A shoe sole gluing track self-adaptive control method based on dynamic scanning and feedback adjustment

Abstract

The application discloses a kind of based on dynamic scanning and feedback regulation's shoe sole glue applying track self-adaptive control method, this method includes: through the multimodal perception of 3D vision sensor and multispectral sensing module integrated in robot end effector, reliable glue applying track is planned by fusing geometry and material information;Based on the track, through the endogenous interference compensation amount of multi-physical field coupling compensation model calculation, and drive robot to execute glue applying, while recording actual control track;Actual glue line track is extracted using downstream quality inspection camera;By comparing actual control track with actual glue line track, the absolute positioning error of robot is calculated, and error mapping table is updated to carry out feedforward compensation to subsequent glue applying instruction.The application effectively overcomes the interference of pollutant and endogenous interference, realizes the high-precision, high-reliability control of glue applying track, and has on-line self-calibration capability, ensures the stability of long-term operation of system.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, specifically an adaptive control method for shoe sole adhesive application trajectory based on dynamic scanning and feedback adjustment. Background Technology

[0002] Shoe sole gluing is a crucial step in the shoe manufacturing process, directly determining the adhesive strength, durability, and aesthetics of the footwear. With the development of industrial automation technology, the use of robots for automated gluing has become an industry trend. Currently, the development of this technology has mainly gone through the following stages and faces corresponding technical bottlenecks: The initial stage of robotic glue application systems employed pre-programmed teaching technology. Operators pre-taught a fixed glue application trajectory, which the robot then repeatedly executed. While this method achieved automation, it lacked flexibility and could not adapt to minor positioning errors of the shoe sole on the fixture, deviations in incoming material dimensions, or deformation of the sole itself. This resulted in the glue line deviating from the preset application channel, leading to unstable bonding quality. Such systems were only suitable for production scenarios with a single product type and low precision requirements.

[0003] To overcome the limitations of pre-programmed technology, an adaptive gluing technology based on 3D vision (such as line lasers and structured light) was introduced. This system uses a 3D vision sensor mounted on the robot's end effector to scan the sole surface in real time, extract the three-dimensional geometric information of the gluing channels, compare it with a preset theoretical trajectory, and generate correction commands to drive the robot to perform tracking gluing.

[0004] However, this technology has the following inherent drawbacks: Susceptible to interference from on-site contaminants: In the production environment, the surface of the shoe sole may be contaminated with residual treatment agents, dust, or residual glue from the previous cycle. These contaminants may have a highly similar three-dimensional geometric shape to the actual glue application channel, causing the vision system to misidentify them and apply glue incorrectly to the contaminants instead of the actual channel, resulting in glue waste and serious quality defects.

[0005] Ignoring "endogeneous interference": This technology only compensates for the external relative pose deviation between the sole and the robot, completely ignoring the systematic errors introduced by the glue application process itself. During the spraying of hot melt adhesive, its heat is transferred to the sole material, causing minor thermal deformation in localized areas. The vision system scans the undeformed sole before glue application, but by the time the glue application valve is activated, the scanned points have already deformed due to heat. This change in the state of the controlled object (sole) caused by the behavior of the control system itself (glue application), i.e., "endogeneous interference," causes deviations in the compensation control based on the initial state, limiting further improvements in accuracy.

[0006] System accuracy cannot be maintained in the long term: After long-term operation, the absolute positioning accuracy of industrial robots will drift due to factors such as gear backlash and link wear. This means that even if visual recognition and trajectory calculation are completely correct, there will still be slight errors in the actual execution position of the robot. The traditional solution is to periodically stop production and use specialized equipment such as laser trackers for on-site calibration. This method is inefficient, costly, and cannot meet the needs of continuous production.

[0007] In summary, existing technologies still face three major challenges: (1) insufficient sensing reliability, making them susceptible to interference from foreign objects; (2) theoretical bottlenecks in control precision, failing to compensate for endogeneous interference; and (3) a lack of long-term self-maintaining precision capability, relying on manual maintenance. Therefore, there is an urgent need in this field for an intelligent glue application control method that can fundamentally solve the above problems and achieve high reliability, high precision, and self-maintaining precision. Summary of the Invention

[0008] The purpose of this invention is to provide an adaptive control method for the adhesive application trajectory of shoe soles based on dynamic scanning and feedback adjustment. This invention effectively overcomes contaminant interference and endogeneous interference, achieves high-precision and high-reliability control of the adhesive application trajectory, and has online self-calibration capability, ensuring the stability of the system's long-term operation.

