A method for controlling online surface quality detection in a hot rolling process of pickling plate

Abstract

The application discloses a pickling plate hot rolling process online surface quality detection control method, relates to the technical field of hot rolling surface detection, and comprises the following steps: based on benchmark synchronization recording, collecting multi-polarization state images, calculating a window pollution state, and performing imaging compensation, quality correction purification processing and gate screening to generate a usable multi-polarization state image group; performing pixel-level alignment and consistency checking on the usable multi-polarization state image group according to a mileage window, constructing a polarization physical feature layer, extracting a defect candidate area, identifying a defect type, and generating a defect list; jointly arbitrating the defect list and the window pollution state, mapping the defect list into online control instructions, simultaneously performing window cleaning and pickling plate hot rolling process line speed limitation, and generating a control execution record. The application realizes instant dynamic linkage of defect detection and production line control, and suppresses defect accumulation.

Description

Technical Field

[0001] This invention relates to the field of hot-rolled surface inspection technology, and in particular to an online surface quality inspection and control method for the hot rolling process of pickled steel plates. Background Technology

[0002] Driven by the high-quality development of the steel industry, the online surface quality inspection technology for hot-rolled pickled steel sheets has undergone a systematic evolution from basic imaging to intelligent integration. The introduction of multispectral imaging technology has enhanced the ability to extract spectral features of defects such as oxide scale, microcracks, and metal inclusions, while the embedding of deep learning algorithms has further optimized the accuracy of defect classification. In recent years, the collaborative innovation of polarization imaging technology and high-speed image sensors has enabled multi-dimensional characterization of surface microstructures. At the same time, the deepening of production line automation has promoted the organic integration of inspection systems and production control, supporting the standardized management and control of pickled steel sheet surface quality, and further promoting the steel manufacturing industry towards the goal of "zero defects," laying a technological foundation for the large-scale production of high-end steel sheets.

[0003] The existing technology system still has key areas for optimization in high-speed continuous production scenarios. First, the dynamic perception and compensation mechanism for window contamination is not yet perfect, resulting in significant fluctuations in image quality. Traditional methods rely on preset static thresholds or basic filtering algorithms (such as Gaussian smoothing), which make it difficult to quantify contamination parameters in real time, leading to a decrease in the signal-to-noise ratio of defect identification and a high false positive rate. Second, the collaborative linkage mechanism between the detection system and production line control is weak, lacking the ability to intervene dynamically based on detection results. It fails to map defect identification with key parameters in real time, resulting in long response delays for window cleaning or speed adjustment, which can easily lead to defect accumulation or production interruption, thus restricting the engineering practicality of online detection. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an online surface quality inspection and control method for the hot rolling process of pickled steel plates, which solves the problems of insufficient detection accuracy and disconnect between detection and control caused by window contamination.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an online surface quality inspection and control method for the hot rolling process of pickled steel sheets. The method includes: establishing a reference reflection reference region and a reference scattering reference region; connecting the mileage signal of the hot rolling process of pickled steel sheets and the production line timestamp signal; constructing a mileage window and binding it with the identity information of the pickled steel sheet coil to generate a reference synchronization record; based on the reference synchronization record, acquiring multi-polarization state images, calculating the window contamination state, and performing imaging compensation, quality correction and purification processing, and gating screening to generate a usable multi-polarization state image group; performing pixel-level alignment and consistency verification on the usable multi-polarization state image group according to the mileage window, constructing a polarization physical feature layer, extracting defect candidate regions, identifying defect types, and generating a defect list; jointly arbitrating the defect list and the window contamination state, and mapping it to online control commands, simultaneously executing window cleaning and limiting the line speed of the hot rolling process of pickled steel sheets to generate a control execution record; and performing online re-inspection of repeated windows in the mileage window using self-compensation acquisition and polarization identification based on the control execution record, marking the control effective mileage segment, and generating a quality inspection control record set.

[0007] As a preferred embodiment of the online surface quality detection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for establishing a reference reflection reference region and a reference scattering reference region, and connecting the mileage signal and production line timestamp signal of the hot rolling process of pickled steel plates, are as follows. Within the imaging field of view, a reference reflection reference area and a reference scattering reference area are fixed, and boundary positioning and pixel coordinate range locking are performed to generate a reference area locking record; Based on the baseline region locking record, the mileage signal of the hot rolling process of pickled plate and the production line timestamp signal are accessed and synchronously collected. At the same time, a corresponding relationship is established, a consistency check is performed, and a time sequence alignment record is generated.

[0008] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for generating the benchmark synchronous record are as follows: Based on the time-series aligned records, set the mileage window length and step parameters, divide the continuous intervals according to the mileage length, and fix the mileage window number and mileage window timestamp range to generate the mileage window definition record; The identification information of the pickled sheet roll is matched with the mileage window definition record by the machine entry timestamp, the mileage window number is located, and the record is aligned and encapsulated as a benchmark synchronization record in combination with the benchmark area locking record.

[0009] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for acquiring multi-polarization state images based on reference synchronous recording are as follows. Based on the baseline synchronization record, configure the trigger queue and solidify the polarization state switching order and initial parameter list to generate a multi-polarization state acquisition task sheet; According to the multi-polarization state acquisition task sheet, multi-polarization state images are acquired, and reference reflection reference area images and reference scattering reference area images are simultaneously captured to generate a multi-polarization state original image group.

[0010] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for generating a group of usable multi-polarization images are as follows: Based on the original multi-polarization state image set, the window contamination state is calculated, the initial parameter list is updated, imaging compensation and fogging suppression are performed, and a compensated multi-polarization state image set is generated. Dark field correction, distortion correction, motion blur suppression, and occlusion removal are performed on the compensated multi-polarization state image group, and gating screening is performed in combination with the window contamination status to generate a usable multi-polarization state image group.

[0011] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for performing pixel-level alignment and consistency verification on the available multi-polarization state image groups according to mileage windows to construct a polarization physical feature layer are as follows. Corner feature points are extracted from the available multi-polarization state image group according to the mileage window, and a correspondence is established. Pixel-level alignment and edge region overlap checks are performed to generate alignment lock records. Based on the alignment lock record, brightness drift check, noise level check, occlusion consistency check and unified dynamic range normalization are performed, and the effective pixel area mask is fixed to generate a consistency check input set. Based on the consistency check input set, multi-polarization state pixel response differential assembly and stability constraint fusion are performed to construct a polarization physical feature layer. Boundary smoothing and outlier removal are then performed to generate a polarization physical feature layer record.

