Numerical control leveler operation state prediction and control method based on digital twinborn model

Abstract

The invention discloses a numerical control leveler operation state prediction and control method based on a digital twinborn model, and relates to the technical field of leveling prediction control. Comprising the following steps: S1, collecting leveling physical data in real time, and carrying out data preprocessing; s2, establishing a multi-roller stress chain model, forming a force chain structure, driving the model to continuously correct, and outputting a fusion stress sequence at the same time; s3, analyzing multi-roller local mutation features to identify an abnormal source roller, evaluating upstream and downstream transmission strength of the abnormal source roller in the multi-roller force chain, and judging whether the abnormality has an amplification trend or not; and S4, according to the abnormal source roller and evolution characteristics, the reduction correction amount is judged, the reduction correction amount is adaptively updated under the safety constraint, and closed-loop regulation and control feedback is carried out. The problems that in the prior art, the force chain transmission process between roller systems cannot be captured, a source roller with abnormal local stress cannot be recognized, and whether the source roller is amplified at the downstream position or not, so that the plate shape is unstable are solved.

Description

Technical Field

[0001] This invention relates to the field of predictive control technology for leveling, specifically to a method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model. Background Technology

[0002] Metal sheets are prone to morphological deviations such as waviness, warping, and uneven residual stress during rolling, cooling, and transportation. To meet the flatness requirements of subsequent processing stages such as shearing, stamping, welding, and forming, industrial production generally relies on leveling machines to adjust the internal stress of the sheet through alternating bending of multiple rollers, gradually restoring it to a flat state. With the increasing variety of materials such as high-strength steel, silicon steel, and aluminum sheets, and the ever-increasing demands of the manufacturing industry for dimensional accuracy and processing consistency, leveling equipment is widely used in industries such as automobiles, sheet metal, photovoltaics, home appliances, and building steel structures. This places higher demands on the stress stability, parameter reliability, and adaptability to different working conditions during the leveling process.

[0003] For example, invention patent CN120706085A discloses a method for constructing a thin plate leveling process parameter model and a leveling method, including: M1. Obtaining the corresponding material properties of the thin plate material to be leveled on-site; M2. Understanding the relevant data of the on-site welding process; M3. Obtaining the initial leveling position, heating time, heating current, and heating frequency based on the material properties of the thin plate and the welding process parameters; M4. Calculating the heat flux density of the Gaussian heat source model at the heating position based on the heating current, selecting the next heating position based on the heat flux density, and obtaining the heating sequence; M5. Repeating steps M3 to M4, further correcting the heating time, heating current, and heating frequency based on actual temperature change data to obtain the final thin plate leveling process parameter model. This invention establishes a multi-input, multi-output thin plate leveling process parameter model that conforms to the on-site working conditions by collecting data on the material properties of the workpiece being leveled and the actual leveling conditions.

[0004] For example, invention patent CN115455656A discloses a parameter calculation method and device for four-point pressure leveling of wide and thick plates. The method obtains the original curvature r0 of the plate to be leveled, the plate's elastic modulus E, yield strength σs, width B, and thickness h; calculates the strain ε at a distance z from the neutral layer when the plate bends in the opposite direction from its original curvature 1 / r0 to curvature 1 / r1 under the action of the leveling force F; calculates the height z0 of the plate's critical layer for elastoplastic deformation from the neutral layer; calculates the external moment M of the plate's rectangular cross-section as an ideal elastoplastic material; expresses the external moment M through the leveling force F, obtaining the relationship between the leveling force F and the curvature 1 / ρw; and finally calculates the leveling force F and the plate's reverse bending amount δ using the formula. This invention can accurately calculate the working parameters required by the leveling machine, improving the leveling machine's working efficiency.

[0005] However, during the continuous rolling process of a multi-roll leveler, there is a significant force chain transmission effect between different roll systems. Minor stress changes in the sheet material generated on each roll are amplified or superimposed along the roll group, forming typical anomalies such as localized wave amplification and reversed secondary waveforms. Traditional equipment, relying solely on single-point pressure or displacement sensors, struggles to capture the dynamic changes of the coupled force chain between rolls, making it impossible to quantify the evolution path of abnormal waveforms across the roll sequence. Furthermore, it is difficult to identify whether a localized anomaly generated by a particular roll will be amplified in the downstream roll group, resulting in difficulty in providing early warning of finished sheet shape defects.

[0006] Therefore, in order to address the above problems, there is an urgent need for a method for predicting and controlling the operating status of CNC leveling machines based on digital twin models. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for predicting and controlling the operating status of CNC leveling machines based on digital twin models. This method solves the problems that traditional technologies cannot capture the force chain transmission process between roller systems, cannot identify the source roller of local stress anomalies and whether they will be amplified downstream, thus leading to plate shape instability.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model, comprising the following steps: S1, real-time acquisition of leveling physical data, and completion of time synchronization, spatial mapping, noise reduction and smoothing, normalization and working condition labeling to obtain preprocessed leveling physical data; S2, establishment of a multi-roller force chain model based on the preprocessed leveling physical data, constructing a force chain structure through contact stiffness, adaptive boundary and bending unit, and driving continuous correction of the model with real-time leveling physical data, while outputting a fused force sequence; S3, analysis of local mutation characteristics of the multi-roller based on the fused force sequence, identification of abnormal source rolls based on the local mutation characteristics of the multi-roller, evaluation of the upstream and downstream transmission strength of the abnormal source roll in the multi-roller force chain, and determination of whether the abnormality has an amplification trend; S4, extraction of historical behavior patterns under the same working conditions based on the abnormal source roll and evolution characteristics, inference of the equivalent sensitivity relationship of the roll system gap, determination of the pressing correction amount, adaptive updating of the pressing correction amount under safety constraints, and closed-loop control feedback.

[0009] Furthermore, the specific process of acquiring preprocessed leveling physical data in real time, and performing time synchronization, spatial mapping, noise reduction and smoothing, normalization, and working condition labeling, is as follows: Real-time acquisition of leveling physical data, including: normal load and roll gap of each leveling roller, sheet speed, sheet passing position, leveling machine structural parameters, and leveling machine static parameters; time synchronization of the leveling physical data according to the equipment's bus timestamp; and mapping of the leveling physical data from the time dimension to the sheet movement dimension by integrating the sheet speed and discretizing the sheet position with a fixed spatial discrete step length. The spatial coordinate sequence of the movement direction is used to make each discrete spatial coordinate correspond to a stress section when the plate passes through the leveling roller group. The normal load, roll gap and motion state of each leveling roller at the same plate position are aligned. Moving average filtering is used to suppress noise and smooth the leveling physical data. The leveling physical data is normalized by z-score. At the same time, working condition labels are generated for each leveling physical data according to the steel grade, plate thickness and plate speed information. An operation predictive control database is established, and the preprocessed leveling physical data with corresponding spatial coordinates and working condition labels are stored in the operation predictive control database.

[0010] Furthermore, based on the preprocessed leveling physical data, a multi-roller force chain model is established. The specific process of constructing the force chain structure through contact stiffness, adaptive boundaries, and bending units is as follows: Based on the roller diameter, roller spacing, number of rollers, and roller shape structure in the leveling machine structural parameters, the roller center position and contact position of each leveling roller are determined in the frame reference coordinates, and the geometric arrangement relationship of the multiple rollers is constructed; the equivalent vertical stiffness of each leveling roller is calculated based on the roller material parameters, effective length of the roller body, and support method in the leveling machine structural parameters; the contact points between the plate and each leveling roller are discretized into force nodes, and the equivalent bending stiffness of the plate is calculated based on the plate thickness and elastic modulus, so that adjacent nodes are connected through plate bending units; the equivalent vertical stiffness, plate bending units, and support boundary conditions determined by the static parameters of the leveling machine are integrated to construct a digital twin model of the multi-roller force chain.