[0009] The technical solution adopted in this invention is as follows: An adaptive control method for sole adhesive application trajectory based on dynamic scanning and feedback adjustment includes the following steps: S1. Multimodal perception and trajectory planning: S11. Using the 3D vision sensor integrated on the robot's end effector, the surface of the shoe sole is scanned in real time to obtain three-dimensional point cloud data; S12. Based on the three-dimensional point cloud data, identify suspected trajectory regions with channel geometric features; S13. Perform material analysis on the suspected trajectory region using a multispectral sensing module to obtain its material characteristic information; S14. The material feature information is compared with the pre-stored material features of the sole body. Only the area where both the geometric features and material features match the sole body is identified as the real glue application trajectory, and a real-time glue application trajectory is planned accordingly. S2. Feedback Adjustment and Adhesive Application Execution: Based on the real-time adhesive application trajectory planned in step S1, control the robot to drive the adhesive application valve to perform adhesive application operations, and record the actual control trajectory data corresponding to the adhesive application operations; S3. Downstream glue line trajectory extraction: At the quality inspection station downstream of the glue application process, obtain an image of the glued shoe sole and extract the actual glue line trajectory from the image; S4. Online error calibration and compensation: Compare the actual control trajectory with the actual glue line trajectory, calculate the robot's absolute positioning error, and perform feedforward compensation for subsequent glue application control commands for the shoe sole based on this error.

[0010] In step S2, when generating control commands for the real-time adhesive application trajectory, a multi-physics coupling compensation model is introduced to compensate for endogeneous interference caused by the adhesive application process itself; the multi-physics coupling compensation model at least considers the influence of the thermal effect of the adhesive on the state of the sole material.

[0011] The multiphysics coupling compensation model is a thermo-mechanical coupling model, which predicts the temperature field change in the adhesive application point area by input adhesive application process parameters, and then calculates the compensation amount for visual measurement error caused by material thermal deformation.

[0012] In step S13, the multispectral sensing module is a near-infrared spectral probe, and the material characteristic information is the reflectance in a specific near-infrared band.

[0013] In step S4, calculating the absolute positioning error of the robot specifically includes: converting the actual glue line trajectory to the robot coordinate system through hand-eye calibration, then performing point cloud registration with the actual control trajectory to calculate the spatial pose deviation.

[0014] An adaptive glue application system for implementing the adaptive control method for the glue application trajectory of the shoe sole includes: The glue application robot unit includes a robot, an end effector, a 3D vision sensor, a multispectral sensing module, and a glue application valve; Downstream quality inspection units, including industrial cameras; The system controller is configured as follows: The 3D vision sensor and the multispectral sensing module are controlled to perform multimodal perception and plan real-time adhesive application trajectory. Record the robot's execution trajectory as the actual control trajectory; Receive the actual adhesive line trajectory sent by the downstream quality inspection unit; Execute online error calibration and compensation algorithms, and correct the robot's control commands accordingly.

[0015] The multispectral sensing module includes a multi-band LED light source and a photoelectric sensor.

[0016] The end effector also integrates a pulsed air curtain protection device, which is linked to the glue application valve. When a glue breakage signal is detected, an airflow is instantly ejected to prevent the glue filaments from getting close to and adhering to the 3D vision sensor lens.

[0017] The system controller includes a pose compensation controller, which incorporates the multiphysics coupling compensation model.

[0018] The system controller is further configured to divide the robot's workspace into a three-dimensional grid, and based on the continuous comparison of the actual control trajectory and the actual glue line trajectory, use a moving average algorithm to update the absolute positioning error value in each grid, forming a dynamically updated error mapping table for feedforward compensation.

[0019] This invention relates to an adaptive control method and system for shoe sole adhesive application trajectory based on dynamic scanning and feedback adjustment. Compared with existing technologies, this invention achieves significant beneficial effects by integrating three technologies: multimodal sensing, endogeneous interference compensation, and system self-calibration. 1. Existing adhesive application systems based on single 3D vision rely solely on geometric features to identify the application trajectory, making them susceptible to misidentification due to interference from contaminants such as residual adhesive and processing agents. This invention introduces a multispectral sensing module to achieve fusion verification of geometric and material features. Only when the target area simultaneously meets both geometric morphology and material spectral requirements is it confirmed as a genuine adhesive application trajectory. This mechanism eliminates misidentification at the source of perception, improving the reliability of trajectory planning to nearly 100%, and fundamentally solving the adhesive application quality problems caused by on-site contaminants in traditional methods.

[0020] 2. Traditional adaptive control only focuses on external disturbances, neglecting the endogenous interferences caused by the adhesive application process itself. This invention establishes a multi-physics coupling compensation model, predicting thermo-mechanical field changes through real-time adhesive application parameters (adhesive temperature, flow rate) and calculating the corresponding deformation compensation. This compensation, combined with the trajectory deviation, generates a comprehensive control command, enabling the system to proactively offset errors introduced by its own process, thus elevating control precision from macroscopic geometric matching to the level of microscopic physical effect compensation.

