A fragile product grabbing stress monitoring and closed-loop control system and method

Abstract

The application discloses a fragile object grabbing stress monitoring and closed-loop control system and method, relates to the field of robot grabbing and optical sensing technology, and comprises an optical emission module, an optical detection module, an artificial intelligence processing module, a control module and a fusion module; wherein the optical emission module is used for generating vortex laser beams with different topological charges; the optical detection module is used for acquiring complex amplitude distribution information corresponding to a target object; the artificial intelligence processing module outputs a stress distribution map and a microscopic deformation field based on a complex domain deep learning network; the control module generates a grabbing safety confidence map according to the stress distribution map and the microscopic deformation field; and the fusion module is used for generating a grabbing safety map and guiding a grabbing path planning. The application improves the detection accuracy and sensitivity of micro-scale stress disturbance, micro-cracks and local deformation of the system, and improves the safety, stability and intelligent level in the robot grabbing process.

Description

Technical Field

[0001] This invention relates to the field of robot grasping and optical sensing technology, and in particular to a stress monitoring and closed-loop control system and method for grasping fragile items. Background Technology

[0002] With the development of industrial automation technology, robotic gripping systems have been widely used in pharmaceutical production, precision optical component assembly, semiconductor manufacturing, and high-precision device handling. In these applications, the objects being gripped typically have characteristics such as thin walls, high brittleness, high surface precision requirements, or sensitivity to internal stress. Examples include glass ampoules, vials, and pre-filled syringes in pharmaceutical production, and precision optical components such as lenses, prisms, and filters in optical manufacturing. During automated gripping, uneven distribution of gripping force or excessive local stress can easily lead to microcracks, localized plastic deformation, or even direct breakage of the target object, thus affecting product quality and production yield.

[0003] Currently, existing robotic grasping systems typically use force sensors, pressure sensors, or tactile sensors to control the grasping force, adjusting the grasping action based on the magnitude of the contact force between the gripper and the target object. However, existing technologies or products still have the following problems in practical use: 1. Existing contact force control solutions require the sensor to be in direct contact with the target object. When grasping fragile items such as glass containers and precision optical components, they are prone to causing additional mechanical disturbance to the surface of the target object, or even causing surface damage or contamination, which makes it difficult to meet the application requirements of high precision and high cleanliness scenarios. 2. Existing force control systems can usually only detect the overall gripping force or local pressure value, making it difficult to obtain high spatial resolution stress distribution information on the surface of the target object in real time. This results in stress concentration in local areas even though the overall gripping force is within a safe range, which can lead to microcracks or local damage. 3. Most existing non-contact optical inspection technologies focus on geometric shape detection or material identification, and cannot directly reflect the real-time stress state and micro-deformation trend during the grasping process, thus making it difficult to provide early warning of potential damage risks; 4. Most existing robot vision systems and grasping control systems operate independently, lacking a fusion mechanism for stress information, deformation information, and 3D point cloud information. This makes it difficult to dynamically adjust grasping parameters based on real-time stress changes, thus failing to form a complete grasping closed-loop control.

[0004] Therefore, how to provide a technical solution that can perform non-contact, high spatial resolution stress monitoring during the grasping process of fragile items, and can dynamically adjust the robot's grasping parameters in combination with real-time stress changes, thereby realizing grasping risk prediction and closed-loop control, has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a fragile item gripping stress monitoring and closed-loop control system and method to solve the problems existing in the background art.

[0006] This invention provides the following technical solution: a fragile item gripping stress monitoring and closed-loop control system and method, comprising: An optical emission module is used to generate a vortex laser beam carrying orbital angular momentum and dynamically modulate the topological charge parameters of the vortex laser beam according to the material properties, geometric features, surface state, or grasping phase of the target object. The optical detection module is used to receive the vortex laser beam scattered or transmitted by the target object and obtain the corresponding complex amplitude distribution information. An artificial intelligence processing module is used to output the stress distribution map and micro-deformation field of the target object through a complex domain deep learning network based on the complex amplitude distribution information. The control module is used to generate a gripping safety confidence map based on the stress distribution map and the micro-deformation field, and to dynamically adjust the gripping parameters of the robot gripper when the gripping safety confidence in a local area is lower than a preset threshold; and The fusion module is used to fuse the stress distribution map, micro-deformation field and the three-dimensional point cloud information output by the robot vision servo system to generate a gripping safety.

[0007] Preferably, the optical emission module includes a spatial light modulator and a phase modulation unit; The spatial light modulator is used to generate vortex laser beams with different topological charges; The phase modulation unit is used to dynamically switch between multiple topological charges, or to synchronously generate a composite vortex optical field with multiple different topological charges. The topological charge parameter ranges from 1 to 5, where the low-order topological charge is used to detect large-scale stress changes in the target object, and the high-order topological charge is used to detect micro-scale stress disturbances, microcracks, or subsurface defects in the target object.

