Method and system for intelligent mechanical stripping and indium enrichment of waste liquid crystal panel

Abstract

This invention relates to the field of resource recycling technology for waste liquid crystal display panels, and in particular to a method and system for intelligent mechanical stripping and indium enrichment of waste liquid crystal panels. The method includes the following steps: separating two glass substrates; mechanically scraping the coated side of the glass substrates while simultaneously performing visual and XRF inspections to acquire mechanical scraping data, and then obtaining judgment indicators to determine the scraping effect; when the judgment result is insufficient scraping, in-situ cyclical re-scraping and judgment are performed; when the judgment result is over-scraping or the maximum number of re-scraping cycles is reached, manual verification is performed; the glass substrates, liquid crystal components, and scraping debris are collected and stored in separate compartments, and a full-process data traceability database is formed using the complete process data of waste liquid crystal panel stripping. This invention realizes an in-situ real-time closed-loop re-scraping mechanism for waste liquid crystal panels through real-time sensing and intelligent algorithms, eliminating the need for physical reflow and rework, and removing the efficiency bottleneck and accuracy loss caused by workpiece handling.

Description

Technical Field

[0001] This invention relates to the field of resource recycling technology for waste liquid crystal display panels, and in particular to a method and system for intelligent mechanical stripping and enriching indium from waste liquid crystal display panels. Background Technology

[0002] Waste liquid crystal display panels typically consist of two glass substrates, an edge-sealing adhesive layer, a polarizer and adhesive layer, liquid crystal material, and an indium-containing transparent conductive film (ITO). Due to variations in usage years, contamination levels, and adhesive layer aging, the morphology of liquid crystal residue and the adhesion of ITO differ significantly between different batches of panels. Existing mechanical stripping or wet recycling methods often lack online quantitative and closed-loop control to ensure proper removal of liquid crystal residue, ITO, or indium, and to prevent over-stripping. This results in both under-stripping and over-stripping, fluctuations in cycle time, and significant variations in downstream raw material quality. In engineering practice, front-end dismantling / stripping typically includes: dismantling and removing accessories, cleaning and drying, heating and softening with controlled deformation, separating the two glass substrates along the edge seal, removing liquid crystal and adhesive layer residue, and mechanically scraping / grinding the coated side to remove ITO and collecting scraping debris. The accompanying downstream recycling may employ methods such as mechanical enrichment, wet leaching, chlorination volatilization / vacuum evaporation, displacement, or electrolytic refining to obtain crude or high-purity indium.

[0003] Although the above-mentioned process chains have various paths, they all share several common pain points in the front-end "peeling-scraping" stage:

[0004] (1) Open-loop parameters lead to both under-peeling and over-peeling. Different models and batches of panels vary significantly in terms of thickness, adhesive formulation and aging degree, contamination adhesion, edge sealant hardness, and heat deformation behavior. Using fixed heating temperature / time, deformation amount, separation speed, and scraping pressure, speed, and number of passes makes it difficult to simultaneously meet the requirements of "complete peeling" and "low damage". This easily leads to problems such as liquid crystal and adhesive residue, ITO residue (under-peeling), glass scratches, edge chipping, substrate thinning, and dust surge (over-peeling), resulting in fluctuations in yield and cycle time.

[0005] (2) Decoupling of detection and execution, and inconsistency in positioning. Although some technical solutions include visual inspection or component detection, these are mostly offline sampling or only used for recording and sorting. The detection coordinates are inconsistent with the tool travel coordinates, and the detection time point is not synchronized with the processing time point. The detection results cannot generate trajectories or parameter adjustment instructions in real time, thus still relying mainly on open-loop processing. Especially in the high-frequency scenario of "local residue", repeated full-width scraping may improve the removal rate, but it will significantly increase the risk of over-stripping and tool wear, and affect the cycle time. Even with online inspection, if the workpiece is moved to other stations or returned to the starting point of the current station after the decision is made, cycle time loss and positioning error will still be introduced, making it impossible to achieve true real-time precise control.

[0006] (3) Multimodal data drift and registration challenges. Visual inspection is easily affected by reflections, light fluctuations, stains, and polarizer residues; XRF online inspection is affected by changes in the distance between the probe and the sample, temperature drift, geometric errors of the scanning mechanism, and deformation of the shielding structure. Without a unified substrate coordinate system, registration residual evaluation, and online self-calibration, visual pixels and XRF sampling points are difficult to correspond, making it difficult to use the detection information for precise control and local peeling.

[0007] (4) Safety and environmental constraints coexist. Mechanical scraping generates dust and shavings, requiring closed negative pressure, filtration, and separate recycling; XRF online detection involves radiation shielding, access control interlocks, and abnormal handling; solvent / heat treatment may lead to volatilization, odor, and energy consumption issues. Engineered equipment needs to strike a balance between safety compliance, cycle efficiency, quality stability, and green and low-consumption. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent mechanical stripping and enriching indium from waste liquid crystal panels.

[0009] To achieve the above objectives, in a first aspect, the present invention provides an intelligent mechanical peeling method for enriching indium from waste liquid crystal panels. The method includes the following steps: cleaning, drying, and controlled deformation treatment of a fixed waste liquid crystal panel in a sealed space, and then separating two glass substrates along the sealing area; mechanically scraping the coated side of the glass substrate while simultaneously performing visual inspection and XRF inspection on the glass substrate and acquiring mechanical scraping data; obtaining a judgment index using the mechanical scraping data, visual inspection results, and XRF inspection results; performing a fusion judgment on the scraping effect based on the judgment index, and when the judgment result is insufficient scraping, performing in-situ cyclic peeling and judgment based on visual inspection and XRF inspection; performing manual verification when the judgment result is over-peeling or the number of peeling attempts reaches the maximum number of peeling attempts; collecting and storing the glass substrate, liquid crystal components, and scraping debris in separate compartments, and storing the entire process data of the waste liquid crystal panel to form a full-process data traceability database. This invention realizes an in-situ real-time closed-loop repair mechanism for waste LCD panels through real-time sensing and intelligent algorithms, eliminating the need for physical reflow and rework, and removing the efficiency bottleneck and precision loss caused by workpiece handling.

[0010] Optionally, the mechanical scraping data includes at least two of the following: visual scratch probability, glass substrate edge cracking probability, scraper blade wear, motor torque fluctuation amplitude, motor torque fluctuation amplitude, root mean square value of robotic arm vibration signal, and probability of the scraper crossing the prohibited area.

[0011] Optionally, the determination index includes an over-peeling risk index, a confidence level for liquid crystal and adhesive layer identification, a liquid crystal peeling degree index, and an indium peeling degree index, wherein the over-peeling risk index is a weighted sum of the normalized results of at least two of the mechanical abrasion data.

[0012] Optionally, a lightweight deep learning segmentation model is used to perform pixel-level segmentation on the visual detection results, outputting the residual masks of the liquid crystal and adhesive layer, as well as the confidence level, and calculating the liquid crystal peeling degree index, which satisfies the following relationship:

[0013]

[0014] in, This refers to the degree of liquid crystal peeling. The residual areas of the liquid crystal and the adhesive layer are determined based on the residual mask. This represents the initial area of ​​the liquid crystal and the adhesive layer.

[0015] Optionally, an indium elemental intensity distribution map is obtained based on the XRF detection results, and then the indium peeling degree index is calculated based on the indium elemental intensity distribution map. The indium peeling degree index satisfies the following relationship:

[0016]

[0017] in, The degree of indium stripping is an indicator. The average indium strength after scraping. The average indium strength before scraping.

[0018] Optionally, the step of fusing and judging the scraping effect based on the judgment index, and when the judgment result is insufficient scraping, performing in-situ cyclic scraping and judgment based on visual inspection and XRF inspection, includes the following steps:

[0019] The scraping effect is judged by the judgment index to obtain the judgment result, which includes insufficient scraping, over-peeling and qualified peeling.

