Method and system for dynamic evaluation of plastic film thickness uniformity based on machine vision

Abstract

The application discloses a kind of based on machine vision's plastic film thickness uniformity dynamic evaluation method and system, it is related to data processing technical field, method includes: obtaining the film image of at least three detection regions sequentially arranged along plastic film running direction, standard reference imaging area image, cumulative running displacement, tension, speed and temperature data;Generate film section identification and carry out the same film section identity confirmation;Obtain correction thickness correlation characteristic value, stable boundary and width state fingerprint;Irreversible fingerprint formation value, optical drift stripping value and process section abnormal increment value are combined, to determine thickness uniformity dynamic evaluation result.The application is used to process the problem that same section corresponding is not accurate across detection region, optical drift misjudgment and abnormal process section is difficult to locate, can be used to identify thickness uniformity anomaly, and distinguish abnormal formation section, abnormal expansion section and abnormal transmission section.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method and system for dynamic evaluation of the thickness uniformity of plastic films based on machine vision. Background Technology

[0002] During continuous production, the film passes through multiple production stations sequentially along the production line. The tension, running speed, temperature, rolling pressure, cooling, and imaging conditions at different stations can all affect the thickness distribution in the width direction of the film. For products such as food packaging films, electronic protective films, optical films, and composite films, thickness uniformity is related to the film's mechanical properties, barrier properties, light transmittance, lamination stability, and winding quality. Therefore, thickness uniformity can be evaluated online during the production process.

[0003] Existing methods for detecting plastic film thickness mainly include offline sampling inspection, single-point thickness measurement, single-frame image analysis, and online alarms based on fixed thresholds. Offline sampling inspection has limited ability to cover continuous production processes, and single-point thickness measurement is not suitable for reflecting local stripe anomalies or edge anomalies in the width direction. Single-frame image analysis can obtain visual data over a large range, but its results may be affected by light source fluctuations, imaging response changes, film tension changes, operating speed changes, and temperature changes, leading to image response changes caused by non-thickness factors being judged as thickness uniformity anomalies.

[0004] Furthermore, in a continuous production line, the same film segment will sequentially pass through at least three inspection areas set up along the direction of plastic film movement. If the data collected from different inspection areas cannot be mapped to the same film segment, the basis for cross-area comparison is insufficient, and judgment errors may occur due to film slippage, expansion and contraction, cumulative displacement errors, or deviations in the timing of data collection. After mapping the same film segment, comparing only the differences in individual thickness characteristics between adjacent inspection areas is not sufficient to determine whether the anomaly was newly formed in the current process segment, expanded in the current process segment, or was transmitted from an upstream process segment. Summary of the Invention

[0005] To address the technical problems existing in the background art, the present invention provides a machine vision-based dynamic evaluation method and system for the thickness uniformity of plastic films. This method maps data from different detection areas to the same film segment and combines the judgment of changes in the state of sub-regions in the width direction, optical drift peeling, and abnormal increments in the process segment to dynamically evaluate the thickness uniformity of the plastic film. This method addresses the problems in existing online plastic film inspection where it is inconvenient to map data from different detection areas to the same film segment, single-frame image analysis is affected by optical drift and changes in operating conditions, local anomaly identification in the width direction is unstable, and it is difficult to distinguish between anomaly formation segments, anomaly expansion segments, and anomaly transmission segments.

[0006] A machine vision-based dynamic evaluation method for the thickness uniformity of plastic films includes: acquiring film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data for at least three detection areas sequentially arranged along the film's running direction; generating film segment identifiers based on the cumulative film displacement data, and verifying the identity of film segments in different detection areas according to the relative optical response order of sub-regions in each width direction in the film image data; and obtaining the corrected thickness correlation features of the same film segment in each detection area and each sub-region in each width direction based on the film image data, standard film sample calibration results, and operating condition data. The system calculates the following values: a stable boundary is determined based on the dynamic stable thin film segment library; a width state fingerprint is generated based on the corrected thickness-related feature value and the stable boundary; an irreversible fingerprint formation value is obtained based on the width state fingerprints of the same thin film segment in the upstream, downstream, and further downstream detection regions; an optical drift stripping value is obtained based on the direction of change of the corrected thickness-related feature value of the same thin film segment in adjacent detection regions and the direction of change of the reference optical response in the standard reference imaging region; and a dynamic evaluation result of thickness uniformity, including anomalous formation segments, anomalous extension segments, or anomalous transmission segments, is determined based on the irreversible fingerprint formation value, the optical drift stripping value, and historical irreversible fingerprint formation values.

[0007] Optionally, the operating condition data includes film tension data, film running speed data, and film temperature data; generating a film segment identifier based on the cumulative film running displacement data includes: reading the cumulative film running displacement data corresponding to the current acquisition time; calculating the ratio between the cumulative film running displacement data and a preset film segment length, and performing a floor function on the ratio calculation result; and using the floor function result as the film segment identifier corresponding to the current acquisition time.

[0008] Optionally, based on the relative optical response order of each width-direction sub-region in the thin film image data, the thin film segments in different detection areas are identified as the same thin film segment. This includes: dividing the thin film image data into multiple width-direction sub-regions along the width direction of the plastic film, and obtaining the original optical response value of each width-direction sub-region; generating an initial visual width state fingerprint according to the arrangement order of the original optical response values ​​of each width-direction sub-region; sequentially matching the initial visual width state fingerprint of the thin film segment to be identified in the downstream detection area with the initial visual width state fingerprint of the candidate thin film segment in the upstream detection area; if the two meet the preset matching conditions, they are identified as belonging to the same thin film segment; if there is an overall misalignment between the two, the thin film segment identification is corrected according to the misalignment direction; if the two do not meet the preset matching conditions, the thin film segment to be identified is marked as having an untrusted identity.

[0009] Optionally, based on thin film image data, standard thin film sample calibration results, and operating condition data, the corrected thickness-related feature values ​​for the same thin film segment in each detection area and each width-direction sub-region are obtained. This includes: based on the thin film image data of the width-direction sub-region, overexposed pixels, underexposed pixels, and edge-occluded pixels are removed, and the original optical response representative value is obtained based on the optical response value corresponding to the remaining effective pixels; based on the standard thin film sample calibration results, the original optical response representative value is converted into the original thickness-related feature value; based on the thin film tension data, the tension compensation rule is queried to obtain the tension compensation amount; based on the thin film running speed data, the speed compensation rule is queried to obtain the speed compensation amount; based on the thin film temperature data, the temperature compensation rule is queried to obtain the temperature compensation amount; based on the original thickness-related feature value, the tension compensation amount, the speed compensation amount, and the temperature compensation amount, the corrected thickness-related feature value is obtained; wherein the tension compensation amount, the speed compensation amount, and the temperature compensation amount are all thickness-related feature compensation amounts that have been pre-converted to the same unit as the original thickness-related feature value through calibration.

[0010] Optionally, the stability boundary is determined based on the dynamic stable film segment library, including: selecting film segments that continuously pass through at least two detection areas as candidate stable film segments; if a candidate stable film segment is not marked as unreliable in identity, the corresponding standard reference imaging area is not marked as unreliable in measurement, and the corresponding width state fingerprint does not show irreversible thickness uniformity anomalies, then the candidate stable film segment is added to the dynamic stable film segment library; for the same detection area and the same width direction sub-region, the corresponding correction thickness associated feature values ​​in the dynamic stable film segment library are arranged in numerical order; when the number of correction thickness associated feature values ​​after arrangement is not less than four, the first and last values ​​are removed, and the first value of the remaining sequence is used as the first stable boundary, and the last value of the remaining sequence is used as the second stable boundary; when the number of correction thickness associated feature values ​​after arrangement is less than four, the first and second stable boundaries corresponding to the previous acquisition cycle are used, or the initial first and initial second stable boundaries obtained from the standard film sample calibration stage are read.

[0011] Optionally, generating a width state fingerprint based on the corrected thickness-related feature value and the stable boundary includes: for each width-direction sub-region of the film segment to be evaluated in each detection area, reading the corresponding corrected thickness-related feature value, the first stable boundary, and the second stable boundary; if the corrected thickness-related feature value is less than the first stable boundary, then marking the corresponding width-direction sub-region as a thinner-side abnormal state; if the corrected thickness-related feature value is greater than or equal to the first stable boundary and less than or equal to the second stable boundary, then marking the corresponding width-direction sub-region as a stable state; if the corrected thickness-related feature value is greater than the second stable boundary, then marking the corresponding width-direction sub-region as a thicker-side abnormal state; and arranging the thinner-side abnormal state, the stable state, and the thicker-side abnormal state in sequence according to the arrangement order of the width-direction sub-regions in the width direction of the plastic film to form a width state fingerprint of the film segment to be evaluated in the corresponding detection area.

[0012] Optionally, based on the width state fingerprints of the same film segment in the upstream detection region, downstream detection region, and further downstream detection region, an irreversible fingerprint formation value is obtained, including: reading the width state fingerprints corresponding to the width-direction sub-regions that can complete the matching of the same film segment; counting the number of width-direction sub-regions that are stable in the upstream detection region, become a thinner-side or thicker-side anomalous in the downstream detection region, and continue to maintain the thinner-side or thicker-side anomalous in the further downstream detection region, to obtain the number of single-region continuous anomalies; counting the number of consecutive width-direction sub-regions that are adjacent stable in the upstream detection region, become adjacent same-side anomalous in the downstream detection region, and continue to maintain the adjacent same-side anomalous in the further downstream detection region, to obtain the number of consecutive anomalies; and obtaining the irreversible fingerprint formation value based on the single-region continuous anomaly number, the consecutive anomaly number, and the number of width-direction sub-regions that can complete the matching of the same film segment.

[0013] Optionally, based on the irreversible fingerprint formation value, the optical drift stripping value, and the historical irreversible fingerprint formation value, the dynamic evaluation result of thickness uniformity including anomalous formation segments, anomalous extension segments, or anomalous transmission segments is determined, including: based on the change direction of the corrected thickness correlation feature value of each width-direction sub-region in the upstream and downstream detection regions of the same film segment, counting the number of width-direction sub-regions whose change direction of the corrected thickness correlation feature value is consistent with the change direction of the reference optical response of the standard reference imaging region, and obtaining the optical drift stripping value based on the number of width-direction sub-regions with consistent direction and the number of width-direction sub-regions that can complete the matching of the same film segment; based on the irreversible fingerprint formation value between the current adjacent detection regions... The system obtains the process segment abnormal increment value by combining the historical irreversible fingerprint formation value with the current adjacent detection area. If the optical drift stripping value meets the optical measurement drift condition, the corresponding width state fingerprint change is determined as optical measurement drift. If the process segment abnormal increment value is greater than zero, and the abnormal state corresponding to the irreversible fingerprint formation value continues to exist in the downstream detection area, the production line process segment between the current adjacent detection areas is determined as an abnormal formation segment or an abnormal expansion segment. If the irreversible fingerprint formation value is greater than zero and the process segment abnormal increment value is zero, the production line process segment between the current adjacent detection areas is determined as an abnormal transmission segment. The system outputs a dynamic evaluation result of thickness uniformity that includes the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment.

