Galvanized steel coil winding method and system based on data twinning

Abstract

The application discloses a galvanized steel coil winding method and system based on data twinning, specifically in the field of metallurgical industry automatic control and process control, comprising obtaining real-time production data of the same steel coil in the winding process, the real-time production data at least including strip line speed, tension roller feedback, winding motor current or torque feedback and roll diameter estimation, and outputting an outer signal sequence for subsequent processing; by applying a first contrast action and a second contrast action that meet the surface intervention amplitude constraint and are mutually mechanism-contrasted within the adjacent winding segments of the same steel coil, response characteristics such as tension recovery time and winding motor current texture intensity are extracted and differentially judged, consistency verification and rollback re-inspection are combined to output causes and confidence, and under an end-to-end collaborative architecture, the causes and confidence are written into a data twinning model to limit the amplitude and gradually update friction parameters or interlayer slip risk parameters to generate a subsequent control action sequence for closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of automatic control and process control technology in the metallurgical industry, and more specifically, to a method and system for coiling galvanized steel coils based on data twins. Background Technology

[0002] In the coiling process of hot-dip galvanized steel coils, the mainstream practice in the industry is to implement closed-loop control of tension and winding torque to suppress loose coiling, uncoiled coiling and end face defects. Usually, a tension set curve is established based on the coil diameter estimation, and parameter tuning and abnormal speed reduction are carried out in conjunction with winding motor current, tension roller feedback and a small amount of end face or deviation monitoring.

[0003] Taking the high-speed winding at the exit section of a continuous hot-dip galvanizing line as an example, the zinc layer on the strip surface is in a state prone to scratching, and the allowable compaction or tension intervention range is limited. At the same time, the continuous increase in coil diameter, oil film, and temperature cause the coefficient of friction to drift within the same coil. On-site, only external signals such as tension roller force, speed, and motor current can be stably obtained, making it difficult to directly observe the interlayer state within the coil. Under the above hard constraints, the mainstream practice will consistently reveal the bottleneck that "the external tension curve seems normal, but defects within the coil are still accumulating": In production, it can be repeatedly observed that the motor current texture gradually coarsens, the tension recovery time becomes longer, or the end face width trend slowly drifts. However, these phenomena are highly homologous between friction drift, strip shape fluctuation, and the initiation of interlayer micro-slippage, leading to control often targeting the wrong object; for example, the initiation of slippage is mistakenly regarded as a change in friction and only tension compensation is performed, which eventually manifests as a tower shape, collapsed edge, or interlayer scratches after several hundred turns and is irreversible.

[0004] The technical problem this application aims to solve is: under an end-to-end collaborative winding control architecture, how to distinguish between friction drift and interlayer micro-slippage initiation within the same roll in real time, and select the correct winding control action accordingly to prevent the accumulation of defects within the roll, without increasing the intervention range that may cause scratches to the galvanized surface and based solely on the existing available outer layer signals, in the context of an end-to-end collaborative winding control architecture. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a data twin-based method and system for winding galvanized steel coils. This method involves applying a first control action and a second control action that satisfy surface intervention amplitude constraints and serve as a mechanism comparison within adjacent winding segments of the same steel coil. It extracts and differentially distinguishes response features such as tension recovery time and winding motor current texture intensity. Combined with consistency verification and backtracking re-detection, it outputs the cause and confidence level, and writes this information into a data twin model to perform a limited, incremental update of friction parameters or interlayer slip risk parameters to generate a subsequent control action sequence for closed-loop control, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a galvanized steel coil winding method based on data twins, comprising: S1. Acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation. Output the outer layer signal sequence for subsequent processing. S2. Based on the outer signal sequence, the winding process of the steel coil is segmented according to a preset number of turns or a preset time window, the adjacent first winding segment and second winding segment are determined, and the segment identifier and corresponding segment data are output. S3. Apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually comparative in at least one of the tension given trajectory, the pinch or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. S4. Calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and output the first feature quantity and the second feature quantity. S5. Perform differential discrimination on the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and output the cause discrimination result. S6. Write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil.

[0007] In a preferred embodiment, S1 includes: S1-1. Using the sampling time of the strip line speed as the reference time axis, resample the tension roller feedback and the winding motor current or torque feedback to the reference time axis by nearest neighbor interpolation or linear interpolation, and output the time-aligned tension roller sequence and motor feedback sequence. S1-2. Calculate the cumulative number of winding turns corresponding to each sampling time based on the estimated roll diameter, and divide the tension roller sequence and motor feedback sequence after time alignment into segments according to the preset number of turns window. Write a segment identifier for each sampling point and output the outer signal sequence with the segment identifier. S1-3. Calculate the absolute value of the difference between adjacent sampling points in the motor feedback sequence within each loop segment and accumulate it in the loop segment to obtain the motor texture intensity feature quantity. At the same time, calculate the number of sampling points experienced by the tension roller sequence from the disturbance peak back to the steady state threshold to obtain the tension recovery time feature quantity, and output the outer layer sensitive feature sequence.

[0008] In a preferred embodiment, S2 includes: S2-1. Using the cumulative number of winding turns in the outer signal sequence as the segment coordinates, generate candidate segment boundaries according to the preset number of turns window and form a continuous set of candidate winding segments. At the same time, write the start number of turns, end number of turns and segment identifier for each candidate winding segment and output the candidate segment index table. S2-2. Calculate the boundary consistency index between the tension roller feedback and the current or torque feedback of the winding motor at the junction of two adjacent candidate winding segments. The boundary consistency index includes the mean difference and fluctuation amplitude difference between the preset sampling point intervals before and after the junction. When the boundary consistency index exceeds the preset threshold, the junction is moved forward or backward by no more than the preset number of turns to avoid the transition disturbance zone. Output the winding segment set after boundary correction. S2-3. Select two adjacent winding segments from the set of boundary-corrected winding segments in chronological order as the first winding segment and the second winding segment, respectively, and extract their corresponding outer signal subsequences as segment data. Output the first segment identifier, the second segment identifier, and the corresponding segment data.

[0009] In a preferred embodiment, S3 includes: S3-1. Read the segment data of the first winding segment and the second winding segment, as well as the corresponding estimated roll diameter and strip speed. Call the constraint set containing the upper limit of the allowable intervention range of the galvanized surface to generate a control action candidate set, and output the action candidate set and its range window. S3-2. Calculate the mechanism comparison scores of the first and second control actions based on the action candidate set. The mechanism comparison score is obtained by combining at least two of the following: the difference in the rate of change of the tension given trajectory, the difference in the pulse energy of the pinch roll pressure or pressure roll pressure given trajectory, and the difference in the holding time of the speed given trajectory. When the mechanism comparison score is lower than a preset threshold, the action candidate is replaced to improve the mechanism comparison. Output the first and second control actions that pass the mechanism comparison gating.