[0021] 3. The absolute positioning accuracy of industrial robots drifts over time, and traditional solutions require periodic shutdowns for manual calibration. This invention innovatively utilizes images of the adhesive lines at downstream quality inspection stations to extract the actual adhesive line trajectory and compares it with the recorded actual control trajectory. A point cloud registration algorithm is then used to calculate the robot's absolute positioning error. A moving average algorithm is employed to continuously update the workspace error mapping table and provide feedforward compensation for subsequent processing commands. This self-calibration mechanism achieves "calibration during production," significantly reducing system maintenance frequency and downtime, and ensuring long-term accuracy stability.

[0022] 4. The pulsed air curtain protection device integrated in this invention transcends the traditional passive cleaning model. This device transforms the inherent "adhesive failure" event during the adhesive application process (detected by a sudden change in air pressure signal) into a trigger condition, actively intercepting contaminants by activating an instantaneous air curtain within milliseconds before they reach the sensor lens. This method of transforming negative process characteristics into positive protective signals achieves a paradigm shift from "post-event handling" to "pre-event prevention," significantly extending the maintenance-free operating time of core sensing components. This effect has an unexpectedly positive impact on ensuring the continuity of high-cycle production.

[0023] 5. The online self-calibration mechanism of this invention constructs a macroscopic closed loop spanning the application and quality inspection stations. Downstream quality information is continuously fed back to the upstream control end to optimize execution accuracy, transforming the originally isolated manufacturing unit into a perception-decision-execution-learning loop similar to a biological "reflex arc." This system exhibits "life-like" characteristics, capable of learning from its own operational experience and continuously optimizing its performance, realizing the evolution from rigid automated equipment to an intelligent system with adaptive and evolutionary capabilities.

[0024] 6. During operation, the system continuously records the "actual control trajectory" and "actual glue line trajectory," generating a massive amount of high-value process data. This data is not only used for real-time calibration but also accurately records the true physical responses of different materials under specific process parameters. The long-term accumulation of data lays a solid foundation for in-depth mining of process patterns, reverse optimization of multiphysics model parameters, and even the autonomous optimization of process parameters. This effect unexpectedly makes this invention a powerful process knowledge mining platform while fulfilling the core task of precise glue application, providing valuable data assets for subsequent intelligent manufacturing upgrades.

[0025] In summary, this invention not only addresses the specific problems of insufficient accuracy, poor reliability, and frequent maintenance in existing technologies in the linear dimension, but also fosters technical effects beyond the expectations of those skilled in the art in the nonlinear dimension, such as preventative maintenance, system-level evolution, and process knowledge mining. These effects demonstrate that this invention is not a simple improvement on existing technologies, but rather a qualitative leap in system capabilities achieved through the deep integration of multiple technologies. It constitutes an organic intelligent system with self-optimization and self-adaptation capabilities, exhibiting outstanding substantive characteristics and significant progress. Attached Figure Description

[0026] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0028] See Figure 1 An adaptive control method for sole adhesive application trajectory based on dynamic scanning and feedback adjustment, the core process of which includes four main steps: S1. Multimodal perception and trajectory planning steps The multimodal perception and trajectory planning steps aim to overcome the limitations of single visual perception in complex industrial environments, which is susceptible to interference from foreign objects, by integrating three-dimensional geometric information and material spectral information, thereby achieving highly reliable identification and planning of the adhesive application trajectory.

[0029] First, a three-dimensional geometric scan is performed (S11). Using a 3D vision sensor integrated into the front end of the robot's end effector, preferably a line laser profilometer, the area of ​​the sole to be glued is continuously scanned in real time as the robot moves along a preset theoretical trajectory to acquire high-density three-dimensional point cloud data. To achieve advanced perception, the scanning direction of the line laser profilometer forms a non-zero angle with the robot's movement direction, thereby enabling advanced detection of the glue application channel.

[0030] Next, preliminary geometric feature identification is performed (S12). The three-dimensional point cloud data is preprocessed, including denoising and filtering operations. Then, feature extraction algorithms, such as curvature analysis or trained deep learning models, are used to identify all suspected trajectory regions with channel geometric features from the point cloud.

[0031] Crucially, multispectral material identification (S13) is performed. To distinguish genuine adhesive channels from contaminants with similar geometric shapes (such as residual adhesive or treatment agent residue), a multispectral sensing module is integrated on the end effector, next to the 3D vision sensor. This module can be specifically implemented as a near-infrared spectral probe, or a simplified system consisting of a multi-band LED light source emitting specific wavelengths (e.g., 850nm, 940nm, 1050nm) and a photoelectric sensor receiving reflected light. When the preliminary geometric feature identification step determines the existence of a suspected trajectory area, the multispectral sensing module is controlled to perform rapid point measurements on that specific area to obtain its spectral reflectance data in different bands, i.e., material characteristic information.