[0012] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for generating the defect list are as follows: The polarization physical feature layer records are subjected to polarization stability threshold segmentation, connected component aggregation, defect candidate region merging and conflict resolution, and geometric morphology parameters are extracted to generate a set of defect candidate regions. The candidate defect regions are grouped and judged to identify the defect type, locate the defect mileage range, and generate a defect list.

[0013] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps of jointly arbitrating the defect list and the contamination status in the viewing window are as follows: Match and align the defect list with the window contamination status according to the mileage window number, write the missing marker, and generate a joint arbitration input package; Based on the mileage window number, perform defect priority sorting and window contamination status level determination on the joint arbitration input package, and conduct joint arbitration within the same mileage window to generate a control strategy selection record.

[0014] As a preferred embodiment of the online surface quality detection and control method for the hot rolling process of pickled steel plates described in this invention, the specific steps for generating the control execution record are as follows: The control strategy selection record is mapped to online control instructions, and amplitude limiting and instruction atomic splitting are performed to generate an online control instruction set; Based on the online control instruction set, the speed limit of the hot rolling process of window cleaning and pickling plate is executed, and the online control instructions are read back to generate control execution records.

[0015] As a preferred embodiment of the online surface quality inspection and control method for the hot rolling process of pickled steel plates described in this invention, the steps of performing online re-inspection of the repeated mileage window based on the control execution record with self-compensation acquisition and polarization identification, marking the control effective mileage segment, and generating a quality inspection and control record set are as follows: Based on the control execution record, establish a mileage window comparison relationship between the re-inspection mileage window and the control mileage window, solidify the re-inspection conditions, and generate an online re-inspection task comparison table; Based on the online re-inspection task comparison table, self-compensation acquisition and polarization recognition are performed to generate a re-inspection defect list; The list of defects to be re-inspected is compared window by window with the online re-inspection task comparison table. A double threshold comparison review is performed to identify and mark the control effective mileage segment and generate a quality inspection control record set.

[0016] The beneficial effects of this invention are as follows: by establishing a reference reflection and scattering benchmark area and connecting it to the mileage signal of the hot rolling process of the pickled plate to construct a mileage window and bind the identity of the coil, the image collected by the image sensor and the production line conditions are synchronized with high precision, which improves the reliability of the detection benchmark and reduces the false judgment rate; by jointly arbitrating and mapping the defect list and the window contamination status into control commands, the real-time dynamic linkage between defect detection and production line control is realized, which suppresses the accumulation of defects. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an online surface quality inspection and control method for the hot rolling process of pickled steel plates.

[0019] Figure 2 A flowchart for generating a usable multi-polarization state image group.

[0020] Figure 3 A flowchart for generating a defect list.

[0021] Figure 4 A flowchart for generating a quality inspection and control record set. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As an embodiment of the present invention, this embodiment provides an online surface quality inspection and control method for the hot rolling process of pickled steel plates, comprising the following steps: S1: Establish reference reflection reference area and reference scattering reference area, connect the mileage signal of the hot rolling process of pickled plate and the production line timestamp signal, construct the mileage window, and bind it with the identity information of pickled plate coil to generate a reference synchronization record; S1.1: Fix the reference reflection reference area and the reference scattering reference area within the imaging field of view, and perform boundary positioning and pixel coordinate range locking to generate a reference area locking record; Specifically, fixed positions of the reference reflection reference region and the reference scattering reference region are selected within the imaging field of view. The correspondence between the imaging field of view coordinate system and the pixel coordinate system is confirmed. The availability of the region is verified based on the visibility and unobstructed state of the reference reflection reference region and the reference scattering reference region within the imaging field of view. The boundaries of the reference reflection reference region and the reference scattering reference region are located separately. The boundary direction is determined by the continuity constraints of the edge brightness change band and the texture change band. The boundary points are sampled point by point along the boundary direction and the connectivity of the boundary points is checked to generate a closed boundary. The stability of the boundary expansion and contraction is compared with the closed boundary to lock the stable range of the boundary position in the pixel coordinate system (e.g., the boundary drift does not exceed 2 pixels). The boundary range is converted into a pixel coordinate range. The pixel coordinate range is locked and the upper and lower boundaries are fixed (e.g., the pixel coordinate range is [100,300]×[50,200]). A reference region locking record is generated.

[0026] It should be noted that the reference reflection reference area refers to a fixed area within the imaging field of view used to provide a stable reference for specular reflection intensity, while the reference scattering reference area refers to a fixed area within the imaging field of view used to provide a stable reference for diffuse reflection scattering intensity.

[0027] S1.2: Based on the baseline region locking record, the mileage signal of the hot rolling process of pickled plate and the production line timestamp signal are accessed and synchronously collected. At the same time, a corresponding relationship is established, a consistency check is performed, and a time sequence alignment record is generated. Specifically, the pixel coordinate ranges of the reference reflection reference area and the reference scattering reference area are read. The pickling plate hot rolling process mileage signal acquisition channel and the production line timestamp signal acquisition channel are connected to the same acquisition trigger queue and the acquisition trigger parameter set is solidified. The pickling plate hot rolling process mileage signal and the production line timestamp signal are synchronously acquired according to the acquisition trigger parameter set, and an acquisition sequence number and acquisition time mark are written for each acquisition (e.g., the acquisition period is 10 milliseconds). The pickling plate hot rolling process mileage signal and the production line timestamp signal obtained by synchronous acquisition are paired and bound according to the acquisition sequence number to establish the correspondence between the pickling plate hot rolling process mileage signal and the production line timestamp signal and write the correspondence entry. Consistency checks are performed on the correspondence entries. The gap is checked according to the continuity of the acquisition sequence number and a gap mark is written. The back jump is checked according to the incrementality of the production line timestamp signal and a back jump mark is written. The abnormal jump is checked according to the incremental continuity of the pickling plate hot rolling process mileage signal and a jump mark is written. A time sequence alignment record is generated.