[0008] Preferably, the optical detection module includes an interferometric measurement unit, a high-speed image acquisition unit, and a phase unwrapping unit; The interferometric measurement unit is used to interfere with the target scattered light and the reference light; The high-speed image acquisition unit is used to acquire interference fringe images; The phase unwrapping unit is used to recover the corresponding complex amplitude distribution information based on the interference fringe image; The optical detection module includes a transmission detection mode and a reflection detection mode. The transmission detection mode is used when the target object is a light-transmitting material, and the off-axis reflection detection mode is used when the target object is a highly reflective material.

[0009] Preferably, the complex domain deep learning network includes: Complex convolutional layers are used to extract complex features from complex amplitude distributions through convolution. The topological load attention module is used to adaptively weight the feature channels corresponding to different topological loads. And a dual-mode decoder for parallel output of stress distribution maps and micro-deformation fields; The convolution kernel in the complex convolutional layer includes real and imaginary parameters trained independently to maintain the phase coupling relationship in the complex amplitude data.

[0010] Preferably, the artificial intelligence processing module trains the complex domain deep learning network based on finite element simulation data and physical optics simulation data; The training data includes stress distribution data under different gripping force conditions, corresponding micro-deformation field data, and corresponding complex amplitude variation data.

[0011] Preferably, the control module generates a local safety score based on the stress distribution map and a local stability score based on the micro deformation field. The control module generates a gripping safety confidence map based on the local safety score and the local stability score, and adjusts the clamping pressure, gripping posture, gripping point position, or gripping force distribution based on the gripping safety confidence map.

[0012] Preferably, the fusion module is used to perform pixel-level spatial alignment of the stress distribution map, micro deformation field and three-dimensional point cloud information to generate a grab safety map that simultaneously contains geometric and mechanical information; The grasping safety map is used for robot grasping path planning and grasping point selection.

[0013] Preferably, the robot gripper includes multiple independently controllable pressure zones or multiple independently driven gripping fingers; The gripping force of each pressure zone or clamping finger can be independently adjusted by the control module.

[0014] A method for monitoring and closed-loop control of gripping stress in fragile items includes the following steps: S1: Acquire the 3D point cloud information and pose information of the target object, and determine the initial grasping area; S2: Generate a vortex laser beam with corresponding topological charge based on the material properties, geometric features, or grasping stage of the target object; S3: Irradiate the target object with the vortex laser beam and collect the complex amplitude distribution information after being scattered or transmitted by the target object; S4: The complex amplitude distribution information is processed based on a complex domain deep learning network to obtain the stress distribution map and micro-deformation field of the target object; S5: Generate a grasping safety confidence map based on the stress distribution map and micro-deformation field; S6: Dynamically adjust the gripping parameters of the robot gripper based on the gripping safety confidence map; S7: The updated stress information is fused with the 3D point cloud information to generate a gripping safety map, and the gripping points or gripping paths are replanned based on the gripping safety map.

[0015] Preferably, the complex domain deep learning network simultaneously outputs a two-dimensional stress distribution map and a microscopic deformation field of the target object; When the deformation rate of the target area is detected to be continuously increasing, even if the current stress value has not reached the material failure threshold, it is still determined that there is a potential risk of damage to the target area, and grasping intervention control is performed in advance.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention generates a vortex laser beam carrying orbital angular momentum through an optical emission module and obtains the complex amplitude distribution information of the target object through an optical detection module, thereby realizing non-contact stress monitoring during the grasping process of fragile items. It effectively avoids the mechanical disturbance, surface scratches and contamination risks caused by direct contact between traditional contact force sensors and the target object, and meets the requirements for automated grasping in high-cleanliness and high-brittleness scenarios such as pharmaceutical glass containers and precision optical components. 2. This invention dynamically modulates the topological charge parameters of a vortex laser beam and processes complex amplitude information using a complex domain deep learning network. This enables real-time output of stress distribution maps and micro-deformation fields of the target object, thereby improving the system's detection accuracy for high spatial resolution stress distribution on the target object surface and its ability to identify microscale stress disturbances. Specifically, low-order topological charges improve the detection effect of overall stress changes, while high-order topological charges enhance the detection sensitivity for microcracks, subsurface defects, and localized micro-deformations, thus strengthening the system's multi-scale synchronous perception capability of stress characteristics at different spatial scales. 3. This invention achieves joint analysis of stress state and deformation trend by simultaneously outputting stress distribution map and micro deformation field. When the deformation rate of a local area is detected to be continuously increasing, even if the current stress value has not yet reached the material failure threshold, the system can still identify potential damage risks in advance and perform grasping intervention control, thereby improving the system's ability to predict and warn of crack initiation and damage risks, and thus improving the safety and stability of the robot grasping process. 4. This invention uses a fusion module to perform pixel-level fusion of stress distribution map, micro-deformation field and 3D point cloud information output by robot vision servo system, forming a grasping safety map that simultaneously contains geometric and mechanical information. Based on the grasping safety map, the position of grasping point, grasping posture and grasping force distribution are dynamically adjusted, thereby improving the accuracy of robot grasping path planning and the level of intelligence of grasping control, and realizing real-time closed-loop collaborative control capability between optical detection, risk assessment and robot grasping control. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the topological load dynamic modulation principle of the present invention.

