High-precision visual alignment and bonding method

Abstract

This invention discloses a high-precision visual alignment and bonding method, comprising the following steps: S1, scanning a flexible workpiece using a TOF camera or structured light sensor to obtain three-dimensional point cloud data of the surface of the flexible workpiece; relating to the field of image recognition technology, this application uses a TOF camera or structured light sensor to scan a flexible workpiece to obtain three-dimensional point cloud data of the surface of the flexible workpiece. After noise reduction, smoothing, and feature extraction, a hyperelastic constitutive model of the flexible workpiece is constructed to generate three-dimensional deformation compensation parameters. Based on the three-dimensional deformation compensation parameters, the robotic arm is controlled to perform alignment adjustment to complete the bonding. This solves the problem that existing 2D vision cannot handle the Z-axis warping, stretching deformation, and angular deviation of flexible workpieces, reduces the error of flexible workpieces when bonding curved surfaces, improves the accuracy of visual alignment and bonding, reduces the occurrence of hollow areas in the bonding of multi-layer flexible workpieces, and thus improves the yield rate of bonding.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a high-precision visual alignment and bonding method. Background Technology

[0002] Visual alignment involves acquiring images of the objects to be bonded using a camera, extracting features from the images, and matching them with a target reference image or model to calculate the positional deviation. The goal is to control the deviation between the position of the object to be bonded, such as a chip, panel, or thin film, and the reference position within a minimal range, ensuring that the bonding accuracy meets process requirements. In the bonding process of flexible workpieces, such as flexible screens and wearable device sensor films, the workpiece is prone to three-dimensional deformations such as Z-axis warping and stretching due to factors like clamping force and temperature stress. Traditional 2D visual alignment technology can only perform positional correction in the XY plane and cannot achieve three-dimensional deformation compensation.

[0003] CN115200480A discloses a visual inspection system for alignment and bonding, comprising: a control terminal, which is the main control terminal of the system and is used to issue execution commands; a transmission module, which is used to transmit the bonding target to the bonding station on the bonding device; a robotic arm module, which is used to grasp the bonding target transmitted by the transmission module to the bonding station on the bonding device; and a second camera module, which is used to collect the illumination and shadow image data of the bonding target. This invention, through module deployment, performs two alignment processes on the target to be aligned and bonded during system operation, thereby effectively improving the accuracy of alignment and bonding. Furthermore, during the second alignment of the target, the multi-directional shadow of the target is used to determine the alignment and bonding position, thereby further improving the accuracy of the aligned and bonded target after alignment correction processing.

[0004] However, the above application still has the following problems: The above application relies on 2D cameras to collect contour images and improves accuracy through two plane corrections, but it can only obtain the XY plane features of the workpiece and cannot perceive the Z-axis depth change or three-dimensional deformation. In the visual alignment and bonding process of flexible workpieces, the inability to handle Z-axis deformation can easily lead to increased bonding errors in the edge area. For example, in the flip-chip bonding process of flexible circuit boards and chips, if the Z-axis warping of the FPC caused by thermal stress is not compensated, it will lead to poor contact between the solder balls and the pads, reducing the yield of flexible workpiece bonding. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention proposes a high-precision visual alignment and bonding method.

[0006] The present invention proposes a high-precision visual alignment and bonding method, comprising the following steps:

[0007] S1. Use a TOF camera or structured light sensor to scan the flexible workpiece and obtain three-dimensional point cloud data of the surface of the flexible workpiece.

[0008] A 3D point cloud refers to a coordinate dataset composed of discrete points in 3D space, where each point contains (x, y, z) spatial coordinates and additional attributes.

[0009] A Time-of-Flight (TOF) camera is an existing 3D imaging device based on the principle of time-of-flight. It calculates the distance between the target object and the camera by measuring the time difference or phase difference between the emission and reception of a light signal, thereby acquiring 3D spatial information of the scene. Its core function is to generate a depth image of the target object in real time, overcoming the limitation of traditional 2D cameras that can only acquire planar pixel information.

[0010] S2. Perform noise reduction and smoothing processing on the three-dimensional point cloud data, extract the three-dimensional contour features of the flexible workpiece, and generate deformation distribution data.