[0028] It should be noted that the mileage signal of the hot rolling process of pickled steel sheet is used to characterize the change in the travel distance of the pickled steel sheet coil along the production line. After establishing a correspondence with the production line timestamp signal, it is used to divide the mileage window, thereby binding the multi-polarization image frame to the mileage position range of the coil.

[0029] S1.3: Based on the time-series aligned records, set the mileage window length and step parameters, divide the continuous intervals according to the mileage length, and fix the mileage window number and mileage window timestamp range to generate the mileage window definition record; Specifically, based on the effective continuous segments of the mileage signal from the hot rolling process of the pickled steel sheet, the start and end points of the available mileage intervals are extracted and the interval boundaries are registered. The mileage window length and step parameters are set and written into the mileage window parameter set (e.g., the mileage window length is 5 meters, and the step parameters specify the sliding update range of the mileage window in the direction of coil movement, so that adjacent mileage windows advance sequentially according to a fixed mileage increment and correspond to the production line timestamp range; an example is 0.5 meters). Using the start point of the available mileage interval as the dividing point, the start and end mileages of the window are generated incrementally according to the step parameters. The mileage signal from the hot rolling process of the pickled steel sheet is covered and checked according to the start and end mileages of the window, and gap markers are included. Write an exception flag to the window marked with a record, bounce flag, or jump flag; assign a mileage window number to each completed window and solidify the numbering order; extract the window start production line timestamp signal and window end production line timestamp signal from the corresponding relationship entries of the window start and end mileages according to the time sequence alignment record and solidify them into the mileage window timestamp range (referring to the start and end time interval defined by the window start production line timestamp signal and window end timestamp signal corresponding to a certain mileage window number in the mileage window definition record); and write the mileage window number, window start and end mileages, mileage window timestamp range, exception flag, and mileage window parameter set into the mileage window definition record.

[0030] S1.4: Match the entry timestamp of the pickled sheet roll material identity information with the mileage window definition record, locate the mileage window number, and combine it with the benchmark area lock record to align and encapsulate it into a benchmark synchronization record.

[0031] Furthermore, the machine entry timestamp of the pickled sheet roll is extracted and the time format is unified. The machine entry timestamp is matched with the mileage window timestamp range in the mileage window definition record. The corresponding mileage window number is located according to the mileage window timestamp range that the machine entry timestamp falls into. If the machine entry timestamp does not fall into any mileage window timestamp range, a matching failure flag is written. Based on the mileage window number obtained by positioning, the window start and end mileage, step parameters and abnormal flags are extracted from the mileage window definition record and bound to the pickled sheet roll identity information and written into the binding entry. Based on the reference area locking record, the pixel coordinate range of the reference reflection reference area and the reference scattering reference area is extracted. The binding entry is aligned and encapsulated with the pixel coordinate range of the reference reflection reference area and the pixel coordinate range of the reference scattering reference area to generate a reference synchronization record.

[0032] It should be noted that the pickled sheet roll identity information refers to a set of roll identification information (such as roll number, batch number, etc.) used to uniquely identify the roll and support binding with the machine entry timestamp and mileage window number.

[0033] S2: Based on the benchmark synchronous recording, acquire multi-polarization state images, calculate the window contamination status, and perform imaging compensation, quality correction and purification processing and gating screening to generate a usable multi-polarization state image group; S2.1: Based on the baseline synchronization record, configure the trigger queue and solidify the polarization state switching order and initial parameter list to generate a multi-polarization state acquisition task sheet; Specifically, based on the reference synchronization record, the mileage window number, mileage window timestamp range, and pixel coordinate range of the reference reflection reference area and the reference scattering reference area are extracted. The mileage window timestamp range is used as the trigger time basis to configure the trigger queue. The trigger queue includes trigger start conditions, trigger intervals, and readback verification conditions (e.g., the trigger interval is 5 milliseconds, and the trigger interval is for multi-polarization state image acquisition). The trigger queue assigns a trigger sequence number to each trigger and binds it to the corresponding mileage window number. According to the polarization state switching order, the polarization state is assigned to each trigger in the trigger queue. The polarization state switching relationship between two adjacent triggers is determined according to a fixed arrangement; the initial value parameter list is used for the initial value setting and state readback verification of polarization state switching. The initial value parameter list is used to complete the value loading before the trigger queue starts and to complete the readback comparison after the trigger is executed. The initial value parameter list is used to uniformly register the average brightness reference value, the scattering fog expansion amplitude reference value, the corresponding dimension mark, the upper and lower bounds of the value, the boundary clipping, and the increment / decrement update step size of each correction as traceable entries, so that there is a reproducible constraint basis when performing value correction based on the window pollution state, and a multi-polarization state acquisition task sheet is generated.

[0034] It should be noted that the trigger start condition refers to the constraint conditions that the trigger queue must meet to start executing the trigger sequence number generation and trigger interval timing. The trigger start condition is used to determine whether the trigger queue is allowed to enter the triggerable state (e.g., it is allowed to start when the entry timestamp falls within the mileage window timestamp range).

[0035] The readback verification condition refers to the constraint condition that the trigger queue must meet to compare the polarization state readback value with the corresponding value in the initial parameter list after the polarization state switching is executed. The readback verification condition is used to determine whether the trigger execution result allows the next trigger (e.g., if the readback polarization state is consistent with the specified polarization state, it passes).

[0036] S2.2: According to the multi-polarization state acquisition task sheet, acquire multi-polarization state images, and simultaneously capture the reference reflection reference area image and the reference scattering reference area image to generate a multi-polarization state original image group; Specifically, the trigger cycle is initiated according to the trigger start conditions, and the polarization state specified in the task sheet is switched upon each trigger arrival. The polarization state is read back and a consistency check is performed according to the readback verification conditions. Trigger numbers that fail the consistency check are marked with an anomaly flag (e.g., an anomaly flag is registered when the readback polarization state is inconsistent with the specified polarization state). Under the trigger number that passes the consistency check, the imaging field of view image is acquired as a multi-polarization state image, and a trigger number, polarization state flag, and acquisition time flag are added to the multi-polarization state image. Based on the pixel coordinate range of the reference reflection reference area and the reference scattering reference area recorded synchronously with the reference, the reference reflection reference area image and the reference scattering reference area image are simultaneously cropped from the multi-polarization state image. During the cropping process, the pixel coordinate range remains unchanged and the start and end coordinates of the cropping are retained. The multi-polarization state image, the reference reflection reference area image, and the reference scattering reference area image are grouped according to the trigger number and arranged according to the polarization state switching order to generate a multi-polarization state original image group.