[0010] In a preferred embodiment, S3 further includes: S3-3. Before applying the first and second comparison actions, the accessibility consistency checker is calculated based on the torque margin of the winding motor and the response time of the clamping or pressure roller actuator. When the consistency check fails, the first and second comparison actions are downgraded to alternative actions with smaller amplitude and shorter duration. The comparison actions that pass the accessibility gate and the downgrade flag are output. S3-4. Apply the first reference action of accessibility gating to the first winding segment and simultaneously collect the feedback amount of the tension roller and the current or torque feedback amount of the winding motor to form the first response data. At the same time, apply the second reference action of accessibility gating to the second winding segment and simultaneously collect the feedback amount of the tension roller and the current or torque feedback amount of the winding motor to form the second response data. Write the degradation mark and action parameters into the reference action record table associated with the segment identifier for subsequent differential discrimination reference. Output the first response data and the second response data.

[0011] In a preferred embodiment, S4 includes: S4-1. Normalize and extract segments of the first response data and the second response data with the start time of the control action as the zero point, and calculate the steady-state baseline and its fluctuation threshold band of the tension roller feedback amount with the preset steady-state window after the end of the control action, and output the first baseline parameters and the second baseline parameters. S4-2. Detect the maximum deviation point of the tension roller feedback amount from the steady-state baseline in the first response data and the second response data respectively as the disturbance peak point, and calculate the duration of the tension roller feedback amount falling back for the first time after the disturbance peak point and continuously remaining within the fluctuation threshold band to obtain the tension recovery time characteristic quantity, and output the first tension recovery time and the second tension recovery time. S4-3. Calculate the absolute value of the difference between adjacent sampling points for the winding motor current or torque feedback in the first response data and the second response data respectively, and accumulate them within the control action duration interval to obtain the motor texture intensity feature. Combine the motor texture intensity feature with the corresponding tension recovery time feature to form the segment response feature to output the first feature and the second feature.

[0012] In a preferred embodiment, S5 includes: S5-1. Read the first feature quantity and the second feature quantity, and read the action parameters and downgrade flags of the first and second control actions from the control action record table. Select the corresponding threshold package and weight package with the downgrade flags, and output the discrimination configuration that matches the current control. S5-2. Calculate the tension recovery time difference and the motor texture intensity difference based on the first feature and the second feature respectively, and normalize the tension recovery time difference and the motor texture intensity difference according to the discrimination configuration to form a dimensionless difference vector, and output the difference vector. S5-3. Input the difference vector into the consistency checker to determine whether the difference direction meets the preset mechanism reference sign constraint. The mechanism reference sign constraint includes: when the first reference action is configured as a friction drift sensitive action and the second reference action is a micro-slip suppression action, the difference vector should present a preset sign direction in at least one dimension; otherwise, a backtracking re-examination token is triggered and a re-examination flag is output.

[0013] In a preferred embodiment, S5 further includes: S5-4. When the re-detection flag is triggered, the first feature quantity and the second feature quantity are sent back to S2 to shorten the fragment window or sent back to S3 to switch to the alternative control action and reacquire the first response data and the second response data. When the re-detection flag is not triggered, the execution continues and the effective difference vector that passes the consistency gating is output. S5-5. Input the effective difference vector into the scoring function to calculate the frictional drift score and the microslip budding score respectively. Generate the cause discrimination result and confidence label based on the difference between the frictional drift score and the microslip budding score and the uncertainty threshold. When the difference is less than the uncertainty threshold, output unresolved and write the unresolved cause code. Otherwise, output the main cause category as frictional drift or interlayer microslip budding.

[0014] In a preferred embodiment, S6 includes: S6-1. Receive the cause identification result and the corresponding confidence level mark, and map the main cause category to the parameter update object in the data twin model. When the main cause category is frictional drift, select the friction parameter as the update object; when the main cause category is interlayer microslip budding, select the interlayer slip risk parameter as the update object; and when the main cause category is unresolved, select the dual-object update mode of simultaneous update but limited amplitude, and output the parameter update instruction. S6-2. Perform a progressive update with confidence weight on the parameter update object according to the parameter update instruction. The progressive update weights and fuses the parameter value at the previous time step with the target parameter increment based on the difference vector and limits the single update amplitude to no more than the preset update limit. Then, write the updated parameter value into the data twin model to output the updated data twin model. S6-3. Based on the updated data twin model, generate at least one of the following within the preset prediction number window: tension given trajectory, clamping or pressure roller given trajectory, and speed given trajectory, as the subsequent winding control action sequence. Write the subsequent winding control action sequence and its applicable winding range into the control action table to output the control input for closed-loop control of the subsequent winding process of the same steel coil.

[0015] In a preferred embodiment, the data twin-based galvanized steel coil winding system includes a data acquisition and alignment module, a segmentation and sheet determination module, a comparison module, a feature extraction module, a differential cause determination module, and an update and distribution module. The acquisition and alignment module is used to acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation, and outputs an outer layer signal sequence for subsequent processing. The segmentation and segmentation module segments the coil winding process according to a preset number of turns or a preset time window based on the outer signal sequence, determines the adjacent first winding segment and second winding segment, and outputs the segment identifier and corresponding segment data. The comparison module is used to apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually compared on at least one of the tension given trajectory, the clamping or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. The feature extraction module is used to calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and outputs the first feature quantity and the second feature quantity. The differential cause determination module is used to perform differential discrimination between the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and outputs the cause determination result. The update and distribution module is used to write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and to generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil.

[0016] The technical effects and advantages of this invention are as follows: By applying control actions that serve as mutual mechanism comparisons within adjacent segments and performing differential discrimination and consistency verification, it is possible to relatively distinguish between frictional drift and interlayer microslip buds under conditions where only the outer layer signal is relied upon and the intervention amplitude is limited, thereby relatively reducing the risk of defect accumulation caused by the control object. By aligning the multi-source sampling signals in time and converting the cumulative number of windings to achieve segmentation, the control segments can be placed at a closer working baseline, thereby relatively improving the comparability and discrimination stability of the first and second feature quantities. By using the boundary consistency index to correct the boundaries of candidate segments within a limited range and setting boundary buffers, the impact of transitional perturbations on segment data can be reduced to a relatively smaller extent, thereby reducing the probability of misjudgment caused by the control action response being masked by boundary transients. By calculating the steady-state baseline and fluctuation threshold band through a steady-state window and extracting the tension recovery time using a continuous hold criterion, and extracting the motor texture intensity in the duration of the control action, the feature quantity can still take values ​​under noise and sampling differences, thereby relatively enhancing the repeatability of the difference vector. When the consistency check is not met, a rollback and re-check are triggered and the data is sent back to the shortened fragment window or the alternative control is switched. This can form a closed-loop review when the fragment boundary is unstable or there is insufficient control, thereby relatively suppressing the misjudgment and uncertainty caused by a single threshold judgment. By writing the cause identification results and confidence levels into the data twin model under the end-to-end collaborative architecture and performing a stepwise update of the parameters with limiting, and then generating the subsequent control trajectory in the prediction loop window and sending it to the winding end for execution, gradual adaptive correction can be achieved without introducing abrupt change control, thereby relatively improving the stability of subsequent winding and surface risk control. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention.

