A cold hard strip straightening flatness defect prediction and re-curl risk assessment system

Abstract

The present application belongs to the field of cold hard strip steel processing and intelligent control technology, and discloses a cold hard strip steel straightening plate shape defect prediction and re-cambering risk assessment system, which comprises a production data acquisition unit, a historical sample management unit, a strip steel section registration unit, a release sensitive feature generation unit, a compound risk prediction unit and a straightening parameter correction unit. The system uses the strip steel material coordinate registration entrance plate shape data, straightening process data and exit plate shape data to generate plate shape changes, plate shape distribution differences, tension changes and straightening action characteristics; historical strip steel sections that are qualified in terms of exit plate shape but exceed the standard after being cut, left to stand or transported for re-inspection are used to establish re-cambering samples, a multi-output neural network model is used to output longitudinal cambering, transverse cambering, edge wave, middle wave and re-cambering risk values, and straightening parameter setting values for subsequent adjustment of strip steel sections are generated according to risk threshold values and a parameter mapping table, and are output after being checked against equipment control limits.

Description

Technical Field

[0001] This invention belongs to the field of cold-hardened strip steel processing and intelligent control technology, specifically a cold-hardened strip steel straightening plate shape defect prediction and warping risk assessment system. Background Technology

[0002] Cold-rolled strip steel typically exhibits high work hardening and uneven residual stress. During rewinding or multi-roll straightening, longitudinal warpage, transverse warpage, edge waviness, and center waviness are usually improved by adjusting the inlet tension, outlet tension, straightening roll gap, and bending roll reduction. Existing production lines generally install shape detection devices on the inlet and outlet sides of the straightening roll system and determine the shape status based on the inlet shape, outlet shape, and straightening process parameters. CN102672003A discloses a scheme for setting tension straightening parameters based on finite element analysis and regression relationships. CN111241657A discloses a scheme for predicting cold-rolled strip steel shape and adjusting process parameters based on deep learning.

[0003] Because the inlet strip shape detection position, the straightening position, and the outlet strip shape detection position are arranged at intervals along the production line, the data obtained at the same sampling time actually correspond to different strip segments. When the strip speed changes, stops, or parameters are switched, associating the data only according to the sampling time can easily cause a mismatch between the inlet strip shape, straightening parameters, and outlet strip shape. In addition, the strip is still constrained by tension, roller system, and curling state during outlet inspection. The outlet strip shape is within the allowable deviation range, but there is still a possibility that the strip shape may change again after the constraints are removed.

[0004] After shearing, settling, or transshipment, some strip steel may exhibit longitudinal warping, transverse warping, edge waviness, or center waviness again as the aforementioned constraints are removed. Existing strip shape prediction and parameter control typically use the exit strip shape as the evaluation result, and subsequent re-inspection data are mostly recorded based on steel coils or plates. A stable correlation has not been established with the specific strip position and corresponding process parameters during straightening. Therefore, it is difficult to identify strip steel sections that pass the exit inspection but subsequently experience warping again, and it is also difficult to form a basis for parameter adjustment corresponding to subsequent strip steel sections. Summary of the Invention

[0005] This invention aims to address the problem that, during the rewinding and multi-roll straightening process of cold-hardened strip steel, the exit shape detection results are within the allowable deviation range, but after subsequent shearing, settling, or transfer, the strip steel exhibits longitudinal warping, transverse warping, edge waviness, center waviness, or re-warping due to the release of tension constraints, roll system constraints, or curling constraints. Existing shape prediction methods mainly predict the exit shape based on the inlet shape and straightening process parameters, or adjust the straightening parameters based on exit shape defects. It is difficult to identify secondary shape changes caused by the release of constraints using the re-inspection results after the exit inspection is qualified, and it is also difficult to convert this risk into straightening parameter settings applicable to subsequent strip steel sections.

[0006] To address the aforementioned technical problems, this invention provides a system for predicting shape defects and assessing warping risks in cold-hardened strip straightening, applied to the multi-roll straightening process in a cold-hardened strip rewinding production line. Following the sequence of strip passing through the straightening production line, the strip segment that has completed exit shape inspection is identified as the reference strip segment, and the strip segment located after the reference strip segment and not yet passed through the straightening roller system is identified as the strip segment to be adjusted.

[0007] The system includes a production data acquisition unit, a historical sample management unit, a strip segment registration unit, a release sensitive feature generation unit, a composite risk prediction unit, a straightening parameter correction unit, and a sample update unit.

[0008] The production data acquisition unit uses the coil number as the association identifier for material data. It reads the steel grade, thickness, and width of the cold-hardened strip from the production management system, and the yield strength, tensile strength, and modulus of elasticity of the corresponding steel coil from the inspection database. It also obtains the inlet shape data from the shape detection device on the inlet side of the straightening roll system and the outlet shape data from the shape detection device on the outlet side of the straightening roll system. Furthermore, it obtains the inlet tension, outlet tension, roll gap setting value of each adjustable straightening roll, bending roll reduction, and strip running speed from the straightener control system.

[0009] The strip section registration unit calculates the cumulative displacement of the strip based on the strip running speed in each sampling period, and establishes the strip length coordinates based on the cumulative displacement. Based on the installation distance between the inlet shape detection position, the straightening roller system position, and the outlet shape detection position, the time difference between the same reference strip section passing through each position in sequence is determined, and the inlet shape data, straightening process data, and outlet shape data are associated with the same strip length coordinates according to the time difference to form a reference strip section straightening sample.

[0010] The release sensitive feature generation unit generates plate shape change features, plate shape distribution difference features, tension change features, and straightening effect features based on the straightening sample of the reference strip segment. The plate shape change features, plate shape distribution difference features, tension change features, and straightening effect features are then combined into the release sensitive features of the reference strip segment according to the predetermined field positions. The current release sensitive features and historical release sensitive features use the same field names, number of fields, and calculation methods.

[0011] The strip shape variation characteristics include: the difference between the inlet strip shape deviation and the outlet strip shape deviation under the same strip length coordinate and the same width detection position; and the change in outlet strip shape deviation between two adjacent sampling positions along the strip length direction.

[0012] The plate shape distribution difference characteristics include the average plate shape deviation of the edge detection point group and the average plate shape deviation of the middle detection point group in the export plate shape data, as well as the difference between the average plate shape deviation of the edge detection point group and the average plate shape deviation of the middle detection point group; the edge detection point group consists of plate shape detection points located in the edge regions on both sides of the strip width direction, and the middle detection point group consists of plate shape detection points located in the middle region of the strip width direction; the width boundaries of the edge region and the middle region are determined according to the effective detection width of the plate shape detection device and the corresponding product's plate shape inspection procedure.

[0013] The tension variation characteristics include the difference between the inlet tension and the outlet tension, the difference between the maximum and minimum inlet tension in the reference strip section, and the difference between the maximum and minimum outlet tension in the reference strip section.

[0014] The straightening action features include the inlet tension strength ratio, the outlet tension strength ratio, the reduction thickness ratio, and the roll gap shape change sequence. The inlet tension strength ratio is the ratio of the inlet tension to the product of the strip cross-sectional area and yield strength; the outlet tension strength ratio is the ratio of the outlet tension to the product of the strip cross-sectional area and yield strength; the strip cross-sectional area is the product of the strip thickness and the strip width; the reduction thickness ratio is the ratio of the reduction amount of the bending roll to the strip thickness; the roll gap shape change sequence consists of the change in the roll gap setting value of each adjustable straightening roll arranged according to the strip length coordinate and the change in the outlet shape deviation.

[0015] The historical sample management unit stores historical benchmark strip straightening samples, historical release sensitive features generated according to the above fields and calculation methods, historical plate shape defect annotation data, and historical strip post-straightening re-inspection data.

[0016] Historical strip shape defect labeling data is generated by comparing the longitudinal warp, transverse warp, edge waviness amplitude, and center waviness amplitude obtained from historical strip steel section export inspections with the allowable deviation range specified in the corresponding product specification documents or strip shape inspection procedures. A longitudinal warp defect label is generated when the longitudinal warp exceeds the allowable deviation range; a transverse warp defect label is generated when the transverse warp exceeds the allowable deviation range; an edge waviness defect label is generated when the edge waviness amplitude exceeds the allowable deviation range; and a center waviness defect label is generated when the center waviness amplitude exceeds the allowable deviation range.

