Film edge detection data processing method and system based on visual compensation

Abstract

The application discloses a kind of film edge detection data processing method and system based on visual compensation, it is related to data processing technical field, method includes: the physical span between monitoring station and transmission line speed are obtained;Extract environmental interference, image feature and vibration state data sequence;The dimensionless film section overall offset scale is calculated;The optical confidence attenuation factor is obtained in combination with the environmental interference sequence of monitoring station;Accordingly, the visual servo gain dynamic correction threshold of monitoring station is determined;The wavefront displacement estimate of upstream film section to be monitored is calculated using vibration data, span and line speed, and is transferred to the current station;Finally, the edge monitoring result is obtained in combination with dynamic correction threshold.The application has the advantages of multi-source fusion, time lag compensation and feedforward prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method and system for processing thin film edge detection data based on visual compensation. Background Technology

[0002] In high-speed film winding and coating production lines, edge position detection and correction are crucial for maintaining winding neatness. Existing technologies typically rely on vision sensors deployed at specific workstations to capture and control local edge coordinates.

[0003] However, as a continuous flexible medium, the tension distribution and physical displacement of a thin film exhibit cross-regional spatiotemporal coupling characteristics. Mechanical vibrations generated by upstream guide rollers or traction mechanisms are transmitted downstream along the film's transport direction as mechanical waves. This transmission process is constrained by the film's physical span and transmission linear velocity, resulting in a wavefront time delay. Existing single-station closed-loop control logic often neglects the wavefront transmission characteristics of continuous media, treating it as a discrete static deviation. This leads to the response of the correction actuator lagging behind the arrival time of the physical deformation, making feedforward intervention impossible.

[0004] Meanwhile, the lighting environment in industrial settings increases the risk of data distortion in detection. The surface of the operating thin film is prone to dynamic specular reflections, and the flickering of workshop lighting equipment and the obstruction from moving mechanical parts create optical noise on the imaging surface of the vision sensor. This optical noise can cause feature extraction algorithms to misinterpret light and shadow fluctuations as actual geometric displacement of the thin film, resulting in spurious bias data. Existing data processing architectures lack mechanisms for evaluating the optical confidence of vision sensors, typically using acquired image feature data as input to the control loop. If the detection area encounters lighting interference, the control system will follow the distorted visual signal, driving the servo motor to produce unnecessary adjustments and vibrations. The combined effect of mechanical wave transmission delays and optical noise-induced misinterpretations makes existing edge detection solutions susceptible to failure under dynamic conditions, leading to thin film edge damage or roll material scrapping. A data processing mechanism capable of removing optical interference and fusing physical time-delay mapping is needed to address these issues. Summary of the Invention

[0005] In view of the technical problems described in the background art, the present invention provides a method and system for processing thin film edge detection data based on visual compensation.

[0006] A visually compensated method for processing thin film edge detection data includes: acquiring the physical span between adjacent monitoring stations and the linear velocity of the thin film during thin film transport operations; dividing the transported thin film into multiple continuous thin film segments along the transport direction; establishing a spatial tracking sequence and denoting it as the [number missing]. The thin film segment, of which the first... Each film segment refers to any one of the plurality of film segments. It is a positive integer, and The total number of thin film segments is less than or equal to the total number of segments. The environmental interference data sequence, image feature data sequence, and vibration state data sequence of the monitoring station within a set time period are obtained. Based on the image feature data sequence of the j-th thin film segment, a dimensionless overall offset scale is obtained. The monitoring station where the j-th thin film segment is located at the current moment is obtained, and based on the environmental interference data sequence of that monitoring station, a j-th optical confidence attenuation factor is obtained. Based on the j-th optical confidence attenuation factor and the j-th overall offset scale, a visual servo gain dynamic correction threshold for the monitoring station where the j-th thin film segment is located at the current moment is obtained. The thin film segment to be monitored upstream of the j-th thin film segment is obtained, and based on the vibration state data sequence of the thin film segment to be monitored, the physical span, and the thin film transmission line velocity, a j-th wavefront displacement estimate is obtained from the thin film segment to be monitored to the monitoring station where the j-th thin film segment is located at the current moment. Based on the j-th wavefront displacement estimate and the visual servo gain dynamic correction threshold, the edge monitoring result of the thin film segment to be monitored is obtained.

[0007] Optionally, obtaining the dimensionless j-th overall offset scale based on the image feature data sequence of the j-th thin film segment includes: extracting the maximum absolute distance from the baseline in the image feature data sequence, calculating the difference between the maximum absolute distance and a set tolerance physical threshold, and performing a ratio operation between the difference and the tolerance physical threshold to obtain a dimensionless instantaneous impulse geometric deviation coefficient; extracting the absolute distance difference between the end time and the beginning time of the image feature data sequence within a set time period, and performing a ratio operation between the absolute distance difference and the tolerance physical threshold to obtain a dimensionless edge drift trend coefficient; if the edge drift trend coefficient is less than zero, setting the j-th overall offset scale to be equal to the instantaneous impulse geometric deviation coefficient; if the edge drift trend coefficient is not less than zero, setting the j-th overall offset scale to the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

[0008] Optionally, obtaining the j-th optical confidence attenuation factor based on the environmental interference data sequence of the monitoring station includes: calculating the standard deviation of the data variation in the environmental interference data sequence; if the standard deviation does not exceed a set optical reference threshold, then obtaining a value of zero as the j-th optical confidence attenuation factor; if the standard deviation exceeds the optical reference threshold, then constructing an exponential decay function based on the difference between the standard deviation and the optical reference threshold, and obtaining a continuous dimensionless value greater than zero and less than one through the exponential decay function as the j-th optical confidence attenuation factor.

[0009] Optionally, obtaining the visual servo gain dynamic correction threshold for the monitoring position of the j-th thin film segment at the current moment, based on the j-th optical confidence attenuation factor and the j-th overall offset scale, includes: using the j-th overall offset scale as a geometric reference scale value, and using the product of the j-th overall offset scale and the j-th optical confidence attenuation factor, which exhibits exponential nonlinearity, as a feedforward adaptive compensation term; performing an algebraic addition of the geometric reference scale value and the feedforward adaptive compensation term to obtain a dimensionless correction multiplier as the visual servo gain dynamic correction threshold, wherein the visual servo gain dynamic correction threshold is used as a feedforward intervention weight to positively amplify the mechanical wave forward displacement estimate when the optical confidence attenuates.

[0010] Optionally, obtaining the estimated j-th wavefront displacement of the monitored thin film segment at the current monitoring station based on the vibration state data sequence of the monitored thin film segment, the physical span, and the thin film transmission linear velocity includes: obtaining the real-time running tension and linear density of the thin film material during the thin film transmission operation, and calculating the independent propagation wave velocity of the transverse mechanical elastic wave in the thin film medium; performing vector superposition of the thin film transmission linear velocity and the independent propagation wave velocity in the same or opposite directions to obtain the equivalent wavefront transmission velocity of the flexible medium; dividing the physical span by the equivalent wavefront transmission velocity of the flexible medium to obtain the time delay of the vibration wavefront transmission compensated for the flexible expansion and contraction characteristics; extracting vibration data within the historical window corresponding to the time delay in the vibration state data sequence, and calculating the root mean square value of the vibration data; constructing a simple harmonic physical mapping model based on the calibrated natural frequency of the mechanical structure, and mapping the root mean square value from the dimension of acceleration to the dimension of spatial length to obtain the estimated j-th wavefront displacement.

[0011] Optionally, obtaining the edge monitoring result of the film segment to be monitored based on the estimated j-th wavefront displacement and the visual servo gain dynamic correction threshold includes: if the estimated j-th wavefront displacement is not greater than zero, obtaining the edge monitoring result in a stable state; if the estimated j-th wavefront displacement is greater than zero, obtaining the effective difference between the estimated j-th wavefront displacement and the set mechanical dead zone threshold; multiplying the effective difference by the visual servo gain dynamic correction threshold, and outputting a target evaluation value that retains the absolute length dimension as the edge monitoring result.