[0018] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Refer to the instruction manual appendix Figure 1-2 The present invention provides a data twin-based method for winding galvanized steel coils, comprising: S1. Acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation. Output the outer layer signal sequence for subsequent processing. This embodiment provides an implementable outer layer signal construction method for step S1. Using the sampling time of the strip linear speed as a unified time reference, the tension roller feedback and the winding motor current or torque feedback are resampled and aligned to this time reference. Then, the roll diameter estimation is used to map the time series to the cumulative winding revolution coordinates, thereby completing the roll segment division and extracting two types of outer layer sensitive features: motor texture intensity and tension recovery time. This implementation process does not rely on new hardware, only using commonly available linear speed, tension roller signals, motor feedback signals, and roll diameter estimation. Missing data processing, amplitude limiting, and steady-state threshold band determination ensure that the features are calculable and reproducible. This implementation process includes the following steps: In S1-1, the sampling time of the strip line speed is used as the reference time axis. The original sampled values ​​and timestamps of the tension roller feedback and the winding motor current or torque feedback are read. For each reference time, the nearest neighbor point or two adjacent points are determined in the corresponding original timestamp. When the nearest neighbor time deviation does not exceed the preset maximum deviation, the nearest neighbor interpolation is used to output the alignment value. When the reference time falls between two adjacent sampling points and the interval does not exceed the preset maximum interval, the linear interpolation is used to output the alignment value. If the above conditions are not met, the reference time is marked as missing and the most recent valid value is used to retain it and written into the retention mark. In this way, the tension roller sequence and motor feedback sequence are obtained after time alignment corresponding one-to-one with the reference time axis. In S1-2, the roll diameter estimation sequence aligned with the reference time axis is read. The strip passage length and corresponding winding circumference are calculated between adjacent reference times. The strip passage length is obtained by multiplying the average linear velocity within the time interval by the time interval, and the winding circumference is converted from the average roll diameter estimation within the time interval. The strip passage length is divided by the winding circumference to obtain the incremental winding number, and the cumulative winding number is accumulated in time sequence. When the roll diameter estimation is less than the preset minimum roll diameter threshold, the minimum roll diameter threshold is used to limit the roll to avoid anomalies in the early stage of roll building. Then, the cumulative winding number is binned according to the preset winding number window to generate a winding segment identifier, and a winding segment identifier and boundary buffer mark are written for each sampling point to form an outer signal sequence with a winding segment identifier. In S1-3, the motor texture intensity feature and tension recovery time feature are calculated within each loop segment. The motor texture intensity feature is obtained by calculating the absolute value of the difference between adjacent sampling points in the motor feedback sequence and accumulating it within the loop segment. The accumulated value is normalized according to the number of effective sampling points or the duration of the loop segment to eliminate sampling rate differences. The tension recovery time feature is determined by identifying the maximum deviation point of the tension roller feedback amount from the steady-state baseline within the loop segment as the disturbance peak point. The steady-state baseline and its threshold band are calculated using the mean of the preset steady-state window at the end of the loop segment. After the disturbance peak point, the moment when the tension roller feedback amount first falls back and remains within the threshold band for no less than the preset number of sampling points is found. The duration from the disturbance peak point to this moment is taken as the tension recovery time. The motor texture intensity feature and tension recovery time feature are associated with the loop segment identifier, stored, and the outer layer sensitive feature sequence is output. Through the above implementation process, outer layer signals from different sampling sources can be unified to the same time reference and mapped to the cumulative roll count coordinates, resulting in an outer layer signal sequence with roll segment identification and its roll segment-level outer layer sensitive feature sequence. This provides reproducible data input for subsequent segment comparison and differential causation. Simultaneously, configurable rules such as maximum deviation and maximum interval, minimum roll diameter threshold, steady-state window, and number of consecutive holding points ensure that calculations can still be completed and results remain consistent even when sampling is asynchronous, initial roll build fluctuations occur, and noise is present. In practical applications, the production line acquisition system uses a linear... Using the speed sampling time as a reference, the tension roller force signal and the driver motor current signal are interpolated and aligned to the same time axis. Then, the roll diameter estimate is used to convert the incremental number of winding turns in each sampling interval and accumulate to obtain the cumulative number of winding turns. A segment identifier is generated every certain number of turns. Subsequently, the absolute value of the motor current difference is calculated and normalized in each segment to obtain the motor texture intensity. At the same time, a threshold band is established with the steady-state window at the end of the segment and the time for the tension to fall from the peak to the threshold band and remain there is calculated as the tension recovery time, forming an outer layer sensitive feature sequence that can be directly used in subsequent steps.