[0020] When the determination result is insufficient scraping, the glass substrate remains in its original position, and the coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed;

[0021] The initial peeling area is determined using the residual mask and the indium element intensity distribution map, and then the final set of peeling areas is determined by morphological dilation, connected component analysis and noise filtering.

[0022] Based on each of the peeling regions in the set of peeling regions, an adaptive peeling trajectory is generated, and the peeling trajectory is simulated and verified to generate a peeling plan.

[0023] The peeling of the peeling area is completed according to the peeling plan, and the scraping effect of the peeling area is fused and determined based on visual detection and XRF detection.

[0024] If the determination result of the peeling area is insufficient, the next peeling is performed, and the peeling ends when any one of the following conditions is met: the determination result is over-peeling, the determination result is qualified peeling, or the maximum number of peelings is reached.

[0025] Optionally, when the determination result is insufficient scraping, the glass substrate remains in its original position, and coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed, including the following steps:

[0026] When the determination result is insufficient scraping, the glass substrate remains in its original position, the extrinsic parameters of the camera coordinate system and the XRF scanning coordinate system are solved, and a mapping table of visual pixels and XRF sampling points to the substrate coordinate system is established.

[0027] Calculate the registration residual from the camera coordinate system to the substrate coordinate system, and when the registration residual is greater than the residual threshold, call the calibration board to perform extrinsic parameter re-estimation;

[0028] The registration residual is recalculated based on the result of the extrinsic parameter re-estimation, and a downgrade scan is performed if the registration residual is still greater than the residual threshold.

[0029] Optionally, the step of determining the initial peeling area using the residual mask and the indium element intensity distribution map, and then determining the final peeling area set through morphological dilation, connected component analysis, and noise filtering, includes the following steps:

[0030] The intensity distribution of the indium element Figure 2 Values ​​are used to obtain the indium-exceeding area, and then the indium element intensity distribution map is fused with the residual mask through a logical OR operation to obtain the initial peeling area;

[0031] Morphological dilation is performed on the initial patched region, and all independent regions are identified through connected component analysis;

[0032] Independent regions with an area smaller than the area threshold are filtered out as noise regions, and the remaining independent regions are used as the supplementary stripping regions to form the supplementary stripping region set.

[0033] Optionally, the step of adaptively generating a stripping trajectory based on each stripping region in the set of stripping regions, simulating and verifying the stripping trajectory, and then generating a stripping plan includes the following steps:

[0034] Extract the regional features of the patched area, including area and shape;

[0035] Based on the regional characteristics, the peeling trajectory of the peeling region is adaptively generated according to a pre-defined trajectory generation rule;

[0036] The stripping trajectory is simulated and verified to predict the area coverage of the stripping area by the stripping trajectory and the over-stripping risk index under the stripping trajectory.

[0037] If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then the re-peeling plan is composed of the re-peeling trajectory and the original scraping parameters.

[0038] If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then the scraping parameters are adjusted, and then the re-peeling plan is formed in combination with the re-peeling trajectory.

[0039] If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then the re-peeling trajectory is adjusted, and the re-peeling plan is formed by combining the scraping parameters.

[0040] If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then the re-peeling trajectory and the scraping parameters are adjusted, and the adjusted re-peeling trajectory and scraping parameters are used to form the re-peeling plan.

[0041] In summary, this method has at least the following beneficial effects:

[0042] 1. By acquiring mechanical scraping data, visual inspection, and XRF detection to obtain judgment indicators to determine the scraping effect, the glass substrate, liquid crystal components, and scraping debris are collected and stored separately. When necessary, in-situ cyclic peeling and judgment are performed based on visual inspection and XRF detection, as well as manual verification. This forms a strongly coupled closed-loop waste liquid crystal panel peeling method, avoiding the problem of only detecting the peeling effect without being able to control it.

[0043] 2. By registering the camera coordinate system with the XRF scanning coordinate system and performing online self-calibration through the registration residual, drift misjudgment can be reduced, the accuracy of the peeling area can be improved, and the reliability of the closed-loop method for peeling off waste LCD panels can be guaranteed.

[0044] 3. Based on a lightweight deep learning segmentation model, residual masks can be quickly identified. Combined with the indium element intensity distribution map, the re-stripping area can be quickly identified and the re-stripping trajectory can be generated in real time. Then, the re-stripping can be performed in place at the current workstation, realizing "inspection and re-stripping at the same time", which greatly reduces the cycle time loss caused by physical reflow and repositioning.

[0045] 4. By limiting the maximum number of re-peeling operations and setting over-peeling conditions, the risk of over-peeling can be controlled, ensuring that the mechanical peeling of waste LCD panels balances proper peeling with a safety margin.

[0046] 5. Separate collection and storage of glass substrates, liquid crystal components and indium-containing scrap can improve back-end stability. Using the full-process data from the stripping of waste liquid crystal panels to form a full-process data traceability database can support various threshold and window version updates, gray-scale releases and rollbacks, which is conducive to large-scale replication and auditing.

[0047] Secondly, the present invention provides an intelligent mechanical peeling and indium enrichment system for waste liquid crystal panels. This system is applicable to the intelligent mechanical peeling and indium enrichment method for waste liquid crystal panels provided by the present invention. The system includes: a cavity, a conveying and positioning module, a cleaning and drying module, a controlled deformation module, a substrate separation module, an intelligent scraping module, a data acquisition module, a graded and compartmented discharge module, a control center, and a data storage module. The control center is electrically connected to the other modules. The cavity provides a sealed space for the intelligent mechanical peeling of waste liquid crystal panels. The conveying and positioning module conveys the waste liquid crystal panels into the cavity and fixes them within the cavity. The cleaning and drying module cleans and dries the waste liquid crystal panels. The controlled deformation module performs controlled deformation processing on the waste liquid crystal panels. The substrate separation module separates two glass substrates along the sealing edge area. The intelligent scraping module... The system is used for mechanically scraping the coated side of a glass substrate. The data acquisition module acquires the mechanical scraping data and performs visual and XRF inspections on the glass substrate. The tiered and compartmentalized discharge module collects and stores the glass substrate, liquid crystal components, and scraping debris in separate compartments. The control center uses the mechanical scraping data, visual inspection results, and XRF inspection results to obtain judgment indicators, including over-peeling risk indicators, liquid crystal peeling degree indicators, and indium peeling degree indicators. The control center performs a fusion judgment on the scraping effect based on the judgment indicators, and when the judgment result is insufficient scraping, it performs in-situ cyclical re-peeling and judgment based on visual and XRF inspections. The control center performs manual verification when the judgment result is over-peeling or the number of re-peeling attempts reaches the maximum number of re-peeling attempts. The data storage module stores the entire process data of the waste liquid crystal panels, forming a full-process data traceability database. Since this system uses this method, it has at least the same beneficial effects as this method. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic flowchart of an intelligent mechanical stripping and indium enrichment method for waste liquid crystal panels according to an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the glass substrate scraping and peeling trajectory according to an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the framework of an intelligent mechanical stripping and indium enrichment system for waste liquid crystal panels according to an embodiment of the present invention. Detailed Implementation

[0052] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0053] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0054] It should be noted in advance that, in one alternative embodiment, except for independent descriptions, the same symbols or letters appearing in all formulas have the same meaning.

[0055] In one optional embodiment, please refer to Figure 1 This invention provides a method for intelligent mechanical stripping and enriching indium from waste liquid crystal panels, the method comprising the following steps:

[0056] S1. The fixed waste LCD panel is cleaned, dried and deformed in a closed space, and then the two glass substrates are separated along the sealing area.

