Dynamic Release Method and System for Intelligent Labeling Station Robot

Abstract

This invention relates to the field of label application control technology, and discloses a dynamic release method and system for a robot at an intelligent label application station. The method acquires visual images, RFID / NFC read / write responses, adsorption negative pressure, conveying status, and robot end-effector status data after the intelligent label to be applied reaches the label picking position. It extracts the sheet material status characteristics of a single label and constructs an application feasibility feature map in the label plane coordinate system. Based on the application feasibility feature map, candidate application actions are generated, and the risks of application offset, edge wrinkling, local bubbles, and electronic response anomalies for each candidate action are predicted. Before robot application, the method determines the status of direct release, compensated release, leveling pending review, or prohibited release, and outputs the corresponding application compensation parameters or stop application command. This invention can reduce the probability of intelligent label application offset, wrinkling, bubbles, and post-application read / write anomalies.

Description

Technical Field

[0001] This invention relates to the field of label application control technology, and more specifically, to a dynamic release method and system for an intelligent label application station robot. Background Technology

[0002] Smart labels typically consist of a flexible face material, an adhesive layer, a release layer, and electronic functional areas located inside or on the surface of the label. These electronic functional areas may include RFID antennas, NFC antennas, chips, conductive lines, or pad connection areas. These types of labels are widely used in smart packaging, clothing hang tags, logistics traceability labels, care labels, and product anti-counterfeiting labels. For smart labels containing electronic functional areas, the adhesion quality depends not only on the accuracy of the label's outer contour but also on whether the antenna area, chip area, and conductive connection area experience bending, pressure, localized wrinkling, air bubble coverage, or positional misalignment during the adhesion process.

[0003] In the smart label application station, the smart labels to be applied are typically conveyed to the label picking position by a conveyor mechanism. After the release film is peeled off by the release film peeling mechanism, the robotic application actuator picks up the label via a suction actuator and applies it to the surface of the target carrier. Because smart labels have a flexible substrate and adhesive layer structure, the label's condition is affected by the peeling angle, adhesive adhesion, conveying tension, conveying speed, and the negative pressure at the end of the suction process during the short period between release film peeling and robotic label picking. If these conditions fluctuate, the label edges may exhibit slight warping, localized arching, adhesive layer dragging deformation, or label picking posture deviation.

[0004] Meanwhile, some smart tags have a transparent coating, a glossy protective layer, or a metal antenna layer on their surface. When the visual acquisition unit identifies the tag's outer contour, positioning marks, antenna area, and chip area, it is easily affected by the reflection of the transparent coating, the highlight of the antenna metal, oblique illumination reflection, and local shadows, resulting in false edges, missing local boundaries, decreased confidence of positioning marks, or deviations in the identification of electronic functional area positions in the image. For ordinary paper tags, attaching control based solely on the center of the tag's outer contour or positioning marks is usually sufficient to meet the attaching requirements; however, for smart tags, even if the tag's outer contour position meets the requirements, if the antenna area or chip area is offset relative to the predetermined area of ​​the target carrier, it may still lead to a shortened RFID / NFC read / write distance, a decreased read / write success rate, or abnormal stress on local electrical connections after attaching.

[0005] Most existing smart label application devices use visual positioning to identify the outer contour or positioning marks of the label and control the robot or labeling mechanism to perform adsorption and application actions according to a preset trajectory. Some devices will perform appearance inspection or RFID / NFC read / write detection after application to determine whether the label is misaligned, wrinkled, has air bubbles, or has read / write abnormalities. This type of method mainly relies on the single visual positioning result before application or the result detection after application, making it difficult to determine whether the current single smart label is suitable for direct application, whether it needs to be compensated for, whether it needs to be leveled and re-evaluated, or whether application should be stopped before the robot's application action is performed.

[0006] Specifically, when the confidence level of the electronic functional area image decreases but the outer contour of the tag can still be identified, the robot may still perform the attachment according to the preset path; when the negative pressure of the adsorption actuator reaches the tag-taking threshold but the negative pressure establishment time is prolonged or the negative pressure fluctuation is increased, existing equipment has difficulty in timely determining whether the tag has insufficient local adsorption or causes posture deviation; when there is slight warping at the edge of the tag, existing equipment usually cannot distinguish whether the warping can be handled by secondary adsorption, pre-pressure leveling, reducing the peeling speed, or extending the holding pressure time; when the RFID / NFC read / write response has experienced continuous failures, prolonged response time, or fluctuating response intensity before attachment, existing equipment also has difficulty in timely associating the abnormal electronic response with the local attachment risk of the antenna area and chip area and the robot's motion parameters.

[0007] Therefore, in flexible intelligent label attaching stations with electronic functional areas, there is a need for a dynamic release method and system that can comprehensively process the label's visual state, electronic response state, adsorption negative pressure state, conveying disturbance state, and robot end state before robot attaching, so as to output attaching compensation parameters, leveling and re-judgment instructions, or stop attaching instructions based on the actual state of the current single label. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for dynamic release of intelligent label-attaching robots at workstations.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for dynamic release of a smart label-attaching robot at a workstation includes the following steps:

[0011] Acquire multi-source workstation data for the smart tag to be attached, including visual images, electronic response data, adsorption status data, conveying status data, and robot end-effector status data;

[0012] Extract the sheet material status characteristics of the current single smart tag based on multi-source workstation data;

[0013] A label plane coordinate system is established based on the current positioning reference point of a single smart label. The material status characteristics are mapped to the label plane coordinate system to construct an attachment feasibility feature map.

[0014] Candidate attachment actions are generated based on the attachment feasibility feature map;

[0015] Before the robot attaches the action, the attaching feasibility feature map, candidate attaching action parameters, robot end state data and transport state data are input into the attaching consequence prediction model to obtain the attaching consequence prediction results corresponding to each candidate attaching action.

[0016] Based on the predicted consequences of affixing, the feasibility feature map of affixing, and the electronic response data, the current dynamic release status of a single smart tag is determined. The dynamic release status includes direct release status, compensated release status, leveling and waiting for re-judgment status, and prohibited release status.