[0014] A machine vision-based dynamic evaluation system for the thickness uniformity of plastic films is also provided, comprising: a data acquisition module for acquiring film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data of at least three detection areas sequentially arranged along the running direction of the plastic film; a film segment identification module for generating film segment identifiers based on the cumulative film displacement data, and identifying film segments in different detection areas as the same film segment based on the relative optical response order of each width direction sub-region in the film image data; and a feature processing module for acquiring the correction of the same film segment in each detection area and each width direction sub-region based on the film image data, standard film sample calibration results, and operating condition data. The system includes a thickness-related feature value module; a fingerprint generation module for determining stable boundaries based on a dynamic stable thin film segment library, and generating width state fingerprints based on the corrected thickness-related feature value and the stable boundaries; and an evaluation and positioning module for obtaining irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream, downstream, and further downstream detection regions, obtaining optical drift stripping values ​​based on the change direction of the corrected thickness-related feature value of the same thin film segment in adjacent detection regions and the change direction of the reference optical response in the standard reference imaging region, and determining the dynamic evaluation result of thickness uniformity including abnormal formation segments, abnormal extension segments, or abnormal transmission segments based on the irreversible fingerprint formation values, the optical drift stripping values, and historical irreversible fingerprint formation values.

[0015] Optionally, the data acquisition module includes a visual data acquisition unit, a reference data acquisition unit, a cumulative displacement data acquisition unit, a film tension data acquisition unit, a film running speed data acquisition unit, and a film temperature data acquisition unit. The visual data acquisition unit is used to acquire film image data. The reference data acquisition unit is used to acquire standard reference imaging area image data and acquire the reference optical response change direction of the standard reference imaging area based on the standard reference imaging area image data. The cumulative displacement data acquisition unit is used to acquire cumulative film running displacement data. The film tension data acquisition unit is used to acquire film tension data. The film running speed data acquisition unit is used to acquire film running speed data. The film temperature data acquisition unit is used to acquire film temperature data. The evaluation and positioning module is also used to determine the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment based on the irreversible fingerprint formation value, optical drift peeling value, and abnormal increment value of the process segment, and send the dynamic evaluation results of thickness uniformity including the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment to the result display module, the communication module, and the production line control equipment.

[0016] The beneficial effects of this invention are reflected in:

[0017] In a machine vision-based dynamic evaluation method for the thickness uniformity of plastic films, film segment identifiers are obtained by accumulating film displacement data and pre-set film segment lengths. The identity of the same film segment is confirmed using an initial visual width state fingerprint, ensuring that data collected from different detection areas at different times corresponds to the same film segment, reducing the probability of mismatches caused by film slippage, stretching, or accumulated displacement deviations. Original thickness-related feature values ​​are obtained through standard film sample calibration results, and corrected thickness-related feature values ​​are obtained through film tension data, film running speed data, and film temperature data, compensating for changes in visual response before thickness uniformity judgment. A dynamic stable film segment library is used to obtain first and second stable boundaries, ensuring that the judgment boundary for width state fingerprints originates from stable film segments in the current production process, reducing reliance on fixed empirical thresholds. Irreversible fingerprint formation values ​​are used to statistically analyze width state changes from upstream stability, downstream new anomalies, and further downstream persistence. Combined with the anomaly formation on the same side of adjacent width direction sub-regions, local persistent anomalies and strip-shaped continuous anomalies are recorded. By using optical drift stripping values ​​to determine whether the direction of change of the correction thickness-related characteristic value is consistent with the direction of change of the reference optical response in the standard reference imaging area, the probability of misjudgment caused by optical measurement drift is reduced. By comparing the irreversible fingerprint formation value between the current adjacent detection area and the historical detection area with the abnormal increment value of the process segment, the dynamic evaluation result of thickness uniformity is located as an abnormal formation segment, an abnormal expansion segment, or an abnormal transmission segment, providing data basis for production line control. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is a schematic diagram illustrating the steps of the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness in this invention.

[0020] Figure 2 This is a schematic diagram of a portion of step S1 in the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness of the present invention.

[0021] Figure 3 This is a schematic diagram of a portion of step S2 in the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness of the present invention.

[0022] Figure 4 This is a schematic diagram of part of step S3 in the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness of the present invention.

[0023] Figure 5 This is a schematic diagram of a portion of step S4 in the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness of the present invention.

[0024] Figure 6 This is a partial structural flowchart of the machine vision-based dynamic evaluation system for plastic film thickness uniformity according to the present invention.

[0025] Figure 7 This is a time-series diagram of the same film segment across the detection area in the machine vision-based dynamic evaluation method for plastic film thickness uniformity of the present invention. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] Similar reference numerals and letters in the accompanying drawings indicate similar items, and reference numerals already described will not be repeated in subsequent drawings. Furthermore, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] This invention provides a machine vision-based method for dynamically evaluating the thickness uniformity of plastic films, such as... Figure 1 As shown, in one embodiment, the method includes:

[0030] S1. Acquire film image data, standard reference imaging area image data, cumulative film displacement data, film tension data, film speed data, and film temperature data of at least three detection areas set sequentially along the running direction of the plastic film; obtain film segment identifiers based on the cumulative film displacement data and the preset film segment length; divide the film image data into multiple width direction sub-regions along the width direction of the plastic film; obtain the initial fingerprint of the visual width state based on the relative optical response order of each width direction sub-region, and verify the identity of the same film segment by identifying the film segment identifiers;

[0031] S2. Obtain the original thickness-related feature value based on the thin film image data of the width-direction sub-region and the calibration results of the standard thin film sample; obtain the corrected thickness-related feature value based on the thin film tension data, thin film running speed data, and thin film temperature data; obtain the first stable boundary and the second stable boundary based on the dynamic stable thin film segment library; obtain the width state fingerprint based on the corrected thickness-related feature value, the first stable boundary, and the second stable boundary;

[0032] S3. Obtain irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream detection region, downstream detection region, and further downstream detection region; obtain optical drift stripping values ​​based on the change direction of the correction thickness-related feature values ​​of the same thin film segment in the upstream and downstream detection regions and the change direction of the reference optical response in the standard reference imaging region.

[0033] S4. Obtain the process segment abnormal increment value based on the irreversible fingerprint formation value and the historical irreversible fingerprint formation value before the current adjacent detection area; determine the thickness uniformity dynamic evaluation result based on the irreversible fingerprint formation value, optical drift peeling value and process segment abnormal increment value; when the corresponding judgment rule is met, determine the production line process segment between the current adjacent detection areas as an abnormal formation segment, abnormal expansion segment or abnormal transmission segment, and output the thickness uniformity dynamic evaluation result.

[0034] In this embodiment, the plastic film moves continuously along the production line's running direction, and at least three detection areas are sequentially arranged along this direction. These at least three detection areas include an upstream detection area, a downstream detection area, and a further downstream detection area. Film image data refers to image data acquired from the plastic film's detection areas. Standard reference imaging area image data refers to image data acquired from the standard reference imaging area, used to obtain the direction of change in the reference optical response of the standard reference imaging area. Cumulative film displacement data refers to the cumulative distance the plastic film has moved along the production line's running direction from a preset starting position. Operating condition data includes film tension data, film running speed data, and film temperature data, used to process the influence of changes in film tension, film running speed, and film temperature on thickness-related characteristics.

[0035] The preset starting position is determined based on fixed mechanical reference points on the production line, preferentially selecting the tangent position of the traction roller at the front end of the inlet detection area or the encoder zeroing position, and is written into the system parameter table during equipment debugging. For example, setting the tangent position of the traction roller 120mm in front of the inlet detection area as the preset starting position means that after the encoder is zeroed at this position, all subsequent accumulated film displacement data will start from this position.

[0036] In this embodiment, the thickness-related feature value is not the absolute thickness value of the plastic film, but a visually related feature value used to characterize the thickness change of the plastic film, obtained by converting the film image data through standard film sample calibration. Since film tension, film running speed, and film temperature can all alter the optical response in the film image data, this invention compensates and corrects the original thickness-related feature value based on film tension data, film running speed data, and film temperature data to obtain a corrected thickness-related feature value. The width-direction sub-region refers to multiple regions obtained by dividing the film image data along the width direction of the plastic film. The initial visual width state fingerprint refers to the initial sequence feature generated based on the relative optical response order of each width-direction sub-region, used to identify the same film segment. The width state fingerprint refers to the state sequence formed by sequentially arranging the corresponding thinner-side abnormal state, stable state, or thicker-side abnormal state of each width-direction sub-region according to their arrangement order along the width direction of the plastic film.

[0037] In this embodiment, it should be noted that in S1, the system first acquires data from at least three detection areas sequentially arranged along the plastic film's running direction and establishes a dual confirmation relationship between the film segment identifier and the initial fingerprint of the visual width state. Taking the production line with an entrance detection area, a processing detection area, and a cooling detection area as an example, if the cumulative film running displacement data is 1240mm and the preset film segment length is 20mm, then the film segment identifier corresponding to the current acquisition time is obtained by rounding down the ratio of the cumulative film running displacement data to the preset film segment length, which corresponds to the 62nd film segment. In the entrance detection area, the system divides the film image data into 8 width-direction sub-regions along the plastic film width direction and acquires the original optical response values ​​of each width-direction sub-region, for example, 128.4, 130.1, 132.6, 133.2, 131.8, 129.7, 127.9, and 126.5 respectively. The system forms the initial fingerprint of the visual width state based on the arrangement order of these original optical response values. When the same film segment arrives at the processing and inspection area, an initial visual width fingerprint is generated again and sequentially matched with the initial visual width fingerprint of the entrance inspection area. If the arrangement order of the two width-direction sub-regions meets the preset matching conditions, they are confirmed to belong to the same film segment; if there is an overall misalignment, the film segment identification is corrected; if the matching conditions are not met, the segment is marked as unreliable. This step allows cross-inspection comparisons to be based on the same film segment, reducing the impact of film slippage, stretching, or cumulative displacement errors on the judgment results.

[0038] In S2, the system converts the thin film image data into corrected thickness-related feature values ​​that can be used for dynamic evaluation of thickness uniformity, and further generates a width state fingerprint. Taking a thin film segment identified as 62 as an example, the system first removes overexposed, underexposed, and edge-occluded pixels from each width-direction sub-region, and then obtains the original optical response representative value. If the original optical response representative value of a certain width-direction sub-region is 132.6, the system converts it into an original thickness-related feature value, such as 0.018 μm, based on the calibration results of a standard thin film sample. If the thin film tension data, thin film running speed data, and thin film temperature data corresponding to the same thin film segment in the detection area are 46 N, 38 m / min, and 42 °C, the system queries the tension compensation rule, speed compensation rule, and temperature compensation rule respectively to obtain a thickness-related feature compensation amount with the same unit as the original thickness-related feature value, and then obtains the corrected thickness-related feature value. The system then reads and sorts multiple corrected thickness-related feature values ​​of the same detection area and the same width-direction sub-region from the dynamically stable thin film segment library to form the first stable boundary and the second stable boundary. If the first stable boundary of a certain width-direction sub-region is -0.015 μm, the second stable boundary is 0.016 μm, and the corrected thickness correlation feature value of the film segment to be evaluated is 0.021 μm, then this width-direction sub-region is marked as an anomalous state with a thicker side. The system performs the same processing on all width-direction sub-regions to form a width-state fingerprint. This step, through calibration, compensation, dynamic boundary, and state processing, transforms the image response into a state sequence that can be compared across detection regions.