[0011] Furthermore, the specific process of continuously correcting the model using real-time leveling physical data and outputting the fused force sequence is as follows: The normal load of each leveling roll is mapped to the corresponding node in the multi-roller force chain digital twin model; the roll gap of each leveling roll is used as the displacement boundary condition of the multi-roller force chain digital twin model; the normal load application location and the stress area of ​​the plate are determined using the plate speed and plate passage position; by comparing the deviation between the multi-roller force change predicted by the multi-roller force chain digital twin model and the actual normal load, the contact stiffness, node transfer coefficient, and roll flexibility parameters in the multi-roller force chain digital twin model are automatically adjusted; the predicted normal load value of each leveling roll is output, and the predicted normal load value and the measured normal load are linearly fused according to a weighting coefficient based on the prediction error to obtain the fused force value of each leveling roll; simultaneously, the latest leveling physical data is continuously read and compared in real-time with the prediction results of the multi-roller force chain digital twin model, and the equivalent vertical stiffness, contact stiffness, and support boundary condition parameters of each leveling roll in the model are updated according to the prediction deviation.

[0012] Furthermore, the specific process of analyzing the local abrupt change characteristics of multi-rollers based on the fused force sequence is as follows: at each spatial coordinate along the direction of plate movement, the fused force value sequence is called and the average value is calculated to obtain the average force value of the multi-rollers; the fused force value of the (i+2)th roll is subtracted from twice the fused force value of the (i+1)th roll, and then the fused force value of the ith roll is added to obtain the second-order difference value of the multi-roller force chain; the second-order difference value of the multi-roller force chain is divided by the average force value of the multi-rollers to obtain the local force abrupt change value; based on the number of rolls at the force section, the local force abrupt change values ​​of all combinations are squared and summed, and then divided by the number of combinations to obtain the overall fluctuation value; the square root of the overall fluctuation value is used to obtain the second-order fluctuation value of the multi-roller force chain.

[0013] Furthermore, the specific process for identifying the abnormal source roller based on the local mutation characteristics of the multi-roller is as follows: For each spatial coordinate obtained by discretization along the direction of material movement, the second-order fluctuation value of the multi-roller force chain is calculated and compared with the fluctuation threshold; when the second-order fluctuation value of the multi-roller force chain is greater than the fluctuation threshold, it is determined that there is a mutation anomaly in the force link at the current spatial position, the current spatial position is marked as the abnormal source segment, and the maximum absolute value of the local force mutation value in the combination is selected, the (i+1)th roller is taken as the abnormal source roller, and the abnormal source roller sequence number is obtained. At the same time, the predictive control database is called to match historical samples with consistent working condition labels, and the median of the historical fused force value of each leveling roller is selected as the reference force value.

[0014] Furthermore, the specific process for evaluating the upstream and downstream transmission strength of the abnormal source roller in the multi-roller force chain is as follows: the set of leveling rollers with sequence numbers less than or equal to the abnormal source roller number is designated as the upstream region, and the set of leveling rollers with sequence numbers greater than the abnormal source roller number is designated as the downstream region. The directional division between the upstream and downstream regions is based solely on the direction of material movement. For each leveling roller in the downstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated, and the absolute deviations of all leveling rollers in the downstream region are summed to obtain the downstream force deviation. Simultaneously, for each leveling roller in the upstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated, and the absolute deviations of all leveling rollers in the upstream region are summed to obtain the upstream force deviation. The downstream force deviation is divided by the upstream force deviation and the smallest positive value to obtain the abnormal amplification evaluation value.

[0015] Further, the specific process for determining whether an anomaly has an amplification trend is as follows: The anomaly amplification assessment value is compared with the attenuation threshold and the amplification threshold. When the anomaly amplification assessment value is greater than the amplification threshold, the anomaly is determined to have an amplification trend in the downstream roller group, and the anomaly amplification marker, the anomaly source roller sequence number, and the corresponding local force mutation value are output. When the anomaly amplification assessment value is within the range of the attenuation threshold and the amplification threshold, the anomaly is determined to be in a stable propagation state, and the stable propagation marker, the anomaly source roller sequence number, and the target convergence direction information corresponding to the anomaly source roller are output. When the fused force value is greater than the reference force value, the target convergence direction is the direction of decreasing force; when the fused force value is less than the reference force value, the target convergence direction is the direction of increasing force. When the anomaly amplification assessment value is less than the attenuation threshold, the anomaly is determined to gradually attenuate downstream, showing a self-reducing trend, and an anomaly attenuation marker is output, while only data monitoring is maintained. The second-order fluctuation value of the multi-roller force chain, the anomaly source segment, the anomaly source roller sequence number, the anomaly amplification assessment value, and the propagation trend marker, along with the operating condition label, are written into the operation predictive control database.

[0016] Furthermore, based on the abnormal source roll and its evolution characteristics, historical behavior patterns under the same working conditions are extracted, and the equivalent sensitivity relationship of the roll gap is inferred. The specific process for determining the reduction correction amount is as follows: When an abnormal amplification marker or a stable propagation marker is received, the target leveling roll set that needs to be reduced is determined according to the abnormal source roll sequence number and the geometric arrangement relationship of the leveling machine roll system. The target leveling roll set includes: the abnormal source roll, the adjacent leveling roll upstream of the abnormal source roll, and the adjacent leveling roll downstream. Historical working condition samples consistent with the current working condition label are selected from the operation predictive control database. The average values ​​of the fused force value sequence and the roll gap sequence of the historical leveling rolls are calculated to obtain the force reference value and the roll gap reference value. For each historical working condition sample, the roll sequence number is used as the alignment basis to... Extract the fusion force value and roll gap of the rolls with the same roll number as the target leveling roll set in the historical working condition sample. Calculate the force difference between the fusion force value and the force reference value, and the gap difference between the roll gap and the roll gap reference value. Multiply the force difference and gap difference to obtain the comprehensive deviation. Sum the comprehensive deviations of all historical working condition samples as the numerator. Sum the squares of the gap differences of all historical working condition samples and add them to the smallest positive value as the denominator. Divide the numerator by the denominator to obtain the roll gap equivalent stiffness value corresponding to each leveling roll in the target leveling roll set. For each leveling roll in the target leveling roll set, calculate the deviation between the current fusion force value and the force reference value, divide it by the roll gap equivalent stiffness value, multiply it by the abnormal amplification evaluation value, and take the negative number to obtain the roll gap correction value.

[0017] Furthermore, the specific process of adaptively updating the pressure correction amount under safety constraints and performing closed-loop control feedback is as follows: the roll gap correction value of each leveling roll in the target leveling roll set is sent to the corresponding roll gap adjustment actuator; when the roll gap correction value is greater than zero, the roll gap of the leveling roll is increased; when the roll gap correction value is less than zero, the roll gap of the leveling roll is decreased; at the same time, a safety constraint is applied to the roll gap correction value of each leveling roll: if the absolute value of the roll gap correction value exceeds the upper limit of the single roll gap adjustment allowed by the leveling machine, the roll gap correction value is limited to the allowable range; if the adjusted roll gap exceeds the mechanical limit gap range... If the range is too large, the adjustment amount is truncated according to the mechanical limit; at the same time, a smoothing constraint is applied to the difference in roll gap correction values ​​between the upstream roll, the abnormal source roll, and the downstream roll; the changes in the adjusted roll gap, the fusion force value, and the overall force distribution of the roll system are fed back to the multi-roll force chain digital twin model for model parameter updates; and the target convergence direction is determined based on the adjusted fusion force value. If the fusion force value moves towards the target convergence direction, the adjustment is deemed effective; otherwise, it is recorded as a negative adjustment sample and the adjustment amplitude of subsequent similar working conditions is reduced. The adjustment and feedback results, as well as the corresponding working condition labels, are written into the operation predictive control database.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention constructs a digital twin model of a multi-roller force chain that includes contact stiffness, bending elements and adaptive boundaries, and drives online correction with real-time straightening physical data. This enables the global dynamic distribution of the force chain between the rollers, breaking through the limitation of traditional equipment that can only observe the pressure of a single roller, and realizing continuous, real-time and quantifiable characterization of the coupled force behavior of multiple rollers.

[0019] (2) This invention, through the local mutation feature index, multi-roller second-order fluctuation degree and anomaly amplification evaluation model constructed based on the fusion force sequence, can capture transient force chain mutations and identify abnormal source rolls on the spatial coordinates along the plate, and further quantify the propagation intensity of the anomaly upstream and downstream, so as to accurately judge the evolution path of the abnormal waveform and enable the anomaly to be discovered and explained in advance.

