A method for detecting and alarming steel strip deviation in a continuous casting and rolling production line

Abstract

The present application relates to the technical field of continuous casting and rolling production equipment control, and particularly relates to a continuous casting and rolling production line steel strip deviation detection and alarm method, comprising: deploying multiple sets of visual detection nodes along the tension transmission path of the steel strip of the continuous casting and rolling production line; performing space-time synchronous calibration on all visual detection nodes, and unifying the sampling time reference of each visual detection node; dynamically constructing a deviation detection virtual reference based on the real-time thermal deformation of the steel strip; obtaining real-time deviation data of the steel strip through each visual detection node; combining the timing law of tension transmission, performing timing correlation fitting on multi-node data, identifying the deviation transmission trend, and then predicting the subsequent deviation risk, generating a full-path risk prediction result, and triggering an alarm of the corresponding level. Through the present application, active prediction and full-path control of steel strip deviation anomalies are realized, and the continuous and stable operation of the continuous casting and rolling production line is ensured, and the production risk is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of continuous casting and rolling production equipment control, and in particular to a method for detecting and alarming steel strip deviation in a continuous casting and rolling production line. Background Technology

[0002] The five-wheel continuous casting machine with a belt is the main equipment in a copper rod continuous casting and rolling production line. The annular thin steel strip is a key structural component of the continuous casting section. After it fits tightly with the crystallization groove on the outer edge of the crystallizing wheel, it together forms an arc-shaped crystallization cavity for the continuous solidification and forming of high-temperature molten copper. The stable alignment of the steel strip directly determines the forming quality of the copper billet and the stability of the continuous operation of the production line. Abnormal deviation of the steel strip can not only cause product quality defects such as copper leakage, billet flash, and uneven solidification, but also, in severe cases, cause the steel strip edge to curl, tear, or even break, leading to unplanned shutdowns of the production line and even posing a safety risk of high-temperature molten copper splashing.

[0003] To address abnormal strip misalignment in continuous casting and rolling production lines, existing methods employ a fixed-position, single-point detection scheme, coupled with a threshold-based alarm mode to detect and control strip misalignment. However, during continuous high-speed production in continuous casting and rolling, the strip undergoes thermal deformation due to temperature variations in the rolling process. Simultaneously, abnormal strip misalignment propagates sequentially along the tension transmission path of the production line. The existing detection methods cannot adapt to the thermal deformation characteristics of the strip and the sequential nature of tension transmission in the production line. In practical applications, these methods suffer from insufficient detection accuracy and delayed anomaly warnings, failing to meet the demands of stable strip operation control in continuous high-speed production. Summary of the Invention

[0004] This invention provides a method for detecting and alarming steel strip offset in a continuous casting and rolling production line, which can effectively solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line, comprising: Multiple sets of visual inspection nodes are deployed along the tension transmission path of the steel strip in the continuous casting and rolling production line; Spatiotemporal synchronization calibration is performed on all visual inspection nodes to unify the sampling time reference of each visual inspection node; A virtual reference for offset detection is dynamically constructed based on the real-time thermal deformation of the steel strip; Real-time offset data of the steel strip is obtained through each visual inspection node; Based on the real-time offset data of multiple nodes of the steel strip after spatiotemporal synchronization, and combined with the temporal pattern of steel strip tension transmission, the real-time offset data is subjected to temporal correlation fitting to identify the offset transmission trend of the steel strip. Based on the identified offset transmission trend of the steel strip, the subsequent offset risk of the steel strip is predicted, and the full-path offset risk prediction result is generated. Based on the full-path offset risk prediction results, the corresponding level of steel strip offset alarm is triggered.

[0006] Furthermore, before dynamically constructing the virtual benchmark for offset detection, the initial process parameters of the steel strip are calibrated, and the basic design parameters and rolling condition parameters of the steel strip are obtained as the initial basis for constructing the virtual benchmark for offset detection. The basic design parameters of the steel strip are the design dimensions and tolerance parameters of the steel strip body, while the rolling condition parameters are the process state parameters of the steel strip during real-time rolling.

[0007] Furthermore, the method for dynamically constructing a virtual benchmark for offset detection includes: Extract the effective region of the steel strip, and generate a center reference fitting line parallel to the running direction of the steel strip based on the boundary fitting of the effective region. Using the central reference fitting line as the axis of symmetry, multi-level detection boundary lines are generated symmetrically to both sides to form a virtual reference for offset detection.

[0008] Furthermore, the multi-level detection boundary lines include safety boundary lines and limit threshold lines. The safety boundary lines enclose the allowable offset range of the steel strip design, and the limit threshold lines enclose the maximum allowable offset range of the steel strip. The spacing between the two safety boundary lines matches the standard bandwidth of the steel strip, and the spacing between the two limit threshold lines and the safety boundary lines on the same side matches the maximum allowable offset of the steel strip.

[0009] Furthermore, methods for obtaining real-time offset data of the steel strip include: The system captures images of the steel strip in operation, identifies the coordinates of the two edges of the steel strip, and calculates the real-time offset data of the steel strip.

[0010] Furthermore, the real-time offset data includes the real-time offset of the steel strip, the edge fluctuation amplitude, the offset direction, and the offset rate.

