Hot rolling flat steel rolling on-line laser measurement closed-loop control method

Abstract

The application discloses a hot-rolled flat steel rolling on-line laser measurement closed-loop control method and belongs to the field of metal rolling automatic control. In order to solve the problems of thick and wide detection lag in high-speed rolling, mismatch between the measured section and the forming state, and unstable feedback caused by measurement error, the application synchronously collects the section profile and the rolling state, generates a credibility state package with material coordinates, reversely matches the historical rack state, updates the model according to the matching error, and jointly solves the horizontal roller and the vertical roller control amount according to the credibility and the matching error, so that the technical effects of suppressing false feedback and stabilizing the finished product size are realized.

Description

Technical Field

[0001] This invention relates to the field of automatic control of metal rolling, and more particularly to a closed-loop control method for online laser measurement in hot-rolled flat steel rolling. Background Technology

[0002] The hot-rolled flat steel rolling process is characterized by high material running speed, continuous changes in hot dimensions, and the coupling between thickness and width. Existing production lines typically perform offline sampling inspections of finished product dimensions at the stand exit or finishing stage, or use only a single thickness measurement result to correct the horizontal roll gap. By the time the inspection results arrive, the corresponding material segment has often left the controllable stand, making it difficult to keep the inspection section, the rolling process state, and subsequent control objects consistent.

[0003] While online laser measurement can increase the sampling frequency of dimensions, coordinate deviations, thermal expansion, optical zero-point drift, acquisition timescale deviations, and rolling disturbances from different measuring heads can collectively contaminate the thickness and width results. If the measurement quality and historical state pairing quality are not distinguished and the dimensional residuals are directly fed into the feedback control, it can easily lead to model parameters being updated by erroneous samples, repeated corrections by horizontal and vertical roll commands, and consequently amplified thickness and width fluctuations. Existing solutions also lack a unified processing chain that simultaneously maps measurement reliability to model update permissions, feedback channel permissions, and single-cycle execution permissions.

[0004] Therefore, there is a need for an online laser measurement closed-loop control method for hot-rolled flat steel rolling that can overcome the shortcomings of the existing technology. Summary of the Invention

[0005] One objective of this invention is to propose a closed-loop control method for online laser measurement in hot-rolled flat steel rolling. Addressing the problems of lag in hot-state thickness and width detection, asynchronous measurement of the cross-section with historical rolling conditions, and unstable feedback due to measurement contamination in existing technologies, this invention proposes a method that synchronously acquires the same cross-sectional profile, reference target results, and rolling conditions. After calibration and correction, a reference temperature dimension is obtained. The method utilizes physical residuals to generate a reliability score and encapsulates the state package with the material coordinates. Historical stand conditions are then matched in reverse according to the material coordinates. The method uses paired error gating to update the model or maintain parameters. Furthermore, based on the reliability score and paired error, the method selects the residual usage channel and jointly solves for the increments of the horizontal and vertical rolls. This invention achieves the technical effect of ensuring correspondence between the measurement point, the causal state, and the controlled object, and reduces the impact of erroneous feedback on thickness and width stability.

[0006] This invention provides a closed-loop control method for online laser measurement in hot-rolled flat steel rolling, comprising:

[0007] S1. Simultaneously collect the surface profile, reference target detection results and rolling status data of the same section on the exit side of the rolling mill stand, perform unified coordinate calibration and temperature drift correction on the surface profile, and obtain the thickness and width at the reference temperature.

[0008] S2. Construct physical residuals based on surface profile, reference target detection results, and rolling state data to generate thickness confidence, width confidence, and joint confidence; integrate the rolling speed to obtain material coordinates, and encapsulate the material coordinates, thickness, width, and confidence into a confidence state package;

[0009] S3. Match the confidence status package to the historical rolling state when the section passes through the stand according to the material coordinates, obtain the historical stand state index and matching error, and write them into the confidence status package.

[0010] S4. Generate thickness residuals and width residuals based on the thickness and width in the confidence state packet and the target thickness and target width; when the pairing error is not greater than the preset pairing error threshold, update the compressed model and the widened model with the residuals; when the pairing error is greater than the threshold, keep the model parameters and use the updated or kept model to predict the thickness and width of the material segment that has not yet passed through the rack.

[0011] S5. Based on the predicted thickness and width and the confidence status package, jointly solve the horizontal roll reduction increment and vertical roll gap increment under preset execution constraints. Determine the use channel of the laser measurement residual based on the relationship between the joint confidence and the pairing error and the corresponding threshold, and output the roll gap control command.

[0012] Optionally, S1 includes:

[0013] The upper and lower laser measuring heads and the left and right laser measuring heads are triggered by a common trigger time mark to collect the contour point set of the same cross section, and the temperature, encoder speed, rolling force, horizontal roll pressing position and vertical roll gap are collected.

[0014] The reference features on the target are measured in a unified reference coordinate system, and the coordinate systems of each laser measuring head are transformed to the unified reference coordinate system.

[0015] The contour point coordinates are corrected based on the difference between the cross-sectional temperature and the reference temperature and the linear expansion coefficient of the hot-rolled flat steel material, and the optical zero drift of each laser measuring head is corrected based on the displacement of the reference target reference feature relative to its calibration position.

[0016] The thickness is determined by the corrected spacing between the upper and lower surface profiles, and the width is determined by the corrected spacing between the side surface profiles.

[0017] Optionally, S2 includes:

[0018] The physical residuals include the geometric closure residual relative to the laser measuring head, the drift residual of the reference target reference feature, the model residual between the size at the reference temperature and the size predicted by the temperature and rolling state data, and the timescale residual between the profile acquisition timescale and the rolling state acquisition timescale of the same cross section.

[0019] Normalized residuals are generated by the ratio of the absolute value of each physical residual to the corresponding residual threshold, wherein the residual threshold is determined by the reference target calibration data and the residual distribution quantile of historical qualified rolling sections.

[0020] Thickness confidence is generated based on the normalized residuals associated with thickness measurement, and width confidence is generated based on the normalized residuals associated with width measurement.

[0021] The thickness-width residual correlation coefficient is estimated based on the thickness normalized residual sequence and width normalized residual sequence in the historical qualified rolled section samples. The joint confidence is generated based on the thickness confidence, width confidence and thickness-width residual correlation coefficient. The dimensional covariance is generated based on the variance of the two normalized residual sequences and the thickness-width residual correlation coefficient.

[0022] Furthermore, the material coordinates are obtained by integrating and summing the adjacent sampling intervals of the common trigger timescale with respect to the rolling speed;

[0023] The credibility status package also includes size covariance, reference target drift status, physical residual, historical rack status index field, feedback channel identifier field, and acquisition time stamp, wherein the historical rack status index field is null before reverse matching, and the feedback channel identifier field is undetermined before control solution.

[0024] Optionally, S3 includes:

[0025] Write the encoder speed, rolling force, horizontal roll pressing position, and vertical roll gap into the historical state sequence according to a common trigger time stamp;

[0026] Based on the material path length between the laser measurement position and the target rolling mill stand, as well as the velocity integral result, the target material coordinates and target time scale of the measured section through the target rolling mill stand are determined.

[0027] Retrieve historical rolling states adjacent to the target material coordinates from the historical state sequence and generate a historical stand state index. Use the weighted sum of the material coordinate difference and time scale difference of the measured section after normalization of their respective allowable errors as the pairing error.

[0028] The historical rolling state is associated with the confidence state package and model update is allowed only when the pairing error is not greater than the pairing error threshold.

[0029] Optionally, S4 includes:

[0030] The thickness residual is generated based on the difference between the thickness at the reference temperature and the target thickness, and the width residual is generated based on the difference between the width at the reference temperature and the target width.

[0031] The parameter update step size of the compression model and the expansion model is limited by the thickness confidence and the width confidence, respectively, and the cross-coupling parameter update step size between the compression model and the expansion model is limited by the joint confidence.

[0032] The updated reduction and width models are applied to the material coordinate range that has not yet passed the target rolling mill stand, and the predicted thickness and width are obtained by combining the current horizontal roll reduction position, vertical roll gap and rolling force corresponding to the material coordinate range.

[0033] Optionally, S5 includes:

[0034] The control target is constructed using the difference between the predicted thickness and the target thickness, and the difference between the predicted width and the target width. The weights of the thickness deviation term, the width deviation term, and the thickness-width coupling term are determined by the thickness confidence, the width confidence, and the joint confidence, respectively.

[0035] The actuator displacement constraint is determined based on the available stroke of the horizontal and vertical rolls, the actuator rate constraint is determined based on the allowable displacement difference between adjacent control cycles, the rolling force constraint is determined based on the allowable upper limit of the rolling force, and the roll gap lower limit constraint is determined based on the allowable minimum clearance to prevent roll collisions.

[0036] The control objective is solved within the constraints according to the preset solution termination condition, and the obtained solution is used as the increment of horizontal roll reduction and vertical roll gap, and the upper limit of the increment of a single control cycle is limited by the joint reliability.

[0037] Furthermore, a first confidence threshold and a second confidence threshold are set within the confidence value range of zero to one. The difference between the first confidence threshold and the second confidence threshold is a positive number. The closed interval from the first confidence threshold to one is used as the first confidence interval, the left-closed and right-open interval from the second confidence threshold to the first confidence threshold is used as the second confidence interval, and the left-closed and right-open interval from zero to the second confidence threshold is used as the third confidence interval.

