Laminating machine defect backtracking and closed-loop control method and system

Abstract

The invention discloses a defect backtracking and closed-loop control method and system for a film laminating machine. The system collects tension, deviation, pressure, temperature, linear speed and other multi-source data and establishes a cache; after defects are detected through online vision, the defect generation moment is inversely calculated based on linear velocity integration, and a backtracking window is constructed in combination with the defect length, the average velocity, the fixed margin and velocity fluctuation; a window is projected to an unwinding section, a pressing section and a winding section according to a transmission distance table to achieve time alignment, the tension change rate root-mean-square, the deviation change rate root-mean-square, correlation coefficients, pressure ripple crest-peak values, temperature insufficiency and other characteristics are calculated, a two-stage discrimination tree is adopted to recognize root causes, and a necessary actuator set is generated; and only solving the minimum change control quantity of the set through limited quadratic programming under the constraint of amplitude limiting and change rate, and issuing the minimum change control quantity. According to the scheme, the defect density, the rework rate and the shutdown risk can be reduced, and the laminating stability and yield are improved.

Description

Technical Field

[0001] This invention relates to the field of laminating machines, and more particularly to a method and system for defect backtracking and closed-loop control of laminating machines. Background Technology

[0002] Lamination technology is widely used in packaging printing, lithium battery separator and electrode protection, optical film and functional film lamination, textile fabric composites, architectural decorative films, and label and release material processing. A typical laminator usually includes an unwinding mechanism, traction and guide roller system, pressure roller group, heating unit, web guiding mechanism, and rewinding mechanism. It achieves continuous lamination of two or more substrates (such as paper / film, film / film, coated substrate / protective film, etc.) through tension control, speed synchronization, web guiding correction, and temperature / pressure regulation. Due to the characteristics of roll-to-roll production, such as high speed, long-distance transmission, strong multi-variable coupling, and significant batch-to-batch material differences, common quality problems during lamination include: wrinkles (creases, ripples), registration deviations caused by misalignment, air bubbles and localized delamination, indentations and streaks caused by uneven pressure, and unstable bonding strength due to insufficient heating. The aforementioned defects are often characterized by "discrete occurrence locations, multiple sources of generation mechanisms, and delayed characterization." Once formed, these defects may be amplified during subsequent winding, causing large sections to be scrapped and resulting in significant material and downtime losses for enterprises. Therefore, how to achieve online defect identification, accurate backtracking of defect occurrence times, and root-cause-oriented closed-loop control during the operation of laminating machines is a continuously focusing technical direction in this field.

[0003] In existing technologies, one type of solution focuses on the tension and registration (or repeat length) control of the laminating machine. This is achieved through closed-loop adjustment of the unwinding, traction, and rewinding drives, maintaining stable tension and synchronization of the material near the pressing point. For example, US Patent 5813587A provides a control approach for laminating machines, focusing on stabilizing parameters such as material tension, speed, and registration / repeat length through sensing and control, thereby reducing quality fluctuations caused by unstable material feeding. The advantage of this type of solution is its direct effect on "continuous feeding stability," making it suitable for solving macroscopic deviation and dimensional errors caused by tension fluctuations and speed asynchrony. However, in actual high-speed lamination, defects are often caused not only by single tension or speed deviations but also by a combination of factors such as dynamic edge deviation, pressing pressure ripple, insufficient thermal conditions, and localized uneven material thickness. Traditional tension / alignment closed-loop control alone is usually insufficient to achieve closed-loop process management of defects, including "event-level location, backtracking, root cause identification, and minimal intervention." On the one hand, defects often form at the pressing point but are only observed downstream, resulting in significant time and space lag. On the other hand, defect occurrence is localized and random, and traditional control tends to prioritize global stability, which can easily lead to the risk of secondary defects caused by "over-adjusting the entire system to correct local defects."

[0004] Secondly, another type of solution focuses on defect detection and location recording during the lamination / coating process. This involves online defect detection through image acquisition and defect recognition algorithms, storing information such as the defect location and detection results for subsequent processing. For example, US Patent 8045151B2 relates to a method for detecting and handling lamination film defects, emphasizing the detection of defects and the recording / storage of inspection information such as the location of defect occurrence. This type of solution can improve quality traceability, providing a basis for subsequent rejection, rework, or process review. However, its focus is usually limited to "discovering and recording defects," lacking deep closed-loop linkage with the core actuators of the laminating machine (tension, correction, pressure, temperature, linear speed, etc.). Even if the location of a defect is known, it may not be possible to accurately trace back to the process state corresponding to the defect's occurrence time. Furthermore, it lacks reproducible judgment logic for the root cause of defects and root cause-oriented control strategy constraints. In scenarios with fluctuating speeds, long transmission distances, and significant dynamic differences between segments, simply calculating delays based on defect locations can easily lead to backtracking errors, resulting in inaccurate selection of adjustment targets or unstable parameter tuning effects.