[0011] Furthermore, methods for identifying the offset transmission trend of steel strips by performing time-series correlation fitting on real-time offset data include: The transmission delay of the steel strip between adjacent nodes is calculated based on the operating parameters of the steel strip and the distance between adjacent visual inspection nodes. Based on transmission delay, time-series matching and correlation fitting are performed on the real-time offset data of upstream and downstream nodes to identify the offset transmission trend of the steel strip.

[0012] Furthermore, the operating parameters of the steel strip are the real-time operating speed parameters of the steel strip; Timing matching and correlation fitting involves correlating and fitting the real-time offset data of the upstream node with the real-time offset data of the downstream node after corresponding transmission delay timing.

[0013] Furthermore, the full-path deviation risk prediction results include risk location identification, risk level, expected cross-boundary time, and deviation transmission path.

[0014] Furthermore, methods for predicting the subsequent offset risk of the steel strip include: By combining historical offset data from each visual inspection node, the expected offset and expected time of the steel strip are calculated to generate a full-path offset risk prediction result.

[0015] The technical solution of this invention can achieve the following technical effects: Multiple sets of visual inspection nodes are deployed along the tension transmission path of the steel strip. Spatiotemporal synchronous calibration ensures that the sampling benchmark of multiple nodes is unified. A virtual benchmark for offset detection is dynamically constructed based on the real-time thermal deformation of the steel strip, which is adapted to the thermal deformation characteristics of the steel strip. By acquiring real-time offset data of the steel strip, the offset transmission trend is identified by combining the tension transmission time sequence law. The subsequent offset risk is predicted by combining historical data and generating a full-path prediction result containing information such as risk location and level, which triggers graded alarms. This solves the problems of existing single-point detection, delayed early warning, and poor adaptability of fixed benchmarks. It accurately controls abnormal steel strip offset, avoids hidden dangers such as copper liquid leakage and steel strip tearing, ensures the continuous and stable operation of the continuous casting and rolling production line, and is adapted to high-speed production conditions.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in 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 illustrating a method for detecting and alarming steel strip offset in a continuous casting and rolling production line according to the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] like Figure 1 As shown, the present invention provides a method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line, which specifically includes the following steps: Step S100: Deploy multiple sets of visual inspection nodes along the tension transmission path of the steel strip in the continuous casting and rolling production line; Step S200: Perform spatiotemporal synchronization calibration on all visual detection nodes to unify the sampling time reference of each visual detection node; Step S300: Based on the real-time thermal deformation of the steel strip, dynamically construct a virtual reference for offset detection; Step S400: Obtain real-time offset data of the steel strip through each visual inspection node; Step S500: Based on the real-time offset data of the multi-node steel strip after spatiotemporal synchronization, and combined with the temporal pattern of steel strip tension transmission, perform temporal correlation fitting on the real-time offset data to identify the offset transmission trend of the steel strip. Step S600: Based on the identified offset transmission trend of the steel strip, predict the subsequent offset risk of the steel strip and generate the full-path offset risk prediction result. Step S700: Based on the full path offset risk prediction result, trigger the corresponding level of steel strip offset alarm.

[0022] In this embodiment, by deploying multiple sets of visual inspection nodes along the tension transmission path, performing spatiotemporal synchronous calibration, dynamically constructing a virtual offset benchmark adapted to thermal deformation, and combining multi-node offset data temporal correlation fitting and offset transmission trend prediction, a spatiotemporally collaborative steel strip offset full-domain detection logic is formed. This upgrades single-point static detection to full-path dynamic trend prediction, which not only improves the accuracy of steel strip offset detection but also achieves early warning of offset risks rather than delayed alarms, avoiding problems such as steel strip damage, copper liquid leakage, and unplanned production line shutdowns caused by untimely warnings.