[0019] Figure 3 This is an architectural diagram of the complex domain deep learning network of the present invention.

[0020] Figure 4 This is a schematic diagram of the process for generating a security graph according to the present invention.

[0021] Figure 5 This is a schematic diagram of an embodiment of the present invention applied to the gripping of pharmaceutical glass containers.

[0022] Figure 6 This is a schematic diagram of an embodiment of the present invention applied to the grasping of precision optical components. Detailed Implementation

[0023] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0024] This invention provides a stress monitoring and closed-loop control system and method for handling fragile items, such as... Figure 1-4As shown, this system can be applied to automated grasping scenarios for pharmaceutical glass containers, precision optical components, semiconductor wafers, thin-walled ceramic products, and other brittle or high-precision devices. It is used to perform non-contact stress monitoring on the target object during robot grasping and dynamically adjust the robot grasping parameters according to the real-time stress state to reduce the risk of local stress concentration, crack initiation, or failure of the target object.

[0025] Specifically, the system includes an optical emission module, an optical detection module, an artificial intelligence processing module, a control module, and a fusion module.

[0026] The optical emission module is used to generate a vortex laser beam carrying orbital angular momentum and dynamically modulate the topological charge parameters of the vortex laser beam according to the material properties, geometric features, surface state, or grasping phase of the target object.

[0027] It should be noted that vortex laser beams possess a helical phase wavefront structure, with a dark nucleus forming in the central region, and different topological charges corresponding to different spatial phase distribution characteristics. When a vortex laser beam irradiates the surface of a target object, minute stress changes within or on the surface of the object can cause local refractive index changes, surface micro-displacement changes, or changes in scattering paths, thereby perturbing the phase structure of the vortex light field. Because vortex lasers are highly sensitive to microscale phase changes, they can achieve highly sensitive detection of minute stress changes and the microcrack initiation process.

[0028] Furthermore, the optical emission module includes a laser, a beam expander, a spatial light modulator, and a phase modulation unit.

[0029] The laser is used to output coherent laser light with a preset wavelength; the beam expander is used to expand and shape the laser beam to improve the coverage of the subsequent phase modulation region; the spatial light modulator is used to load a spiral phase distribution to generate vortex laser beams with different topological charges; and the phase modulation unit is used to switch different topological charge parameters in real time according to control commands, or to generate multiple composite vortex light fields with different topological charges at the same time.

[0030] In some implementations, the topological load parameter can take values ​​ranging from l=1 to 5.

[0031] Among them, the low-order topological charged vortex beam has a larger main spot coverage and lower spatial frequency sensitivity, making it more suitable for detecting overall force changes and large-scale stress distribution of target objects; the high-order topological charged vortex beam has a higher spatial phase gradient and higher detection sensitivity for local small deformations, microcracks and subsurface defects, thus improving the system's ability to detect weak disturbances in the crack initiation stage.

[0032] Furthermore, during the initial contact phase with the target object, the system prioritizes the use of low-order topological loads for overall force detection to establish an initial stress distribution model for the target object; during the stabilization phase or in areas of drastic stress changes, it switches to high-order topological loads for local fine detection, thus balancing the overall detection range with local detection accuracy.

[0033] Furthermore, the optical detection module is used to receive the vortex laser beam scattered or transmitted by the target object and obtain the corresponding complex amplitude distribution information.

[0034] The optical detection module includes a beam splitting unit, a reference optical path, an interferometric measurement unit, a high-speed image acquisition unit, and a phase unwrapping unit.

[0035] The beam splitting unit is used to output a portion of the laser light as a reference light; the interferometry unit is used to interfere the target scattered light with the reference light to form interference fringes; the high-speed image acquisition unit is used to acquire interference fringe images in real time; and the phase unwrapping unit is used to recover the phase distribution and amplitude distribution of the target light field based on the interference fringes, thereby reconstructing complete complex amplitude information.

[0036] It should be noted that, compared to traditional methods that only utilize light intensity distribution for detection, this embodiment, by restoring complete complex amplitude information, can simultaneously preserve both amplitude and phase characteristics. Phase information is more sensitive to minute stress disturbances, thus effectively improving stress detection accuracy and microcrack identification capabilities.