[0005] Furthermore, existing technologies propose integrated solutions for visual inspection lines and mechanisms for roll / film materials, combining visual inspection, feeding and correction, and tension adjustment mechanisms to meet the online inspection needs of materials such as optical films. For example, Chinese patent CN107449785B provides an intelligent visual inspection mechanism for optical films, demonstrating the engineered configuration of the visual inspection unit with feeding, correction, and tension adjustment mechanisms within the same system. Such solutions facilitate the coordinated arrangement of inspection and conveying control on the production line and possess good engineering feasibility. However, for quality control of laminating machines, a simple combination of "inspection mechanism + correction / tension mechanism" is not equivalent to a "defect-driven closed-loop control method." In actual working conditions, different defect types have different sensitivities to the controlled object: for example, wrinkles are often related to the coupling of tension change rate and deviation change rate; bubbles and debonding are more affected by insufficient temperature and pressure ripple; edge defects caused by deviation are related to correction overshoot or uneven edge friction. Without a clear and reproducible root cause identification mechanism and actuator selection mechanism, the system is prone to falling into the experience-based operation of "adjusting multiple parameters simultaneously after detecting defects", which leads to mutual interference of control actions and may even cause new defects (such as overcorrection causing secondary wrinkling, pressure / temperature sudden change causing local indentation or uneven bonding), ultimately making it difficult to steadily improve yield.

[0006] In summary, while existing technologies offer beneficial solutions in areas such as tension / registration control of laminating machines (e.g., US5813587A), lamination defect detection and position recording (e.g., US8045151B2), and integration of vision inspection and feeding control mechanisms (e.g., CN107449785B), However, the following common shortcomings still exist: First, defects are mostly formed at the pressing point and detected downstream, resulting in a time and space lag. Existing solutions often use simplified time delay relationships, making it difficult to accurately trace back the defect generation time and its corresponding segmented process state under speed fluctuation conditions. Second, there is a lack of discrimination logic to establish a causal relationship between defect type / location and multivariate process signals (tension, deviation, pressure, temperature, speed), making it difficult to reliably distinguish root cause categories such as "correction-driven, tension-driven, pressure fluctuation, and insufficient thermal performance". Third, there is a lack of actuator selection and minimum intervention adjustment mechanisms oriented towards root causes, which easily leads to problems such as "excessive adjustment, mutual coupling, and secondary defects". Fourth, there is insufficient version binding and traceable evidence chain between defect handling and control elements such as process parameters, thresholds, and weights, making it difficult to implement a closed loop of review and continuous improvement. Therefore, it is necessary to propose a comprehensive technical solution that can realize "defect backtracking - root cause identification - actuator selection - constrained minimum modification control - version traceability" during the operation of the laminating machine, so as to improve the defect suppression effect and process reproducibility, reduce the risk of secondary defects caused by experience-based parameter tuning, and improve the efficiency of quality traceability and process optimization. Summary of the Invention

[0007] This invention aims to address the problems of delayed defect detection, difficulty in accurately tracing the defect generation time, difficulty in reproducible root cause identification, and the tendency of empirical linkage of multiple actuators to cause secondary defects in high-speed roll-to-roll operation of laminating machines. It provides a method for back-calculating the defect generation time and constructing a backtracking window based on online defect detection. It combines segmented projection alignment, two-level discriminant tree root cause identification, and determination of the set of required actuators. Under the constraints of amplitude and rate of change, it implements minimum modification closed-loop control and versioned traceability storage, thereby achieving precise suppression of defects such as wrinkles, deviation, bubbles / debonding and stable yield improvement.

[0008] To achieve the objectives of this invention, the following technical solution is adopted:

[0009] A closed-loop control method for defect backtracking, root cause identification, and actuator selection in a laminating machine is disclosed. This method is applicable to laminating machines with unwinding, pressing, and rewinding sections, a web guiding mechanism, a heating unit, and pressing rollers. It collects data on tension, web deviation, pressing pressure, temperature, and linear velocity. Visual inspection outputs the defect type, defect length, and defect position in the width and feed direction. The defect generation time is calculated by inversely using the formula "the integral of linear velocity over time equals the sum of the distance from the camera to the pressing point and the defect's position in the feed direction." The backtracking window is determined by dividing the defect length by the average window speed + a fixed margin + an expansion proportional to the root mean square of the speed fluctuation. The backtracking window is projected onto the unwinding, pressing, and rewinding sections according to the transmission distance table for each section, and time alignment is achieved. Within the backtracking window, the root mean square of the tension change rate, the root mean square of the web deviation change rate, the correlation coefficient between tension and web deviation, the peak-to-peak value of the pressure ripple, and the temperature deficit are calculated. A two-level discrimination tree is used for... Root cause identification: First, distinguish between correction-dominated and tension-dominated types based on correlation coefficient thresholds. Then, under correction-dominated types, further subdivide them into pressure fluctuation type and thermal insufficiency type based on pressure ripple threshold or temperature deficit threshold, and generate a set of mandatory actuators accordingly. When correction-dominated, the set must contain at least a correction actuator; when tension-dominated, the set must contain at least an unwinding or rewinding tension actuator. Only for this set, a constrained quadratic programming solution is used to minimize the control change, minimizing the sum of the "control change square term and defect risk function". The defect risk function includes the product of tension change rate and deviation change rate, the absolute value of the correlation coefficient, the pressure ripple term, and the temperature deficit term. The weights are determined by a discrete weight table of "width partition × segment" and can only be updated during roll switching or batch switching. Independent amplitude and rate of change constraints are applied to each actuator. Corrections are issued, and defects, windows, root causes, weight table version numbers, and control trajectories are bound and stored.

[0010] As a further improvement, the width partition includes at least a left zone, a middle zone, and a right zone, wherein the left zone is 0-20% of the effective width, the middle zone is 20-80%, and the right zone is 80-100%. When the defect is located in the left zone or the right zone, the discrete weight table increases the weight related to deviation in the defect risk function and reduces the upper limit of the maximum rate of change of the deviation correction actuator.