[0012] A visually compensated thin-film edge detection data processing system is also provided, comprising: an acquisition module for acquiring the physical span between adjacent monitoring stations and the linear velocity of the thin-film transmission during thin-film transport operations; dividing the transmitted thin film into multiple continuous thin-film segments along the transmission direction, establishing a spatial tracking sequence and denoting it as the j-th thin-film segment, where the j-th thin-film segment refers to any one of the multiple thin-film segments, j is a positive integer, and j is less than or equal to the total number of thin-film segments; and acquiring the environmental interference data sequence, image feature data sequence, and vibration state data sequence of the monitoring station within a set time period; and a data processing module for acquiring a dimensionless j-th overall offset scale based on the image feature data sequence of the j-th thin-film segment; and acquiring the monitoring station where the j-th thin-film segment is located at the current moment. Based on the environmental interference data sequence of the monitoring station, the j-th optical confidence attenuation factor is obtained; the coefficient calculation module is used to obtain the visual servo gain dynamic correction threshold of the j-th film segment at the current monitoring station based on the j-th optical confidence attenuation factor and the j-th overall offset scale; the trend monitoring module is used to obtain the film segment to be monitored upstream of the j-th film segment, and based on the vibration state data sequence of the film segment to be monitored, the physical span, and the film transmission line velocity, obtain the j-th wavefront displacement estimate transmitted from the film segment to be monitored to the monitoring station of the j-th film segment at the current time; and based on the j-th wavefront displacement estimate and the visual servo gain dynamic correction threshold, obtain the edge monitoring result of the film segment to be monitored.

[0013] Optionally, the data processing module is further configured to: extract the maximum absolute distance from the baseline in the image feature data sequence, calculate the difference between the maximum absolute distance and a set tolerance physical threshold, and perform a ratio operation between the difference and the tolerance physical threshold to obtain a dimensionless instantaneous impulse geometric deviation coefficient; extract the absolute distance difference between the end time and the beginning time of the image feature data sequence within a set time period, and perform a ratio operation between the absolute distance difference and the tolerance physical threshold to obtain a dimensionless edge drift trend coefficient; if the edge drift trend coefficient is less than zero, set the j-th overall offset scale to be equal to the instantaneous impulse geometric deviation coefficient; if the edge drift trend coefficient is not less than zero, set the j-th overall offset scale to be the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

[0014] Optionally, the data processing module is further configured to: calculate the standard deviation of the data variation in the environmental interference data sequence; if the standard deviation does not exceed a set optical reference threshold, obtain a value of zero as the j-th optical confidence attenuation factor; if the standard deviation exceeds the optical reference threshold, construct an exponential decay function based on the difference between the standard deviation and the optical reference threshold, and obtain a continuous dimensionless value greater than zero and less than one through the exponential decay function as the j-th optical confidence attenuation factor.

[0015] Optionally, the trend monitoring module is also used to: acquire the real-time operating tension and linear density of the thin film material during the thin film transport operation, and calculate the independent propagation wave velocity of the transverse mechanical elastic wave in the thin film medium; perform vector superposition of the thin film transport linear velocity and the independent propagation wave velocity in the same or opposite directions to obtain the equivalent wavefront propagation velocity of the flexible medium; divide the physical span by the equivalent wavefront propagation velocity of the flexible medium to obtain the time delay of the vibration wavefront propagation compensated for the flexible expansion and contraction characteristics; extract the vibration data in the historical window corresponding to the time delay in the vibration state data sequence, and calculate the root mean square value of the vibration data; construct a simple harmonic physical mapping model based on the calibrated natural frequency of the mechanical structure, and map the root mean square value from the dimension of acceleration to the dimension of spatial length to obtain the estimated value of the j-th wavefront displacement.

[0016] The beneficial effects of this invention are reflected in: In the entire visual compensation-based thin-film edge detection data processing method, the instantaneous impulse geometric deviation coefficient and edge drift trend coefficient are extracted to transform physical variables with length units into a dimensionless overall offset scale, thus solving the dimensional differences in the data fusion process. Simultaneously, an exponential decay function based on the standard deviation of reflected light intensity is introduced to quantify the optical confidence attenuation factor, enabling the identification of optical noise interference caused by on-site reflections or flicker, avoiding logical abrupt changes in existing binarization judgments. Based on this, the scheme dynamically calculates the visual servo gain dynamic correction threshold and establishes a weight transformation mechanism. When the optical environment deteriorates, this correction threshold is increased to adjust the control weights of subsequent feedforward data, suppressing erroneous actions caused by servo mechanisms following optical pseudo-deviations. For the mechanical transmission characteristics of continuous flexible media, this method utilizes the thin film physical span and transmission line velocity to extract the wavefront propagation time delay. Furthermore, the scheme constructs a simple harmonic physical mapping model based on the inherent dominant frequency of the mechanical structure, mapping the upstream vibration root mean square value from an acceleration dimension to a wavefront displacement estimate with a spatial length dimension. This frequency domain mapping, combined with a preset mechanical dead zone constraint, filters out conventional background vibrations. Before the effective displacement reaches the downstream station, it is multiplied with the aforementioned optical correction threshold to output a target evaluation value that retains the length dimension. This closed-loop processing enables feedforward intervention through upstream mechanical vibration time-delay mapping when the visual sensor is interfered with by ambient light and shadow, improving the stability and anti-interference capability of edge correction control during winding operations. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a schematic diagram illustrating the steps of the visual compensation-based thin film edge detection data processing method of the present invention; Figure 2 This is a schematic diagram of a portion of steps S1 in the visual compensation-based thin film edge detection data processing method of the present invention; Figure 3 This is a schematic diagram of a portion of step S2 in the visual compensation-based thin film edge detection data processing method of the present invention; Figure 4 This is a schematic diagram of part of step S4 in the visual compensation-based thin film edge detection data processing method of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] This invention provides a method for processing thin film edge detection data based on visual compensation, such as... Figure 1 As shown, in one specific embodiment, the method includes: S1. Obtain the physical span between adjacent monitoring stations and the linear velocity of the thin film during the thin film transport operation; divide the thin film in the transport along the transport direction into multiple continuous thin film segments, establish a spatial tracking sequence and record it as the j-th thin film segment, where the j-th thin film segment refers to any one of the multiple thin film segments, j is a positive integer, and j is less than or equal to the total number of thin film segments; obtain the environmental interference data sequence, image feature data sequence and vibration state data sequence of the monitoring station within a set time period.

[0023] S2. Based on the image feature data sequence of the j-th thin film segment, obtain the dimensionless j-th global offset scale; obtain the monitoring position of the j-th thin film segment at the current moment, and obtain the j-th optical confidence attenuation factor based on the environmental interference data sequence of the monitoring position. The optical confidence attenuation factor refers to a quantitative assessment value of the quality degradation of imaging data when it is affected by external fluctuations; a larger value indicates lower data reliability.

[0024] S3. Based on the j-th optical confidence attenuation factor and the j-th overall offset scale, obtain the visual servo gain dynamic correction threshold for the monitoring position of the j-th thin film segment at the current moment. Here, the visual servo gain dynamic correction threshold refers to the dimensionless multiplier used to adjust the back-end control tolerance after comprehensively considering geometric offset and environmental confidence.

[0025] S4. Obtain the monitored thin film segment upstream of the j-th thin film segment. Based on the vibration state data sequence, physical span, and thin film transmission linear velocity of the monitored thin film segment, obtain the estimated j-th wavefront displacement of the monitored thin film segment transmitted to the monitoring station of the j-th thin film segment at the current moment. Based on the estimated j-th wavefront displacement and the visual servo gain dynamic correction threshold, obtain the edge monitoring result of the monitored thin film segment. The estimated wavefront displacement is calculated by treating the thin film as a continuous waveguide medium, representing the expected lateral spatial deformation when the mechanical vibration wave generated at the upstream station is transmitted to the downstream station along with the thin film.

[0026] In this embodiment, it should be noted that in S1, at the initial stage of thin film edge detection, a unified spatiotemporal mapping benchmark is first established to address the technical problem of isolated and difficult-to-synchronize physical data between different monitoring stations in the industrial field. It should also be noted that the first... A thin film segment refers to any one of a plurality of thin film segments, wherein It is a positive integer, and The total number of thin film segments is less than or equal to the total number of segments, determined by introducing variables. It enables serial tracking of the state of thin films at different times and locations on a continuous production line.