[0011] S3. Construct a hyperelastic constitutive model of the flexible workpiece. Based on the deformation distribution data, generate three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle, and roll compensation angle.

[0012] S4. Based on the three-dimensional deformation compensation parameters, control the robotic arm to perform alignment adjustments and complete the bonding.

[0013] Preferably, in S1, when scanning the flexible workpiece using a TOF camera or a structured light sensor, an initial image of the flexible workpiece is acquired, and the initial image of the flexible workpiece is semantically segmented using a lightweight convolutional neural network to identify the planar and curved regions of the flexible workpiece.

[0014] Preferably, S1 further includes: performing layered scanning of the curved surface region of the flexible workpiece using a combination of binocular vision and laser line scanning to obtain three-dimensional point cloud data of the curved surface region of the flexible workpiece;

[0015] The goal is to acquire more 3D point cloud data to compensate for the lack of detail in TOF cameras or structured light sensors.

[0016] Preferably, in step S2, the three-dimensional point cloud data is subjected to noise reduction and smoothing processing to extract the three-dimensional contour features of the flexible workpiece and generate deformation distribution data, as follows:

[0017] Voxel mesh filtering is used to downsample the 3D point cloud data of the surface of the flexible workpiece and the 3D point cloud data of the curved area of ​​the flexible workpiece to generate noise-reduced 3D point cloud data.

[0018] Outliers in the denoised 3D point cloud data are removed using the RANSAC algorithm to generate smoothed 3D point cloud data.

[0019] The RANSAC algorithm is an existing robust algorithm for estimating mathematical model parameters through iterative random sampling, primarily used to handle data fitting problems with a large number of noisy points. Its core idea is to use a probabilistic sampling strategy to select valid points from the observed data that fit the model, eliminating outlier interference and thus obtaining reliable model parameters.

[0020] Based on the smoothed 3D point cloud data, the edge contour points and warping feature points of the flexible workpiece are extracted to generate deformation distribution data.

[0021] In S2, noise reduction and smoothing are performed on the three-dimensional point cloud data, which includes the three-dimensional point cloud data of the surface of the flexible workpiece and the three-dimensional point cloud data of the curved area of ​​the flexible workpiece.

[0022] Preferably, in S2, based on the smoothed 3D point cloud data, the edge contour points and warping feature points of the flexible workpiece are extracted, as follows:

[0023] Based on the smoothed 3D point cloud data, a deep learning algorithm is used to extract the edge contour points of the flexible workpiece.

[0024] Based on the smoothed 3D point cloud data, machine learning algorithms are used to extract the warping feature points of the flexible workpiece.

[0025] Deep learning algorithms are existing technologies that simulate the hierarchical structure of neurons in the human brain to construct multi-layered nonlinear computational models, enabling automatic feature extraction and pattern recognition from complex data. Their core lies in utilizing the deep structure of multi-layered neural networks to abstract high-order semantic features layer by layer from raw data, overcoming the limitations of manually designed features in traditional machine learning.

[0026] Machine learning algorithms are existing algorithmic frameworks that enable computer systems to improve task performance without explicit programming by automatically learning patterns, rules, or decision rules from data. The core idea is to allow machines to extract knowledge from observed data and generalize that knowledge to unknown scenarios, solving complex problems that are difficult to handle with traditional programming.

[0027] Preferably, in step S3, a hyperelastic constitutive model of the flexible workpiece is constructed, and three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process are generated based on the deformation distribution data, as follows:

[0028] A hyperelastic constitutive model of a flexible workpiece is established through finite element analysis. A geometric mesh model of the flexible workpiece is constructed based on a preset hyperelastic constitutive model type. The material parameters and boundary conditions of the flexible workpiece are obtained as inputs. The deformation field distribution function of the flexible workpiece under different clamping forces is calculated through a finite element solver.

[0029] Finite element analysis (FEM) is an existing numerical computation method that discretizes a continuum into a finite number of elements and solves engineering and physics problems based on variational principles. Its core lies in transforming complex problems in mechanics, thermodynamics, and electromagnetism into systems of algebraic equations, which are then solved efficiently by computers. It is widely used in engineering design, simulation, and optimization.