[0037] Furthermore, the optical detection module includes both transmission detection mode and reflection detection mode.

[0038] When the target object is a glass ampoule, a transparent optical element, or other light-transmitting material, a transmission detection mode can be used to improve the detection capability of internal stress changes. When the target object is a highly reflective metal part, a coated optical device, or an opaque ceramic device, an off-axis reflection detection mode is used to reduce specular reflection interference and improve the stability of the interference signal.

[0039] Furthermore, the artificial intelligence processing module is used to output the stress distribution map and micro-deformation field of the target object through a complex domain deep learning network based on the complex amplitude distribution information.

[0040] The complex domain deep learning network includes complex convolutional layers, topological attention modules, and a bimodal decoder.

[0041] Complex convolutional layers are used to extract complex features from complex amplitude data; topological load attention modules are used to adaptively assign weights to feature channels corresponding to different topological loads; and dual-modal decoders are used to output two-dimensional stress distribution maps and micro-deformation fields, respectively.

[0042] It should be noted that, since complex amplitude data contains both real and imaginary parts, traditional real-valued convolutional networks struggle to fully preserve phase coupling relationships. Therefore, in this embodiment, the complex convolutional layers employ independently trained real and imaginary convolutional parameters to achieve joint modeling of complex domain features, thereby improving the ability to identify minute phase perturbations.

[0043] Furthermore, the artificial intelligence processing module trains the complex domain deep learning network based on finite element simulation data and physical optics simulation data.

[0044] Specifically, the stress distribution and micro-deformation inside the target object under different grasping forces can be simulated first based on the finite element model, and then the corresponding complex amplitude change characteristics can be simulated based on the physical optics propagation model, thereby establishing the mapping relationship between "stress state - light field change" to form a network training dataset.

[0045] By adopting the above method, the problem of relying solely on a large number of real-world destruction experiments for training can be avoided, thereby reducing training costs and improving the network's ability to generalize to target objects of different materials and structures.

[0046] Furthermore, the control module is used to generate a gripping safety confidence map based on the stress distribution map and the micro deformation field, and dynamically adjust the gripping parameters of the robot gripper when the gripping safety confidence in a local area is lower than a preset threshold.

[0047] Specifically, the control module generates a local safety score based on the stress distribution map and a local stability score based on the micro deformation field, and then weights and fuses the two to form a capture safety confidence map.

[0048] Among them, the local safety score is used to reflect the degree of stress risk in the current area; the local stability score is used to reflect whether there is a continuous deformation trend or crack propagation trend in the local area.

[0049] When the system detects a continuous increase in the deformation rate of a local area, even if the current stress value has not yet reached the material failure threshold, the system can still determine that there is a potential risk of damage in the corresponding area and perform grabbing intervention control in advance.

[0050] For example, the control module can perform control operations such as reducing local gripping pressure, adjusting the gripper posture, replanning the gripping point position, changing the gripping force distribution, or reducing the robot's movement speed.

[0051] By using the above methods, the problem that traditional systems can only be adjusted after damage has occurred can be effectively avoided, thereby improving the security of data capture.

[0052] Furthermore, the fusion module is used to perform pixel-level spatial alignment of the stress distribution map, micro-deformation field, and 3D point cloud information output by the robot's vision servo system to generate a gripping safety map that simultaneously contains geometric and mechanical information.

[0053] Among them, the three-dimensional point cloud information is used to characterize the spatial contour, pose and surface structure of the target object; the stress distribution map is used to characterize the local stress state; and the micro deformation field is used to characterize the local deformation trend.

[0054] After pixel-level fusion, the robot can not only identify the grasping position, but also identify which part of the grasping position is safer, thereby achieving joint optimization of geometric graspability and mechanical safety.

[0055] Furthermore, the robot gripper includes multiple independently controllable pressure zones or multiple independently driven gripping fingers, and the gripping force of each pressure zone or gripping finger can be independently adjusted by the control module.

[0056] For example, when the system detects localized stress concentration in a certain area, it can reduce the clamping force only in the corresponding clamping area without reducing the overall gripping force, thereby reducing the risk of localized damage while ensuring gripping stability.

[0057] In this embodiment, a method for stress monitoring and closed-loop control of fragile items is provided. The method can be applied to automated gripping scenarios of glass ampoules, vials, pre-filled syringes, precision optical components, semiconductor wafers, thin-walled ceramic parts, and other brittle workpieces. It is used to perform non-contact stress monitoring on the target object during robot gripping and dynamically adjust the robot gripping parameters according to the real-time stress state to reduce the risk of local stress concentration, crack initiation, or failure of the target object during gripping.

[0058] Specifically, the method includes the following steps: S1: Acquire the 3D point cloud information and pose information of the target object, and determine the initial grasping area.