[0017] The re-inspection data after straightening includes the shape test data obtained after the historical strip steel section has undergone shearing, standing, or transshipment; the allowable deviation range of the export shape is determined according to the export shape control requirements in the product specification document; the allowable deviation range of the shape after shearing, the allowable deviation range of the shape after standing, and the allowable deviation range of the shape after transshipment are determined according to the shape inspection procedure adopted for the corresponding product status.

[0018] When the exit shape data of a historical strip segment is within the allowable deviation range of the exit shape, and at least one of the re-inspection shape data corresponding to the actual shearing, static or transshipment state experienced by the historical strip segment exceeds the corresponding allowable deviation range of the re-inspection shape, the historical strip segment is marked as a re-inspection positive sample.

[0019] When the export shape data of a historical strip segment is within the allowable deviation range of the export shape, and all the re-inspection shape data actually obtained for that historical strip segment are within the corresponding allowable deviation range of the re-inspection shape, the historical strip segment is marked as a re-inspection negative sample.

[0020] The positive and negative labels of the repeated warping samples are associated and stored with the historical release sensitive features corresponding to the same steel coil number and the same strip length coordinate, respectively.

[0021] The composite risk prediction unit is equipped with a multi-output neural network model; the multi-output neural network model includes a shared feature extraction layer, a longitudinal warping risk output layer, a transverse warping risk output layer, a side wave risk output layer, a middle wave risk output layer, and a complex warping risk output layer.

[0022] The shared feature extraction layer receives historical release-sensitive features or release-sensitive features of the reference strip segment, and extracts the correlation data between plate shape change features, plate shape distribution difference features, tension change features and straightening action features.

[0023] The longitudinal warping risk output layer is trained using historical longitudinal warping defect labels, the lateral warping risk output layer is trained using historical lateral warping defect labels, the edge wave risk output layer is trained using historical edge wave defect labels, the mid-wave risk output layer is trained using historical mid-wave defect labels, and the complex warping risk output layer is trained using complex warping positive sample labels and complex warping negative sample labels.

[0024] The multi-output neural network model receives the release sensitivity features of the reference strip segment and outputs longitudinal warping risk value, lateral warping risk value, edge wave risk value, middle wave risk value and re-warping risk value with values ​​ranging from zero to one. The risk value reflects the probability of occurrence of the corresponding plate shape defect or subsequent re-warping state of the reference strip segment under the current material properties, specification data and straightening conditions.

[0025] The risk thresholds for longitudinal warping, lateral warping, edge wave, mid-wave, and re-warping are determined based on historical validation samples. When determining the risk thresholds, multiple candidate thresholds are set in the range of zero to one. The model classification results corresponding to each candidate threshold are compared with the actual defect labels of the historical validation samples, and the candidate threshold with the lowest classification error rate is determined as the corresponding risk threshold.

[0026] The straightening parameter correction unit compares each risk value of the reference strip segment with its corresponding risk threshold. When at least one risk value exceeds the corresponding risk threshold, it calculates the difference between each excess risk value and its corresponding risk threshold, and determines the defect type corresponding to the risk value with the largest difference as the target defect type. When none of the risk values ​​exceed the corresponding risk threshold, it maintains the original straightening parameter setting value of the strip segment to be adjusted.

[0027] The straightening parameter correction unit sets a parameter mapping table; each record in the parameter mapping table includes the defect type, adjustable straightening parameter, parameter adjustment direction, single adjustment amount, and re-inspection confirmation result; the parameter mapping table is established based on the unit process card and historical parameter adjustment records of the straightening production line, and retains historical parameter adjustment records where the re-inspection result of the adjusted plate shape meets the corresponding allowable deviation range of the plate shape and the adjustment process does not trigger the control limit of the straightening machine.

[0028] Adjustable straightening parameters include inlet tension setting value, outlet tension setting value, roll gap setting value of each adjustable straightening roll, and bending roll reduction amount. After determining the target defect type, the straightening parameter correction unit reads the adjustable straightening parameters, parameter adjustment direction, and single adjustment amount corresponding to the target defect type from the parameter mapping table, and combines the current straightening parameter setting value of the strip segment to be adjusted with the single adjustment amount to obtain the candidate straightening parameter setting value.

[0029] The straightening parameter correction unit is equipped with a safety verification module. The safety verification module compares the candidate straightening parameter setpoint with the upper and lower control limits of the corresponding parameter in the straightening machine control system. When the candidate straightening parameter setpoint is between the corresponding upper and lower control limits, the candidate straightening parameter setpoint is retained. When the candidate straightening parameter setpoint exceeds the upper or lower control limit, it is restricted to the control limit closest to the candidate straightening parameter setpoint. When the corresponding straightening parameter is in a locked state or does not have the conditions for online adjustment, the adjustment of that parameter is canceled.

[0030] The straightening parameter settings, after safety verification, are output to the straightening machine control system or operating terminal and applied to the strip section to be adjusted.

[0031] After the strip segment to be adjusted completes straightening and exit inspection, it is identified as the new reference strip segment. The production data acquisition unit acquires the material property data, specification data, inlet shape data, straightening process data, and outlet shape data of the strip segment. The strip segment registration unit associates the data with the same strip length coordinates to form an updated strip segment straightening sample. The release sensitive feature generation unit generates updated release sensitive features according to the same field names, number of fields, and calculation method as the historical release sensitive features.

[0032] After the strip section has actually undergone shearing, settling, or transshipment, updated warp sample labels are generated based on the obtained re-inspection strip shape data, and updated strip shape defect labeling data are generated based on the longitudinal warp amount, transverse warp amount, edge wave amplitude, and center wave amplitude obtained from the exit inspection. The updated release sensitive features, updated strip shape defect labeling data, and updated warp sample labels are stored in association with the coil number and strip length coordinates, and used as incremental training samples for the multi-output neural network model.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. This invention uses the coordinates of the strip material as an index to register the entry strip shape data, straightening process data, and exit strip shape data, so that data obtained from different acquisition positions correspond to the same actual strip segment, reducing data mismatch caused by changes in the spacing of detection devices, strip speed, and straightening parameter switching; and constructs release-sensitive features based on the changes in strip shape before and after straightening, the strip shape distribution in the width direction, tension changes, and the state of straightening action of the same strip segment, so that the strip shape risk assessment is based on production data with actual material position correspondence.

[0035] 2. This invention marks historical strip steel sections whose exit shape is within the allowable deviation range but whose shape exceeds the corresponding allowable deviation range after shearing, settling, or transshipment as positive samples of re-warping. The re-warping sample labels are associated with the release sensitivity features of the corresponding historical strip steel sections, enabling the multi-output neural network model to simultaneously evaluate longitudinal warping, transverse warping, edge waviness, center waviness, and re-warping risk. As a result, the strip shape evaluation is no longer limited to the instantaneous state at the straightening exit, but can also reflect the strip shape stability after the tension constraint, roll system constraint, or curling constraint of the strip steel is released, providing a basis for identifying strip steel sections that pass the exit inspection but subsequently experience re-warping.

[0036] 3. This invention determines the target defect type based on the risk values ​​of each strip segment and their excess amount relative to the risk threshold. It generates candidate straightening parameter settings by combining the unit process card and historical parameter adjustment records confirmed by strip shape re-inspection. Safety verification is performed under the control limits of the straightener, parameter locking state, and online adjustment conditions, so that the risk assessment results can be transformed into straightening parameters applicable to the adjusted strip segment that has not yet entered the straightening roller system. After the adjusted strip segment completes straightening and re-inspection, its production data, release sensitive features, and defect labels continue to form incremental training samples, so that the model data keeps corresponding to the actual operating state of the production line. Attached Figure Description

[0037] Figure 1 This is the overall flowchart of the cold-hardened strip straightening plate shape defect prediction and warping risk assessment system of the present invention;

[0038] Figure 2 This is a flowchart of the process for registering the coordinates of the strip steel material and generating the sensitive release feature according to the present invention;

[0039] Figure 3 This is a flowchart of the composite risk prediction and straightening parameter closed-loop correction process of the present invention.