[0017] Dynamic release control results are generated based on the dynamic release status. Specifically, direct release status and compensated release status correspond to the generation of attachment control parameters, leveling pending re-judgment status corresponds to the generation of transport correction parameters, and prohibited release status corresponds to the generation of stop attachment command.

[0018] In a preferred embodiment, acquiring the multi-source workstation data to be attached to the smart tag includes:

[0019] The reference time is determined by the tag placement signal or the robot tag-picking trigger signal;

[0020] Establish a preset time window that includes the reference time;

[0021] Within a preset time window, visual images, electronic response data, adsorption status data, delivery status data, and robot end effector status data are collected.

[0022] The visual images, electronic response data, adsorption status data, delivery status data, and robot end-effector status data within the preset time window are associated with the current single smart tag according to the timestamp.

[0023] In a preferred embodiment, the smart tag to be attached includes an antenna area and a chip area;

[0024] The sheet material state characteristics include edge warping characteristics, local wrinkling characteristics, visual confidence of electronic functional areas, positioning reference point deviation, read / write response stability, adsorption establishment stability, and transport disturbance level. Among them, the electronic response data includes the number of successful read / write operations, response intensity, and response time obtained by performing electronic function read / write operations on the antenna area and the chip area.

[0025] The adsorption state data includes the negative pressure establishment and maintenance data of the adsorption actuator;

[0026] The conveying status data includes the operating status data of the conveying mechanism and the release film peeling mechanism.

[0027] In a preferred embodiment, the construction of the attachment feasibility feature map includes:

[0028] The current single smart tag is divided into multiple two-dimensional grid units in the tag plane coordinate system;

[0029] Based on the position of each two-dimensional grid unit in the current single smart tag, determine whether the area to which the two-dimensional grid unit belongs is the edge area, the center area, the adhesive layer transition area, the antenna area, or the chip area;

[0030] Based on the visual risk, electronic response risk, adsorption risk and transport disturbance risk corresponding to each two-dimensional grid cell, the attachment risk value of that two-dimensional grid cell is generated.

[0031] The attachment feasibility feature map is formed by the attachment risk values ​​of multiple two-dimensional grid cells. In a preferred embodiment, the generation of the attachment risk values ​​includes:

[0032] For the two-dimensional grid cells corresponding to the antenna region and the chip region, an electronic response risk weight higher than that of the central region is configured.

[0033] For the two-dimensional grid cells corresponding to the edge region and the adhesive layer transition region, configure a higher visual risk weight or adsorption risk weight than that of the center region.

[0034] In a preferred embodiment, the candidate attachment actions include direct attachment actions, pose-compensated attachment actions, pressure-compensated attachment actions, extended pressure-holding attachment actions, secondary adsorption and leveling actions, and release-prevention actions.

[0035] The candidate attachment action parameters include the robot end effector's lateral compensation amount, longitudinal compensation amount, rotation angle compensation amount, attachment pressure, attachment speed, pressure holding time, and whether secondary adsorption and leveling are performed.

[0036] In a preferred embodiment, the predicted attachment consequences include risks of attachment misalignment, edge wrinkling, localized bubbles, and abnormal electronic response.

[0037] The attachment consequence prediction model is trained based on historical attachment condition data, which includes multi-source workstation data before attachment, attachment feasibility feature map, attachment action parameters actually performed by the robot, and post-attachment detection results. The post-attachment detection results include actual attachment position deviation, edge wrinkling detection results, local bubble detection results, and electronic function retest results.

[0038] The prediction results of the attachment consequences are used to determine the dynamic release status, and the dynamic release status is used to generate attachment control parameters, transmit correction parameters or stop attachment commands.

[0039] In a preferred embodiment, the dynamic release state includes:

[0040] When the risks of attachment offset, edge wrinkling, local bubble, and abnormal electronic response corresponding to the direct attachment action are all lower than the corresponding direct release threshold, the current single smart tag is determined to be in the direct release state.

[0041] When any of the risks of attachment offset, edge wrinkling, and local bubble exceeds the corresponding direct release threshold, but none of the risks of attachment offset, edge wrinkling, and local bubble exceed the corresponding compensation upper limit threshold, and the read / write response stability is not lower than the preset lower limit, the current single smart tag is determined to be in the compensation release state.

[0042] When the high-risk area in the attachment feasibility feature map is located in the edge area or adhesive layer transition area, the proportion of high-risk area in the antenna area and chip area does not exceed the preset upper limit, and the read / write response stability is not lower than the preset lower limit, the current single smart tag is determined to be in the flattening and re-judgment state, and after performing secondary adsorption, pre-pressing flattening, reducing peeling speed or pausing conveying, the multi-source station data is re-acquired.

[0043] The current single-chip smart tag is determined to be in a prohibited release state when any of the following conditions are met: the read / write response stability is lower than the preset lower limit and the electronic response data is continuously abnormal; the proportion of high-risk area in at least one of the antenna area and chip area exceeds the preset upper limit; the negative pressure cannot be established within the preset time; or at least one of the antenna area and chip area contains a high-risk area that meets the prohibited release conditions.

[0044] In a preferred embodiment, the method further includes: after the bonding is completed, acquiring post-bonding images and electronic response retest data; and associating pre-bonding multi-source workstation data, bonding feasibility feature map, dynamic release status, bonding control parameters, transport correction parameters, post-bonding images and electronic response retest data as bonding feedback samples.

[0045] The calibration parameters of the adhesion consequence prediction model, the weight parameters of the adhesion feasibility feature map, or the discrimination threshold of the dynamic release status are updated based on the adhesion feedback samples.

[0046] Each time an update is performed, the calibration parameters, weight parameters, or discrimination thresholds from before the update are saved. If the updated parameters cause an increase in the abnormal adhesion index, the calibration parameters, weight parameters, or discrimination thresholds saved before the update are restored.

[0047] A dynamic release system for an intelligent label-attaching robot includes a controller, which is communicatively connected to a vision acquisition unit, an electronic function reading and writing unit, a negative pressure detection unit, a conveying status detection unit, a robot control unit, and a conveying control unit. The controller is equipped with a data acquisition module, a status feature extraction module, an attachment feasibility construction module, a candidate action generation module, a consequence prediction module, a dynamic release module, a compensation output module, and a feedback update module, wherein:

[0048] The data acquisition module is used to acquire multi-source workstation data to be attached to the smart tag, and output the multi-source workstation data to the status feature extraction module;

[0049] The status feature extraction module is used to extract the current status features of a single smart tag from multi-source workstation data;

[0050] The adhesion feasibility construction module is used to map the sheet material state characteristics to the label plane coordinate system and construct an adhesion feasibility feature map.