[0059] Specifically, the robot vision servo system scans the target object using a binocular camera, structured light camera, or laser scanning device to acquire its 3D point cloud information. Based on this point cloud information, it performs spatial reconstruction of the target object to determine its outer contour shape, surface curvature, spatial pose, and edge regions. Furthermore, the system preferentially selects areas with minimal curvature change, high structural strength, and distance from edges as the initial grasping area to improve stability during subsequent grasping processes. It should be noted that in this embodiment, the initial grasping area is only a pre-selected area when the robot first contacts the target object; the system can dynamically adjust the grasping point position and grasping posture based on real-time stress detection results.

[0060] S2: Generate a vortex laser beam with corresponding topological charge based on the material properties, geometric features, or grasping stage of the target object.

[0061] Specifically, the system determines the material type of the target object based on its material database or visual recognition results, and automatically selects the corresponding detection mode according to different material types. For glass or transparent optical components, the transmission detection mode is preferred to improve the detection capability of internal stress changes; for metal-coated parts, opaque ceramic parts, or highly reflective workpieces, the off-axis reflection detection mode is preferred to reduce specular reflection interference and improve the stability of interference signals.

[0062] Furthermore, the control module dynamically adjusts the topological charge parameters of the vortex laser beam according to the current grasping stage of the target object. Specifically, low-order topological charges are preferentially used during the robot's approach to the target object and the initial contact stage to improve the detection range of the overall stress field. In the stable holding stage or in the local stress anomaly area, high-order topological charges are switched to improve the detection sensitivity of micro-cracks, local deformations and subsurface defects, thereby taking into account both the overall detection range and the local fine detection capability.

[0063] Furthermore, the spatial light modulator loads the corresponding spiral phase distribution according to the control command to generate a vortex laser beam with the corresponding topological charge. In some embodiments, the system can also generate multiple composite vortex light fields with different topological charges at the same time to synchronously acquire stress characteristic information at different spatial scales, thereby improving the detection accuracy and detection stability under complex working conditions.

[0064] It should be noted that vortex laser beams possess a helical phase wavefront structure, with different topological charges corresponding to different spatial phase distribution characteristics. When minute stress changes occur inside or on the surface of a target object, they can cause local refractive index changes, surface micro-displacement changes, or scattering path changes, thereby perturbing the phase structure of the vortex light field. Because vortex light fields are highly sensitive to microscale phase changes, they can achieve highly sensitive detection of minute stress changes and crack initiation processes.

[0065] S3: Irradiate the target object with a vortex laser beam and collect the complex amplitude distribution information after being scattered or transmitted by the target object.

[0066] Specifically, after the vortex laser beam irradiates the surface of the target object, changes in internal stress, local deformation, and microcracks within the target object cause changes in the amplitude and phase distribution of the scattered light. Further, the interferometry unit interferes with the target scattered light and the reference light to form an interference fringe image. The high-speed image acquisition unit acquires the interference fringes in real time and inputs them into the phase unwrapping unit. The phase unwrapping unit processes the interference fringes based on a phase retrieval algorithm to recover the amplitude and phase information corresponding to the target light field, thereby obtaining complete complex amplitude distribution information.

[0067] It should be noted that, compared with the traditional method of detection using only light intensity information, this embodiment can effectively improve the system's detection sensitivity to minute stress disturbances and crack initiation stages by simultaneously recovering amplitude and phase information. In particular, phase information is more sensitive to microscale deformation and local stress concentration, thus improving stress inversion accuracy and microcrack identification capability.

[0068] S4: Based on a deep learning network in the complex domain, the complex amplitude distribution information is processed to obtain the stress distribution map and micro-deformation field of the target object.

[0069] Specifically, the system inputs complex amplitude distribution information into a complex domain deep learning network. Complex convolutional layers perform joint convolution operations on the real and imaginary features of the complex amplitude data to preserve the phase coupling relationship. Furthermore, the topological load attention module dynamically allocates weights based on the importance of feature channels corresponding to different topological loads. When the system detects local high-frequency disturbances, it increases the feature weights of higher-order topological load channels; conversely, when the system detects overall force changes, it increases the feature weights of lower-order topological load channels, thereby enhancing adaptability to different detection scenarios.

[0070] Furthermore, the dual-modal decoder outputs a two-dimensional stress distribution map and a micro-deformation field, respectively. The two-dimensional stress distribution map is used to characterize the real-time stress state of different regions of the target object, while the micro-deformation field is used to characterize the deformation of local regions of the target object and the deformation trend. By jointly analyzing the stress state and deformation trend, it is possible to predict potential damage risks in advance.

[0071] Furthermore, the complex domain deep learning network in this embodiment is jointly trained based on finite element simulation data and physical optics simulation data. The finite element model is used to generate stress distribution data and deformation data under different gripping force conditions, and the physical optics propagation model is used to generate corresponding complex amplitude change data, thereby establishing a mapping relationship between "stress state - deformation state - light field characteristics" to improve the network's generalization ability to target objects with different materials and structures.