[0040] Figure 4 This is a schematic diagram illustrating the working principle of the cold-hardened strip straightening plate shape defect prediction and warping risk assessment system of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] like Figures 1 to 4 As shown, this embodiment provides a system for predicting shape defects and assessing warping risk in cold-hardened strip straightening. The system is applied to a cold-hardened strip recoiling production line equipped with a multi-roll straightener. For independent multi-roll straightening production lines with the same equipment layout and data acquisition conditions, the data processing method described in this embodiment can also be used.

[0043] Along the strip's running direction, the production line is sequentially equipped with an inlet shape detection device, a multi-roll straightener, and an outlet shape detection device. The multi-roll straightener includes an inlet tension control mechanism, an outlet tension control mechanism, multiple straightening rolls, a roll gap adjustment mechanism, and a bending roll pressing mechanism. The inlet shape detection device obtains the inlet shape data of the strip before it enters the multi-roll straightener, and the outlet shape detection device obtains the outlet shape data of the strip after it leaves the multi-roll straightener.

[0044] The system includes a production data acquisition unit, a historical sample management unit, a strip segment registration unit, a release sensitive feature generation unit, a composite risk prediction unit, a straightening parameter correction unit, and a sample update module.

[0045] The production data acquisition unit, strip registration unit, release sensitive feature generation unit, composite risk prediction unit, and straightening parameter correction unit are deployed in the production line industrial computer. The historical sample management unit is deployed in the industrial database server. The multi-output neural network model can be trained in the model training server, and the validated model parameters, field configuration files, standardized parameters, and risk thresholds are published to the production line industrial computer.

[0046] The production line industrial computer interacts with the production management system, quality inspection database, inlet strip shape detection device, outlet strip shape detection device, and multi-roll straightener control system via an industrial communication network. The production line industrial computer includes at least a processor, memory, data communication interface, and program storage medium. The memory stores programs used for performing strip segment registration, feature generation, risk prediction, and parameter verification.

[0047] In this manual, L-shaped warpage refers to longitudinal warpage of the strip along the rolling direction; C-shaped warpage refers to transverse warpage of the strip along the width direction; edge waviness refers to periodic undulations in the edge area of ​​the strip along the width direction; center waviness refers to periodic undulations in the middle area of ​​the strip along the width direction; and recurved warpage refers to a state where the strip shape at the straightening exit is within the allowable deviation range of the exit shape, but after subsequent shearing, settling, or transshipment, the shape exceeds the allowable deviation range of the corresponding re-inspection shape.

[0048] To avoid adjusting the same strip that has already been straightened after obtaining the exit strip shape test results, this embodiment defines the strip that has passed through the multi-roll straightener and completed the exit strip shape test as the evaluation strip, and the strip that is located after the evaluation strip, has not yet entered the multi-roll straightener and can be adjusted as the adjustment strip.

[0049] The composite risk prediction unit generates various risk values ​​based on the evaluated strip segment, and the straightening parameter correction unit generates straightening parameter settings for adjusting the strip segment based on these risk values. The evaluated strip segment and the adjusted strip segment belong to the same steel coil, and there are no changes in steel type, thickness, or width between them.

[0050] When the strip section is close to the end of the coil and there are no adjustment strip sections that have not yet entered the multi-roll straightener, the system does not send parameter adjustment instructions to the current coil. Instead, it sends the risk value, material properties, specification data, and the straightening parameter settings that have been safety-checked to the operation terminal, and records them as process references for subsequent similar coils.

[0051] Example 2: Production Data Acquisition and Data Validation Processing

[0052] The production data acquisition unit includes a material data interface, a plate shape data interface, and a straightening process data interface.

[0053] The material data interface uses the steel coil number as the associated identifier to read the steel type, strip thickness, and strip width from the production management system, and the yield strength, tensile strength, and elastic modulus from the quality inspection database.

[0054] Yield strength is used to calculate the inlet tension strength ratio and outlet tension strength ratio. Steel grade, thickness, and width are used to select the appropriate model version and retrieve parameter mapping records. Tensile strength and modulus of elasticity are used to determine whether the current steel coil is within the applicable range of material properties for the selected model version.

[0055] The model configuration file records the steel grade, thickness range, width range, yield strength range, tensile strength range, and elastic modulus range corresponding to the model. When any material property data of the current steel coil exceeds the corresponding model configuration range, the system can continue to display the plate shape data and risk analysis results, but will not automatically send parameter settings to the multi-roll straightener control system. Instead, it will output a prompt message indicating that the material properties exceed the applicable range of the model.

[0056] When there are material property test results from multiple sampling locations for the same steel coil, the different material property data are associated with the corresponding strip length range according to the sampling locations in the test record. When the test record only provides material property data for the steel coil level, the same set of material property data is used within the range of that steel coil.

[0057] The plate shape data interface reads inlet and outlet plate shape data from the inlet and outlet plate shape detection devices, respectively. Both inlet and outlet plate shape data include at least the coil number, sampling time, detection position in the width direction, plate shape deviation value, and data validity identifier.

[0058] The inlet and outlet plate shape data must represent the same physical quantity of plate shape and use the same unit of measurement. If the data types or units of measurement output by the inlet and outlet plate shape detection devices are different, they should be converted into the uniform plate shape deviation specified by the production line according to the calibration documents of the respective detection devices before subsequent difference calculations are performed.

[0059] For example, if both the inlet and outlet plate shape detection devices output plate shape units, the plate shape units should be used directly as the unified unit of measurement. If one detection device outputs the difference in elongation in the length direction and the other outputs plate shape units, the two should be converted to the same plate shape representation based on the calibration relationship of the detection devices. Data that have not undergone unit conversion should not be directly subtracted.

[0060] When the number or location of width detection points on the inlet and outlet strip shape detection devices are inconsistent, a unified width coordinate system is established within the effective detection width of the strip. This unified width coordinate system includes mm width detection positions, arranged in order from the operating side edge to the transmission side edge.

[0061] The inlet and outlet plate shape data are interpolated to a uniform width coordinate system. For uniform width positions located between adjacent original detection points, linear interpolation is used; for width positions exceeding the effective detection range of the original detection device, extrapolation is not used, but the width position is marked as invalid.

[0062] The straightening process data interface reads the inlet tension, outlet tension, roll gap setting value of each adjustable straightening roll, bending roll reduction, strip running speed, upper control limit of parameters, lower control limit of parameters, parameter lock status, online adjustment permission status and equipment alarm status from the multi-roll straightener control system.

[0063] The production data acquisition unit checks the data according to the steel coil number, sampling time, equipment operating status, and data validity identifier.

[0064] A valid strip straightening sample will not be formed for the corresponding strip length range when any of the following conditions occur: the coil number cannot be matched; the inlet or outlet strip shape data is invalid; the strip running speed is missing; the inlet and outlet strip shape data cannot be converted into a unified physical quantity; the strip is in the state of threading, reversing, or running in reverse; or the multi-roll straightener is under maintenance, manually jogging, or abnormally stopped.

[0065] For continuous quantities such as inlet tension, outlet tension, and strip running speed, linear interpolation is used to fill in the gaps when valid data exist at both ends of the continuous missing interval and the missing length does not exceed the maximum allowable interpolation length specified in the production line configuration file. When the missing length exceeds the maximum allowable interpolation length, the corresponding strip length interval is marked as invalid, and no model prediction or model training is performed.

[0066] Example 3: Establishment and Data Registration of Strip Material Coordinates

[0067] The inlet strip shape detection device, the multi-roll straightener, and the outlet strip shape detection device are located at different positions on the production line. The inlet strip shape data, straightening process data, and outlet strip shape data obtained at the same sampling time correspond to different actual strip material positions, and therefore cannot be directly combined according to the sampling time.

[0068] The strip section registration unit uses the head of the steel coil as the starting point of the material coordinates and calculates the cumulative running length of the strip based on the running speed of the strip.