[0037] S2.3: Based on the original multi-polarization state image group, calculate the window contamination state, update the initial parameter list, perform imaging compensation and fogging suppression, and generate a compensated multi-polarization state image group; Specifically, the original multi-polarization state image group is expanded according to the trigger sequence number. The reference reflection reference region image and reference scattering reference region image corresponding to each trigger sequence number are extracted one by one. The offset of the average brightness of the reference reflection reference region image relative to the average brightness reference value in the initial parameter list is normalized to obtain the average brightness offset normalization amount. The difference between the scattering haze expansion amplitude of the reference scattering reference region image and the scattering haze expansion amplitude reference value in the initial parameter list is checked and normalized to obtain the scattering haze expansion amplitude normalization amount. The consistency of differences between adjacent trigger sequences is checked and converted to obtain the difference consistency inverse characterization item. The scaling factor is used to perform an exponential mapping on a combination of the normalized average brightness offset, the normalized scattering fog expansion, and the inverse characterization term of difference consistency to obtain the window contamination state. Based on the window contamination state, the initial parameter list is corrected, including increment / decrement adjustments and boundary clipping while preserving the adjustment trajectory. Imaging compensation is performed on the original multi-polarization image set based on the updated initial parameter list, with the imaging compensation performing pull-back correction for brightness drift, vignetting, and polarization state response deviation. Fog suppression is performed on the imaging compensation record, with the suppression intensity selected based on the window contamination state while preserving edge details, generating a compensated multi-polarization image set.

[0038] The formula for calculating the window contamination status is:

[0039] in, Indicates the window is contaminated. Indicates the scaling factor. Indicates the trigger sequence number. Indicates the first The normalized amount of the average brightness shift of the reference reflection reference area image under the second trigger. Indicates the first The normalized amount of the scattering haze spread amplitude of the reference scattering reference region image under secondary triggering. Indicates the first The reverse characterization of the consistency of adjacent trigger sequence numbers under each trigger.

[0040] It should be noted that the window contamination state is compressed into a state value between zero and one by the contamination index through monotonic mapping, while maintaining the correspondence that the higher the contamination index, the higher the contamination state. The contamination index is obtained by combining the average brightness offset normalization, the scattering fog expansion amplitude normalization, and the trigger sequence number difference consistency inverse characterization term according to the scaling factor.

[0041] The scaling factor refers to the normalization amount used to adjust the average brightness shift of the reference reflection reference area image to the window contamination state, the normalization amount of the scattering fog expansion amplitude of the reference scattering reference area image, and the sensitivity of the reverse characterization term to the consistency of the difference between adjacent trigger numbers.

[0042] S2.4: Perform dark field correction, distortion correction, motion blur suppression and occlusion removal on the compensated multi-polarization state image group, and perform gating screening in combination with the window contamination status to generate a usable multi-polarization state image group.

[0043] Furthermore, the compensated multi-polarization state image group is expanded frame by frame according to the trigger sequence while maintaining the polarization state switching order. Dark field correction is performed on each frame, subtracting the dark level bias from the pixel response and truncating and pulling back negative responses (e.g., the truncation lower limit is Example 0). Distortion correction is performed on each frame, which performs coordinate inverse mapping of radial curvature and tangential offset and completes pixel resampling. Neighborhood interpolation is performed on resampling hole regions and interpolation marks are written. Motion blur suppression is performed on each frame, which estimates the blur direction. The blur length estimation constrains the sharpening intensity and preserves the edge gradient; occlusion culling is performed on each frame. Occlusion culling extracts the occlusion mask by determining the occlusion region and sets the occluded pixels as invalid pixels. Gating screening is performed in combination with the window contamination status. Gating screening can mark the image frames corresponding to the trigger sequence number of the window contamination status as unusable and remove them from the compensated multi-polarization state image group according to the level of window contamination status. The remaining image frames are retained as usable multi-polarization state image groups to generate usable multi-polarization state image groups.

[0044] It should be noted that the occlusion region determination is achieved by locating connected regions in the compensated multi-polarization state image group that exhibit long-term pixel response saturation, sudden brightness drop across areas, near-disappearance of texture contrast, or continuous zero edge gradient. These regions are then compared with the synchronous changes in the reference reflection reference region image and the reference scattering reference region image. Connected regions that meet the occlusion characteristics are identified as occlusion regions.

[0045] S3: Perform pixel-level alignment and consistency verification on the available multi-polarization state image group according to the mileage window, construct a polarization physical feature layer, extract defect candidate regions, identify defect types, and generate a defect list; S3.1: Extract corner feature points from the available multi-polarization state image group according to the mileage window, establish the correspondence, perform pixel-level alignment and edge region overlap check, and generate alignment lock record; Specifically, frames falling within the same mileage window timestamp range are selected from the available multi-polarization state image group according to the mileage window number and arranged in the order of polarization state switching. A reference frame within the same mileage window is selected as the alignment reference frame. Corner feature points are extracted from both the reference frame and the frame to be aligned. Candidate points for intersection structures with obvious changes along the brightness gradient are extracted from the corner feature points, and dense points are removed by minimum spacing constraints. Pixel coordinates and intensity scores are added to the corner feature points. A correspondence is established between the reference frame and the frame to be aligned. Matching is completed by sorting the neighborhood similarity of the corner feature point pixel coordinates and intensity scores. One-to-many and many-to-one matches are eliminated, and one-to-one matching pairs are retained. Pixel displacement is obtained, and an alignment transformation is generated. The alignment transformation is used to map the frame to be aligned to the coordinate system of the reference frame and perform pixel-level resampling. An edge region overlap check is performed on the mapped frame to be aligned and the reference frame. The edge region overlap check uses the boundary of the effective pixel area mask as the edge region and measures the overlap area ratio (e.g., overlap not less than 0.90 in the example). The pixel-level alignment results, the number of matching pairs, pixel displacement statistics, and edge region overlap check results are summarized to generate an alignment lock record.