[0023] In a specific implementation, as one example, given that the tension on the steel strip in the continuous casting and rolling production line is transmitted step by step from the preceding casting and rolling zone to the subsequent coiling zone along the running direction, the offset anomaly does not occur synchronously throughout the entire path. Instead, it starts at the position with the most intense tension fluctuation and propagates along the transmission path at a limited speed. Therefore, multiple sets of visual inspection nodes need to be deployed along the tension transmission path so that the field of view of each node covers the full width of the steel strip at the corresponding position, in order to capture the process quantity information of the offset propagating along the path. This embodiment obtains the process quantity data of the offset propagating along the path by deploying multiple visual inspection nodes along the tension transmission path, so that the field of view of each node covers the full width of the steel strip at the corresponding position. The details are as follows: Step S110: Identify the key node positions on the steel strip tension transmission path; in the production line, the steel strip sequentially passes through the guide roller on the exit side of the crystallizing wheel, the free arc section before entering the tensioning wheel, the exit side of the tensioning wheel, and the correction roller before entering the finishing mill; each of the above positions corresponds to a node with independent boundary constraints on the tension transmission path; at the guide roller on the exit side of the crystallizing wheel, the steel strip tension is affected by the periodic fluctuations of the crystallizing wheel rotation; in the free arc section before the tensioning wheel, the steel strip is in a state without rigid support, and the tension fluctuation is mainly manifested as low-frequency drift; at the exit side of the tensioning wheel, the steel strip tension tends to stabilize after being adjusted by the tensioning wheel; at the correction roller, the steel strip is subjected to forced correction, and the offset trend may be corrected in the opposite direction; Step S120: Determine the field of view width for each visual inspection node. The field of view of each node must cover the full width of the steel strip at the corresponding position, that is, the boundary of the field of view width direction exceeds the theoretical outer edge of the steel strip on both sides under the maximum allowable offset state. The maximum allowable offset is determined by the standard bandwidth of the steel strip and the width of the crystallization groove of the crystallization wheel: the width of the crystallization groove minus the standard bandwidth of the steel strip and divided by two, yields the allowable offset on one side. Taking the standard bandwidth of the steel strip as B and the allowable offset on one side as D, the static coverage width is at least B+2D. On the basis of this static coverage width, a redundancy R needs to be added to both sides to form the final field of view width W=B+2D+2R. The redundancy R is used to accommodate the dynamic overrun of the steel strip due to vibration, thermal warping or instantaneous tension fluctuations during high-speed operation, which exceeds the static allowable offset range. The value of R depends on the support method of the steel strip at the location of the inspection node: in the free arc section, the steel strip has no rigid support and the dynamic swing amplitude is large, so R takes a larger value; at the guide roller or correction roller, the steel strip is constrained by the roller surface and the swing amplitude is small, so R takes a smaller value. Step S130: Configure the hardware of each vision inspection node. Each vision inspection node includes an industrial area array camera, a fixed-focus lens, and a bracket. The camera is mounted on the normal direction of the running plane of the steel strip via the bracket, and the optical axis of the lens is perpendicular to the surface of the steel strip. The working distance between the camera and the steel strip is determined according to the required field of view width W and the lens focal length: First, measure the theoretical coordinates of the two edges of the steel strip at the installation position in the state of no offset. Then, calculate the limit coordinates of the two edges after the steel strip reaches the maximum allowable offset on one side and the redundancy R is added. Adjust the working distance so that the camera field of view boundary covers the limit coordinate range. The camera resolution selection should ensure that the imaging width of the steel strip edge is not less than 15 pixels in the full field of view. This threshold is used for sub-pixel edge detection. The camera frame rate selection is based on the time constant of the steel strip offset change: continuously collect 1000 frames of images on the production line, calculate the offset of each frame, perform Fourier analysis on the offset sequence, and take the reciprocal of the highest frequency component with an energy ratio of more than 95% as the time constant. Set the frame rate to 5 to 10 times the frequency corresponding to this time constant. Step S140: Perform on-site calibration for each visual inspection node, including: a. Field of view boundary verification: Use a standard width calibration plate with a width of B+2D+2R. Place the calibration plate in the running plane of the steel belt and move it to the extreme offset positions on both sides to confirm that the edges of both sides of the calibration plate are inside the camera's field of view. b. Confirmation of optical axis perpendicularity: Use a laser line projector to project a reference line parallel to the running direction of the steel strip onto the surface of the steel strip. Compare the parallelism between the edge of the steel strip and the reference line in the image captured by the camera. If the two are not parallel, adjust the camera installation angle until they are parallel. c. Frame rate and speed matching verification: When the production line is running stably at the lowest and highest operating speeds respectively, offset data is continuously collected. The offset sequence is analyzed in the time domain to check for aliasing caused by insufficient sampling. If the offset change between two adjacent frames exceeds 30% of the allowable offset D on one side, the frame rate is determined to be insufficient and the frame rate needs to be increased until the aliasing disappears.

[0024] In this embodiment, the node at the guide roller on the crystallizing wheel exit side outputs the real-time coordinates of the two edges of the steel strip at that position; the node at the free arc section before the tensioning wheel inlet outputs the oscillation trajectory of the steel strip in a state without rigid support; the node at the tensioning wheel exit side outputs the position deviation of the steel strip after adjustment by the tensioning wheel; and the node at the straightening roller before the finishing mill outputs the residual offset after forced correction. Each of the four nodes continuously outputs offset data at a preset frame rate, forming four sets of time series arranged along the tension transmission path. Based on this, when the offset of the upstream node begins to increase but has not yet reached the safety boundary line, the time for it to reach the downstream node can be predicted according to its rate of change and direction, thus achieving early warning. This prediction relies on the accurate calculation of the steel strip transmission delay between adjacent nodes, and the delay calculation requires knowledge of the node spacing and the time correspondence between the offset sequences of the two nodes. If there is no multi-point offset data distributed along the path, the time correspondence cannot be established, resulting in the inability to calculate the transmission delay, the inability to perform time-series matching between the upstream and downstream offset data, and the inability to identify the offset transmission trend.