[0027] Furthermore, the physical span between the upstream tension roller to be monitored and the downstream target coating die is obtained, specifically set to 2.5 meters, and the film transmission linear velocity set under the current process is recorded in real time as 5.0 meters per second. Within a fixed sampling period of 0.1 seconds, the multi-source sensing devices deployed on-site are synchronously driven to perform discrete sampling, extracting image feature data sequences containing absolute physical coordinates in the lateral direction, environmental interference data sequences recording grayscale fluctuations in surface reflected light intensity, and vibration state data sequences capturing the amplitude of mechanical lateral jitter. The core computational logic of this step lies in strictly aligning the three different dimensions of physical states—visual imaging, optical interference, and mechanical motion—on the time axis through a unified timestamp, eliminating the data misalignment phenomenon caused by inconsistent acquisition frequencies in existing single-point monitoring. The establishment of this mechanism transforms the originally independent sensor signals into a low-level data stream with spatiotemporal coupling, providing reliable parameter input support for subsequent analysis of the wavefront propagation characteristics of continuous flexible media and the removal of light and shadow interference, ensuring the effectiveness of subsequent fusion calculations.

[0028] In S2, after acquiring the synchronous data stream, the local geometric deviation and optical imaging quality are quantized in a dimensionless manner to address dimensional conflicts and visual pseudo-bias caused by light and shadow fluctuations during multi-source data fusion. First, the image feature sequence is processed, extracting the maximum absolute distance from the baseline as 1.4 mm. This difference is then calculated by subtracting it from a tolerance physical threshold of 1.0 mm, and finally, a ratio is calculated between this value and the threshold to obtain the instantaneous impulse geometric deviation coefficient, which is 0.4. The instantaneous impulse geometric deviation coefficient represents the relative ratio between the maximum absolute physical distance from the unbiased baseline in the feature data and the set tolerance boundary, used to quantitatively characterize the instantaneous geometric offset amplitude caused by physical impact.

[0029] Simultaneously, the distance difference between the last 1.2 mm and the initial 0.9 mm in this time period was extracted, and the edge drift trend coefficient was calculated to be 0.3 using the same ratio. The edge drift trend coefficient refers to the relative ratio of the change in the beginning and end of the feature data within the observation window to the tolerance boundary, and is used to characterize whether the lateral slip behavior tends to converge or diverge.

[0030] Given that the trend value is divergent, the two are added together to obtain an overall offset scale of 0.7, completing the dimensionless transformation of physical length. At the optical level, the standard deviation of light intensity fluctuation is calculated to be 35. Since this value exceeds the optical reference threshold of 15, a noise resistance constant of 40 is used, and the result is a smoothing calculation using an exponential decay model, yielding an optical confidence decay factor of 0.3935. This calculation logic removes the dimensional constraints of the basic data and transforms environmental interference into a continuous weighted evaluation variable. Its beneficial effect is that it not only unifies the scale of subsequent fusion operations but also objectively measures the degree of confidence reduction caused by workshop flicker interference in the visual sensor.

[0031] In S3, based on the dimensionless processing completed in the previous step, evaluation variables from both geometric and optical dimensions are further integrated to address the technical problem of misadjustment and excessive tremor caused by the actuator following visual interference signals. The core calculation logic uses the previously derived overall offset scale of 0.7 as the base evaluation value and multiplies this base evaluation value by an optical confidence attenuation factor of 0.3935 to calculate an additional compensation value of 0.27545. Subsequently, the base evaluation value and the additional compensation value are algebraically added to obtain a visual servo gain dynamic correction threshold of 0.97545. This calculation process essentially constructs an environment-adaptive weight adjustment mechanism. When the detection area encounters strong environmental reflections or changes in lighting that cause visual feature degradation, this calculation logic automatically amplifies this dimensionless correction multiplier proportionally. The beneficial effect of this data processing step is that it establishes a data barrier at the algorithm level, blocking the direct drive of the servo motor by the local distorted image signal. By outputting this dynamic correction threshold, it guides the control to actively increase the tolerance to error signals when the visual environment deteriorates, and shifts the weight of the control reference towards the direction of mechanical vibration feedforward prediction, thus ensuring the smooth operation of edge tracking.

[0032] In S4, after parameter correction, a feedforward prediction mechanism based on mechanical wave propagation characteristics is executed to address the response lag problem between physical deformation propagation and correction actions in continuous flexible media. The computational logic first derives a wavefront propagation time delay of 0.5 seconds by dividing the 2.5-meter physical span by a linear velocity of 5.0 meters per second. Based on this, it backtracks to extract the root mean square value of vibration per square second over a historical window of 1500 millimeters. To avoid drift errors caused by continuous integration of discrete signals, a frequency-domain physical mapping model is used, combining a 10 Hz mechanical natural frequency and a lateral deformation constant of 1.2 to convert the acceleration dimension into a wavefront displacement estimate of 0.456 millimeters. Since this value is greater than zero, a mechanical dead zone threshold of 0.2 millimeters is subtracted, extracting an effective difference of 0.256 millimeters. Finally, this difference is multiplied by a dynamic correction threshold of 0.97545 to output a target evaluation value of 0.2497 millimeters. The beneficial effects of this step are that frequency domain mapping replaces calculus to ensure the stability of the operation within the industrial controller, while dead zone constraints filter out harmless background vibrations; this mechanism enables feedforward prediction before the physical displacement reaches the downstream station, and combined with optical compensation weights, outputs intervention commands with absolute length dimensions, improving the real-time response capability of winding control.

[0033] In summary, the entire visual compensation-based thin-film edge detection data processing method solves the dimensional differences in the data fusion process by extracting the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient, transforming physical variables with length units into a dimensionless overall offset scale. Simultaneously, an exponential decay function based on the standard deviation of reflected light intensity is introduced to quantify the optical confidence attenuation factor, enabling the identification of optical noise interference caused by on-site reflections or flicker, avoiding logical abrupt changes in existing binarization judgments. Based on this, the scheme dynamically calculates the visual servo gain dynamic correction threshold and establishes a weight transformation mechanism. When the optical environment deteriorates, this correction threshold is increased to adjust the control weights of subsequent feedforward data, suppressing erroneous actions caused by servo mechanisms following optical pseudo-deviations. For the mechanical transmission characteristics of continuous flexible media, this method uses the thin film physical span and transmission line velocity to extract the wavefront propagation time delay. Furthermore, the scheme constructs a simple harmonic physical mapping model based on the inherent dominant frequency of the mechanical structure, mapping the upstream vibration root mean square value from an acceleration dimension to a wavefront displacement estimate in the spatial length dimension. This frequency domain mapping, combined with a preset mechanical dead zone constraint, filters out conventional background vibrations. Before the effective displacement reaches the downstream station, it is multiplied with the aforementioned optical correction threshold to output a target evaluation value that retains the length dimension. This closed-loop processing enables feedforward intervention through upstream mechanical vibration time-delay mapping when the visual sensor is interfered with by ambient light and shadow, improving the stability and anti-interference capability of edge correction control during winding operations.

[0034] like Figure 2 As shown, in one specific embodiment, S1 includes: S11, obtaining physical mapping parameters. The thin-film conveyor line is divided into multiple monitoring stations, and the physical span between adjacent monitoring stations is obtained, with the dimension of the physical span being spatial length. The thin-film conveyor line speed during the operation is obtained, with the dimension of the thin-film conveyor line speed being length divided by time.

[0035] The set thin-film transmission linear speed is determined based on the physical material properties of the target thin film, its tensile strength limit, and the overall production line's capacity requirements. It is optimized through orthogonal experiments using historical process windows. For example, for lithium-ion battery separators of specific thicknesses, transmission tests at different speed gradients were conducted during the pilot production phase, and the yield was recorded. Historical test data showed that when the linear speed exceeded 6.0 m / s, uneven stress distribution within the film significantly increased the risk of edge tearing and wavy edges; while below 4.0 m / s, although operation was stable, it failed to meet economic capacity targets. Considering the rated power of the equipment's traction motor and the servo response bandwidth of the tension closed-loop control, the optimal thin-film transmission linear speed for this specific process, ensuring continuous and stable transmission, was ultimately set at 5.0 m / s.