[0038] The laser measurement residual integral state of the bias update channel is formed by accumulating the thickness residual and width residual according to the coordinate increments of adjacent materials, respectively;

[0039] When the joint confidence level is within the first confidence level range and the pairing error is not greater than the pairing error threshold, the laser measurement residual is allowed to enter the cross-section feedback channel and the bias update channel.

[0040] When the joint confidence level is in the second confidence level range and the pairing error is not greater than the pairing error threshold, the laser measurement residual is prohibited from entering the cross-sectional feedback channel, and the laser measurement residual is allowed to enter the bias update channel with an update step size limited by the joint confidence level.

[0041] When the joint confidence level is in the third confidence level range or the pairing error is greater than the pairing error threshold, the current value of the laser measurement residual integral state is maintained and its update is prohibited. The system is then switched to the model feedforward channel that generates roll gap instructions based on the rolling force, the horizontal roll pressing position and the vertical roll gap.

[0042] After the joint confidence of the cross sections that have continuously reached the threshold for the number of restored cross sections are all within the first confidence interval and the pairing error is not greater than the pairing error threshold, the update of the laser measurement residual integral state is restored.

[0043] Furthermore, the roll gap control command includes a horizontal roll pressing command and a vertical roll gap command. The limiting and speed limiting processing of the roll gap control command includes: trimming the target positions of the horizontal roll pressing command and the vertical roll gap command according to the actuator displacement constraints, trimming the command change amount of adjacent control cycles according to the actuator speed constraints, and performing a monotonically increasing mapping on the upper limit of the command change amount of a single control cycle according to the joint confidence.

[0044] Calculate the changes in the horizontal roll pressing command and the vertical roll gap command in the current control cycle and the previous control cycle respectively. When the changes of the same command component have opposite signs in the two control cycles, the corresponding command component is allowed to reverse only after the cumulative value of the thickness residual corresponding to the horizontal roll pressing command or the cumulative value of the width residual corresponding to the vertical roll gap command exceeds the corresponding reversal dead zone boundary and the number of continuous sections reaches the reversal confirmation number threshold. The reversal dead zone boundary is determined based on the corresponding component in the dimensional covariance and the position resolution of the corresponding actuator.

[0045] Furthermore, the material coordinates, thickness and width at reference temperature, physical residual, thickness confidence, width confidence, joint confidence, dimensional covariance, historical rack status index, pairing error, feedback channel identifier, model parameter version, horizontal roll pressing command, vertical roll gap command, actual position of horizontal roll and actual position of vertical roll for each confidence status package are written into the traceability record.

[0046] The absolute value of the difference between the horizontal roll pressing command and the actual position of the horizontal roll is used as the horizontal roll tracking error, and the absolute value of the difference between the vertical roll gap command and the actual position of the vertical roll is used as the vertical roll tracking error. Any tracking error exceeding the corresponding execution tracking error threshold or the rolling force reaching the upper limit of the rolling force is determined as an execution tracking abnormality.

[0047] When an execution tracking anomaly occurs, the feedback channel identifier is switched to the model feedforward channel and the current value of the laser measurement residual integration state is maintained. After both tracking errors are not greater than the corresponding execution tracking error threshold and the rolling force has not reached the upper limit of the rolling force, the laser measurement residual usage channel is restored according to the recovery conditions.

[0048] The beneficial effects of this invention are:

[0049] 1. By generating thickness confidence, width confidence, and joint confidence through unified coordinate calibration, temperature drift correction, and physical residuals, measurement deviations from different sources can be transformed into determinable data quality information before entering the model and controller.

[0050] 2. Obtain the material coordinates through velocity integration, and reverse match the measured section to its historical rolling state when it passes through the rolling mill stand. Use the matching error to limit the model update, so that the downstream dimensional residuals correspond to the causes of the pressing position, vertical roll gap and rolling force, etc., and reduce the feedback phase mismatch in high-speed rolling.

[0051] 3. By jointly selecting the cross-section feedback, bias update, or model feedforward channel through joint confidence and pairing error, and jointly solving the control quantities of the horizontal and vertical rolls under execution constraints, the measurement quality simultaneously limits the model update and actuator action, reducing malfunctions and command reversals caused by measurement contamination. Attached Figure Description

[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0053] Fig. 1 This is a flowchart of the online laser measurement closed-loop control method for hot-rolled flat steel rolling according to the present invention.

[0054] Fig. 2 This is a flowchart of step S5 of the present invention. Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0056] refer to Figs. 1-2 A closed-loop control method for online laser measurement in hot-rolled flat steel rolling, including:

[0057] S1. Simultaneously collect the surface profile, reference target detection results and rolling status data of the same section on the exit side of the rolling mill stand, perform unified coordinate calibration and temperature drift correction on the surface profile, and obtain the thickness and width at the reference temperature.

[0058] S2. Construct physical residuals based on surface profile, reference target detection results, and rolling state data to generate thickness confidence, width confidence, and joint confidence; integrate the rolling speed to obtain material coordinates, and encapsulate the material coordinates, thickness, width, and confidence into a confidence state package;

[0059] S3. Match the confidence status package to the historical rolling state when the section passes through the stand according to the material coordinates, obtain the historical stand state index and matching error, and write them into the confidence status package.

[0060] S4. Generate thickness residuals and width residuals based on the thickness and width in the confidence state packet and the target thickness and target width; when the pairing error is not greater than the preset pairing error threshold, update the compressed model and the widened model with the residuals; when the pairing error is greater than the threshold, keep the model parameters and use the updated or kept model to predict the thickness and width of the material segment that has not yet passed through the rack.

[0061] S5. Based on the predicted thickness and width and the confidence status package, jointly solve the horizontal roll reduction increment and vertical roll gap increment under preset execution constraints. Determine the use channel of the laser measurement residual based on the relationship between the joint confidence and the pairing error and the corresponding threshold, and output the roll gap control command.

[0062] In this specific embodiment, S1 includes:

[0063] The online measurement controller uses a common trigger to generate the acquisition timescale. As a unique index for the same cross section The cross-section numbers are ordered in ascending order of collection. The unit is seconds (s) and originates from the rolling mill precision clock protocol master clock. The common trigger period is set to 10ms. After the trigger edge arrives, the upper laser measuring head, lower laser measuring head, left laser measuring head, and right laser measuring head respectively output the measuring head number, point number, local coordinates, reflection intensity, and... The profile point set, while reading the section temperature from the rolling control system. Encoder speed Rolling force Horizontal roller pressing position and vertical roller gap ,in The unit is Celsius. The unit is mm / s. The unit is MN. and The unit is mm;

[0064] The controller is the first Each laser measuring head stores an external parameter rotation matrix. Translation vector , These represent the top, bottom, left, and right laser measuring heads, respectively. The matrix is ​​a 3x3 dimensionless orthogonal matrix obtained from the shutdown calibration. For a three-dimensional vector in mm, the first vector acquired by the measuring head... Local contour points implement ,in The point number is the point index within the same contour point set. and under a unified reference coordinate system All are three-dimensional column vectors in mm, using a unified reference coordinate system. Shaft along the rolling direction, The shaft is along the width direction of the flat steel and The shaft is along the thickness direction;

[0065] Two heat-resistant reference target features are set at the edge of the field of view of each laser measurement head. The reference target recognition input structure consists of a local image, template number, and unified reference coordinates. The recognizer performs normalized correlation matching using a grayscale template generated from the calibration sample and outputs feature coordinates, template number, and confidence score. The controller will... The first time to be identified Reference feature coordinates With stop calibration coordinates Find the difference and the mean, according to The optical zero-point drift vector is obtained, where The reference feature number, For the first The number of effective reference features involved in the calculation of each measuring head. , and The units are all mm and all are in a unified reference coordinate system. Write the zero-point correction set, when If the confidence level is less than 2 or the recognition confidence level of any benchmark feature is lower than the threshold of 0.95 determined by the reference target calibration samples, no update is made. The low-confidence identification results and the invalid drift state of the reference target are then written into the subsequent synchronous section record;

[0066] The controller uses a reference temperature Celsius and the coefficient of linear expansion corresponding to the current steel grade Implement temperature drift correction, The unit is per degree Celsius and is read from the material grade parameter record. In this specific embodiment, Q235B corresponds to... / degrees Celsius, for relative to the center of the cross section coordinate component execution ,in A three-dimensional vector, in mm, defined by the center of the bounding box of four sets of contour points. To be converted to The unit is mm for the corrected profile points, and the denominator of the formula is the dimensionless thermal expansion ratio, which is limited to a positive number.

[0067] For the four groups First, the surface is cropped according to the lower limit of reflection intensity and the effective field of view of the measuring head. Then, isolated pulses are removed by median filtering of five adjacent points. The residual of local quadratic curve fitting is no greater than 0.08 mm as the criterion for effective points on the surface. When the number of effective points of the same measuring head is less than 80% of the number of calibrated points, the effective profile mark of that head is set to zero and the original point set is retained for traceability. A complete cross-sectional profile record is generated only when all four measuring heads are effective. The complete cross-sectional profile record includes... ,each , , , , , and the drift state of the reference target;

[0068] The controller performs linear interpolation on the upper and lower surfaces on a common width coordinate sampling grid, according to... Determine the thickness at the reference temperature, where This is a set of grid indices with valid interpolation points on both the upper and lower surfaces. For the grid index in this set, and These are the height coordinates of the upper and lower surfaces, respectively, in mm. The unit is mm. Linear fitting is performed on the effective points of the left and right surfaces respectively, according to... Determine the width at the reference temperature, where and These are the sets of valid point indices for the right and left surfaces, respectively. and For the point index in the corresponding set, , and The units are all mm;

[0069] when If the number of points is less than 70% of the common sampling grid points or the linear fitting residual on either side of the surface exceeds 0.10 mm, the controller will still calculate based on valid points. and However, the corresponding size validity flag is set to zero, allowing S2 to reduce confidence through physical residuals. When the temperature field is missing, the most recent valid value within 100ms is used. The temperature hold flag is written to the record, and the reference temperature dimension of this section is not output if there is no valid temperature after 100ms.