[0021] S2. Based on the outer signal sequence, the winding process of the steel coil is segmented according to a preset number of turns or a preset time window, the adjacent first winding segment and second winding segment are determined, and the segment identifier and corresponding segment data are output. To ensure that subsequent comparison actions are based on comparable segments with similar operating conditions, this embodiment provides an implementable segmentation and segmentation process for step S2. This process uses the cumulative number of winding turns in the outer signal sequence as segmentation coordinates. First, continuous candidate winding segments are formed according to a preset number of turns window. Then, a boundary consistency index is calculated at the boundary of adjacent candidate segments, and the boundary is corrected within a limited range to ensure that the segment boundary avoids the transition zone formed by acceleration / deceleration, control switching, or transient disturbances. Finally, adjacent first and second winding segments and their segment data are output. This implementation process includes the following steps: In S2-1, the cumulative number of winding turns corresponding to each sampling point in the outer signal sequence is read. A preset window size for the number of turns and a minimum effective number of turns threshold are set. The sampling points are binned according to the cumulative number of winding turns from smallest to largest, so that sampling points whose cumulative number of winding turns falls within the same window range form a candidate winding segment. The starting number of the candidate winding segment is set as the minimum cumulative number of winding turns in the segment, and the ending number of the candidate winding segment is set as the maximum cumulative number of winding turns in the segment. When the effective number of a candidate winding segment is less than the minimum effective number of turns threshold, the candidate winding segment is merged with its adjacent candidate winding segments, and the starting number of winding turns and the ending number of winding turns are updated. A segment identifier is generated for each candidate winding segment, and a candidate segment index table containing the segment identifier, the starting number of winding turns, the ending number of winding turns, and the sampling point index range is output. In S2-2, boundary correction is performed on the boundary between two adjacent candidate take-up segments in the candidate segment index table. Specifically, a preset sampling point interval is taken before and after the boundary as a boundary window. The mean and fluctuation amplitude of the tension roller feedback and the current or torque feedback of the take-up motor are calculated in the boundary window before and after the boundary, respectively. The difference between the mean and the fluctuation amplitude before and after the boundary constitutes the boundary consistency index. When the boundary consistency index does not exceed a preset threshold, the boundary is retained as the segment boundary. When the boundary consistency index exceeds the preset threshold, the candidate boundary positions are traversed within a search range of no more than a preset number of turns before and after the boundary, and the boundary consistency index is calculated repeatedly. The candidate boundary with the smallest boundary consistency index that does not exceed the preset threshold is selected as the corrected segment boundary. If the preset threshold is not met within the search range, the candidate boundary with the smallest boundary consistency index is selected and written as a boundary correction mark. After all boundary corrections are completed, the set of take-up segments with boundary correction and its updated index table are output. In S2-3, the starting number of the winding segments after boundary correction is sorted from smallest to largest. Two adjacent winding segments are selected as the first winding segment and the second winding segment. A boundary buffer number of turns is set to form the effective interval of the segment. The effective interval of the segment is the range of turns after deducting the boundary buffer number from the starting and ending turns of each segment. The outer signal subsequences corresponding to the first winding segment and the second winding segment within the effective interval of the segment are extracted as segment data. The segment data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, roll diameter estimation, and cumulative winding turns. The first segment identifier, the second segment identifier, and the corresponding segment data are output. Through the above implementation process, continuous candidate segments can be obtained by using the cumulative number of winding turns as segment coordinates. By using the boundary consistency index and limited range correction, the segment boundaries can be made to avoid the transition disturbance zone, thereby obtaining the first winding segment and the second winding segment with higher working condition consistency and stronger comparability, providing an executable data basis for subsequent comparison actions and differential judgment. In practical applications: For the outer layer signal sequence of the same steel coil, candidate winding segments are first generated according to the preset number of turns window and a candidate segment index table is established. Then, for each boundary, the same length boundary window is taken before and after the boundary to calculate the mean difference and fluctuation amplitude difference of the tension roller feedback and the motor feedback to form the boundary consistency index. When the boundary consistency index exceeds the preset threshold, a new boundary position is searched within the limited number of turns before and after the boundary, and the correction is completed and written into the boundary correction mark. Finally, two adjacent segments are selected in chronological order as the first winding segment and the second winding segment, and the boundary buffer turns are removed before outputting the corresponding segment data for subsequent steps.

[0022] S3. Apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually comparative in at least one of the tension given trajectory, the pinch or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. To ensure that the first and second winding segments are comparable without increasing the risk to the galvanized surface, this embodiment provides a specific process for generating and applying the control action in step S3. This process takes segment data, estimated coil diameter, and strip speed as inputs. First, it constructs a candidate set of control actions based on the upper limit of the allowable intervention range for the galvanized surface and determines the range window. Then, it filters out the first and second control actions as mutual mechanism controls based on the mechanism control score. Subsequently, it performs reachability verification by combining the torque margin of the winding motor and the response time of the pinch roll or pressure roll actuator, and downgrades the control if necessary. Finally, it issues the control action, simultaneously collects response data, and records action parameters for subsequent differential discrimination. This implementation process includes the following steps: In S3-1, the segment data of the first and second winding segments are read to obtain the corresponding estimated coil diameter and strip speed. The estimated coil diameter and strip speed are used to determine the conversion of the number of continuous turns in the control application interval and the instruction issuance cycle. A control action candidate set is generated by calling the constraint set containing the upper limit of the allowable intervention amplitude of the galvanized surface. The constraint set includes at least the upper limit of the change rate of the tension given trajectory, the upper limit of the pulse amplitude and the upper limit of the pulse duration of the pinch roll pressure or pressure roll pressure given trajectory, and the upper limit of the holding time of the speed given trajectory. The above upper limits are selected from the preset table according to the steel type or zinc coating grade and combined with the current strip speed range. Within the upper limit constraints, tension given trajectory candidates, pressure given trajectory candidates and speed given trajectory candidates are constructed respectively, and an amplitude window is determined for each candidate action. The action candidate set and its amplitude window are output. In S3-2, candidate control action pairs are formed based on the action candidate set, and mechanism control scores for the first control action and the second control action are calculated respectively. Specifically, at least two of the following are calculated for the candidate control action pairs: the difference in the rate of change of tension given trajectory, the difference in pulse energy of the pinch roll pressure or pressure roller pressure given trajectory, and the difference in the duration of the speed given trajectory. The difference is normalized and then input into the scoring function to obtain the mechanism control score. When the mechanism control score is lower than a preset threshold, the candidate control action pair is replaced and recalculated until a candidate control action pair that meets the threshold is obtained or the preset maximum number of replacements is reached. If the threshold is not met even after reaching the preset maximum number of replacements, the candidate control action pair with the highest mechanism control score is selected and marked as insufficient control. The first control action and the second control action that pass the mechanism control gating are output. In S3-3, before applying the first and second comparison actions, an execution accessibility check is performed. Based on the current roll diameter estimate and tension level, the current torque requirement of the take-up motor is calculated and compared with the maximum allowable torque to obtain the torque margin of the take-up motor. At the same time, the response time of the pinch roll or pressure roll actuator is read and compared with the pressure trajectory pulse rise time requirement to form an execution accessibility consistency checker. When the torque margin of the take-up motor is lower than the preset margin threshold or the actuator response time is higher than the preset response threshold, the consistency check is determined to fail, and the first and second comparison actions are downgraded to alternative actions with smaller amplitude and shorter duration. The downgrade rules include proportionally reducing the tension change rate, proportionally reducing the pressure pulse amplitude or duration, or canceling the speed holding segment and replacing it with a smooth transition segment, until the consistency check passes or the preset maximum number of downgrades is reached. The comparison actions that pass the accessibility gating and the downgrade flag are output. In S3-4, the first and second comparison actions, respectively, through accessibility gating, are issued within the comparison application intervals of the first and second winding segments. At least one of the tension set trajectory, pinch roll pressure or pressure roll pressure set trajectory, and speed set trajectory is issued point by point as a control command according to the sampling period. The tension roll feedback and the winding motor current or torque feedback are collected simultaneously to form the first and second response data, and the two response data are associated with the corresponding segment identifiers and stored. The degradation mark and the key parameters of the comparison action are written into the comparison action record table associated with the segment identifier. The key parameters include at least the tension change rate, pressure pulse amplitude and number of cycles, speed holding time, and the start and end times of the comparison application interval. The first and second response data are output. Through the above implementation process, a verifiable mechanism-control action pair can be formed under the constraint of the allowable intervention range on the galvanized surface. The stability of the control action is ensured through reachability verification and degradation rules. At the same time, the action parameters and response data are established according to the segment identifier, so that the subsequent differential judgment can refer to the consistent comparison basis, thereby improving the reproducibility of cause identification and subsequent control updates. In practical application: For the determined first and second winding segments, the upper limit of tension change rate, upper limit of pressure pulse amplitude, upper limit of continuous number of revolutions, and upper limit of speed holding time are selected from the constraint set according to steel type and zinc layer grade to generate a set of action candidates. Then, the mechanism comparison score of the candidate control action pair is calculated to select the first and second control actions. Then, the reachability is verified according to the torque margin of the winding motor and the response time of the actuator, and the action is adjusted according to the degradation rules. Finally, the action is issued in the comparison application interval of the two segments, and the tension roller feedback and motor feedback are collected simultaneously to form two segments of response data. At the same time, the degradation mark and action parameters are recorded for subsequent differential judgment.