[0025] In some embodiments of the present invention, the deployed multiple visual inspection nodes independently acquire images of the steel strip running and output offset data. Each node's internal clock source has an inherent frequency deviation, and the sampling start times of each node are inconsistent, resulting in a lack of a unified reference standard between the timestamps carried in the offset data output by different nodes. When the steel strip runs at a high speed, even millisecond-level time alignment errors can cause spatial deviations in the offset between upstream and downstream nodes to reach centimeter levels, far exceeding the allowable offset of the steel strip. Therefore, it is necessary to perform spatiotemporal synchronization calibration on all visual inspection nodes to unify the sampling time reference of each node, ensuring that the synchronization error of the sampling timestamps of each node is controlled within a preset accuracy range that meets the timing matching requirements of multi-node data. Specifically, the following operations are performed: Step S210: Deploy an external timing source, i.e., a pulse signal generator. The periodic drift of its output pulse signal does not exceed a preset threshold within 24 hours of continuous operation. This preset threshold is determined based on the timing matching accuracy required for transmission delay calculation. The transmission delay is equal to the distance between adjacent nodes divided by the real-time running speed of the steel strip. The speed of the steel strip fluctuates during production, and the delay value is not a fixed constant. To ensure that the correlation coefficient between the upstream and downstream offset sequences reaches an acceptable level after delay compensation, the misalignment of the two sequences on the time axis must not exceed the time required for the steel strip to move a distance corresponding to one allowable offset. This value is used as the upper limit of the sampling timestamp synchronization error, i.e., the upper boundary of the preset accuracy range. The periodic drift threshold of the external timing source is set to one-tenth of this upper limit value to ensure that the long-term stability of the timing source does not become the main contributor to the synchronization error. The external timing source outputs a synchronization pulse signal, which is transmitted in parallel to each visual inspection node. Step S220: Each visual inspection node's camera controller is equipped with an external trigger input port, which has interrupt response capability. When the rising edge of the synchronization pulse signal arrives, the camera controller immediately pauses any non-critical tasks currently being executed, records the current moment as a sampling timestamp, and sends an exposure start command to the image sensor. The transmission delay of the exposure start command, i.e., the time interval from the arrival of the rising edge of the synchronization pulse to the start of exposure by the image sensor, must remain constant. This constant value is achieved through a hardware timer inside the camera controller: at the arrival of the rising edge, the hardware timer locks a count value, which triggers exposure start after a fixed number of clock cycles. The duration corresponding to the fixed number of clock cycles is, for example, 2 microseconds, which is much smaller than the upper limit of the preset accuracy range and can be ignored. Each visual inspection node uses the same hardware timer configuration to ensure that the time delay from receiving the synchronization pulse to starting exposure is consistent for each node. Step S230: Measure the transmission delay difference of each visual inspection node. Each node uses the same hardware timer configuration, but the physical transmission path length from the output of the pulse signal generator to the external trigger input port of each node is different, resulting in differences in the signal propagation time in the cable. Furthermore, the response time of the signal conditioning circuit within each node also exhibits device variability. It is necessary to measure the total transmission delay of each node, including cable propagation delay and internal circuit response delay. This is achieved by connecting an oscilloscope probe in parallel at the output of the pulse signal generator and another oscilloscope probe in parallel at the external trigger input port of the node, with both probes connected to the same dual-channel oscilloscope. The pulse signal generator outputs a single pulse, and the oscilloscope records the time difference between the rising edges of the two channels; this time difference is the total transmission delay of the node. Repeat the above measurement for each visual inspection node to obtain the total transmission delay value for each node. Step S240: Set the sampling timestamp compensation value for each node; take the maximum value of the total transmission delay of each node as the reference value. For each node, calculate the difference between the reference value and the total transmission delay of that node. This difference is the sampling timestamp compensation value for that node. After a node completes an image acquisition under the trigger of the synchronization pulse, when the camera controller generates the frame offset data, it sets the sampling timestamp to the actual time when the synchronization pulse arrives at the external trigger input port plus the compensation value of that node. After compensation, the sampling timestamps carried by the offset data output by each node are corrected to a unified time axis starting from the same reference time. This reference time is taken as the time when the synchronization pulse is emitted from the output of the pulse signal generator. After compensation and correction, the sampling timestamp synchronization error between each node is controlled within a preset accuracy range. The upper limit of this preset accuracy range is the time required for the steel belt to move a distance corresponding to an allowable offset, and the lower limit is zero.

[0026] In this embodiment, hardware synchronization pulses are used to trigger sampling. Each node acquires images at the same physical moment, and the time correspondence between offset sequences is directly determined by the rising edge of the synchronization pulse. By measuring the total transmission delay of each node and compensating for it node by node, the synchronization error caused by the difference in signal transmission path length and the discreteness of the device is eliminated. This makes the synchronization accuracy limited only by the periodic drift of the pulse signal generator itself and the resolution of the hardware timer. The magnitudes of the two error sources are much smaller than the upper limit of the preset accuracy range, so that the sampling timestamp synchronization error of all visual inspection nodes is controlled within the preset accuracy range.