[0036] S12. Acquire time-series sampling data. Within a uniform set time period, acquire image feature data sequences characterizing the absolute physical coordinates of the thin film in the lateral direction; acquire environmental disturbance data sequences characterizing the surface reflected light intensity; and acquire vibration state data sequences characterizing the lateral vibration amplitude.

[0037] The set time period (i.e., the fixed sampling time period) is determined through signal analysis and system integration, based on the lower limit of the maximum sampling rate of the multi-source sensors, the communication bandwidth of the industrial control bus, and the actual control cycle of the correction servo system. For example, during the system integration phase, the highest frame rate of the field-deployed vision sensors was assessed at 60 frames per second, and the sampling rate of the high-frequency vibration sensors was 1000 times per second. To ensure that at least several effective discrete image frames and hundreds of vibration data samples can be accumulated within a data processing cycle to meet the statistical calculation requirements of standard deviation and root mean square values, without causing congestion in the underlying bus transmission or severe lag in control decisions due to excessively large data packets, and to cover a complete fluctuation cycle of the mechanical's inherent main frequency (e.g., 10 Hz), the optimal value for this set time period was determined to be 0.1 seconds after a limited number of closed-loop tracking tests.

[0038] In this embodiment, it should be noted that in S11 and S12, a unified physical mapping parameter and timing sampling benchmark are first established to solve the problem of loosely distributed multi-source sensors and difficulty in synchronizing data in industrial settings. The film conveyor line is divided into multiple monitoring stations, and the physical span between the upstream tension roller and the downstream coating die is obtained as 2.5 meters, and the current film conveyor line speed is recorded as 5.0 meters per second.

[0039] Subsequently, a uniform time period of 0.1 seconds was set to synchronously drive the visual sensors, laser displacement sensors, and high-frequency vibration acceleration sensors deployed on-site. Within this time period, image feature data sequences representing the absolute physical coordinates in the lateral direction, environmental interference data sequences representing the surface reflected light intensity, and vibration state data sequences representing the lateral vibration amplitude were acquired, respectively. This operational logic, through the strict alignment of physical space distance measurement and the time axis, eliminated data misalignment caused by differences in sampling frequency and spatial location between workstations. This transformed the originally isolated optical, image, and mechanical signals into a low-level data stream with spatiotemporal coupling, providing basic parameter inputs for subsequent analysis of wavefront propagation characteristics.

[0040] like Figure 3As shown, in one specific embodiment, S2 includes: S21, obtaining the instantaneous impulse geometric deviation coefficient. Setting the unbiased baseline coordinates and a set tolerance physical threshold. Converting the real-time coordinates in the image feature data sequence into absolute distances from the baseline. When there is a maximum absolute distance in the sequence exceeding the tolerance physical threshold, extracting the difference between the maximum absolute distance and the tolerance physical threshold. Ratioing the absolute distance difference in the image feature data sequence to the set tolerance physical threshold, eliminating length units through division, and obtaining the dimensionless instantaneous impulse geometric deviation coefficient.

[0041] The unbiased baseline coordinates are set based on the mechanical centerline position of the winding coating machine and the ideal transmission trajectory of the film under standard tension, obtained through static calibration and multi-point alignment procedures during system initialization. For example, with the equipment stopped and the tape threaded under stable rated tension, a high-precision laser line marker projects the absolute physical centerline of the machine frame onto the film surface. Subsequently, the mounting posture and field of view of the vision sensor are finely adjusted so that the origin of the image coordinate system or a specified center pixel column precisely coincides with this physical centerline. Under this static, unbiased condition, the camera is continuously triggered to acquire 50 reference images and extract the true edge coordinates of the film. The mathematical expectation value of these 50 coordinates is calculated to filter out residual errors from static assembly and minor environmental fluctuations. Finally, this absolute expectation value is set as the unbiased baseline coordinates during system operation.

[0042] The set tolerance physical threshold is based on the characteristics of the thin film material and the requirements of process precision. It is obtained by collecting lateral fluctuation amplitude data of good films from a limited number of historical production runs, and then extracting the upper limit of its confidence interval after distribution fitting. For example, in the historical production prototype test, lateral absolute displacement data of 1000 good films were continuously collected during transmission. Statistical analysis showed that 99.7% of the data fell between 0.1 mm and 0.9 mm. To ensure that the system does not generate false alarms and to take into account engineering margins, the upper limit of the normal distribution interval of 0.9 mm was added to a safety margin of 0.1 mm, and the final tolerance physical threshold was determined to be 1.0 mm.

[0043] S22. Obtain the edge drift trend coefficient. Extract the absolute distance difference between the end time and the beginning time of the image feature data sequence within the current set time period, and calculate the ratio of this difference to the set tolerance physical threshold to obtain the dimensionless edge drift trend coefficient.

[0044] S23. Obtain the j-th overall offset scale. If the edge drift trend coefficient is less than zero, set the j-th overall offset scale to be equal to the instantaneous impulse geometric deviation coefficient; if the edge drift trend coefficient is not less than zero, set the j-th overall offset scale to be the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

[0045] S24. Calculate the standard deviation of the data. Calculate the standard deviation of the data variation in the environmental interference data sequence for this monitoring station, which is used to measure the level of local high-frequency light and shadow fluctuations.

[0046] S25. Obtain the j-th optical confidence attenuation factor. If the standard deviation does not exceed the set optical reference threshold, then the value of zero is obtained as the j-th optical confidence attenuation factor. If the standard deviation exceeds the optical reference threshold, then an exponential decay function is constructed based on the difference between the standard deviation and the optical reference threshold. A continuous dimensionless value greater than zero and less than one is obtained through the exponential decay function as the j-th optical confidence attenuation factor. The specific formula is as follows:

[0047] In the formula, Let be the j-th optical confidence attenuation factor, which is a dimensionless value; It is an exponential function composed of natural constants; The standard deviation of the data variation in the environmental disturbance data sequence; The set optical reference threshold maintains the same units of measurement as the standard deviation; The optical constants characterizing noise resistance are kept in the same dimension as the standard deviation to ensure that the ratios within the exponential term are dimensionless parameters.

[0048] The optical reference threshold and noise immunity constant were determined based on an on-site anechoic chamber interferometry experiment. These were obtained by collecting a limited number of historical optical background noise data points, followed by spectral analysis and mean filtering. For example, with basic workshop lighting maintained, 500 consecutive ambient reflected light intensity grayscale data points were collected. The mean standard deviation of the background was calculated to be 12. Considering sensor temperature drift compensation of 3 grayscale levels, the optical reference threshold was set to 15. Simultaneously, in a stroboscopic light source calibration experiment, the maximum light intensity fluctuation standard deviation was measured to be 55. The difference of 40 between this maximum value and the reference threshold of 15 was extracted and directly used as the noise immunity constant setting value for the smooth attenuation function.

[0049] In this embodiment, it should be noted that in S21, the instantaneous impulse geometric deviation coefficient is obtained to address the length dimension conflict and transient physical impact quantification issues during multi-source data fusion. First, an unbiased baseline coordinate is set, and a set tolerance physical threshold of 1.0 mm is defined. When processing the image feature data sequence, the maximum absolute distance deviating from the baseline is extracted to be 1.4 mm. Then, the difference between this maximum absolute distance of 1.4 mm and the tolerance physical threshold of 1.0 mm is extracted, yielding an excess of 0.4 mm. To eliminate the constraints of physical dimensions, this 0.4 mm difference is compared with the set tolerance physical threshold of 1.0 mm, and the length unit is eliminated through division, thus obtaining a dimensionless instantaneous impulse geometric deviation coefficient of 0.4. This calculation logic transforms specific physical displacements into relative deviation ratios, quantifies the degree of transient physical impact on the film, and unifies the calculation scale for subsequent multi-dimensional data algebraic fusion through dimensionless processing, improving the algorithm's compatibility with different operating parameters.