[0023] S4. Calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and output the first feature quantity and the second feature quantity. To ensure that the response results of the first and second control actions can be stably compared using a unified standard and used for subsequent differential discrimination, this embodiment provides a process for extracting segment response feature quantities in step S4. This process takes the first and second response data as input. First, it performs segment normalization on the two data segments at the start time of the control action, unifying the zero point. Then, within the steady-state window after the control action ends, it calculates the steady-state baseline and its fluctuation threshold band of the tension roller feedback quantity. Next, it uses the steady-state baseline to locate the disturbance peak point and calculates the duration required for the tension to fall back and stably enter the threshold band as the tension recovery time feature quantity. Simultaneously, it calculates and normalizes the cumulative motor feedback difference within the duration of the control action to obtain the motor texture intensity feature quantity, and finally outputs the first and second feature quantities. This implementation process includes the following steps: In S4-1, the first response data and the second response data are read respectively, and the start time and end time of the control action are obtained from the control action record table. The normalized time is obtained by subtracting the corresponding start time of the control action from the timestamp of each sampling point. The response subsequence containing the duration of the control action and the preset steady-state window after its end is extracted according to the normalized time. The steady-state baseline and fluctuation amplitude of the tension roller feedback are calculated within the steady-state window. The steady-state baseline is the mean or median within the steady-state window, and the fluctuation amplitude is the standard deviation or peak-to-peak value within the steady-state window. The fluctuation threshold band is constructed by the steady-state baseline and the fluctuation amplitude, and the first baseline parameter and the second baseline parameter are output. In S4-2, the deviation of the tension roller feedback is calculated based on the corresponding steady-state baseline. Within the preset observation interval, the sampling point corresponding to the largest deviation is selected as the disturbance peak point. After the disturbance peak point, the tension roller feedback is checked point by point to see if it falls into the fluctuation threshold band. The threshold for the number of consecutive points is set to be that no less than the preset number of sampling points are all within the fluctuation threshold band. The normalized time difference between the starting sampling point that first meets the threshold for the number of consecutive points and the disturbance peak point is used as the tension recovery time feature quantity and output as the first tension recovery time and the second tension recovery time. If it is not met within the observation interval, the preset upper limit value is output and a non-convergence mark is written. In S4-3, within the corresponding action duration intervals of the first and second response data, the absolute value of the difference between adjacent sampling points of the winding motor current or torque feedback is calculated and accumulated to obtain the texture cumulative value. The texture cumulative value is then divided by the number of effective sampling points or the duration within the corresponding action duration interval to obtain the motor texture intensity feature. The motor texture intensity feature and the corresponding tension recovery time feature are combined to form the segment response feature, and the first feature and the second feature are output respectively. Both feature and feature use the same field and the same calculation method for subsequent steps. Through the above implementation process, tension recovery time and motor texture intensity can be extracted from two response data segments under a unified zero point, unified steady-state window, and unified threshold band criterion, forming comparable feature quantities. This ensures that the features are quantifiable, calculable, and reproducible, thereby providing stable input for subsequent differential discrimination and reducing the impact of noise, sampling rate, and segment length differences. In practical applications: the two response data segments obtained from each control application are first normalized and truncated at the start time of the control action, and the steady-state baseline and threshold band are calculated in the steady-state window after the control action ends. Then, the disturbance peak point is located in the observation interval, and the time it falls back and remains continuously within the threshold band is calculated as the tension recovery time. At the same time, the absolute values ​​of adjacent differences in the motor feedback are accumulated within the duration of the control action and normalized according to the number of effective sampling points to obtain the motor texture intensity. Finally, the two features are combined into the first feature quantity and the second feature quantity for subsequent differential discrimination.