[0072] S5: Generate a grasping safety confidence map based on the stress distribution map and the micro deformation field.

[0073] Specifically, the control module generates a local safety score based on the stress distribution map and a local stability score based on the micro-deformation field. The local safety score and the local stability score are then fused to form a capture safety confidence map. Areas with high capture safety confidence indicate that the current area is under stable stress and is not prone to damage, while areas with low capture safety confidence indicate that the corresponding area has stress concentration, crack propagation risk, or a tendency for local instability.

[0074] Furthermore, when the system detects that the deformation rate of the target area continues to increase, even if the current stress value has not yet reached the material failure threshold, the system can still determine that there is a potential risk of failure in the corresponding area and perform grasping intervention control in advance, so as to realize the transformation from a post-failure response to a pre-failure risk prediction control method.

[0075] S6: Dynamically adjust the gripping parameters of the robot gripper based on the gripping safety confidence map.

[0076] Specifically, when the system detects that the grasping safety confidence level of a local area is lower than a preset threshold, the control module dynamically adjusts the grasping parameters of the robot gripper, such as reducing local gripping pressure, adjusting gripping posture, changing gripping force distribution, reselecting the gripping point position, or reducing the robot's movement speed, thereby reducing the risk of damage to the local area.

[0077] Furthermore, in some embodiments, the robotic gripper includes multiple independently controllable pressure zones or multiple independently driven gripping fingers, and the gripping force of each pressure zone or gripping finger can be independently adjusted by the control module. When the system detects local stress concentration in a certain area, it can reduce the gripping force of the corresponding gripping area only, without reducing the overall gripping force of the robot, thereby reducing the risk of local breakage while ensuring gripping stability.

[0078] S7: The updated stress information is fused with the 3D point cloud information to generate a gripping safety map, and the gripping points or gripping paths are replanned based on the gripping safety map.

[0079] Specifically, the fusion module performs pixel-level spatial alignment of stress distribution maps, micro-deformation fields, and 3D point cloud information to generate a gripping safety map that simultaneously contains geometric and mechanical information. The 3D point cloud information is used to characterize the spatial contour, pose, and surface structure of the target object, the stress distribution map is used to characterize the local stress state, and the micro-deformation field is used to characterize the local deformation trend. After the above information is fused, the robot can not only identify the grippable area of ​​the target object, but also identify a safer grippable area, thereby achieving joint optimization of geometric grippability and mechanical safety.

[0080] Furthermore, the robot control system re-plans the gripping point position, gripping path, gripping posture, and gripping force distribution strategy based on the gripping safety map, so as to reduce the risk of local stress concentration while ensuring gripping stability, thereby improving the safety, stability, and intelligence level of the robot during the gripping process.

[0081] Furthermore, during the robot's grasping process, the system can repeatedly execute steps S2 to S7 to update the stress state and grasping safety state of the target object in real time, thereby realizing real-time closed-loop optimization control of the robot's grasping process of fragile items.

[0082] Example 1: Grabbing of pharmaceutical glass containers like Figure 5 As shown, taking a common 2mL vial in pharmaceutical production lines as an example, this container is made of borosilicate glass with a wall thickness of about 0.8mm and a diameter of about 16mm, making it a typical thin-walled container with high brittleness. On automated filling production lines, robots need to pick up vials from the conveyor belt to the filling station. If stress control is not properly performed during the picking process, microcracks can easily occur at the transition point between the bottle shoulder and the bottom.

[0083] In this embodiment, the robot system first activates the visual servo module to acquire the 3D point cloud and precise pose of the vial using a binocular camera, initially determining the grasping area to be the middle section of the vial. Based on the object attribute database (material: borosilicate glass, wall thickness: 0.8mm, smooth surface), the system automatically configures initial detection parameters: selecting alternating transmission of topological load sequences l=1 and l=2, setting the frame rate to 50Hz, and setting the stress warning threshold to 70% of the borosilicate glass's breaking strength.

[0084] The optical emission module generates a vortex beam carrying a topological charge l=1 using a spatial light modulator. Alternatively, instead of alternating emission, it can simultaneously generate a composite vortex light field carrying multiple different topological charges to synchronously acquire multi-scale stress information. After beam expansion and collimation, the beam is projected onto the middle section of the vial. Since the glass vial is a transmissive material, the optical detection module employs a transmission measurement method, setting up a receiving optical path on the opposite side of the vial to collect the complex amplitude distribution of the transmitted beam. Before the clamp contacts the vial, the artificial intelligence processing module first performs a background measurement to establish a baseline complex amplitude distribution under zero stress.