[0051] The candidate action generation module is used to generate candidate attachment actions based on the attachment feasibility feature map;

[0052] The consequence prediction module is used to output the prediction results of the attachment consequences for each candidate attachment action based on the attachment feasibility feature map, candidate attachment action parameters, robot end state data and delivery state data.

[0053] The dynamic release module is used to determine the current dynamic release status of a single smart tag based on the prediction results of the affixing consequences, the affixing feasibility feature map, and the electronic response data.

[0054] The compensation output module is used to output attachment control parameters to the robot control unit, output delivery correction parameters to the delivery control unit, and output a stop attachment command to the robot control unit according to the dynamic release status.

[0055] The feedback update module is used to generate attachment feedback samples based on the post-attachment images and electronic response retest data, and update the calibration parameters of the attachment consequence prediction model, the weight parameters of the attachment feasibility feature map, or the discrimination threshold of the dynamic release status.

[0056] By adopting the above technical solution, the beneficial effects of the present invention are as follows:

[0057] By associating visual images, RFID / NFC read / write responses, adsorption negative pressure, conveying status, and robot end-effector status with the same smart tag's data window before the robot attaches the tag, the release decision is based on the same tag-taking stage data of the current tag, reducing attachment misjudgments caused by mismatched timing of different data sources.

[0058] By mapping the local states of the tag's outer contour, adhesive layer transition area, antenna area, and chip area into an attachment feasibility feature map, and improving the risk impact of the antenna area and chip area in the release judgment, the system can identify problems such as electronic functional area positioning deviation, read / write response fluctuation, and increase in local high-risk areas before attachment, thereby reducing the probability of read / write failure, shortened read / write distance, or abnormal pressure on the electronic functional area after attachment.

[0059] Candidate attaching actions are generated based on the attaching feasibility feature map, and the attaching offset risk, edge wrinkling risk, local bubble risk and electronic response abnormality risk corresponding to each candidate action are predicted. This allows for selection among direct release, compensated release, leveling and re-judgment, and prohibition of release, so that the robot end pose, attaching pressure, attaching speed, holding time and conveying status can be adjusted according to the actual state of a single label.

[0060] Feedback samples are generated by using images after affixing and RFID / NFC retest data. Version updates and rollbacks are performed on the release judgment threshold or prediction model calibration parameters. This can adapt to changes in the adhesion of the label adhesive layer, the reflectivity of the film, the conveying tension, and the state of the adsorption actuator in different batches, and reduce the adverse effects of abnormal parameter updates on the control results of the affixing station. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the dynamic release method for the intelligent label attaching station robot of the present invention;

[0062] Figure 2 This is a schematic diagram of the intelligent label attaching station robot dynamic release system module of the present invention;

[0063] Figure 3 This is a schematic diagram of the multi-source data acquisition and execution control relationship at the intelligent label attaching station of the present invention. Detailed Implementation

[0064] In the following description, numerous technical details are presented to enable the reader to better understand the present invention. However, those skilled in the art will understand that the technical solutions claimed in this invention can be implemented even without these technical details and with various variations and modifications based on the following embodiments.

[0065] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0066] Example 1:

[0067] This embodiment provides a dynamic release control method based on a smart label attaching station. This method is applied to, for example... Figure 3The smart label application station shown is equipped with a smart label to be applied, a conveying mechanism, a release film peeling mechanism, a robotic application actuator, an adsorption actuator, a target carrier, a negative pressure detection unit, a conveying status detection unit, a vision acquisition unit, an RFID / NFC read / write unit, and a controller. The smart label to be applied includes a label outline, an adhesive layer transition area, an antenna area, and a chip area.

[0068] In this embodiment, the conveying mechanism delivers the smart label to be attached to the robot's label-picking position, the release film peeling mechanism peels off the release film before the robot picks up the label, and the robot attaching execution mechanism adsorbs the smart label to be attached via an adsorption actuator and attaches the smart label to the target carrier. The vision acquisition unit acquires image data of the label's outer contour, adhesive layer transition area, antenna area, and chip area; the RFID / NFC read / write unit acquires electronic response data corresponding to the antenna area and chip area; the negative pressure detection unit acquires data on the establishment and maintenance of negative pressure by the adsorption actuator; and the conveying status detection unit acquires operating status data of the conveying mechanism and the release film peeling mechanism. The controller receives the above data and outputs control parameters to the robot attaching execution mechanism, the conveying mechanism, and the release film peeling mechanism based on the release judgment result.

[0069] like Figure 1 As shown, the method in this embodiment includes the following steps:

[0070] S1, acquire multi-source workstation data of the smart label to be attached.

[0071] When the smart tag to be attached reaches the robot's tag-picking position, the controller establishes a data window for the current single tag based on the tag arrival signal or the robot's tag-picking trigger signal, and obtains the multi-source workstation data corresponding to the current single tag within the data window.

[0072] Multi-source workstation data includes visual images, electronic response data, adsorption status data, conveying status data, and robot end-effector status data. Visual images include frontal images, oblique illumination images, and polarized images; electronic response data includes the number of successful RFID / NFC read / write operations, response intensity, response time, and continuous read / write fluctuations; adsorption status data includes the negative pressure build-up time, negative pressure holding value, negative pressure fluctuation amplitude, and negative pressure attenuation slope of the adsorption actuator; conveying status data includes conveying speed, conveying tension, peeling angle, and peeling speed; and robot end-effector status data includes the coordinates of the end-effector adsorption point, end-effector attitude angle, end-effector approach speed, and adsorption contact time.

[0073] Specifically, a time window is established with the tag arrival time or the robot tag-picking trigger time as the base time t0. .in, The timeframe can be from 50ms to 200ms. The timeframe can range from 300ms to 800ms; in production lines with faster labeling cycles, it can also be adjusted based on the camera frame rate, reader sampling period, negative pressure sampling period, and robot control period. The controller associates the visual frames, RFID / NFC read / write response sequences, negative pressure curves, conveying status curves, and robot end-effector status records within this time window into multi-source workstation data for the current single tag.