[0027] In a specific implementation, as one example, the steel strip undergoes thermal deformation due to the rolling process temperature, leading to changes in bandwidth. Simultaneously, mechanical wear of rotating components causes theoretical alignment position drift. If a fixed physical reference or a preset, unchanging virtual reference is used for offset detection, the bandwidth change will introduce spurious offset components. Furthermore, the fixed reference requires manual recalibration during operating condition switching, making the detection function unreliable during calibration. This embodiment independently constructs its own offset detection virtual reference for each visual inspection node based on the real-time thermal deformation of the steel strip. This allows the detection reference to adaptively adjust with the actual state of the steel strip, adapting to the differences in thermal deformation at each node's location. The specific implementation steps are as follows: Step S310: Obtain the basic design parameters and rolling condition parameters of the steel strip as the initial basis for constructing the virtual benchmark for offset detection. The basic design parameters of the steel strip include the standard bandwidth and allowable offset of the steel strip, which are obtained from the steel strip factory design drawings and factory inspection reports, and verified in conjunction with the sampling inspection results of the steel strip actually used in the production line. The rolling condition parameters include the steel strip rolling process temperature, which is obtained through a temperature sensor. This sensor is deployed at key locations on the steel strip rolling path and corresponds to the visual inspection nodes to ensure that the obtained temperature data corresponds one-to-one with the steel strip area captured by the visual inspection nodes. Step S320: After the production line is restarted for the first time after the last shutdown and maintenance, and the steel belt is in a stable running state without offset, N frames of images are continuously collected, and the inter-frame difference algorithm and sub-pixel edge detection algorithm are executed on each frame of image. Inter-frame difference algorithm filters out interference from roller vibration, rolling water mist, and oxide scale on the steel strip surface in the monitoring image. Roller vibration, rolling water mist, and oxide scale can cause false edges in the image. If not filtered, it will lead to incorrect steel strip edge recognition. The inter-frame difference algorithm can retain the moving area of ​​the steel strip by comparing the pixel differences between two adjacent frames. The sub-pixel edge detection algorithm identifies the sub-pixel coordinates of the two sides of the steel strip. Sub-pixel level detection can improve the recognition accuracy of edge coordinates to below the pixel level, ensuring the accuracy of the steel strip edge coordinates. By identifying the edge coordinates of the two sides of the steel strip, abnormal points on the edges are eliminated, and valid edge coordinates are retained, thereby determining the effective area of ​​the steel strip. The effective area is the area between the two effective edges of the steel strip. This area completely covers the actual operating width of the steel strip and does not contain any interference areas. Step S330: For each frame, the left edge coordinate sequence and the right edge coordinate sequence are fitted with least squares lines to obtain the left edge fitting line and the right edge fitting line respectively. The midline of the two lines is calculated as the center reference fitting line of the frame. The arithmetic mean of the center reference fitting lines of N frames is taken as the initial center reference fitting line. The initial center reference fitting line remains unchanged in the subsequent production process and serves as the absolute reference for the offset. Step S340: Using the initial center reference fitting line as the axis of symmetry, generate safety boundary lines and limit threshold lines in the vertical directions to both sides: offset the initial center reference fitting line to the left by half the standard width of the steel strip to obtain the left safety boundary line; offset it to the right by the same distance to obtain the right safety boundary line; continue to offset the left side of the left safety boundary line by the design allowable offset distance to obtain the left limit threshold line; continue to offset the right side of the right safety boundary line by the design allowable offset distance to obtain the right limit threshold line; The standard bandwidth and design allowable offset of the steel strip used in the above offset operation need to be dynamically corrected according to the actual bandwidth of the current frame: calculate the difference between the actual bandwidth of the current frame and the initially calibrated standard bandwidth of the steel strip, and adjust the initially calibrated design allowable offset proportionally according to the ratio of the actual bandwidth to the standard bandwidth: if the actual bandwidth is greater than the standard bandwidth, the design allowable offset decreases by the same proportion; if the actual bandwidth is less than the standard bandwidth, the design allowable offset increases by the same proportion; the basis for the proportional adjustment is that the width of the crystallization groove is a fixed value. When the steel strip bandwidth increases, the gap between the edge of the steel strip and the inner wall of the crystallization groove decreases, and the allowable offset decreases accordingly; when the steel strip bandwidth decreases, the gap increases, and the allowable offset increases accordingly; the corrected design allowable offset is used to generate the position of the limit threshold line of the current frame; the initial center reference fitting line, the safety boundary line, and the limit threshold line together form the virtual reference for offset detection of the current frame.

[0028] In this embodiment, the initial center reference fitting line is determined and fixed during the first calibration of the production line, serving as the absolute reference for the offset. The deviation of the current frame's steel strip center line from this fixed reference is the real-time offset. The safety boundary line and the limit threshold line are dynamically adjusted according to the actual bandwidth of the current frame, but the position of the boundary line is adjusted, not the position of the center reference fitting line. This ensures that the offset calculation has a fixed reference and that the detection boundary can adapt to the bandwidth changes caused by the thermal deformation of the steel strip.