[0070] The reference target recognizer first constructs a candidate reference feature input structure using the calibration template number, field of view area, and geometric spacing. Then, it sorts these candidates from highest to lowest recognition confidence. If the confidence levels are the same, they are sorted from smallest to largest by the Euclidean distance between the feature coordinates and the calibration position. The top-ranked candidate is then selected. The first non-overlapping candidate is the... The baseline features of each measuring head are identified. The confidence threshold of 0.95 is determined by the intersection of correct and incorrect matches in the calibration sample. If the second candidate is still tied, the point with the earliest number is selected. The selected feature number, sorting value, output coordinates and reasons for not being selected are written into the reference target detection results. Low-confidence candidates are not included in the zero-point correction and are used for subsequent confidence calculation.

[0071] The controller sets the reference temperature to , The corrected four sets of surface profile point sets and the reference target drift vector Its status, rolling status field, and all valid flags are written in section number. The key's synchronous cross-section record, along with the acquisition time stamp, is recorded. The physical residuals and credibility calculation queues are sent to S2.

[0072] In this specific embodiment, S2 includes:

[0073] The reliability calculator reads the synchronous cross-section record generated by S1, performs nearest neighbor registration on adjacent measurement headpoint sets with overlapping fields of view, and follows... Generate the first The geometrical closure residuals of the group relative to the laser measuring head, among which For the group index of the overlapping fields of view of top left, top right, bottom left, or bottom right, This represents the number of valid pairings in the group. and To unify two paired points in the reference coordinate system, with the unit being mm, The unit is mm. When the number of paired points is less than 60% of the number of calibration points in the overlapping field of view, the corresponding... Set the value to twice the residual threshold of this group and mark geometric closure anomalies;

[0074] For the reference target features in S1, the confidence calculator follows... Generate the first The drift residuals of each measuring head, among which The optical zero-point drift vector is in mm and , The unit is mm. When the reference target drift state is invalid, Set it to twice the corresponding drift residual threshold so that the anomaly is clearly included in the subsequent normalized residual instead of replacing the detection result of this section with the historical value;

[0075] The reliability calculator reads the regression coefficients corresponding to the current steel grade and specification from the rolling state prediction parameter record, and then... and Obtain the state prediction size and form and ,in , , and The unit is mm. and The unit is mm. and This indicates the change in size (in mm) corresponding to one degree Celsius change in temperature. and This indicates the change in size (in mm) corresponding to a 1MN change in rolling force. and These are dimensionless coefficients, which are obtained by least-squares fitting of no less than 500 cross sections from the most recent qualified rolling batch and saved with the version number.

[0076] The credibility calculator uses the contour point set timescale Original timescale of rolling state in the same synchronous section record The difference As time-scaled residuals, , and The units are all in seconds (s). When the rolling state is aligned via interpolation... Take the time-stamped weighted value of the two state records involved in the interpolation, and when the interpolation span exceeds 20ms, The absolute value is set to twice the time-scaled residual threshold;

[0077] The residual threshold record stores the threshold using the residual type and measurement header index as a combined key. Unit, valid steel grade specifications, calibration batch and version number, among which For the type index of geometric closure, reference target drift, thickness model, width model, or time-scaled residual, the first... Class residuals Determine the threshold. It is the 99.5th percentile of the absolute value of the residuals of the three most recent qualified rolling batches; when When it is a geometrically closed or reference target drift type, Take the standard deviation of the corresponding residual sequence obtained from no less than 1000 static samplings of the reference target around its sample mean. When it is a thickness model, width model, or time-scaled residual type, Take the standard deviation of the corresponding residual sequence around its sample mean within the three most recent qualified rolling batches of the same steel grade and target specification. Each residual sequence is first grouped according to the residual type and measurement head group joint key, and samples with invalid quality marks are removed. , and Having the same residual unit and When there are insufficient samples or the joint key record is not hit, the model update is prohibited and the previous valid version is used.

[0078] For each physical residual calculate ,in Indicates any of the aforementioned units and Consistent physical residuals For dimensionless normalized residuals in the range of 0 to 2, For residual type index and The section number is used as the confidence calculator. and Generate thickness and width confidence scores; set of thickness residual types. The members are uniquely limited to all geometrically closed residuals. Drift residuals of upper and lower measuring heads and Thickness model residual and time-scaled residuals Width residual type set The members are uniquely limited to all geometrically closed residuals. Drift residuals of left and right measuring heads and Width model residual and time-scaled residuals The weight of residual types not included in the corresponding set in this confidence channel is zero; , These are dimensionless non-negative numbers determined by sensitivity analysis of historical qualified cross sections, and their sum is 1. and All are dimensionless numbers between 0 and 1; the larger the value, the more reliable the measurement.

[0079] The reliability calculator uses the thickness-normalized model residual sequence of the most recent 500 historical qualified rolling sections. and width normalized model residual sequence Calculate the Pearson correlation coefficient ,in This refers to the historical sample sequence number. and This corresponds to the mean of the dimensionless sequence. The variance is a dimensionless number ranging from -1 to 1; the sum of the two variances in the denominator of the correlation coefficient is respectively compared with the dimensionless positive threshold. Comparison, when the sum of any variance is not greater than Continue to use the previous effective When there is no previous valid value, the command is given. Simultaneously, write the correlation coefficient preservation flag and prohibit updating the relevant parameters in this window; then calculate. ,in The dimensionless correlation penalty coefficient is calibrated by the combined misjudgment rate of thickness and width of qualified batches. The joint confidence level is between 0 and 1, with a larger value indicating a more reliable joint thickness and width measurement.

[0080] The credibility calculator uses the same history window. , Calculate the dimensionless standard deviation and and in accordance with Generate the size covariance, where and These are the residual thresholds for the thickness model and the width model, respectively, in mm. It is a 2x2 matrix with diagonal components. , Off-diagonal components The units are all mm squared. This construction restores the normalized residual variance to the dimension of size by using a threshold.

[0081] Residual threshold records, rolling state regression parameter records, and confidence weight records are established by the same calibration batch before production: the reference target is statically sampled no less than 1000 times to form geometric closure and drift calibration samples, historical qualified rolling sections are grouped according to steel type and target specifications to form model and time-scaled residual samples, and the controller generates corresponding fields according to the aforementioned quantile values, standard deviations, and sensitivity rules. All three types of records save steel type, target thickness and width, sample start and end time scales, effective range, version number, and effective time scale. Reconstruction is triggered when the reference target is maintained, the laser head is reinstalled, or the residual distribution of 50 consecutive qualified sections exceeds the original 99.5% quantile boundary. If the joint key is not hit during operation, the previous effective version is used and model updates are prohibited.

[0082] After the size covariance calculation is completed, the confidence calculator verifies the result by ensuring that the eigenvalues ​​are not less than zero. The semi-positive definiteness of the matrix means that if rounding produces negative eigenvalues, these negative eigenvalues ​​are pruned to zero and the matrix is ​​reconstructed using the original eigenvectors. If the historical valid window contains fewer than 500 samples, the previous valid window is used. , and The covariance preservation flag is written into the status packet, and the covariance version is only updated after the window is restored to 500 valid samples.

[0083] The material coordinate calculator uses the first valid synchronous section record as its origin. ,according to Accumulated material coordinates, where and The unit is mm. , The unit is mm / s. The unit is seconds (s); the material coordinate calculator also reads the cumulative pulse count, which is derived from the encoder speed. The displacement conversion factor per pulse is given by the encoder resolution and the effective diameter calibration record of the working roll, and the unit is mm per pulse, and it is a positive number. When the adjacent time difference exceeds 30ms, the encoder speed field is invalid, or the cumulative pulse count is missing or reversed, the encoder enters an interrupt state, the material coordinate validity flag is set to zero, and the previous time is frozen and saved. The corresponding cumulative pulse count and time stamp, this section retains its size and reliability but does not participate in the S3 model update pairing;

[0084] In encoder interruption mode, velocity interpolation for abnormal intervals is not used, nor is the last velocity before the abnormality and the first velocity after the abnormality substituted into the trapezoidal integral across intervals. After receiving a synchronization section with two consecutive valid velocity fields, adjacent time difference not exceeding 30ms, and cumulative pulse count monotonically increasing, the material coordinate calculator multiplies the difference between the current cumulative pulse count and the cumulative pulse count at the breakpoint by the per-pulse displacement conversion coefficient to obtain the displacement of the abnormal interval, and adds it to the saved breakpoint material coordinates to complete the re-anchoring. Then, starting from the second consecutive valid section, the calculation continues according to the aforementioned recursive breakpoint. Only when the re-anchoring displacement is non-negative and does not exceed the equipment's allowable displacement... The material coordinate valid flag is reset to one only when the product of the maximum linear velocity and the abnormal duration is 1.05 times, the calculated material coordinates are monotonically increasing, and the single section increment does not exceed 1.05 times the product of the maximum allowable linear velocity of the equipment and the corresponding time difference. Otherwise, the continuous valid count is cleared and the interrupted state is maintained until the accumulated pulse becomes valid again or the breakpoint is reset by a traceable manual reference record. Before the recovery confirmation, S3 keeps the historical rack status index empty, S4 keeps the model update permission prohibited, and S5 prohibits section feedback, offset update and residual integration and processes it as an existing model feedforward or safe hold branch.