[0046] It should be noted that the minimum spacing constraint refers to setting a lower limit on the pixel coordinate distance between any two corner feature points during the corner feature point extraction process, in order to exclude overly dense corner feature points in the same area and maintain a uniform spatial distribution of corner feature points (e.g., the pixel coordinate distance is not less than 5 pixels in the example).

[0047] S3.2: Based on the alignment lock record, perform brightness drift check, noise level check, occlusion consistency check and unified dynamic range normalization, fix the effective pixel area mask, and generate consistency check input set; Specifically, within the same mileage window, the available multi-polarization state image groups are aligned pixel-level to achieve coordinate unification while maintaining the polarization state switching order; a brightness drift check is performed within the effective pixel area, using the average brightness difference between frames as a reference to a drift threshold, and frames exceeding the drift threshold are marked with a drift mark; a noise level check is performed within the effective pixel area, using the grayscale fluctuation amplitude of flat areas as a reference to a noise threshold, and frames exceeding the noise threshold are marked with a noise mark; an occlusion consistency check is performed, verifying the spatial overlap of the occlusion masks for each frame, and marking frames with insufficient overlap as occlusion inconsistencies (e.g., overlap less than 0.95 in the example); unified dynamic range normalization is performed on the frames that pass the check, using the target grayscale range constraint to achieve linear stretching and truncation; the effective pixel area mask is determined based on the intersection of the edge region overlap check and the occlusion mask, and the effective pixel area mask is kept unchanged; the frames that pass the check, the drift mark, the noise mark, the occlusion inconsistency mark, and the effective pixel area mask are summarized to generate a consistency verification input set.

[0048] It should be noted that the drift threshold is defined based on the historical statistical distribution of the average brightness difference between adjacent frames within the same mileage window in the consistency check input set. It is used to determine abnormal drift in brightness drift checks, with an example range of 0.02 to 0.08 normalized brightness difference.

[0049] The noise threshold is defined based on the historical statistical distribution of grayscale fluctuation amplitude in the flat area of ​​the same mileage window within the consistency check input set, and is used to determine abnormal noise in the noise level check (for example, the noise threshold is taken as the grayscale standard deviation of Examples 1 to 5).

[0050] S3.3: Based on the consistency check input set, perform differential assembly of multi-polarization state pixel response and stability constraint fusion to construct a polarization physical feature layer, and perform boundary smoothing and outlier removal to generate a polarization physical feature layer record; Specifically, based on the consistency check input set, frames that have passed the check are expanded according to the mileage window number, while maintaining the polarization state switching order. The pixel positions participating in the processing are limited according to the effective pixel area mask. For the same pixel position, multi-polarization state grayscale responses are extracted according to the polarization state switching order and multi-polarization state pixel response differential assembly is performed. The multi-polarization state pixel response differential assembly generates differential entries by differentiating adjacent polarization state grayscale responses and cross-polarization state grayscale responses, and the differential entries are bound to pixel coordinates and polarization state markers. The differential entries are then fused according to stability constraints. Using inter-frame difference consistency and neighborhood difference continuity as constraints, weighted fusion is performed on difference entries with significant fluctuations and deviations, and neighborhood padding is performed on missing difference entries. The fused difference entries are backfilled according to pixel coordinates to form a polarization physical feature layer. Boundary smoothing is performed on the polarization physical feature layer, and neighborhood weighted transition is used at the mask boundary of the effective pixel area to suppress breakage. Abnormal isolated point removal is performed on the polarization physical feature layer. Abnormal isolated point removal locates isolated pixel connected points that are significantly inconsistent with the neighborhood difference and performs neighborhood replacement processing to generate polarization physical feature layer records.

[0051] S3.4: Perform polarization stability threshold segmentation, connected component aggregation, defect candidate region merging and conflict resolution on the polarization physical feature layer records, and extract geometric morphology parameters to generate a set of defect candidate regions; Furthermore, the polarization physical feature layer records are expanded according to the mileage window number, and the processing range is limited based on the effective pixel area mask. Threshold segmentation is performed on the polarization physical feature layer records based on the polarization stability threshold. Pixels that meet the polarization stability threshold condition are marked as candidate pixels, and pixels that do not meet the polarization stability threshold condition are set as background. Connected component aggregation is performed on the candidate pixels. Connected component aggregation generates connected components according to the pixel adjacency relationship and assigns a connected component number to each connected component. Defect candidate region merging is performed on the connected components. Defect candidate region merging checks the boundary distance of adjacent connected components and merges the connected components and updates the connected component number when the boundary distance meets the merging condition. For defects... Conflict resolution is performed on the candidate regions. Conflict resolution determines the assignment of overlapping parts of the defect candidate regions (referring to the allocation determination process of overlapping pixels in the defect candidate regions during conflict resolution. By comparing the polarization stability in the neighborhood of overlapping pixels with the consistency of the geometric morphology parameters of the defect candidate regions, overlapping pixels are assigned to defect candidate regions with higher consistency). Overlapping pixels are also assigned to defect candidate regions with larger areas or higher polarization stability. Geometric morphology parameters are extracted one by one for the defect candidate regions that have completed conflict resolution. Geometric morphology parameters include area, perimeter, bounding rectangle, principal axis direction, and shape compactness. The defect candidate region boundaries, connected component numbers, and geometric morphology parameters are summarized to generate a set of defect candidate regions.

[0052] It should be noted that the polarization stability threshold is defined based on the statistical distribution of the differences in the responses of multiple polarization states of pixels within the same mileage window in the polarization physical feature layer record, in terms of inter-frame consistency and neighborhood continuity. It is used to determine whether the polarization response of a pixel position satisfies a stable state (for example, the polarization stability threshold is taken as 0.70 to 0.90).

[0053] Pixel adjacency refers to the adjacency determination method within the effective area mask of a pixel based on the adjacent pixel coordinates. It is used to determine whether candidate pixels are connected in the vertical, horizontal, left-right, or diagonal directions and to form a connected region accordingly.