[0085] The gripper slowly approaches the bottle. When the contact force reaches a preset initial value, the system enters real-time monitoring mode. The optical emission module switches the topological charge to l=2, at which point the phase gradient of the vortex beam is steeper, making it more sensitive to minute deformations on the bottle surface. The detection module collects the complex amplitude of the scattered light every 20 milliseconds and inputs it into a complex domain deep learning network. This network has been trained in advance through finite element simulation, and the training data covers the stress distribution and deformation field of the vial under different gripping forces and contact positions, as well as the corresponding changes in the complex amplitude of the vortex beam.

[0086] During the grasping process, the complex domain deep learning network outputs a stress distribution map of the bottle surface in real time. The system observed localized stress concentration at the transition point of the bottle shoulder, where the stress value had reached 65% of the destructive strength, approaching the warning threshold. Simultaneously, the microscopic deformation field showed a continuous deformation rate of 0.2 micrometers per second in this area, indicating that the stress was still increasing. Based on these two indicators, the control module generated a grasping safety confidence map and determined that the confidence level of the bottle shoulder area was below the safety threshold.

[0087] The control module then issues a command to adjust the pressure element in the corresponding bottle shoulder area of ​​the gripper, reducing the pressure in that area by 15%. Simultaneously, the fusion module overlays the real-time stress distribution map with the 3D point cloud to generate an updated gripping safety map. Based on this map, the system fine-tunes the gripper's posture, ensuring a more even distribution of contact force across the middle section of the bottle. After adjustment, the stress distribution map shows that the stress in the bottle shoulder area has decreased to 45% of the breaking strength, the deformation rate approaches zero, and the gripping process returns to a safe state. To achieve this refined force control, the gripper used in this invention has multiple independently controllable pressure areas or independently driveable gripping fingers, and the gripping force of each area / finger tip can be individually adjusted by the control module.

[0088] The entire gripping process lasted approximately 1.5 seconds, during which the system completed about 75 stress monitoring and updates, achieving precise closed-loop control of the gripping force. After gripping, no new microcracks were found on the surface of the vial, and the filling process proceeded normally.

[0089] Example 2: Precision Optical Component Gripping like Figure 6 As shown, taking a precision optical lens with a diameter of 25mm and a thickness of 3mm as an example, the lens surface is coated with an anti-reflective film, requiring extremely high surface quality and internal stress control. During the assembly of the optical system, the robot needs to pick up the lens from the tray and place it into the lens barrel. Any local stress concentration during the picking process may cause changes in the lens surface shape, affecting the imaging quality.

[0090] In this embodiment, before the grasping action begins, the system first performs a pre-grabbing stress scan. The optical emission module uses alternating high-order topological charges l=3 and l=4 for emission, because high-order vortex beams have higher detection sensitivity for micron-level surface defects and subsurface stresses. The detection module uses an off-axis reflective receiving method to adapt to the high reflectivity of the lens surface.

[0091] A vortex beam is incident on the lens surface at a 45-degree angle, and the detection module receives the complex amplitude of the scattered light. A complex-domain deep learning network analyzes the collected data and outputs a stress distribution map of the lens surface. The system detects a micro-stress zone at approximately the 2 o'clock position on the lens edge, with a stress value of about 30% of the destructive strength. Although this does not reach the warning threshold, analysis of the micro-deformation field reveals abnormal high-frequency disturbance characteristics in this area. The artificial intelligence processing module marks this area as a potential risk zone, assigning it a low safety confidence score.

[0092] The fusion module performs pixel-level fusion of the stress distribution map with the 3D point cloud of the lens output by the vision servo system to generate a gripping safety map. The system selects the area with the highest confidence level on this map as the gripping point, specifically approximately 5mm below the lens center and at least 10mm away from the micro-stress area. Based on this plan, the gripper grips the lens using three contact points, ensuring that the force application avoids high-risk areas.

[0093] During the grasping process, the system continued real-time monitoring with a topological load of l=3. Since the grasping point avoided the micro-stress zone, the stress distribution map showed uniform stress near the contact point, with the maximum stress value remaining below 20% of the failure strength, and no obvious anomalies in the micro-deformation field. The lens was smoothly grasped and assembled into the lens barrel.

[0094] After the data capture is complete, the system stores the captured data (including stress distribution map, topological load modulation sequence, and micro-stress zone location information) in the database for subsequent training data expansion and model iterative optimization. For high-value products such as optical components, the non-contact stress monitoring of this system effectively avoids component scrapping caused by improper capture, while providing quantitative mechanical data support for production process optimization.

[0095] Several points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection" and "linkage" should be interpreted broadly, and can be mechanical or electrical connection, or internal connection between two components, or direct connection. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationship. When the absolute position of the described object changes, the relative positional relationship may change.