[0020] (3) This invention, through cross-condition alignment of samples under the same working condition, construction of comprehensive deviation and solution of equivalent stiffness, can learn the roll gap force relationship of different roll sequences and obtain the roll gap equivalent sensitivity of the target roll, so that the amount of reduction correction no longer depends on human experience, but is automatically inferred from historical behavior patterns, realizing the quantification, self-adaptation and low human dependence of reduction compensation.

[0021] (4) This invention ensures the safety and reliability of the pressing adjustment process by combining multiple protection mechanisms such as abnormal evolution state, safety limit, mechanical limit and smooth constraint between rolls when performing roll gap compensation; and by feeding back the adjusted state to the digital twin model, the model parameters are updated and the adjustment effect is reversed, so that the leveling control forms a closed-loop adaptive mechanism, which significantly improves the stability of plate shape quality and the robustness of leveling control.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 The flowchart shows the method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model. Figure 2 Flowchart for the construction and online calibration of a digital twin model of a multi-roller force chain; Figure 3 This is a bar chart showing the distribution of sudden changes in local force values ​​in a multi-roller force chain. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, 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.

[0025] Please see Figures 1-3 This invention provides a technical solution: a method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model, such as... Figure 1 As shown, the process includes the following steps: S1, real-time acquisition of leveling physical data, and completion of time synchronization, spatial mapping, denoising and smoothing, normalization and working condition labeling to obtain preprocessed leveling physical data; S2, establishment of a multi-roller force chain model based on the preprocessed leveling physical data, forming a force chain structure through contact stiffness, adaptive boundary and bending unit, and driving continuous correction of the model with real-time leveling physical data, while outputting a fused force sequence; S3, analysis of local mutation characteristics of multi-rollers based on the fused force sequence, identification of abnormal source rolls based on the local mutation characteristics of multi-rollers, evaluation of the upstream and downstream transmission strength of abnormal source rolls in the multi-roller force chain, and determination of whether the abnormality has an amplification trend; S4, extraction of historical behavior patterns under the same working conditions based on abnormal source rolls and evolution characteristics, inference of the equivalent sensitivity relationship of roll system gap, determination of the pressing correction amount, adaptive update of the pressing correction amount under safety constraints, and closed-loop control feedback.

[0026] Specifically, the process of acquiring and preprocessing the physical data of the leveling machine in real time, including time synchronization, spatial mapping, noise reduction and smoothing, normalization, and working condition labeling, is as follows: Real-time acquisition of physical data includes: the normal load and roll gap of each leveling roller, the speed of the sheet material, the position of the sheet material passing through, the structural parameters of the leveling machine, and the static parameters of the leveling machine equipment. The normal load of each leveling roller is acquired by a pressure sensor installed on the roller support, reflecting the contact pressure between the roller and the sheet material in real time. The roll gap of each leveling roller is acquired by an encoder to record changes in the mechanical clearance between rollers. The speed of the sheet material is acquired by a speed encoder installed on the frame side. The position of the sheet material passing through is obtained by integrating the speed signal over time. The structural parameters of the leveling machine include: the number of rollers, roller diameter, roll gap, roller shape, effective length of the roller body, support method, and the spatial arrangement of each roller in the frame reference coordinate system, used to describe the geometric and mechanical characteristics of the leveling system. The static parameters of the leveling machine include: the inherent stiffness of each leveling roller support system, the frame constraint boundary conditions, the maximum adjustment stroke of the actuator, the mechanical limits of the roller gap adjustment mechanism, the rated load range of the leveling machine, and the basic calibration parameters of the roller support assembly. These parameters characterize the inherent static mechanical properties of the equipment. The leveling physical data is synchronized according to the equipment's bus timestamp; the bus timestamp is uniformly issued by the leveling machine control system, ensuring that multi-source data have the same time reference. Time synchronization avoids data misalignment caused by inconsistent sensor sampling delays, ensuring that the leveling physical data at each moment accurately reflects the current stress and motion state of the sheet metal. To ensure the continuity and physical meaning of spatial coordinates, the plate speed is integrated to map the time point to the actual physical position of the plate within the leveling area. The plate position is then discretized using a fixed spatial offset length, which is determined based on the equipment sampling rate and plate speed, preferably 2 to 10 mm. The original continuous spatial position is divided into multiple equidistant spatial coordinates. The leveling physical data is mapped from the time dimension to a spatial coordinate sequence along the plate's movement direction, so that each discrete spatial coordinate corresponds to a force-bearing section of the plate as it passes through the leveling roller group. The normal load, roller gap, and motion state of each leveling roller at the same plate position are aligned to form a complete multi-roller synchronous force data. A moving average filter is used to suppress and smooth the leveling physical data. The length of the sliding window is adaptively determined based on the sensor noise level and plate speed to balance noise suppression and dynamic response capabilities, ensuring that force fluctuation information is not lost due to excessive smoothing.The physical data of the leveling rolls is z-score normalized to avoid model training and calculation biases caused by differences in dimensions. Simultaneously, based on steel grade, plate thickness, and plate speed information, a working condition label is generated for each piece of physical data. This label distinguishes data distribution differences under different production conditions and provides precise filtering conditions in subsequent steps such as anomaly identification, historical sample matching, and target reduction correction calculation. This allows the system to compare and learn under the same working conditions, thereby improving prediction accuracy and control effectiveness. An operational predictive control database is established, storing the preprocessed physical data along with corresponding spatial coordinates and working condition labels. The database structure includes a time index, a spatial coordinate index, a leveling roll number index, and a working condition label index, ensuring rapid data retrieval and access.

[0027] In this implementation scheme, through time synchronization, spatial mapping, noise reduction, smoothing, and normalization, the normal load, roll gap, and plate motion state during the leveling process are uniformly aligned to fixed spatial coordinates, ensuring consistency and comparability of multi-roll force information on the same stress cross section. Furthermore, by combining steel grade, plate thickness, and speed, operating condition labels are generated to differentiate data under different production conditions. The resulting predictive control database has a clear data structure and strong searchability, providing a high-quality, accurate, and traceable data foundation for subsequent multi-roll force chain modeling, local mutation identification, anomaly propagation assessment, and reduction correction calculation.

[0028] Specifically, a multi-roller force chain model is established based on the preprocessed leveling physical data. The specific process of constructing the force chain structure through contact stiffness, adaptive boundaries, and bending elements is as follows: Based on the roller diameter, roller spacing, number of rollers, and roller shape structure in the leveling machine structural parameters, the roller center position and contact position of each leveling roller are determined in the frame reference coordinate system, and the geometric arrangement relationship of the multi-roller is constructed. The frame reference coordinate system is jointly formed by the leveling machine installation reference plane and the column centerline, ensuring that the spatial position of all rollers has a unified reference standard, so that the geometric arrangement relationship can accurately reflect the actual roller system structure. The equivalent vertical stiffness of each leveling roller is calculated based on the roller material parameters, effective roller length, and support method in the leveling machine structural parameters. The calculation of the equivalent vertical stiffness is based on an elasticity model, using the Young's modulus of the roller material, the moment of inertia of the section, and the support boundary conditions for solution. The results are used to describe the force-displacement response characteristics of each leveling roller under vertical load, providing a foundation for subsequently establishing the nodal force relationships. The contact points between the sheet metal and each leveling roller are discretized into force-bearing nodes. The equivalent bending stiffness of the sheet metal is calculated based on the sheet thickness and elastic modulus. Adjacent nodes are connected through sheet metal bending units. The discretization length of the force-bearing nodes is set according to the sheet thickness and the roller spacing of the leveling machine, preferably so that one node corresponds to the actual working range of one leveling roller, to ensure that the stress characteristics of the sheet metal can be accurately discretized and are countable. The calculation of the equivalent bending stiffness of the sheet metal is based on the classical plate-beam bending theory, which ensures that the mechanical meaning of the bending units between adjacent nodes is clear. To describe the contact relationship between the sheet metal and the rollers, contact stiffness elements are introduced at the stress nodes. The preferred contact stiffness is determined based on Hertz contact theory, which calculates the displacement response under unit contact pressure based on the roller surface curvature radius, sheet metal thickness, elastic modulus of the materials on both sides, and Poisson's ratio. In actual calculations, a linear approximation is made near the current working condition to obtain an equivalent linear contact stiffness that can be directly used in finite element solutions. For working conditions with gaps, opening and closing judgment conditions can be superimposed before the contact stiffness. When the normal load is below the contact threshold, the contact stiffness is zero; when the normal load exceeds the contact threshold, it is used in the calculation as a linear contact stiffness, thus taking into account the influence of the roller gap gap on the continuity of the force chain. The equivalent vertical stiffness, sheet metal bending elements, and support boundary conditions determined by the static parameters of the leveling machine are integrated to construct a digital twin model of the multi-roller force chain. The support boundary conditions include the rigid constraints of the frame, the roller system support method such as double support or cantilever support, and the static limit position of the roller gap adjustment mechanism. Based on these parameters, the fixed, sliding, and elastic boundary forms of the nodes can be determined. For the ends of the plates at the entrance and exit of the leveling area, it is preferable to use simply supported or elastic simply supported boundaries for modeling. That is, the outer conveying roller of the leveling machine and the self-weight of the plate are equivalent to vertical elastic supports, and their stiffness is calculated according to the cross-sectional characteristics of the plate and the support spacing, thereby limiting the reasonable range of end rotation angle and displacement.After integration, a force chain finite element structure consisting of multiple force nodes and their connecting units can be established, realizing an accurate expression of the force transmission characteristics of the actual leveling process, and forming a multi-roller force chain digital twin model that can be used for prediction and correction.