[0011] As a further improvement, the segment discrete weight table shall at least satisfy the following: the weight of the pressing segment for the pressure ripple term and the temperature deficit term is higher than that of the unwinding segment and the winding segment; the weight of the unwinding segment for the tension-deviation coupling term is higher than that of the pressing segment; and the weight of the winding segment for the tension-deviation coupling term is not lower than that of the pressing segment.

[0012] As a further improvement, the correlation coefficient threshold, pressure ripple threshold, and temperature underestimation threshold are determined by the statistical quantiles of historical qualified rolls and are updated once when the roll is switched or the batch is switched. The correlation coefficient threshold is taken as the 90th percentile of the absolute value of the correlation coefficient, the pressure ripple threshold is taken as the 90th percentile of the peak-to-peak value of the pressure ripple, and the temperature underestimation threshold is taken as the 90th percentile of the temperature underestimation.

[0013] As a further improvement, the independent limit and rate of change constraints include setting maximum change and maximum rate of change for four types of actuators: correction, tension, pressure and temperature. When correction is dominant, the maximum rate of change of the correction actuator is no higher than 0.7 times the maximum rate of change of the tension actuator.

[0014] Another aspect of the present invention provides a closed-loop control system for defect backtracking, root cause identification, and actuator selection in a laminating machine, applied to the laminating machine. The system includes: a data acquisition module, a visual inspection module, a backtracking time calculation and window generation module, a segment projection and time alignment module, a feature calculation module, a two-level discriminant tree root cause identification module, a mandatory actuator set generation module, a constrained quadratic programming solution module, and a traceability storage module. The backtracking time calculation and window generation module is used to calculate the defect generation time based on the linear velocity integral relationship and generate a backtracking window according to the defect length, average window speed, fixed margin, and speed fluctuation. The two-level discriminant tree root cause identification module is used to output the root cause category based on the correlation coefficient threshold, pressure ripple threshold, and temperature under-limit threshold. The mandatory actuator set generation module is used to generate a mandatory set that includes at least a correction actuator or an unwinding / rewinding tension actuator. The constrained quadratic programming solution module is used to solve for the minimum modification control quantity only on the mandatory set and apply amplitude and rate of change constraints. The traceability storage module is used to bind and store the defect, window, root cause, weight table version number, and control trajectory.

[0015] As a further improvement, the discrete weight table is fixed in the form of a configuration file, which includes a version number and activation conditions, and can only be loaded or updated during roll switching or batch switching; the segment projection and time alignment module stores the transmission distance table of each segment, and projects the same backtracking window onto the sampling clocks of the unwinding segment, the pressing segment and the rewinding segment respectively, and outputs the aligned feature calculation data stream.

[0016] As a further improvement, the root cause discrimination module outputs a confidence level. When the confidence level is lower than the threshold, the mandatory actuator set generation module switches the mandatory set to a conservative set. The conservative set includes at least linear velocity reduction and temperature compensation, and triggers at least one of alarm or flag removal. The traceability storage module generates associated records of defect location, defect length, defect generation time, traceability window, root cause category, threshold parameter, weight table version number, and control trajectory, and supports playback of defect map and control trajectory overlay data by volume number.

[0017] A third aspect of the present invention provides a computer device, the computer device including a processor, a graphics processing unit (GPU) and a memory, the memory storing a computer program, which, when executed by the processor and the GPU, causes the computer device to perform the method described thereon.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program causing the computer device to perform the method when executed by a computer device.

[0019] This invention, through a closed-loop link of "defect location—variable speed integral backtracking—defect length / speed fluctuation adaptive windowing—segmented projection alignment—two-level discriminant tree root cause identification—mandatory actuator set—constrained quadratic programming minimum modification control—versioned traceability," enables the laminating machine to accurately locate the time period of defect formation at the pressing point and its corresponding unwinding, pressing, and rewinding segmented process states, even when defects are detected downstream. This significantly reduces backtracking errors caused by speed fluctuations and transmission distances. Simultaneously, based on a two-level discriminant tree of causal characteristics such as tension change rate, deviation change rate, tension-deviation correlation, pressure ripple peak-to-peak value, and temperature deficiency, it achieves reproducible classification of root causes such as "deviation correction dominance, tension dominance, pressure fluctuation, and insufficient thermal performance," and forcibly generates a mandatory actuator set, thereby avoiding mutual interference caused by simultaneous adjustment of multiple parameters in traditional methods. Furthermore, this invention employs constrained quadratic programming that solves only for the set of required actuators. Under constraints of amplitude and rate of change, it optimizes the control quantity with minimal modifications. It also utilizes a discrete weighting table of "width partition × segment" to differentiate the weighting of the defect risk function. This allows the control strategy to adaptively suppress defects such as wrinkles, bubbles / debonding, indentation streaks, and registration deviations, taking into account process characteristics such as the tendency for deviation in edge areas and the greater sensitivity of the pressing section to pressure and heat. This reduces secondary wrinkling caused by over-adjustment, correction overshoot, or pressing fluctuations, thereby improving yield and production stability. Simultaneously, by binding and storing defects, backtracking windows, root cause categories, threshold / weighting table version numbers, and control trajectories, an auditable chain of evidence is formed. This facilitates rapid review, identification of process shortcomings, and continuous optimization, achieving a closed-loop quality improvement from "defect discovery" to "cause localization and stable correction." Attached Figure Description

[0020] Figure 1 A schematic diagram of the closed-loop control system for defect backtracking, root cause analysis, and actuator selection in a laminating machine. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0022] I. Terminology and Abbreviations

[0023] 1. Laminating machine: A device that, under continuous feeding conditions, unwinds, guides, heats, presses, laminates, and rewinds roll-to-roll materials (film / paper / coated substrate) into finished products. It typically includes an unwinding section, a pressing section, and a rewinding section, and is usually equipped with a web guiding mechanism, tension control, pressing pressure control, and temperature control (see [link to documentation]). Figure 1 ).