[0032] Finally, fusion decision-making and trajectory generation are performed (S14). The system has a pre-stored database of standard spectral characteristics of standard sole materials and common contaminants. The measured material characteristic information obtained in step S13 is compared with the pre-stored standard spectral characteristics of the sole body material in the database to calculate the matching degree. This step introduces a fusion decision-making logic: only when a region simultaneously meets the dual conditions of "its geometry is identified as a channel feature" and "its material characteristics match the sole body material with a degree higher than a preset threshold" is the region confirmed as a true adhesive application trajectory point. Based on all confirmed true trajectory points, the system plans and generates a real-time, highly reliable actual adhesive application trajectory, providing accurate path guidance for subsequent adaptive adhesive application.

[0033] The system determines whether the material of the measured point matches the material of the shoe sole by calculating spectral similarity. ; in, Similarity coefficient. Values ​​range from -1 to 1. A value closer to 1 indicates a higher similarity between the sample spectrum and the database spectrum. At wavelength Below, the spectral reflectance of the measured point (suspected trajectory area) is shown. This value is obtained in real time by the multispectral sensing module. At wavelength Below, the standard spectral reflectance of the shoe sole material is pre-stored in the database. The average reflectance of the measured point across n wavelengths. The average reflectance of the standard spectrum across n wavelengths in the database. The specific number of wavelengths used for comparison (e.g., if three bands are used: 850nm, 940nm, and 1050nm, then n=3). Wavelength index.

[0034] Calculated similarity coefficient It will be compared with a preset threshold. If If the value is greater than this threshold, the material is considered to be a match.

[0035] This multimodal perception mechanism elevates the judgment of a single geometric shape to the collaborative verification of geometric and material properties, fundamentally solving the problem of trajectory misidentification caused by on-site contaminants, and significantly improving the robustness and decision-making accuracy of the system.

[0036] S2, Feedback Adjustment and Sizing Execution Steps The feedback adjustment and sizing execution steps are designed to solve the endogenous interference problem caused by the sizing process itself, and to record key data for system-level accuracy calibration while executing trajectory planning.

[0037] First, based on the real-time adhesive application trajectory planned in step S1, the system generates basic motion commands for the robot. Crucially, to compensate for endogeneous interference, the system simultaneously activates a multi-physics coupling compensation model. This model is a mathematical model built on the principle of thermo-mechanical coupling, and its input parameters include real-time adhesive application process parameters, preferably including adhesive temperature, spray flow rate, and flow velocity. Based on these input parameters, the model predicts in real-time the amount of heat input from the adhesive as a heat source to the local area of ​​the sole during adhesive application, the resulting transient temperature field distribution, and the changes in the thermal stress field caused by material thermal expansion. Finally, it calculates the spatial position deviation of the measurement points on the sole surface caused by the thermal deformation; this deviation is the endogeneous interference compensation amount.

[0038] The model predicts the surface deformation of the sole caused by the heat effect of glue application (endogenous disturbance) and estimates the linear deformation under local temperature rise using a simplified thermal expansion formula.

[0039] ; in, Endogenous disturbance compensation (displacement deviation in the normal direction, unit: mm or μm). This is a key output that the model ultimately needs to calculate for trajectory compensation. Characteristic thickness of the sole in the affected area (unit: mm). This is a process parameter related to the sole model, which can be measured in advance and stored in a database. The linear expansion coefficient of the shoe sole material (unit: 1 / ℃). This is a key thermophysical parameter of the material, and the value varies for different materials (such as EVA and rubber). It is derived from the material parameter sub-model library mentioned above. Predicted temperature rise (in °C) in the local area of ​​the adhesive application point. This is the result calculated by the multiphysics model based on adhesive application parameters (adhesive temperature, flow rate, contact time) and material thermal conductivity. The formula shows that the predicted deformation is proportional to the material's coefficient of thermal expansion, local temperature rise, and material thickness. The model predicts this using real-time process parameters. Then call the material parameters and The amount of displacement that needs to be compensated can then be calculated. .

[0040] Next, a comprehensive trajectory correction is performed. The trajectory deviation calculated in step S1 based on vision and material perception (used to compensate for external disturbances) is vector-superimposed with the endogenous disturbance compensation amount calculated by the multiphysics coupling compensation model to generate a comprehensive correction command. This command drives the robot to move the glue valve, which, in addition to compensating for the initial pose error of the sole, further pre-compensates for the deformation of the working surface that will be caused by its own glue application behavior, thereby achieving extremely high-precision trajectory tracking.