[0029] In this implementation scheme, by constructing the geometric arrangement of the roller system based on the structural parameters of the leveling machine, calculating the equivalent vertical stiffness of each leveling roller, and discretizing the contact relationship between the sheet and the rollers into force-bearing nodes, this method forms a digital twin force model that can realistically reflect the force chain transmission characteristics during the leveling process. Under unified boundary condition constraints, this model systematically expresses the force coupling between multiple rollers, the bending response of the sheet, and the mechanical relationships between nodes. It can accurately describe the distribution and transmission law of load in the roller group, thus providing a physically consistent foundation for subsequent anomaly identification, force prediction, and pressure reduction control, significantly improving the model's prediction accuracy and the reliability of the control strategy.

[0030] Specifically, the process of continuously correcting the model using real-time straightening physical data and simultaneously outputting a fused force sequence is as follows: Figure 2 The diagram shows the flowchart for the construction and online calibration of the multi-roller force chain digital twin model. The normal loads of each leveling roll are mapped to corresponding nodes in the multi-roller force chain digital twin model. The normal load data is a real-time pressure signal after time synchronization and spatial alignment, accurately reflecting the stress magnitude on the sheet at different roll positions. The roll gaps of each leveling roll are used as the displacement boundary conditions of the multi-roller force chain digital twin model, realistically describing the geometric constraints between rolls and the overall opening variation. The normal load application location and the sheet stress area are determined using the sheet speed and through position. That is, based on the current position of the sheet obtained after spatial mapping, the load application point is accurately mapped to the stress node position where the sheet contacts each roll, thereby ensuring... The model's force distribution is consistent with the actual physical process. By comparing the deviation between the multi-roller force chain digital twin model's predicted multi-roller force changes and the actual normal load, the model automatically adjusts the contact stiffness, node transfer coefficient, and roll flexibility parameters in the multi-roller force chain digital twin model. This enables the model to maintain adaptability under different steel grades, plate thicknesses, incoming material shapes, and equipment conditions, achieving dynamic consistency correction for real working conditions. The model outputs the predicted normal load values ​​for each leveling roll and linearly fuses the predicted and measured normal load values ​​according to a weighting coefficient based on the prediction error to obtain the fused force values ​​for each leveling roll. Specifically: ,in, Indicates the fusion force value. Indicates the predicted value of the normal load. Indicates the measured normal load. The weighting coefficients are determined by constructing a normalized error index to adaptively allocate the weights between model predictions and measured values ​​based on the magnitude of the prediction error. When the prediction error is small, the model prediction is prioritized, giving it a higher weight and the measured value a lower weight. As the prediction error increases, the weight of the model prediction decreases while the weight of the measured value increases, making the fused result closer to the actual force level. To avoid completely discarding model prediction information when the error is large, the weight of the model prediction has a minimum lower limit, ensuring that the basic participation ratio is maintained under all circumstances, thus guaranteeing the continuity of the force chain field. The integrity of the system is not compromised; the fusion process effectively reduces errors caused by single sensor noise, local sliding interference, or dynamic impacts, making the output force sequence more stable and reliable. The prediction error is measured using feasible numerical forms, including: preferably using the L2 norm (sum of squared errors) as the main error index to improve sensitivity to samples with large deviations; switching to Huber loss to enhance robustness when local impacts or burr-type noise are present; and using the L1 norm to reduce the impact of outliers when rapidly correcting local deviations of a single roller, thus enabling flexible adaptation of error assessment to different working conditions. Error calculation is based on the normal load data within the latest spatial coordinate range, preferably using a sliding window statistical analysis of 3 to 10 force sections to balance real-time performance and stability. Simultaneously, the system continuously reads the latest leveling physical data and compares it in real time with the prediction results of the multi-roller force chain digital twin model. Based on the prediction deviation, it updates the equivalent vertical stiffness, contact stiffness, and support boundary condition parameters of each leveling roll in the model, maintaining the model's online learning capability. This allows for automatic parameter correction when equipment vibration, temperature rise, support wear, and other conditions change, achieving stable long-term prediction performance. Parameter updates use a decrease in prediction error as the convergence criterion; if the error does not decrease, the parameters from the previous cycle are retained. The model calibration frequency is adaptively set based on the equipment sampling rate and the plate speed to avoid rapid deformation causing a disconnect between prediction and actual force. When the operating conditions are stable and the prediction error remains below the threshold, the calibration cycle can be extended, thereby reducing the computational load. All calibration cycles are executed through a fixed scheduler to ensure stable operation of the prediction, fusion, and parameter update processes.

[0031] In this implementation scheme, by continuously mapping and correcting real-time leveling physical data to a multi-roller force chain digital twin model, the model can adaptively update contact stiffness, node transmission coefficient, and boundary conditions according to actual force changes, thereby maintaining dynamic consistency with real working conditions. Through weighted fusion of predicted results and measured values, sensor noise and local impact interference can be effectively suppressed, improving the stability and reliability of the force sequence. The fused force values ​​retain the continuity of the overall force chain structure of the model and reflect the real instantaneous force changes, making subsequent mutation identification, abnormal source roll positioning, and pressure correction calculation more accurate and reliable, laying the foundation for building a leveling force chain prediction and control system that can operate online for a long time.

[0032] Specifically, the process of analyzing the local abrupt change characteristics of multi-rollers based on the fused force sequence is as follows: At each spatial coordinate along the direction of material movement, the fused force value sequence is called, and the average value is calculated to obtain the average force value of the multi-rollers. Each spatial coordinate corresponds to a force cross section of the material passing through the leveling roll group. The cross section contains the synchronous fused force values ​​of all leveling rolls at the current position, thus ensuring that the data has spatial consistency and comparability. The fused force value of the (i+2)th roll is subtracted from twice the fused force value of the (i+1)th roll, and then added to the fused force value of the ith roll to obtain the second-order difference value of the multi-roller force chain. The second-order difference can reflect the local curvature change of the force between adjacent roll positions, and can sensitively capture the abrupt force change characteristics caused by misaligned rolls, local hard points, and abrupt changes in material shape. The local stress mutation value is obtained by dividing the second-order difference value of the multi-roll force chain by the average stress value of the multi-rollers, eliminating the influence of differences in steel grades, plate thicknesses, and overall load levels, making the mutation index more comparable. Based on the number of rolls at the stress section, the local stress mutation values ​​of all combinations are squared and summed, then divided by the number of combinations to obtain the overall fluctuation value. The number of combinations is automatically determined by the number of rolls, ensuring that the fluctuation calculation results of a uniform structure can still be obtained when the number of rolls changes, and the local mutations are quantified as a whole. The square summation method can amplify the contribution of significant mutation points to the overall fluctuation, making abnormal segments easier to identify. The square root of the overall fluctuation value is used to obtain the second-order fluctuation value of the multi-roll force chain, providing a quantitative basis for subsequent identification of abnormal source rolls and amplification trend assessment.

[0033] The specific formula for the second-order variability value of the multi-roller force chain is as follows: ; In the formula, This represents the second-order variability value of the multi-roller force chain, used to measure the degree of local discontinuity of the multi-roller force chain at a certain spatial coordinate in the direction of sheet movement; Indicates the number of rollers; This represents the fusion force value of the i-th roller; This represents the fusion force value of the (i+1)th roller; This represents the fusion force value of the (i+2)th roller; This indicates the average force value across multiple rollers.