[0024] 2. Unwinding section U / Pressing section P / Rewinding section W: These are process sections divided according to the material path. Unwinding section U is responsible for unwinding and establishing initial tension; pressing section P is the key section for defect formation (temperature, pressure, and deviation dynamics are superimposed here); rewinding section W is responsible for stabilizing the end tension and winding shape.

[0025] 3. Defect events: Quality anomalies identified by the vision inspection module, including fields such as defect type, length, width and location, and feed direction.

[0026] 4. Defect generation time The time point at which the defect actually forms at the pressing point (or the critical location for defect formation). Since the camera is usually located downstream of the pressing point, the detection time... With the generation time There is a lag.

[0027] 5. Backtracking Window :around The constructed time interval is used to extract signals related to defect formation, such as tension, deviation, pressure, temperature, and velocity, from the buffer zone and to perform root cause identification and control solutions.

[0028] 6. Transfer Distance Table: Describes the equivalent transfer distance between key measuring points in different process sections and the pressing point. It is used to project the same defect backtracking window onto the sampling clock of the U / P / W segment.

[0029] 7. Feature set A set of reproducible features calculated within the backtracking window, including at least: root mean square of tension change rate, root mean square of deviation change rate, correlation coefficient between tension and deviation, peak-to-peak value of pressure ripple, and underestimation of temperature.

[0030] 8. Two-level discrimination tree: a root cause classification logic with a fixed structure. The first level distinguishes between correction-dominated and tension-dominated types based on correlation thresholds; the second level further subdivides the correction-dominated type into pressure fluctuation type or thermal insufficiency type based on pressure ripple thresholds or temperature deficit thresholds.

[0031] 9. Required set of actuators : A subset of actuators that must be involved in regulation for the root cause category. When correction is dominant, the set must contain at least correction actuators; when tension is dominant, the set must contain at least unwinding or winding tension actuators; when pressure fluctuates, the set must contain at least pressure actuators; when thermally deficient, the set must contain at least temperature or linear velocity actuators.

[0032] 10. Constrained Quadratic Programming: An optimization method for finding the "minimum control change" under given amplitude and rate of change constraints. This invention is limited to... The solution aims to avoid secondary defects caused by empirical linkage of multiple actuators.

[0033] 11. Discrete Weight Table: The width is divided into left / middle / right partitions, and the process is segmented into U / P / W segments. The weights of each item in the defect risk function are given using a discrete table. This table can only be updated during roll switching or batch switching and is linked to a version number for traceability.

[0034] II. System Structure

[0035] 2.1 Mechanisms and Actuators (see...) Figure 1 )

[0036] The laminating machine includes, in order of material feeding direction: unwinding section U (unwinding motor or magnetic powder brake, unwinding tension roller, guide roller group), pressing section P (heating roller / hot plate, pressing roller group, pressure loading mechanism), winding section W (winding motor, winding tension controller, roll diameter estimation unit), correction mechanism (swing roller or transverse guide roller), and online vision inspection frame (camera + light source + mounting reference ruler).

[0037] The following is an example equipment configuration: Unwinding / rewinding servo drive: 1.5~3 kW servo motor + encoder; Pressing pressure loading: any one of cylinder / hydraulic / electric screw loading; Correction actuator: swing roller type EPC or transverse guide roller type EPC; Heating unit: heat transfer oil heating roller or electric heating roller (with temperature control module).

[0038] 2.2 Sensors and Data Acquisition

[0039] At least includes: tension: unwinding tension , winding tension (Optional pre-pressing tension) Deviation: Edge offset (mm); Pressure: Pressing pressure (MPa or kPa) and its ripple; Temperature: pressing temperature (°C) and setting Linear velocity: (m / s or m / min); Visual defect events: defect type, defect length Width and position Material feeding direction and position Detection time Sampling recommendations: 50–200 Hz; The frequency range is 10–50 Hz; the visual data is an event stream. All data is timestamped and enters a circular buffer.

[0040] 2.3 Controller, Edge Computing and Traceability

[0041] The system adopts a combination of "PLC / motion controller + industrial PC / edge controller": the PLC is responsible for basic interlocking, drive control, temperature control, pressure control and safety shutdown; the edge controller is responsible for: defect event access, speed integral backtracking, window construction, segmented projection alignment, feature calculation, discriminant tree root cause identification, mandatory set generation, QP solution and traceability storage.

[0042] 2.4 Configuration File and Version Binding

[0043] The system must maintain at least the following configuration files: F-TRF: Transmission Distance Table (version number TRF-Vx); F-WGT: Discrete Weight Table (version number WGT-Vx); F-THR: Threshold Strategy or Threshold Set (version number THR-Vx); F-LIM: Actuator Limiting and Rate of Change Constraint Table (version number LIM-Vx). Each roll must be fixed with the following records upon unwinding: roll number, batch number, and TRF / WGT / THR / LIM version number; changes are not allowed within the roll (unless the system enters "conservative mode" according to the S6 exception handling procedure and records the event code).