[0096] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A stress monitoring and closed-loop control system for handling fragile items, characterized in that, include: An optical emission module is used to generate a vortex laser beam carrying orbital angular momentum and dynamically modulate the topological charge parameters of the vortex laser beam according to the material properties, geometric features, surface state, or grasping phase of the target object. The optical detection module is used to receive the vortex laser beam scattered or transmitted by the target object and obtain the corresponding complex amplitude distribution information. An artificial intelligence processing module is used to output the stress distribution map and micro-deformation field of the target object through a complex domain deep learning network based on the complex amplitude distribution information. The control module is used to generate a gripping safety confidence map based on the stress distribution map and the micro deformation field, and to dynamically adjust the gripping parameters of the robot gripper when the gripping safety confidence in a local area is lower than a preset threshold. as well as The fusion module is used to fuse the stress distribution map, micro-deformation field and the three-dimensional point cloud information output by the robot vision servo system to generate a gripping safety.

2. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The optical emission module includes a spatial light modulator and a phase modulation unit; The spatial light modulator is used to generate vortex laser beams with different topological charges; The phase modulation unit is used to dynamically switch between multiple topological charges, or to synchronously generate a composite vortex optical field with multiple different topological charges. The topological charge parameter has a value range of l=1~5, where the low-order topological charge is used to detect large-scale stress changes of the target object, and the high-order topological charge is used to detect micro-scale stress disturbances, microcracks or subsurface defects of the target object.

3. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The optical detection module includes an interferometric measurement unit, a high-speed image acquisition unit, and a phase unwrapping unit; The interferometric measurement unit is used to interfere with the target scattered light and the reference light; The high-speed image acquisition unit is used to acquire interference fringe images; The phase unwrapping unit is used to recover the corresponding complex amplitude distribution information based on the interference fringe image; The optical detection module includes a transmission detection mode and a reflection detection mode. The transmission detection mode is used when the target object is a light-transmitting material, and the off-axis reflection detection mode is used when the target object is a highly reflective material.

4. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The complex domain deep learning network includes: Complex convolutional layers are used to extract complex features from complex amplitude distributions through convolution. The topological load attention module is used to adaptively weight the feature channels corresponding to different topological loads. And a dual-mode decoder for parallel output of stress distribution maps and micro-deformation fields; The convolution kernel in the complex convolutional layer includes real and imaginary parameters trained independently to maintain the phase coupling relationship in the complex amplitude data.

5. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The artificial intelligence processing module trains the complex domain deep learning network based on finite element simulation data and physical optics simulation data; The training data includes stress distribution data under different gripping force conditions, corresponding micro-deformation field data, and corresponding complex amplitude variation data.

6. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The control module generates a local safety score based on the stress distribution map and a local stability score based on the micro deformation field. The control module generates a gripping safety confidence map based on the local safety score and the local stability score, and adjusts the clamping pressure, gripping posture, gripping point position, or gripping force distribution based on the gripping safety confidence map.

7. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The fusion module is used to perform pixel-level spatial alignment of the stress distribution map, micro deformation field and three-dimensional point cloud information to generate a grab safety map that simultaneously contains geometric and mechanical information. The grasping safety map is used for robot grasping path planning and grasping point selection.

8. The fragile item gripping stress monitoring and closed-loop control system according to claim 1, characterized in that: The robot gripper includes multiple independently controllable pressure zones or multiple independently driven gripping fingers; The gripping force of each pressure zone or clamping finger can be independently adjusted by the control module.

9. A method for monitoring and closed-loop control of gripping stress in fragile items, based on the stress monitoring and closed-loop control system for gripping fragile items as described in any one of claims 1-8, characterized in that: Includes the following steps: S1: Acquire the 3D point cloud information and pose information of the target object, and determine the initial grasping area; S2: Generate a vortex laser beam with corresponding topological charge based on the material properties, geometric features, or grasping stage of the target object; S3: Irradiate the target object with the vortex laser beam and collect the complex amplitude distribution information after being scattered or transmitted by the target object; S4: The complex amplitude distribution information is processed based on a complex domain deep learning network to obtain the stress distribution map and micro-deformation field of the target object; S5: Generate a grasping safety confidence map based on the stress distribution map and micro-deformation field; S6: Dynamically adjust the gripping parameters of the robot gripper based on the gripping safety confidence map; S7: The updated stress information is fused with the 3D point cloud information to generate a gripping safety map, and the gripping points or gripping paths are replanned based on the gripping safety map.

10. The method for monitoring and closed-loop control of gripping stress in fragile items according to claim 9, characterized in that: The complex domain deep learning network simultaneously outputs a two-dimensional stress distribution map and a microscopic deformation field of the target object. When the deformation rate of the target area is detected to be continuously increasing, even if the current stress value has not reached the material failure threshold, it is still determined that there is a potential risk of damage to the target area, and grasping intervention control is performed in advance.

Images