[0041] In parallel, during the glue application process, the robot controller continuously records the theoretical pose data of the end effector corresponding to the control commands sent to each servo axis of the robot in the robot's base coordinate system at a predetermined high-frequency sampling frequency, forming a set of actual control trajectory data points synchronized with the actual glue application process. This actual control trajectory data is bound to a unique identifier of the currently processed shoe sole and stored in the system cache or database, providing an accurate data source for subsequent online error calibration and compensation steps.

[0042] This step, by introducing a forward-looking multiphysics compensation mechanism, elevates the control precision to a level that can offset the influence of its own process, and through detailed process data recording, lays a solid foundation for building an intelligent machining system with self-learning and self-calibration capabilities.

[0043] S3, Downstream Adhesive Line Trajectory Extraction Steps The downstream adhesive trajectory extraction step aims to provide an objective, truth data source based on the final adhesive application result for system self-calibration. This step is independent of the upstream adhesive application process, ensuring the objectivity of the evaluation benchmark.

[0044] The glued soles are transferred to a designated quality inspection station downstream of the production line. Using the industrial camera and lighting system already installed at this station, a high-resolution digital image containing the complete glue line is captured when the sole reaches the predetermined position.

[0045] After image acquisition, image processing and trajectory extraction algorithms are executed. First, the image is preprocessed, including grayscale conversion, filtering and denoising, and contrast enhancement to optimize image quality. Then, based on the contrast characteristics of the rubber line and the shoe sole background, an adaptive algorithm is used to extract the rubber line region: for high-contrast scenes, edge detection operators such as Canny are preferred to obtain the two edges of the rubber line; for scenes with moderate contrast but uniform rubber line regions, threshold segmentation algorithms (such as the Otsu method) can be used to directly distinguish rubber line pixels from background pixels. Subsequently, morphological closing operations are performed on the extracted binarized regions to connect breakpoints and smooth contours, ensuring the connectivity of the rubber line region.

[0046] Crucially, to obtain the centerline of the adhesive line, a skeleton extraction algorithm is performed on the binarized region, iteratively removing boundary pixels until a centerline skeleton with a single pixel width is obtained. Subsequently, the skeleton is sequentially tracked, recording a series of ordered two-dimensional pixel coordinates. Finally, based on pre-calibrated camera parameters, the sequence of two-dimensional pixel coordinates is transformed into the coordinate system of the quality inspection station, generating actual adhesive line trajectory data representing the true position of the adhesive line. This trajectory data is then bound to a unique identifier corresponding to the shoe sole and transmitted to the system's main controller.

[0047] This step, through independent image acquisition and processing, transforms the physical shape of the adhesive line into a precise, quantifiable digital trajectory, providing an indispensable objective basis for subsequent comparison and calibration.

[0048] S4. Online Error Calibration and Compensation Steps The online error calibration and compensation steps aim to estimate and compensate for the robot's absolute positioning error online by comparing theoretical instructions with actual results, thereby autonomously maintaining system accuracy during continuous production.

[0049] First, data association and coordinate system one are performed. Using the unique identifier on the sole, the actual control trajectory data recorded in step S2 is associated with the actual adhesive line trajectory data extracted in step S3. Then, using the coordinate transformation matrix obtained beforehand through hand-eye calibration technology, the actual adhesive line trajectory, which is in the quality inspection camera coordinate system, is transformed to the robot's basic coordinate system, allowing for direct comparison with the actual control trajectory in the same coordinate system.

[0050] Subsequently, absolute positioning error is calculated. The actual control trajectory and the actual adhesive trajectory, both within the same robot base coordinate system, are considered as two 3D point clouds. A point cloud registration algorithm, preferably an iterative nearest-point algorithm, is used to calculate the spatial transformation parameters that best match the two point clouds. The calculated transformation parameters characterize the systematic deviation of the robot's actual executed pose from the commanded pose; this deviation is defined as the robot's absolute positioning error at the corresponding working point.

[0051] Crucially, the error mapping table is constructed and updated. The robot's workspace is logically divided into a three-dimensional grid array, with each grid corresponding to a storage unit. A moving average filtering algorithm is used to continuously update and smooth the absolute positioning error data within each grid, mathematically expressed as: New error estimate = (Original error estimate × Number of historical updates + Current error measurement) / (Number of historical updates + 1) Specifically, the formula is: ; in, : The new error estimate at the 3D mesh coordinates (x, y, z). This is a six-degree-of-freedom pose deviation (ΔX, ΔY, ΔZ, ΔRx, ΔRy, ΔRz), with each component calculated independently. The old error estimate at the 3D grid coordinates (x, y, z) (i.e., the value after the last update). : The cumulative number of updates for this specific grid throughout its history. : The currently calculated absolute positioning error value. This is fresh error data calculated by comparing the "actual control trajectory" and the "actual adhesive track" near the current working point. New estimate. It is a weighted average of historical data, which can effectively filter out random noise in a single measurement. With each update... The increase, newly collected individual data The impact on the overall estimate gradually diminishes, and the mapping table tends to stabilize. However, the algorithm continues to fine-tune, thus tracking the slow changes (drift) in robot accuracy over time. Each 3D mesh is updated independently, thereby establishing a detailed "error map" across the robot's entire workspace.