[0050] In step S22, an edge drift trend coefficient is obtained to address the technical challenge of assessing the dynamic trajectory of the thin film's lateral slippage during transmission. Within a set timeframe of 0.1 seconds, the absolute distance at the end of the image feature data sequence is extracted as 1.2 mm, and the absolute distance at the beginning is extracted as 0.9 mm. The difference between these two values ​​is calculated, yielding a distance increment of 0.3 mm. Subsequently, this 0.3 mm difference is compared with a set tolerance physical threshold of 1.0 mm, and the dimensionless edge drift trend coefficient of 0.3 is obtained by eliminating the length dimension through division. This calculation logic determines the current lateral slippage behavior of the thin film by comparing the displacement changes at the beginning and end of the observation window. A positive trend coefficient indicates that the thin film is moving away from the baseline and is in a divergent state; a negative value indicates that the thin film is converging towards the baseline. This processing method dynamically captures the continuous cumulative trend of edge displacement, providing a quantitative basis for assessing the urgency of corrective intervention.

[0051] In S23, the j-th overall offset scale is obtained by fusing the above parameters, aiming to solve the problem that a single transient deviation or a single trend data cannot fully reflect the true geometric deviation state of the thin film. The specific value of the edge drift trend coefficient is determined based on the pre-calculated value. In this scenario, the calculated edge drift trend coefficient is 0.3. Since 0.3 is not less than zero, it indicates that the lateral slip of the thin film is currently in a divergent state, facing the risk of continuous offset. Based on this determination, the previously obtained instantaneous impulse geometric deviation coefficient of 0.4 is algebraically added to the edge drift trend coefficient of 0.3, resulting in the j-th overall offset scale of 0.7. If the edge drift trend coefficient is less than zero, it indicates the existence of physical return inertia, and the overall offset scale is only equal to the instantaneous impulse geometric deviation coefficient to avoid repeated penalties. This conditional fusion logic ensures that a penalty term is only added when the offset condition worsens, and transient records are retained when the situation tends to stabilize, outputting a dimensionless value characterizing the basic gain requirement in the pure geometric dimension.

[0052] In step S24, after quantifying the geometric dimensions, the standard deviation of the data is calculated. The main purpose is to address the quantification of visual sensing interference caused by the ambient lighting environment and high-frequency reflections from the thin-film surface. The environmental interference data sequence collected by the monitoring station within a set time period is retrieved. This sequence consists of grayscale values ​​of surface reflected light intensity obtained from a laser displacement sensor. The standard deviation of these light intensity data over time is calculated, yielding a standard deviation of 35 for the current data variation. This calculation logic utilizes the standard deviation model in statistics to measure the dispersion of the light intensity signal around its average value; a larger dispersion indicates more severe fluctuations in the current ambient light and shadow. Through this calculation, the stability of the ambient lighting is transformed into a precise numerical parameter, enabling the successful capture and quantification of flicker or dynamic specular reflections. This provides direct data support for the subsequent objective evaluation of the visual sensor's imaging quality and signal-to-noise ratio.

[0053] In S25, the j-th optical confidence attenuation factor is obtained, aiming to address the issue of smooth confidence transition when optical noise causes visual feedback distortion. Specifically, an exponential attenuation function is used. The calculation logic is designed to obtain the j-th optical confidence attenuation factor in order to solve the problem of visual sensor data distortion caused by sudden high-frequency light and shadow fluctuations in industrial sites, and to avoid control oscillations caused by directly discarding data or using simple binarization judgment.

[0054] In the formula, Let be the j-th optical confidence attenuation factor, which is a dimensionless value and its value ranges from 1 to 2. between; It is an exponential function composed of natural constants; The standard deviation of the data variation in the environmental disturbance data sequence reflects the severity of light fluctuations. The set optical reference threshold is the tolerable background noise level; These are optical constants that characterize noise resistance.

[0055] Furthermore, the reason for adopting As a molecule, the difference is positive only when the actual light fluctuation exceeds the tolerance threshold, thus activating the attenuation mechanism. Dividing this difference by... This not only unified the grayscale dimensions, making it a dimensionless parameter, but also... The sensitivity of the attenuation was adjusted: The smaller the value, the more severe the decay.

[0056] Meanwhile, the outer layer adopts The form of the attenuation factor causes it to rise rapidly when the noise just exceeds the threshold. As the noise increases further, its value smoothly approaches 1 infinitely. This objectively reflects the physical process by which the reliability of visual sensors decreases non-linearly until it approaches zero (completely unreliable) when they are interfered with.

[0057] In a specific battery separator coating and winding scenario, the standard deviation of light intensity within the current time window is calculated. Set the optical reference threshold Grayscale, noise immunity constant Grayscale. Substituting into the formula, the numerator difference is... grayscale, ratio is After exponential operation, we get... .this The attenuation factor quantifies the unreliability of the current visual data, providing a precise basis for smoothly adjusting the dynamic correction threshold of the servo gain in the subsequent step S31, and preventing the control from blindly following the erroneous correction commands generated by noise.

[0058] In one specific implementation, S3 includes: S31, obtaining the visual servoing gain dynamic correction threshold. The j-th overall offset scale is used as the geometric reference scale value, and the product of the j-th overall offset scale and the j-th optical confidence attenuation factor, which exhibits exponential nonlinearity, is used as the feedforward adaptive compensation term. The geometric reference scale value and the feedforward adaptive compensation term are algebraically added to obtain a dimensionless correction multiplier as the visual servoing gain dynamic correction threshold. This visual servoing gain dynamic correction threshold is used as the feedforward intervention weight to positively amplify the mechanical wavefront displacement estimate when the optical confidence attenuates.

[0059] In this embodiment, it should be noted that the purpose of obtaining the visual servo gain dynamic correction threshold in S31 is to address the misadjustment problem caused by follower pseudo-deviations in the actuator when local optical imaging quality deteriorates. The previously calculated j-th overall offset scale, with a value of 0.7, is used as the geometric reference scale value. Subsequently, this reference scale value of 0.7 is multiplied by the j-th optical confidence attenuation factor, which exhibits exponential nonlinearity and has a value of 0.3935, to obtain a feedforward adaptive compensation term of approximately 0.27545. The geometric reference scale value of 0.7 is added to the feedforward adaptive compensation term of 0.27545 to obtain a correction multiplier of 0.97545, which is then used as the visual servo gain dynamic correction threshold. This calculation logic constructs an environment-adaptive weight adjustment mechanism: when enhanced optical interference causes the attenuation factor to increase, this correction threshold is proportionally amplified synchronously. It forms a data barrier at the algorithm level. By outputting the dynamic correction threshold, it guides the controller to actively increase the amplification factor of the basic deviation when the visual confidence deteriorates, and shifts the weight of the correction basis to the mechanical vibration feedforward end, thus ensuring the stability of the control under complex lighting conditions.

[0060] like Figure 4 As shown, in one specific embodiment, S4 includes: S41, obtaining the time delay and root mean square value of vibration wavefront propagation. This involves obtaining the real-time operating tension and linear density of the film material during the film transport operation, calculating the independent propagation velocity of the transverse mechanical elastic wave within the film medium; vector superimposing the film transport linear velocity and the independent propagation velocity in the same or opposite directions to obtain the equivalent wavefront propagation velocity of the flexible medium; dividing the physical span by the equivalent wavefront propagation velocity of the flexible medium to obtain the time delay of vibration wavefront propagation compensated for the flexible expansion and contraction characteristics. Vibration data within the historical window corresponding to the time delay in the vibration state data sequence is extracted, and the root mean square value of the vibration data is calculated.

[0061] S42. Obtain the estimated displacement of the j-th wavefront. Based on the calibrated natural frequency of the mechanical structure, construct a simple harmonic physical mapping model. Map the root mean square value from the dimension of acceleration to the dimension of spatial length to obtain the estimated displacement of the j-th wavefront. The specific formula is as follows:

[0062] In the formula, Let be the estimated displacement of the j-th wavefront, with the dimension being spatial length; is the transverse deformation transfer constant of the thin film, which is a dimensionless value; The root mean square value of the vibration data corresponding to the time delay is expressed in terms of acceleration. Pi is a constant. The calibrated mechanical structure has a natural frequency, measured in units of frequency.

[0063] The calibrated mechanical structure's inherent dominant frequency and the film's lateral deformation transfer constant were calculated based on modal impact experiments of the target guide roller mechanism, combined with historical test data of the film's Young's modulus. For example, a force hammer was used to strike the tension roller, and its decaying oscillation waveform was collected. The first-order resonant peak frequency was extracted using a fast Fourier transform, and the mechanical inherent dominant frequency was determined to be 10 Hz. Simultaneously, a 1 mm lateral static offset excitation was applied to the film at a baseline velocity, and the actual displacement measured by the downstream sensor was recorded as 1.2 mm. The ratio of the actual deformation to the theoretical excitation was calculated, and the average value was taken after 10 repeated experiments, calibrating the film's lateral deformation transfer constant to be 1.2.