[0074] This process ensures that data from different sources correspond to the same smart tag to be attached and the same tag acquisition stage, avoiding the use of electronic response data from the previous tag or negative pressure data from the next sampling cycle for the release judgment of the current tag.

[0075] S2, Extract the sheet material state characteristics of individual labels.

[0076] The controller processes the multi-source workstation data acquired by S1 and extracts the sheet material status characteristics of the current single label.

[0077] Specifically, the controller identifies the tag's outer contour, positioning marks, adhesive layer transition area, antenna area, and chip area based on visual images, and establishes a tag planar coordinate system using the tag's outer contour center point, positioning marks, or antenna reference point. The controller obtains the edge offset based on the deviation between the tag edge and the preset reference contour; it obtains the edge warping characteristics based on the local curvature of the edge line segments, edge shadow changes, and contour breakage areas; and it obtains the visual confidence level of the electronic functional area based on the boundary integrity, reflective area, image clarity, and positioning confidence level of the antenna area and chip area.

[0078] Furthermore, the controller calculates read / write response stability based on the number of successful RFID / NFC read / write operations, response intensity fluctuations, and response time deviations. Read / write response stability can be normalized to a value between 0 and 1; a higher value indicates a more stable electronic response. If the number of consecutive read / write failures increases, response intensity fluctuations increase, or the response time exceeds a preset range, the read / write response stability decreases.

[0079] The controller calculates the adsorption establishment stability based on the negative pressure build-up time, negative pressure holding value, negative pressure fluctuation amplitude, and negative pressure decay slope. Adsorption establishment stability can be normalized to a value between 0 and 1; a higher value indicates that the adsorption actuator's adsorption state for the current tag is closer to the preset tag-taking state. If the negative pressure build-up time is prolonged, the negative pressure holding value is lower than the set range, the negative pressure fluctuation amplitude increases, or the negative pressure decay slope is abnormal, the adsorption establishment stability decreases.

[0080] The controller determines the conveyor disturbance level based on conveyor speed fluctuations, conveyor tension changes, peel angle deviations, and peel speed variations. The conveyor disturbance level characterizes the degree of mechanical disturbance experienced by the current individual label during release film peeling and conveying.

[0081] Through the above processing, the obtained sheet material state characteristics include edge warping characteristics, local wrinkle characteristics, visual confidence of electronic functional areas, positioning reference point deviation, read / write response stability, adsorption establishment stability, and conveying disturbance level.

[0082] S3, Constructing an attachment feasibility feature map

[0083] The controller maps the sheet material status features obtained by S2 to the label plane coordinate system to construct a feature map of the current single smart label's attachment feasibility.

[0084] Specifically, the smart tag is divided into multiple two-dimensional grid units in the tag plane coordinate system, and each two-dimensional grid unit is assigned to the edge area, center area, adhesive layer transition area, antenna area, and chip area according to the tag's physical structure. For each grid unit, the controller calculates the attachment risk value based on the visual risk, electronic response risk, adsorption risk, and transport disturbance risk corresponding to that grid unit.

[0085] In this embodiment, the first Attachment risk value of each grid cell It can be determined in the following way:

[0086] when ;when ;when hour, .

[0087] in, Indicates the first The visual risk value corresponding to each grid cell Indicates the first The electronic response risk value corresponding to each grid cell Indicates the first The adsorption risk value corresponding to each grid cell Indicates the first The transmission disturbance risk value corresponding to each grid cell; , , , These are the weighting coefficients corresponding to visual risk, electronic response risk, adsorption risk, and transport disturbance risk, respectively. , , and All are normalized to the range of 0 to 1. , , and All are non-negative numbers and satisfy the following conditions: + + + =1.

[0088] In one specific implementation, for the grid cells corresponding to the antenna region and the chip region, It can be taken as 0.35 to 0.55. It can be between 0.15 and 0.35. It can be between 0.10 and 0.30. It can be taken as 0.05 to 0.20; for the grid cells corresponding to the edge region and the adhesive layer transition region, It can be between 0.30 and 0.50. It can be between 0.15 and 0.35. It can be taken as 0.05 to 0.20. The values ​​can range from 0.05 to 0.20. Each weighting coefficient can be determined based on the label specifications, the location of the electronic functional area, the reflectivity of the coating, and the production line calibration results.

[0089] For grid cells whose feature values ​​are not directly acquired, the controller interpolates and completes the data based on the attachment risk values ​​of adjacent grid cells. For grid cells corresponding to the antenna area and chip area, the controller configures a higher electronic response risk weight than that for ordinary edge areas and center areas, so that the abnormal state of the electronic functional area has a greater impact on the release decision.

[0090] The adhesion feasibility feature map is used to characterize the overall adhesion risk of the current label, as well as the degree of influence of different local areas of the label on adhesion misalignment, edge wrinkling, local bubbles, and abnormal electronic response. Through this feature map, the controller can determine that the risks mainly originate from edge warping, abnormal adhesive layer transition areas, abnormal antenna area read / write operations, decreased visual confidence in the chip area, insufficient adsorption, or transport disturbances.

[0091] S4, Generate candidate attachment actions

[0092] The controller generates candidate attachment actions based on the attachment feasibility feature map. Candidate attachment actions include direct attachment actions, pose-compensated attachment actions, pressure-compensated attachment actions, extended pressure-holding attachment actions, secondary adsorption and leveling actions, and release-prevention actions.

[0093] Direct attachment refers to the robot attaching the label according to a preset attachment path, preset attachment pressure, and preset holding time. Position-compensated attachment refers to attaching the label after compensating for deviations in the positioning reference point, antenna area offset direction, or edge warping direction, adjusting the robot's end effector's lateral position, longitudinal position, or rotation angle. Pressure-compensated attachment refers to adjusting the attachment pressure or speed based on risks of edge warping, air bubbles, or adhesive layer transition zones. Extended holding time attachment refers to extending the holding time of the robot's end effector after the label contacts the target carrier. Secondary adsorption and leveling refers to the robot or attachment station first adsorbing the label again, pre-pressing and leveling it, locally flattening it, reducing the peeling speed, or pausing the conveyor before re-collecting multi-source station data and performing a re-evaluation. Prohibition of release refers to stopping the robot's attachment of the current single label and marking the label as abnormal.