[0039] In S3, the system obtains irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream, downstream, and further downstream detection regions. It also obtains optical drift stripping values ​​based on the direction of change of the correction thickness-related feature values ​​of the same thin film segment in the upstream and downstream detection regions, as well as the direction of change of the reference optical response in the standard reference imaging region. Taking a thin film segment identified as 62 as an example, the width state fingerprint of the entry detection region is [0,0,0,0,0,0,0,0], the width state fingerprint of the processing detection region is [0,0,1,1,0,0,0,0], and the width state fingerprint of the cooling detection region is [0,0,1,1,0,0,0,0]. At this time, the 3rd and 4th width-direction sub-regions are in a stable state in the upstream detection region, become an aberrant state on the thicker side in the downstream detection region, and continue to maintain the aberrant state on the thicker side in the further downstream detection region, with both being adjacent aberrant states on the same side. The system counts the number of persistent and consecutive anomalies in a single region, and then obtains the irreversible fingerprint formation value based on the number of width-direction sub-regions that can complete the matching of the same thin film segment. Simultaneously, the system reads the direction of change of the correction thickness-related characteristic values ​​of each width-direction sub-region in the entry detection area and the processing detection area, and reads the direction of change of the reference optical response in the standard reference imaging area. If the direction of change of the correction thickness-related characteristic values ​​of most width-direction sub-regions is consistent with the direction of change of the reference optical response, it indicates that the change is more likely to originate from optical measurement drift; if only a localized anomalous sub-region is inconsistent with the direction of change of the reference optical response, and persists in the downstream detection area, it indicates that the change is more likely to correspond to an irreversible thickness uniformity anomaly.

[0040] In S4, the system obtains the process segment anomaly increment value based on the irreversible fingerprint formation value and the historical irreversible fingerprint formation value before the current adjacent detection area, and determines the dynamic evaluation result of thickness uniformity based on the irreversible fingerprint formation value, optical drift stripping value, and process segment anomaly increment value. Taking the area from the entrance detection area to the processing detection area as the current adjacent detection area as an example, if the current irreversible fingerprint formation value is 0.25, and there is no historical irreversible fingerprint formation value before the current adjacent detection area, then 0.25 is taken as the process segment anomaly increment value. If the optical drift stripping value does not meet the optical measurement drift condition, and the abnormal state corresponding to the irreversible fingerprint formation value continues to exist in the cooling detection area, then the production line process segment between the entrance detection area and the processing detection area is determined as an abnormal formation segment. If the irreversible fingerprint formation value between the processing detection area and the cooling detection area further increases, for example, from 0.25 to 0.375, then the production line process segment between the processing detection area and the cooling detection area is determined as an abnormal extension segment. If the subsequent detection area still maintains the same width of fingerprint state, but the process segment anomaly increment value is 0, then the subsequent production line process segment is determined as an abnormal transmission segment. This step further implements the dynamic assessment results of thickness uniformity into the production line process, enabling the system to distinguish between anomaly formation, anomaly propagation, and anomaly transmission.

[0041] In summary, the machine vision-based dynamic evaluation method for plastic film thickness uniformity firstly obtains film segment identifiers by accumulating film displacement data and pre-set film segment lengths. The same film segment is then identified using an initial visual width state fingerprint, ensuring that data collected from different detection areas at different times corresponds to the same film segment, reducing mismatches caused by film slippage, stretching, or accumulated displacement deviations. Original thickness-related feature values ​​are obtained through standard film sample calibration results, and corrected thickness-related feature values ​​are obtained using film tension data, film running speed data, and film temperature data, compensating for changes in visual response before thickness uniformity judgment. A dynamic stable film segment library is used to obtain first and second stable boundaries, ensuring that the width state fingerprint judgment boundary originates from stable film segments in the current production process, reducing reliance on fixed empirical thresholds. Irreversible fingerprint formation values ​​are used to statistically analyze upstream stability, downstream new anomalies, and further downstream persistent width state changes. Combined with the anomaly formation on the same side of adjacent width direction sub-regions, local persistent anomalies and strip-shaped continuous anomalies are recorded. By using optical drift stripping values ​​to determine whether the direction of change of the correction thickness-related characteristic value is consistent with the direction of change of the reference optical response in the standard reference imaging area, the probability of misjudgment caused by optical measurement drift is reduced. By comparing the irreversible fingerprint formation value between the current adjacent detection area and the historical detection area with the abnormal increment value of the process segment, the dynamic evaluation result of thickness uniformity is located as an abnormal formation segment, an abnormal expansion segment, or an abnormal transmission segment, providing data basis for production line control.

[0042] like Figure 2 As shown, in one specific embodiment, S1 includes:

[0043] S11. Acquire data from at least three detection areas;

[0044] S12. Obtain the film segment identifier based on the cumulative film displacement data and the preset film segment length;

[0045] S13. Based on the initial fingerprint of the visual width state, the identity of the same thin film segment is confirmed.

[0046] In this embodiment, it should be noted that in S11, film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data are acquired in at least three detection areas sequentially arranged along the running direction of the plastic film. The operating condition data includes film tension data, film running speed data, and film temperature data. Film image data is used to acquire the optical response of each width-direction sub-region; standard reference imaging area image data is used to acquire the direction of change of the reference optical response of the standard reference imaging area; cumulative film displacement data is used to acquire film segment identifiers; and operating condition data is used to acquire correction thickness-related feature values. Within the same detection area, film image data and standard reference imaging area image data correspond to the same acquisition time. Operating condition data is matched with film image data according to the acquisition time. If the acquisition frequencies of various data types are different, they are matched according to the acquisition time so that the image data and operating condition data of the same film segment correspond within the same detection area. For example, if the inlet detection area, processing detection area, and cooling detection area all acquire thin film image data with a sampling period of 100ms, and the thin film tension data is acquired with a sampling period of 50ms, then the system selects the tension data closest to the thin film image acquisition time as the thin film tension data of that thin film segment in the detection area.

[0047] The thin-film image data sampling period is determined based on the thin-film running speed and the preset thin-film segment length, ensuring that the thin-film movement distance between adjacent sampling times does not exceed one preset thin-film segment length. For example, if the preset thin-film segment length is 20mm and the thin-film running speed is 200mm / s, then the time required for the thin-film to move 20mm is 0.1s. The system can set the thin-film image data sampling period to 0.1s or less.

[0048] The sampling period for operating condition data is determined based on the response speed of the operating condition sensors and the image acquisition period, and is usually set to be no greater than the sampling period for thin film image data to reduce timing matching errors. For example, if the thin film image data sampling period is 0.1s, and the tension sensor, speed sensor, and temperature sensor all support sampling at 0.02s, then the system will set the operating condition data sampling period to 0.02s and select the operating condition data closest to the image acquisition time.

[0049] In step S12, the cumulative film displacement data corresponding to the current acquisition time is read; the ratio of the cumulative film displacement data to the preset film segment length is calculated, and the ratio result is rounded down; the rounded result is used as the film segment identifier corresponding to the current acquisition time. The preset film segment length is used to divide the continuous plastic film into multiple film segments arranged along the running direction. The cumulative film displacement data and the preset film segment length use the same length unit and can be compared. Through the film segment identifier, data located in different detection areas can be established as corresponding to the same film segment. For example, if the cumulative film displacement data is 1240mm, the preset film segment length is 20mm, the ratio is 62, and the film segment identifier after rounding down is 62. If the cumulative film displacement data is 1258mm, the ratio is 62.9, and after rounding down it still corresponds to the film segment identifier 62, it indicates that the acquisition time is still within the same film segment range.

[0050] The preset film segment length is determined based on the imaging length of the detection area, the film running speed, and the camera acquisition cycle, ensuring that the same film segment can be completely distinguished within adjacent acquisition cycles. When determining the value, first obtain the effective imaging length of the camera along the running direction, then select an integer length not greater than this effective imaging length and write it into the system parameter table. For example, if the effective imaging length of the camera is 25mm, the film running speed is 200mm / s, and the acquisition cycle is 0.1s, then the running distance per cycle is 20mm, and the preset film segment length can be set to 20mm.

[0051] In step S13, the film image data is divided into multiple width-direction sub-regions along the width direction of the plastic film, and the original optical response value of each width-direction sub-region is acquired. The original optical response value can be derived from grayscale response, transmittance response, or reflection response, but the same optical response acquisition method is used in the same detection area. An initial visual width state fingerprint is generated according to the arrangement order of the original optical response values ​​of each width-direction sub-region. The initial visual width state fingerprint includes the width-direction sub-region number sequence obtained by arranging the width-direction sub-regions in ascending order of their original optical response values. The system sequentially matches the initial visual width state fingerprint of the film segment to be confirmed in the downstream detection area with the candidate film segment corresponding to the current film segment identifier and the candidate film segment corresponding to the adjacent film segment identifier in the upstream detection area. If the number of consecutive identical width-direction sub-region numbers in the two initial visual width state fingerprints reaches a preset matching number, or the proportion of consecutive identical width-direction numbers reaches a preset matching proportion, then the two are considered to meet the preset matching conditions. The preset matching number or preset matching proportion is determined by the calibration data when the same film segment passes through adjacent detection areas under normal production conditions and is written into the system parameter table. If the candidate film segment corresponding to the current film segment identifier does not meet the preset matching conditions, but meets the preset matching conditions with the candidate film segment corresponding to the adjacent film segment identifier, then an overall misalignment is identified, and the film segment identifier of the film segment to be confirmed is corrected to a candidate film segment identifier that meets the preset matching conditions. If the preset matching conditions are not met within the search range of candidate film segment identifiers, the film segment to be confirmed is marked as having an untrusted identity. Film segments with untrusted identities are not included in the calculation of irreversible fingerprint formation values ​​between adjacent detection areas.

[0052] The number of width-direction sub-regions is determined based on the film width, camera lateral resolution, and minimum required strip width. The division ensures that the actual width covered by each width-direction sub-region does not exceed the minimum required strip width. For example, if the film width is 800mm, the camera has 1600 effective lateral pixels, and the minimum required strip width is 100mm, then the area can be divided into 8 width-direction sub-regions, each corresponding to 100mm and 200 lateral pixels.

[0053] The method for obtaining the original optical response is determined based on the light transmittance of the thin film material and the detection optical path. During the equipment calibration stage, the distinguishability of grayscale response, transmittance response, and reflection response for samples of known thickness is tested respectively, and a method with monotonic response change and good repeatability is selected. For example, transmittance responses are collected for standard thin film samples of 18μm, 20μm, and 22μm respectively. If the transmittance response decreases steadily with increasing thickness and the repeatability error is less than 2%, the transmittance response is selected as the original optical response value.

[0054] The preset matching quantity and preset matching ratio are determined through a limited number of same-segment calibration data under normal production conditions. During calibration, multiple sets of initial visual width state fingerprints confirmed to belong to the same film segment are collected, and the number and proportion of relatively consistent sequential order are statistically analyzed. The lower statistical value in the stable samples is taken as the lower limit of matching. For example, if 50 sets of same-segment samples are collected, and the minimum number of relatively consistent sequential order in the 8 width direction sub-regions is 6, corresponding to a consistency ratio of 75%, then the preset matching quantity can be set to 6, and the preset matching ratio can be set to 75%.

[0055] The search range for candidate film segment identifiers is determined based on the cumulative displacement error, film slippage, and preset film segment length. Typically, the current film segment identifier and one or two adjacent film segment identifiers before and after it are selected. For example, if the maximum cumulative displacement error of the encoder is 12mm and the preset film segment length is 20mm, and the error is less than the length of one film segment, then the search range for candidate film segment identifiers can be set as the current film segment identifier, the previous film segment identifier, and the next film segment identifier.

[0056] In S1, the system can also timestamp the data for each detection area and use the film segment identifier, detection area name, acquisition time, and width-direction sub-region number as a data index. Using this index, subsequent steps read image data, status fingerprint data, and compensation correction data for the same film segment in different detection areas. If a film segment cannot be confirmed in a detection area due to occlusion, image loss, or unreliable segment identity, the system marks the data for that film segment in the corresponding detection area as unusable for cross-region comparison to reduce calculation errors in subsequent irreversible fingerprint formation values.

[0057] like Figure 3 As shown, in one specific embodiment, S2 includes:

[0058] S21. Obtain the correction thickness correlation characteristic value based on the thin film image data, standard thin film sample calibration results and operating condition data;

[0059] S22. Obtain the first and second stability boundaries based on the dynamic stable thin film segment library;

[0060] S23. Obtain the width state fingerprint based on the corrected thickness-related feature value, the first stable boundary, and the second stable boundary.