[0085] Before encapsulation, integrity checks are performed on the dimensions, reliability, covariance, physical residuals, and material coordinates for the same cross-section number. The check rules require reference temperature dimensions. and All have been output and the size validity mark is 1. , and All are located between 0 and 1. The dimension is 2x2 and the material coordinate timescale is consistent with the synchronous section timescale; if any necessary dimension or field is missing, the physical residual calculation dependent on the input is not performed, the corresponding confidence is set to zero, and if the verification fails, the status package quality flag is set to invalid, the historical rack status index is kept as null, the model update permission is set to prohibited, and the feedback channel identifier is preset to model feedforward; if the verification passes, the quality flag is set to valid.

[0086] The state wrapper will , , , , , , , , ,each ,each The reference target drift state, material coordinate validity flag, dimension validity flag, and state package quality flag are written into the confidence state package. The historical rack state index field is initialized to null. The model update permission is initialized to prohibited or pending according to the quality flag, and the feedback channel identifier is initialized to model feedforward or pending. Each confidence state package also carries the threshold record version, rolling state prediction parameter version, and non-repeatable section number. Write to The status packets are organized in ascending order and sent to S3.

[0087] In this specific embodiment, S3 includes:

[0088] History State Manager at each public trigger time marker Read the encoder speed of the target rolling mill stand Rolling force Horizontal roller pressing position and vertical roller gap And record the status. Write a circular historical state sequence with a length covering twice the maximum transport time from the laser measurement position to the target rolling mill stand, in which... This is a historical state sequence number. The unit is s. The unit is mm and obtained according to the same integration rule as S2. The units of the remaining fields are consistent with S1. The circular sequence is... Create an ordered index;

[0089] The calibration length along the material path from the laser measurement position to the center of the roll gap of the target rolling mill stand is denoted as... , The unit is mm, and it is obtained by taking the average of three measurements along the rolling centerline when the machine is stopped. In this specific embodiment... mm, the first effective material coordinate A confidence status packet; the pairer calculates the target material coordinates. ,in The cumulative material coordinates corresponding to the laser measurement position and These are the cumulative material coordinates that the cross-section should correspond to when it passes through the target rolling mill stand; both are in mm.

[0090] The pairer retrieves matching conditions from the historical state sequence. Adjacent records, and according to Determine the target timescale, where and These are the lower and upper historical state indices that enclose the target material coordinates, respectively. Immediately afterwards , The unit is s. When the denominator is not greater than 0.1mm corresponding to the encoder single pulse or no enclosing record is found, the pairing state is set to no candidate and the model parameter update of this package is prohibited.

[0091] After the target timescale is formed, the pairer verifies the interpolation results for speed consistency, according to... Calculate the average velocity across the bounding record interval, where The unit is mm / s. and For enclosing record indexes, , The unit is mm and , The unit is s, when When the speed exceeds 2% of the rated speed, the two enclosing records are removed from the candidate set and retrieved again. If the speed consistency is still not satisfied after the retrieval, the state of no candidates is output and the original interpolation diagnostic value is retained.

[0092] For each candidate historical state calculate ,in This represents the dimensionless candidate pairing error. mm represents the allowable material coordinate error determined by the encoder resolution and the upper limit of the speed integral error. s is the allowable timescale error obtained by calibration of common trigger and state communication delay. and The candidate set is composed of dimensionless weights summing to 1. Three historical records before and after, in order Select the smallest item in ascending order; if the errors are the same, select the smallest item. The smallest record;

[0093] The selected historical state index is denoted as The corresponding pairing error is denoted as ,in Integer index in the historical state sequence and For dimensionless numbers, the pairing error threshold is... From historical qualified cross-sections, correctly matched samples The 99.5th percentile value is calibrated and saved in the pairing parameter version record. The larger the value, the worse the consistency between the measured cross section and the historical causal state;

[0094] Only when the status packet quality flag is valid and At that time, the pairer will In , , and With the Establish a read-only association for each credibility status packet, and write the historical rack status index field as... And set the model update permission field to allowed, and save the association at the same time. , , and Provide S4 to verify the pairing quality; when the quality flag is invalid, historical state associations must not be established, the historical index must remain empty, the pairing status must be written as no candidates, and model update permission must remain prohibited;

[0095] Perform field validity filtering before writing historical status: It must be non-negative and not exceed the rated speed of the rolling mill. Must be between zero and the device's allowed upper limit. and They must be located at their respective travel boundaries. If any field is invalid, the original record is retained and the corresponding bit of the valid mask is cleared. Candidate pairings are constructed only from records in which all four rolling status fields are valid. If there are no valid records for more than 50ms, the historical sequence status is set to interrupt.

[0096] The ordered index uses material coordinates as the primary key and time scale as the secondary key, and the matcher uses binary search for location. After determining the insertion position, three valid records are read from each side of the insertion position to form a candidate set. Records with duplicate material coordinates and time scales are deleted. Candidates are then sorted by... Sort by size from smallest to largest. If the errors are the same, sort by absolute value of time standard deviation from smallest to largest. If they are still the same, sort by historical status number. Sort by size from smallest to largest, with the first item as the unique pair output and the remaining candidate errors written into the pair diagnosis field;

[0097] Pairing parameter record saving , , , , , and their respective units, calibration batches, and version numbers, After the measurement frame or rack position is adjusted, the measurement is repeated. The allowable error and weight are obtained by minimizing the mismatch rate of no less than 1,000 historical cross sections with manual confirmation indexes. When the material coordinate difference or time scale difference of 20 consecutive qualified cross sections exceeds their respective allowable errors, a version review is triggered. During the review, the previous valid version is maintained and new samples are prohibited from updating the model.

[0098] When the status packet quality flag is invalid, When the material coordinate validity flag is zero or there is no historical record surrounding the target material coordinates, the matcher will keep the historical rack status index field empty, set the model update permission field to disabled, and... If the actual value exceeds the threshold or the preset no candidate value is 2, the branch does not delete the confidence state packet. S4 only uses the previous valid model parameters to generate a restricted prediction and must not write it into the identification sample or update the parameters. S5 accordingly maintains the model feedforward and the current safety instructions.

[0099] After pairing is complete, the pairer will... , The paired parameter version, model update permission field, and associated historical state summary are written back to the same confidence state packet in material coordinates. Maintain the packet identity and push the updated state packet to the S4 model update queue.

[0100] In this specific embodiment, S4 includes:

[0101] The model updater reads the first output of S3. Each reliability status packet is verified before any size subtraction, checking the status packet quality mark, reference temperature, and thickness. Width of reference temperature Existence, corresponding size valid markings, and target thickness in the current product specification record. With target width Only when the quality mark is valid and Only when both exist, both size valid flags are one, and both target dimensions are valid positive numbers, will the operation proceed according to the specified conditions. and Generate thickness residuals and width residuals, where , All dimensions mentioned above are in mm. A positive residual indicates that the measured dimension exceeds the target dimension, and a negative residual indicates that the measured dimension is lower than the target dimension. The residual is then compared with the material coordinates in the status package. Write the size residual sequence together; if any precondition is not met, write the thickness residual state and the width residual state as unavailable, do not perform subtraction, do not write identification samples and do not enter parameter update, and write the specific missing field or invalid flag as the reason for the unavailable residual into the model update state field of the state package and pass it to S5.

[0102] The compression model and the expansion model use the same dimensionless state feature vector. ,in The historical state index obtained by pairing S3. , and These are the horizontal roll position reference, vertical roll gap reference, and rolling force reference set for the current specification rolling mill, respectively, with units of mm, mm, and MN. The parameters must be taken from the current specification rolling process parameter record and must be greater than zero and within the allowable rolling force range of the equipment. and The normalized dimensions of the non-zero positions identified through trial rolling are in mm. Each component is dimensionless and stored in the model input record; construction or the section of material to be rolled Pre-check For valid positive numbers within the aforementioned range, if the verification fails, rolling force normalization is not performed, corresponding model input is not generated, residual prediction is not calculated, and model parameters are not updated. The model update status field of the status packet records that the rolling force normalization denominator is invalid and continues to use the previous valid model parameters. The corresponding prediction record is marked as restricted prediction. Only when the model feedforward mapping and its bound version verified for the current specification are available, the safety handling flag is written as model feedforward priority. Otherwise, it is written as safety hold and handed over to S5 for processing according to the mutual exclusion branch.

[0103] The model is based on and This represents the predicted thickness residual and the predicted width residual caused by the paired historical state, where and These are the four-dimensional dimensionless parameter vectors for the compression model and the expansion model, respectively. and The unit is mm. The initial parameter values ​​are obtained by least squares identification of the state and dimensional residuals of the three most recent qualified batches of the same steel type and specification through ridge term. The parameter record saves the model parameter version, applicable specifications, sample range and identification time scale.