[0094] In this embodiment, candidate attachment actions are generated based on the risk sources in the attachment feasibility feature map. For example, when high-risk areas are concentrated in the edge area and the electronic response is normal, the controller generates pre-pressure leveling, extended pressure holding, or pressure compensation attachment actions; when the antenna area read / write response is continuously abnormal and the proportion of high-risk area exceeds a preset upper limit, the controller generates a prohibition on release action; when the positioning reference point deviation is within the compensable range, the controller generates a pose compensation attachment action.

[0095] S5, predicts the attachment consequences of candidate attachment actions.

[0096] The controller inputs the attachment feasibility feature map, candidate attachment action parameters, robot end state data and delivery state data into the attachment consequence prediction model to obtain the attachment consequence prediction results corresponding to each candidate attachment action.

[0097] Candidate attachment parameters include end-to-end lateral compensation, longitudinal compensation, rotation angle compensation, attachment pressure, attachment speed, holding time, and whether secondary adsorption leveling is performed. Attachment consequence predictions include risks of attachment misalignment, edge wrinkling, localized bubbles, and abnormal electronic response; each risk value can be normalized to a value between 0 and 1.

[0098] In this embodiment, the attachment consequence prediction model is trained based on historical attachment condition data. This historical data includes pre-attachment multi-source workstation data, attachment feasibility feature maps, actual attachment action parameters performed by the robot, and post-attachment detection results. The post-attachment detection results include actual attachment position deviation, edge wrinkling detection results, local bubble detection results, and RFID / NFC retest results.

[0099] In one specific implementation, the attachment consequence prediction model employs a gradient boosting decision tree model. The model input is a feature vector, which includes the overall mean and variance of the attachment feasibility feature map, the maximum risk value in the antenna area, the maximum risk value in the chip area, the proportion of high-risk grid areas in the edge area, the proportion of high-risk grid areas in the adhesive layer transition area, the negative pressure establishment time, the negative pressure holding value, the negative pressure fluctuation amplitude, the read / write response stability, the transport speed fluctuation, the peeling angle deviation, the end-effectoral compensation amount, the end-effector longitudinal compensation amount, the end-effector rotation angle compensation amount, the attachment pressure, the attachment speed, and the holding time. The model output consists of four risk values, corresponding to the risks of attachment offset, edge wrinkling, localized bubbles, and abnormal electronic response, respectively.

[0100] During training, historical attachment condition samples were divided into training and validation sets. Each set of historical attachment condition samples included pre-attachment input features, actual attachment action parameters, and post-attachment test results. The post-attachment test results were obtained through post-attachment image detection and RFID / NFC retesting, and were converted into attachment offset risk tags, edge wrinkling risk tags, local bubble risk tags, and electronic response anomaly risk tags. After training, the validation set was used to check the deviation between the model's output risk and the post-attachment test results. When the deviation met the preset requirements, the model was used to predict the consequences of candidate actions during the production line operation phase.

[0101] In another implementation, the attachment consequence prediction model can employ a support vector regression model, a classification regression model, or a lightweight neural network model. Regardless of the model used, its inputs are the physical condition data of the attachment station and the robot's motion parameters, and its outputs are the risk values ​​corresponding to the physical inspection results after attachment.

[0102] Preferably, the model output can be expressed as:

[0103]

[0104] in, Indicates the first Prediction results of attachment consequences for each candidate attachment action. This indicates the feasibility feature map to be attached. Indicates the first Candidate attachment action parameters This represents the robot's end-effector state data. This indicates the transmission status data. These include risks such as attachment misalignment, edge wrinkling, localized bubbles, and abnormal electronic response.

[0105] S6, Determine the dynamic release status and output compensation parameters.

[0106] Based on the predicted consequences of affixing, the distribution of high-risk areas in the affixing feasibility feature map, and electronic response data, the controller determines the current dynamic release status of a single smart tag and outputs the corresponding compensation parameters.

[0107] In this embodiment, the dynamic release status includes direct release status, compensated release status, leveling and waiting for reassessment status, and prohibited release status.

[0108] When the risks of attachment offset, edge wrinkling, local bubble formation, and abnormal electronic response associated with the direct attachment action are all below the corresponding direct release thresholds, the controller determines that the current single smart tag is in a direct release state and controls the robot to perform attachment according to a preset attachment path. In one specific embodiment, the direct release thresholds can be set as follows: attachment offset risk no greater than 0.15 to 0.25, edge wrinkling risk no greater than 0.20 to 0.30, local bubble formation risk no greater than 0.18 to 0.30, and abnormal electronic response risk no greater than 0.10 to 0.20.

[0109] When the risk of attachment misalignment, edge wrinkling, or localized air bubbles exceeds the direct release threshold but does not exceed the compensation upper limit threshold, and the RFID / NFC read / write response is within an acceptable range, the controller determines that the current single smart tag is in a compensated release state. Based on the source of the risk, it outputs one or more of the following compensation amounts: lateral compensation, longitudinal compensation, rotation angle compensation, attachment pressure compensation, attachment speed compensation, and holding time compensation. In one specific implementation, the compensation upper limit threshold can be set as follows: attachment misalignment risk not exceeding 0.45 to 0.60, edge wrinkling risk not exceeding 0.45 to 0.60, and localized air bubble risk not exceeding 0.45 to 0.60. An acceptable RFID / NFC read / write response means that the read / write response stability is not lower than 0.70, and the number of failures does not exceed 1 in 3 to 5 consecutive read / write attempts.

[0110] When edge warping characteristics, adsorption establishment stability, or transport disturbance level indicate that the current single tag has a recoverable adhesion risk, the controller determines that the current single smart tag is in a leveling and re-judgment state. A recoverable adhesion risk refers to a risk mainly distributed in the edge area or adhesive layer transition area, RFID / NFC read / write response stability not lower than a preset lower limit, the proportion of high-risk areas in the antenna area and chip area not exceeding a preset upper limit, and the negative pressure establishment process being able to meet the tag removal requirements after secondary adsorption or pre-pressure leveling. When in the leveling and re-judgment state, the controller first triggers secondary adsorption, pre-pressure leveling, reduction of peeling speed, or transport pause actions, then re-acquires the multi-source station data for the single tag and re-executes S2 to S6.