[0030] The hyperelastic constitutive model is an existing nonlinear elastic theoretical model that describes the stress-strain relationship of materials under large deformation.

[0031] The clamping force data of the robotic arm when grasping the flexible workpiece is obtained, and combined with the deformation distribution data, it is input into the hyperelastic constitutive model to solve for the three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle, and roll compensation angle.

[0032] Preferably, when acquiring the clamping force data when the robotic arm grasps the flexible workpiece, a six-dimensional force sensor is used to collect the clamping force data when the robotic arm grasps the flexible workpiece, and the six-dimensional force sensor should be installed on the end effector of the robotic arm.

[0033] Preferably, S3 further includes:

[0034] Acquire historical flexible workpiece bonding data, which includes three-dimensional deformation compensation parameters, actual bonding error data, and working condition parameters;

[0035] Based on historical flexible workpiece bonding data, a spatiotemporal sequence model is constructed, which is used to predict bonding errors.

[0036] Obtain the current three-dimensional deformation compensation parameters and working condition parameters, input the current three-dimensional deformation compensation parameters and working condition parameters into the spatiotemporal sequence model, and the spatiotemporal sequence model outputs the fitting error prediction data;

[0037] Based on the current 3D deformation compensation parameters and bonding error prediction data, composite compensation parameters are generated. These composite compensation parameters are used as the final 3D deformation compensation parameters to guide the alignment adjustment of the robotic arm or bonding platform before bonding.

[0038] The high-precision visual alignment and bonding method proposed in this invention has the following beneficial technical effects:

[0039] 1. This application uses a TOF camera or structured light sensor to scan a flexible workpiece, obtain three-dimensional point cloud data of the workpiece surface, and after noise reduction, smoothing, and feature extraction, constructs a hyperelastic constitutive model of the flexible workpiece to generate three-dimensional deformation compensation parameters. Based on the three-dimensional deformation compensation parameters, the robotic arm is controlled to perform alignment adjustment to complete the bonding. This solves the problem that existing 2D vision cannot handle the Z-axis warping, stretching deformation, and angular deviation of flexible workpieces, reduces the error of flexible workpiece bonding on curved surfaces, improves the accuracy of visual alignment bonding, reduces the occurrence of hollow areas in the bonding of multi-layer flexible workpieces, and thus improves the bonding yield.

[0040] 2. A lightweight convolutional neural network is used to perform semantic segmentation on the initial image of the flexible workpiece, identifying planar and curved regions. This avoids indiscriminate scanning of the entire image, reduces data redundancy, and improves point cloud acquisition efficiency. By combining binocular vision with laser line scanning, the curved regions of the flexible workpiece are scanned in layers, effectively compensating for the blind spots of a single TOF camera or structured light sensor in areas with complex curvature changes, and improving the accuracy of contour restoration of the curved regions of the flexible workpiece.

[0041] 3. By establishing a hyperelastic constitutive model of the flexible workpiece through finite element analysis, three-dimensional deformation compensation parameters are generated. Compared with the traditional 2D scheme that only compensates for translation, it can correct pitch, yaw, and roll angle deviations caused by uneven clamping force, reduce fitting angle errors, and improve fitting accuracy.

[0042] 4. By acquiring historical flexible workpiece bonding data containing three-dimensional deformation compensation parameters, actual bonding error data, and working condition parameters, a spatiotemporal sequence model is constructed. Based on the current parameters, the bonding error can be predicted and composite compensation parameters can be generated. The X-axis compensation displacement, Y-axis compensation displacement, and Z-axis compensation displacement can be calibrated to further improve the bonding accuracy. Dynamic calibration of the three-dimensional deformation compensation parameters can be achieved, solving the problem that traditional static compensation cannot adapt to changes in working conditions, and further reducing the bonding displacement error.

[0043] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0044] Figure 1 This is a flowchart of a high-precision visual alignment and bonding method according to the present invention. Detailed Implementation

[0045] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0046] like Figure 1 The high-precision visual alignment and bonding method shown includes the following steps:

[0047] S1. Use a TOF camera or structured light sensor to scan the flexible workpiece and obtain three-dimensional point cloud data of the surface of the flexible workpiece.