[0057] Specifically, in this embodiment, the waste LCD panel enters the conveyor mechanism (conveyor belt) via the loading platform and is then conveyed into a sealed space. The positioning clamp at the end of the conveyor mechanism uses a flexible clamping method to fix the waste LCD panel within the sealed space, ensuring it is parallel to the horizontal plane. After fixing the waste LCD panel, a spray device sprays it with 25-40°C pure water or 0.1-0.5wt% neutral cleaning solution at a spray pressure of 0.1-0.3MPa for 30-120 seconds to remove surface contaminants. Simultaneously, a brushing unit uses a flexible nylon brush at a low rotation speed (50-100rpm) to brush the surface of the waste LCD panel, removing stubborn stains remaining after spraying. After cleaning, the waste LCD panels enter the hot air / air knife unit. The hot air unit uses an array of hot air nozzles to blow out hot air at 50-80°C at a wind speed of 3-8m / s to dry the waste LCD panels for 60-180 seconds. At the same time, the air knife unit uses high-pressure airflow to peel off the water film, especially targeting areas prone to water accumulation such as panel edges and structural grooves, to ensure that no liquid droplets remain.

[0058] Furthermore, after the waste LCD panels are air-dried, the bottom surface of the waste LCD panels is heated for 30-180 seconds using a heating plate at a temperature of 80-140°C. Simultaneously, on the other side of the waste LCD panels, with a pressure roller pressing in 0.2-2.0mm, a pressure roller is used to roll the waste LCD panels from one end to the other at a speed of 5-30mm / s. This process allows for controlled deformation of the waste LCD panels, creating a controllable stress difference between the sealing adhesive layer and the glass substrate interface, thus facilitating the subsequent separation of the two glass substrates.

[0059] After the controlled deformation of the waste LCD panel is completed, an initial cut of 0.3-1.0 mm depth is first made at the edge sealing adhesive layer of the waste LCD panel using a diamond cutting wheel, with the cut length covering the short side of the waste LCD panel. Then, a wedge-shaped cutter head is inserted from the cut and advanced along the edge sealing adhesive layer at a speed of 2-20 mm / s, causing relative displacement between the two glass substrates through mechanical wedging force. When a gap of 1-5 mm is formed at the separation interface, a vacuum suction cup is used to adsorb the surface of the upper glass substrate and vertically lift it at a speed of 5-50 mm / s to widen the separation gap, prevent the two glass substrates from re-adhering, and allow the liquid crystal components to flow to a dedicated collection chamber under the waste LCD panel under gravity, providing operating space for subsequent scraping.

[0060] More specifically, the enclosed space can be a cavity with a lead plate at least 3mm thick installed on its inner wall as a radiation shielding structure. The cavity is also equipped with an access control interlock device, a negative pressure exhaust system (negative pressure ≥50Pa), and an emergency stop button. In case of abnormal access control interlock status, insufficient negative pressure, or an emergency, the power can be cut off via the control center or by pressing the emergency stop button to forcibly prevent further actions, ensuring radiation safety and dust control.

[0061] S2. Mechanically scrape the coated side of the glass substrate, while simultaneously performing visual and XRF inspections on the glass substrate and acquiring mechanical scraping data.

[0062] Specifically, in this embodiment, please refer to Figure 2 In (a), after separating the two glass substrates, the coated side of the glass substrate is placed on the cutting platform with the coating side facing up. The robotic arm maintains the angle between the scraper and the coated side of the glass substrate at 25-65° through the angle adjustment mechanism. The constant force actuator controlled by air pressure provides a stable normal pressure of 10-120N (equivalent contact pressure of 0.02-0.25MPa) during the scraping process. The three-axis motion mechanism completes the two-dimensional cutting and Z-axis fine adjustment at a linear speed of 0.05-0.5m / s, and scrapes the coated side of the glass substrate 1-6 times.

[0063] During the mechanical scraping of the coated side of the glass substrate, a cross-polarized light source or an ultraviolet light source is used to enhance the contrast of the residual liquid crystal or adhesive layer, and visual inspection and XRF inspection are performed on the glass substrate. Specifically, an image of the coated side of the glass substrate is obtained using a camera, and an X-ray fluorescence spectrometry (XRF) probe is used to scan the coated side of the glass substrate with a step size of 0.5-5 mm and a residence time of 5-50 ms to obtain the energy and intensity data of the characteristic fluorescent X-rays of indium on the coated side of the glass substrate.

[0064] In addition, mechanical scratching data is required. This data should include at least two of the following: visual scratch probability, glass substrate edge chipping probability, scraper blade wear, motor torque fluctuation amplitude, motor torque fluctuation amplitude, root mean square value of robotic arm vibration signal, and the probability of the scraper crossing a prohibited area.

[0065] In this embodiment, scratch areas in the acquired glass substrate coated side images are labeled, and a scratch recognition dataset is constructed using a large number of labeled glass substrate coated side images. The scratch recognition dataset is then divided into training and validation sets in a 7:3 ratio to complete the training and validation of the U-Net network, resulting in a scratch area recognition model. The scratch area recognition model uses the glass substrate coated side images as input and the scratch area recognition results as output. During training, cross-entropy loss and the Adam optimizer are used, with an initial learning rate of 0.0001 and a batch size of 8. During validation, accuracy and recall are used as evaluation metrics. The scratch area recognition model is input with real-time acquired glass substrate coated side images to identify scratch areas, and the visual scratch probability can be calculated according to the following formula. :

[0066]

[0067] in, The area of ​​the scratch. This represents the area of ​​the image on the coated side of the glass substrate. Specifically, the area is represented by the number of pixels in the scratched area, and the area is represented by the number of pixels in the image on the coated side of the glass substrate.

[0068] Similar to building a scratch recognition model, a crack recognition model can be constructed using a U-Net network. Its input is an image of the coated side of the glass substrate, and its output is the recognition result of cracks at the edge of the glass substrate. The crack recognition model is input with real-time acquired images of the coated side of the glass substrate to identify cracks, and the ratio of the crack area to the area of ​​the coated side image is used as the edge chipping probability. Similarly, the area can be represented by the number of pixels when calculating the edge chipping probability.

[0069] A tool measurement reference station set up next to the scraping station is used to measure the "protrusion" of the cutting edge relative to the tool holder in real time. The difference between the initial protrusion and the protrusion at different times is then used as the wear amount of the scraper cutting edge. The tool measurement reference station and the use of the tool measurement reference station to measure tool wear are existing technologies and will not be described in more detail here.

[0070] The torque and force sensor and torque sensor are used to monitor the motor torque and torque during the scraping process, respectively, and then the motor torque fluctuation amplitude and motor torque fluctuation amplitude are calculated according to the following formulas:

[0071]

[0072]

[0073] in, This refers to the amplitude of motor torque fluctuation. For maximum motor torque, To minimize motor torque, This refers to the fluctuation range of motor torque. This is the maximum motor torque. This represents the minimum motor torque.

[0074] Vibration sensors are used to collect and sample the vibration signals of the robotic arm. The root mean square error of the vibration amplitude is then calculated using the vibration amplitude at each sampling point, which is used as the root mean square value of the robotic arm's vibration signal.

[0075] The movement trajectory of the scraper is monitored using path sensors, such as the displacement sensors of a robotic arm. No-peel zones are defined, for example... Figure 2 In the edge region of the glass substrate shown in (b), each time the scraper enters these forbidden peeling areas during the cutting process, it is considered a violation event. The probability of the scraper crossing the forbidden peeling area is then calculated according to the following formula:

[0076]

[0077] in, The probability of illegally crossing the prohibited area with a knife. The number of violations This represents the total number of cuts.

[0078] S3. Obtain judgment indicators using the mechanical scraping data, visual inspection results, and XRF inspection results.

[0079] Specifically, in this embodiment, the determination indicators include the over-peeling risk indicator, the confidence level of liquid crystal and adhesive layer identification, the liquid crystal peeling degree indicator, and the indium peeling degree indicator.