[0064] S43. Obtain edge monitoring results. If the estimated displacement of the j-th wavefront is not greater than zero, obtain the edge monitoring results for the steady state. If the estimated displacement of the j-th wavefront is greater than zero, obtain the effective difference between the estimated displacement of the j-th wavefront and the set mechanical dead zone threshold; multiply the effective difference by the visual servo gain dynamic correction threshold, and output the target evaluation value that retains the absolute length dimension as the edge monitoring result. The specific formula is as follows:

[0065] In the formula, The target evaluation value is measured in terms of spatial length. This is a function to find the maximum value, used to output the larger of the two values ​​within parentheses; Let be the estimated displacement of the j-th wavefront, with the dimension being spatial length; The set mechanical dead zone threshold is measured in terms of spatial length. The target evaluation value is the dynamic correction threshold for visual servo gain, which is a dimensionless value. The final edge detection result is determined based on this target evaluation value.

[0066] The mechanical dead zone threshold is set based on the closed-loop resolution and transmission error of the servo actuator. It is obtained by analyzing historical background vibration data from a limited number of no-load operations of the equipment, combined with an assessment of the process tolerance threshold. For example, in the system's no-load state, the amplitude of disordered high-frequency vibration of the correction guide rail is continuously measured synchronously within 1 minute, collecting 600 background data points. Statistical analysis shows that the maximum mechanical backlash coupled with vibration displacement is approximately 0.18 mm. To avoid the servo motor frequently and ineffectively following minute, high-frequency background vibrations that do not affect quality, thus reducing lead screw wear, this maximum value is rounded and a small margin is added, determining the set mechanical dead zone threshold to be 0.2 mm.

[0067] In this embodiment, it should be noted that in S41, the time delay and root mean square (RMS) value of vibration wavefront propagation are obtained to address the engineering challenge of the delay in the physical deformation of the thin film as a continuous medium reaching the downstream station during transmission. Real-time operating tension and the linear density of the thin film material are acquired, and the independent propagation velocity of the transverse elastic wave is calculated. This velocity is then vector-superimposed with the current thin film transmission linear velocity of 5.0 m / s to obtain the equivalent wavefront propagation velocity. Using the pre-acquired physical span of 2.5 meters, this velocity is divided by the equivalent wavefront propagation velocity, resulting in a calculated time delay of 0.5 seconds to compensate for the flexible expansion and contraction characteristics. Based on this time delay parameter, the vibration acceleration data corresponding to a 0.5-second historical window in the vibration state data sequence is extracted and the RMS value of the vibration data within this historical period is calculated to be 1500 mm / s². This calculation logic clarifies the time required for upstream disturbances to propagate to the downstream target point. This time-delay aligned data tracing eliminates the asynchronous state of data between independent workstations, extracts the effective mechanical vibration energy that will affect downstream processes, and provides kinematic parameters that conform to physical causality for subsequent feedforward prediction.

[0068] In S42, the estimated j-th wavefront displacement is obtained to address the problem of accumulated drift error that easily arises when directly performing discrete calculus on high-frequency digital signals. Specifically, a simple harmonic physical mapping model is used. Obtaining the estimated displacement of the j-th wavefront is primarily to address the engineering challenge of accumulated drift errors arising from discrete calculus caused by high-frequency upstream vibrations in high-speed transmission films. In the formula, Let be the estimated displacement of the j-th wavefront, with the dimension being spatial length; is the transverse deformation transfer constant of the thin film, which is a dimensionless value and is related to the material of the thin film. The root mean square value of the vibration data corresponding to the time delay is expressed in terms of acceleration. Pi is a constant. The calibrated mechanical structure has a natural frequency, measured in units of frequency.

[0069] Furthermore, this computational logic abandons the existing practice of performing two consecutive discrete integrals of acceleration in the time domain, as this easily amplifies data noise and leads to severe cumulative errors in the displacement calculation results. Instead, it assumes the vibration to be simple harmonic motion with a stable dominant frequency, utilizing the frequency domain transformation relationship: displacement amplitude equals acceleration amplitude divided by angular frequency. The square of. Root mean square value. This represents the effective vibrational energy level within the corresponding time delay window. The acceleration dimension (e.g., Divide by the square of the angular frequency (dimension 1) The result is a strict dimensionality reduction mapping to spatial length units ( Finally, multiply by a constant. It is used to correct deformation transmission loss caused by the material's own tension or damping.

[0070] Based on the aforementioned battery separator scenario, the inherent dominant frequency of the upstream tension roller structure is calibrated as follows: The transverse deformation transfer constant is set to Extract the corresponding The root mean square value of vibration was calculated from historical window data of time delay. First, calculate the squared term of the angular frequency. .Will Divide by The basic displacement amplitude is approximately Multiply by the deformation constant Finally, the predicted wavefront displacement is obtained. This frequency domain mapping method, while ensuring strict dimensional uniformity, stably predicts the physical deformations that will occur at the downstream coating die.

[0071] In S43, the final edge monitoring result is obtained, resolving the technical problem of minute high-frequency mechanical vibrations being mistakenly identified as targets requiring correction, thus causing servo motor oscillation. Specifically, this is achieved through an expression... Output target evaluation values ​​to address the comprehensive evaluation issues caused by excessive mechanical wear due to minute high-frequency background vibrations in the servo mechanism response and the combined effects of multiple interference sources.

[0072] In the formula, The target evaluation value is measured in terms of spatial length. This is a function to find the maximum value, used to output the larger of the two values ​​within parentheses; Let be the estimated displacement of the j-th wavefront, with the dimension being spatial length; The set mechanical dead zone threshold is measured in terms of spatial length. The threshold for dynamic correction of visual servo gain is a dimensionless value.

[0073] Furthermore, set mechanical dead zones. The initial idea was that all industrial equipment has inherent background vibrations that cannot be completely eliminated and do not affect process quality. If the wavefront prediction is not filtered, it will lead to frequent and ineffective start-stop adjustments of the servo motor, increasing mechanical wear. The operation ensures that the effective difference value requiring compensation is extracted only when the predicted displacement exceeds the process tolerance threshold; if it does not exceed the threshold, the output is 0, maintaining the current state. After obtaining the effective difference value, it is then compared with the visual servo gain dynamic correction threshold. Multiplication. This correction threshold combines the geometric offset and the optical confidence attenuation factor, which increases as the visual signal deteriorates, thereby proportionally amplifying the weight of the feedforward prediction.

[0074] In practical applications, the predicted wavefront displacement has already been obtained. Set the mechanical dead zone threshold to... Calculate the difference The value is greater than 0, therefore it is extracted. The effective difference. Combined with the dynamic correction threshold obtained in the previous steps. (Based on a base offset of 0.7 and an optical attenuation factor of 0.3935), perform multiplication: Finally, the output is a quantity with absolute length dimensions. As the target evaluation value, this calculation logic, based on filtering out invalid jitter, comprehensively considers the weight allocation under visual interference and outputs a stable and dimensionlessly consistent feedforward correction command.

[0075] This invention also provides a visually compensated thin-film edge detection data processing system. The system is used to implement a visually compensated thin-film edge detection data processing method. The system includes: an acquisition module for acquiring the physical span between adjacent monitoring stations and the thin-film transmission linear velocity during thin-film transport operations; and acquiring environmental interference data sequences, image feature data sequences, and vibration state data sequences of the monitoring stations within a set time period; a data processing module for acquiring a dimensionless j-th overall offset scale based on the image feature data sequence of the j-th thin-film segment; acquiring the monitoring station where the j-th thin-film segment is located at the current moment, and processing the data based on the environmental interference data sequence of that monitoring station. The system acquires the j-th optical confidence attenuation factor; a coefficient calculation module is used to acquire the visual servo gain dynamic correction threshold of the j-th film segment at the current monitoring position based on the j-th optical confidence attenuation factor and the j-th overall offset scale; a trend monitoring module is used to acquire the film segment to be monitored upstream of the j-th film segment, and acquire the j-th wavefront displacement estimate transmitted from the film segment to be monitored to the current monitoring position of the j-th film segment based on the vibration state data sequence, physical span, and film transmission line velocity of the film segment to be monitored; and acquire the edge monitoring result of the film segment to be monitored based on the j-th wavefront displacement estimate and the visual servo gain dynamic correction threshold.