[0024] S5. Perform differential discrimination on the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and output the cause discrimination result. To stably distinguish between frictional drift and interlayer microslip initiation using comparable responses generated by the first and second control actions within the same steel coil, this embodiment provides a specific process for differential discrimination in step S5. This process uses the first and second feature quantities as inputs, and the action parameters and degradation markers in the control action record table as configuration criteria. First, a discrimination configuration matching the current control is generated. Then, a dimensionless difference vector is calculated and normalized. Subsequently, a consistency checker verifies whether the difference direction satisfies the mechanism control sign constraint. If not, a backtracking re-examination token is triggered to reconstruct the fragment or reselect the control action. After consistency is passed, the effective difference vector is input into a scoring function to output the causation discrimination result and confidence level marker. This implementation process includes the following steps: In S5-1, the first feature quantity and the second feature quantity are read, and the action parameters and downgrade markers of the first and second control actions are read from the control action record table. Based on the downgrade markers, a threshold package and a weight package are selected to form a discrimination configuration. The threshold package includes at least a normalized scaling factor, a sign constraint parameter, an uncertainty threshold, and a re-detection threshold. The weight package includes at least a differential component weight coefficient. The discrimination configuration is then associated with the current segment identifier, stored, and output. In S5-2, the tension recovery time difference is calculated as the first tension recovery time minus the second tension recovery time, and the motor texture intensity difference is calculated as the first motor texture intensity minus the second motor texture intensity. The two difference values ​​are scaled and limited according to the normalization scale factor configured by the discrimination, forming a dimensionless difference vector and outputting the difference vector. In S5-3, the difference vector is input into the consistency checker. The mechanism configuration is determined based on the control action type mark in the control action record table, and the corresponding symbol constraint parameters are selected. When the first control action is configured as a friction drift sensitive action and the second control action is a microslip suppression action, it is determined whether the difference vector presents a preset symbol direction in at least one dimension and satisfies the minimum number of satisfied dimensions threshold. If it does not satisfy the threshold, a backtracking re-examination token is triggered and a re-examination mark is output. If it satisfies the threshold, a consistency pass mark is output and the difference vector is marked as a candidate valid difference vector. In S5-4, when the re-detection flag is triggered, a backtracking re-detection is performed. If the re-detection flag indicates that the segment boundary is unstable, it is sent back to S2 and the preset number of loops window is shortened or the number of boundary buffer loops is increased to redetermine the first and second winding segments. If the re-detection flag indicates that the control action is not sufficiently controllable or is excessively downgraded, it is sent back to S3 and switched to an alternative control action to reacquire the first and second response data. After the backtracking is completed, S4 and S5-1 to S5-3 are re-executed and the maximum number of re-detection threshold is set. If the maximum number of re-detection threshold is reached and a re-detection is still triggered, an unresolved output is output and a re-detection failure reason code is written. If the re-detection flag is not triggered, the valid difference vector that passes the consistency gating is output. In S5-5, the effective difference vector is input into the scoring function and the frictional drift score and microslip budding score are calculated according to the weight package. The difference between the two is calculated and compared with the uncertainty threshold to generate the cause discrimination result and confidence label. When the difference is less than the uncertainty threshold, the unresolved result is output and the unresolved cause code is written. Otherwise, the main cause category is output as frictional drift or interlayer microslip budding and a confidence label is generated according to the difference size for subsequent parameter updates. Through the above implementation process, the difference vector is used as the core, and a sign constraint and backtracking re-examination mechanism consistent with the control action configuration are introduced, so that the discrimination result can be verified and backtracked. The risk of misjudgment is reduced when there is perturbation of fragment boundaries, downgrading of control action, or insufficient difference magnitude. At the same time, the pending cause code and confidence label are output to support the progressive update of the subsequent data twin model. In practical application: the obtained first feature quantity and second feature quantity are first selected according to the downgrading label to generate the discrimination configuration, then the difference vector is calculated and normalized to form a difference vector and a consistency check is performed. If the sign constraint is not met, it is sent back to S2 to shorten the fragment window or sent back to S3 to switch the alternative control action and recalculate the features and re-examine. If the consistency is passed, the two types of scores are calculated and the difference is compared with the uncertainty threshold to output the main cause category or pending and the corresponding confidence label.

[0025] S6. Write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller pressure setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil. To transform the main causal categories obtained from differential discrimination into executable winding control corrections, this embodiment provides a process for writing and updating the data twin model and generating subsequent control actions in step S6. This process takes the causal discrimination results and confidence level labels as input. First, it determines the parameter update object and generates parameter update instructions. Then, it performs a progressive update on the selected parameters according to the confidence level weights and writes them back to the data twin model with limited amplitude. Finally, within a preset prediction loop window, it generates a sequence of subsequent winding control actions based on the updated data twin model and writes it into the control action table for subsequent winding closed-loop control of the same steel coil. This implementation process includes the following steps: In S6-1, the cause determination result and the corresponding confidence level flag are received, and two types of updatable parameter fields, friction parameters and interlayer slip risk parameters, are preset in the data twin model. When the main cause category is frictional drift, the friction parameters are selected as the parameter update object and a single-object update instruction is generated. When the main cause category is interlayer micro-slip budding, the interlayer slip risk parameters are selected as the parameter update object and a single-object update instruction is generated. When the main cause category is unresolved, a dual-object update mode that updates both friction parameters and interlayer slip risk parameters is selected and a limit amplitude flag is written. The parameter update instruction is output, which includes at least the parameter field identifier, update mode, and limit amplitude flag. In S6-2, the parameter values ​​from the previous time step are read according to the parameter update instruction, and the difference vector is converted into the target parameter increment according to the preset mapping relationship. The preset mapping relationship adopts a lookup table method or a piecewise linear method and maintains the consistency of the direction of the difference components and the target parameter increment. The update weight is determined according to the confidence level label, and the parameter values ​​from the previous time step and the target parameter increment are weighted and fused to obtain the candidate update values. The candidate update values ​​are subjected to single update amplitude limiting processing, and a smaller preset update upper limit is used when the amplitude limiting label exists. The amplitude-limited update values ​​are written into the corresponding parameter fields of the data twin model and the updated data twin model is output. In S6-3, a preset prediction coil number window length is set, and the current coil diameter estimate and the current strip speed are used as initial conditions. The updated data twin model is called to generate the subsequent winding control action sequence within the prediction coil number window. When the main cause category is frictional drift, a tension or speed given trajectory is generated to compensate for frictional changes. When the main cause category is interlayer micro-slippage budding, a pinch roll pressure or pressure roller given trajectory is generated to improve the bite threshold, and a smoother rate of change constraint is applied to the tension given trajectory. When the main cause category is unresolved, a combined trajectory with limited amplitude is generated and the smoothness constraint is improved. Under the end-to-end collaborative architecture, at least one of the tension given trajectory, pinch roll pressure or pressure roller pressure given trajectory, and speed given trajectory is written into the control action sub-publishing with the applicable coil segment range, and the control input for closed-loop control of the subsequent winding process of the same steel coil is output. Through the above implementation process, the cause identification results can be written into the data twin model in the form of parameter update objects, and controllable incremental updates can be achieved with confidence weights and amplitude limiting rules. This avoids tension fluctuations or increased surface risks caused by a one-time large correction. At the same time, a control action sequence consistent with the cause category is generated and issued within the prediction cycle window, so that the winding control can continuously adaptively correct within the same steel coil and enhance closed-loop stability. In practical applications: when the output main cause category is friction drift, the friction parameters are incrementally updated and the tension given trajectory is issued to compensate for friction changes. When the output main cause category is the initiation of interlayer micro-slip, the interlayer slip risk parameters are incrementally updated and the pressure given trajectory is issued, along with a smoother tension trajectory to suppress slip risk. When the output is undecided, dual-object amplitude limiting updates are used and a conservative combined trajectory is issued to ensure winding stability and leave convergence space for subsequent comparison and identification.