[0034] In this embodiment, Table 1 is a data table of local stress mutation values. When the average stress value of the multi-rollers is 100, the table records in detail the fusion stress value of the i-th roll, the fusion stress value of the (i+1)-th roll, the fusion stress value of the (i+2)-th roll, and the local stress mutation value corresponding to the five three-roller sequence combinations; among them, the fusion stress value of the i-th roll corresponding to combination 1 is 92.7, the fusion stress value of the (i+1)-th roll is 101.3, the fusion stress value of the (i+2)-th roll is 119.8, and the local stress mutation value is 0.99; the fusion stress value of the i-th roll corresponding to combination 2 is 101.5, the fusion stress value of the (i+1)-th roll is 121.8, the fusion stress value of the (i+2)-th roll is 149.2, and the local stress mutation value is 0.0. 71; The fusion force value of the i-th roller corresponding to combination 3 is 119.6, the fusion force value of the (i+1)-th roller is 110.4, the fusion force value of the (i+2)-th roller is 99.7, and the local force mutation value is −0.015; The fusion force value of the i-th roller corresponding to combination 4 is 101.1, the fusion force value of the (i+1)-th roller is 79.6, the fusion force value of the (i+2)-th roller is 70.9, and the local force mutation value is 0.128; The fusion force value of the i-th roller corresponding to combination 5 is 131.4, the fusion force value of the (i+1)-th roller is 100.8, the fusion force value of the (i+2)-th roller is 89.5, and the local force mutation value is 0.193.

[0035] Table 1 Data on Local Stress Sudden Changes

[0036] like Figure 3 The image shows a bar chart illustrating the distribution of local force abrupt changes in a multi-roller force chain. It displays the local force abrupt changes calculated at the current position for five sets of three-roller force sequences. The horizontal axis represents the sequence combination number, and the vertical axis represents the corresponding degree of local force abrupt change. The bar height reflects the second-order difference normalization result of the three-roller force changes in different combinations, used to quantify the degree of local discontinuity in the multi-roller force chain. (Based on Table 1 and...) Figure 3 It can be seen that combination 1 has the highest mutation value, much higher than other combinations, indicating that the corresponding three-roller stress sequence has the most significant local stress mutation, which is a typical feature of the abnormal source segment. The mutation values ​​of combinations 2, 4, and 5 are relatively small. Although there is some discontinuity in stress, the amplitude is mild and can be regarded as mild disturbance or initial anomaly. The mutation value of combination 3 is negative, indicating that the overall three-roller stress sequence is smooth and without significant mutation, which is a normal operating stress state. The overall distribution shows strong non-uniformity: only a few sequences have drastic mutations, while the rest of the sequences change slowly, which is a physical feature of local anomalies and chain transmission during the straightening process.

[0037] In this implementation scheme, by performing second-order difference analysis on the fused force sequence in discrete spatial coordinates and constructing normalized local force mutation values, local abnormal changes in the multi-roll force chain can be stably identified under different plate speeds, different overall load levels, and different steel grades. At the same time, the overall fluctuation formed by the sum of squares can effectively amplify the disturbance of abnormal roll position on the overall force chain, enabling rapid and accurate capture of force anomalies caused by plate shape mutations, misaligned rolls, and local hard points, improving the sensitivity and robustness of anomaly detection, and providing a reliable quantitative basis for subsequent anomaly source roll location and amplification trend judgment.

[0038] Specifically, the process of identifying the abnormal source roller based on the local abrupt change characteristics of the multi-roller is as follows: For each spatial coordinate discretized along the direction of plate movement, the second-order fluctuation value of the multi-roller force chain is calculated and compared with the fluctuation threshold; when the second-order fluctuation value of the multi-roller force chain is greater than the fluctuation threshold, it is determined that there is an abrupt change in the force chain at the current spatial position, and the current spatial position is marked as the abnormal source segment. At this time, all three-roller force combinations of the corresponding section are called, the local force abrupt change values ​​are analyzed one by one, and the maximum absolute value of the local force abrupt change value in the combination is selected, and the (i+1)th value is selected. Roller is used as the source of the abnormality, and the sequence number of the source roller is obtained. Among them, i, i+1, and i+2 are three-roller sequences arranged in the order of rollers. The larger the absolute value of the local force mutation value, the stronger the disturbance of the roller position on the force chain. Therefore, the roller with the largest absolute value is selected as the source roller of the abnormality, which can ensure that the abnormality location has the most direct mechanical basis. At the same time, the second-order difference of the three-roller sequence essentially reflects the discontinuity of the force on the middle roller, i.e., the i+1 roller, relative to the front and rear rollers. When the local force mutation value is the largest, it indicates that the disturbance of the middle roller on the force chain is the strongest. Therefore, the i+1 roller is determined to be the source roller of the abnormality. To avoid misjudgments due to transient impacts and local measurement noise caused by multiple nearby sources, spatial connectivity and abrupt change characteristic persistence constraints are further introduced during the identification of abnormal source rolls. Specifically, it is required that among a series of spatial coordinates along the plate movement direction, preferably 2 to 5, the absolute value of the local force abrupt change corresponding to the same three-roll combination is the largest, and the corresponding (i+1)th roll is continuously identified as a candidate abnormal source roll within the spatial interval. Only then is the roll sequence ultimately confirmed as the abnormal source roll of the spatial segment, thus filtering out false triggering phenomena caused by single-point extreme values ​​and multiple nearby sources. Simultaneously, the predictive control database is invoked, historical samples with consistent working condition labels are matched, and the median of the historical fused force values ​​for each leveling roll is selected as the reference force value. The working condition labels include parameters such as steel grade, plate thickness, and plate speed, ensuring that the reference force value remains consistent with the current working condition. The median is used to reduce the influence of extreme samples, improve the stability and representativeness of the benchmark force, and thus provide a reliable basis for subsequent anomaly amplification assessment and pressure correction.

[0039] In this implementation scheme, the abnormal source roll is accurately identified by locating the abrupt force changes at each spatial cross-section. Based on the mechanical characteristics reflecting the discontinuity of force on the intermediate roll using second-order difference, the (i+1)th roll with the largest absolute value of local abrupt changes is identified as the abnormal source roll. Simultaneously, by combining historical samples with operating condition tags and extracting the median reference force, the anomaly identification process exhibits stronger noise resistance, consistency, and operating condition adaptability. This mechanism enables high-confidence localization of the anomaly source, providing stable and reliable basic data for subsequent anomaly propagation assessment and compression correction.

[0040] Specifically, the process for evaluating the upstream and downstream transmission strength of the abnormal source roller in the multi-roller force chain is as follows: the set of leveling rollers with sequence numbers less than or equal to the abnormal source roller number is designated as the upstream region, and the set of leveling rollers with sequence numbers greater than the abnormal source roller number is designated as the downstream region. The directional division of the upstream and downstream regions is based solely on the actual movement direction of the material. When the material is being fed in reverse or retracting, the upstream and downstream regions are automatically redefined based on the real-time movement direction of the material to ensure that the force transmission analysis is always consistent with the physical process. The upstream region corresponds to the force transmission path of the material before it passes the abnormal source roller, and the downstream region corresponds to the force response area of ​​the material after it passes the abnormal source roller, reflecting whether the abnormality propagates backward along the force chain. The division of the upstream and downstream regions is automatically completed based on the roller sequence number, ensuring compatibility with leveling machines of different structural forms. For each leveling roll in the downstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated. The absolute deviations of all leveling rolls in the downstream region are summed to obtain the downstream force deviation. At the same time, for each leveling roll in the upstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated. The absolute deviations of all leveling rolls in the upstream region are summed to obtain the upstream force deviation. The reference force value is derived from the median trend of historical working conditions and can serve as a force benchmark when there are no abnormalities. Therefore, the larger the deviation, the stronger the abnormal disturbance. The summation of the absolute deviations can reflect the degree of force offset in the overall region without being affected by the randomness of individual roll fluctuations, making the measurement of abnormal propagation intensity more stable and robust. The anomaly amplification assessment value is obtained by dividing the downstream force deviation by the upstream force deviation and the minimum positive value. The minimum positive value is used to avoid the division by zero problem when the upstream deviation is close to zero, thus improving the stability of the calculation. An anomaly amplification assessment value greater than 1 indicates that the downstream deviation is relatively larger, that is, the anomaly has a tendency to be amplified. An assessment value close to 1 indicates that the anomaly is stably transmitted. An assessment value less than 1 indicates that the downstream disturbance is smaller than the upstream disturbance, and the anomaly has a tendency to be attenuated.