[0061] In this embodiment, it should be noted that in S21, for each width-direction sub-region of each detection area, the corresponding film image data is read. Overexposed pixels, underexposed pixels, and edge-occluded pixels are first removed. Then, the original optical response representative value is obtained based on the optical response values ​​corresponding to the remaining valid pixels. The original optical response representative value can be obtained from the median value of the valid pixels, the representative value of the stable range, or the representative value after anomaly removal. In one specific embodiment, the system arranges the optical response values ​​corresponding to the remaining valid pixels in numerical order and uses the optical response value in the middle position as the original optical response representative value; when the number of valid pixels is even, the average of the two middle optical response values ​​is used as the original optical response representative value. The same method for obtaining the original optical response representative value is used for the same detection area and the same batch of plastic films.

[0062] The overexposure and underexposure pixel thresholds are determined based on the camera's grayscale depth and the grayscale distribution of normal imaging samples. Taking an 8-bit grayscale image as an example, the grayscale value ranges from 0 to 255. During equipment calibration, normal thin-film images are acquired and the effective imaging grayscale range is statistically analyzed. If the grayscale of normal samples mainly falls between 35 and 230, then pixels with a grayscale value no greater than 30 can be identified as underexposure pixels, and pixels with a grayscale value no less than 240 can be identified as overexposure pixels.

[0063] The range of pixels to be determined by edge occlusion is based on the film edge positioning results and the fixed positions of the fixture, guide roller, or obstruction in the image. During equipment calibration, images of defect-free films are acquired and the width of the edge interference area is recorded. For example, if approximately 8 pixels on each of the left and right edges of the film are affected by the shadow of the fixture, in order to avoid the edge shadow from being included in the thickness-related feature calculation, the first 10 pixels and the last 10 pixels of each width-direction sub-region closest to the film edge can be discarded as edge occlusion pixels.

[0064] Furthermore, based on the calibration results of the standard thin film samples, the original representative optical response value is converted into the original thickness-related feature value. The calibration results of the standard thin film samples are obtained from standard thin film samples with known thickness or known thickness deviation, and are used to establish the correspondence between the original representative optical response value and the thickness-related feature value. The calibration results of the standard thin film samples include a correspondence table between the original representative optical response value and the thickness-related feature value. If the original representative optical response value to be processed is the same as the calibration point in the correspondence table, the corresponding thickness-related feature value is directly read; if the original representative optical response value to be processed is located between two adjacent calibration points, the original thickness-related feature value is obtained based on the linear interpolation result between the two adjacent calibration points; if it exceeds the calibration range, the sub-region in the width direction is marked as data outside the calibration range and is not included in the subsequent calculation of irreversible fingerprint formation values.

[0065] The calibration results for standard thin film samples are generated using a finite number of standard thin film samples with known thicknesses. During calibration, the original representative optical response values ​​of each standard thin film sample are collected under the same light source, camera exposure, and detection distance, and a correspondence table is established between these values ​​and the corresponding known thickness or thickness deviation. For example, if the original representative optical response values ​​for 18μm, 20μm, and 22μm standard thin film samples are 176, 168, and 159 respectively, these three sets of data are written into the correspondence table for subsequent interpolation and conversion.

[0066] Operating condition data includes film tension data, film running speed data, and film temperature data. The system queries the tension compensation rule based on the film tension data to obtain the tension compensation amount, queries the speed compensation rule based on the film running speed data to obtain the speed compensation amount, queries the temperature compensation rule based on the film temperature data to obtain the temperature compensation amount, and obtains the corrected thickness correlation characteristic value based on the original thickness correlation characteristic value, tension compensation amount, speed compensation amount, and temperature compensation amount.

[0067] Tension compensation rules, speed compensation rules, and temperature compensation rules are generated during the equipment calibration phase. During calibration, the same standard film sample is used, and a baseline thickness-related characteristic value is obtained under standard tension, standard operating speed, and standard temperature. Subsequently, while keeping the other two operating conditions constant, the film tension, film operating speed, or film temperature are changed respectively to obtain the thickness-related characteristic value under the corresponding condition. The offset of this thickness-related characteristic value relative to the baseline thickness-related characteristic value is recorded as the thickness-related characteristic compensation amount for the corresponding condition. The system writes the film tension data, film operating speed data, and film temperature data, along with the corresponding compensation amounts, into the compensation rule table. When the actual operating condition value falls between two adjacent calibration points, the corresponding compensation amounts for the two adjacent calibration points are read and obtained by linear interpolation. For example, if the baseline thickness-related characteristic value is 0.021 μm at a standard tension of 40 N, and the thickness-related characteristic value is 0.024 μm at a tension of 46 N with constant speed and temperature, then the tension compensation amount corresponding to a tension of 46 N is recorded as 0.003 μm.

[0068] Among them, the tension compensation, speed compensation, and temperature compensation are all thickness-related feature compensation values ​​that have been pre-converted to units consistent with the original thickness-related feature values ​​through calibration. When obtaining the corrected thickness-related feature values, the film tension data, film running speed data, and film temperature data are not directly added together. Instead, the corresponding effects are first converted into thickness-related feature compensation values, and then the original thickness-related feature values ​​are compensated and corrected.

[0069] Specifically, the tension compensation, speed compensation, and temperature compensation are all signed thickness-related characteristic compensation values. The corrected thickness-related characteristic value is obtained by sequentially subtracting the tension compensation, speed compensation, and temperature compensation from the original thickness-related characteristic value. When a compensation value is positive, it indicates that the corresponding operating condition causes a positive shift in the original thickness-related characteristic value, which needs to be subtracted from the original thickness-related characteristic value. When a compensation value is negative, it indicates that the corresponding operating condition causes a negative shift in the original thickness-related characteristic value, and subtracting the negative value is equivalent to replenishing the original thickness-related characteristic value. Thus, the calculation direction between the corrected thickness-related characteristic value and the original thickness-related characteristic value, tension compensation, speed compensation, and temperature compensation remains definite.

[0070] The tension compensation rule is generated through single-factor experiments during the equipment calibration phase. During calibration, the film running speed and film temperature are kept constant, while the film tension is varied and the thickness-related characteristic value of the same standard film sample is collected. The offset of this thickness-related characteristic value relative to the standard tension is used as the tension compensation amount. For example, if the thickness-related characteristic value is 0.021 μm at a standard tension of 40 N and 0.024 μm at a tension of 46 N, then the tension compensation amount corresponding to 46 N is 0.003 μm.

[0071] The speed compensation rule is generated during the equipment calibration phase by changing the film running speed. During calibration, the film tension and temperature are kept constant. Thickness-related characteristic values ​​of the same standard film sample are collected at multiple running speeds, and the offset of these values ​​relative to the thickness-related characteristic values ​​at the standard running speed is used as the speed compensation amount. For example, if the thickness-related characteristic value is 0.021 μm at a standard running speed of 200 mm / s and 0.022 μm at 240 mm / s, then the speed compensation amount corresponding to 240 mm / s is 0.001 μm.

[0072] The temperature compensation rule is generated during the equipment calibration phase by changing the film temperature. During calibration, the film tension and film running speed are kept constant. Thickness-related characteristic values ​​of the same standard film sample are collected at multiple temperature points, and the offset of these values ​​relative to the thickness-related characteristic values ​​at the standard temperature is used as the temperature compensation amount. For example, if the thickness-related characteristic value is 0.021 μm at the standard temperature of 25℃ and 0.023 μm at the film temperature of 35℃, then the temperature compensation amount corresponding to 35℃ is 0.002 μm.

[0073] Specifically, when the actual film tension data, film running speed data, or film temperature data does not fall on the calibration point of the compensation rule table, the system performs linear interpolation based on the compensation amounts corresponding to two adjacent calibration points to obtain the actual compensation amount. For example, a tension of 40N corresponds to a tension compensation amount of 0μm, a tension of 50N corresponds to a tension compensation amount of 0.005μm, and when the actual tension is 46N, the tension compensation amount is taken as 0.003μm.

[0074] In S22, thin film segments that continuously pass through at least two detection areas are designated as candidate stable thin film segments. If a candidate stable thin film segment is not marked as unreliable due to its segment identity, and the corresponding standard reference imaging area is not marked as unreliable for measurement, and the corresponding width-state fingerprint does not exhibit irreversible thickness uniformity anomalies, then the candidate stable thin film segment is added to the dynamic stable thin film segment library. The measurement reliability status of the standard reference imaging area is generated based on the standard reference imaging area image data. If the number of effective pixels in the standard reference imaging area image data is lower than the preset number of effective pixels, or the total number of overexposed pixels, underexposed pixels, and edge-occluded pixels exceeds the preset number of invalid pixels, or the reference optical response value of the standard reference imaging area exceeds the allowable response range formed by the standard reference imaging area during the calibration stage, then the corresponding standard reference imaging area is marked as unreliable for measurement; otherwise, the corresponding standard reference imaging area is marked as reliable for measurement. The preset number of effective pixels, the preset number of invalid pixels, and the allowable response range are statistically obtained from multiple calibration images of the standard reference imaging area under normal imaging conditions and written into the system parameter table.

[0075] The measurement reliability of the standard reference imaging area is determined based on the image data of the standard reference imaging area. If the number of effective pixels is lower than the preset number of effective pixels, or the total number of overexposed, underexposed, and edge-occluded pixels exceeds the preset number of invalid pixels, or the reference optical response value exceeds the allowable response range formed during the calibration phase, the measurement is marked as unreliable. For example, if the standard reference imaging area has 1000 pixels and the preset number of effective pixels is 850, but the actual number of effective pixels is only 820, then the standard reference imaging area is marked as unreliable.

[0076] The preset effective pixel count and preset invalid pixel count are obtained by statistically analyzing multiple calibration images under normal imaging conditions. The system acquires multiple frames of normal images of the standard reference imaging area, counts the effective pixel count and invalid pixel count for each frame, and uses the lowest effective pixel count and highest invalid pixel count under stable conditions as the determination criteria. For example, if 30 frames of normal images are acquired, and the lowest effective pixel count is 870 and the highest invalid pixel count is 130, then the preset effective pixel count can be set to 850 and the preset invalid pixel count can be set to 150.

[0077] The permissible response range of the standard reference imaging area is obtained statistically from the reference optical response values ​​under normal imaging conditions. The system acquires multiple frames of images under stable light source, stable camera exposure, and unobstructed standard reference imaging area conditions. The minimum and maximum values ​​of the reference optical response are then expanded by a small tolerance to determine the permissible response range. For example, if the reference optical response values ​​of 30 calibration images are distributed between 151 and 158, expanding by two grayscale values ​​allows the permissible response range to be set to 149 to 160.

[0078] For the same detection area and the same width-direction sub-region, the corresponding correction thickness-related feature values ​​in the dynamic stable thin film segment library are arranged in numerical order. When the number of correction thickness-related feature values ​​after arrangement is not less than four, the first and last values ​​are removed, and the first value of the remaining sequence is used as the first stable boundary, and the last value of the remaining sequence is used as the second stable boundary. When the number of correction thickness-related feature values ​​after arrangement is less than four, the first and second stable boundaries corresponding to the previous acquisition cycle are used. Both the first and second stable boundaries are derived from the correction thickness-related feature values ​​corresponding to the same detection area and the same width-direction sub-region in the dynamic stable thin film segment library, and can be compared with the correction thickness-related feature values ​​of the thin film segment to be evaluated. For example, if the correction thickness-related feature values ​​of a certain width-direction sub-region in the dynamic stable thin film segment library are -0.013μm, -0.008μm, 0.002μm, 0.011μm, and 0.018μm respectively, after removing the first and last values, the first stable boundary is -0.008μm, and the second stable boundary is 0.011μm.