[0044] III. Specific Technical Route for Implementing the Method of the Invention

[0045] 1) Collect multi-source process data and establish a cache;

[0046] 2) Visually identify defects and output defect events;

[0047] 3) Use linear velocity integration to inversely calculate the defect generation time. And build a backtracking window ;

[0048] 4) Project the window onto the U / P / W segment using the transfer distance table and align it in time;

[0049] 5) Calculate the fixed feature set ;

[0050] 6) The two-level discriminant tree outputs the root cause categories and generates a set of required executors. ;

[0051] 7) Only for To solve for the minimum control change using constrained QP, the weights are derived from a discrete weight table of "width partition × segment".

[0052] 8) Issue corrections and create an auditable traceable record of "defect - window - root cause - version - control track".

[0053] IV. Detailed Implementation of Method Steps (S1~S6)

[0054] S1: Multi-source data acquisition, unified timestamp and cache management

[0055] The goal of S1 is to ensure that under high-speed roll-to-roll conditions, multi-source signals can be uniformly acquired, aligned, and subsequently recalled, avoiding engineering problems such as "defect detection downstream, parameters upstream, and timing asynchrony leading to backtracking deviations".

[0056] S1.1 Sampling Channels and Sampling Strategies

[0057] The system establishes a sampling channel: tension Deviation ,pressure ,temperature ,speed Sampling period recommendation: for (100 Hz) for (20 Hz). All channels must be timestamped. The timestamps come from a unified clock source (PLC master clock or PTP synchronous clock) to avoid clock drift between different acquisition boards.

[0058] S1.2 Signal Preprocessing and Quality Marking

[0059] To enable subsequent calculations , The rate of change characteristics are more stable, so a "lightweight filtering + outlier marking" strategy is adopted: for , , A moving average was used (window 0.1–0.2 s); for Median filtering (5-point window) is used to suppress occasional spikes; for A first-order low-pass filter (time constant 0.5–1.0 s) was used. A quality marker was added to each sampling point. : This indicates that the statement is valid; This indicates a missing / exceeded limit / communication error. If the speed... In the event of a short-term missing value, the most recent valid value is used to retain and record the event code.

[0060] S1.3 Circular Cache and Index

[0061] The system establishes a buffer that covers at least: the maximum transmission time from the camera to the pressing point + the maximum backtracking window + safety redundancy. Example: distance from the camera to the pressing point... Speed ​​range The slowest latency is approximately 1.6 seconds; if the longest defect length corresponds to a window of approximately 1.2 seconds, speed fluctuations expand the window by approximately 0.6 seconds, and a redundancy of 1 second is added, then a buffer size of ≥5 seconds is recommended. The buffer uses "timestamp index + sequential storage" and supports... Quickly slice and extract interval data.

[0062] S1.4 Process Section Calibration and Loading of "Transfer Distance Table"

[0063] During equipment installation or annual maintenance, the distances to key points are calibrated on-site using a measuring tape / laser rangefinder, and a Transfer Distance Table (TRF) is generated. An example table is shown in Table 1. This table is loaded and locked with version number TRF-Vx upon unwinding each roll, and changes are not permitted within the roll.

[0064] Table 1. Examples of transmission distance representation

[0065]

[0066] After S1 is completed, the system has a "traceable data base" to provide reliable input for S3 to S6.

[0067] S2: Online visual defect event generation, field standardization, and deduplication / merging

[0068] The goal of S2 is to standardize "image recognition results" into defective events that can be controlled, and to ensure that the events are not over-regulated due to repeated recognition.

[0069] S2.1 Defect Event Field

[0070] Each defect event must include at least: defect type (Enumeration codes, such as WR: wrinkle; BB: bubble; DL: debonding; ST: streaks / indentations, etc.); Defect length (mm); Width Position (mm, 0~) ); material feeding direction position (mm, defined as the downstream distance of the defect relative to the camera baseline); Detection time (Unified clock with S1).

[0071] S2.2 Defect Length Calculation and Calibration

[0072] For linear scan cameras, the result can be obtained by multiplying the number of rows spanned by the defective connected component by the row spacing. For array cameras, the pixel length of the defect bounding box in the feed direction can be converted. Pixel-to-millimeter calibration must be completed during the commissioning phase, and the calibration parameters must be written into the vision configuration file and associated with the roll number traceability.

[0073] S2.3 Event Deduplication and Merging

[0074] The system merges defects using the "temporal nearest neighbor + spatial nearest neighbor" rule: when two adjacent events satisfy... and If the defect types are the same, they are merged into the same event. Take the length of the union of the two sets. The first detection time is selected. This rule can significantly reduce the frequent adjustments caused by "the same wrinkle being identified multiple times in consecutive frames".

[0075] S2.4 Defect Severity Classification and Counting Standards

[0076] To match the style of the project materials, the system allows you to set the severity: Major / Minor. Example: Wrinkles Major if it is a major, otherwise Minor; equivalent diameter of the bubble For Major, etc.

[0077] S3: Defect Generation Timing Back Calculation and Backtracking Window Construction

[0078] When speed During fluctuations, the arrival time of the defect from the contact point to the camera no longer satisfies the simple... The generation time must be determined by "velocity integral inverse calculation" and the window must be determined by "defect length + velocity fluctuation" to improve the accuracy and repeatability of backtracking.