[0054] Merging conditions refer to the constraints on the similarity of the boundary distance between adjacent connected regions, the pixel state of the interval region, and the geometric morphological parameters during the merging process of defect candidate regions. They are used to determine whether adjacent connected regions are allowed to be merged into the same defect candidate region (e.g., the boundary distance does not exceed 3 pixels in Example and the background ratio of the interval region does not exceed 0.20 in Example).

[0055] S3.5: Group and distinguish the candidate defect regions, identify the defect type, locate the defect mileage range, and generate a defect list.

[0056] Furthermore, the defect candidate region set is grouped by mileage window number while maintaining the traceability of connected component numbers. For each defect candidate region, geometric morphological parameters and boundary coordinates are extracted. Grouping discrimination is performed based on the value range of geometric morphological parameters and boundary shape features, and defect candidate regions that meet the same discrimination conditions are grouped into the same group. For each group, the distribution of geometric morphological parameters and boundary orientation features of the defect candidate regions within the group are summarized, and each is compared with the defect type discrimination conditions to complete the defect type assignment. According to the mileage window definition, the start and end mileage of the window corresponding to the mileage window number is recorded. The minimum and maximum coordinates of the defect candidate region boundary in the mileage window direction are mapped to the defect mileage location interval. For defect candidate regions that cross the mileage window boundary, interval truncation and interval splicing are performed to maintain the continuity of the defect mileage location interval (for example, the defect mileage location interval is 12.3 meters to 12.8 meters). The mileage window number, connected component number, defect type, defect mileage location interval, and geometric morphological parameters are summarized to generate a defect list.

[0057] It should be noted that the discrimination criteria refer to the comparison between the value range of geometric morphological parameters and the boundary shape characteristics of the defect candidate region, which is used to divide and group the defect candidate region.

[0058] Defect type discrimination criteria refer to the comparison between the distribution of geometric morphological parameters and boundary orientation characteristics within a group to determine the defect type (such as holes, sticky roller marks, roller marks, and edge cracks).

[0059] S4: Jointly arbitrate the defect list and the window contamination status, and map them into online control instructions. At the same time, execute the window cleaning and pickling plate hot rolling process line speed limit, and generate control execution records. S4.1: Match and align the defect list with the window contamination status according to the mileage window number, write the missing marker, and generate a joint arbitration input package; Specifically, the defect list is expanded by mileage window number, and the mileage window number, defect type, and defect mileage location range corresponding to each defect list item are extracted to form a defect list index. The window contamination status is expanded by mileage window number, and the mileage window number and window contamination status value corresponding to each window contamination status are extracted to form a window contamination status index. The defect list index and the window contamination status index are matched and aligned item by item using the mileage window number as the matching key. When the match is successful, the defect type, defect mileage location range, and window contamination status under the same mileage window number are merged into the same alignment item. When the match fails, a missing mark is set for mileage window numbers that exist in the defect list index but not in the window contamination status index, and a missing mark is set for mileage window numbers that exist in the window contamination status index but not in the defect list index. The alignment items and missing marks are summarized and packaged to generate a joint arbitration input package.

[0060] S4.2: Based on the mileage window number, perform defect priority sorting and window contamination status level determination on the joint arbitration input package, and conduct joint arbitration within the same mileage window to generate a control strategy selection record; Specifically, the joint arbitration input packet is traversed by mileage window number. For each mileage window number, the defect type, defect mileage location range, geometric parameters, window contamination status, and missing marker are extracted. Multiple defect list entries under the same mileage window number are prioritized. The priority ranking is based on the priority correspondence between defect types and the correspondence between geometric parameter values, generating a sorting key. The results are then sorted from highest to lowest according to the sorting key. A window contamination status level determination is performed, comparing the window contamination status with the grading threshold and outputting a level marker (when the window contamination status value is less than the lowest grading threshold). Output the lowest level marker; output the intermediate level marker when the window contamination status value falls between two adjacent grade thresholds; output the highest level marker when the window contamination status value is greater than the highest grade threshold. Perform joint arbitration within the same mileage window number, compare the header defect entries of the sorting results with the level markers, determine the available arbitration items based on the missing markers, and select a control strategy. The control strategy selection includes a combination of window cleaning, pickling plate hot rolling process line speed limit, and window cleaning and pickling plate hot rolling process line speed limit. Summarize the mileage window number, header defect entries, level markers, missing markers, and control strategies to generate a control strategy selection record.

[0061] It should be noted that the priority mapping relationship refers to the set of correspondences between defect types and priorities, which is used as a sorting key to convert defect types into defect priority sorting.

[0062] The geometric morphology parameter value correspondence refers to the set of correspondences between the value range of geometric morphology parameters and the sorting bonus, which is used to convert geometric morphology parameters into sorting key components for defect priority sorting.

[0063] The grading threshold is determined based on the distribution quantiles of window contamination status in historical sampling. The example range is the normalized cutoff value of 0.20 to 0.80, because the change range of window contamination status below 0.20 and above 0.80 is prone to saturation and the grading distinction is insufficient.

[0064] S4.3: Map the control strategy selection record to online control instructions, and perform amplitude limiting and instruction atomic splitting to generate an online control instruction set; Furthermore, the control strategy selection record is expanded according to the mileage window number, and mapping is completed based on the correspondence between control strategies and controlled objects. Window cleaning is mapped to window cleaning instructions, and the speed limit of the hot rolling process of pickled steel plate is mapped to speed limit instructions. Window cleaning and the speed limit of the hot rolling process of pickled steel plate are mapped to window cleaning instructions and speed limit instructions generated concurrently. The instruction parameters of the window cleaning instructions and speed limit instructions are completed respectively. The instruction parameters determine the effective mileage segment based on the defect mileage location interval, the effective start and end time based on the mileage window timestamp range, and the execution intensity based on the level mark. Each instruction is assigned an instruction sequence number and a target mileage window number. For window cleaning instructions, a throttling process is performed, limiting the execution time and frequency to within an allowable range (e.g., execution time not exceeding 10 seconds in the example). For pickling plate hot rolling process line speed limit instructions, a throttling process is also performed, limiting the target speed and speed reduction to within an allowable range (e.g., target speed not lower than 0.6 times the baseline speed in the example). Instructions that have undergone throttling are then atomically split, breaking each instruction into a start instruction, a hold instruction, and a release instruction, while maintaining the instruction sequence number association. These are then aggregated to generate an online control instruction set.

