When the signal amplitude of the stress wave data exceeds the preset signal amplitude threshold, the waveform is triggered for acquisition. The energy, peak amplitude, rise time, and arrival time difference of the signals received by the sensors at both ends of the waveform are calculated in real time to form stress wave characteristic data. The instantaneous torque command value, actual torque feedback value, and motor speed of the winding motor are collected synchronously. The following error between the actual torque and the command torque is calculated, and the energy spectral density change of the following error in a specific frequency band is analyzed to form drive load response characteristic data.
5. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 1, characterized in that: The internal defect is the slippage of interlayers within the composite material; In the defect tracking module, the internal defect state estimation is represented by a state vector, which consists of four elements: the position of the slip along the axial direction, the depth of the slip along the radial direction, the magnitude of the slip, and the rate of change of the magnitude of the slip. The state estimation model includes a state transition model for describing the evolution of the state vector over time.
6. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 5, characterized in that: In the defect tracking module, the pre-built state estimation model includes a feature observation model; The construction methods of the feature observation model include: On the slitting test bench, a high-speed microscopic camera is used to capture and quantize the slip state vector of the edge of the slid composite material as the ground value, and the multidimensional feature data is collected simultaneously to form a training dataset consisting of the slip state vector ground value and the corresponding multidimensional feature data. Based on the training dataset, a feature observation model is constructed using a deep neural network. The feature observation model is used to simulate the mapping relationship between internal defect states and multidimensional feature data. The feature observation model is used to predict the corresponding theoretical feature values based on the input state vector during real-time estimation.
7. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 6, characterized in that: In the defect tracking module, sequential importance sampling and resampling iterations are performed to estimate the state vector of internal defects in the composite material in real time, including: Initially, N particles are generated, each particle representing an assumption about the state vector; At each time step, the state transition model is used to predict the state of all particles from the previous time step to the current time step. For each particle, the theoretical feature value corresponding to its predicted state is calculated using the feature observation model. The theoretical feature value is compared with the multidimensional feature data to calculate the likelihood probability. The likelihood probability is used as the confidence weight of the particle. The optimal estimate of the state vector at the current moment is obtained by taking a weighted average of the state vectors represented by all particles. When the number of effective particles is lower than the preset effective particle number threshold, resampling is performed.
8. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 7, characterized in that: In the defect tracking module, after obtaining the optimal estimate of the state vector, the following steps are also included: Based on the optimal estimate of the state vector, an internal defect risk index is calculated. The internal defect risk index is positively correlated with the slip magnitude in the optimal estimate and also positively correlated with the absolute value of the slip magnitude change rate in the optimal estimate. The internal defect risk index is obtained by multiplying the slip magnitude and the absolute value of the slip magnitude change rate by a pre-set normalization coefficient and then summing them.
9. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 8, characterized in that: The collaborative decision-making module generates control instructions aimed at balancing the internal defect state estimate and the external quality indicators, including: Construct a multi-objective optimization function, which includes at least a first optimization term related to the internal defect state estimation and a second optimization term related to the external quality index; Based on the real-time internal defect state estimation and the external quality indicators, the multi-objective optimization function is solved to generate a control command that makes the multi-objective optimization function optimal.
10. The multi-process parameter adaptive intelligent control system for slitting machines as described in claim 9, characterized in that: In the execution control module, the control commands include the winding tension adjustment amount, the pressure roller pressure adjustment amount, and the slitting speed adjustment amount; Executing the control commands includes: adjusting the winding tension according to the winding tension adjustment amount, adjusting the pressure roller pressure according to the pressure roller pressure adjustment amount, and adjusting the slitting speed according to the slitting speed adjustment amount.