[0111] When RFID / NFC read / write responses are continuously abnormal, the proportion of high-risk areas in the antenna or chip area exceeds a preset upper limit, negative pressure cannot be established within a preset time, or there are unrecoverable high-risk areas in the attachment feasibility feature map, the controller determines that the current single smart tag is in a prohibited release state, stops the robot's attachment action, and marks the single smart tag as an abnormal tag. An unrecoverable high-risk area refers to a high-risk area located in the antenna, chip, or conductive connection area, and this area simultaneously meets one or more of the following conditions: visual confidence level is below a preset threshold, read / write response stability is below a preset threshold, or negative pressure adsorption cannot cover the area. In one specific embodiment, the preset upper limit for the proportion of high-risk areas in the antenna or chip area can be set to 10% to 25%, and the overall unrecoverable high-risk area threshold for the tag can be set to 15% to 30% of the total tag area.

[0112] Based on the above determination, the risk value output by the prediction model is transformed into specific control results such as robot end-effector pose compensation, attachment pressure compensation, pressure holding time compensation, conveying speed correction, or prohibition of attachment.

[0113] S7, executes the attachment action and collects feedback results after attachment.

[0114] When the current single smart tag is in direct release or compensated release state, the robot control unit controls the robot's attachment actuator to move according to the attachment path and compensation parameters output by the controller, so that the adsorption actuator attaches the smart tag to the target carrier.

[0115] After the affixing is completed, the vision acquisition unit acquires the image after affixing, and the RFID / NFC read / write unit acquires the electronic response retest data after affixing. The controller obtains the actual affixing position deviation, edge wrinkling results, and local bubble results based on the image after affixing; and obtains the read / write success status, response strength, and response time based on the electronic response retest data after affixing.

[0116] S8, Update the release judgment parameters based on the feedback results.

[0117] The controller correlates the multi-source station data before attachment, the attachment feasibility feature map, the dynamic release status, the robot compensation parameters, and the feedback results after attachment to form an attachment feedback sample.

[0118] Feedback updates are triggered when the cumulative number of processed labels reaches a preset quantity, or when any of the following indicators—adhesion position deviation rate, edge wrinkling rate, local bubble rate, or electronic response anomaly rate—increases over multiple consecutive statistical periods. These multiple consecutive statistical periods can be 2 to 5 periods, and each period can be determined based on a preset number of labels or a preset production time, such as 500 labels or 30 minutes per period.

[0119] In one specific implementation, the feedback update includes adjusting the output layer calibration parameters of the attachment consequence prediction model, or updating the direct release threshold, compensated release threshold, leveling pending review threshold, and prohibited release threshold based on an exponential moving average method. A certain release threshold can be updated according to the following formula:

[0120]

[0121] in, This indicates the updated threshold. This represents the threshold value before the update. This indicates a reference threshold determined based on the most recent batch of sticker feedback samples. This is the smoothing coefficient. The value can range from 0.70 to 0.95, preferably 0.85. The threshold can be determined based on the quantile of the corresponding risk value in the most recent batch of attached feedback samples; for example, the 90th or 95th percentile of the corresponding risk value in the most recent batch of samples can be used as a reference threshold.

[0122] Preferably, each feedback update generates a corresponding parameter version number, and saves the thresholds, model calibration parameters, and statistical indicators before and after the update. When the updated parameters cause an increase in the attachment position deviation rate, edge wrinkling rate, local bubble rate, or electronic response anomaly rate, the controller reverts to the release judgment parameters of the previous version.

[0123] Through steps S1 to S8, this embodiment establishes a closed-loop control process encompassing pre-attachment multi-source data acquisition, sheet material state feature extraction, attachment feasibility feature map construction, candidate action consequence prediction, dynamic release judgment, robot compensation execution, and post-attachment feedback updates. Through this processing, visual images, RFID / NFC read / write responses, negative pressure detection data, conveying status data, and robot attachment control parameters are established within the same single-tag data window. The pre-attachment risk assessment results can be transformed into control parameters for robot end-effector pose, attachment pressure, attachment speed, holding time, and conveying status.

[0124] Example 2:

[0125] This embodiment provides a dynamic release system for a smart label attaching robot, such as... Figure 2 and Figure 3 As shown, the system includes a data acquisition module, a status feature extraction module, an attachment feasibility construction module, a candidate action generation module, a consequence prediction module, a dynamic release module, a compensation output module, and a feedback update module.

[0126] The system in this embodiment is communicatively connected to a vision acquisition unit, an RFID / NFC read / write unit, a negative pressure detection unit, a transport status detection unit, a robot control unit, and a transport control unit. The vision acquisition unit, RFID / NFC read / write unit, negative pressure detection unit, and transport status detection unit respectively input the visual image, electronic response data, adsorption status data, and transport status data of the current single smart tag into the system; the robot control unit and the transport control unit receive attachment compensation parameters, transport correction parameters, or stop-attachment commands output by the system.

[0127] The data acquisition module is used to acquire multi-source workstation data of the smart tag to be attached. This multi-source workstation data includes visual images, electronic response data, adsorption status data, conveying status data, and robot end-effector status data. The data acquisition module establishes a data window for the current single tag based on the tag arrival signal or the robot tag-picking trigger signal, and outputs the data from different acquisition sources after timestamping and then to the status feature extraction module.

[0128] The state feature extraction module is used to extract the current single tag's sheet state features from multi-source workstation data. Specifically, the state feature extraction module extracts the tag's outer contour, positioning marks, adhesive layer transition area, antenna area, and chip area based on visual images, and outputs edge warping features, local wrinkle features, visual confidence level of electronic functional areas, and positioning reference point deviation; it outputs read / write response stability based on electronic response data; it outputs adsorption establishment stability based on adsorption state data; and it outputs conveying disturbance level based on conveying state data.