[0069] Cumulative running length, number The cumulative running length of the strip corresponding to each sampling time is determined by the following formula:

[0070]

[0071] In the formula, For the first The cumulative running length of the strip at each sampling time, in meters; The cumulative running length of the strip steel at the previous sampling time is expressed in meters. and These are the strip running speeds at the previous and current sampling times, respectively, in meters per second. The time interval between two adjacent sampling times is expressed in seconds.

[0072] The product of speed and time interval in the formula is in meters, which is consistent with the unit of the cumulative running length of the strip.

[0073] Establish equipment coordinates at a fixed reference position on the production line. The installation coordinates of the inlet shape detection position, the straightening action position of the multi-roll straightener, and the outlet shape detection position relative to the fixed reference position are denoted as follows: , and ;

[0074] No. Each sampling time at the device location The coordinates of the strip material corresponding to the obtained data are determined by the following formula:

[0075]

[0076] In the formula, For the first Each sampling time at the device location The obtained data corresponds to the coordinates of the strip steel material, in meters; For the first The cumulative running length of the strip at each sampling time, in meters; For equipment location Installation distance relative to a fixed reference position, in meters; These respectively indicate the inlet plate shape detection position, the straightening action position of the multi-roll straightener, or the outlet plate shape detection position.

[0077] The strip section registration unit resamples the inlet plate shape data, straightening process data, and outlet plate shape data according to the strip material coordinates, so that the data corresponding to the same material coordinates form a registration data point.

[0078] When the coordinates of the material to be registered are located between two adjacent original material coordinates, the continuous data is interpolated according to the following formula:

[0079]

[0080] In the formula, Material coordinates The corresponding interpolated data; and To and Two adjacent original material coordinates, and satisfying ; and Material coordinates and The corresponding raw data.

[0081] For the control settings such as the roll gap setting and the bending roll reduction, which are maintained in segments, the control system is associated with the strip material coordinates according to the actual effective range of the control system, and the data on both sides of the setting switching point are not linearly interpolated.

[0082] In this embodiment, a continuous 5-meter strip steel is divided into one strip steel segment, and resampling is performed at a material coordinate interval of 0.1 meters. Each strip steel segment includes 50 sampling positions in the length direction. The 5-meter strip steel segment length and the 0.1-meter resampling interval are one implementation configuration, which can be adjusted according to the minimum evaluation length specified in the strip shape inspection procedure and the sampling capacity of the production line. However, the same strip steel segment length and resampling interval should be used for model training and online prediction.

[0083] To determine the adjustable strip segment that can accept parameter adjustments, the strip segment registration unit calculates the earliest adjustable material coordinates based on the control system response time.

[0084]

[0085] In the formula, The earliest material coordinates that can accept adjustments to the current parameters; The cumulative running length of the strip at the time of control response completion predicted based on the current strip running speed or the speed setpoint of the control system; The current moment; The response time of the multi-roll straightener control system from receiving new parameters to the actual effect of the parameters; These are the installation coordinates for the straightening action position of the multi-roller straightener.

[0086] Material coordinates less than The strip steel section is not used as an adjustment strip steel section; the material coordinates are not less than Strips that have not yet entered the multi-roller straightener and belong to the same steel coil as the evaluation strip can be identified as adjustment strips.

[0087] Construction of releasing sensitive features

[0088] The release sensitive feature generation unit generates plate shape change features, plate shape distribution difference features, tension change features, and straightening effect features based on the registered strip straightening sample.

[0089] The release-sensitive feature does not directly measure or calculate the residual stress inside the strip. Instead, it combines the changes in strip shape before and after straightening, the strip shape distribution in the width direction, the tension state, and the straightening action state of the same strip segment to characterize the risk conditions of strip shape changes after the tension constraint, roll system constraint, or curling constraint of the strip is released.

[0090] Let one evaluation strip include Each length direction sampling position and The detection positions are located in the width direction. The inlet plate shape deviation and outlet plate shape deviation are denoted as follows:

[0091]

[0092] In the formula, This represents the sampling position number along the length of the strip. This refers to the detection position number in the width direction of the strip. For the first The length position and the first The inlet plate shape deviation corresponding to each width position; The outlet plate shape deviation corresponds to the same material location; the inlet plate shape deviation and the outlet plate shape deviation use the same unit of measurement.

[0093] The change in plate shape before and after straightening for the same material location is:

[0094]

[0095] In the formula, For the first The length position and the first The change in bridge straightness at each width position is measured in the same unit as the deviation of the inlet and outlet slab shapes.

[0096] The change in strip shape at adjacent exit positions along the length of the strip is:

[0097]

[0098] In the formula, For the first The change in the outlet plate shape between the current length position and the previous length position. Its unit of measurement is the same as the export plate shape deviation.

[0099] The shape variation characteristics are evaluated from all within the strip segment. and They are arranged in a fixed order according to the sampling positions in the length direction and the detection positions in the width direction.

[0100] The differences in plate shape distribution distinguish the left edge region, the right edge region, and the middle region; the detection point sets for the left edge, the right edge, and the middle region are denoted as follows: , and .

[0101] The boundaries of each region are preferentially defined according to the product shape inspection procedure or the region configuration file of the shape inspection device. When the corresponding document does not explicitly define the region division, in one implementation, the inspection points on the left side of the effective inspection width are assigned to the left edge inspection point set, the inspection points on the right side of the effective inspection width are assigned to the right edge inspection point set, and the remaining 60% of the inspection points are assigned to the middle inspection point set. This division result is stored in the model configuration file and remains unchanged within the same model version.

[0102] No. The average deviation of the left side plate shape at each length position is:

[0103]

[0104] In the formula, For the first The average deviation of the left side plate shape at each length position; The number of detection points in the detection point set on the left side; the average plate shape deviation on the right side. Mean deviation of the middle plate shape The same method was used to determine it.

[0105] The characteristics of the plate shape distribution differences include the mean plate shape deviation of the left side, the mean plate shape deviation of the right side, the mean plate shape deviation of the middle section, the difference between the mean plate shape deviations of the left and right sides, and the difference between the average of the mean plate shape deviations of the two sides and the mean plate shape deviation of the middle section.

[0106] The difference between the mean deviations of the left and right side plate shapes is:

[0107]

[0108] The deviation in plate shape between the two sides and the middle is:

[0109]

[0110] In the formula, Used to preserve information about the asymmetrical shape of the left and right sides; This data is used to represent the difference in fabric shape between the edge and center regions; the above data uses the same unit of measurement as the export fabric shape deviation.

[0111] The tension variation characteristics include the difference between the inlet tension and the outlet tension under the same material coordinates, the difference between the maximum and minimum inlet tension within the strip segment, the difference between the maximum and minimum outlet tension, and the variation sequence of the inlet tension and outlet tension along the strip length.

[0112] The characteristics of the straightening action include the inlet tension strength ratio, the outlet tension strength ratio, the reduction thickness ratio, and the roll gap shape change sequence.

[0113] The inlet tension strength ratio is:

[0114]

[0115] The outlet tension strength ratio is:

[0116]

[0117] In the formula, Inlet tension, measured in Newtons (N); The tension at the outlet is expressed in Newtons (N). The width of the strip; For strip thickness; This represents the yield strength of the strip steel.

[0118] When using units commonly used in the production site, and The unit is millimeters. The unit is Newtons per square millimeter. The unit is Newton; the inlet tension strength ratio and the outlet tension strength ratio are both dimensionless; millimeters, meters, Pascals and Newtons per square millimeter must not be mixed in the same calculation.

[0119] When the tension detection device outputs the tension per unit width, the tension per unit width is first multiplied by the strip width to convert it into total tension, and then the tension strength ratio is calculated.

[0120] For the A bending roller, with a thickness reduction ratio of:

[0121]

[0122] In the formula, For the first The reduction thickness ratio corresponding to each bending roller; For the first The amount of pressure applied by each bending roller relative to the zero position calibrated by the equipment; For strip thickness; and Since the same length unit is used, the reduction thickness ratio is a dimensionless quantity.

[0123] The zero point and positive direction of the reduction amount of the equipment calibration are determined by the multi-roller straightener control system and recorded in the model configuration file; the positive direction of the reduction amount remains consistent within the same model version.