[0076] In one specific implementation, the data processing module is further configured to: perform a ratio calculation between the absolute distance in the image feature data sequence and a set tolerance physical threshold to obtain the dimensionless instantaneous impulse geometric deviation coefficient and edge drift trend coefficient; if the edge drift trend coefficient is less than zero, set the j-th overall offset scale to be equal to the instantaneous impulse geometric deviation coefficient; if the edge drift trend coefficient is not less than zero, set the j-th overall offset scale to be the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

[0077] In one specific implementation, the data processing module is further configured to: calculate the standard deviation of the data variation in the environmental interference data sequence; if the standard deviation does not exceed the set optical reference threshold, obtain the value of zero as the j-th optical confidence attenuation factor; if the standard deviation exceeds the optical reference threshold, construct an exponential decay function based on the difference between the standard deviation and the optical reference threshold, and obtain a continuous dimensionless value greater than zero and less than one as the j-th optical confidence attenuation factor through the exponential decay function.

[0078] In one specific implementation, the trend monitoring module is also used to: divide the physical span by the thin film transmission line velocity to obtain the time delay of the vibration wavefront propagation; extract the vibration data within the historical window corresponding to the time delay in the vibration state data sequence, and calculate the root mean square value of the vibration data; construct a simple harmonic physical mapping model based on the calibrated natural frequency of the mechanical structure, and map the root mean square value from the dimension of acceleration to the dimension of spatial length to obtain the estimated displacement of the j-th wavefront.

[0079] To further clarify the operating mechanism and physical quantification process of the technical solution of the present invention in actual industrial production lines, the following analysis will be conducted in detail on the underlying derivation logic of the optical distortion detection data processing method, using a high-frequency bending fatigue test scenario of a flexible folding screen cover containing specific physical parameters and test data.

[0080] In a specific lithium battery separator high-speed coating machine winding application scenario, the system begins by executing step S1 to obtain physical mapping parameters and timing sampling data. The physical span between the film segment to be monitored (upstream tension roller) and the monitoring station (downstream coating die) where the j-th film segment is located is set as... The current process setting for thin film transport linear speed is The system-defined sampling data time period is [length missing]. During this time sequence, the visual sensor, laser displacement sensor, and high-frequency vibration acceleration sensor simultaneously perform high-frequency discrete sampling, outputting image feature data sequences in millimeters and grayscale parameters, respectively. to The system consists of environmental disturbance data sequences in units of millimeters per square second and vibration state data sequences in units of millimeters per square second. These synchronized multi-source physical data form the input basis for subsequent discretized control algorithms.

[0081] Entering the feature quantization stage in step S2, the system first calculates the dimensionless j-th global offset scale. The unbiased baseline coordinates of the visual sensor are set as zero, and the system's set tolerance physical threshold is... In the set Within a given time period, the maximum absolute distance from the baseline in the image feature data sequence was measured as follows: The system extracts the difference between the maximum absolute distance and the tolerance physical threshold, i.e. Then, the difference is compared with the tolerance physical threshold, i.e. After eliminating the dimension of length, the geometric deviation coefficient of the instantaneous impulse is obtained as follows: Next, the system extracts the absolute distance of the last time point within that time period. The absolute distance at the initial moment is The difference between the two is The difference is then compared to the tolerance physical threshold. The dimensionless edge drift trend coefficient is obtained as follows: Due to the edge drift trend coefficient This indicates that the lateral slip is in a divergent state. Based on the judgment rules, the system sets the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient as the j-th overall offset scale, and calculates the overall offset scale as follows: .

[0082] After completing the dimensionality reduction calculation, the system continues with step S2 to obtain the attenuation state of the local optical imaging. The system retrieves the environmental interference data sequence acquired by the laser displacement sensor and calculates the standard deviation of the data variation caused by workshop lighting flicker and high-frequency vibration of the thin film surface. At this time, the optical reference threshold set within the system... And optical constants used to characterize the noise immunity of hardware Due to the actual standard deviation detected. Greater than the benchmark threshold The system is substituted with the exponential decay function. Perform the calculation. The specific calculation process is as follows: After simplification, it becomes The final value is obtained through exponential calculation using the natural constant, which represents the j-th optical confidence attenuation factor. .

[0083] Based on the derived geometric and optical dimensionless coefficients, the system proceeds to step S3 to calculate the dynamic correction parameters. The system then uses the j-th global offset scale obtained in the previous step. Set this as the base value. Then, scale the overall offset. The obtained j-th optical confidence attenuation factor Perform multiplication to obtain the product. This is then set as an additional value. By adding the base value to the additional value, the system obtains a dimensionless correction multiplier, specifically calculated as follows: The system will The numerical results are used as the dynamic correction threshold for visual servo gain. And update it in the buffer register. To establish a clear weight allocation mechanism, the total corrective control output of the system is defined as: The feedback adjustment amount obtained by the vision sensor Feedforward target evaluation value extracted by mechanical channel Superposition, that is Due to subsequent steps It is the wavefront displacement prediction quantity and The direct product of these values ​​reflects the system's output amplification under the current specific intensity of optical interference. (0.97545) directly increases the algebraic level. In the total control output The absolute proportion of the signal. This explicit multiplicative weighting transformation matrix allows the system to physically reduce the impact of distorted signals as visual confidence decays. This dependence enables deterministic amplification of feedforward intervention weights.

[0084] The system then executes step S4, initiating a feedforward prediction mechanism based on a physical model. The system extracts the real-time operating tension of the thin film using the field sensor bus. and the linear density of thin film materials Based on the transverse wave equation of a flexible string, the independent propagation wave velocity of the mechanoelastic wave within the thin-film medium is calculated. Combine it with the current The equivalent wavefront propagation velocity of the flexible medium is obtained by vector superposition of the thin film transmission linear velocities. If a basic rigid body model is used, the time delay is... However, by introducing flexible scaling parameters, the system utilizes the 2.5m physical span divided by the actual equivalent wavefront propagation velocity. The calculated wavefront time delay that accurately compensates for the flexible tension characteristics is... Based on this real spatiotemporal delay parameter, the system traces back along the historical timeline, extracts the vibration state data sequence within the historical sampling window prior to 0.036 seconds, calculates the root mean square value of the vibration data within this interval, and then calculates the equivalent vibration intensity. .

[0085] After obtaining the equivalent vibration intensity, the system uses the physical mapping formula. Obtain the estimated displacement of the j-th wavefront. The pre-calibrated natural frequency of the mechanical guide roller structure. Horizontal deformation transfer constant of thin film The system calculates the denominator. Using root mean square value Divide by the denominator term and then multiply by the constant. Calculate the predicted displacement of the j-th wavefront. .because If the value is greater than zero, the system determines that the corresponding edge monitoring result needs to be obtained and substituted into the formula. In the middle. Set the mechanical dead zone threshold. The system obtains the valid difference. Finally, this effective difference is compared with the visual servo gain dynamic correction threshold. Multiplication, that is Output the target evaluation value while retaining the absolute length dimension. .Should The value is output as the edge monitoring result and is used to drive the correction controller to perform precise displacement compensation.

[0086] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0087] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0088] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for processing thin film edge detection data based on visual compensation, characterized in that, The methods include: The physical span between adjacent monitoring stations and the linear velocity of the thin film during the thin film transport operation are obtained, and the thin film in the transport is divided into multiple continuous thin film segments along the transport direction. Acquire the environmental disturbance data sequence, image feature data sequence, and vibration state data sequence of the monitoring station within a set time period; Based on the image feature data sequence of the j-th thin film segment, obtain the dimensionless j-th global offset scale; Obtain the monitoring station where the j-th thin film segment is located at the current moment, and obtain the j-th optical confidence attenuation factor based on the environmental interference data sequence of the monitoring station; Based on the j-th optical confidence attenuation factor and the j-th overall offset scale, obtain the visual servo gain dynamic correction threshold of the monitoring station where the j-th thin film segment is located at the current moment; Obtain the membrane segment to be monitored that is upstream of the j-th membrane segment. Based on the vibration state data sequence of the membrane segment to be monitored, the physical span, and the membrane transmission line velocity, obtain the estimated j-th wavefront displacement of the membrane segment to be monitored to the monitoring position of the j-th membrane segment at the current moment. Based on the estimated j-th wavefront displacement and the visual servo gain dynamic correction threshold, the edge monitoring result of the thin film segment to be monitored is obtained.