[0041] The specific formula for the abnormal amplification assessment value is as follows: ; In the formula, This represents the amplification assessment value, used to measure the degree of amplification of abnormal forces as they propagate from upstream to downstream in a multi-roller force chain. Indicates the roller sequence number of the abnormal source; This represents the fusion force value of the j-th leveling roller in the downstream region; This represents the reference force value corresponding to the j-th leveling roller; This represents the fusion force value of the i-th leveling roller in the upstream region; The reference force value corresponding to the i-th leveling roller; Represents a very small positive value, taking values ​​of .

[0042] In this implementation scheme, by using the abnormal source roller as the dividing point, the degree of force deviation in the upstream and downstream regions is quantified, and the ratio of the two is used to construct an amplified evaluation value for the abnormality, thus achieving a quantifiable determination of the propagation trend of the abnormality along the multi-roller force chain. This method can simultaneously reflect the local disturbance intensity and overall transmission characteristics of the abnormality, avoiding misjudgments caused by relying solely on single-point force. By fusing the deviation comparison between the force value and the reference force value, the evaluation process has good stability and noise resistance. Combined with the introduction of extremely small positive numbers, the calculation is guaranteed to have numerical stability and feasibility under various operating conditions, thereby providing a reliable basis for subsequent pressure reduction compensation strategies and improving the timeliness and accuracy of abnormality control.

[0043] Specifically, the process for determining whether an anomaly has an amplification trend is as follows: The anomaly amplification assessment value is compared with an attenuation threshold and an amplification threshold. These thresholds, set based on historical sample statistics and equipment characteristics, reflect the natural attenuation range and the boundary between the anomaly amplification range of the force chain disturbance, and are used to classify the propagation characteristics of anomalies of different intensities. The comparison process is executed in real-time by the control system at each stress section, ensuring timely identification of the anomaly propagation trend. When the anomaly amplification assessment value exceeds the amplification threshold, the anomaly is determined to have an amplification trend in the downstream roller assembly, and an anomaly amplification marker is output. The abnormal source roll number and its corresponding local force mutation value are immediately triggered, prompting subsequent compensation logic to quickly suppress the abnormality before it amplifies further. When the abnormality amplification evaluation value is within the attenuation threshold and amplification threshold range, the abnormality is determined to be in a stable propagation state. A stable propagation marker, the abnormal source roll number, and the target convergence direction information corresponding to the abnormal source roll are output. The target convergence direction is determined by comparing the fused force value with the reference force value and is used to guide the next roll gap adjustment direction, giving the control a clear mechanical convergence direction. When the fused force value is greater than the reference force value, the target convergence direction is to decrease the force. The direction of the force; when the fused force value is less than the reference force value, the target convergence direction is the direction of increasing the force; to avoid frequent switching of the target convergence direction between the fused force value and the reference force value within a small fluctuation range, a convergence tolerance band is set when judging the relationship between the two; when the absolute value of the difference between the two is lower than the tolerance band threshold, preferably 1% to 3% of the reference force value, the convergence direction of the previous sampling period remains unchanged, thereby avoiding jitter in direction determination caused by sensor noise or board perturbation, and improving the stability and continuity of the control action; at this time, the subsequent compensation logic is also triggered; when abnormal amplification evaluation When the value is less than the attenuation threshold, the anomaly is determined to be gradually attenuating downstream, showing a self-reducing trend. An anomaly attenuation flag is output, and only data monitoring is maintained. In such cases, no pressure intervention is required, which can reduce unnecessary adjustment actions and avoid over-control. The second-order fluctuation value of the multi-roller force chain, the anomaly source segment, the anomaly source roll sequence number, the anomaly amplification evaluation value, and the propagation trend flag, along with the working condition label, are written into the operation predictive control database. This ensures that each anomaly identification event is saved in a traceable form, which is convenient for subsequent roll gap equivalent sensitivity inference and behavior pattern learning, and achieves continuous optimization and robust convergence in long-term operation.

[0044] In this implementation scheme, an anomaly propagation judgment mechanism constrained by attenuation and amplification thresholds is constructed to accurately identify the amplification, stabilization, or attenuation trends of abnormal forces in a multi-roller force chain. During the stable propagation stage of anomalies, the target convergence direction is further output, giving subsequent roll gap control a clear mechanical orientation. During the attenuation stage of anomalies, unnecessary control actions are avoided, thereby reducing the frequency of intervention and improving system stability. At the same time, the complete anomaly identification results are written into the operation predictive control database, which can continuously accumulate behavioral patterns and strengthen subsequent adjustment strategies during long-term operation, significantly improving the accuracy of anomaly location, the rationality of control response, and the overall stability of the leveling process.

[0045] Specifically, based on the abnormal source roll and its evolution characteristics, the process of extracting historical behavior patterns under the same working conditions, inferring the equivalent sensitivity relationship of the roll gap, and determining the reduction correction amount is as follows: When an abnormal amplification mark or a stable propagation mark is received, the target leveling roll set that needs to be reduced is determined according to the abnormal source roll sequence number and the geometric arrangement relationship of the leveling machine roll system. The target leveling roll set includes: the abnormal source roll, the adjacent leveling roll upstream of the abnormal source roll, and the adjacent leveling roll downstream, with the neighborhood radius limited to ±1. Historical working condition samples consistent with the current working condition label are selected from the operation prediction and control database. The average values ​​of the fused force value sequence and the roll gap gap sequence of the historical leveling rolls are calculated to obtain the force reference value and the roll gap gap reference value. Among them, the historical working condition samples are selected with completely consistent steel grade, plate thickness, and plate speed working condition labels to ensure that the true statistical law of the roll gap force relationship is obtained under the same material and working conditions. The force reference value and the gap reference value are calculated using the mean value, which can stably reflect the central state in the historical behavior pattern and avoid the reference deviation caused by a single abnormal sample. For each historical working condition sample, using the roll sequence number as the alignment basis, the fused force value and roll gap of the rolls with the same roll sequence number as the target leveling roll set in the current working condition are extracted. The force difference between the fused force value and the force reference value, and the gap difference between the roll gap and the roll gap reference value are calculated respectively. The force difference and the gap difference are multiplied to obtain the comprehensive deviation. The comprehensive deviations of all historical working condition samples are summed as the numerator. The squares of the gap differences of all historical working condition samples are summed and added to the smallest positive value as the denominator. The numerator is divided by the denominator to obtain the equivalent stiffness value of the roll gap corresponding to each leveling roll in the target leveling roll set. Among them, the comprehensive deviation is used to reflect the sensitivity of roll gap changes to force changes, that is, the empirical statistical significance of equivalent stiffness. Using the square of the gap difference as the denominator can avoid the cancellation of positive and negative values, making the result more stable. Adding the smallest positive number avoids the division by zero problem, making the algorithm computable in the actual engineering environment. The equivalent stiffness value essentially describes the structural stiffness and force chain response characteristics of each leveling roll under operating conditions, and can be used to guide the actual adjustment range. For each leveling roll in the target leveling roll set, the deviation between the current fused force value and the force reference value is calculated, divided by the roll gap equivalent stiffness value, and then multiplied by the anomaly amplification evaluation value and the inverse number is taken to obtain the roll gap correction value.The calculation involves dividing by the equivalent stiffness value of the roll gap to represent the roll position's response sensitivity to roll gap adjustment; then multiplying by the anomaly amplification evaluation value to dynamically amplify or suppress the adjustment amplitude; simultaneously, it avoids introducing additional disturbances due to compensation applied to the far-end roll. The anomaly amplification evaluation value participates as a dynamic adjustment intensity factor in the compensation calculation, but to avoid excessive adjustment in extreme cases, a saturation constraint is applied before using the anomaly amplification evaluation value, preferably limited to between 0.5 and 2.0. This reflects the intensity of anomaly propagation without causing excessive compression or oscillating control, thus ensuring the controllability of the adjustment action and the safety of engineering execution; finally, taking the opposite number represents the actual adjustment performed according to the force chain convergence direction, thereby causing the force state to return to the reference direction. This makes the compression compensation amount both physically meaningful and quantifiablely adaptive.