[0026] Furthermore, it also includes: a data twin-based galvanized steel coil winding system, comprising a data acquisition and alignment module, a segmentation and sheet determination module, a comparison module, a feature extraction module, a differential cause determination module, and an update and distribution module. The acquisition and alignment module is used to acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation, and outputs an outer layer signal sequence for subsequent processing. The segmentation and segmentation module segments the coil winding process according to a preset number of turns or a preset time window based on the outer signal sequence, determines the adjacent first winding segment and second winding segment, and outputs the segment identifier and corresponding segment data. The comparison module is used to apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually compared on at least one of the tension given trajectory, the clamping or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. The feature extraction module is used to calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and outputs the first feature quantity and the second feature quantity. The differential cause determination module is used to perform differential discrimination between the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and outputs the cause determination result. The update and distribution module is used to write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and to generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil.

[0027] Working principle: The winding process is made into a closed-loop control mechanism within the same steel coil. First, linear speed, tension roller feedback, motor current or torque, and estimated coil diameter are collected. Time alignment is achieved, and the cumulative number of winding turns is calculated. Based on this, the entire coil is divided into adjacent first and second winding segments according to the number of turns. Then, a pair of mutually controlling actions that satisfy the allowable intervention range constraints on the galvanized surface are applied to each segment, simultaneously acquiring the response data of the two segments. The response is then extracted with the start time of the controlling action as the zero point. The baseline and threshold band are calculated in the steady-state window, and features such as tension recovery time and motor texture intensity are extracted. The system generates a first feature quantity and a second feature quantity. Then, it performs differential normalization on the two features to obtain a difference vector. First, it performs a consistency check. If the sign constraint is not met, it backtracks and re-checks by shortening the segment window or changing the alternative control action. If the constraint is met, it outputs the friction drift or interlayer micro-slip bud and confidence level using a scoring function. Finally, the discrimination result is written into the data twin model. The friction parameter or interlayer slip risk parameter is updated incrementally with a limited amplitude according to the confidence level. At least one of the following is generated within the prediction cycle window: the tension given trajectory, the pinch roll pressure or pressure roll pressure given trajectory, and the speed given trajectory as the closed-loop control input. In actual winding, if early signals appear such as coarser motor current texture and slower tension recovery but not yet obvious end face, the system will execute two comparative actions of the mechanism in two adjacent segments and collect the response. The tension recovery time and motor texture intensity are calculated using the same caliber, and then the difference is performed and the difference direction is checked to see if it matches the expected relationship of the comparative action. If it does not match, it will automatically re-check to avoid misjudgment caused by segment boundary disturbance or insufficient comparative action. After passing the test, the main cause and confidence level can be given, and the twin parameters can be updated step by step and the subsequent control trajectory can be issued. For example, when it is judged as friction drift, the tension or speed trajectory is more inclined to compensate for friction changes. When it is judged as micro-slippage, the pressure of the pinch roll or pressure roll is more inclined to increase and make the tension change smoother. In this way, the risk is suppressed in the early stage in the coil without exceeding the intervention range of the galvanized surface.

[0028] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for coiling galvanized steel coils based on data twins, characterized in that, include: S1. Acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation. Output the outer layer signal sequence for subsequent processing. S2. Based on the outer signal sequence, the winding process of the steel coil is segmented according to a preset number of turns or a preset time window, the adjacent first winding segment and second winding segment are determined, and the segment identifier and corresponding segment data are output. S3. Apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually comparative in at least one of the tension given trajectory, the pinch or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. S4. Calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and output the first feature quantity and the second feature quantity. S5. Perform differential discrimination on the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and output the cause discrimination result. S6. Write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil.

2. The galvanized steel coil winding method based on data twins according to claim 1, characterized in that: S1 includes: S1-1. Using the sampling time of the strip line speed as the reference time axis, resample the tension roller feedback and the winding motor current or torque feedback to the reference time axis by nearest neighbor interpolation or linear interpolation, and output the time-aligned tension roller sequence and motor feedback sequence. S1-2. Calculate the cumulative number of winding turns corresponding to each sampling time based on the estimated roll diameter, and divide the tension roller sequence and motor feedback sequence after time alignment into segments according to the preset number of turns window. Write a segment identifier for each sampling point and output the outer signal sequence with the segment identifier. S1-3. Calculate the absolute value of the difference between adjacent sampling points in the motor feedback sequence within each loop segment and accumulate it in the loop segment to obtain the motor texture intensity feature quantity. At the same time, calculate the number of sampling points experienced by the tension roller sequence from the disturbance peak back to the steady state threshold to obtain the tension recovery time feature quantity, and output the outer layer sensitive feature sequence.

3. The galvanized steel coil winding method based on data twins according to claim 2, characterized in that: S2 includes: S2-1. Using the cumulative number of winding turns in the outer signal sequence as the segment coordinates, generate candidate segment boundaries according to the preset number of turns window and form a continuous set of candidate winding segments. At the same time, write the start number of turns, end number of turns and segment identifier for each candidate winding segment and output the candidate segment index table. S2-2. Calculate the boundary consistency index between the tension roller feedback and the current or torque feedback of the winding motor at the junction of two adjacent candidate winding segments. The boundary consistency index includes the mean difference and fluctuation amplitude difference between the preset sampling point intervals before and after the junction. When the boundary consistency index exceeds the preset threshold, the junction is moved forward or backward by no more than the preset number of turns to avoid the transition disturbance zone. Output the winding segment set after boundary correction. S2-3. Select two adjacent winding segments from the set of boundary-corrected winding segments in chronological order as the first winding segment and the second winding segment, respectively, and extract their corresponding outer signal subsequences as segment data. Output the first segment identifier, the second segment identifier, and the corresponding segment data.

4. The galvanized steel coil winding method based on data twins according to claim 3, characterized in that: S3 includes: S3-1. Read the segment data of the first winding segment and the second winding segment, as well as the corresponding estimated roll diameter and strip speed. Call the constraint set containing the upper limit of the allowable intervention range of the galvanized surface to generate a control action candidate set, and output the action candidate set and its range window. S3-2. Calculate the mechanism comparison scores of the first and second control actions based on the action candidate set. The mechanism comparison score is obtained by combining at least two of the following: the difference in the rate of change of the tension given trajectory, the difference in the pulse energy of the pinch roll pressure or pressure roll pressure given trajectory, and the difference in the holding time of the speed given trajectory. When the mechanism comparison score is lower than a preset threshold, the action candidate is replaced to improve the mechanism comparison. Output the first and second control actions that pass the mechanism comparison gating.