[0052] In this way, an absolute positioning error mapping table is formed that is dynamically updated and whose accuracy improves as data accumulates.

[0053] Finally, feedforward compensation control is implemented. When performing adaptive gluing of the sole, after generating the comprehensive correction command in step S2, the system queries the error estimate corresponding to the three-dimensional grid of the current target point in the absolute positioning error mapping table, and uses this value as the feedforward compensation amount, which is superimposed on the final control command, thereby realizing real-time cancellation of the absolute positioning error of the robot body at the software level.

[0054] This step, by creating an autonomous loop of "execution-measurement-comparison-learning-compensation," enables the system to learn from its own operational results and optimize its performance, achieving accuracy self-calibration and long-term stability maintenance under conditions of continuous production.

[0055] See Figure 2 The implementation of the method described in this invention relies on a dedicated hardware system that works collaboratively. This system consists of four core parts: a glue-applying robot unit, a sensing and execution module, a downstream quality inspection unit, and a system controller. Through precise coordination and data interaction among these components, the adaptive control function is achieved.

[0056] The dispensing robot unit, as the main execution component of the system, consists of a six-degree-of-freedom industrial robot, a robot controller, and an end effector. The six-degree-of-freedom industrial robot provides the necessary flexible movement capabilities and high repeatability in a three-dimensional workspace. The robot controller receives high-level trajectory commands and translates them into drive signals for the servo motors of each joint. The end effector is a rigid structural component fixed to the robot's sixth-axis flange, used to integrate and precisely calibrate the relative pose relationships of each functional module.

[0057] The perception and execution module, integrated into the end effector, is key to achieving adaptive perception and operation, and includes: The 3D vision sensor, preferably a line laser profilometer, has its optical plane at an angle to the robot's direction of motion, and is used to acquire three-dimensional point cloud data of the shoe sole surface in real time during movement.

[0058] The multispectral sensing module can be implemented as a combination of a near-infrared spectral probe or a multi-band LED light source of a specific wavelength and a photoelectric sensor to collect the material spectral characteristics of the suspected trajectory area.

[0059] The glue application valve is a hot melt adhesive spray valve equipped with a temperature and flow control unit to perform precise glue spraying under control.

[0060] The downstream quality inspection unit is independently located downstream of the adhesive application station and consists of a fixed industrial area array camera and a matching lighting system. Its function is to trigger the capture of a clear image of the adhesive line when the adhesive flows through a predetermined position on the sole, providing an independent truth data source for system self-calibration.

[0061] The system controller, as the central hub for system computation and command, is typically an industrial-grade computer or a high-performance programmable logic controller (PLC). It runs integrated control software with functions encompassing: Process 3D point cloud and multispectral data, and perform multimodal fusion trajectory planning; Run the multiphysics coupling compensation algorithm to calculate the compensation amount for endogeneous disturbances; Schedule the robot's movement and glue application actions, and record the actual control trajectory; Process downstream camera images to extract the actual adhesive line trajectory; Execute the online error calibration algorithm, update the error mapping table, and implement feedforward compensation.

[0062] The hardware system operates collaboratively in the following manner: a robot unit carrying a sensing and execution module moves along the sole of the shoe, while 3D vision and multispectral sensors simultaneously acquire data; the system controller plans the trajectory based on this data and adds compensation amounts before driving the glue application valve; simultaneously, the downstream quality inspection unit acquires images of the glue line; finally, the controller completes online calibration and compensation of the system accuracy by comparing the actual control trajectory with the glue line trajectory. The entire system, through the precise coordination of the hardware units and the intelligent decision-making of the controller, forms a complete adaptive control closed loop.

[0063] Furthermore, to improve the long-term robustness and adaptability of the system in complex industrial environments, the present invention has refined and optimized the following key details: A dedicated anti-contamination device is integrated into the end effector. This device consists of a ring-shaped air nozzle and a high-speed solenoid valve, with the ring-shaped air nozzle coaxially fitted around the lens of the 3D vision sensor. The device is connected to a high-response air pressure sensor located on the adhesive application path. Its operating logic is as follows: When an adhesive line breaks during the application process, the air pressure sensor detects the unique pressure change signal within the adhesive path in real time. The system controller triggers the high-speed solenoid valve within a millisecond delay, causing it to generate a brief and powerful compressed air pulse. This pulsed airflow is discharged through the ring-shaped air nozzle and forms a transient conical air curtain directly in front of the lens, effectively dispersing the scattered adhesive strands caused by the adhesive breakage, thereby preventing contaminants from adhering to the lens surface and ensuring that the 3D vision sensor continuously obtains clear scanning data. This design achieves predictive proactive intervention against contamination sources, rather than reactive post-event cleaning.