[0079] S3.1 Inverse calculation of velocity integral to generate time

[0080] The material travel distance from the pressing point to the camera baseline is [missing information]. The following relationship is used to inversely calculate the time. :

[0081] ;

[0082] Parameter definition: This is the moment when the defect is generated; For the detection time; Linear velocity; The distance from the camera to the pressing point; This indicates the location of the defective material in the feed direction. Within the buffer area, from... Perform numerical integration forward to accumulate distance. First satisfaction When interpolation is performed, the value is obtained. .when When the error is derived from pixel conversion and contains errors, the error is covered by a fixed margin. Absorption in S3.2 to avoid affecting stability.

[0083] S3.2 Backtracking Window Length

[0084] Backtracking window length Construct it according to the following formula:

[0085] ;

[0086] Parameter definition: The length of the defect; The average speed of the window (can be taken as...) (Average within 0.3 seconds before and after). This is a fixed margin (to cover visual recognition boundary errors, camera-triggered jitter, etc.). This is the window expansion factor; It is the root mean square of the rate of change of velocity, used to reflect the intensity of velocity fluctuations.

[0087] The root mean square of the rate of change of velocity is calculated using discrete methods.

[0088] ;

[0089] Parameter definition: These are velocity sample values; The sampling period; To calculate the number of points within the window.

[0090] Project Terms:

[0091] A time of 0.15–0.30 s is recommended. A value of 0.10–0.20 is recommended. The recommended calculation window is... In the first 0.5 seconds, when abnormal speed fluctuations occur (such as during winding joints or correction impacts), the window expansion term automatically increases to counteract backtracking errors.

[0092] S3.3 Backtracking Window Start and End and Boundary Clipping

[0093] The default value for the backtracking window is... .when When the cache exceeds its available range, the principle of "forward expansion first, backward pruning" should be followed to ensure at least one layer of coverage. Find the nearby key interval and record the clipping in the event code.

[0094] S4: Perform segmented projection and time alignment according to the transmission distance table.

[0095] S4.1 Segmented Projection

[0096] Known defects form windows at the pressure points We need to find: the unwinding segment window: the signal that is transmitted to the pressing point at the unwinding test point to form a defect; the winding segment window: the signal that is transmitted to the winding test point after the defect is formed; and map the three segments to the same "material segment" semantics for S5 to extract features.

[0097] S4.2 Time inverse calculation with distance as constraint

[0098] Taking the unwinding section as an example, the distance from the unwinding measuring point to the pressing point is... The starting point of the corresponding window for the unrolled segment. satisfy:

[0099] ;

[0100] Parameter definition: This is the starting point of the roll segment window; This represents the distance from the unwinding measurement point to the pressing point in the distance table. (Window endpoint) Can be used in the same way The result is obtained by reverse calculation. The same logic applies to the retractable segment window. Solve by backward calculation or forward integration.

[0101] S4.3 Time Alignment and Resampling

[0102] because and With different sampling frequencies, the system resamples the data for each window segment using a uniform time axis (e.g., 10 ms): low-frequency signals are interpolated linearly; missing points are processed using... Mark and trigger exception handling strategies (e.g., downgrade to a conservative set, or abandon the calculation of certain features and record event codes).

[0103] S5: Feature Calculation and Data Quality Inspection

[0104] S5.1 Root Mean Square of Tension Change

[0105] ;

[0106] Parameter definition: For tension sampling within the window; The sampling period; The number of points.

[0107] S5.2 Root Mean Square Rate of Deviation

[0108] ;

[0109] Parameter definition: These are the sampled values ​​that deviated from the expected value.

[0110] S5.3 Correlation coefficient between tension and deviation

[0111] ;

[0112] Parameter definition: , These are the window mean values, respectively.

[0113] S5.4 Pressure ripple peak-to-peak value

[0114] ;

[0115] Parameter definition: For window pressure sampling.

[0116] S5.5 Temperature underestimation

[0117] ;

[0118] Parameter definition: To set the temperature; This is the actual temperature.

[0119] S5.6 Data Quality Check

[0120] If the percentage of valid samples within the window is less than 95% (e.g.) If there are insufficient points, the event is marked as "insufficient data" and enters the S6 conservative strategy.

[0121] S6: Root cause identification, mandatory set generation, QP solution, distribution and tracing

[0122] S6.1 Two-level discriminant tree

[0123] Level 1: If It is judged to be the main focus of correction; if If it is determined to be tension-dominant, then it is a composite type.

[0124] Level Two: Under the guidance of corrective measures, if It is judged to be a pressure fluctuation type; otherwise, if If it is classified as insufficient thermal performance, it is otherwise classified as dynamic instability due to deviation.

[0125] S6.2 Required Actuator Set

[0126] Corrective measures are the primary focus: It must contain at least a correction actuator. Tension-driven: It must include at least an unwinding tension actuator. Or winding tension actuator Pressure fluctuation type: Includes at least a pressure actuator Insufficient heat treatment type: Includes at least a temperature actuator or linear velocity actuator .

[0127] S6.3 Constrained QP Solution

[0128] The objective of QP is:

[0129] ;

[0130] Parameter definition: For set Corresponding control increment; To weigh the parameters (from the configuration table); This is the defect risk function.