[0065] S4.4: Based on the online control instruction set, execute the speed limit of the hot rolling process of window cleaning and pickling plate, and issue online control instructions for readback to generate control execution records.

[0066] Specifically, based on the online control instruction set, the start instruction, hold instruction, and release instruction are retrieved sequentially according to the instruction sequence number. The target mileage window number is checked to be consistent with the current mileage window number, and the mileage window timestamp range is checked to be within the effective period. Instructions that do not meet the effective period are marked as skipped. For window cleaning instructions, window cleaning is performed. The start instruction triggers the cleaning action and records the start time. The hold instruction maintains the cleaning action and records the hold interval. The release instruction stops the cleaning action and records the stop time. For speed limit instructions, the speed limit of the pickling plate hot rolling process line is executed. The start instruction adjusts the production line speed to the target speed and records the adjustment time. The hold instruction maintains the target speed and records the hold interval. The release instruction restores the production line speed and records the restore time. For each executed instruction, an online control instruction is issued for readback. The online control instruction is readback to retrieve the instruction sequence number, target mileage window number, instruction parameters, and execution status, and compares them item by item with the online control instruction set. If there is a discrepancy, the difference item is recorded. The online control instruction set, the readback online control instruction, the skip mark, and the difference item are summarized to generate a control execution record.

[0067] S5: Based on the control execution record, perform online re-inspection of the repeated windows of the mileage window with self-compensation acquisition and polarization identification, mark the mileage segment where the control is effective, and generate a quality inspection control record set.

[0068] S5.1: Based on the control execution record, establish a mileage window comparison relationship between the re-inspection mileage window and the reference mileage window, solidify the re-inspection conditions, and generate an online re-inspection task comparison table; Specifically, based on the control execution record, the execution status, start time, stop time, recovery time, skip marker, and difference entries of the window cleaning instruction and speed limit instruction are merged according to the target mileage window number. The target mileage window number with the execution status of completion and no skip marker is selected as the re-inspection mileage window, and the mileage window timestamp range and the window start and end mileage are used as the re-inspection boundary. Based on the mileage window definition record, the reference mileage window is located according to the mileage window number adjacency relationship. The adjacent mileage window number before or after the re-inspection mileage window is selected as the reference mileage window, and it is verified that there is no record of the window cleaning instruction and speed limit instruction execution status being completed in the control execution record. The reference mileage window that fails the verification is replaced with the next adjacent mileage window number (e.g., offset by 1 mileage window number). The re-inspection mileage window and the reference mileage window are bound into a mileage window comparison relationship, and the re-inspection conditions are determined as the control strategy corresponding to the re-inspection mileage window, the mileage window timestamp range, the window start and end mileage, and the difference entry verification requirements. An online re-inspection task comparison table is generated.

[0069] S5.2: Based on the online re-inspection task comparison table, perform self-compensation acquisition and polarization recognition to generate a re-inspection defect list; Specifically, self-compensation acquisition is performed within the timestamp range of the mileage window corresponding to the re-inspection mileage window. The self-compensation acquisition adjusts the trigger interval, polarization state switching order, and initial parameter list according to the re-inspection conditions, and performs value pullback based on the brightness shift and scattering spread amplitude of the reference reflection reference area image and the reference scattering reference area image, acquiring multi-polarization state images to obtain a re-inspection multi-polarization state image group; polarization recognition is performed on the re-inspection multi-polarization state image group, which sequentially completes pixel-level alignment and consistency verification, polarization physical feature layer recording construction, polarization stability threshold segmentation, connected component aggregation, defect candidate region merging and conflict resolution, geometric morphology parameter extraction, defect type assignment, and defect mileage location interval positioning; the same process is performed on the comparison mileage window to obtain comparison records, and a re-inspection defect list is generated.

[0070] S5.3: Compare the list of defects to be re-inspected with the online re-inspection task comparison table window by window, perform double threshold comparison verification, identify and mark the control effective mileage segment, and generate a quality inspection control record set.

[0071] Specifically, the list of defects to be re-inspected is expanded according to the re-inspection mileage window number, and the defect type, defect mileage location range, and geometric parameters (such as defect area parameters, defect perimeter parameters, and defect circumscribed rectangle size parameters) are extracted. Simultaneously, the reference mileage window number and re-inspection conditions are located from the online re-inspection task comparison table according to the re-inspection mileage window number. Window-by-window comparison entries are established according to the re-inspection mileage window number, containing the defect entries corresponding to the re-inspection mileage window number and the defect entries corresponding to the reference mileage window number. A double-threshold comparison review is performed on the window-by-window comparison entries. The double-threshold comparison review verifies the consistency of defect types and checks several... The differences in morphological parameters are compared (the differences in defect area, defect perimeter, and the size of the defect's circumscribed rectangle are calculated item by item, and the double threshold comparison is deemed successful when any parameter simultaneously meets the change range threshold and the minimum change threshold); when the double threshold comparison is successful, the defect mileage location interval in the re-inspection mileage window is marked as the control effective mileage segment; when the double threshold comparison is unsuccessful, the re-inspection mileage window is marked as ineffective; the re-inspection mileage window number, the comparison mileage window number, the control effective mileage segment marking, the defect type verification conclusion, and the geometric morphological parameter difference judgment are summarized to generate a quality inspection control record set.

[0072] It should be noted that the dual threshold is defined based on the historical distribution of the differences in geometric morphological parameters between the re-inspection defect list and the control defect list under the same defect type. The dual threshold consists of a change magnitude threshold and a minimum change threshold, which are used to simultaneously constrain relative changes and absolute changes (for example, the change magnitude threshold is 0.10 to 0.30 and the minimum change threshold is 3 to 10 pixels).