[0104] When the status packet quality flag is valid, and both the thickness residual status and the width residual status are available, And when model update permission is enabled, the model updater calculates... and ,in and The innovation quantity of the dimensionless model; define the dimensionless fourth-order identity matrix. The non-cross component selection matrix of the compression model And the selection matrix of non-cross components of the extended model ,in This indicates that a diagonal matrix is ​​formed using the listed values ​​as its main diagonal elements. Zero element exclusion The vertical roller gap component in the middle, Zero element exclusion The horizontal roller position component; first press and Only update non-crossing components, where the superscript is updated. This represents a temporary parameter vector for the cross components that have not yet been updated. and The dimensionless reference step size is determined to bring the prediction mean square error to converge through playback of qualified batches. To prevent dimensionless positive numbers with a denominator of zero, thickness reliability is required. and width credibility The actual update step size of non-crossing components is limited to zero to the corresponding baseline step size;

[0105] The model updater will The vertical roll gap component and the horizontal roll position component are identified as cross-coupled parameters in the pressing model and the width expansion model, respectively, and are then processed according to... Limit its update step size, where The dimensionless reference cross step size is determined based on the thickness-width coupling trial rolling data. The joint credibility generated for S2 and The range is 0 to 0.01; the relevant strength used for the relevant penalty is The penalized cross step length is The penalty factor is Define the actual crossover step size. ,when season ,when season ,in The correlation coefficient of the dimensionless thickness-width model residuals obtained from S2. This is the dimensionless actual cross step size after applying the relevant penalty; then according to... and Only update the respective cross components. and Each parameter component retains only one corresponding cross component; thus, each parameter component is updated exactly once within the same cross-section update cycle. Non-cross components no longer accept cross step sizes, and cross components no longer accept cross step sizes. or The step size, the updated model parameters must still satisfy the parameter boundary constraints, and if the constraints are not satisfied or the update is not feasible, the original parameters are retained;

[0106] After each update, the parameter components are trimmed to the boundary of their respective 95% confidence intervals obtained from trial rolling identification, which are then doubled. The version acceptance criterion is whether the normalized sum of squares of innovations of the most recent 20 allowed update sections decreases. Upon successful acceptance, an incremental version of the model parameters is generated. And save , And the sample index; if the acceptance fails, the parameter update will be revoked and the current version will continue to be used. ,in and This serves as a model parameter version identifier, not a numerical computational cost; after version acceptance or retention determination, the model updater writes the actual model parameter version used in this package back to the actual model parameter version field of the same credibility state package, and writes it upon successful acceptance. If the acceptance fails, the gating is updated, or the prediction is restricted, the previous valid value will be written to the actual usage. When a write-back fails, the model update status is recorded as a version write-back failure and the safety handling flag is set to safe hold.

[0107] When the status packet quality flag is invalid, or the thickness residual status or width residual status is unavailable... When the historical rack status index is empty or the model update permission is disabled, the model updater skips the corresponding residual calculation and identification sample writing; if and If a state packet that has passed size pre-verification fails to be generated, but subsequent quality or pairing gating fails, neither of these will be written into the identification sample window; this branch remains... and The model update state field in the state packet records the specific reasons for missing dimensions, invalid dimension validity flags, and retention in quality gating or pairing gating; the model parameter version for this branch remains unchanged. Only the previous effective model is used to generate limited predictions, and the unusable state of the residuals and the reasons therefor are passed to S5 along with the size prediction record. This allows S5 to choose model feedforward or safety hold based on this and not to consume undefined residuals, thus preventing invalid or mismatched sections from contaminating the model.

[0108] For the first rolling mill stand that has not yet passed the target rolling mill stand and is within the control range For each material segment, the model updater updates the model based on the current horizontal roller position of that material segment. vertical roller gap And rolling force prediction structure and in accordance with and The predicted thickness and predicted width are obtained, where This is an index of the material segments to be rolled, arranged in ascending order of material coordinates. and The unit is mm. It is determined by the median of the five most recent effective rolling force samples and the force increment corresponding to the set reduction change, with the unit being MN;

[0109] The model updater updates the material coordinate range of each material segment to be rolled. , , The adopted or The parameter update status and source credibility status package number are written to the size prediction queue, and the thickness residual status, width residual status, prediction status, model parameter version, and safety handling flag are also written to each prediction record. The model parameter version of each prediction record is copied from the actual model parameter version field of the source credibility status package. After copying, the consistency between the two is verified byte by byte. If the version write-back or copy verification fails, a normal prediction record is not generated and the safety handling flag is set to safe hold. The safety handling flag uses three mutually exclusive enumerated values: allow optimization, model feedforward priority, and safe hold. When a missing or unrolled material segment exceeds the valid range of the model parameter record, the prediction record is marked as a restricted prediction. If the current specification's verified model feedforward mapping and its bound version are both available, the safety handling flag is set to "model feedforward priority"; otherwise, it is set to "safety hold". Restricted records are still sent to S5 in the material coordinate order along with the original confidence status package. They must not be deleted from the queue, skipped, or bypassed by subsequent material segments. The safety handling flag is set to "optimization allowed" only when the prediction input is complete, the unrolled material segment is within the valid range of the specification, the actual model parameter version of the source status package is successfully written back, the prediction record copy version is successfully verified byte by byte, and the prediction status is "normal prediction". Optimization allowed is not allowed if any condition is not met. S5 can only enable optimization when the first unrolled segment record is a normal prediction, the safety handling flag matches word by word and allows optimization, and the other first-segment controls are passed.

[0110] In this specific embodiment, S5 includes:

[0111] The joint controller operates on a 100ms control cycle. Each cycle begins by using the material coordinates and source status package number of the first section to be rolled in the S4 size prediction queue as the joint key to read the unique corresponding [data / data]. A confidence status packet is generated, and the target thickness and target width are verified in a fixed order first, and the thickness residual status and width residual status are both valid positive numbers. and All conditions are met; the predicted status of the first section to be rolled is normal; the safety handling mark of the record is matched word by word, allowing optimization; and the version of the model parameters carried by the prediction record exists and is consistent with its source status package. The quality mark is then verified. , Threshold version and paired material coordinates; the threshold version verification objects include the residual threshold version, confidence weight version, and paired parameter version carried in the status packet. All three must exist, their version timestamps must cover the current control time, and they must be correlated with the generated current physical residual and residual status, respectively. and The threshold version verification is considered passed only if the recorded versions are identical word by word. If any version is missing, expired, or inconsistent, cross-sectional feedback and bias updates are prohibited, and the specific failed object is written into the control solution state. If the model feedforward mapping, mapping input, and their bound versions verified in the current specification are all available, the system enters the model feedforward phase; otherwise, it enters the safety hold phase and retains the most recently confirmed safety instruction. The controller uses the first confidence threshold within the confidence range of 0 to 1. Second confidence threshold Both are dimensionless numbers determined by replaying historical qualified cross-sections to ensure that the false feedback rate does not exceed 0.5%, and satisfy the following conditions: The pairing error threshold follows the dimensionless value of S3. This status packet With the first segment of the prediction time domain One-to-one correspondence, Material segments only use S4 model predictions and do not use the current section residuals; when any residual is unavailable, section feedback and offset updates are prohibited and a model feedforward channel is selected; when the first segment to be rolled is a restricted prediction, the first segment must not be skipped or replaced by a subsequent material segment. When the safety handling flag is model feedforward priority, the model feedforward mapping verified for the current specification is called. When called, its input and binding versions are verified again. If the verification fails, the most recently confirmed safety instruction is retained. When the safety handling flag is safety hold, the most recently confirmed safety instruction is directly retained; when the target size is invalid, the prediction input is missing, or the model parameter version is missing or inconsistent, this control cycle is uniformly handled as safety hold, and model feedforward or quadratic programming must not be called; only after all the above-mentioned head integrity access control and threshold version verifications have passed, is the status package valid. and First, select the cross-section feedback and offset update channels, and ensure the status packet is valid. and First, select the offset update channel only. Invalid status packet, mismatched material coordinates, etc. or The model feedforward channel is selected directly at the time; only after the channel is determined is the feedforward channel in the prediction time domain sorted by material coordinates selected. The first material segment, and the first Segment prediction deviation composition ,in To predict the material segment index in the time domain, Both components are in mm. and The product specification record from the corresponding material segment predicts a time domain coverage of 0.6s, and each control only executes the first increment of the allowed optimization branch solution sequence;

[0112] The controller employs a linearized thickness-width response only for branches at the beginning of the cycle where the cross-section feedback and bias update channels are selected, or only the bias update channel is selected. ,in For the control state vector, The motion control quantity is expressed in mm. It is a 2x2 dimensionless residual transfer matrix. This is a 2x2 dimensionless roll gap sensitivity matrix. This is a disturbance vector in mm, calculated using the S4 model from the deviation of the current rolling force from the reference value. and From the current model parameter version Calculated and corrected using the measured responses of the 20 most recently allowed updated sections, matrix validity is determined by the absolute value of the determinant not being less than The elements are jointly determined when they are within the trial rolling identification boundary. The control objective is to minimize the objective function and satisfy the execution constraints in each control cycle. When the matrix is ​​abnormal or the constraints are infeasible, the system switches to model feedforward and the current optimization control quantity is not effective.