[0080] The over-peeling risk index is a weighted sum of the normalized results of at least two mechanical scratching data points. For the probability of visual scratches, the probability of glass substrate edge breakage, and the probability of non-compliance, the calculation results from step S2 can be directly used during the weighted summation without normalization, as the probabilities themselves are within the range [0,1]. For any one of the following data points: scraper blade wear, motor torque fluctuation amplitude, motor torque fluctuation amplitude, and the root mean square value of the robotic arm vibration signal, outlier identification and removal are first performed based on the interquartile range method, then missing values ​​are filled using the median filling method, and finally... It is normalized and used to calculate the over-exploitation risk index, where, For the normalized i-th type of mechanical scraping data, For the i-th type of mechanical scraping data, The 0.95 quantile of the data for the i-th type of mechanical scraping. It is the 0.995 quantile of the i-th type of mechanical scraping data.

[0081] Furthermore, when calculating the over-peeling risk index, the weights of various mechanical scratching data are determined as follows: Given the selected types of mechanical scratching data for calculating the over-peeling risk index, historical mechanical scratching data is collected, and the presence or absence of over-peeling on the corresponding glass substrate is determined through manual inspection, with each substrate labeled accordingly. A logistic regression model is then used to fit the data. , This refers to the number of glass substrates labeled as "over-peeled". This represents the total number of glass substrates. For the sigmoid function, For the cutoff item, Let be the coefficient of the i-th type of mechanical scraping data; and let be The weights are mapped to the corresponding mechanical scraping data, i.e. , is the weight of the i-th type of mechanical scraping data.

[0082] A large number of images of the coated side of glass substrates, containing liquid crystal and adhesive layer residues with varying degrees of peeling, were collected. These images were preprocessed (noise reduction, cropping) and then pixel-level labeled to identify which pixels belonged to liquid crystal residue, adhesive layer residue, or background. This allowed for the generation of accurate residue masks for liquid crystal and adhesive layers. A dataset was constructed using the coated side images of the glass substrates and the residue masks. The dataset was divided into training and validation sets in a 7:3 ratio. The coated side images of the glass substrates were used as input, and the corresponding residue masks and prediction confidence scores were used as outputs. The U-Net model was trained and validated using the training and validation sets, resulting in a lightweight deep learning segmentation model. During training, cross-entropy loss and the Adam optimizer were used, with an initial learning rate of 0.0001 and a batch size of 8. During validation, the evaluation metrics included the Dice coefficient, accuracy, and recall.

[0083] After obtaining the lightweight deep learning segmentation model, it is used to perform pixel-level segmentation on the visual inspection results, outputting the residual masks of liquid crystal and adhesive layer, as well as the confidence scores, and calculating the liquid crystal peeling degree index. The visual inspection results are the coated side images of the glass substrate. The liquid crystal peeling degree index satisfies the following relationship:

[0084]

[0085] in, This is an indicator of the degree of liquid crystal peeling. To determine the residual areas of the liquid crystal and the adhesive layer based on the residual mask, This represents the initial area of ​​the liquid crystal and the adhesive layer. Specifically, the image area can be calculated based on the size of the image on the glass substrate coating side. Simultaneously, the ratio of the number of residual mask pixels to the number of pixels in the glass substrate coating side image can be calculated. Multiplying this ratio by the image area yields the final image area. ; This represents the area of ​​the liquid crystal and adhesive layer before this scraping process.

[0086] The indium elemental intensity distribution map is obtained based on the XRF detection results, and then the indium peeling degree index is calculated based on the indium elemental intensity distribution map. The XRF detection results refer to the energy and intensity data of the characteristic fluorescent X-rays of indium. The indium peeling degree index satisfies the following relationship:

[0087]

[0088] in, As an indicator of the degree of indium stripping, The average indium strength after scraping. This represents the average indium strength before scraping. Specifically, This is the average value of the indium intensity at each point in the indium intensity distribution map after this scraping. This is the average value of the indium intensity at each point in the indium intensity distribution map before this scraping.

[0089] S4. The scraping effect is fused and judged according to the judgment index, and when the judgment result is insufficient scraping, in-situ cyclic scraping and judgment are performed based on visual detection and XRF detection.

[0090] Step S4 specifically includes the following steps:

[0091] S41. The scraping effect is judged by the judgment index to obtain the judgment result, which includes insufficient scraping, over-peeling and qualified peeling.

[0092] Specifically, in this embodiment, when the confidence level Less than the confidence threshold or At that time, the judgment result was over-peeling. Rover is the over-exploitation risk indicator, and Tover is the risk threshold. The risk threshold is not a fixed constant but is designed as an "adaptive threshold." It depends on the normalized over-exploitation risk indicator and the quantile threshold calculated using traceability data, specifically satisfying the following:

[0093]

[0094] in, It is the 0.95 quantile. It is the 0.995th quantile. This is a sample set of 500 to 2000 recently qualified glass substrates that have been peeled off.

[0095] Furthermore, the following specific example illustrates the method for obtaining the risk threshold.

[0096] Assuming that two types of mechanical abrasion data—visual scratch probability and root mean square value of robotic arm vibration signal—are used to calculate the over-peeling risk index, with corresponding weights of 0.65 and 0.35, respectively, and... It contains a sample set of the most recent 1000 qualified glass substrates after peeling. Calculate the 0.95 quantile of the root mean square value of the robotic arm vibration signal. and the 0.995 quantile ; obtain The probability of visual scratch on a glass substrate and the root mean square (RMS) value of the robotic arm vibration signal are calculated, and the RMS value of the robotic arm vibration signal is normalized. The probability of visual scratch is set to 0.5, and the RMS value of the robotic arm vibration signal is... The normalized root mean square value of the robotic arm vibration signal was 0.667. Using the visual scratch probability, the normalized root mean square value of the robotic arm vibration signal, and their weights, the over-peel risk index was calculated, yielding Rover = 0.558. Finally, the risk threshold was calculated using the over-peel risk index of 1000 glass substrates. , The 0.995 quantile of the over-exploitation risk indicator is specifically calculated to be 0.5.

[0097] when and and At that time, the judgment result was that the peeling was qualified. The threshold value for the degree of liquid crystal peeling. , The threshold for the degree of indium stripping. If the conditions for qualified stripping are not met, but and The result was determined to be insufficient scraping.

[0098] S42. When the determination result is insufficient scraping, the glass substrate remains in its original position, and the coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed.

[0099] Step S42 specifically includes the following steps:

[0100] S421. When the determination result is insufficient scraping, the glass substrate remains in its original position, the extrinsic parameters of the camera coordinate system and the XRF scanning coordinate system are solved, and a mapping table of visual pixels and XRF sampling points to the substrate coordinate system is established.

[0101] Specifically, in this embodiment, reference marks are set at the edge of the glass substrate. When the determination result is insufficient scraping, the glass substrate remains in place, and the edge points, corner points, and reference marks of the substrate are acquired by the camera. Then, the extrinsic parameter transformation matrix from the camera coordinate system to the glass substrate coordinate system is solved. Simultaneously, based on the geometric parameters of the displacement mechanism and the alignment sampling points, the extrinsic parameter transformation matrix from the XRF scanning coordinate system to the glass substrate coordinate system is solved. The displacement mechanism is a mechanical device used to precisely control the spatial position of the X-ray emitting / receiving probe. Its function is to move the probe to the target area of ​​the glass substrate through multi-dimensional motion (X / Y / Z axis translation and rotation). The alignment sampling points are markers with known physical positions on the surface of the glass substrate, used to establish the spatial relationship between the XRF scanning coordinate system and the glass substrate coordinate system, including the substrate edge points, corner points, and reference marks. The specific process of solving the above extrinsic parameters is an existing technical method and will not be described in detail here.

[0102] After solving for the extrinsic parameters of the camera coordinate system and the XRF scanning coordinate system to obtain the corresponding extrinsic parameter transformation matrix, we can know the spatial position of each pixel on the glass substrate coating side image and the corresponding XRF sampling point on the glass substrate coating side, and then establish a mapping table from visual pixels to XRF sampling points to the glass substrate coordinate system.

[0103] S422. Calculate the registration residual from the camera coordinate system to the substrate coordinate system, and when the registration residual is greater than the residual threshold, call the calibration board to perform extrinsic parameter re-estimation.