[0064] To optimize the accuracy of the multiphysics coupling compensation model on different shoe sole materials, the system incorporates a material parameter database. This database pre-stores multiple material parameter sub-models corresponding to different shoe sole materials (such as rubber, EVA, and PU), each containing precise thermophysical parameters (such as specific heat capacity, thermal conductivity, and coefficient of thermal expansion) for that material. When the production system switches shoe types, the system controller receives shoe sole material type information from the upper-level production management system and automatically calls the matching material parameter sub-model from the database, loading it into the multiphysics coupling compensation algorithm. This ensures that the material parameters used in the thermo-mechanical coupling calculation are accurately matched with the current processing object, significantly improving the accuracy of endogeneous interference prediction and compensation, enabling the system to adapt to different production process requirements.

[0065] The above-mentioned detailed solutions, from the two dimensions of hardware protection and software self-adaptation, further enhance the stability and accuracy of the system when facing actual production disturbances and product changes, which is an important guarantee for the high engineering practical value of this invention.

[0066] To objectively evaluate the actual performance of this invention, we designed a comparative test scheme. The test aims to quantify its improvement over traditional 3D vision-based adaptive dispensing technology in key indicators such as trajectory recognition accuracy, long-term operational stability, and overall production efficiency. Comparison objects: The experimental group used the complete system of this invention (including multimodal sensing, multiphysics compensation, and online self-calibration).

[0067] The control group used a traditional adaptive sizing system (based solely on 3D visual geometric scanning and trajectory tracking, without material identification, endogenous compensation, or system self-calibration functions).

[0068] The test was conducted in a real shoe manufacturing production line environment, where unavoidable dust and a small amount of residual treatment agents from previous processes were present.

[0069] The test sample consisted of 600 EVA athletic shoe soles from the same batch. The first 100 were used for initial testing, and the remaining 500 were used for long-term stability testing. Simulated contaminants (small spots of transparent treatment agent) were intentionally placed on 20% of the soles.

[0070] The experimental and control groups used the same model of six-degree-of-freedom industrial robot and basic line laser 3D sensor to ensure that the hardware benchmarks were consistent.

[0071] The evaluation criteria include: Track offset error: The average deviation between the actual glue line center and the theoretical glue application groove center of the shoe sole is measured by sampling using a high-precision coordinate measuring machine (CMM).

[0072] Misapped adhesive rate: The percentage of shoe soles where incorrect adhesive was applied near contaminants.

[0073] System stability: Observe the change in trajectory offset error after running 500 shoe soles continuously (whether manual intervention calibration is required).

[0074] The table below summarizes the comparative data for key indicators.

[0075] The average trajectory deviation error of the experimental group was significantly lower than that of the control group. This is mainly due to the fact that the multiphysics coupling compensation model effectively counteracted the endogeneous interference caused by the sizing thermal effect, achieving higher precision control.

[0076] The experimental group had a 0% misapplied adhesive rate, while the control group had a rate as high as 18%. This fully demonstrates the necessity of multispectral material discrimination. Traditional systems cannot distinguish between genuine channels and contaminants, resulting in nearly one-fifth of the products having quality risks. This invention completely solves this problem through material verification.

[0077] After continuous operation, the control group's error drifted from 0.35 mm to 0.62 mm, a drift of +0.27 mm. This is because the robot's absolute positioning accuracy drifts with time and temperature, and traditional systems cannot sense and compensate for this change.

[0078] The experimental group's error drifted by only +0.01mm, which is almost negligible. This is entirely due to the online self-calibration mechanism. By continuously comparing the "instructions" with the "results," the system automatically updates the error mapping table, silently compensating for the robot's accuracy drift and achieving "self-maintained" high precision.

[0079] The control group experienced three downtimes and one manual calibration during testing due to quality issues and decreased accuracy, which severely impacted production cycle time and efficiency.

[0080] The experimental group achieved "zero unplanned downtime." High-quality trajectory recognition and the system's self-calibration capabilities ensured the continuity, stability, and efficiency of the production process.

[0081] This comparative test was conducted under conditions close to actual production. The data shows that: Compared to traditional adaptive application techniques, this invention represents a significant leap forward in accuracy, reliability, and overall efficiency. Its core innovations—multimodal sensing, endogenous interference compensation, and online self-calibration—have proven to be effective means of overcoming the bottlenecks of traditional technologies.