[0113] Model parameter version During deployment, samples are selected from the historical records where command feedback pairing was completed at the time of deployment, and the response matrix records are generated and frozen synchronously. The controller first follows Get the first Initial values ​​of the model sensitivity for segment, where , Versions The normalized components of the horizontal roller position and the normalized components of the vertical roller gap in the parameter vector of the pressing model are respectively... , For the corresponding components of the same version of the extended model, , The target size follows the definition mentioned above. The units of length for both the numerator and denominator of each element are mm, and all are dimensionless after simplification. A fixed window is formed from the 20 most recently updated historical cross-sections of the same steel type and specification, whose updates have been completed and whose command feedback has been paired. The model version used in historical control is not required to be the same as the newly released one. But must Pairing was completed before publication, and adjacent states were saved for each sample. and Actual motion, consisting of the actual position difference within the same period and measured in mm. Disturbance in mm and the corresponding sample specifications ,in The unit is per mm; for each sample within the window The controller reads the target thickness stored in the sample. Target width Location normalization scale and And uniformly read the parameter version of the newly released model. The four corresponding parameter components, Prediction segment index in matrix Replace with sample index Then recalculate the sample sensitivity. The old sensitivity calculated using the historical control version of the sample must not be used; the sample steel type or specification must be consistent with... Inconsistency, missing target size or normalization scale, incomplete command feedback pairing, incorrect pairing timing, or excessive sample input. When the sample falls within the valid range of the specifications, the sample is taken from... Excluded samples without replacing them with other specifications; identify the three most recent qualified batches of the same steel grade and specification using a two-by-two dimensionless matrix with the same regression structure. To pass initial values ​​to the matrix and calculate sequentially , , as well as Then set the prediction time domain segments. and ;in For by sample Save input and version The calculated sensitivity matrix of the 2x2 dimensionless model. For the sample The regression residual vector, with each component in mm. This is the dimensionless residual vector after sample tolerance normalization. Let the dimensionless objective function consist of the normalized sum of squared residuals and two ridge penalties. This is a 2x2 dimensionless sensitivity correction matrix. and To determine and write through closed-loop playback of qualified batches The positive dimensionless ridge coefficient, and Let L2 and Frobenius norms be represented respectively. The normalized residuals and both ridge penalties are dimensionless, and the positive ridge coefficients ensure the uniqueness of the correction rule. When the window has fewer than 20 paired sections after removing unqualified samples, the previous valid section of the same specification in the parent model version is used. and If no previous valid record exists, it will be Mark as unavailable and prohibit secondary planning; save Window start and end section numbers, original historical control version and usage of each sample The indicator of recalculated sensitivity , Ridge coefficients, identification timescales, and check hashes are frozen, and samples generated after the release time are no longer accepted. Subsequent new samples are only used to generate new response matrix records when the next model parameter version is released; each size prediction record in S4 also carries a string strictly bound to its model parameter version. The identifier, S5 requires prediction of all records in the time domain. and They are consistent and read according to the record version. , If any record version is missing, inconsistent, or fails to be verified, the quadratic programming solution must not be solved and the model feedforward solution must be switched. If the feedforward bound version is also unavailable, the most recently confirmed safety instruction must be retained.

[0114] Construct dimensionless thickness and width bias costs for the prediction time domain respectively. and Dimensionless action cost and in accordance with Form control objectives, among which and for The thickness and width deviations are in mm. and Positive dimensional tolerances given for product specifications, in mm. and The rated single-cycle displacement of the actuator is expressed in mm. and The dimensionless motion penalty weight is calibrated through closed-loop playback. , and All are dimensionless numbers. The smaller the value, the lower the combined cost of the predicted size deviation and the actuator movement.

[0115] Weighting of thickness deviation term The width deviation term is weighted as follows: The weight of the thickness-width coupling term is taken. ,in , and For dimensionless benchmark weights, , and The confidence state packet from the most recent control time and corresponding to the material coordinates is used to ensure the positive semi-definiteness of the objective function. Limit to no more than Increased reliability leads to a monotonically increasing correction weight for the corresponding dimensional deviation in the solution;

[0116] Actuator displacement constraints are written as and The rate constraint is written as and ,in and This is the current command position in mm. , , , The boundary values ​​in mm are given for the equipment travel record. and The actuator's permissible speed is expressed in mm / s. s is the control period; used for the first segment of the prediction time domain. and Take the increment that was actually issued in the previous control cycle and has been limited in amplitude and speed. When the system starts, the control is reset, or there is no previous valid cycle, both are taken as 0 and written into the solver input record of the current cycle.

[0117] The controller uses the current state collected in real time. and motion control quantity , ,according to Predicting and applying rolling force At the same time, apply ,in Current rolling force and the upper limit of rolling force The units are all MN. and The force sensitivity identified by trial rolling of the same specification is expressed in MN, where the change in rolling force corresponding to each 1 mm actuator displacement is measured. To determine the minimum anti-collision clearance in mm, calibrated based on the vertical roll type, thermal expansion allowance, and mechanical protection distance, the stroke and [other parameters] are remeasured after equipment maintenance. When switching specifications, update the force sensitivity version and remove the corresponding action if the prediction exceeds the rolling force limit or the anti-collision constraint is not feasible.

[0118] The joint credibility uses a monotonically increasing mapping to the upper limit of the single-cycle action control quantity. and and constraints , The joint confidence level is the input state. , The minimum confidence level action limit, expressed in mm, is calibrated and stored in the control parameter version record by the corresponding actuator's minimum stable step size, position resolution, and closed-loop playback of abnormal samples, and must satisfy the following conditions: and ; , The rated single-cycle operation limit is expressed in mm. The joint confidence level is between 0 and 1. The control objective is to make the change monotonically limited within the constraints of this period. When the mapping is abnormal or the constraints are infeasible, it switches to the minimum action limit and takes effect in the next period.

[0119] After the initial channel determination, cross-sectional feedback state correction, or bias state update of the cycle are completed, and only when optimization is allowed for that branch, the controller will assemble the above model, objective function, and constraints into a convex quadratic programming problem. Using the current command and the solution of the previous cycle as initial values, the active set solver will iterate no more than 30 times within each 100ms control cycle. The solver will divide the position residual by the corresponding stroke width, the rate residual by the corresponding single-cycle allowable displacement, and the rolling force residual by the positive allowable upper limit. Divide the anti-collision gap residual by a positive number The resulting constraint residuals are all dimensionless, with the maximum violation of the normalized constraint not exceeding [a certain value]. And the change between two consecutive dimensionless target values ​​does not exceed As a termination condition, the obtained solution is used as the increment of horizontal roll reduction and vertical roll gap, and only the first increment of the solution sequence is issued. If the solution does not converge within 50ms, the matrix validity judgment fails, or the constraint is infeasible, the current optimization increment is not issued. The feedback channel identifier is set as the model feedforward channel and the current model steady-state roll gap setting is used with the current model with amplitude and speed limited.

[0120] The first confidence threshold used in the periodic header Second credibility threshold and dimensionless pairing error threshold Following the definition in the first paragraph of this step, the first confidence interval is: The second confidence interval is The third confidence interval is and with Verify the order and interval exclusivity of the two confidence thresholds;

[0121] The bias update channel saves the thickness residual integral state. and width residual integral state and with and Update, in which The coordinate increment of adjacent materials is in mm. , The unit is mm. , and positive number limit , All units are mm squared; system startup, control reset, and first package order for new specifications. The previous valid integral state is inherited only when the specifications are the same and the control parameter versions are consecutive. and These are determined by the product of the product's allowable long-term thickness deviation, allowable long-term width deviation, and the upper limit of the effective material window length, and are stored in the control parameter version record. The dimensionless channel update coefficient is between 0 and 1. This means clipping the integral state to the given upper and lower limits;

[0122] When the beginning of the cycle is determined to be a cross-section feedback and bias update channel, the controller is set... ,Will and Simultaneously write to the cross-section feedback channel and allow updates. and And before constructing the objective function and starting the solver, Replace the first segment of the predicted time domain state with the current measured residual; the bias update channel follows... and The slowly varying bias is obtained, where The difference between the first and last material coordinates of at least 20 consecutive effective sections, trimmed to the range of 500mm to 5000mm given in the control parameter version record, in mm; if there are fewer than 20 consecutive effective sections, the previous effective section is used. There is neither a sufficient continuous cross section nor a previous effective section. When this happens, cancel the integral increment for this period and maintain... and If a previous valid bias exists that is bound to the current specification and control parameter version, then that bias will be retained. and Otherwise, set both to 0 until a valid result is obtained. Bias updates were only resumed later. and The unit is mm. Positive offset indicates that the long-term measured deviation of the corresponding size is positive. When the periodic head is determined to be an offset-only update channel, the original prediction is explicitly maintained. Do not perform current residual replacement, and set This ensures that the laser measurement residuals only enter the bias update channel with a step size limited by joint confidence.

[0123] After completing the cross-sectional feedback state replacement and bias update or hold, and before constructing the objective function, the controller uses the slowly varying bias as the sole consumption entry point for the predicted state: in the cross-sectional feedback and bias update channels, it holds the state already replaced by the current measured residual. No longer superimposed bias, only for implement In the bias-only update channel, the current section residual is not replaced. , and for implement ;in , and Following the definition used in this step, both components of the bias vector are in mm, and are positive. or A negative value will necessarily increase the prediction bias after correction, while a negative value will necessarily decrease it. The corrected values ​​will then... Enter directly and The controller simultaneously saves the data used in the current cycle in the traceability record. state, , Before revision And after correction This allows for the reproduction of the actual impact of the bias on the control quantity. Only after the above corrections are completed can the optimization problem be constructed and solved.