[0079] The initial first stable boundary and the initial second stable boundary are obtained during the standard thin film sample calibration stage. Under normal operating conditions, the system continuously acquires multiple corrected thickness-related feature values ​​of the same detection area and the same width direction sub-region. After arranging them in numerical order, the lowest and highest values ​​are discarded. The lowest value of the remaining sequence is used as the initial first stable boundary, and the highest value of the remaining sequence is used as the initial second stable boundary. For example, calibration yields -0.010μm, -0.008μm, 0.002μm, 0.011μm, and 0.014μm. After discarding the first and last values, the initial first stable boundary is -0.008μm, and the initial second stable boundary is 0.011μm.

[0080] Setting the minimum number of values ​​to four ensures that after removing the lowest and highest values, at least two corrected thickness-related feature values ​​are retained to determine the stability boundary, avoiding the determination of the stability range based on a single sample. For example, a sub-region in the width direction of a dynamic stable thin film segment library has four corrected thickness-related feature values, ordered as -0.010μm, -0.006μm, 0.008μm, and 0.013μm. After removing the first and last values, the first stability boundary of -0.006μm and the second stability boundary of 0.008μm can be obtained.

[0081] Irreversible thickness uniformity anomalies are determined based on the width state fingerprints of the same film segment in the upstream, downstream, and further downstream detection regions. If a certain width-direction sub-region changes from a stable state upstream to an anomalous state on the thinner or thicker side downstream, and continues to maintain the same anomalous state on the same side further downstream, then the corresponding width state fingerprint is considered to have an irreversible thickness uniformity anomaly. For example, if the third width-direction sub-region is stable in the inlet detection region but exhibits an anomalous state on the thicker side in both the processing and cooling detection regions, then this film segment will not be added to the dynamically stable film segment library.

[0082] In step S23, for each width-direction sub-region of the film segment to be evaluated in each detection area, the corresponding corrected thickness associated feature value, the first stable boundary, and the second stable boundary are read, and the width state value is obtained according to the segmentation rules. If the corrected thickness associated feature value is less than the first stable boundary, the width-direction sub-region is marked as a thin-side abnormal state; if the corrected thickness associated feature value is between the first and second stable boundaries, the width-direction sub-region is marked as a stable state; if the corrected thickness associated feature value is greater than the second stable boundary, the width-direction sub-region is marked as a thick-side abnormal state. The system arranges the thin-side abnormal state, stable state, and thick-side abnormal state in sequence according to the arrangement order of the width-direction sub-regions in the width direction of the plastic film, forming the width state fingerprint of the film segment to be evaluated in the corresponding detection area. For example, if a film segment has the following states in the eight width-direction sub-regions of the processing detection area: stable state, stable state, thick-side abnormal state, thick-side abnormal state, stable state, stable state, stable state, stable state, stable state, then its width state fingerprint can be recorded as [0,0,1,1,0,0,0,0]. This state fingerprint is used to record local stripe anomalies. For example, consecutive thicker-side anomalies in the third and fourth width-direction sub-regions can be preserved in the state fingerprint without being merged into a single average with other stable regions.

[0083] In S23, the width state value is obtained according to the following segmentation rules:

[0084]

[0085] in, This represents the width state value of the film segment identified as s in the j-th detection region and located in the r-th width direction sub-region; j represents the detection region index; s represents the film segment identifier; r represents the width direction sub-region index; This represents the corrected thickness-related feature value of the thin film segment identified as s in the j-th detection region and located in the r-th width-direction sub-region; This represents the first stable boundary corresponding to the r-th width-direction sub-region in the j-th detection region; This represents the second stable boundary corresponding to the r-th width-direction sub-region in the j-th detection region; -1 indicates an abnormal state on the thinner side; 0 indicates a stable state; and 1 indicates an abnormal state on the thicker side. In the above expression, , and All of them are thickness-related characteristic values ​​under the same calibration system, which can be compared numerically. These are state values ​​used to form the width state fingerprint. When this segmentation rule is adopted, the thinner side anomalous state, the stable state, and the thicker side anomalous state correspond to different state values, which are used to record the anomalous direction and width position, and provide state data for subsequent irreversible fingerprint formation value calculation.

[0086] like Figure 4 As shown, in one specific embodiment, S3 includes:

[0087] S31. Read the width state fingerprint corresponding to the width direction sub-region that can complete the matching of the same thin film segment;

[0088] S32. Obtain the irreversible fingerprint formation value based on the width state fingerprint;

[0089] S33. Obtain the optical drift stripping value based on the direction of change of the correction thickness-related characteristic value and the direction of change of the reference optical response of the standard reference imaging area.

[0090] In this embodiment, it should be noted that in S31, the width state fingerprint corresponding to the width-direction sub-region that can complete the matching of the same film segment is read in the upstream detection region, downstream detection region, and further downstream detection region of the same film segment. The width-direction sub-region that can complete the matching of the same film segment refers to a width-direction sub-region that has not been marked as untrustworthy during the identity verification process of the same film segment and can establish a corresponding relationship between adjacent detection regions. If a certain width-direction sub-region is occluded in the downstream detection region or the image quality does not meet the usage conditions, then that width-direction sub-region will not be included in the calculation set of irreversible fingerprint formation values. It should be noted that the calculation of irreversible fingerprint formation values ​​requires that the same film segment has usable width state fingerprint data in the upstream detection region, downstream detection region, and further downstream detection region. For the current adjacent detection regions between the p-th and q-th detection regions, the system calculates the irreversible fingerprint formation value between the current adjacent detection regions only if the q+1-th detection region exists and the same film segment can reach the q+1-th detection region; if the q+1-th detection region does not exist, the system does not perform the irreversible fingerprint formation value calculation for the current adjacent detection region, or waits for the downstream detection data of the same film segment to be completed before performing the calculation.

[0091] The number of width-direction sub-regions that can complete the matching of the same film segment is determined based on the segment identity verification results, image quality status, and occlusion status. The system first considers all width-direction sub-regions as candidates, and then eliminates width-direction sub-regions that are untrustworthy in terms of segment identity, have missing images, are occluded, or are outside the calibration range. For example, if the film image is divided into 8 width-direction sub-regions, and the 8th width-direction sub-region is occluded in the cooling detection area, then the number of width-direction sub-regions that can complete the matching of the same film segment is 7.

[0092] For example, if the entry detection area, processing detection area, and cooling detection area can all complete the matching of the same film segment in 8 width-direction sub-regions, then the number of width-direction sub-regions participating in the calculation is 8. If the 8th width-direction sub-region is occluded in the cooling detection area, the number of width-direction sub-regions participating in the calculation is adjusted to 7 to reduce the impact of missing data on irreversible fingerprint formation values.

[0093] In S32, the system obtains irreversible fingerprint formation values ​​based on the width state fingerprints of the same film segment in the upstream detection region, downstream detection region, and further downstream detection region. When obtaining irreversible fingerprint formation values, it first determines whether each width-direction sub-region capable of matching the same film segment meets the continuous anomaly formation condition, and then determines whether it meets the adjacent same-side anomaly formation condition with its adjacent width-direction sub-regions. For the same width-direction sub-region, if both conditions are met simultaneously, it is counted only once. The continuous anomaly formation condition is used to identify cases where a single width-direction sub-region changes from an upstream stable state to a downstream anomalous state and continues to maintain the anomalous state further downstream. The adjacent same-side anomaly formation condition is used to identify cases where two adjacent width-direction sub-regions change from an upstream adjacent stable state to a downstream adjacent same-side anomalous state and continue to maintain this same-side anomalous state further downstream. This process enables irreversible fingerprint formation values ​​to record local continuous anomalies and strip-shaped continuous anomalies, and reduces the repeated counting of the same anomaly.

[0094] In this context, adjacent width-direction sub-regions are determined according to their numbering order along the film width direction, and two width-direction sub-regions with a numbering difference of 1 are considered to be adjacent width-direction sub-regions. For example, if the width-direction sub-regions are numbered sequentially from 1 to 8, then the 3rd width-direction sub-region is adjacent to the 4th width-direction sub-region; if both are in a stable state upstream and in a thicker-side anomalous state downstream and further downstream, then the conditions for the formation of adjacent same-side anomalous regions are met.

[0095] In S32, the irreversible fingerprint formation value is obtained according to the following expression:

[0096]

[0097] in, This represents the irreversible fingerprint formation value of the same film segment identified as s between the p-th and q-th detection regions; p represents the upstream detection region index; q represents the downstream detection region index; s represents the film segment identifier; and r represents the sub-region index in the width direction. This represents the set of width-direction sub-regions in the p-th and q-th detection regions that can complete the matching of the same thin film segment; Represents a set The number of sub-regions in the medium-width direction; Represents a set Summation is performed on sub-regions of each width direction; This indicates a persistent anomaly formation marker in the r-th width-direction sub-region; This indicates the formation of adjacent same-side anomalies in the r-th width-direction sub-region; This indicates taking the larger of the two marked values.

[0098] in, The value selection rules are as follows: when the thin film segment in the p-th detection region is identified as s and the width state value of the sub-region in the r-th width direction is stable, the width state value of the sub-region in the same width direction in the q-th detection region is an abnormal state on the thinner side or the thicker side, and the sub-region in the same width direction in the (q+1)-th detection region maintains the abnormal state on the same side as the q-th detection region, Select 1; otherwise select 0.

[0099] in, The value selection rule is as follows: when there is an adjacent width-direction sub-region in the r-th width direction that can complete the matching of the same film segment, and both of these adjacent width-direction sub-regions in the p-th detection region are in a stable state, both of these adjacent width-direction sub-regions in the q-th detection region become an aberrant state on the same side, and this aberrant state on the same side continues to exist in the q+1-th detection region, Select 1; otherwise select 0.

[0100] In the above expression, and All are counting markers generated based on the width state value; This represents the number of sub-regions in the width direction. The expression only counts, takes the maximum value, sums, and normalizes the number; it does not involve direct addition or subtraction of data in different units such as thickness, temperature, tension, and speed. This calculation logic is used to record state changes that form in the downstream detection area and continue to exist in the next downstream detection area. Short-term noise, single-frame acquisition errors, or instantaneous optical disturbances generally do not meet the conditions of upstream stability, new downstream anomalies, and continuation on the same side in the next downstream area. Strip-shaped thickness anomalies can manifest as anomalies on the same side of adjacent sub-regions in the width direction; therefore, a marker is set for adjacent anomalies on the same side to record continuous anomalies in the width direction. Taking the larger of the two marker values ​​for the same sub-region in the width direction can reduce the repeated counting of continuous anomalies and adjacent anomalies on the same side.

[0101] In step S33, for each width-direction sub-region of the same thin film segment that can be matched in both the upstream and downstream detection regions, the corresponding correction thickness correlation feature value is read. For each width-direction sub-region, the direction of change of the correction thickness correlation feature value in the downstream detection region relative to the correction thickness correlation feature value in the upstream detection region is compared. The direction of change includes increasing, decreasing, or no change. The reference optical response value of the standard reference imaging region corresponding to the same thin film segment in both the upstream and downstream detection regions is read, and the direction of change of the reference optical response of the standard reference imaging region is obtained. The number of width-direction sub-regions whose direction of change of correction thickness correlation feature value is consistent with the direction of change of reference optical response of the standard reference imaging region is counted, and the optical drift stripping value is obtained based on the number of width-direction sub-regions with consistent direction and the number of width-direction sub-regions that can be matched in both directions.