[0104] Specifically, in this embodiment, the registration residual from the camera coordinate system to the substrate coordinate system is calculated, which specifically involves calculating the root mean square reprojection error (RMSRE) from the camera coordinate system to the substrate coordinate system. If the registration residual is less than the residual threshold, the coordinate registration result is considered reliable; otherwise, the checkerboard calibration plate is placed at the location of the glass substrate and extrinsic parameter reestimation is performed.

[0105] More specifically, in this embodiment, the residual threshold is set to 0.8 mm. In other alternative embodiments, extrinsic parameter estimation can also be performed when the ambient temperature changes by more than 3°C, the vibration intensity of the robotic arm exceeds 5 times the gravitational acceleration, equipment maintenance and replacement are completed, and the shift for stripping waste LCD panels is changed. Furthermore, the alignment position deviation or inlier ratio can be used to characterize the effect of coordinate registration.

[0106] S423. Recalculate the registration residual based on the result of the external parameter re-estimation, and perform a downgrade scan if the registration residual is still greater than the residual threshold.

[0107] Specifically, in this embodiment, at least 5 pixels are sampled and selected based on the extrinsic transformation matrix obtained by extrinsic re-estimation to recalculate the registration residual. If the registration residual is still greater than the residual threshold, a downgraded scanning strategy is triggered, specifically by increasing the XRF step size to 3-5mm and reducing the upper limit of the scraping pressure to 30N, and then using the extrinsic transformation matrix obtained by extrinsic re-estimation to perform subsequent operations; otherwise, the extrinsic transformation matrix obtained by extrinsic re-estimation is used directly to perform subsequent operations.

[0108] S43. The initial peeling area is determined using the residual mask and the indium element intensity distribution map, and then the final peeling area set is determined by morphological dilation, connected component analysis and noise filtering.

[0109] Step S43 specifically includes the following steps:

[0110] S431, Distribute the indium element intensity Figure 2 Values ​​are used to obtain regions where indium exceeds the standard. Then, the indium element intensity distribution map is fused with the residual mask through a logical OR operation to obtain the initial peeling area.

[0111] Specifically, in this embodiment, an indium intensity threshold is set. When the indium intensity at an XRF sampling point is greater than or equal to the indium intensity threshold, it is considered that the indium intensity at that XRF sampling point exceeds the limit and is represented by 1; otherwise, it is represented by 0. This allows for the distribution of indium intensity. Figure 2 The indium intensity distribution map is binarized and recorded as the indium binary distribution map. The indium exceeding the standard area can be determined by the indium binary distribution map.

[0112] Furthermore, both the residual mask and the indium element binary distribution map are binary images. In the residual mask, 1 indicates the presence of liquid crystal / adhesive layer residue at the pixel, and 0 indicates the absence of liquid crystal / adhesive layer residue at the pixel. Therefore, the indium element intensity distribution map and the residual mask can be fused using a logical OR operation to obtain the initial peeling area. Specifically, according to the mapping table obtained in step S421, for any pixel, if the value of that pixel is 1, or the value of the corresponding XRF sampling point on the indium element binary distribution map is 1, then that pixel is considered a peeling point and recorded as 1; otherwise, it is recorded as 0. This yields a binary image that fuses the indium element intensity distribution map and the residual mask, and this binary image is recorded as the fused image. The initial peeling area is the non-removable region with a value of 1 on the fused image. It is understood that there may be multiple initial peeling areas.

[0113] S432. Morphological dilation is performed on the initial patched region, and all independent regions are identified through connected component analysis.

[0114] Specifically, in this embodiment, a 3×3 or 5×5 rectangular structuring element is used to dilate the initial patching region by 1-2 pixels (corresponding to a physical size of 0.1-0.4mm). This operation can compensate for small residuals in coordinate registration, connect adjacent small residual points to form a coherent region, and smooth the boundary of the initial patching region, which is beneficial for planning the patching trajectory. After morphological dilation of the initial patching region, connected component analysis is performed, specifically identifying all independent regions (white) on the fused image. However, not all of these white independent regions can be considered as patching regions, as some white independent regions may be noise regions.

[0115] S433. Independent regions with an area smaller than the area threshold are filtered out as noise regions, and the remaining independent regions are used as the peeling regions to form the peeling region set.

[0116] Specifically, in this embodiment, after identifying all independent regions on the fused image, independent regions with fewer than 20 pixels are filtered out as noise regions, and the remaining independent regions are used as peeling regions to form the peeling region set.

[0117] S44. Generate a peeling trajectory adaptively based on each peeling region in the set of peeling regions, and perform simulation verification on the peeling trajectory to generate a peeling plan.

[0118] Step S44 specifically includes the following steps:

[0119] S441. Extract the regional features of the patched area, the regional features including area and shape.

[0120] Specifically, in this embodiment, the method for obtaining the area of ​​the re-peeling region can be referred to The method for obtaining the data is as follows: the area to be peeled is represented by the compactness of its shape. The calculation method for compactness is a current technical method and will not be described in detail here.

[0121] S442. Based on the regional characteristics, adaptively generate the peeling trajectory of the peeling region according to the pre-set trajectory generation rules.

[0122] Specifically, in this embodiment, based on the regional characteristics, the trajectory type is adaptively selected according to the pre-set trajectory generation rules, and then the peeling trajectory of the peeling region is adaptively generated. The peeling trajectory specifically includes the contour offset trajectory and the filling scan trajectory.

[0123] Contour offset trajectory, suitable for larger areas (area greater than 100 mm). 2For relatively regular-shaped (compactness greater than 0.7) peeling areas (ROIs), a circular path is generated by offsetting 1-3mm inward along the boundary of the peeling area, peeling layer by layer. The advantages are continuous path, low tool vibration, and low risk of over-peeling. The fill scan trajectory is suitable for small, irregularly shaped, or discrete point-like peeling areas, including zigzag and spiral trajectories. Zigzag trajectory: suitable for rectangular or near-rectangular ROIs, providing uniform coverage and simple code implementation. Spiral trajectory: suitable for circular or elliptical ROIs, expanding outward from the center without sharp reversals, resulting in more uniform tool wear. The step size of the fill scan is typically set to 0.2-0.5mm to ensure high coverage (≥95%). For contour offset trajectories and fill scan trajectories, please refer to [link to relevant documentation]. Figure 2 (b) in the middle.

[0124] S443. Simulate and verify the peeling trajectory to predict the area coverage of the peeling area by the peeling trajectory and the over-peeling risk index under the peeling trajectory.

[0125] Specifically, in this embodiment, a digital twin model of the glass substrate scraping is constructed as a simulation system. The scraping trajectory is input into the simulation system, and a virtual scraping process is generated based on the scraping trajectory. This simulates the movement path of the scraper on the scraping area and analyzes potential risks, namely, calculating the area coverage rate and over-scraping risk index, for simulation verification to ensure the scraping effect and feasibility. The area coverage rate is the ratio of the area scraped according to the scraping trajectory during the simulation to the area of ​​the scraping region.

[0126] The construction method of the digital twin model of glass substrate scraping is an existing technology and will not be described in detail here. In other alternative embodiments, the simulation system can also be a lightweight virtual process simulation / proxy model. Of course, simulation verification can also be omitted, and the glass substrate can be scraped directly according to the scraping trajectory and the original scraping parameters.

[0127] If the area coverage is less than 0.95, the trajectory generation strategy needs to be optimized and the scraper path planning adjusted. If the over-peeling risk index is greater than 0.5, it indicates that the risk of re-peeling is too high, and the scraping parameters such as the scraping pressure and linear speed of the scraper need to be adjusted. Through simulation verification, problems of missed scraping caused by improper re-peeling trajectory design can be identified and optimized in advance, ensuring that the scraping coverage meets the set standard. It can also predict the over-peeling risk that may occur in actual operation, helping to adjust and optimize the scraping operation and avoid damage to the substrate.