[0124] When the beginning of the cycle is determined to be a model feedforward channel, the controller is set... and maintain , It directly bypasses the aforementioned model construction, objective construction, and solution steps of quadratic programming, does not use invalid or mismatched size states, and does not read, generate, or consume data. and The recovery counter only works if the following conditions are met simultaneously across a continuous section: the state package quality flag is valid, both the target thickness and target width are valid, both the thickness residual state and width residual state are available, the first section to be rolled is in normal prediction, and the versions of the model parameters and response matrix records in the prediction records exist and are consistent. and The count increments sequentially; otherwise, it is reset to zero. The count continues until the threshold for the number of restored cross-sections is reached. Only then will the residual usage and integral status update be resumed in the next control cycle. Determined by the length of the shortest stable segment in the replay of abnormal samples;

[0125] The model feedforward channel reads the target thickness from the current product specification record. and target width Real-time rolling force Current horizontal roller pressing position Current vertical roller gap Previous effective model parameter version And the version of the feedforward control parameters bound to it; the steady-state roll gap calibration record of this version uses steel grade, specification, target thickness, and target width as search keys. The value fields include reference rolling force, steady-state position of horizontal roll, steady-state gap of vertical roll, horizontal roll force correction coefficient, vertical roll force correction coefficient, applicable thickness and width range, applicable rolling force range, and version time scale. Among them, the position and gap are in mm, the rolling force is in MN, and the force correction coefficient is in mm / MN. This coefficient only completes the calibration conversion from rolling force difference to roll gap displacement, and does not involve torque, lever arm, mass, or gravity conversion. Each record is formed by taking the median of roll gap and rolling force from qualified trial rolling sections of the same steel grade and specification when the dimensions enter the tolerance and are continuously stable for no less than 20 sections. The record is frozen during model version acceptance and is linked to the model version. Jointly released; during runtime, bilinear interpolation is performed between adjacent valid grid records of the target thickness and target width to obtain two steady-state roll gap base values, and then the current values ​​are calculated. The difference between the interpolated reference rolling force and the stress correction factor is multiplied to compensate, and then the current value is used. , Starting from the point of change, reproducible horizontal roll feedforward targets and vertical roll feedforward targets are obtained;

[0126] When the interpolated horizontal roll feedforward target and the vertical roll feedforward target simultaneously touch the constraints due to thickness-width coupling, the controller handles the conflict in the following order: anti-collision constraint, rolling force upper limit, equipment position and rate boundary, thickness tolerance control, and width tolerance control. It first projects or tightens lower priority components; if conflicts still exist, it maintains the corresponding current safety command. The target thickness, target width, or... When the application range of the calibration record is exceeded, boundary recording is only used when the thickness, width and rolling force distance of the most recent valid record of the same steel type and specification do not exceed the calibration grid spacing of the record. Otherwise, both components maintain the current safety instruction. The maintenance is also performed when the record is missing, the version verification fails or any input field is invalid. The two targets obtained by feedforward are checked in sequence by equipment position limit, single cycle rate limit, minimum clearance for vertical roll anti-collision and upper limit of predicted rolling force before they are issued as candidate positions. This branch always uses the calibration mapping result as the candidate position and does not refer to non-existent optimization increments.

[0127] For branches that allow optimization, the first increment obtained from the solution is added to the current instruction to form a candidate position. and For the model feedforward branch, and The horizontal roll feedforward target and the vertical roll feedforward target generated by the aforementioned calibration mapping are directly taken; the controller then executes uniformly according to the equipment displacement boundary. and The units for both the candidate position and the position after clipping are mm. This section indicates that values ​​less than the lower bound are taken from the lower bound, values ​​greater than the upper bound are taken from the upper bound, and all others retain their original values.

[0128] After position limiting, the controller calculates... and ,in , All are single-cycle variation upper limits in mm. Therefore, the greater the joint confidence, the greater the allowable variation upper limit is monotonically increasing, but it will never exceed the actuator rate constraint.

[0129] To suppress instruction reversal between adjacent cycles, the controller calculates the current cycle's and the previous cycle's instructions respectively. , When the product of the same components is less than zero, the inversion confirmation of that component is initiated, and the thickness inversion dead zone is activated. Width inversion dead zone ,in For the confirmed material length in mm, To reverse the confirmation quantity threshold, This represents the average material coordinate increment of the three most recent effective cross-sections, in mm. and The unit is mm square. and The actuator position resolution is expressed in millimeters. and The horizontal roll pair thickness diagonal component and the vertical roll pair width diagonal component of the roll gap sensitivity matrix corresponding to the current model parameter version are taken respectively. These are identified by trial rolling of the same specification and bound to the model version. Both are dimensionless numbers within the valid range of trial rolling identification. Reversal of the corresponding components is prohibited when there is no valid version or the range is exceeded. and All units are mm squared;

[0130] The horizontal and vertical rollers each maintain independent thickness reversal confirmation counters. Width Reversal Confirmation Counter Both are 0 to Integer; for horizontal rollers, when When the request symbol is inverted relative to the previously issued non-zero horizontal component, it is only valid in the current cross-sectional state and the request is non-zero and Only on a continuous cross section where all three authorization conditions are simultaneously satisfied can it be allowed Add one to the previous cross-section count; when At this time, the horizontal component of the current period is set to zero but the count is retained, reaching the desired level. Only allow the horizontal reversal and then immediately reset to zero. If any authorization condition is not met, the horizontal component of this period will be set to zero and cleared. ; opposite rollers, when When the request symbol is reversed relative to the previously issued non-zero vertical roller component, it is only valid in the current cross-sectional state and the request is non-zero and Only on a continuous cross section where all three authorization conditions are simultaneously satisfied can it be allowed Add one to the previous cross-section count; when At this time, the vertical roller component of the current cycle is set to zero but the count is retained, until the target is reached. Only allow the vertical roller to reverse once and then immediately reset to zero. If any authorization condition is not met, the vertical roller component for this cycle will be set to zero and cleared. When the corresponding component request is zero, has no explicit reversal symbol, or the request direction is restored to the previously issued direction, the counter of that component is cleared only because there is no reverse request in the current cycle, and its non-reversal component is processed according to the normal amplitude and speed limiting results; when degraded branches such as model feedforward, safety protection, feedback anomaly, and execution tracking anomaly are triggered, the corresponding reverse component in the current cycle is set to zero and its counter is cleared, and subsequent reverse requests must start counting continuously from the beginning; the counting, zeroing, release, or clearing of any component must not change the state of another component, and the final horizontal roll pressing instruction is issued. And vertical roller gap command Write to the distribution buffer; the unit for both instructions is mm.

[0131] Each time a command is issued, the controller generates a globally unique and monotonically increasing dimensionless command cycle identifier for the paired commands of the horizontal and vertical rollers. Command timestamps recorded in milliseconds And generate sections with serial numbers Material coordinates and For traceability records with a union key and a feedback status of pending pairing; control status inputs include reference temperature. , All physical residuals , , , , Historical rack status index and pairing error The motion control quantity field includes the feedback channel identifier and the model parameter version. , and The offset used in this cycle and the state before and after offset correction; the position sampling point within the device's allowable response time. Upon expiration or at a pre-fixed feedback sampling point in the next control cycle, the control parameter version will select one of these two options for execution. The unit is milliseconds (ms), and the 99.5th percentile of the time required for the device to reach the command position tolerance during an unloaded step test is taken. Feedback messages must be sent back. and feedback timescale in milliseconds Only when the feedback cycle identifier is consistent with Consistent and Only then is the actual position of the horizontal roller read by the position sensor. actual position of vertical roller Write the pairing record and mark it as paired. This control cycle aims to minimize the instruction tracking error and satisfy the execution constraints, while also saving the threshold version, control solution status, and acquisition time stamp. ;

[0132] The controller only calculates the tracking error for paired records that have consistent command cycle identifiers and valid timing; to unify the recursive notation, the original control cycle... Issued The completed paired records, consisting of the instruction and its subsequent feedback within the same period, are uniformly re-recorded as an index during error judgment. Therefore, the controller uses the absolute value of the difference between the horizontal roller pressing command and the actual position of the horizontal roller as the horizontal roller tracking error, and the absolute value of the difference between the vertical roller gap command and the actual position of the vertical roller as the vertical roller tracking error, and executes accordingly. and ,in , The units for both command position and actual position are mm, and the execution tracking error threshold is... , Take the maximum value between the 99.5th percentile of the no-load step test error and twice the position resolution, in mm. , or If any of these conditions are met, it is determined to be an execution trace anomaly;

[0133] Feedback has not arrived even after the timeout period specified in the control parameter version, or the feedback cycle identifier is missing or inconsistent with... If there is inconsistency, the feedback timescale is earlier than the allowed sampling time, the actual location field is missing, or the trace record writing fails, the controller will mark the corresponding command cycle as a feedback anomaly. This feedback must not be directly subtracted from the current or subsequent new instructions, nor should it be generated. and The abnormal feedback branch immediately disables the new optimization increment, retains the most recently confirmed safety instruction, and sets... and maintain and The current value will tighten the upper limit of the change in the next cycle to... and until continuous The safety hold is only exited after feedback confirming that all command cycle identifiers are consistent and the timing is correct; when an execution tracking anomaly occurs in a completed pairing record, the controller immediately switches the motion control quantity to the model feedforward channel, using tracking error, rolling force, confidence level, and pairing error as input states. and maintain and The current value, while tightening the upper limit of change for the next cycle to and The goal of this control cycle is to restore tracking stability within the execution constraints. If the recovery conditions are not met, the degraded state is maintained. All cross sections satisfy , , , and At that time, the residual use channel is reopened in the next cycle according to the aforementioned recovery conditions. Finally, the horizontal roll pressing command and the vertical roll gap command, which have been confirmed by channel judgment, amplitude limit, speed limit and reversal, are sent to the target rolling stand for execution.

[0134] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0135] This invention establishes a continuous processing chain comprising online laser cross-section reconstruction, material coordinate pairing, reliability assessment, and joint thickness-width control. The reliability state packet maintains the same material coordinates across all processing stages and carries dimensions, reliability, historical state index, and channel status, enabling the controller to distinguish between measurement data suitable for model updates and closed-loop feedback, and measurement data only suitable for model feedforward.

[0136] Compared to control structures that directly adjust the roll gap based solely on the current dimensional deviation, this invention adds pairing error gating, model update or parameter retention branches, and reliability grading channels, and respectively limits the changes and execution tracking states of horizontal and vertical roll commands. This allows the model parameters and control actions to adapt to the measurement quality, and maintains a traceable degradation control path in the event of measurement or execution anomalies.