[0102] In one specific implementation, the optical drift stripping value is obtained according to the following rule:

[0103]

[0104] in, This represents the optical drift stripping value of the same film segment identified as s between the p-th and q-th detection regions; p represents the upstream detection region index; q represents the downstream detection region index; and s represents the film segment identifier. This represents the number of width-direction sub-regions between the p-th and q-th detection regions where the direction of change of the correction thickness-related feature value is consistent with the direction of change of the reference optical response of the standard reference imaging region. This represents the number of width-oriented sub-regions in the p-th and q-th detection regions that can achieve matching of the same thin film segment. When the corrected thickness-related characteristic value of a certain width-oriented sub-region increases in the q-th detection region relative to the p-th detection region, and the reference optical response of the standard reference imaging region also increases, it is considered oriented consistent; when both decrease, it is also considered oriented consistent; when one increases while the other decreases, it is considered oriented inconsistent; when neither changes, it is not included in the number of oriented consistent regions.

[0105] When the change direction of most width-direction sub-regions that can complete the matching of the same thin film segment is consistent with the change direction of the reference optical response of the standard reference imaging region, the corresponding width state fingerprint change is judged as optical measurement drift; when the change direction of a local width-direction sub-region is inconsistent with the change direction of the reference optical response of the standard reference imaging region, the judgment of irreversible thickness uniformity anomaly is entered.

[0106] The optical drift stripping value is calculated as the ratio of the number of width-direction sub-regions with consistent orientation to the number of width-direction sub-regions capable of matching the same thin film segment. When the direction of change of the correction thickness-related characteristic value is consistent with the direction of change of the reference optical response of the standard reference imaging area, the consistent orientation is included. For example, if the number of width-direction sub-regions capable of matching the same thin film segment is 8, and the direction of change of 6 of these width-direction sub-regions is consistent with the direction of change of the reference optical response, then the optical drift stripping value is 6 / 8 = 0.75.

[0107] In S3, the system can simultaneously record irreversible fingerprint formation values ​​and optical drift stripping values ​​into the thin film segment evaluation data table. The thin film segment evaluation data table includes at least the thin film segment identifier, upstream detection area, downstream detection area, further downstream detection area, number of width-direction sub-regions involved in the calculation, irreversible fingerprint formation values, optical drift stripping values, and width state fingerprints. Using this data table, the system can read historical irreversible fingerprint formation values ​​prior to the current adjacent detection area in S4, and further obtain process segment anomaly increment values.

[0108] The thin film segment evaluation data table is generated after each thin film segment completes the comparison of adjacent detection areas. The data recorded in the table is directly written from the processing results of S1 to S3, and an index is established according to the combination of thin film segment identifier and detection area. For example, if the thin film segment identifier is 62, the number of sub-regions in the calculation width direction from the entrance detection area to the processing detection area is 8, the irreversible fingerprint formation value is 0.25, and the optical drift stripping value is 0.25, then the above data will be written into the thin film segment evaluation data table as an evaluation record.

[0109] like Figure 5 As shown, in one specific embodiment, S4 includes:

[0110] S41. Obtain the abnormal incremental value of the process segment based on the irreversible fingerprint formation value and the historical irreversible fingerprint formation value;

[0111] S42. Determine the dynamic evaluation results of thickness uniformity based on the irreversible fingerprint formation value, optical drift peeling value, and abnormal increment value of the process segment.

[0112] S43, Output the dynamic evaluation results of thickness uniformity.

[0113] In this embodiment, it should be noted that in S41, for the production line process segment between the current adjacent detection areas, the irreversible fingerprint formation value between the current adjacent detection areas is read. Then, the historical irreversible fingerprint formation value before the current adjacent detection area is read. If no historical irreversible fingerprint formation value exists, the irreversible fingerprint formation value between the current adjacent detection areas is used as the process segment's abnormal increment value. If a historical irreversible fingerprint formation value exists, the largest historical irreversible fingerprint formation value is selected from the historical irreversible fingerprint formation values ​​before the current adjacent detection area. If the irreversible fingerprint formation value between the current adjacent detection areas is greater than the largest historical irreversible fingerprint formation value, the difference between the two is used as the process segment's abnormal increment value; if the irreversible fingerprint formation value between the current adjacent detection areas is less than or equal to the largest historical irreversible fingerprint formation value, the process segment's abnormal increment value is recorded as 0. For example, if the irreversible fingerprint formation value between the entry detection area and the processing detection area is 0.25, and no historical irreversible fingerprint formation value exists before the current adjacent detection area, then the process segment's abnormal increment value is 0.25. If the irreversible fingerprint formation value between the processing inspection area and the cooling inspection area is 0.375, and the maximum historical irreversible fingerprint formation value is 0.25, then the process segment anomaly increment value is 0.125. If the irreversible fingerprint formation value between subsequent inspection areas is still 0.375, then the process segment anomaly increment value is 0.

[0114] The historical irreversible fingerprint formation value is derived from the upstream adjacent detection area records that have already been evaluated in the thin film segment evaluation data table. During reading, only records with the same thin film segment identifier and located before the current adjacent detection area are selected. For example, if the current adjacent detection area is from the processing detection area to the cooling detection area, and the thin film segment identifier is 62, and if an irreversible fingerprint formation value of 0.25 has already been recorded from the entrance detection area to the processing detection area, then this 0.25 is used as the historical irreversible fingerprint formation value before the current adjacent detection area.

[0115] The maximum historical irreversible fingerprint formation value is obtained by comparing the historical irreversible fingerprint formation values ​​of the same film segment in each adjacent detection area before the current adjacent detection area. It is used to determine whether a new anomaly has occurred in the current process segment. For example, if the historical irreversible fingerprint formation value of the same film segment from the entrance detection area to the processing detection area is 0.25, and the historical irreversible fingerprint formation value in the upstream adjacent detection area is 0.125, then the maximum historical irreversible fingerprint formation value is 0.25.

[0116] Specifically, if the irreversible fingerprint formation value between currently adjacent detection areas does not exceed the maximum historical irreversible fingerprint formation value, it indicates that no new irreversible anomalies have formed in the current process segment, and the system records the process segment anomaly increment value as 0. For example, if the historical maximum irreversible fingerprint formation value is 0.375, and the current irreversible fingerprint formation value between adjacent detection areas is still 0.375, then the process segment anomaly increment value is 0, which can be further determined as an anomaly transmission segment.

[0117] In S42, if the optical drift stripping value meets the optical measurement drift condition, the corresponding width state fingerprint change is determined as optical measurement drift, and the dynamic evaluation result of thickness uniformity corresponding to the optical measurement drift is output. The optical measurement drift condition is: the number of width-direction sub-regions whose change direction of the correction thickness associated feature value is consistent with the change direction of the reference optical response of the standard reference imaging area is greater than the number of width-direction sub-regions whose change direction is inconsistent; in other words, when When the system determines that the optical measurement drift condition is met; when At that time, the system determines that the optical measurement drift condition is not met. If the abnormal increment value of the process segment is greater than 0, and the abnormal state corresponding to the irreversible fingerprint formation value continues to exist in the downstream detection area, then the production line process segment between the current adjacent detection areas is determined to be either an abnormal formation segment or an abnormal expansion segment, depending on whether an irreversible thickness uniformity abnormality already existed before the current adjacent detection area. If there is no irreversible thickness uniformity abnormality before the current adjacent detection area, then the production line process segment between the current adjacent detection areas is determined to be an abnormal formation segment; if there is an irreversible thickness uniformity abnormality before the current adjacent detection area, then the production line process segment between the current adjacent detection areas is determined to be an abnormal expansion segment. If the irreversible fingerprint formation value If the abnormal increment value of the process segment is 0, then the production line process segment between the current adjacent detection areas is determined as the abnormal transmission segment.

[0118] The optical measurement drift condition is determined using the majority consensus principle. When the optical drift stripping value is greater than 0.5, it indicates that most width-direction sub-regions change in the same direction as the standard reference imaging area, and this is considered optical measurement drift. When the optical drift stripping value is not greater than 0.5, it is not considered optical measurement drift. For example, if the number of width-direction sub-regions involved in the judgment is 8, and the number of consistent directions is 5, then the optical drift stripping value is 0.625, which satisfies the optical measurement drift condition.

[0119] The distinction between anomaly formation and anomaly expansion is based on whether an irreversible thickness uniformity anomaly existed previously in the adjacent detection area. If no irreversible thickness uniformity anomaly existed previously, and the current process segment's anomaly increment value is greater than 0, it is determined to be an anomaly formation segment. If an irreversible thickness uniformity anomaly existed previously, and the current process segment's anomaly increment value continues to be greater than 0, it is determined to be an anomaly expansion segment. For example, if the process segment anomaly increment value of 0.25 first appears between the entry detection area and the processing detection area, it is determined to be an anomaly formation segment; if it further increases by 0.125 between the processing detection area and the cooling detection area, it is determined to be an anomaly expansion segment.

[0120] In S43, the dynamic evaluation results for thickness uniformity include film segment identification, detection area, width status fingerprint, irreversible fingerprint formation value, optical drift peeling value, process segment abnormal increment value, abnormal formation segment, abnormal extension segment, or abnormal transmission segment. When the dynamic evaluation results for thickness uniformity include an abnormal formation segment, abnormal extension segment, or abnormal transmission segment, the dynamic evaluation results for thickness uniformity are sent to the result display module, communication module, and production line control equipment. The production line control equipment can execute prompts, speed reduction, shutdown, or verification requests based on the dynamic evaluation results for thickness uniformity. For example, when multiple consecutive film segments are identified as abnormal formation segments between the entrance detection area and the processing detection area, the production line control equipment prompts the user to check the tension setting, roller gap, heating status, or traction status between the entrance detection area and the processing detection area; when an abnormality is identified as an abnormal transmission segment, the system retains the record and does not treat the subsequent process segment as an abnormal formation location.

[0121] In S4, the system can also generate process segment statistical records based on the dynamic evaluation results of the thickness uniformity of multiple continuous film segments. If a process segment on the same production line is identified as an abnormal formation segment or an abnormal expansion segment in five consecutive film segments, the system increases the review priority of that production line process segment. If a film segment is identified as having optical measurement drift, but adjacent film segments do not show irreversible fingerprint formation values, the system retains the drift record and continues to collect subsequent data. This supplementary implementation does not change the core algorithm logic; it is only used to improve readability and management convenience in actual production lines.

[0122] like Figure 6 and Figure 7 As shown, the present invention also provides a machine vision-based dynamic evaluation system for the thickness uniformity of plastic films, including a data acquisition module, a film segment confirmation module, a feature processing module, a fingerprint generation module, and an evaluation and positioning module.

[0123] The data acquisition module is used to acquire film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data from at least three detection areas arranged sequentially along the direction of plastic film movement. The operating condition data includes film tension data, film speed data, and film temperature data.

[0124] The thin film segment identification module is used to generate a thin film segment identifier based on the cumulative thin film running displacement data, and to identify the same thin film segment in different detection areas according to the relative optical response order of each width direction sub-region in the thin film image data.

[0125] The feature processing module is used to obtain the correction thickness correlation feature values ​​of the same film segment in each detection area and each width direction sub-region based on film image data, standard film sample calibration results and operating condition data.

[0126] The fingerprint generation module is used to determine the stable boundary based on the dynamic stable thin film segment library, and generate a width state fingerprint based on the corrected thickness associated feature value and the stable boundary.

[0127] The evaluation and positioning module is used to obtain irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream detection area, downstream detection area, and further downstream detection area; to obtain optical drift stripping values ​​based on the direction of change of the corrected thickness-related feature values ​​of the same thin film segment in adjacent detection areas and the direction of change of the reference optical response in the standard reference imaging area; and to determine the dynamic evaluation result of thickness uniformity, including abnormal formation segments, abnormal extension segments, or abnormal transmission segments, based on the irreversible fingerprint formation values, the optical drift stripping values, and historical irreversible fingerprint formation values.