[0124] Assume a multi-roller straightening machine has The first adjustable straightening roller, the... The adjustable straightening roller is in the first The roll gap setting value corresponding to each length position is... The change in the roll gap setpoint is:

[0125]

[0126] In the formula, For the first The change in the roll gap setting value between two adjacent length positions of an adjustable straightening roller is measured in the same unit as the roll gap setting value.

[0127] No. The data on the variation of the roll gap shape at each length position are as follows:

[0128]

[0129] In the formula, For the first Data on the variation of roll gap shape at each length position; The number of straightening rollers is adjustable;

[0130] This represents the number of effective detection points in the width direction.

[0131] The change in roll gap setting and the change in exit plate shape are not directly added together; after each field is standardized, they are input into the multi-output neural network model according to the fixed field positions.

[0132] For the For a continuous numeric field, the standardized value is determined according to the following formula:

[0133]

[0134] In the formula, For the first The original values ​​of each field; For the training set The average of each field; For the training set Standard deviation of each field; This is the standardized dimensionless value.

[0135] When the standard deviation of a certain field is 0, it means that the field has not changed in the training samples. The field is removed from the input field of the current model version and no division by 0 is performed.

[0136] The field order, field mean, field standard deviation, width detection location, and region division results obtained during training are stored together with the model version; during online prediction, the standardized parameters of the corresponding model version are used, and the mean and standard deviation are not recalculated using the current evaluation strip segment.

[0137] Establishment of historical samples and recurrence labels

[0138] The historical sample management unit uses the steel coil number and the coordinate range of the strip material as the sample association identifier, and stores historical strip section straightening samples, historical release sensitive features, historical plate shape defect annotation data, and post-straightening re-inspection data.

[0139] Historical plate shape defect labeling data is read from the quality inspection database, and the online output results of the inlet plate shape detection device or the outlet plate shape detection device are not directly equated with all plate shape defect labels.

[0140] The L-shaped warpage, C-shaped warpage, edge wave amplitude, and center wave amplitude are obtained by the testing station, support condition, testing length, and testing method specified in the product sheet shape inspection procedure.

[0141] When the L-shaped warpage exceeds the allowable deviation range, an L-shaped warpage defect label is generated; when the C-shaped warpage exceeds the allowable deviation range, a C-shaped warpage defect label is generated; when the edge wave amplitude exceeds the allowable deviation range, an edge wave defect label is generated; when the mid-wave amplitude exceeds the allowable deviation range, a mid-wave defect label is generated.

[0142] When a defect exists, the corresponding label is set to 1; when a defect does not exist, the corresponding label is set to 0. If no defect detection is performed or the detection result is invalid, the corresponding label is marked as a missing label and is not used as 0 in training.

[0143] The allowable deviation range of the exported sheet shape is determined according to the product specification document or the export sheet shape control procedure; the allowable deviation range of the sheet shape after shearing, standing, and transshipment is determined according to the sheet shape inspection procedure of the corresponding product state, and the allowable deviation range of the exported sheet shape is not directly used as the judgment standard for all re-inspection states.

[0144] The re-inspection after settling should be carried out in accordance with the settling time and support method specified in the plate shape inspection procedure; the model configuration file stores the settling time, placement direction and support method; data obtained without following the specified settling conditions shall not be used to generate the re-inspection label after settling.

[0145] Post-transfer re-inspection should be carried out at the final inspection station of the specified transfer route; the model configuration file stores the corresponding transfer route identifier and the final inspection station identifier; data for which the transfer route or inspection station cannot be confirmed will not be used to generate post-transfer re-inspection labels.

[0146] When historical strip steel is sheared into individual sheets, the sheet number is associated with the material coordinate range in the original steel coil based on the shearing plan, the material coordinates at the start of shearing, and the material coordinates at the end of shearing. The re-inspection results after shearing are traced back to the corresponding historical strip steel segment based on the sheet number.

[0147] When historical strip steel is kept in coil state for static or transshipment, the relationship between the re-inspection record and the original steel coil material coordinate range is established based on the coil number, steel coil number and production batch number; the coil-level re-inspection results that cannot be located to the specific material coordinate range are only used for steel coil-level statistics and are not used as strip segment-level re-warping training labels.

[0148] When the re-inspection record clearly states that the abnormal plate shape was caused by collision, hoisting compression, abnormal stacking support, or other external mechanical damage, the record shall not be used to form a re-warped sample label.

[0149] When the exit shape data of a historical strip segment is within the allowable deviation range of the exit shape, and at least one re-inspection shape data corresponding to the actual completed shearing, static or transshipment state of the historical strip segment exceeds the corresponding allowable deviation range of the re-inspection shape, the historical strip segment is marked as a re-curved positive sample, and the re-curved sample label is 1.

[0150] When the export shape data of a historical strip segment is within the allowable deviation range of the export shape, and all the re-inspection shape data actually obtained for that historical strip segment are within the corresponding allowable deviation range of the re-inspection shape, the historical strip segment is marked as a negative sample with a re-curved shape label of 0.

[0151] Shearing, standing, or transfer states that have not been actually completed are not included in the positive and negative sample judgment; historical strip steel sections with only export plate shape data and no post-straightening re-inspection data are not used to train the warping risk output layer.

[0152] Training samples are formed by associating historical release sensitive features, L-shaped warping defect labels, C-shaped warping defect labels, edge wave defect labels, middle wave defect labels, and complex warping sample labels with steel coil number and material coordinate range.

[0153] Construction and training of multi-output neural network models

[0154] The composite risk prediction unit is equipped with a multi-output neural network model; the model includes an input processing layer, a shared feature extraction layer, an L-shaped warp risk output layer, a C-shaped warp risk output layer, a side wave risk output layer, a middle wave risk output layer, and a complex warp risk output layer.

[0155] The input to the multi-output neural network model is the release-sensitive features, including plate shape change features, plate shape distribution difference features, tension change features, and straightening effect features.

[0156] Steel grade, tensile strength, and modulus of elasticity are used for model version selection, model applicability verification, and parameter mapping record retrieval, and are not used as independent model input fields; yield strength, thickness, and width participate in the construction of straightening action features through inlet tension strength ratio, outlet tension strength ratio, and compression thickness ratio.

[0157] In this embodiment, the evaluation strip length is 5 meters, the length direction resampling interval is 0.1 meters, forming 50 length direction sampling positions; the inlet plate shape data and outlet plate shape data are unified to 20 width direction detection positions.

[0158] The strip shape change sequence, tension change sequence, and roll gap strip shape change sequence are used as sequence inputs; the mean strip shape distribution, the difference between the maximum and minimum tension values, the inlet tension strength ratio, the outlet tension strength ratio, and the reduction thickness ratio within the strip segment are used as fixed-length numerical inputs.

[0159] The sequence input passes through two one-dimensional convolutional layers in sequence; the first one-dimensional convolutional layer has 32 convolutional channels and a kernel length of 3 sampling positions in the length direction; the second one-dimensional convolutional layer has 64 convolutional channels and a kernel length of 3 sampling positions in the length direction; a linear rectified activation function is set after each one-dimensional convolutional layer, and a global average pooling layer is set after the second one-dimensional convolutional layer.

[0160] The fixed-length numerical input passes through two fully connected layers with 64 nodes and 32 nodes respectively, with a linear rectified activation function set after each fully connected layer.

[0161] The features obtained after global average pooling of the sequence input are concatenated with the features obtained after the fixed-length numerical input is processed through a fully connected layer, and then input into a shared feature extraction layer with 64 nodes.

[0162] The output of the shared feature extraction layer is connected to five risk output branches. Each risk output branch includes a fully connected layer with 16 nodes and an output node. The output node uses a logical function to transform the output value to the range of 0 to 1.

[0163] The output value is the standardized risk value of the corresponding defect, which is used to compare with the corresponding risk threshold and is not directly used as the actual defect probability without calibration.

[0164] The L-shaped warp risk output layer is trained using L-shaped warp defect labels, the C-shaped warp risk output layer is trained using C-shaped warp defect labels, the edge wave risk output layer is trained using edge wave defect labels, the middle wave risk output layer is trained using middle wave defect labels, and the complex warp risk output layer is trained using complex warp sample labels.