[0029] In a specific implementation, as one example, the accuracy and real-time performance of the steel strip offset detection alarm method directly depend on the quality of the real-time offset data. The real-time offset data includes at least the real-time offset amount, edge fluctuation amplitude, offset direction, and offset rate of the steel strip. The real-time offset amount is the deviation distance of the steel strip centerline relative to the offset detection virtual reference at the current moment, which is the direct basis for determining whether the safety boundary line or the limit threshold line has been touched. The edge fluctuation amplitude is the root mean square value of the deviation of the steel strip's two edges from their average position within a preset time window, reflecting the stability of the steel strip edges. The offset direction is the deviation of the steel strip centerline relative to the reference, used to determine the direction of the correction action. The offset rate is the change in offset amount per unit time, used to predict the remaining time to reach the boundary line. This embodiment achieves comprehensive and accurate acquisition of real-time steel strip offset data through the design of adapting to multiple visual detection nodes and related algorithm optimizations. The specific implementation steps are as follows: Step S410: Determine the preset sampling frequency for each visual inspection node and acquire the running image of the steel strip; the determination of the sampling frequency needs to be combined with the time characteristics of the steel strip running speed and offset change. The time constant of the steel strip offset change is obtained through on-site testing. The sampling frequency is set to 5 to 10 times the frequency corresponding to the time constant to ensure that the dynamic change process of the steel strip offset can be completely captured, while controlling the amount of data processing to ensure the real-time performance of the inspection. Step S420: Perform interference filtering on the acquired steel strip running screen; Roller vibration, rolling water mist, and oxide scale on the steel strip surface in the monitoring screen can form false edges, leading to incorrect steel strip edge recognition, which in turn causes deviation in offset data calculation and affects detection accuracy; Interference filtering processes the acquired continuous frame images through an inter-frame difference algorithm, and retains the steel strip movement area by comparing the pixel differences between two adjacent frames, ensuring that only the effective area of ​​the steel strip is retained in the image; Step S430: The subpixel edge detection algorithm is used to process the image after interference filtering to identify the subpixel coordinates of the two sides of the steel strip. During the identification process, the image is first converted to grayscale and noise is reduced to reduce the impact of image noise on edge recognition. Then, the subpixel edge detection algorithm is used to locate the precise coordinates of the two sides of the steel strip, improving the edge coordinate recognition accuracy to below the pixel level and ensuring the accuracy of edge coordinate recognition. Abnormal coordinate points caused by residual image interference or algorithm errors are removed, and the effective edge coordinate sequence is retained to ensure the reliability of edge coordinates. Step S440: Based on the extracted effective region, perform least squares line fitting on the left edge coordinate sequence and the right edge coordinate sequence respectively to obtain the left edge fitting line and the right edge fitting line. Calculate the center line of the two lines, which represents the actual position of the geometric center of the steel strip in the current frame. Step S450: Compare the position of the steel strip centerline with the initial center reference fitting line and calculate the vertical distance between them; the vertical distance is output with a sign to indicate the offset direction, a positive sign indicates that the steel strip centerline is located to the right of the initial center reference fitting line, and a negative sign indicates that it is located to the left. Step S460: For the sub-pixel coordinate sequence of the left edge, calculate the residual of all coordinate points in the sequence relative to the fitted line of the left edge, and take the standard deviation of the residual as the fluctuation amplitude of the left edge. Calculate the fluctuation amplitude of the right edge in the same way; take the larger value of the fluctuation amplitudes of the left and right edges as the edge fluctuation amplitude of the current frame for output. Step S470: Calculate the difference between the real-time offset of the current frame and the real-time offset of the previous frame, and divide the difference by the inter-frame time interval to obtain the offset rate.

[0030] In this embodiment, by reasonably determining the preset sampling frequency, it is possible to fully capture the dynamic changes of the steel strip offset while avoiding excessive data processing load, thus balancing detection real-time performance and data integrity. Interference is filtered through the inter-frame difference algorithm, which can effectively eliminate various interferences in the complex environment of the production line without additional hardware investment, ensuring image quality. The sub-pixel edge detection algorithm improves the edge coordinate recognition accuracy, ensuring the accuracy of offset data calculation. By calculating key real-time offset parameters, the dynamic characteristics of the steel strip offset are fully reflected, avoiding misjudgment or omission of offset due to data loss or errors.

[0031] In a specific implementation, as an example, existing methods for detecting steel strip offset are limited to independent judgment of a single node, lacking data correlation between nodes. In a continuous casting and rolling production line, the tension on the steel strip is transmitted step by step from the preceding casting and rolling zone to the subsequent coiling zone along the running direction. The offset anomaly starts at the position with the most intense tension fluctuation and propagates along the transmission path at a limited speed. There is an inherent temporal causal relationship between the offset data of different nodes. The offset change of the upstream node will produce a correlated change in the downstream node after a certain transmission delay. If the temporal correspondence between the offset data of upstream and downstream nodes cannot be established, it is impossible to distinguish the offset propagation from local noise, resulting in the inability to identify the offset transmission trend. This embodiment combines the temporal sequence law of steel strip tension transmission and performs temporal correlation fitting on the real-time offset data of multiple nodes after spatiotemporal synchronization to achieve accurate identification of the steel strip offset transmission trend. The specific implementation steps are as follows: Step S510: Obtain the relevant parameters required for calculating the steel belt transmission delay, including the real-time running speed parameters of the steel belt and the distance between adjacent visual inspection nodes. The real-time running speed parameters of the steel belt determine the transmission time of the steel belt between adjacent nodes. These parameters are collected in real time by speed sensors deployed on the production line. The speed sensors are deployed in correspondence with the visual inspection nodes to ensure that the collected speed data corresponds to the offset data of each node in time and space. The collection frequency is consistent with the sampling frequency of the visual inspection nodes to ensure that the speed data can track the fluctuations of the steel belt running status in real time. The distance between adjacent visual inspection nodes is obtained through on-site measurement. During measurement, the projection point of the camera optical axis of each visual inspection node on the running plane of the steel belt is used as the reference. Multiple measurements are taken using a laser rangefinder, and the average value is taken to eliminate measurement errors. Step S520: Calculate the transmission delay of the steel belt between adjacent visual inspection nodes based on the acquired parameters. The transmission delay is the time required for the steel belt to propagate from the upstream visual inspection node to the downstream adjacent visual inspection node. It is calculated based on the real-time running speed of the steel belt and the distance between adjacent nodes, obtained by dividing the distance between adjacent nodes by the real-time running speed of the steel belt. For the working condition where the instantaneous speed of the steel belt fluctuates greatly, speed data smoothing is added. By averaging the speed data of multiple consecutive frames, the impact of instantaneous speed fluctuations on the delay calculation is reduced, ensuring the stability and accuracy of the transmission delay. Step S530: Based on the calculated transmission delay, perform time-series matching on the real-time offset data of the upstream and downstream visual detection nodes. The time-series matching operation aligns the real-time offset data of the upstream visual detection node with the real-time offset data of the downstream adjacent visual detection node after the corresponding transmission delay time sequence. That is, the offset data of the upstream node at a certain moment is matched with the offset data of the downstream node at that moment plus the transmission delay, ensuring that the offset data of the upstream and downstream nodes correspond to the actual process of offset transmission in time sequence. Step S540: Perform time-series correlation fitting on the real-time offset data of multiple nodes after time-series matching to identify the offset transmission trend of the steel strip. During the fitting process, based on the offset data of upstream and downstream nodes after time-series matching, a correlation model of the offset data is established using a linear fitting method. The input of the model is the real-time offset data of the upstream node, and the output is the predicted value of the offset data of the downstream node at the corresponding time. By comparing the predicted value with the actual offset data collected by the downstream node, the fitting model parameters are adjusted to ensure that the model can accurately reflect the correlation law of the offset data of upstream and downstream nodes. Through this correlation model, the direction, rate, and change law of offset transmission, i.e., the offset transmission trend, can be identified.