[0129] The adhesion feasibility construction module is used to construct an adhesion feasibility feature map based on the sheet material state characteristics output by the state feature extraction module. Specifically, the adhesion feasibility construction module establishes a label planar coordinate system based on the label's outer contour center point, positioning mark, or antenna reference point, dividing the label into edge area, center area, adhesive layer transition area, antenna area, and chip area, and further dividing them into two-dimensional grid units. The adhesion feasibility construction module generates an adhesion risk value based on the visual risk, electronic response risk, adsorption risk, and transport disturbance risk corresponding to each grid unit, and outputs the adhesion feasibility feature map.

[0130] The candidate action generation module is used to generate candidate attachment actions based on the attachment feasibility feature map. The candidate attachment actions include direct attachment actions, pose-compensated attachment actions, pressure-compensated attachment actions, extended pressure-holding attachment actions, secondary adsorption and leveling actions, and release-prevention actions.

[0131] The consequence prediction module is used to predict the attachment consequences corresponding to each candidate attachment action. The module receives attachment feasibility feature maps, candidate attachment action parameters, robot end-effector state data, and transport state data, and outputs the risks of attachment offset, edge wrinkling, localized bubbles, and abnormal electronic response. The prediction model in the consequence prediction module is trained using historical pre-attachment data, actual action parameters, and post-attachment measurement results.

[0132] The dynamic release module is used to determine the current dynamic release status of a single smart tag based on the output of the consequence prediction module, the distribution of high-risk areas in the affixing feasibility feature map, and electronic response data. The dynamic release status includes direct release, compensated release, leveling and awaiting reassessment, and prohibited release.

[0133] The compensation output module generates corresponding control outputs based on the judgment results of the dynamic release module. When the dynamic release state is direct release, the compensation output module outputs preset attachment parameters; when the dynamic release state is compensated release, the compensation output module outputs one or more of the following: robot end-effector lateral compensation, longitudinal compensation, rotation angle compensation, attachment pressure compensation, attachment speed compensation, and holding time compensation; when the dynamic release state is leveling and awaiting re-judgment, the compensation output module outputs control parameters corresponding to secondary adsorption, pre-pressure leveling, reduced peeling speed, or paused delivery; when the dynamic release state is prohibited release, the compensation output module outputs a stop attachment command and abnormal label marking information.

[0134] The feedback update module receives post-attachment images and electronic response retest data, and updates the release judgment parameters based on the post-attachment inspection results. Specifically, the feedback update module associates pre-attachment multi-source workstation data, attachment feasibility feature maps, dynamic release status, robot compensation parameters, post-attachment images, and electronic response retest data as attachment feedback samples. When the cumulative number of processed tags reaches a preset number, or when any of the following indicators—attachment position deviation rate, edge wrinkling rate, local bubble rate, or electronic response anomaly rate—increases within several consecutive statistical periods, the feedback update module updates the model calibration parameters of the consequence prediction module, the weight parameters of the attachment feasibility construction module, or the threshold parameters of the dynamic release module.

[0135] Furthermore, the feedback update module generates a parameter version number for each update and saves the model calibration parameters, threshold parameters, and statistical indicators before and after the update. When the updated parameters cause an increase in the attachment position deviation rate, edge wrinkling rate, local bubble rate, or electronic response anomaly rate, the feedback update module sends a rollback command to the consequence prediction module, attachment feasibility construction module, or dynamic release module to restore the system to the previous version of parameters.

[0136] Through the above system structure, visual images, RFID / NFC read / write responses, negative pressure detection data, conveying status data, and robot end-effector status data are associated within the data window of the current single tag. After sequentially undergoing status feature extraction, attachment feasibility construction, candidate action generation, consequence prediction, and dynamic release processing, control parameters are formed that act on the robot's attachment actuator, adsorption actuator, conveying mechanism, or release film peeling mechanism.

[0137] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic release of a robot at an intelligent label attaching station, characterized in that, Includes the following steps: Acquire multi-source workstation data for the smart tag to be attached, including visual images, electronic response data, adsorption status data, conveying status data, and robot end-effector status data; Extract the sheet material status characteristics of the current single smart tag based on multi-source workstation data; A label plane coordinate system is established based on the current positioning reference point of a single smart label. The material status characteristics are mapped to the label plane coordinate system to construct an attachment feasibility feature map. Candidate attachment actions are generated based on the attachment feasibility feature map; Before the robot attaches the action, the attaching feasibility feature map, candidate attaching action parameters, robot end state data and transport state data are input into the attaching consequence prediction model to obtain the attaching consequence prediction results corresponding to each candidate attaching action. Based on the predicted consequences of affixing, the feasibility feature map of affixing, and the electronic response data, the current dynamic release status of a single smart tag is determined. The dynamic release status includes direct release status, compensated release status, leveling and waiting for re-judgment status, and prohibited release status. Dynamic release control results are generated based on the dynamic release status. Specifically, direct release status and compensated release status correspond to the generation of attachment control parameters, leveling pending re-judgment status corresponds to the generation of transport correction parameters, and prohibited release status corresponds to the generation of stop attachment command.

2. The method for dynamic release of a robot at an intelligent label attaching station according to claim 1, characterized in that, The acquisition of multi-source workstation data for the smart label to be attached includes: The reference time is determined by the tag placement signal or the robot tag-picking trigger signal; Establish a preset time window that includes the reference time; Within a preset time window, visual images, electronic response data, adsorption status data, delivery status data, and robot end effector status data are collected. The visual images, electronic response data, adsorption status data, delivery status data, and robot end-effector status data within the preset time window are associated with the current single smart tag according to the timestamp.

3. The method for dynamic release of a robot at an intelligent label attaching station according to claim 2, characterized in that, The smart tag to be attached includes an antenna area and a chip area; The sheet material state characteristics include edge warping characteristics, local wrinkling characteristics, visual confidence of electronic functional areas, positioning reference point deviation, read / write response stability, adsorption establishment stability, and transport disturbance level. Among them, the electronic response data includes the number of successful read / write operations, response intensity, and response time obtained by performing electronic function read / write operations on the antenna area and the chip area. The adsorption state data includes the negative pressure establishment and maintenance data of the adsorption actuator; The conveying status data includes the operating status data of the conveying mechanism and the release film peeling mechanism.