[0128] The data acquisition module includes a visual data acquisition unit, a reference data acquisition unit, a cumulative displacement data acquisition unit, a film tension data acquisition unit, a film running speed data acquisition unit, and a film temperature data acquisition unit. The visual data acquisition unit acquires film image data. The reference data acquisition unit acquires image data of a standard reference imaging area and, based on this image data, obtains the direction of change in the reference optical response of the standard reference imaging area. The cumulative displacement data acquisition unit acquires cumulative film running displacement data. The film tension data acquisition unit acquires film tension data. The film running speed data acquisition unit acquires film running speed data. The film temperature data acquisition unit acquires film temperature data. The evaluation and positioning module also determines the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment based on the irreversible fingerprint formation value, optical drift peeling value, and process segment abnormal increment value, and sends the dynamic evaluation results of thickness uniformity, including the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment, to the result display module, communication module, and production line control equipment.

[0129] The output of the dynamic assessment results for thickness uniformity is determined based on the anomaly type and corresponding process segment generated by the assessment and positioning module. It includes at least the film segment identifier, anomaly type, corresponding detection area, corresponding production line process segment, irreversible fingerprint formation value, and process segment anomaly increment value. For example, if the film segment identifier is 62, the area between the inlet detection area and the processing detection area is determined to be an anomaly formation segment, and the irreversible fingerprint formation value is 0.25, the result display module will show that anomaly formation exists in the film segment from the inlet to the processing process segment.

[0130] The modules in the above system can be implemented using industrial computers, programmable logic controllers (PLCs), field-programmable gate arrays (FPGAs), image processing boards, or combinations thereof. The data flow between modules is consistent with the steps described above. During system execution, no training model is used, nor is semantic segmentation performed on defective pixels. The system completes the dynamic evaluation of thickness uniformity based on data acquisition, calibration transformation, compensation correction, boundary generation, state fingerprint generation, irreversible fingerprint formation value acquisition, optical drift stripping value acquisition, process segment anomaly increment value acquisition, and evaluation and positioning steps. Furthermore, the aforementioned calibration transformation, compensation rule query, stable boundary generation, state fingerprint generation, irreversible fingerprint formation value acquisition, optical drift stripping value acquisition, and process segment anomaly increment value acquisition all have defined input data, processing rules, and output data, and do not rely on model parameters obtained through iterative training of samples.

[0131] To further clarify the operational mechanism and physical quantification process of the technical solution of this invention, the following describes the underlying derivation logic of the machine vision-based dynamic evaluation method for the uniformity of plastic film thickness, using a scenario containing specific parameters and data. For example... Figure 6 and Figure 7As shown, in a continuous plastic film production scenario, the production line is equipped with an entry detection area, a processing detection area, a cooling detection area, and a winding detection area. The nominal thickness of the plastic film is 50 μm, the film running speed is 38 m / min, and the preset film segment length is 20 mm. At a certain acquisition moment, the cumulative film running displacement data is 1240 mm. The system calculates the ratio of the cumulative film running displacement data to the preset film segment length, obtaining 1240 / 20 = 62, and performs a round-down operation on the comparison result, resulting in a film segment identifier of 62. In the entry detection area, the system divides the film image data into 8 width-direction sub-regions along the width direction of the plastic film. The original optical response values ​​of each width-direction sub-region are 128.4, 130.1, 132.6, 133.2, 131.8, 129.7, 127.9, and 126.5, respectively, and obtains the initial fingerprint of the visual width state based on the relative optical response order. When the film segment identified as 62 reaches the processing detection area, the system again acquires the initial fingerprint of the visual width state. If the initial fingerprints of the visual width state of the processing detection area and the entrance detection area meet the preset matching conditions, it is confirmed that they belong to the same film segment. This process corresponds to S1, and its function is to ensure that the data participating in the subsequent comparison comes from the same film segment, rather than from accidental image differences from different film segments.

[0132] In S2, taking the third width-direction sub-region of the processing and inspection area as an example, after removing overexposed, underexposed, and edge-occluded pixels from the thin film image data, the system obtains a raw optical response representative value of 134.8. According to the calibration results of the standard thin film sample, this raw optical response representative value corresponds to a raw thickness-related characteristic value of 0.024 μm. At the same acquisition time, the thin film segment identified as 62 in the processing and inspection area has a thin film tension of 46 N and a thin film running speed of... The film temperature data is 42℃. Based on the tension compensation rule, speed compensation rule, and temperature compensation rule, the system obtains a tension compensation amount of 0.003μm, a speed compensation amount of 0.001μm, and a temperature compensation amount of 0.002μm, respectively. These compensation amounts are all thickness-related feature compensation amounts with units consistent with the original thickness-related feature value and can be used for compensation correction. In this embodiment, the corrected thickness-related feature value is obtained by subtracting the tension compensation amount, speed compensation amount, and temperature compensation amount from the original thickness-related feature value. For example, if the original thickness-related feature value is 0.024μm, the tension compensation amount is 0.003μm, the speed compensation amount is 0.001μm, and the temperature compensation amount is 0.002μm, then the corrected thickness-related feature value is: The corrected thickness correlation feature value is used for comparison with the first and second stable boundaries corresponding to the same detection area and the same width-direction sub-region. Similarly, the corrected thickness correlation feature value for the fourth width-direction sub-region is 0.020 μm, and the corrected thickness correlation feature values ​​for the remaining width-direction sub-regions are 0.003 μm, 0.006 μm, 0.004 μm, 0.003 μm, 0.002 μm, and 0.001 μm, respectively. This processing avoids the problem of directly mixing film tension data, film running speed data, and film temperature data with image response calculations.

[0133] Furthermore, the system obtains the width state fingerprint based on the corrected thickness correlation feature value, the first stable boundary, and the second stable boundary. Taking the third width-direction sub-region of the processing detection area as an example, the corrected thickness correlation feature values ​​corresponding to the same detection area and the same width-direction sub-region in the dynamic stable thin film segment library are -0.013μm, -0.008μm, 0.002μm, 0.011μm, and 0.018μm, respectively. After arranging the values ​​in numerical order, the system removes the first and last values, taking the first value of the remaining sequence, -0.008μm, as the first stable boundary, and the last value of the remaining sequence as the second stable boundary. This serves as the second stable boundary. The width state value is obtained according to the following segmentation rule: For the third width-direction sub-region of the processing and inspection area, , , ,because Therefore, the width state value of the third width-direction sub-region is 1, corresponding to the anomalous state on the thicker side. The corrected thickness correlation feature value of the fourth width-direction sub-region is... It is also greater than the corresponding second stability boundary. Therefore, it is also marked as an abnormal state on the thicker side. The eight width-direction sub-regions of the processing detection area ultimately form a width state fingerprint [0,0,1,1,0,0,0,0]. The corresponding width state fingerprint of the entry detection area is [0,0,0,0,0,0,0,0], and the corresponding width state fingerprint of the cooling detection area is [0,0,1,1,0,0,0,0].

[0134] In S3, the system obtains irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream detection region, downstream detection region, and further downstream detection region. For a thin film segment identified as 62, the entry detection region is designated as the p-th detection region, the processing detection region as the q-th detection region, and the cooling detection region as the q+1-th detection region; the set of width-direction sub-regions capable of matching the same thin film segment contains 8 width-direction sub-regions, i.e. The irreversible fingerprint value is obtained according to the following expression: For the third width-direction sub-region, the width state value of the inlet detection region is 0, the width state value of the processing detection region is 1, and the width state value of the cooling detection region is 1. Therefore, it meets the conditions for continuous anomaly formation. Meanwhile, the third width-direction sub-region is adjacent to the fourth width-direction sub-region. Both are in a stable state in the entry detection area, but both are in an abnormal state on the thicker side in the processing detection area. Furthermore, this same-side abnormal state continues to exist in the cooling detection area. Therefore, the third width-direction sub-region satisfies the conditions for the formation of adjacent same-side abnormalities. For the third width-direction sub-region, The fourth width-direction sub-region is counted only once. Similarly, the fourth width-direction sub-region is counted once. The remaining six width-direction sub-regions do not meet the conditions for sustained anomaly formation or adjacent side-by-side anomaly formation, therefore they are all counted as 0. Therefore: .

[0135] The calculation results indicate that among the eight width-direction sub-regions capable of matching the same thin film segment, two width-direction sub-regions exhibited an irreversible fingerprint change characterized by upstream stability, downstream novel anomalies, and then continuous same-side downstream anomalies. This change includes adjacent same-side anomalous features. (The expression is derived from...) By merging persistent anomaly formation markers and adjacent ipsilateral anomaly formation markers, sub-regions in the same width direction will not be counted repeatedly even if they simultaneously satisfy both local persistent anomalies and adjacent strip anomalies. In this way, the irreversible fingerprint formation value can reflect both local persistent anomalies and strip-shaped continuous anomalies, while avoiding the amplification of evaluation results due to the same anomaly being counted multiple times.

[0136] During the calculation of optical drift stripping values, the system compares the direction of change of the correction thickness-related characteristic values ​​of each width-direction sub-region in the entry detection area and the processing detection area. Assuming there are 8 width-direction sub-regions, the 3rd width-direction sub-region is composed of... Change to The direction of change is increasing; the fourth width direction sub-region is composed of... Change to The direction of change is increasing; the remaining six width-direction sub-regions are in a stable state. The reference optical response value of the standard reference imaging area in the entry detection area is 120.2, and the reference optical response value of the standard reference imaging area in the processing detection area is 120.1. The direction of change of the reference optical response of the standard reference imaging area is decreasing. The direction of change of the corrected thickness associated characteristic value of the 3rd and 4th width-direction sub-regions is inconsistent with the direction of change of the reference optical response of the standard reference imaging area, and this local anomaly continues to exist in the cooling detection area. Therefore, the system does not determine the width state fingerprint change as optical measurement drift. If the corrected thickness associated characteristic value of most width-direction sub-regions in another thin film segment changes in the same direction as the reference optical response of the standard reference imaging area, the system will determine the corresponding width state fingerprint change as optical measurement drift and will not proceed to the anomaly formation stage.

[0137] In S4, the irreversible fingerprint formation value between the entry detection area and the processing detection area is 0.25. Since there are no historical irreversible fingerprint formation values ​​before the current adjacent detection area, the system uses 0.25 as the abnormal increment value for the process segment. Because the optical drift stripping value does not meet the optical measurement drift condition, and the abnormal state on the thicker side corresponding to the irreversible fingerprint formation value continues to exist in the cooling detection area, the system identifies the production line process segment between the entry detection area and the processing detection area as an abnormal formation segment. If the width state fingerprint between the processing detection area and the cooling detection area does not add any new abnormal width direction sub-regions and only remains [0,0,1,1,0,0,0,0], then the current abnormal increment value for the process segment is 0, and the system identifies the production line process segment between the processing detection area and the cooling detection area as an abnormal transmission segment. If the width state fingerprint in the subsequent detection area further changes to [0,0,1,1,1,0,0,0], and the newly added 5th width direction sub-region continues to maintain the abnormal state of being thicker on the side in the downstream detection area, then the system obtains a new process segment abnormal increment value based on the formation of the newly added irreversible fingerprint, and determines the corresponding production line process segment as the abnormal extension segment.