[0046] The specific formula for the roll gap correction value is as follows: ; In the formula, This represents the roll gap correction value. It comprehensively calculates the roll gap correction amount that needs to be adjusted for each target leveling roll, ensuring that the abnormality gradually converges and returns to the normal stress state. This indicates the fusion force value corresponding to the current leveling roller; Indicates the reference value of the force; Indicates the reference value for the roll gap; This represents the fused stress value corresponding to the historical working condition sample; This indicates the roll gap corresponding to the historical operating condition sample; Indicates the number of historical operating condition samples; Indicates the abnormally amplified assessment value; Represents a very small positive value, taking values ​​of .

[0047] In this implementation plan, a quantifiable equivalent stiffness of the roll gap is constructed by extracting statistical sensitivity relationships of roll gap forces from historical samples under the same working conditions. This allows for the automatic calculation of the reduction correction amount for each roll based on the current force deviation. Combined with anomaly amplification evaluation values, the adjustment range can be dynamically amplified or suppressed, ensuring that the reduction compensation is based on the true force chain response and has adaptive capabilities that change with working conditions. This mechanism significantly improves the accuracy and stability of the compensation strategy, avoids over-adjustment or under-adjustment problems, makes the leveling process more controllable and reliable, and can continuously maintain the consistency of sheet shape quality and the stability of the production process.

[0048] Specifically, the process of adaptively updating the pressure correction amount under safety constraints and performing closed-loop control feedback is as follows: The roll gap correction value of each leveling roll in the target leveling roll set is sent to the corresponding roll gap adjustment actuator. The roll gap adjustment actuator is an electric servo cylinder or a hydraulic servo adjustment unit, which can achieve precise fine-tuning of the roll gap opening. When the roll gap correction value is greater than zero, the roll gap of the leveling roll is increased; when the roll gap correction value is less than zero, the roll gap of the leveling roll is decreased. At the same time, a safety constraint is applied to the roll gap correction value of each leveling roll: if the absolute value of the roll gap correction value exceeds the allowable single roll gap of the leveling machine... Adjusting the upper limit restricts the roll gap correction value to the allowable range. The upper limit for a single roll gap adjustment is provided by the equipment manufacturer and is usually in the range of 0.02 to 0.10 mm to prevent transient impacts and mechanical interference caused by over-adjustment. If the roll gap exceeds the mechanical limit clearance range after adjustment, the adjustment amount is cut off according to the mechanical limit. The mechanical limit clearance range is determined by the equipment structure and includes the minimum closing gap and the maximum opening gap. The cut-off strategy can effectively prevent equipment damage, roll neck overload or plate breakage risks caused by excessive pressing or excessive opening, and ensure that the adjustment amount is always within the safe physical range. Simultaneously, a smoothing constraint is applied to the difference in roll gap correction values ​​between the upstream roll, the abnormal source roll, and the downstream roll. The smoothing constraint is used to avoid transient jumps in the force chain caused by excessive differences in the adjustment amounts of adjacent rolls. Specifically, an upper limit for the difference in correction amounts between adjacent rolls is set, preferably 0.01 to 0.05 mm. When the correction value of the abnormal source roll changes significantly while the correction value of the upstream or downstream roll changes less significantly, if the correction difference between the two rolls exceeds the upper limit for the difference in correction amounts between adjacent rolls, the excess portion is compressed to ensure that it does not exceed the threshold. This avoids sudden interruptions in the force chain or step-like disturbances within the roll system, ensuring that the impact of the pressing adjustment action on the plate shape and force curve is more stable and controllable. The adjusted roll gap, fused force value, and overall force distribution of the roll system are fed back to the multi-roll force chain digital twin model for parameter updates. This updates the model's internal contact stiffness, vertical flexibility, and boundary constraints in real time, ensuring the mechanical response of the multi-roll force chain digital twin model remains consistent with the physical equipment and enabling online model self-calibration. The target convergence direction is determined based on the adjusted fused force value. If the fused force value moves towards the target convergence direction, the adjustment is considered effective; otherwise, it is recorded as a negative adjustment sample, and the adjustment amplitude for subsequent similar operating conditions is reduced. The negative adjustment sample records potential over-adjustment sensitivity under the current operating condition, allowing for appropriate reduction of the adjustment gain in future similar operating conditions, forming an empirical adaptive control mechanism. The adjustment and feedback results, along with the corresponding operating condition labels, are written into the predictive control database, achieving closed-loop control and long-term self-learning capabilities.

[0049] In this implementation scheme, adaptive updates of the roll gap correction amount are achieved under safety constraints, and a feedback-based closed-loop control mechanism is constructed, enabling the pressing adjustment process to simultaneously possess safety, stability, and self-learning capabilities. By setting a single adjustment upper limit and mechanical limit cutoff, equipment impact and roll damage caused by over-adjustment are avoided; adjacent roll smoothing constraints ensure the continuity of the adjustment process and prevent abrupt changes in the force chain; and the adjusted roll gap state and fusion force changes are fed back in real time to the multi-roll force chain digital twin model, enabling online correction of contact stiffness and flexibility parameters, ensuring the model remains consistent with the real equipment over the long term. Simultaneously, the adjustment effect is automatically distinguished between effective and negative adjustments, forming accumulative experience adjustment samples to gradually optimize the adjustment amplitude under different operating conditions, improving the adaptive capability of the pressing strategy. Overall, this method significantly enhances the stability, safety, and intelligent control level of the leveling process, achieving continuously optimized self-closed-loop pressing control.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model, characterized in that, Includes the following steps: S1 collects leveling physical data in real time and completes time synchronization, spatial mapping, noise reduction and smoothing, normalization and working condition labeling to obtain preprocessed leveling physical data; S2, a multi-roller force chain model is established based on the pre-processed leveling physical data. The force chain structure is formed by contact stiffness, adaptive boundary and bending unit. The model is continuously corrected by real-time leveling physical data, and the fused force sequence is output. S3, based on the analysis of local mutation characteristics of multi-rollers through fusion force sequence, identify the abnormal source roll according to the local mutation characteristics of multi-rollers, evaluate the upstream and downstream transmission intensity of the abnormal source roll in the multi-roller force chain, and determine whether the abnormality has an amplification trend. S4 extracts historical behavior patterns under the same working conditions based on the abnormal source roll and its evolution characteristics, infers the equivalent sensitivity relationship of the roll system clearance, determines the reduction correction amount, realizes adaptive update of the reduction correction amount under safety constraints, and performs closed-loop control feedback.

2. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of acquiring real-time leveling physical data and completing time synchronization, spatial mapping, noise reduction and smoothing, normalization and working condition labeling to obtain preprocessed leveling physical data is as follows: Real-time acquisition of leveling physical data, including: normal load and roll gap of each leveling roller, plate speed, plate passing position, leveling machine structural parameters and leveling machine equipment static parameters; The leveling physical data is synchronized in time according to the bus timestamp of the equipment; the plate speed is integrated and the plate position is discretized with a fixed spatial discrete step length, mapping the leveling physical data from the time dimension to a spatial coordinate sequence along the plate movement direction, so that each discrete spatial coordinate corresponds to a force section of the plate when it passes through the leveling roller group, and the normal load, roller gap, and motion state of each leveling roller at the same plate position are aligned; moving average filtering is used to suppress noise and smooth the leveling physical data; the leveling physical data is normalized by z-score; at the same time, working condition labels are generated for each leveling physical data according to the steel grade, plate thickness, and plate speed information; an operation predictive control database is established, and the preprocessed leveling physical data with corresponding spatial coordinates and working condition labels is stored in the operation predictive control database.

3. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of establishing a multi-roller force chain model based on preprocessed leveling physical data, and constructing the force chain structure through contact stiffness, adaptive boundaries, and bending elements, is as follows: Based on the roller diameter, roller spacing, number of rollers and roller shape structure in the structural parameters of the leveling machine, the roller center position and contact position of each leveling roller are determined in the frame reference coordinates, and the geometric arrangement relationship of the multiple rollers is constructed. The equivalent vertical stiffness of each leveling roller is calculated based on the roller material parameters, effective length of the roller body, and support method in the structural parameters of the leveling machine; the contact points between the plate and each leveling roller are discretized into force nodes, and the equivalent bending stiffness of the plate is calculated based on the plate thickness and elastic modulus, so that two adjacent nodes are connected through the plate bending unit. By integrating the equivalent vertical stiffness, the plate bending unit, and the support boundary conditions determined by the static parameters of the leveling machine, a digital twin model of the multi-roller force chain is constructed.

4. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of driving the model to continuously correct itself using real-time straightening physical data, while simultaneously outputting a fused force sequence, is as follows: The normal load of each leveling roll is mapped to the corresponding node in the digital twin model of the multi-roller force chain. The roll gap of each leveling roll is used as the displacement boundary condition of the digital twin model of the multi-roller force chain. The normal load application position and the stress area of ​​the plate are determined by the plate speed and the plate passing position. By comparing the deviation between the multi-roller force change predicted by the multi-roller force chain digital twin model and the actual normal load, the contact stiffness, node transmission coefficient and roll body flexibility parameters in the multi-roller force chain digital twin model are automatically adjusted; the predicted normal load value of each leveling roll is output, and the predicted normal load value and the measured normal load are linearly fused according to the weighting coefficient based on the prediction error to obtain the fused force value of each leveling roll; Meanwhile, the latest leveling physical data is continuously read and compared in real time with the prediction results of the multi-roller force chain digital twin model. Based on the prediction deviation, the equivalent vertical stiffness, contact stiffness and support boundary condition parameters of each leveling roll in the model are updated.

5. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of analyzing the local mutation characteristics of multi-rollers based on fused stress sequence is as follows: At each spatial coordinate along the direction of plate movement, the fused force value sequence is called and the average value is calculated to obtain the average force value of the multi-roller; the fused force value of the (i+2)th roll is subtracted from twice the fused force value of the (i+1)th roll, and then the fused force value of the ith roll is added to obtain the second-order difference value of the multi-roller force chain; the second-order difference value of the multi-roller force chain is divided by the average force value of the multi-roller to obtain the local force mutation value; based on the number of rolls at the force section, the local force mutation values ​​of all combinations are squared and summed, and then divided by the number of combinations to obtain the overall fluctuation value; the square root of the overall fluctuation value is used to obtain the second-order fluctuation value of the multi-roller force chain.

6. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process for identifying the abnormal source roller based on the local mutation characteristics of multiple rollers is as follows: For each spatial coordinate discretized along the direction of plate movement, the second-order variability value of the multi-roller force chain is calculated and compared with the variability threshold. When the second-order variability value of the multi-roller force chain is greater than the variability threshold, it is determined that there is a sudden change in the force chain at the current spatial position. The current spatial position is marked as the abnormal source segment, and the maximum absolute value of the local force change value in the combination is selected. The (i+1)th roll is taken as the abnormal source roll, and the abnormal source roll sequence number is obtained. At the same time, the predictive control database is called to match historical samples with consistent working condition labels, and the median of the historical fused force value of each leveling roll is selected as the reference force value.

7. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process for evaluating the upstream and downstream transmission strength of the abnormal source roller in the multi-roller force chain is as follows: The set of leveling rolls with numbers less than or equal to the abnormal source roll number is designated as the upstream region, and the set of leveling rolls with numbers greater than the abnormal source roll number is designated as the downstream region. The directional division between the upstream and downstream regions is based solely on the direction of material movement. For each leveling roll in the downstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated. The absolute deviations of all leveling rolls in the downstream region are summed to obtain the downstream force deviation. Simultaneously, for each leveling roll in the upstream region, the absolute deviation between the fused force value and the corresponding reference force value is calculated. The absolute deviations of all leveling rolls in the upstream region are summed to obtain the upstream force deviation. The abnormal amplification assessment value is obtained by dividing the downstream force deviation by the upstream force deviation and the smallest positive value.

8. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process for determining whether an anomaly has an amplification trend is as follows: The abnormal amplification assessment value is compared with the attenuation threshold and the amplification threshold. When the abnormal amplification assessment value is greater than the amplification threshold, it is determined that the abnormality shows an amplification trend in the downstream roller group, and the abnormal amplification mark, the abnormal source roller number and the corresponding local stress change value are output. When the abnormal amplification evaluation value is within the range of the attenuation threshold and the amplification threshold, the abnormality is determined to be in a stable propagation state. The stable propagation mark, the abnormal source roller number, and the target convergence direction information corresponding to the abnormal source roller are output. When the fusion force value is greater than the reference force value, the target convergence direction is the direction of decreasing force; when the fusion force value is less than the reference force value, the target convergence direction is the direction of increasing force. When the abnormal amplification evaluation value is less than the attenuation threshold, it is determined that the abnormality is gradually attenuating downstream and showing a self-reducing trend. An abnormal attenuation mark is output, and only data monitoring is maintained. Write the second-order fluctuation value of the multi-roller force chain, the abnormal source segment, the abnormal source roll sequence number, the abnormal amplification evaluation value, and the propagation trend marker along with the operating condition label into the operation prediction control database.

9. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of extracting historical behavior patterns under the same working conditions based on the abnormal source roll and its evolution characteristics, inferring the equivalent sensitivity relationship of the roll system clearance, and determining the reduction correction amount is as follows: When an abnormal amplification mark or a stable propagation mark is received, the target set of leveling rolls that need to be pressed down is determined based on the abnormal source roll number and the geometric arrangement of the leveling machine roll system. The target set of leveling rolls includes: the abnormal source roll, the adjacent leveling roll upstream of the abnormal source roll, and the adjacent leveling roll downstream. Historical operating condition samples that match the current operating condition label are selected from the operation prediction and control database. The average values ​​of the fusion force value sequence and the roll gap sequence of the historical leveling rolls are calculated to obtain the force reference value and the roll gap reference value. For each historical working condition sample, using the roll sequence number as the alignment basis, the fusion force value and roll gap of the roll sequence number with the same number as the target leveling roll set in the current working condition are extracted from the historical working condition sample. The force difference between the fusion force value and the force reference value, and the gap difference between the roll gap and the roll gap reference value are calculated respectively. The force difference and the gap difference are multiplied to obtain the comprehensive deviation. The comprehensive deviations of all historical working condition samples are accumulated as the numerator. The squares of the gap differences of all historical working condition samples are accumulated and added to the smallest positive value as the denominator. The numerator is divided by the denominator to obtain the roll gap equivalent stiffness value corresponding to each leveling roll in the target leveling roll set. For each leveling roll in the target leveling roll set, calculate the deviation between the current fusion force value and the force reference value, divide it by the roll gap equivalent stiffness value, multiply it by the abnormal amplification evaluation value, and take the opposite number to obtain the roll gap correction value.

10. The method for predicting and controlling the operating status of a CNC leveling machine based on a digital twin model according to claim 1, characterized in that, The specific process of adaptively updating the compression correction amount under safety constraints and performing closed-loop control feedback is as follows: The roll gap correction value of each leveling roll in the target leveling roll set is sent to the corresponding roll gap adjustment actuator; when the roll gap correction value is greater than zero, the roll gap of the leveling roll is increased; when the roll gap correction value is less than zero, the roll gap of the leveling roll is decreased. At the same time, safety constraints are applied to the roll gap correction value of each leveling roll: if the absolute value of the roll gap correction value exceeds the upper limit of the roll gap adjustment allowed by the leveling machine in a single operation, the roll gap correction value is limited to the allowable range; if the roll gap after adjustment exceeds the mechanical limit clearance range, the adjustment amount is cut off according to the mechanical limit; at the same time, a smoothing constraint is applied to the difference in roll gap correction values ​​between the upstream roll, the abnormal source roll, and the downstream roll. The changes in the adjusted roll gap, fusion force value, and overall force distribution of the roll system are fed back to the digital twin model of the multi-roll force chain to update the model parameters. The target convergence direction is determined based on the adjusted fusion force value. If the fusion force value moves toward the target convergence direction, the adjustment is deemed effective. Otherwise, it is recorded as a negative adjustment sample, and the adjustment amplitude of subsequent similar working conditions is reduced. The adjustment and feedback results, as well as the corresponding working condition labels, are written into the operation predictive control database.

Images