[0165] Since not every training sample has all five valid labels, a valid label is assigned to each task for each sample during training; the total loss of the multi-output neural network model is determined according to the following formula:

[0166]

[0167] In the formula, This represents the total loss of the model; This represents one of the five risk output tasks; For the first The weight of each risk output task in the total loss; This represents the number of samples in the current training batch. For the first The sample at the th The label validity identifier for each risk output task is set to 1 if there is a valid label and 0 if there is no valid label. The first in the current training batch The number of samples with valid labels for each risk output task; For the first The sample at the th The actual label for each risk output task is either 0 or 1. This corresponds to the risk value output by the model. and The first The positive and negative sample weights for each risk output task.

[0168] when At that time, the current training batch does not count the first... To mitigate the loss of each risky output task, avoid dividing by zero.

[0169] The weights of positive and negative samples are determined based on the number of positive and negative samples for the corresponding task in the training set. When the number of positive samples is less than the number of negative samples, the weight of positive samples is increased to reduce the bias caused by the imbalance of sample numbers.

[0170] Historical samples are divided according to steel coils. Different strips of the same steel coil are only included in one set of the training set, validation set, or test set. Adjacent strips of the same steel coil are not simultaneously assigned to different sets.

[0171] In this embodiment, 70% of the steel coils are assigned to the training set, 15% to the validation set, and 15% to the test set according to the number of steel coils. The division is made so that the distribution of each product group and each defect label in the three sets is comparable.

[0172] The model was trained using the adaptive moment estimation optimization method, with an initial learning rate of 0.001, a batch size of 64 training samples, and a maximum training epoch of 300. Training was stopped when the total loss of the validation set did not decrease for 10 consecutive training epochs, and the model parameters corresponding to the lowest total loss of the validation set were retained.

[0173] The number of convolutional channels, the number of fully connected layer nodes, the learning rate, and the number of training epochs mentioned above are a set of feasible parameters used in this embodiment and do not constitute a limitation on other feasible network configurations.

[0174] Determination of risk threshold

[0175] The risk thresholds for L-shaped warping, C-shaped warping, edge wave risk, mid-wave risk, and recurring warping are determined based on the validation set.

[0176] For the Each risk output task arranges the corresponding risk values ​​in the validation set in ascending order, and uses the midpoint between two adjacent different risk values ​​as the candidate risk threshold.

[0177] No. Each risk output task at the candidate risk threshold The weighted classification error is:

[0178]

[0179] In the formula, For the first Each risk output task at the candidate risk threshold Weighted classification error rate; This represents the number of missed samples where the actual label is 1 but the risk value does not exceed the candidate risk threshold. This represents the number of misclassified samples where the actual label is 0 but the risk value exceeds the candidate risk threshold. Weights that were missed during the evaluation; This is a misjudgment of weights.

[0180] The weights for missed judgments and wrong judgments are pre-configured based on the product quality control documents. When production control places more emphasis on avoiding missed judgments, the weight for missed judgments is greater than the weight for wrong judgments.

[0181] The candidate risk threshold that minimizes the weighted classification error is determined as the risk threshold for the c-th risk output task. When multiple candidate risk thresholds correspond to the same minimum weighted classification error, the selection is made according to the rule of prioritizing missed judgments or wrong judgments as specified in the product quality control document.

[0182] The risk threshold is released along with the model version and is not changed temporarily based on individual steel coils during online prediction.

[0183] Establishment of parameter mapping table

[0184] The parameter mapping table is established based on the process card of the multi-roller straightener unit and the historical parameter adjustment records confirmed by the plate shape inspection results.

[0185] The parameter mapping table is indexed according to steel type, thickness range, width range, target defect type and current parameter range; each parameter mapping record includes at least the adjustable straightening parameter, parameter adjustment direction, single adjustment amount, parameter setting value before adjustment, parameter setting value after adjustment, export plate shape inspection result after adjustment and re-inspection result after straightening.

[0186] Only historical parameter adjustment records that meet the following conditions are considered valid parameter mapping records: the steel type, thickness range, and width range are consistent before and after the adjustment; no strip breakage, slippage, equipment alarm, or parameter exceeding limits occurred during the adjustment process; the output plate shape after adjustment is within the allowable deviation range of the output plate shape; and when straightening re-inspection data has been obtained, the re-inspected plate shape after straightening is within the allowable deviation range of the corresponding re-inspection plate shape.

[0187] Historical parameter adjustment records for which no post-straightening re-inspection data has been obtained can be used to form parameter mapping records corresponding to L-shaped warp, C-shaped warp, edge waves, and center waves, but not to form parameter mapping records corresponding to complex warp.

[0188] The parameter mapping table does not pre-specify that all production lines use the same parameter adjustment direction; the adjustment direction of inlet tension, outlet tension, roll gap setting value and bending roll reduction amount shall be based on the unit process card of the corresponding multi-roll straightener and the historical records confirmed by re-inspection.

[0189] When there are multiple valid historical records with the same adjustment direction under the same index condition, the median of the corresponding single adjustment amount can be taken as the single adjustment amount in the current parameter mapping record; when there are conflicting parameter adjustment directions in the historical records under the same index condition, automatic parameter mapping records are not generated, and corresponding records are established after confirmation by process personnel.

[0190] When there is no valid record in the parameter mapping table that matches the current steel type, thickness range, width range, and target defect type, the system only outputs the target defect type and the corresponding risk value, and does not automatically generate straightening parameter settings.

[0191] Online risk prediction and parameter adjustment

[0192] During the production process, after an evaluation strip completes the exit shape detection, the production data acquisition unit checks the data integrity and model applicability of that evaluation strip.

[0193] When the data is valid and the material properties are within the applicable range of the current model version, the strip registration unit forms the evaluation strip straightening sample, the release sensitive feature generation unit forms the release sensitive feature, and the multi-output neural network model outputs L-shaped warping risk value, C-shaped warping risk value, edge wave risk value, middle wave risk value and re-warping risk value.

[0194] The straightening parameter correction unit compares the five risk values ​​with their corresponding risk thresholds. When none of the risk values ​​exceed their corresponding risk thresholds, no parameter adjustment is generated, and the original straightening parameter settings are used for the adjusted strip segment.

[0195] When at least one risk value exceeds the corresponding risk threshold, the excess amount between each excess risk value and the corresponding risk threshold is calculated; when the excess amount of one risk value is significantly greater than the excess amounts of other risk values, the defect type corresponding to that risk value is determined as the target defect type.

[0196] When the excess amounts of two or more risk values ​​are equal after being taken to the decimal places specified in the model configuration file, the straightening parameter setting value is not automatically generated. Instead, the parallel defect types, corresponding risk values, and available parameter mapping records are sent to the operation terminal.

[0197] After determining the target defect type, the straightening parameter correction unit reads the adjustable straightening parameters, parameter adjustment direction, and single adjustment amount from the parameter mapping table.

[0198] The candidate straightening parameter settings are determined according to the following formula:

[0199]

[0200] In the formula, Set values ​​for candidate straightening parameters; To adjust the original straightening parameter settings of the strip section; This is the single adjustment amount in the parameter mapping table, and it takes a non-negative value. Adjust the direction of the parameter by taking +1 when increasing the parameter and -1 when decreasing the parameter.

[0201] , and The same unit of measurement is used; the inlet tension, outlet tension, roll gap setting value and bending roll reduction are calculated separately, and the values ​​are not added between different physical quantities.

[0202] The safety verification module reads the upper control limit, lower control limit, parameter lock status, online adjustment permission status, and equipment alarm status of each adjustable parameter from the multi-roller straightener control system.

[0203] If the candidate straightening parameter setting value is between the corresponding upper and lower control limits, and the corresponding parameter is in an unlocked state, online adjustment is allowed, and there are no related alarms in the equipment, the candidate straightening parameter setting value is retained.

[0204] When the candidate straightening parameter setting value is higher than the upper control limit, the candidate straightening parameter setting value is restricted to the upper control limit; when it is lower than the lower control limit, the candidate straightening parameter setting value is restricted to the lower control limit.