5. The galvanized steel coil winding method based on data twins according to claim 4, characterized in that: S3 also includes: S3-3. Before applying the first and second comparison actions, the accessibility consistency checker is calculated based on the torque margin of the winding motor and the response time of the clamping or pressure roller actuator. When the consistency check fails, the first and second comparison actions are downgraded to alternative actions with smaller amplitude and shorter duration. The comparison actions that pass the accessibility gate and the downgrade flag are output. S3-4. Apply the first reference action of accessibility gating to the first winding segment and simultaneously collect the feedback amount of the tension roller and the current or torque feedback amount of the winding motor to form the first response data. At the same time, apply the second reference action of accessibility gating to the second winding segment and simultaneously collect the feedback amount of the tension roller and the current or torque feedback amount of the winding motor to form the second response data. Write the degradation mark and action parameters into the reference action record table associated with the segment identifier for subsequent differential discrimination reference. Output the first response data and the second response data.

6. The galvanized steel coil winding method based on data twins according to claim 5, characterized in that: S4 includes: S4-1. Normalize and extract segments of the first response data and the second response data with the start time of the control action as the zero point, and calculate the steady-state baseline and its fluctuation threshold band of the tension roller feedback amount with the preset steady-state window after the end of the control action, and output the first baseline parameters and the second baseline parameters. S4-2. Detect the maximum deviation point of the tension roller feedback amount from the steady-state baseline in the first response data and the second response data respectively as the disturbance peak point, and calculate the duration of the tension roller feedback amount falling back for the first time after the disturbance peak point and continuously remaining within the fluctuation threshold band to obtain the tension recovery time characteristic quantity, and output the first tension recovery time and the second tension recovery time. S4-3. Calculate the absolute value of the difference between adjacent sampling points for the winding motor current or torque feedback in the first response data and the second response data respectively, and accumulate them within the control action duration interval to obtain the motor texture intensity feature. Combine the motor texture intensity feature with the corresponding tension recovery time feature to form the segment response feature to output the first feature and the second feature.

7. The galvanized steel coil winding method based on data twins according to claim 6, characterized in that: S5 includes: S5-1. Read the first feature quantity and the second feature quantity, and read the action parameters and downgrade flags of the first and second control actions from the control action record table. Select the corresponding threshold package and weight package with the downgrade flags, and output the discrimination configuration that matches the current control. S5-2. Calculate the tension recovery time difference and the motor texture intensity difference based on the first feature and the second feature respectively, and normalize the tension recovery time difference and the motor texture intensity difference according to the discrimination configuration to form a dimensionless difference vector, and output the difference vector. S5-3. Input the difference vector into the consistency checker to determine whether the difference direction meets the preset mechanism reference sign constraint. The mechanism reference sign constraint includes: when the first reference action is configured as a friction drift sensitive action and the second reference action is a micro-slip suppression action, the difference vector should present a preset sign direction in at least one dimension; otherwise, a backtracking re-examination token is triggered and a re-examination flag is output.

8. The galvanized steel coil winding method based on data twins according to claim 7, characterized in that: S5 also includes: S5-4. When the re-detection flag is triggered, the first feature quantity and the second feature quantity are sent back to S2 to shorten the fragment window or sent back to S3 to switch to the alternative control action and reacquire the first response data and the second response data. When the re-detection flag is not triggered, the execution continues and the effective difference vector that passes the consistency gating is output. S5-5. Input the effective difference vector into the scoring function to calculate the frictional drift score and the microslip budding score respectively. Based on the difference between the frictional drift score and the microslip budding score and the uncertainty threshold, generate the cause discrimination result and confidence label. When the difference is less than the uncertainty threshold, output "unresolved" and write the unresolved cause code. Otherwise, output the main cause category as frictional drift or interlayer microslip budding.

9. The galvanized steel coil winding method based on data twins according to claim 8, characterized in that: S6 includes: S6-1. Receive the cause identification result and the corresponding confidence level mark, and map the main cause category to the parameter update object in the data twin model. When the main cause category is frictional drift, select the friction parameter as the update object; when the main cause category is interlayer microslip budding, select the interlayer slip risk parameter as the update object; and when the main cause category is unresolved, select the dual-object update mode of simultaneous update but limited amplitude, and output the parameter update instruction. S6-2. Perform a progressive update with confidence weight on the parameter update object according to the parameter update instruction. The progressive update weights and fuses the parameter value at the previous time step with the target parameter increment based on the difference vector and limits the single update amplitude to no more than the preset update limit. Then, write the updated parameter value into the data twin model to output the updated data twin model. S6-3. Based on the updated data twin model, generate at least one of the following within the preset prediction number window: tension given trajectory, clamping or pressure roller given trajectory, and speed given trajectory, as the subsequent winding control action sequence. Write the subsequent winding control action sequence and its applicable winding range into the control action table to output the control input for closed-loop control of the subsequent winding process of the same steel coil.

10. A galvanized steel coil winding system based on data twins, comprising a data acquisition and alignment module, a segmentation and sheet determination module, a comparison application module, a feature extraction module, a differential cause determination module, and an update and distribution module, characterized in that: The acquisition and alignment module is used to acquire real-time production data of the same steel coil during the winding process. The real-time production data includes at least the strip line speed, tension roller feedback, winding motor current or torque feedback, and coil diameter estimation, and outputs an outer layer signal sequence for subsequent processing. The segmentation and segmentation module segments the coil winding process according to a preset number of turns or a preset time window based on the outer signal sequence, determines the adjacent first winding segment and second winding segment, and outputs the segment identifier and corresponding segment data. The comparison module is used to apply a first comparison action to the first winding segment and a second comparison action to the second winding segment. The first comparison action and the second comparison action are mutually compared on at least one of the tension given trajectory, the clamping or pressure roller given trajectory or the speed given trajectory, and both satisfy the constraint on the allowable intervention range of the galvanized surface, and output the first response data and the second response data. The feature extraction module is used to calculate the segment response feature quantity for the first response data and the second response data respectively. The segment response feature quantity includes at least one of tension recovery time, winding motor current texture intensity or response loop area, and outputs the first feature quantity and the second feature quantity. The differential cause determination module is used to perform differential discrimination between the first feature quantity and the second feature quantity to obtain the main cause category. The main cause category includes at least frictional drift or interlayer microslip initiation, and outputs the cause determination result. The update and distribution module is used to write the cause identification results into the data twin model to update the corresponding friction parameters or interlayer slip risk parameters, and to generate the subsequent winding control action sequence based on the updated data twin model as the control input and output for tension setting, clamping or pressure roller setting or speed setting, so as to implement closed-loop control for the subsequent winding process of the same steel coil.

Images