2. The method for processing thin film edge detection data based on visual compensation according to claim 1, characterized in that, The step of obtaining the dimensionless j-th global offset scale based on the image feature data sequence of the j-th thin film segment includes: Extract the maximum absolute distance from the baseline in the image feature data sequence, calculate the difference between the maximum absolute distance and the set tolerance physical threshold, and perform a ratio operation between the difference and the tolerance physical threshold to obtain the dimensionless instantaneous impulse geometric deviation coefficient. Extract the absolute distance difference between the end time and the beginning time of the image feature data sequence within a set time period, and perform a ratio calculation between the absolute distance difference and the tolerance physical threshold to obtain a dimensionless edge drift trend coefficient; If the edge drift trend coefficient is less than zero, the j-th overall offset scale is set to be equal to the instantaneous impulse geometric deviation coefficient; If the edge drift trend coefficient is not less than zero, the j-th overall offset scale is set to the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

3. The method for processing thin film edge detection data based on visual compensation according to claim 1, characterized in that, The step of obtaining the j-th optical confidence attenuation factor based on the environmental interference data sequence of the monitoring station includes: Calculate the standard deviation of the data variation in the environmental disturbance data sequence; If the standard deviation does not exceed the set optical reference threshold, then the value of zero is obtained as the j-th optical confidence attenuation factor; If the standard deviation exceeds the optical reference threshold, an exponential decay function is constructed based on the difference between the standard deviation and the optical reference threshold. A continuous dimensionless value greater than zero and less than one is obtained through the exponential decay function as the j-th optical confidence decay factor.

4. The method for processing thin film edge detection data based on visual compensation according to claim 1, characterized in that, The step of obtaining the visual servo gain dynamic correction threshold of the monitoring position of the j-th thin film segment at the current moment based on the j-th optical confidence attenuation factor and the j-th overall offset scale includes: The j-th overall offset scale is used as the geometric reference scale value, and the product of the j-th overall offset scale and the j-th optical confidence attenuation factor, which has an exponential nonlinear characteristic, is used as the feedforward adaptive compensation term. The geometric reference ratio value is algebraically added to the feedforward adaptive compensation term to obtain a dimensionless correction multiplier as the visual servo gain dynamic correction threshold. The visual servo gain dynamic correction threshold is used as the feedforward intervention weight to positively amplify the mechanical wavefront displacement estimate when the optical confidence decreases.

5. The method for processing thin film edge detection data based on visual compensation according to claim 1, characterized in that, The step of obtaining the estimated j-th wavefront displacement of the monitored thin film segment at the current monitoring position based on the vibration state data sequence of the monitored thin film segment, the physical span, and the thin film transmission linear velocity includes: To obtain the real-time operating tension and linear density of the thin film material during thin film transport operations, and to calculate the independent propagation wave velocity of transverse mechanical elastic waves in the thin film medium; The equivalent wavefront propagation speed of the flexible medium is obtained by vector superimposing the thin film transmission line velocity and the independent propagation wave velocity in the same or opposite directions. Divide the physical span by the equivalent wavefront propagation velocity of the flexible medium to obtain the time delay of vibration wavefront propagation that compensates for the flexible stretching characteristics. Extract vibration data from the historical window corresponding to the time delay in the vibration state data sequence, and calculate the root mean square value of the vibration data. Based on the calibrated natural frequency of the mechanical structure, a simple harmonic physical mapping model is constructed, and the root mean square value is mapped from the dimension of acceleration to the dimension of spatial length to obtain the estimated value of the j-th wavefront displacement.

6. The method for processing thin film edge detection data based on visual compensation according to claim 1, characterized in that, The step of obtaining the edge monitoring result of the thin film segment to be monitored based on the estimated j-th wavefront displacement and the visual servo gain dynamic correction threshold includes: If the estimated displacement of the j-th wavefront is not greater than zero, obtain the edge monitoring result of the steady state. If the estimated displacement of the j-th wavefront is greater than zero, obtain the effective difference between the estimated displacement of the j-th wavefront and the set mechanical dead zone threshold. The effective difference is multiplied by the visual servo gain dynamic correction threshold, and the target evaluation value that retains the absolute length dimension is output as the edge detection result.

7. A thin film edge detection data processing system based on visual compensation, characterized in that, include: The acquisition module is used to acquire the physical span between adjacent monitoring stations and the linear velocity of the thin film transport during the thin film transport operation. The thin film in transmission is divided into multiple continuous thin film segments along the transmission direction; and the environmental interference data sequence, image feature data sequence, and vibration state data sequence of the monitoring station within a set time period are acquired. The data processing module is used to obtain the dimensionless j-th global offset scale based on the image feature data sequence of the j-th thin film segment; obtain the monitoring position of the j-th thin film segment at the current moment; and obtain the j-th optical confidence attenuation factor based on the environmental interference data sequence of the monitoring position. The coefficient calculation module is used to obtain the visual servo gain dynamic correction threshold of the monitoring position of the j-th thin film segment at the current moment based on the j-th optical confidence attenuation factor and the j-th overall offset scale. The trend monitoring module is used to acquire the monitored film segment located upstream of the j-th film segment, and based on the vibration state data sequence of the monitored film segment, the physical span, and the film transmission linear velocity, to acquire the estimated j-th wavefront displacement of the monitored film segment transmitted to the monitoring position of the j-th film segment at the current moment; and based on the estimated j-th wavefront displacement and the visual servo gain dynamic correction threshold, to acquire the edge monitoring result of the monitored film segment.

8. The visual compensation-based thin film edge detection data processing system according to claim 7, characterized in that, The data processing module is also used for: Extract the maximum absolute distance from the baseline in the image feature data sequence, calculate the difference between the maximum absolute distance and the set tolerance physical threshold, and perform a ratio operation between the difference and the tolerance physical threshold to obtain the dimensionless instantaneous impulse geometric deviation coefficient. Extract the absolute distance difference between the end time and the beginning time of the image feature data sequence within a set time period, and perform a ratio calculation between the absolute distance difference and the tolerance physical threshold to obtain a dimensionless edge drift trend coefficient; If the edge drift trend coefficient is less than zero, the j-th overall offset scale is set to be equal to the instantaneous impulse geometric deviation coefficient; If the edge drift trend coefficient is not less than zero, the j-th overall offset scale is set to the sum of the instantaneous impulse geometric deviation coefficient and the edge drift trend coefficient.

9. The visual compensation-based thin film edge detection data processing system according to claim 7, characterized in that, The data processing module is also used for: Calculate the standard deviation of the data variation in the environmental disturbance data sequence; If the standard deviation does not exceed the set optical reference threshold, then the value of zero is obtained as the j-th optical confidence attenuation factor; If the standard deviation exceeds the optical reference threshold, an exponential decay function is constructed based on the difference between the standard deviation and the optical reference threshold. A continuous dimensionless value greater than zero and less than one is obtained through the exponential decay function as the j-th optical confidence decay factor.

10. The visual compensation-based thin film edge detection data processing system according to claim 7, characterized in that: The coefficient calculation module is also used to calculate the first... The overall offset scale is used as the geometric reference ratio value, and its product with the optical confidence attenuation factor is used as the feedforward adaptive compensation term. The visual servo gain dynamic correction threshold is obtained by algebraic addition. The trend monitoring module is also used to acquire the real-time operating tension and linear density of the thin film and calculate the independent propagation wave velocity. This velocity is then superimposed with the thin film transmission linear velocity vector to obtain the equivalent wavefront propagation velocity. Based on this, the time delay compensated for the flexible stretching characteristics is obtained, and then combined with the simple harmonic physical mapping model to obtain the first... The estimated displacement of the wavefront.

Images