[0032] In this embodiment, the problem of timing mismatch between upstream and downstream node offset data is solved by acquiring the steel strip running speed and node spacing in real time and calculating the transmission delay, ensuring the basic reliability of the correlation fitting; timing matching eliminates the timing difference that still exists after spatiotemporal synchronization, so that the upstream and downstream node offset data can accurately correspond to the actual process of offset transmission; and a correlation model of upstream and downstream node offset data is established through timing correlation fitting to achieve accurate identification of offset transmission trend and adapt to the working conditions of continuous high-speed production of continuous casting and rolling.

[0033] In a specific implementation, as one example, existing offset detection methods can only trigger an alarm after the steel strip offset reaches a limit threshold, which is a passive alarm mode, preventing operators from taking timely preventive measures. This embodiment combines historical offset data from each visual detection node with the identified offset propagation trend, and uses the LSTM (Long Short-Term Memory) time-series prediction algorithm to predict the subsequent offset risk of the steel strip and generate the full-path offset risk prediction result. The specific implementation steps are as follows: Step S610: Obtain the basic data required for prediction, including historical offset data of each visual inspection node and the identified steel strip offset transmission trend data; historical offset data is the basis for predicting the subsequent offset pattern of the steel strip, and is extracted from the storage module of each visual inspection node. It needs to cover the offset situation under different rolling conditions and different hot deformation states of the steel strip. Step S620: Preprocess the acquired basic data to eliminate data noise and outliers. The preprocessing operation includes outlier removal and data standardization. Outlier removal is achieved by comparing the fluctuation range of a single set of data with historical data under the same working conditions. Data exceeding the reasonable fluctuation range is identified as outliers and removed. The reasonable fluctuation range is determined based on the statistical results of historical offset data under the same working conditions. Data standardization transforms offset data of different dimensions into the same numerical range to eliminate the impact of differences in the magnitude of different parameters on the prediction model. Step S630: Configure the parameters of the LSTM (Long Short-Term Memory) temporal prediction algorithm and construct the offset risk prediction model. The LSTM algorithm can effectively retain long-term dependency information in the temporal data through the gating mechanism, adapting to the temporal variation pattern of the steel strip offset. The time step of the algorithm is determined according to the transmission delay of the steel strip between adjacent visual detection nodes and the sampling frequency of the visual detection nodes, ensuring that the time step can cover the complete time of the offset anomaly from one node to the next node. The number of hidden layers is determined according to the complexity of the offset data. The more dimensions and the more complex the fluctuations of the offset data, the more the number of hidden layers should be increased appropriately to ensure that the model can fully learn the variation pattern of the offset data. The number of iterations is determined according to the convergence of the model during training. When the model prediction error decreases to the preset threshold and no longer changes significantly, the iteration is stopped to avoid underfitting due to insufficient iterations and overfitting due to excessive iterations. Step S640: Train the offset risk prediction model using the preprocessed training set, and verify the model's prediction accuracy using the test set. During training, use historical offset data and offset propagation trend data from the training set as input, and the actual offset data at the corresponding time as output. Adjust the model parameters using the backpropagation algorithm to gradually reduce the error between the model's predicted value and the actual value to a preset threshold. During the test set verification process, input the test set data into the trained model, compare the predicted offset data output by the model with the actual offset data in the test set, calculate the prediction error, and if the prediction error exceeds the preset threshold, return to readjust the algorithm parameters or optimize the data preprocessing process until the prediction accuracy meets the requirements. Step S650: Input the acquired real-time offset transmission trend data and real-time offset data into the prediction model. Based on the temporal correlation law, the model outputs the expected offset of the steel strip at each visual detection node within a preset time period. The calculation of the expected cross-boundary time is based on the virtual benchmark of offset detection. The positional relationship between the expected offset of each node and the limit threshold line and the safety boundary line is compared. When the expected offset reaches the limit threshold line or the safety boundary line, the corresponding time point is recorded, which is the expected cross-boundary time. If the expected offset never touches the safety boundary line, it is determined that there is no risk of cross-boundary, and the expected cross-boundary time is marked as none. During the calculation process, the calculation accuracy of the expected cross-boundary time is dynamically adjusted in combination with the steel strip offset transmission rate to ensure that the expected cross-boundary time can accurately reflect the development rhythm of the offset anomaly and reserve sufficient prevention and control time for operators. Step S660: Determine the risk level, risk location markers, and offset propagation path to generate a full-path offset risk prediction result. The risk level is classified based on the distance between the expected offset and the limit threshold line and the safety boundary line, as well as the expected cross-boundary time: if the expected offset is close to the limit threshold line and the expected cross-boundary time is short, it is judged as high risk; if the expected offset is close to the safety boundary line and the expected cross-boundary time is long, it is judged as medium risk; if the expected offset does not touch the safety boundary line, it is judged as no risk. The risk location markers correspond to the location information of each visual detection node, clearly marking the nodes with offset risk and their corresponding detection sections. Combined with the offset propagation trend, the subsequent propagation path of the offset anomaly is determined, clearly marking the order of the offset anomaly from the current node to subsequent nodes and the expected arrival time. The full-path offset risk prediction result integrates the risk location markers, risk level, expected cross-boundary time, and offset propagation path, and outputs them in a recognizable format to ensure that operators can quickly obtain key risk information and clarify the key points of prevention and control.