[0138] This application scenario illustrates that this invention does not merely perform anomaly detection on a single frame image of a specific detection area, but rather continuously analyzes the width state fingerprint changes of the same film segment across at least three detection areas sequentially set along the plastic film's running direction. Film segment identification and initial visual width state fingerprints address the issue of corresponding segments; corrected thickness-related feature values ​​address visual response offsets caused by tension, speed, and temperature effects; a dynamically stable film segment library and stable boundaries address the insufficient adaptability of fixed thresholds; the width state value segmentation rule in S23 transforms the corrected thickness-related feature values ​​into thinner-side anomaly states, stable states, and thicker-side anomaly states that can be compared across detection areas; the irreversible fingerprint formation value in S32 further utilizes upstream stability, downstream new anomalies, downstream same-side persistence, and adjacent same-side anomaly formation rules to count and normalize true thickness uniformity anomalies; optical drift stripping values ​​resolve the confusion between optical measurement drift and true thickness uniformity anomalies; and process segment anomaly increment values ​​distinguish between anomaly formation, anomaly expansion, and anomaly propagation. Each of the above steps has clear inputs and outputs, and those skilled in the art can implement it based on image acquisition, data calibration, compensation and correction, sorting, state discrimination, counting and statistics, and rule judgment.

[0139] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0140] 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, the present invention will not describe the various possible combinations separately.

[0141] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

[0142] 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 they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these 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 machine vision-based dynamic evaluation method for the thickness uniformity of plastic films, characterized in that, The methods include: Acquire film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data from at least three detection areas set sequentially along the running direction of the plastic film. Based on the accumulated thin film displacement data, a thin film segment identifier is generated, and based on the relative optical response order of each width direction sub-region in the thin film image data, the thin film segments in different detection areas are identified as the same thin film segment. Based on thin film image data, standard thin film sample calibration results, and operating condition data, the correction thickness correlation characteristic values ​​of the same thin film segment in each detection area and each width direction sub-region are obtained; The stability boundary is determined based on the dynamic stable thin film segment library, and the width state fingerprint is generated based on the corrected thickness-related feature value and the stability boundary. Irreversible fingerprint formation values ​​are obtained based on the width state fingerprints of the same thin film segment in the upstream detection region, downstream detection region, and further downstream detection region. The optical drift stripping value is obtained by considering the direction of change of the correction thickness-related characteristic value of the same thin film segment in adjacent detection areas and the direction of change of the reference optical response in the standard reference imaging area. Based on the irreversible fingerprint formation value, the optical drift stripping value, and the historical irreversible fingerprint formation value, determine the dynamic evaluation result of thickness uniformity including abnormal formation segments, abnormal extension segments, or abnormal transmission segments.

2. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, The operating condition data includes film tension data, film running speed data, and film temperature data; Based on the accumulated thin film displacement data, a thin film segment identifier is generated, including: Read the cumulative thin film displacement data corresponding to the current acquisition time; The cumulative film displacement data is compared with the preset film segment length, and the comparison result is rounded down. The result of the rounding down process will be used as the identifier of the thin film segment corresponding to the current acquisition time.

3. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, Based on the relative optical response order of sub-regions in each width direction in the thin film image data, the identity of the same thin film segment is confirmed for thin film segments in different detection regions, including: The thin film image data is divided into multiple width-direction sub-regions along the width direction of the plastic film, and the original optical response value of each width-direction sub-region is obtained respectively; Generate the initial visual width state fingerprint according to the arrangement order of the original optical response values ​​of each width direction sub-region; The initial fingerprint of the visual width state of the thin film segment to be confirmed in the downstream detection area is sequentially matched with the initial fingerprint of the visual width state of the candidate thin film segment in the upstream detection area. If the two meet the preset matching conditions, they are confirmed to belong to the same film segment; if the two are misaligned, the film segment identification is corrected according to the misalignment direction; if the two do not meet the preset matching conditions, the film segment to be confirmed is marked as untrustworthy.

4. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, Based on thin film image data, standard thin film sample calibration results, and operational condition data, the corrected thickness correlation characteristic values ​​of the same thin film segment in each detection area and each width-direction sub-region are obtained, including: Based on the thin film image data of the width direction sub-region, overexposed pixels, underexposed pixels, and edge-occluded pixels are removed, and the original optical response representative value is obtained based on the optical response value corresponding to the remaining effective pixels. Based on the calibration results of the standard thin film sample, the original optical response representative value is converted into the original thickness-related characteristic value; The tension compensation amount is obtained by querying the tension compensation rule based on the film tension data, the speed compensation amount is obtained by querying the speed compensation rule based on the film running speed data, and the temperature compensation amount is obtained by querying the temperature compensation rule based on the film temperature data. Based on the original thickness-related feature value, the tension compensation amount, the speed compensation amount, and the temperature compensation amount, the corrected thickness-related feature value is obtained; The tension compensation, speed compensation, and temperature compensation are all thickness-related feature compensation amounts that have been pre-converted to units consistent with the original thickness-related feature values ​​through calibration.

5. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, The stability boundary is determined based on the dynamic stable thin film segment library, including: Thin film segments that pass continuously through at least two detection regions are considered as candidate stable thin film segments; If a candidate stable thin film segment is not marked as untrustworthy in the same segment identity, the corresponding standard reference imaging area is not marked as untrustworthy in measurement, and the corresponding width state fingerprint does not show irreversible thickness uniformity abnormalities, then the candidate stable thin film segment will be added to the dynamic stable thin film segment library. For the same detection area and the same width direction sub-region, the corresponding correction thickness correlation feature values ​​in the dynamic stable thin film segment library are arranged in numerical order; When the number of corrected thickness-related feature values ​​after arrangement is not less than four, the first and last values ​​are removed, and the first value of the remaining sequence is taken as the first stable boundary, and the last value of the remaining sequence is taken as the second stable boundary. When the number of corrected thickness-related feature values ​​after arrangement is less than four, the first and second stable boundaries corresponding to the previous acquisition cycle are used, or the initial first and second stable boundaries obtained from the standard thin film sample calibration stage are read.

6. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, Generate a width state fingerprint based on the corrected thickness-related feature value and the stable boundary, including: For each width-direction sub-region of the film segment to be evaluated in each detection area, the corresponding corrected thickness-related feature value, first stable boundary, and second stable boundary are read. If the corrected thickness-related feature value is less than the first stable boundary, the corresponding width-direction sub-region is marked as a thinner side anomalous state. If the corrected thickness-related feature value is greater than or equal to the first stable boundary and less than or equal to the second stable boundary, then the corresponding width-direction sub-region is marked as stable. If the corrected thickness-related feature value is greater than the second stable boundary, the corresponding width-direction sub-region is marked as an anomalous state on the thicker side. According to the arrangement order of the width-direction sub-regions in the width direction of the plastic film, the abnormal state of the thinner side, the stable state, and the abnormal state of the thicker side are arranged in sequence to form the width state fingerprint of the film segment to be evaluated in the corresponding detection area.

7. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, Irreversible fingerprint formation values ​​are obtained based on the width state fingerprints of the same thin film segment in the upstream detection region, downstream detection region, and further downstream detection region, including: Read the width state fingerprint corresponding to the width direction sub-region that can complete the matching of the same thin film segment; The number of sub-regions in the width direction that are stable in the upstream detection area, become abnormal on the thinner side or the thicker side in the downstream detection area, and continue to remain abnormal on the thinner side or the thicker side in the next downstream detection area are counted to obtain the number of continuous anomalies in a single region. The number of consecutive anomalies is obtained by counting the number of consecutive width-direction sub-regions that are in a stable state in the upstream detection area, become an abnormal state on the same side in the downstream detection area, and continue to be in an abnormal state on the same side in the next downstream detection area. The irreversible fingerprint formation value is obtained based on the number of persistent anomalies in a single region, the number of consecutive anomalies, and the number of width-direction sub-regions that can complete the matching of the same thin film segment.

8. The machine vision-based dynamic evaluation method for the uniformity of plastic film thickness according to claim 1, characterized in that, Based on the irreversible fingerprint formation value, the optical drift stripping value, and the historical irreversible fingerprint formation value, determine the dynamic evaluation result of thickness uniformity including anomalous formation segments, anomalous extension segments, or anomalous transmission segments, including: Based on the direction of change of the correction thickness correlation feature value of each width direction sub-region in the upstream and downstream detection areas of the same film segment, the number of width direction sub-regions whose direction of change of correction thickness correlation feature value is consistent with the direction of change of reference optical response of the standard reference imaging area is counted, and the optical drift stripping value is obtained based on the number of width direction sub-regions with consistent direction and the number of width direction sub-regions that can complete the matching of the same film segment. Based on the irreversible fingerprint formation value between the current adjacent detection areas and the historical irreversible fingerprint formation value before the current adjacent detection areas, obtain the abnormal incremental value of the process segment; If the optical drift stripping value meets the optical measurement drift condition, then the corresponding width state fingerprint change is determined to be optical measurement drift; If the abnormal increment value of the process segment is greater than zero, and the abnormal state corresponding to the irreversible fingerprint formation value continues to exist in the downstream detection area, then the production line process segment between the current adjacent detection areas is determined as the abnormal formation segment or the abnormal extension segment. If the irreversible fingerprint formation value is greater than zero and the abnormal increment value of the process segment is zero, then the production line process segment between the current adjacent detection areas is determined as an abnormal transmission segment. The output includes dynamic evaluation results of thickness uniformity for abnormal formation segments, abnormal expansion segments, or abnormal transmission segments.

9. A machine vision-based dynamic evaluation system for the thickness uniformity of plastic films, characterized in that, include: The data acquisition module is used to acquire film image data, standard reference imaging area image data, cumulative film displacement data, and operating condition data of at least three detection areas set sequentially along the running direction of the plastic film. The thin film segment identification module is used to generate a thin film segment identifier based on the accumulated thin film running displacement data, and to identify the same thin film segment in different detection areas based on the relative optical response order of each width direction sub-region in the thin film image data. The feature processing module is used to obtain the correction thickness correlation feature values ​​of the same film segment in each detection area and each width direction sub-region based on film image data, standard film sample calibration results and operating condition data. The fingerprint generation module is used to determine the stable boundary based on the dynamic stable thin film segment library, and generate a width state fingerprint based on the corrected thickness-related feature value and the stable boundary. The evaluation and positioning module is used to obtain irreversible fingerprint formation values ​​based on the width state fingerprints of the same thin film segment in the upstream detection area, downstream detection area, and further downstream detection area; to obtain optical drift stripping values ​​based on the change direction of the corrected thickness-related feature values ​​of the same thin film segment in adjacent detection areas and the change direction of the reference optical response in the standard reference imaging area; and to determine the dynamic evaluation result of thickness uniformity, including abnormal formation segments, abnormal extension segments, or abnormal transmission segments, based on the irreversible fingerprint formation values, the optical drift stripping values, and historical irreversible fingerprint formation values.

10. The machine vision-based dynamic evaluation system for the uniformity of plastic film thickness according to claim 9, characterized in that: The data acquisition module includes a visual data acquisition unit, a reference data acquisition unit, a cumulative displacement data acquisition unit, a film tension data acquisition unit, a film running speed data acquisition unit, and a film temperature data acquisition unit; The visual data acquisition unit is used to acquire thin film image data; The reference data acquisition unit is used to acquire image data of the standard reference imaging area and to obtain the direction of reference optical response change of the standard reference imaging area based on the image data of the standard reference imaging area. The cumulative displacement data acquisition unit is used to acquire cumulative thin film displacement data; The thin film tension data acquisition unit is used to acquire thin film tension data; The thin film running speed data acquisition unit is used to acquire thin film running speed data; The thin film temperature data acquisition unit is used to acquire thin film temperature data; The evaluation and positioning module is also used to determine the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment based on the irreversible fingerprint formation value, optical drift peeling value, and abnormal increment value of the process segment, and to send the dynamic evaluation results of the thickness uniformity including the abnormal formation segment, abnormal expansion segment, or abnormal transmission segment to the result display module, communication module, and production line control equipment.

Images