[0131] The risk function always contains four terms:

[0132] ;

[0133] Parameter definition: The weights given for the discrete weight table.

[0134] S6.4 Constraint Structure

[0135] For each actuator Apply:

[0136] ;

[0137] Parameter definition: The maximum change; This represents the maximum rate of change.

[0138] S6.5 Discrete Weight Table (Width Partition × Segment) and Update Rules (Unchanged Within Volume)

[0139] Width Division: Left Side Central District right side area Segment type: U / P / W. Weights are derived from the discrete table WGT-Vx and can only be updated during volume or batch switching; locked within the volume.

[0140] S6.6 Issuance and Tracing

[0141] The system only supports The correction is issued, and other executors maintain their original settings or slowly revert to avoid "global misadjustments." Simultaneously, the stored defect event fields are traced back. , U / P / W projection window, features Root cause category The record includes the content, QP solution, amplitude and rate of change constraints, TRF / WGT / THR / LIM version number, and execution result (defect density trend). This traceability record supports playback of "defect map + control trajectory overlay" by volume number.

[0142] V. Exception Event Code Table

[0143] Table 2 Examples of Exception Event Codes

[0144]

[0145] VI. Examples and Comparative Examples

[0146] 6.1 General pilot-scale conditions (consistent across all groups)

[0147] Materials: PET 25 μm + protective film 20 μm; effective width Single roll length Each group consists of 10 consecutive volumes.

[0148] set up: , , , Target speed (Short-term speed reduction is allowed, but not below) ).

[0149] Defect statistics: Major defects are counted as 1, and Minor defects are counted as 0.5. Defect density. Defined as: in This represents the number of critical defects. This represents the general number of defects. The length of the roll is in meters (m).

[0150] rework rate : in The length of rework / rejection (m).

[0151] 6.2 Discrete Weight Table and Version

[0152] Table 3 Discrete Weight Table

[0153]

[0154] Table 4 Discrete Weight Table

[0155]

[0156] 6.2 Threshold Determination Strategy (THR-V1, updated once during volume switch)

[0157] Pick 90th percentile; Pick 90th percentile; Pick The 90th percentile. Quantile samples are taken from the feature library of the previous batch of "qualified volumes" (e.g., the most recent 30 volumes). Calculated and locked when each volume is opened.

[0158] 6.3 Examples E1-E4 and Comparative Examples C1-C4

[0159] Comparative Example C1

[0160] The vision system only alarms and does not close the loop; it lacks inverse integral calculation, windowing, segment projection, discrimination tree, mandatory set, QP, and weight table version.

[0161] Comparative Example C2

[0162] A defect map is generated and stored, but the control remains a traditional PID controller; no defect-driven adjustment is performed.

[0163] Comparative Example C3

[0164] The generation time is converted using the average velocity; the window is fixed at 0.5 s; there is no velocity fluctuation window expansion; there is no segment projection; and multiple executors are simultaneously fine-tuned according to empirical rules.

[0165] Comparative Example C4

[0166] It has the ability to backtrack and make judgments, but the weights are updated every 30 seconds, which causes the control to be unstable.

[0167] Example E1

[0168] , ; Calculation window 0.5 s; correction-dominant time .

[0169] Example E2

[0170] Based on E1, only The value was increased to 0.18 to verify the robustness of backtracking under velocity fluctuation conditions.

[0171] Example E3

[0172] Based on E1, only WGT-V2 (roll-to-load switching) is replaced, and the pressing section pays more attention to pressure ripple and temperature under-limit.

[0173] Example E4

[0174] Based on E1, the edge region is improved. The upper limit of the correction change rate was lowered (0.6 times that of the middle zone for the edge zone).

[0175] Table 5. Original Records (10 volumes per group, summary of defects and downtime)

[0176]

[0177] Table 6 Statistical results (mean ± standard deviation, n=10 volumes)

[0178]

[0179] Table 7 Defect density (points / km, mean) broken down by defect type

[0180]

[0181] 6.4 Correspondence of Results

[0182] Compared to C1 / C2, E1 to E4 of this invention will reduce the defect density. Reduce rework rate by approximately 50% to 60%. The number of downtimes and downtime durations decreased significantly by approximately 60% to 75%, indicating that the "defect backtracking - root cause identification - mandatory set - minimum modification of restricted QP" approach can reduce over-adjustment and secondary defects.

[0183] Compared to C3, this invention uses "speed integral inverse calculation + The "+speed fluctuation expansion window+segment projection" significantly reduces backtracking error, thus making the adjustment object more precise and the defect suppression more stable.

[0184] Compared to C4, this invention suppresses control oscillations caused by intra-volume weight drift through "discrete weight table + volume / batch switching update + version binding", thus resulting in significantly less downtime.

[0185] The gains of E2 / E3 / E4 correspond to: windowing robustness, enhanced risk term of the compression segment, and enhanced edge region suppression, respectively.