[0128] S444. If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then the re-peeling plan is composed of the re-peeling trajectory and the original scraping parameters.

[0129] S445. If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then adjust the scraping parameters and combine them with the re-peeling trajectory to form the re-peeling plan.

[0130] Specifically, in this embodiment, if the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, the simulation verification will fail, and the scraping parameters such as scraping pressure and linear speed need to be adjusted to ensure the effectiveness and stability of subsequent operations.

[0131] The scraping pressure of the squeegee determines the contact force between the squeegee and the glass substrate. Excessive scraping pressure may lead to over-peeling or damage to the glass substrate, while insufficient pressure may result in inadequate scraping. Therefore, when the risk of over-peeling is too high ( To reduce damage to the glass substrate, the scraping pressure needs to be reduced. This adjustment can be achieved by decreasing the normal pressure applied by the scraper (by 10% to 20%). In actual operation, if the current scraping pressure is 50N, then if the simulation verification fails, the scraping pressure should be reduced by 10% from 50N to 45N to ensure that over-peeling is not caused. However, the scraping pressure can only be reduced to a minimum of 10N.

[0132] The linear speed of the scraper has a significant impact on the stripping process. Excessively high linear speeds may cause the scraper to pass through certain areas too quickly, failing to fully remove the indium material. Conversely, excessively low linear speeds may lead to unnecessary prolonged contact, increasing the risk of over-stripping. Therefore, when simulation verification fails, the scraper's linear speed can be appropriately reduced to increase the scraping time for each area requiring additional stripping, ensuring thorough removal of the indium material. In actual operation, if the current scraper linear speed is 0.3 m / s, then when simulation verification fails, the linear speed should be reduced by 0.05 m / s (0.25 m / s) to reduce the risk of over-stripping and improve coverage. However, the minimum linear speed reduction is 0.05 m / s.

[0133] S446. If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then adjust the re-peeling trajectory and combine it with the scraping parameters to form the re-peeling plan.

[0134] Specifically, in this embodiment, if the area coverage rate is not greater than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, the simulation verification fails, and the supplementary peeling trajectory needs to be adjusted to improve the area coverage rate. The following provides a detailed explanation of how to adjust the supplementary peeling trajectory.

[0135] First, locate the bounding rectangle of the ROI and lay a mesh within this bounding rectangle, with each mesh square having a side length of 15mm. For any given mesh square, calculate the ratio of the area of ​​the peeling region to the total area of ​​the mesh square. If this ratio is greater than or equal to 30%, treat the ROI portion within that mesh square as a sub-ROI; otherwise, consider the ROI portion within that mesh square too fragmented to be processed and discard it. For any sub-ROI of an ROI, if its area is less than 50mm², it is considered a sub-ROI. 2 If a sub-ROI is considered too small, it will lead to frequent toolpath reversals and a higher risk of over-stripping. Therefore, it will not be processed separately and needs to be merged into the nearest sub-ROI (prioritizing merging with the sub-ROI with the longest contact boundary; if neither is in contact, merging with the sub-ROI with the shortest Euclidean distance from its centroid). Otherwise, the sub-ROI is considered suitable for separate processing. Through this series of operations, several sub-ROIs of "moderate size and not too fragmented shape" are obtained. Then, according to steps S441 and S442, the re-stripping trajectory for each sub-ROI of the ROI is replanned. Therefore, the re-stripping trajectory of the adjusted ROI is actually a collection of the re-stripping trajectories of its various sub-ROIs.

[0136] Furthermore, if it is necessary to readjust the stripping trajectory of the ROI later, simply adjust the mesh placement to avoid the current mesh completely overlapping with the mesh used when adjusting the stripping trajectory.

[0137] After splitting the ROI, multiple new ROIs can be obtained, and then steps S441 to S446 can be re-executed.

[0138] S447. If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then adjust the re-peeling trajectory and the scraping parameters, and then use the adjusted re-peeling trajectory and scraping parameters to form the re-peeling plan.

[0139] Specifically, in this embodiment, if the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, the re-peeling trajectory and scraping parameters need to be adjusted simultaneously. The adjustment method has been explained in steps S445 and S446, and will not be repeated here.

[0140] Even with continuous adjustments to the scraping parameters or peeling trajectory, simulation verification may still fail. Therefore, under the conditions that the scraping pressure can be reduced to a minimum of 10N and the linear velocity can be reduced to a minimum of 0.05m / s, upper limits are set for the number of adjustments and the ROI splitting level (i.e., the maximum number of times the ROI can be re-split). When any one or more of the following conditions are met—scraping pressure reduced to 10N, linear velocity reduced to 0.05m / s, reaching the upper limit of the number of adjustments, or reaching the upper limit of the ROI splitting level—but the simulation verification still fails, adjustments to the scraping parameters or peeling trajectory are stopped, and the glass substrate is deemed over-peeled to ensure that the process does not fall into an infinite loop and that the risk is controllable. The upper limit of the number of adjustments can be set to 4 times, and the upper limit of the ROI splitting level can be set to 2 times.

[0141] S45. Complete the peeling of the peeling area according to the peeling plan, and at the same time, perform a fusion judgment on the scraping effect of the peeling area based on visual detection and XRF detection.

[0142] Specifically, in this embodiment, the process of fusing visual detection and XRF detection to determine the scraping effect of the patching area can be referred to steps S2 to S41, and will not be elaborated here.

[0143] S46. If the determination result of the peeling area is insufficient, then perform the next peeling, and end the peeling when any one of the following conditions is met: the determination result is over-peeling, the determination result is qualified peeling, or the maximum number of peelings is reached.

[0144] Specifically, in this embodiment, the maximum number of peeling operations is set to 3-5 times. If the peeling result for the peeling area is insufficient, the next peeling operation is performed. At this time, it is necessary to re-obtain the residual mask, the binary distribution map of indium, the set of peeling areas, and the peeling plan based on the peeled glass substrate. The method for obtaining the residual mask, the binary distribution map of indium, the set of peeling areas, and the peeling plan has been described in the aforementioned steps, so it will not be repeated here.

[0145] S5. When the judgment result is over-peeling or the number of re-peeling times reaches the maximum number of re-peeling times, manual review shall be performed.

[0146] S6. Collect and store the glass substrate, liquid crystal components and scrap separately, and store the full process data of the waste liquid crystal panels to form a full process data traceability database.

[0147] Specifically, in this embodiment, the qualified glass substrate is placed in a clean glass substrate storage chamber using a robotic arm; when separating two glass substrates, the liquid crystal component collection chamber below the waste liquid crystal panel is used to collect the outflowing liquid crystal component; such as Figure 2As shown in (a), a negative pressure suction port located above the glass substrate is used to draw in scraper debris, which is then sent to a cyclone separator through a pipe to filter large scraper debris. The large scraper debris is then sent to an indium-rich scraper debris collection chamber through a discharge port. Meanwhile, cleaner air flows out from the central outlet at the top of the cyclone separator, enters a filter through a pipe, filters small scraper debris, and sends the small scraper debris to a low-indium scraper debris collection chamber through a discharge port.