[0073] In summary, this invention achieves high-precision synchronization between images collected by image sensors and production line conditions by: establishing a reference reflection and scattering benchmark area, connecting it to the mileage signal of the hot rolling process of pickled steel sheets, constructing a mileage window, and binding the coil identity; improving the reliability of the detection benchmark and reducing the false judgment rate; and by jointly arbitrating and mapping the defect list and the contamination status of the window into control commands, it realizes real-time dynamic linkage between defect detection and production line control, and suppresses defect accumulation.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online surface quality inspection and control in the hot rolling process of pickled steel sheets, characterized in that: include, Establish a reference reflection reference area and a reference scattering reference area, connect the mileage signal of the hot rolling process of pickled steel plate and the production line timestamp signal, construct a mileage window, and bind it with the identity information of the pickled steel plate coil to generate a reference synchronization record; Based on the baseline synchronous recording, multi-polarization state images are acquired, the contamination state of the window is calculated, and imaging compensation, quality correction and purification processing and gating screening are performed to generate a group of usable multi-polarization state images. Pixel-level alignment and consistency verification are performed on the available multi-polarization state image group according to the mileage window, a polarization physical feature layer is constructed, and candidate defect regions are extracted, defect types are identified, and a defect list is generated. The defect list and the contamination status of the window are jointly arbitrated and mapped into online control instructions. At the same time, the window cleaning and hot rolling process speed limits of the pickled plate are executed, and control execution records are generated. Based on the control execution records, the online re-inspection of the repeated windows of the mileage window is carried out with self-compensation acquisition and polarization identification, and the mileage segments in which the control is effective are marked, generating a quality inspection control record set.

2. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 1, characterized in that: The specific steps for establishing the reference reflection reference region and the reference scattering reference region, and connecting the pickling plate hot rolling process mileage signal and the production line timestamp signal are as follows. Within the imaging field of view, a reference reflection reference area and a reference scattering reference area are fixed, and boundary positioning and pixel coordinate range locking are performed to generate a reference area locking record; Based on the baseline region locking record, the mileage signal of the hot rolling process of pickled plate and the production line timestamp signal are accessed and synchronously collected. At the same time, a corresponding relationship is established, a consistency check is performed, and a time sequence alignment record is generated.

3. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 2, characterized in that: The specific steps for generating the benchmark synchronization record are as follows. Based on the time-series aligned records, set the mileage window length and step parameters, divide the continuous intervals according to the mileage length, and fix the mileage window number and mileage window timestamp range to generate the mileage window definition record; The identification information of the pickled sheet roll is matched with the mileage window definition record by the machine entry timestamp, the mileage window number is located, and the record is aligned and encapsulated as a benchmark synchronization record in combination with the benchmark area locking record.

4. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 3, characterized in that: The specific steps for acquiring multi-polarization state images based on reference synchronous recording are as follows. Based on the baseline synchronization record, configure the trigger queue and solidify the polarization state switching order and initial parameter list to generate a multi-polarization state acquisition task sheet; According to the multi-polarization state acquisition task sheet, multi-polarization state images are acquired, and reference reflection reference area images and reference scattering reference area images are simultaneously captured to generate a multi-polarization state original image group.

5. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 4, characterized in that: The specific steps for generating usable multi-polarization state image groups are as follows. Based on the original multi-polarization state image set, the window contamination state is calculated, the initial parameter list is updated, imaging compensation and fogging suppression are performed, and a compensated multi-polarization state image set is generated. Dark field correction, distortion correction, motion blur suppression, and occlusion removal are performed on the compensated multi-polarization state image group, and gating screening is performed in combination with the window contamination status to generate a usable multi-polarization state image group.

6. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 5, characterized in that: The steps for performing pixel-level alignment and consistency verification on the available multi-polarization state image groups according to the mileage window, and constructing a polarization physical feature layer, are as follows: Corner feature points are extracted from the available multi-polarization state image group according to the mileage window, and a correspondence is established. Pixel-level alignment and edge region overlap checks are performed to generate alignment lock records. Based on the alignment lock record, brightness drift check, noise level check, occlusion consistency check and unified dynamic range normalization are performed, and the effective pixel area mask is fixed to generate a consistency check input set. Based on the consistency check input set, multi-polarization state pixel response differential assembly and stability constraint fusion are performed to construct a polarization physical feature layer. Boundary smoothing and outlier removal are then performed to generate a polarization physical feature layer record.

7. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 6, characterized in that: The specific steps for generating the defect list are as follows: The polarization physical feature layer records are subjected to polarization stability threshold segmentation, connected component aggregation, defect candidate region merging and conflict resolution, and geometric morphology parameters are extracted to generate a set of defect candidate regions. The candidate defect regions are grouped and judged to identify the defect type, locate the defect mileage range, and generate a defect list.

8. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 7, characterized in that: The specific steps for jointly arbitrating the defect list and the window contamination status are as follows. Match and align the defect list with the window contamination status according to the mileage window number, write the missing marker, and generate a joint arbitration input package; Based on the mileage window number, perform defect priority sorting and window contamination status level determination on the joint arbitration input package, and conduct joint arbitration within the same mileage window to generate a control strategy selection record.

9. The online surface quality inspection and control method for the hot rolling process of pickled steel plates as described in claim 8, characterized in that: The specific steps for generating the control execution record are as follows: The control strategy selection record is mapped to online control instructions, and amplitude limiting and instruction atomic splitting are performed to generate an online control instruction set; Based on the online control instruction set, the speed limit of the hot rolling process of window cleaning and pickling plate is executed, and the online control instructions are read back to generate control execution records.

10. The online surface quality inspection and control method for the hot rolling process of pickled steel plate as described in claim 9, characterized in that: The online re-inspection of the repeated windows in the mileage window based on the control execution record, which involves self-compensation acquisition and polarization identification, and marking the control-effective mileage segments to generate a quality inspection control record set, is described in the following steps. Based on the control execution record, establish a mileage window comparison relationship between the re-inspection mileage window and the control mileage window, solidify the re-inspection conditions, and generate an online re-inspection task comparison table; Based on the online re-inspection task comparison table, self-compensation acquisition and polarization recognition are performed to generate a re-inspection defect list; The list of defects to be re-inspected is compared window by window with the online re-inspection task comparison table. A double threshold comparison review is performed to identify and mark the control effective mileage segment and generate a quality inspection control record set.

Images