[0186] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

Claims

1. A closed-loop control method for defect tracing, root cause identification, and actuator selection in a laminating machine, applicable to laminating machines having an unwinding section, a pressing section, a winding section, a web guiding mechanism, a heating unit, and a pressing roller assembly, characterized in that: This method includes collecting data on tension, deviation, pressing pressure, temperature, and linear velocity; visual inspection outputs defect type, defect length, and defect position in width and feed direction; calculating the defect generation time based on the principle that "the integral of linear velocity over time equals the sum of the distance from the camera to the pressing point and the defect's position in the feed direction," and determining the backtracking window by dividing the defect length by the average window speed + fixed margin + expansion proportional to the root mean square of speed fluctuation; projecting the backtracking window onto the unwinding, pressing, and rewinding sections according to the transfer distance table for each section and completing time alignment; and calculating features such as the root mean square of tension change rate, root mean square of deviation change rate, correlation coefficient between tension and deviation, peak-to-peak value of pressure ripple, and temperature underestimation within the backtracking window. A two-level discriminant tree is used for root cause identification: first, the system distinguishes between correction-dominant and tension-dominant systems based on correlation coefficient thresholds; then, under correction-dominant systems, it further subdivides into pressure fluctuation type and thermal insufficiency type based on pressure ripple thresholds or temperature deficit thresholds, and generates a set of mandatory actuators accordingly. When correction-dominant systems are dominant, the set must contain at least a correction actuator; when tension-dominant systems are dominant, the set must contain at least an unwinding or rewinding tension actuator. Only for this set, a constrained quadratic programming problem is used to solve for the minimum control change, minimizing the sum of the "control change square term and defect risk function". The defect risk function includes a weighted sum of the product of tension change rate and deviation change rate, the absolute value of the correlation coefficient, the pressure ripple term, and the temperature deficit term. The weights are determined by a discrete weight table of "width partition × segment" and are only allowed to be updated during roll switching or batch switching. Independent amplitude and rate of change constraints are applied to each actuator. Corrections are issued, and defects, windows, root causes, weight table version numbers, and control trajectories are bound and stored.

2. The method according to claim 1, characterized in that: The width partition includes at least a left zone, a middle zone, and a right zone, wherein the left zone is 0-20% of the effective width, the middle zone is 20-80%, and the right zone is 80-100%. When the defect is located in the left zone or the right zone, the discrete weight table increases the weight related to deviation in the defect risk function and reduces the upper limit of the maximum change rate of the deviation correction actuator.

3. The method according to claim 1, characterized in that: The segment discrete weight table must at least satisfy the following: the weight of the pressing segment for the pressure ripple term and the temperature deficit term is higher than that of the unwinding segment and the winding segment; the weight of the unwinding segment for the tension-deviation coupling term is higher than that of the pressing segment; and the weight of the winding segment for the tension-deviation coupling term is not lower than that of the pressing segment.

4. The method according to claim 1, characterized in that: The correlation coefficient threshold, pressure ripple threshold, and temperature underestimation threshold are determined by the statistical quantiles of historical qualified rolls and are updated once when the roll is switched or the batch is switched. The correlation coefficient threshold is taken as the 90th percentile of the absolute value of the correlation coefficient, the pressure ripple threshold is taken as the 90th percentile of the peak-to-peak value of the pressure ripple, and the temperature underestimation threshold is taken as the 90th percentile of the temperature underestimation.

5. The method according to claim 1, characterized in that: The independent limit and rate of change constraints include setting maximum change and maximum rate of change for four types of actuators: correction, tension, pressure and temperature. When correction is dominant, the maximum rate of change of the correction actuator shall not be higher than 0.7 times the maximum rate of change of the tension actuator.

6. A closed-loop control system for defect backtracking, root cause identification, and actuator selection in a laminating machine, characterized in that... include: The system comprises a data acquisition module, a visual inspection module, a backtracking time calculation and window generation module, a segment projection and time alignment module, a feature calculation module, a two-level discriminant tree root cause discrimination module, a mandatory actuator set generation module, a constrained quadratic programming solution module, and a traceability storage module. The backtracking time calculation and window generation module calculates the defect generation time based on the linear velocity integral relationship and generates a backtracking window according to defect length, average window velocity, fixed margin, and velocity fluctuation. The two-level discriminant tree root cause discrimination module outputs the root cause category based on correlation coefficient thresholds, pressure ripple thresholds, and temperature underestimation thresholds. The mandatory actuator set generation module is used to generate a mandatory set that includes at least a correction actuator or an unwinding / rewinding tension actuator; the constrained quadratic programming solution module is used to solve for the minimum modification control quantity only on the mandatory set and apply amplitude and rate of change constraints; the traceability storage module is used to bind and store defects, windows, root causes, weight table version numbers and control trajectories.

7. The system according to claim 6, characterized in that: The discrete weight table is fixed in the form of a configuration file, which includes a version number and activation conditions, and can only be loaded or updated during roll switching or batch switching; the segment projection and time alignment module stores the transmission distance table of each segment, and projects the same backtracking window onto the sampling clocks of the unwinding segment, the pressing segment and the rewinding segment respectively, and outputs the aligned feature calculation data stream.

8. The system according to claim 6, characterized in that: The root cause discrimination module outputs a confidence level. When the confidence level is lower than the threshold, the mandatory actuator set generation module switches the mandatory set to a conservative set. The conservative set includes at least linear velocity reduction and temperature compensation, and triggers at least one of alarm or flag removal. The traceability storage module generates associated records of defect location, defect length, defect generation time, traceability window, root cause category, threshold parameter, weight table version number, and control trajectory, and supports playback of defect map and control trajectory overlay data by volume number.

9. A computer device, characterized in that, The computer device includes a processor, a graphics processing unit (GPU), and a memory, wherein the memory stores a computer program that, when executed by the processor and the GPU, causes the computer device to perform the method as described in any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a computer device, causing the computer device to perform the method as described in any one of claims 1 to 5.

Images