[0148] Furthermore, the entire process data of waste LCD panels is stored to form a full-process data traceability database. The complete process data for the waste LCD panels includes at least the following: batch number of waste LCD panels, model number of waste LCD panels, processing timestamp of waste LCD panels, spray pressure, cleaning time, rotation speed of flexible nylon brush, hot air temperature, airflow pressure of air knife unit, heating plate temperature, heating plate heating time, pressure roller infeed, pressure roller speed, wedge head advance speed, vertical lifting speed of vacuum suction cup, negative pressure of negative pressure exhaust device; the angle between the scraper and the coating side of the glass substrate during the first scraping and subsequent scraping, the normal pressure of the constant force actuator, the cutting speed, the number of scrapings, mechanical scraping data, visual inspection results, XRF inspection results, residual mask, over-peeling risk index, confidence level of liquid crystal and adhesive layer identification, liquid crystal peeling degree index, indium peeling degree index, indium element intensity distribution map, indium element binary distribution map, fused image, set of re-peeling areas, regional characteristics, and judgment results of scraping effect, etc.; the number of re-peelings, re-peeling sequence number, and re-peeling trajectory during re-scraping. Some of the aforementioned data can be manually set directly at the control center, such as spray pressure, cleaning time, flexible nylon brush speed, hot air temperature, airflow pressure of the air knife unit, heating plate temperature, heating plate heating time, pressure roller infeed, pressure roller speed, wedge head advance speed, vacuum suction cup vertical lifting speed, negative pressure of the negative pressure exhaust device, angle between the scraper and the coating side of the glass substrate, normal pressure of the constant force actuator, cutting speed, and number of scraping operations. Some data can be directly collected by sensors, such as visual inspection results and XRF inspection results. Some data are obtained by data processing at the control center based on the collected data, such as residual mask, over-peeling risk index, confidence level of liquid crystal and adhesive layer identification, liquid crystal peeling degree index, indium peeling degree index, indium element intensity distribution map, indium element binary distribution map, fused image, set of re-peeling areas, number of re-peeling operations during re-scraping, re-peeling sequence number, re-peeling trajectory, area characteristics, and judgment results of scraping effect, etc.

[0149] Furthermore, accumulated data is periodically used to optimize key thresholds and process parameters such as the liquid crystal peeling degree index threshold and the indium peeling degree index threshold. For example, for all substrates that have recently passed peeling, the 95th percentile of the corresponding liquid crystal peeling degree index is calculated as the recommended value for the liquid crystal peeling degree index threshold. Recent data can be used periodically to establish regression relationships between process parameters such as the included angle of the glass substrate coating side, the normal pressure of the constant force actuator, and the feed rate, and the liquid crystal peeling degree index and the indium peeling degree index, and this is used to optimize the process parameters. The optimized process parameters are first released in a grayscale test on some equipment or production lines. If the test results on some equipment or production lines show that the peeling pass rate increases by more than 2% and the over-peeling rate does not increase under the optimized process parameters, then all equipment or production lines will use the optimized process parameters to achieve a full-line update of the process parameters; otherwise, a rollback operation is performed, that is, the equipment or production line using the optimized process parameters reverts to the previous version of the process parameters. All process parameter version changes, test results, and rollback operations are recorded in detail and stored in a full-process data traceability database to ensure that the process is auditable.

[0150] It should be noted that in some cases, the actions described in the specification can be performed in different orders and still achieve the desired results. In this embodiment, the order of steps is given only to make the embodiment clearer and easier to explain, and not to limit it.

[0151] In one optional embodiment, please refer to Figure 3 This invention provides an intelligent mechanical stripping and indium enrichment system for waste liquid crystal panels. The system is applicable to the intelligent mechanical stripping and indium enrichment method for waste liquid crystal panels provided by this invention. The system includes: a cavity, a conveying and positioning module, a cleaning and drying module, a controlled deformation module, a substrate separation module, an intelligent scraping module, a data acquisition module, a graded and compartmented material discharge module, a control center, and a data storage module.

[0152] The cavity provides a sealed space for the intelligent mechanical peeling of waste LCD panels; the conveying and positioning module conveys the waste LCD panels into the cavity and fixes them inside; the cleaning and drying module cleans and dries the waste LCD panels; the controlled deformation module performs controlled deformation processing on the waste LCD panels; the substrate separation module separates two glass substrates along the sealing area; the intelligent scraping module mechanically scrapes the coated side of the glass substrate; the data acquisition module acquires mechanical scraping data and performs visual and XRF inspection on the glass substrate; and the graded and compartmented discharge module separates the glass substrate, liquid crystal components, and scraping debris. The system involves separate collection and storage in designated areas. The control center utilizes the mechanical scraping data, visual inspection results, and XRF inspection results to obtain judgment indicators, including over-peeling risk indicators, liquid crystal peeling degree indicators, and indium peeling degree indicators. The control center performs a fusion judgment on the scraping effect based on these indicators, and when the judgment result indicates insufficient scraping, it performs in-situ cyclical re-peeling and judgment based on visual inspection and XRF inspection. The control center performs manual verification when the judgment result indicates over-peeling or the number of re-peeling attempts reaches the maximum number of re-peeling attempts. The data storage module stores the entire process data of the waste liquid crystal panels, forming a full-process data traceability database.

[0153] The inner wall of the cavity is lined with a lead plate at least 3mm thick as a radiation shielding structure; the conveying and positioning module includes at least a loading platform, a conveying mechanism, and a positioning fixture; the cleaning and drying module includes at least a spray device, a brushing unit, and a hot air / air knife unit; the controlled deformation module includes at least a heating plate and a pressure roller; the substrate separation module includes at least a diamond cutting wheel, a wedge blade, and a vacuum suction cup; the intelligent scraping module includes at least an angle adjustment mechanism, a scraper, a constant force actuator, and a three-axis motion mechanism; the data acquisition module includes at least a cross-polarized light source, a camera, and an XRF probe; the graded and compartmented discharge module includes at least a clean glass substrate storage compartment, a liquid crystal component collection compartment, a negative pressure suction port, pipes, a cyclone separator, a filter, an indium-rich scraper collection compartment, and a low-indium scraper collection compartment. All of the above equipment and mechanisms are existing technologies, and their models or types can be selected according to actual needs.

[0154] The positioning fixture, cleaning and drying module, controlled deformation module, substrate separation module, intelligent scraping module, data acquisition module, clean glass substrate storage compartment, liquid crystal component collection compartment, negative pressure suction port and other equipment are located inside the cavity.

[0155] The control center is electrically connected to other modules to control their operation, thereby enabling intelligent mechanical stripping and indium enrichment of waste LCD panels. It can also trigger voice prompts when necessary to remind relevant personnel to conduct manual verification.

[0156] In summary, this method has at least the following beneficial effects: By collecting mechanical scraping data, visual inspection, and XRF detection to obtain judgment indicators for determining the scraping effect, the glass substrate, liquid crystal components, and scraping debris are collected and stored separately. When necessary, in-situ cyclic peeling and judgment are performed based on visual inspection and XRF detection, along with manual verification, thus forming a strongly coupled closed-loop method for peeling waste liquid crystal panels, avoiding the problem of only detecting the peeling effect without control. By registering the camera coordinate system and the XRF scanning coordinate system, and performing online self-calibration through the registration residual, drift misjudgment can be reduced, and the accuracy of the peeling area can be improved, thereby ensuring the reliability of the closed-loop method for peeling waste liquid crystal panels. Based on lightweight depth... The degree-learning segmentation model quickly identifies residual masks, and combined with the indium intensity distribution map, it can quickly identify the re-peeling area and generate the re-peeling trajectory in real time. This allows for in-situ re-peeling at the current workstation, achieving "inspection and re-peeling simultaneously," significantly reducing cycle time losses caused by physical reflow and repositioning. By limiting the maximum number of re-peeling operations and setting over-peeling conditions, the risk of over-peeling is controllable, ensuring that the mechanical peeling of waste LCD panels balances thorough peeling with safety margins. Separate collection and storage of glass substrates, liquid crystal components, and indium-containing scraping shavings improves backend stability. Using the complete process data from waste LCD panel peeling to form a full-process data traceability database supports various threshold and window version updates, gray-scale releases, and rollbacks, facilitating large-scale replication and auditing. Furthermore, since this system uses this method, it possesses at least the same beneficial effects as the method.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for the intelligent mechanical stripping and enrichment of indium from waste liquid crystal panels, characterized in that, Includes the following steps: In a closed space, the fixed waste LCD panel is cleaned, dried and deformed in a controlled manner, and then the two glass substrates are separated along the sealing area. Mechanical scraping is performed on the coated side of the glass substrate, while visual inspection and XRF inspection are performed on the glass substrate, and mechanical scraping data is obtained. The judgment criteria are obtained using the mechanical scraping data, visual inspection results, and XRF inspection results. The scraping effect is judged by the judgment index to obtain the judgment result, which includes insufficient scraping, over-peeling and qualified peeling. When the determination result is insufficient scraping, the glass substrate remains in its original position, and the coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed; The initial peeling area was determined by using the residual mask and indium intensity distribution map, and then the final peeling area set was determined by morphological dilation, connected component analysis and noise filtering. Based on each of the peeling regions in the set of peeling regions, an adaptive peeling trajectory is generated, and the peeling trajectory is simulated and verified to generate a peeling plan. The peeling of the peeling area is completed according to the peeling plan, and the scraping effect of the peeling area is fused and determined based on visual detection and XRF detection. If the determination result of the peeling area is insufficient, the next peeling is performed, and the peeling ends when any one of the following conditions is met: the determination result is over-peeling, the determination result is qualified peeling, or the maximum number of peelings is reached. When the judgment result is over-peeling or the number of re-peeling times has reached the maximum number of re-peeling times, manual review shall be performed. Glass substrates, liquid crystal components, and scrap are collected and stored in separate compartments, and the entire process data of waste liquid crystal panels is stored to form a full-process data traceability database.