[0205] If the corresponding parameter is locked, online adjustment is not allowed, or the device has an alarm state that affects the execution of the parameter, the parameter adjustment item will be canceled and other parameters will not be automatically used as substitutes.

[0206] The straightening parameter settings after safety verification are sent to the multi-roll straightener control system only when the evaluation strip and the adjustment strip belong to the same steel coil, the material properties and specifications have not changed, and the adjustment strip has not yet entered the multi-roll straightener; otherwise, the straightening parameter settings after safety verification are sent to the operation terminal for the operator to use for the next applicable strip.

[0207] Example of calculation process

[0208] This embodiment is used to illustrate the coordinates of strip materials, tensile strength ratio, reduction thickness ratio, and risk handling process. The specific values ​​in this embodiment are not to be used as fixed control parameters for all production lines.

[0209] In one deployment configuration, the data sampling period is 0.1 seconds, and the cumulative running length of the strip at the previous sampling time is... The length is 100,000 meters, and the strip speed at the previous sampling time is... The current sampling time is 2.0 meters per second, representing the strip speed. It is 2.1 meters per second.

[0210] The cumulative running length of the strip at the current sampling time is:

[0211]

[0212] The installation distances of the inlet plate shape detection position, the straightening action position of the multi-roller straightener, and the outlet plate shape detection position relative to the fixed reference position are 2.0 meters, 5.0 meters, and 8.0 meters, respectively.

[0213] At the current sampling moment, the strip material coordinates corresponding to the inlet strip shape detection data are 98.205 meters, the strip material coordinates corresponding to the multi-roll straightener straightening process data are 95.205 meters, and the strip material coordinates corresponding to the outlet strip shape detection data are 92.205 meters.

[0214] The strip section registration unit does not directly combine the above three sets of current time data. Instead, it searches for the inlet plate shape data, straightening process data and outlet plate shape data when the material coordinates are all 92.205 meters in the historical sampling sequence, and then interpolates to form registration data points under the same material coordinates.

[0215] In one example of a tension strength ratio calculation, the strip width is 1000 mm, the strip thickness is 0.8 mm, the strip yield strength is 650 N / mm², the inlet tension is 60000 N, and the outlet tension is 55000 N.

[0216] The cross-sectional area of ​​the strip is:

[0217]

[0218] The product of the strip cross-sectional area and the yield strength is:

[0219]

[0220] The inlet tension strength ratio is:

[0221]

[0222] The outlet tension strength ratio is:

[0223]

[0224] When the bending roller reduction is 0.12 mm, the reduction thickness ratio is:

[0225]

[0226] All three ratios mentioned above are dimensionless quantities.

[0227] In one risk assessment example, the L-shaped warping risk value, C-shaped warping risk value, edge wave risk value, middle wave risk value, and recurring warping risk value output by the multi-output neural network model are 0.41, 0.28, 0.63, 0.22, and 0.76, respectively, with corresponding risk thresholds of 0.55, 0.50, 0.60, 0.60, and 0.65.

[0228] The edge wave risk value exceeds the edge wave risk threshold by 0.03, the re-warping risk value exceeds the re-warping risk threshold by 0.11, and the remaining risk values ​​do not exceed their corresponding risk thresholds. Because the re-warping risk value exceeds the threshold by the largest amount, re-warping is identified as the target defect type.

[0229] The system then retrieves the parameter mapping table based on the current production line, steel type, thickness range, width range, and warping defect type; the parameter adjustment direction and single adjustment amount are provided by the parameter mapping records already confirmed for the production line, without directly calculating the tension, roll gap, or reduction adjustment direction based on the above risk values.

[0230] Incremental sample updates and model version management

[0231] After the straightening parameter settings, which have undergone safety verification, are applied to the adjusting strip section, the adjusting strip section becomes the new evaluation strip section when the exit strip shape test is completed.

[0232] The production data acquisition unit reacquires the material property data, specification data, inlet shape data, straightening process data, and outlet shape data of the strip steel segment; the strip steel segment registration unit generates updated strip steel segment straightening samples according to the same material coordinate generation method; and the sensitive feature release generation unit generates updated sensitive features according to the field names, number of fields, field order, and calculation method corresponding to the current model version.

[0233] After the strip steel section completes the plate shape quality inspection, corresponding defect labels are generated based on the L-shaped warpage, C-shaped warpage, edge wave amplitude, and center wave amplitude.

[0234] After the strip steel section has been actually sheared, left to stand or transferred and the re-inspection plate shape data has been obtained, a re-inspection sample label is formed based on the export plate shape status and the re-inspection plate shape status.

[0235] Update and release sensitive features, L-shaped warp defect labels, C-shaped warp defect labels, edge wave defect labels, middle wave defect labels, and complex warp sample labels, and store them in association according to steel coil number and material coordinate range.

[0236] When retest data is not yet available, the sample can be used for subsequent training of L-shaped warp, C-shaped warp, edge wave, or mid-wave risk output layers with valid labels, but not for training of complex warp risk output layers.

[0237] Incremental samples do not immediately change the online model after obtaining a single data point; the historical sample management unit summarizes incremental samples that have passed the validity check according to the predetermined model update cycle, re-divides the training set, validation set, and test set, and re-executes model training and risk threshold determination.

[0238] After the updated model has been checked on an independent test set, it is released as a new model version. The model version information includes at least the model number, input field configuration, standardized parameters, model parameters, risk threshold, applicable steel grade category, applicable thickness range, applicable width range, and release date.

[0239] The original model version and its corresponding configuration file are preserved, allowing historical risk output results to be traced based on the model version.

[0240] Production verification methods

[0241] To verify the applicability of the system on a specific production line, production batches with consistent steel grades, thickness ranges, width ranges, and downstream re-inspection conditions can be selected for offline verification or controlled online verification.

[0242] During offline verification, the evaluation strip data is input in the order of historical production occurrence, and data generated after the evaluation point is not allowed to be used as the current model input; the risk results output by the system are compared with the subsequent actual quality inspection results, and the number of missed judgments, misjudgments, and valid label samples for L-shaped warping, C-shaped warping, edge waves, middle waves, and compound warping are counted respectively.

[0243] The incidence of recurrence is determined by the following formula:

[0244]

[0245] In the formula, The re-inspection rate of qualified strip steel sections exported; The number of strip steel segments whose export shape is within the allowable deviation range of export shape, but whose subsequent re-inspection shape exceeds the allowable deviation range of re-inspection shape. This refers to the total number of strip steel sections whose export shape is within the allowable deviation range and for which valid re-inspection data has been obtained.

[0246] During online verification, the observation operation mode is first adopted, which only outputs risk values ​​and parameter adjustment suggestions and does not automatically write to the multi-roller straightener control system. After the process personnel confirm that the parameter mapping records and safety verification results are consistent with the actual control logic of the production line, automatic parameter writing is then enabled.

[0247] The online verification record should include at least the steel coil number, the material coordinates of the evaluated strip segment, the material coordinates of the adjusted strip segment, the model version, each risk value, the corresponding risk threshold, the target defect type, the parameter setting before adjustment, the parameter setting after adjustment, the export strip shape inspection results, and the re-inspection results after straightening.

[0248] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0249] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for predicting shape defects and assessing warping risk in cold-hardened strip straightening, characterized in that, For cold-hardened strip recoiling or multi-roll straightening production lines, including a production data acquisition unit, a historical sample management unit, a strip segment registration unit, a release sensitive feature generation unit, a composite risk prediction unit, and a straightening parameter correction unit; The production data acquisition unit acquires the inlet shape data, straightening process data, and outlet shape data of the evaluation strip that has completed export inspection; The strip segment registration unit uses the strip material coordinates as an index to register the data as an evaluation sample for strip straightening. The release sensitive feature generation unit generates release sensitive features from the sample, including changes in strip shape, differences in strip shape distribution, changes in tension, and the straightening effect. The historical sample management unit establishes historical samples based on historical release sensitive features, defect labeling data, and post-straightening re-inspection data, and marks historical strip steel sections whose export shape is qualified but whose shape exceeds the tolerance after shearing, standing, or transportation as re-inspection samples with positive warping. The composite risk prediction unit includes a multi-output neural network model trained from the historical samples, which takes the release sensitivity characteristics of the strip segment as input and outputs longitudinal warping, lateral warping, edge wave, middle wave and complex warping risk values. The straightening parameter correction unit determines the target defect type based on the threshold excess of each risk value, generates candidate straightening parameter settings for the adjustment strip section located after the evaluation strip section and not yet entering the straightening roll system within the same steel coil, and outputs them after verification according to the straightening machine control limits.

2. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 1, characterized in that: The production data acquisition unit also acquires material property data and specification data for evaluating the strip steel section; The material property data includes steel grade, yield strength, tensile strength and elastic modulus; the specification data includes strip thickness and strip width; and the straightening process data includes inlet tension, outlet tension, roll gap setting value of each adjustable straightening roll, reduction amount of each bending roll and strip running speed. The composite risk prediction unit matches the model version of the multi-output neural network model according to the steel grade, strip thickness and strip width, and judges the applicability of the model based on whether the yield strength, tensile strength and elastic modulus are within the value range of the training samples corresponding to the model version. If material property data exceeds the corresponding value range, the automatic output of straightening parameter setting values ​​to the straightening machine control system will stop.

3. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 2, characterized in that: The production data acquisition unit includes a material data interface, a plate shape data interface, and a straightening process data interface; The material data interface reads the steel grade, strip thickness and strip width from the production management system according to the steel coil number, and reads the yield strength, tensile strength and elastic modulus from the quality inspection database. The plate shape data interface reads the inlet plate shape data and the outlet plate shape data from the plate shape detection devices at the inlet and outlet sides of the straightening roller system, respectively. The straightening process data interface reads the straightening process data from the straightening machine control system; The strip shape data interface and the straightening process data interface also read the sampling time and data validity identifier. Strip length intervals with invalid data validity identifiers are not used to form evaluation strip straightening samples.

4. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 3, characterized in that: The strip section registration unit determines the cumulative running length of the strip based on the strip running speed in each sampling period, and determines the strip material coordinates of the corresponding data as the difference between the cumulative running length of the strip and the installation coordinates of each data acquisition position relative to the fixed reference position. The strip section registration unit resamples the inlet shape data, straightening process data, and outlet shape data according to the same strip material coordinates, associates the data corresponding to the same strip material coordinates as registration data points, and forms an evaluation strip section straightening sample by combining multiple registration data points that are continuous along the strip length direction. The continuously changing data is interpolated using the data corresponding to the coordinates of adjacent strip materials. The roll gap setting and bending roll reduction are associated with the strip material coordinates according to their actual effective range in the straightening machine control system.

5. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 4, characterized in that: The plate shape variation includes the difference between the exit plate shape deviation and the inlet plate shape deviation under the same strip material coordinates and the same width detection position, as well as the difference between the exit plate shape deviation under adjacent strip material coordinates; The plate shape distribution difference includes the average exit plate shape deviation of each of the left side detection point group, the right side detection point group, and the middle detection point group; the difference between the average exit plate shape deviation of the left side detection point group and the right side detection point group; and the difference between the average of the average exit plate shape deviation of the two side detection point groups and the average exit plate shape deviation of the middle detection point group. The tension variation includes the difference between the inlet tension and the outlet tension under the same strip material coordinates, as well as the difference between the maximum and minimum values ​​of the inlet tension and the outlet tension within the strip segment.

6. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 5, characterized in that: The left side inspection point group, right side inspection point group, and middle inspection point group are respectively composed of plate shape inspection points located in the left side region, right side region, and middle region in the width direction of the strip. The width boundary of each region is determined according to the plate shape inspection procedure of the corresponding product or the region configuration file of the plate shape inspection device. The straightening action includes the inlet tension strength ratio, the outlet tension strength ratio, the reduction thickness ratio, and the roll gap shape change sequence; The inlet tension strength ratio is the ratio of the inlet tension to the product of the strip cross-sectional area and yield strength; the outlet tension strength ratio is the ratio of the outlet tension to the product of the strip cross-sectional area and yield strength; and the strip cross-sectional area is the product of the strip thickness and the strip width. The reduction thickness ratio is the ratio of the reduction amount of each bending roller to the thickness of the strip. The roll gap shape change sequence consists of the roll gap setting value change of each adjustable straightening roll arranged according to the strip material coordinates and the corresponding exit shape deviation change.

7. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 6, characterized in that: The multi-output neural network model includes a shared feature extraction layer and five risk output layers corresponding to the five risk values ​​described in claim 1. The defect labeling data is formed by comparing the longitudinal warp, transverse warp, edge wave amplitude, and center wave amplitude of historical strip steel sections with the corresponding allowable deviation ranges. The longitudinal warp, transverse warp, edge wave amplitude, and center wave amplitude are obtained from the quality inspection database. Each risk output layer outputs a risk value in the range of 0 to 1; if any type of valid defect labeling data is missing for a historical strip segment, that historical strip segment will not participate in the loss calculation of the corresponding risk output layer.

8. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 7, characterized in that: The export plate shape being qualified means that the export plate shape data is within the allowable deviation range of the export plate shape specified in the product specification document. The re-inspection plate shape exceeding the tolerance means that the re-inspection plate shape data after shearing, standing, or transportation exceeds the allowable deviation range of the re-inspection plate shape for the corresponding product state. If the export strip shape of a historical strip is qualified and all the actual re-inspection strip shape data are within tolerance, the historical strip is marked as a negative sample. When the re-inspection record indicates that the abnormal plate shape was caused by collision, hoisting compression, or stacking support abnormality, the re-inspection record shall not be used to form a warping sample; historical strip segments for which no post-straightening re-inspection data has been obtained shall not participate in the training of the warping risk output layer. The positive and negative samples of repeated warping are associated and stored with the corresponding historical release sensitive features according to the steel coil number and the coordinate range of the strip material.

9. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 8, characterized in that: For any risk output layer, the midpoint between adjacent risk output values ​​of historical verification samples is taken as the candidate risk threshold, and the candidate risk threshold that minimizes the weighted sum of the number of missed samples and the number of misjudged samples is determined as the corresponding risk threshold. The threshold excess is the difference between the risk value and the corresponding risk threshold; When two or more risk values ​​have the same maximum threshold exceedance, candidate straightening parameter settings are not automatically generated. The straightening parameter correction unit generates candidate straightening parameter settings through a parameter mapping table. The parameter mapping table is indexed according to steel type, thickness range, width range, target defect type and current straightening parameter range, and includes adjustable straightening parameters, parameter adjustment direction and single adjustment amount. The parameter mapping table is established based on the unit process card of the straightening production line and the historical parameter adjustment records confirmed by the exit plate shape detection or the re-inspection after straightening. The adjustable straightening parameters include the inlet tension setting value, the outlet tension setting value, the roll gap setting value of each adjustable straightening roller, and the pressing amount of each bending roller.

10. The cold-hardened strip straightening plate shape defect prediction and warping risk assessment system according to claim 9, characterized in that: The straightening parameter correction unit includes a safety verification module, and the system also includes a sample update unit; The safety verification module combines the current straightening parameter setting value of the adjusted strip segment with the corresponding single adjustment amount according to the parameter adjustment direction to obtain the candidate straightening parameter setting value; When the candidate straightening parameter setting value is between the upper control limit and the lower control limit of the corresponding parameter, the candidate straightening parameter setting value is retained; when the candidate straightening parameter setting value exceeds the corresponding control limit, it is restricted to the corresponding control limit; when the corresponding parameter is in a locked state, online adjustment is not allowed, or the straightening machine has an alarm state that affects the execution of the parameter, the adjustment of the corresponding parameter is cancelled. After the strip section completes straightening and exit inspection, the sample update unit generates updated strip section straightening samples and updated release sensitive features. It also generates updated defect labeling data based on quality inspection data. After obtaining the re-inspection data after straightening, it generates updated warped sample labels. The updated release sensitive features, updated defect labeling data, and updated warped sample labels are associated and stored as incremental training samples.

Images