[0034] In this embodiment, by combining the offset propagation trend with the LSTM time-series prediction algorithm, the subsequent offset risk is actively predicted instead of passively waiting for the boundary alarm. The generated full-path prediction result can clearly provide the risk location, level, expected boundary time and propagation path, providing a comprehensive basis for operators to take timely prevention and control measures, effectively avoiding the abnormal expansion of steel strip offset, reducing unplanned shutdowns and safety risks of the production line, and ensuring the continuous and stable operation of the continuous casting and rolling production line.

[0035] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line, characterized in that, include: Multiple sets of visual inspection nodes are deployed along the tension transmission path of the steel strip in the continuous casting and rolling production line; Spatiotemporal synchronization calibration is performed on all the aforementioned visual detection nodes to unify the sampling time reference of each visual detection node; Based on the real-time thermal deformation of the steel strip, a virtual reference for offset detection is dynamically constructed. Before dynamically constructing the virtual reference for offset detection, the initial process parameters of the steel strip are calibrated, and the basic design parameters and rolling condition parameters of the steel strip are obtained as the initial basis for constructing the virtual reference for offset detection. The basic design parameters of the steel strip are the design dimensions and tolerance parameters of the steel strip body, and the rolling condition parameters are the process state parameters of the steel strip during real-time rolling. A method for dynamically constructing a virtual reference for offset detection includes: extracting the effective area of ​​the steel strip; generating a central reference fitting line parallel to the running direction of the steel strip based on the boundary of the effective area; and generating multi-level detection boundary lines symmetrically to both sides of the central reference fitting line as the axis of symmetry to form the virtual reference for offset detection. The running images of the steel strip are collected through each of the aforementioned visual detection nodes, the coordinates of the two sides of the steel strip are identified, and the real-time offset data of the steel strip is calculated. Based on the real-time offset data of the multi-node steel strip after spatiotemporal synchronization, and combined with the temporal pattern of steel strip tension transmission, the real-time offset data is subjected to temporal correlation fitting to identify the offset transmission trend of the steel strip. This includes: calculating the transmission delay of the steel strip between adjacent nodes based on the operating parameters of the steel strip and the distance between adjacent visual detection nodes; and performing temporal matching and correlation fitting on the real-time offset data of upstream and downstream nodes based on the transmission delay to identify the offset transmission trend of the steel strip. Based on the identified offset transmission trend of the steel strip, the subsequent offset risk of the steel strip is predicted, and a full-path offset risk prediction result is generated, including risk location identification, risk level, expected cross-boundary time and offset transmission path. Based on the full-path offset risk prediction results, the corresponding level of steel strip offset alarm is triggered.

2. The method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line according to claim 1, characterized in that, The multi-level detection boundary line includes a safety boundary line and a limit threshold line. The safety boundary line encloses the allowable offset range of the steel strip design, and the limit threshold line encloses the maximum allowable offset range of the steel strip. The spacing between the two safety boundary lines matches the standard bandwidth of the steel strip, and the spacing between the two limit threshold lines and the safety boundary line on the same side matches the maximum allowable offset of the steel strip.

3. The method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line according to claim 1, characterized in that, The real-time offset data includes the real-time offset of the steel strip, the edge fluctuation amplitude, the offset direction, and the offset rate.

4. The method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line according to claim 1, characterized in that, The operating parameters of the steel strip are the real-time operating speed parameters of the steel strip; The timing matching and correlation fitting involves correlating and fitting the real-time offset data of the upstream node with the real-time offset data of the downstream node corresponding to the transmission delay timing sequence.

5. The method for detecting and alarming steel strip misalignment in a continuous casting and rolling production line according to claim 1, characterized in that, A method for predicting the risk of subsequent displacement of the steel strip includes: By combining historical offset data from each visual inspection node, the expected offset and expected time of the steel strip are calculated to generate a full-path offset risk prediction result.

Images