2. The method for intelligent mechanical stripping and enriching indium from waste liquid crystal panels according to claim 1, characterized in that: The mechanical scraping data includes at least two of the following: visual scratch probability, glass substrate edge cracking probability, scraper blade wear, motor torque fluctuation amplitude, motor torque fluctuation amplitude, root mean square value of robotic arm vibration signal, and probability of the scraper crossing the prohibited peeling zone.

3. The method for intelligent mechanical stripping and enriching indium from waste liquid crystal panels according to claim 2, characterized in that: The determination criteria include an over-peeling risk index, confidence level of liquid crystal and adhesive layer identification, liquid crystal peeling degree index, and indium peeling degree index. The over-peeling risk index is a weighted sum of the normalized results of at least two of the mechanical abrasion data.

4. The intelligent mechanical stripping and indium enrichment method for waste liquid crystal panels according to claim 3, characterized in that: A lightweight deep learning segmentation model is used to perform pixel-level segmentation on the visual detection results, outputting the residual masks of liquid crystal and adhesive layer, as well as the confidence score, and calculating the liquid crystal peeling degree index, which satisfies the following relationship: in, This refers to the degree of liquid crystal peeling. The residual areas of the liquid crystal and the adhesive layer are determined based on the residual mask. This represents the initial area of ​​the liquid crystal and the adhesive layer.

5. The intelligent mechanical stripping and indium enrichment method for waste liquid crystal panels according to claim 4, characterized in that: The indium elemental intensity distribution map is obtained based on the XRF detection results. Then, the indium peeling degree index is calculated based on the indium elemental intensity distribution map. The indium peeling degree index satisfies the following relationship: in, The degree of indium stripping is an indicator. The average indium strength after scraping. The average indium strength before scraping.

6. The method of claim 1, wherein the method is characterized by: When the determination result is insufficient scraping, the glass substrate remains in its original position, and coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed, including the following steps: When the determination result is insufficient scraping, the glass substrate remains in its original position, the extrinsic parameters of the camera coordinate system and the XRF scanning coordinate system are solved, and a mapping table of visual pixels and XRF sampling points to the substrate coordinate system is established. Calculate the registration residual from the camera coordinate system to the substrate coordinate system, and when the registration residual is greater than the residual threshold, call the calibration board to perform extrinsic parameter re-estimation; The registration residual is recalculated based on the result of the extrinsic parameter re-estimation, and a downgrade scan is performed if the registration residual is still greater than the residual threshold.

7. The intelligent mechanical stripping and indium enrichment method for waste liquid crystal panels according to claim 1, characterized in that, The process of determining the initial peeling area using a residual mask and indium element intensity distribution map, and then determining the final peeling area set through morphological dilation, connected component analysis, and noise filtering, includes the following steps: The indium intensity distribution map is binarized to obtain the indium-exceeding area, and then the indium intensity distribution map is fused with the residual mask by a logical OR operation to obtain the initial peeling area; Morphological dilation is performed on the initial patched region, and all independent regions are identified through connected component analysis; Independent regions with an area smaller than the area threshold are filtered out as noise regions, and the remaining independent regions are used as the supplementary stripping regions to form the supplementary stripping region set.

8. The method of claim 1, wherein the method is characterized by: The step of adaptively generating a stripping trajectory based on each stripping region in the set of stripping regions, simulating and verifying the stripping trajectory, and then generating a stripping plan includes the following steps: Extract the regional features of the patched area, including area and shape; Based on the regional characteristics, the peeling trajectory of the peeling region is adaptively generated according to a pre-defined trajectory generation rule; The stripping trajectory is simulated and verified, and then the area coverage of the stripping area by the stripping trajectory and the over-stripping risk index under the stripping trajectory are predicted. If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then the re-peeling plan is composed of the re-peeling trajectory and the original scraping parameters. If the area coverage rate is greater than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then the scraping parameters are adjusted, and then the re-peeling plan is formed in combination with the re-peeling trajectory. If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is less than the risk threshold, then the re-peeling trajectory is adjusted, and the re-peeling plan is formed by combining the scraping parameters. If the area coverage rate is less than the coverage rate threshold and the over-peeling risk index is greater than the risk threshold, then the re-peeling trajectory and the scraping parameters are adjusted, and the adjusted re-peeling trajectory and scraping parameters are used to form the re-peeling plan.

9. A smart mechanical stripping and indium enrichment system for waste liquid crystal panels, the system being applicable to the smart mechanical stripping and indium enrichment method for waste liquid crystal panels as described in any one of claims 1-8, characterized in that, The system includes: The system includes a cavity, a conveying and positioning module, a cleaning and drying module, a controlled deformation module, a substrate separation module, an intelligent scraping module, a data acquisition module, a graded and compartmented material discharge module, a control center, and a data storage module. The control center is electrically connected to the other modules. The cavity is used to provide a sealed space for the intelligent mechanical stripping of waste LCD panels; The conveying and positioning module is used to convey the waste LCD panel into the cavity and fix it in the cavity; The cleaning and drying module is used to clean and dry waste LCD panels; The controlled deformation module is used for controlled deformation processing of waste LCD panels; The substrate separation module is used to separate two glass substrates along the sealing area; The intelligent scraping module is used to mechanically scrape the coated side of the glass substrate. The data acquisition module is used to acquire mechanical scraping data and perform visual inspection and XRF inspection on the glass substrate; The graded and compartmented material discharge module is used to collect and store glass substrates, liquid crystal components and scrap in separate compartments; The control center uses the mechanical scraping data, visual inspection results, and XRF inspection results to obtain judgment indicators, including over-peeling risk indicators, liquid crystal peeling degree indicators, and indium peeling degree indicators. The control center performs a fusion judgment on the scraping effect based on these indicators to obtain a judgment result, which includes insufficient scraping, over-peeling, and qualified peeling. When the judgment result is insufficient scraping, the glass substrate remains in place, and coordinate registration between the camera coordinate system and the XRF scanning coordinate system is completed. The initial peeling area is determined using the residual mask and indium element intensity distribution map, and then the final peeling area is determined through morphological dilatation, connected component analysis, and noise filtering. A set of regions is defined; a peeling trajectory is adaptively generated for each peeling region in the set, and the peeling trajectory is simulated and verified to generate a peeling plan; the peeling of the regions is completed according to the peeling plan, and the scraping effect of the regions is fused and judged based on visual inspection and XRF inspection; if the judgment result of the peeling region is insufficient scraping, the next peeling is performed, and the peeling ends when any one of the following conditions is met: the judgment result is over-peeling, the judgment result is qualified peeling, or the maximum number of peelings is reached; the control center performs manual verification when the judgment result is over-peeling or the number of peelings reaches the maximum number of peelings. The data storage module is used to store the entire process data of waste LCD panels, forming a full-process data traceability database.

Images