4. The method for dynamic release of a robot at an intelligent label attaching station according to claim 3, characterized in that, The constructed attachment feasibility feature map includes: The current single smart tag is divided into multiple two-dimensional grid units in the tag plane coordinate system; Based on the position of each two-dimensional grid unit in the current single smart tag, determine whether the area to which the two-dimensional grid unit belongs is the edge area, the center area, the adhesive layer transition area, the antenna area, or the chip area; Based on the visual risk, electronic response risk, adsorption risk and transport disturbance risk corresponding to each two-dimensional grid cell, the attachment risk value of that two-dimensional grid cell is generated. The attachment feasibility feature map is formed by the attachment risk values ​​of multiple two-dimensional grid cells.

5. The method for dynamic release of a robot at an intelligent label attaching station according to claim 4, characterized in that, The generation of the attachment risk value includes: For the two-dimensional grid cells corresponding to the antenna region and the chip region, an electronic response risk weight higher than that of the central region is configured. For the two-dimensional grid cells corresponding to the edge region and the adhesive layer transition region, configure a higher visual risk weight or adsorption risk weight than that of the center region.

6. The method for dynamic release of a robot at an intelligent label attaching station according to claim 1, characterized in that, The candidate attachment actions include direct attachment actions, position compensation attachment actions, pressure compensation attachment actions, extended pressure holding attachment actions, secondary adsorption and leveling actions, and release prohibition actions. The candidate attachment action parameters include the robot end effector's lateral compensation amount, longitudinal compensation amount, rotation angle compensation amount, attachment pressure, attachment speed, pressure holding time, and whether secondary adsorption and leveling are performed.

7. The method for dynamic release of a robot at an intelligent label attaching station according to claim 6, characterized in that, The predicted results of the attachment consequences include the risk of attachment misalignment, risk of edge wrinkling, risk of localized bubbles, and risk of abnormal electronic response. The attachment consequence prediction model is trained based on historical attachment condition data, which includes multi-source workstation data before attachment, attachment feasibility feature map, attachment action parameters actually performed by the robot, and post-attachment detection results. The post-attachment detection results include actual attachment position deviation, edge wrinkling detection results, local bubble detection results, and electronic function retest results. The results of the attachment consequence prediction are used to determine the dynamic release status, and the dynamic release status is used to generate attachment control parameters, transmit correction parameters, or stop attachment commands.

8. The method for dynamic release of a robot at an intelligent label attaching station according to claim 7, characterized in that, The dynamic release status includes: When the risks of attachment offset, edge wrinkling, local bubble, and abnormal electronic response corresponding to the direct attachment action are all lower than the corresponding direct release threshold, the current single smart tag is determined to be in the direct release state. When any of the risks of attachment offset, edge wrinkling, and local bubble exceeds the corresponding direct release threshold, but none of the risks of attachment offset, edge wrinkling, and local bubble exceed the corresponding compensation upper limit threshold, and the read / write response stability is not lower than the preset lower limit, the current single smart tag is determined to be in the compensation release state. When the high-risk area in the attachment feasibility feature map is located in the edge area or adhesive layer transition area, the proportion of high-risk area in the antenna area and chip area does not exceed the preset upper limit, and the read / write response stability is not lower than the preset lower limit, the current single smart tag is determined to be in the flattening and re-judgment state, and after performing secondary adsorption, pre-pressing flattening, reducing peeling speed or pausing conveying, the multi-source station data is re-acquired. The current single-chip smart tag is determined to be in a prohibited release state when any of the following conditions are met: the read / write response stability is lower than the preset lower limit and the electronic response data is continuously abnormal; the proportion of high-risk area in at least one of the antenna area and chip area exceeds the preset upper limit; the negative pressure cannot be established within the preset time; or at least one of the antenna area and chip area contains a high-risk area that meets the prohibited release conditions.

9. The method for dynamic release of a robot at an intelligent label attaching station according to claim 1, characterized in that, The method also includes: after the bonding is completed, acquiring post-bonding images and electronic response retest data; and associating pre-bonding multi-source workstation data, bonding feasibility feature map, dynamic release status, bonding control parameters, transport correction parameters, post-bonding images and electronic response retest data as bonding feedback samples. The calibration parameters of the adhesion consequence prediction model, the weight parameters of the adhesion feasibility feature map, or the discrimination threshold of the dynamic release status are updated based on the adhesion feedback samples. Each time an update is performed, the calibration parameters, weight parameters, or discrimination thresholds from before the update are saved. If the updated parameters cause an increase in the abnormal adhesion index, the calibration parameters, weight parameters, or discrimination thresholds saved before the update are restored.

10. A dynamic release system for an intelligent label-attaching robot, characterized in that, It includes a controller, which is communicatively connected to a vision acquisition unit, an electronic function reading and writing unit, a negative pressure detection unit, a conveying status detection unit, a robot control unit, and a conveying control unit; The controller is configured with a data acquisition module, a state feature extraction module, an attachment feasibility construction module, a candidate action generation module, a consequence prediction module, a dynamic release module, a compensation output module, and a feedback update module, among which: The data acquisition module is used to acquire multi-source workstation data to be attached to the smart tag, and output the multi-source workstation data to the status feature extraction module; The status feature extraction module is used to extract the current status features of a single smart tag from multi-source workstation data; The adhesion feasibility construction module is used to map the sheet material state characteristics to the label plane coordinate system and construct an adhesion feasibility feature map. The candidate action generation module is used to generate candidate attachment actions based on the attachment feasibility feature map; The consequence prediction module is used to output the prediction results of the attachment consequences for each candidate attachment action based on the attachment feasibility feature map, candidate attachment action parameters, robot end state data and delivery state data. The dynamic release module is used to determine the current dynamic release status of a single smart tag based on the prediction results of the affixing consequences, the affixing feasibility feature map, and the electronic response data. The compensation output module is used to output attachment control parameters to the robot control unit, output delivery correction parameters to the delivery control unit, and output a stop attachment command to the robot control unit according to the dynamic release status. The feedback update module is used to generate attachment feedback samples based on the post-attachment images and electronic response retest data, and update the calibration parameters of the attachment consequence prediction model, the weight parameters of the attachment feasibility feature map, or the discrimination threshold of the dynamic release status.

Images