Claims

1. A closed-loop control method for online laser measurement in hot-rolled flat steel rolling, characterized in that, include: S1. Simultaneously collect the surface profile, reference target detection results and rolling status data of the same section on the exit side of the rolling mill stand, perform unified coordinate calibration and temperature drift correction on the surface profile, and obtain the thickness and width at the reference temperature. S2. Construct physical residuals based on surface profile, reference target detection results, and rolling state data to generate thickness confidence, width confidence, and joint confidence; integrate the rolling speed to obtain material coordinates, and encapsulate the material coordinates, thickness, width, and confidence into a confidence state package; S3. Match the confidence status package to the historical rolling state when the section passes through the stand according to the material coordinates, obtain the historical stand state index and matching error, and write them into the confidence status package. S4. Generate thickness residuals and width residuals based on the thickness and width in the credibility state packet and the target thickness and target width; When the pairing error is not greater than the preset pairing error threshold, the residual is used to update the compressed model and the widened model. When the pairing error is greater than the threshold, the model parameters are kept and the thickness and width of the material segment that has not yet passed through the rack are predicted using the updated or kept model. S5. Based on the predicted thickness and width and the confidence status package, jointly solve the horizontal roll reduction increment and vertical roll gap increment under preset execution constraints. Determine the use channel of the laser measurement residual based on the relationship between the joint confidence and the pairing error and the corresponding threshold, and output the roll gap control command.

2. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 1, characterized in that, S1 includes: The upper and lower laser measuring heads and the left and right laser measuring heads are triggered by a common trigger time mark to collect the contour point set of the same cross section, and the temperature, encoder speed, rolling force, horizontal roll pressing position and vertical roll gap are collected. The reference feature on the reference target is measured in a unified reference coordinate system, and the coordinate system of each laser measuring head is transformed to the unified reference coordinate system. The contour point coordinates are corrected according to the difference between the cross section temperature and the reference temperature and the linear expansion coefficient of the hot-rolled flat steel material, and the optical zero drift of each laser measuring head is corrected according to the displacement of the reference feature of the reference target relative to its calibration position. The thickness is determined by the corrected upper and lower surface contour spacing, and the width is determined by the corrected side surface contour spacing.

3. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 1, characterized in that, In step S2, the physical residuals include the geometric closure residual relative to the laser measuring head, the drift residual of the reference target reference feature, the model residual between the dimension at the reference temperature and the dimension predicted by the temperature and rolling state data, and the timescale residual between the contour acquisition timescale and the rolling state acquisition timescale of the same cross section. Normalized residuals are generated by the ratio of the absolute value of each physical residual to the corresponding residual threshold, wherein the residual threshold is determined by the reference target calibration data and the residual distribution quantile of historical qualified rolled cross sections. Thickness confidence is generated based on the normalized residuals associated with thickness measurement, and width confidence is generated based on the normalized residuals associated with width measurement. The thickness-width residual correlation coefficient is estimated based on the thickness normalized residual sequence and the width normalized residual sequence in the historical qualified rolled cross section samples. Joint confidence is generated based on thickness confidence, width confidence, and thickness-width residual correlation coefficient. Dimensional covariance is generated based on the variance of the two normalized residual sequences and the thickness-width residual correlation coefficient.

4. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 3, characterized in that, In step S2, the material coordinates are obtained by integrating and accumulating the adjacent sampling intervals of the common trigger timescale with the rolling speed; the confidence status package also includes the size covariance, the reference target drift state, the physical residual, the historical rack status index field, the feedback channel identifier field, and the acquisition timescale, wherein the historical rack status index field is null before reverse matching, and the feedback channel identifier field is undetermined before control solution.

5. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 4, characterized in that, S3 includes: writing encoder speed, rolling force, horizontal roll pressing position, and vertical roll gap into a historical state sequence according to a common trigger timescale; determining the target material coordinates and target timescale of the measured section passing through the target rolling mill based on the material path length between the laser measurement position and the target rolling mill stand and the speed integral result; retrieving historical rolling states adjacent to the target material coordinates from the historical state sequence and generating a historical mill stand state index; using the weighted sum of the material coordinate difference and timescale difference of the measured section after normalization by their respective allowable errors as the pairing error; associating the historical rolling state with the credibility state package and allowing model updates only when the pairing error is not greater than the pairing error threshold.

6. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 5, characterized in that, S4 includes: The thickness residual is generated based on the difference between the thickness at the reference temperature and the target thickness, and the width residual is generated based on the difference between the width at the reference temperature and the target width. The parameter update step size of the reduction model and the width expansion model is limited by the thickness confidence level and the width confidence level, respectively. The cross-coupling parameter update step size between the reduction model and the width expansion model is limited by the joint confidence level. The updated reduction model and width expansion model are applied to the material coordinate range that has not yet passed the target rolling mill stand. The predicted thickness and predicted width are obtained by combining the current horizontal roll reduction position, vertical roll gap and rolling force corresponding to the material coordinate range.

7. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 6, characterized in that, In step S5, a control objective is constructed using the difference between the predicted thickness and the target thickness, and the difference between the predicted width and the target width. The weights of the thickness deviation term, the width deviation term, and the thickness-width coupling term are determined by the thickness confidence level, the width confidence level, and the combined confidence level, respectively. The actuator displacement constraint is determined based on the available stroke of the horizontal roll and the vertical roll, the actuator rate constraint is determined based on the allowable displacement difference between adjacent control cycles, the rolling force constraint is determined based on the allowable upper limit of the rolling force, and the roll gap lower limit constraint is determined based on the allowable minimum gap to prevent roll collisions. The control objective is solved within the constraints according to the preset solution termination condition, and the obtained solution is used as the horizontal roll reduction increment and the vertical roll gap increment. The upper limit of the increment for a single control cycle is determined by the combined confidence level.

8. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 7, characterized in that, In step S5, a first confidence threshold and a second confidence threshold are set within the confidence value range of zero to one. The difference between the first confidence threshold and the second confidence threshold is a positive number. The closed interval from the first confidence threshold to one is used as the first confidence interval, the left-closed and right-open interval from the second confidence threshold to the first confidence threshold is used as the second confidence interval, and the left-closed and right-open interval from zero to the second confidence threshold is used as the third confidence interval. The laser measurement residual integral state of the bias update channel is formed by accumulating the thickness residual and width residual according to the coordinate increments of adjacent materials, respectively. When the joint confidence is in the first confidence interval and the pairing error is not greater than the pairing error threshold, the laser measurement residual is allowed to enter the cross-section feedback channel and the bias update channel. When the joint confidence is in the second confidence interval and the pairing error is not greater than the pairing error threshold, the laser measurement residual is prohibited from entering the cross-section feedback channel, and the laser measurement residual is allowed to enter the bias update channel with an update step size limited by the joint confidence. When the joint confidence is in the third confidence interval or the pairing error is greater than the pairing error threshold, the current value of the laser measurement residual integral state is maintained and its update is prohibited. The channel is switched to the model feedforward channel that generates roll gap commands based on rolling force, horizontal roll pressing position and vertical roll gap. After the joint confidence of the cross-sections that continuously reach the recovery cross-section number threshold are all in the first confidence interval and the pairing error is not greater than the pairing error threshold, the update of the laser measurement residual integral state is restored.

9. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 8, characterized in that, The roll gap control command includes a horizontal roll pressing command and a vertical roll gap command. The limiting and speed limiting processing of the roll gap control command includes: trimming the target positions of the horizontal roll pressing command and the vertical roll gap command according to the actuator displacement constraints; trimming the command change amount of adjacent control cycles according to the actuator speed constraints; and performing a monotonically increasing mapping on the upper limit of the command change amount of a single control cycle according to the joint reliability; calculating the horizontal roll pressing command change amount and the vertical roll gap command change amount of the current control cycle and the previous control cycle respectively; when the change amount of the same command component has opposite signs in the two control cycles, the corresponding command component is allowed to reverse only after the cumulative value of the thickness residual corresponding to the horizontal roll pressing command or the cumulative value of the width residual corresponding to the vertical roll gap command exceeds the corresponding reversal dead zone boundary and the number of continuous sections reaches the reversal confirmation number threshold. The reversal dead zone boundary is determined according to the corresponding component in the dimensional covariance and the position resolution of the corresponding actuator.

10. The online laser measurement closed-loop control method for hot-rolled flat steel rolling according to claim 9, characterized in that, The material coordinates, thickness and width at reference temperature, physical residual, thickness confidence, width confidence, joint confidence, dimensional covariance, historical rack status index, pairing error, feedback channel identifier, model parameter version, horizontal roll pressing command, vertical roll gap command, actual horizontal roll position, and actual vertical roll position of each confidence status package are written into the traceability record. The absolute value of the difference between the horizontal roll pressing command and the actual horizontal roll position is used as the horizontal roll tracking error, and the absolute value of the difference between the vertical roll gap command and the actual vertical roll position is used as the vertical roll tracking error. Any tracking error exceeding the corresponding execution tracking error threshold or the rolling force reaching the upper limit of the rolling force is determined as an execution tracking anomaly. When an execution tracking anomaly occurs, the feedback channel identifier is switched to the model feedforward channel and the current value of the laser measurement residual integration state is maintained. After both tracking errors are not greater than the corresponding execution tracking error threshold and the rolling force has not reached the upper limit of the rolling force, the laser measurement residual usage channel is restored according to the recovery conditions.

Images