[0082] Specifically: Multimodal perception elevates trajectory recognition from "geometric uniqueness" to "geometric-material dual verification," fundamentally eliminating misidentification.

[0083] Endogenous disturbance compensation improves control precision from the "macro" level to the "microscopic physical effect" level.

[0084] Online self-calibration enables the system to move from "high precision" to "long-term high stability", significantly reducing maintenance needs and human intervention.

[0085] Therefore, this invention not only significantly improves the quality of a single application of adhesive, but more importantly, it constructs an intelligent production unit that can adapt to complex industrial environments and continuously maintain optimal performance, providing a reliable technical foundation for achieving true "intelligent manufacturing".

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive control method for shoe sole adhesive application trajectory based on dynamic scanning and feedback adjustment, characterized in that, Includes the following steps: S1. Multimodal perception and trajectory planning: S11. Using the 3D vision sensor integrated on the robot's end effector, the surface of the shoe sole is scanned in real time to obtain three-dimensional point cloud data; S12. Based on the three-dimensional point cloud data, identify suspected trajectory regions with channel geometric features; S13. Perform material analysis on the suspected trajectory region using a multispectral sensing module to obtain its material characteristic information; S14. The material feature information is compared with the pre-stored material features of the sole body. Only the area where both the geometric features and material features match the sole body is identified as the real glue application trajectory, and a real-time glue application trajectory is planned accordingly. S2. Feedback Adjustment and Adhesive Application Execution: Based on the real-time adhesive application trajectory planned in step S1, control the robot to drive the adhesive application valve to perform adhesive application operations, and record the actual control trajectory data corresponding to the adhesive application operations; S3. Downstream glue line trajectory extraction: At the quality inspection station downstream of the glue application process, obtain an image of the glued shoe sole and extract the actual glue line trajectory from the image; S4. Online error calibration and compensation: Compare the actual control trajectory with the actual glue line trajectory, calculate the robot's absolute positioning error, and perform feedforward compensation for subsequent glue application control commands for the shoe sole based on this error.

2. The adaptive control method for sole adhesive application trajectory according to claim 1, characterized in that, In step S2, when generating control commands for the real-time adhesive application trajectory, a multi-physics coupling compensation model is introduced to compensate for endogeneous interference caused by the adhesive application process itself; the multi-physics coupling compensation model at least considers the influence of the thermal effect of the adhesive on the state of the sole material.

3. The adaptive control method for sole adhesive application trajectory according to claim 2, characterized in that, The multiphysics coupling compensation model is a thermo-mechanical coupling model, which predicts the temperature field change in the adhesive application point area by input adhesive application process parameters, and then calculates the compensation amount for visual measurement error caused by material thermal deformation.

4. The adaptive control method for sole adhesive application trajectory according to claim 1, characterized in that, In step S13, the multispectral sensing module is a near-infrared spectral probe, and the material characteristic information is the reflectance in a specific near-infrared band.

5. The adaptive control method for sole adhesive application trajectory according to claim 1, characterized in that, In step S4, calculating the absolute positioning error of the robot specifically includes: converting the actual glue line trajectory to the robot coordinate system through hand-eye calibration, then performing point cloud registration with the actual control trajectory to calculate the spatial pose deviation.

6. An adaptive glue application system for implementing the adaptive control method for the glue application trajectory of the shoe sole according to any one of claims 1-5, characterized in that, include: The glue application robot unit includes a robot, an end effector, a 3D vision sensor, a multispectral sensing module, and a glue application valve; Downstream quality inspection units, including industrial cameras; The system controller is configured as follows: The 3D vision sensor and the multispectral sensing module are controlled to perform multimodal perception and plan real-time adhesive application trajectory. Record the robot's execution trajectory as the actual control trajectory; Receive the actual adhesive line trajectory sent by the downstream quality inspection unit; Execute online error calibration and compensation algorithms, and correct the robot's control commands accordingly.

7. The adaptive sizing system according to claim 6, characterized in that, The multispectral sensing module includes a multi-band LED light source and a photoelectric sensor.

8. The adaptive sizing system according to claim 6, characterized in that, The end effector also integrates a pulsed air curtain protection device, which is linked to the glue application valve. When a glue breakage signal is detected, an airflow is instantly ejected to prevent the glue filaments from getting close to and adhering to the 3D vision sensor lens.

9. The adaptive sizing system according to claim 6, characterized in that, The system controller includes a pose compensation controller, which incorporates the multiphysics coupling compensation model.

10. The adaptive sizing system according to claim 6, characterized in that, The system controller is further configured to: divide the robot's workspace into a three-dimensional grid, and based on the continuous comparison of the actual control trajectory and the actual glue line trajectory, use a moving average algorithm to update the absolute positioning error value in each grid, forming a dynamically updated error mapping table for feedforward compensation.

Images