[0048] A 3D point cloud refers to a coordinate dataset composed of discrete points in 3D space, where each point contains (x, y, z) spatial coordinates and additional attributes.

[0049] A Time-of-Flight (TOF) camera is an existing 3D imaging device based on the principle of time-of-flight. It calculates the distance between the target object and the camera by measuring the time difference or phase difference between the emission and reception of a light signal, thereby acquiring 3D spatial information of the scene. Its core function is to generate a depth image of the target object in real time, overcoming the limitation of traditional 2D cameras that can only acquire planar pixel information.

[0050] In an optional embodiment, in S1, when scanning the flexible workpiece using a TOF camera or a structured light sensor, an initial image of the flexible workpiece is acquired, and the initial image of the flexible workpiece is semantically segmented using a lightweight convolutional neural network to identify the planar region and curved region of the flexible workpiece.

[0051] In an optional embodiment, S1 further includes: performing layered scanning of the curved surface region of the flexible workpiece by combining binocular vision and laser line scanning to obtain three-dimensional point cloud data of the curved surface region of the flexible workpiece.

[0052] The goal is to acquire more 3D point cloud data to compensate for the lack of detail in TOF cameras or structured light sensors;

[0053] A lightweight convolutional neural network is used to perform semantic segmentation on the initial image of the flexible workpiece, identifying planar and curved areas. This avoids indiscriminate scanning of the entire image, reduces data redundancy, and improves point cloud acquisition efficiency. By combining binocular vision with laser line scanning, the curved areas of the flexible workpiece are scanned in layers, effectively compensating for the blind spots of a single TOF camera or structured light sensor in areas with complex curvature changes, and improving the accuracy of contour restoration of the curved areas of the flexible workpiece.

[0054] S2. Perform noise reduction and smoothing processing on the three-dimensional point cloud data, extract the three-dimensional contour features of the flexible workpiece, and generate deformation distribution data.

[0055] In an optional embodiment, in S2, the three-dimensional point cloud data is subjected to noise reduction and smoothing processing to extract the three-dimensional contour features of the flexible workpiece and generate deformation distribution data, as follows:

[0056] Voxel mesh filtering is used to downsample the 3D point cloud data of the surface of the flexible workpiece and the 3D point cloud data of the curved area of ​​the flexible workpiece to generate noise-reduced 3D point cloud data.

[0057] Outliers in the denoised 3D point cloud data are removed using the RANSAC algorithm to generate smoothed 3D point cloud data.

[0058] The RANSAC algorithm is an existing robust algorithm for estimating mathematical model parameters through iterative random sampling, primarily used to handle data fitting problems with a large number of noisy points. Its core idea is to use a probabilistic sampling strategy to select valid points from the observed data that fit the model, eliminating outlier interference and thus obtaining reliable model parameters.

[0059] Based on the smoothed 3D point cloud data, the edge contour points and warping feature points of the flexible workpiece are extracted to generate deformation distribution data.

[0060] In S2, noise reduction and smoothing are performed on the three-dimensional point cloud data, which includes the three-dimensional point cloud data of the surface of the flexible workpiece and the three-dimensional point cloud data of the curved area of ​​the flexible workpiece.

[0061] In an optional embodiment, in S2, the edge contour points and warping feature points of the flexible workpiece are extracted based on the smoothed 3D point cloud data, as follows:

[0062] Based on the smoothed 3D point cloud data, a deep learning algorithm is used to extract the edge contour points of the flexible workpiece.

[0063] Based on the smoothed 3D point cloud data, machine learning algorithms are used to extract the warping feature points of the flexible workpiece.

[0064] Deep learning algorithms are existing technologies that simulate the hierarchical structure of neurons in the human brain to construct multi-layered nonlinear computational models, enabling automatic feature extraction and pattern recognition from complex data. Their core lies in utilizing the deep structure of multi-layered neural networks to abstract high-order semantic features layer by layer from raw data, overcoming the limitations of manually designed features in traditional machine learning.

[0065] Machine learning algorithms are existing algorithmic frameworks that enable computer systems to improve task performance without explicit programming by automatically learning patterns, rules, or decision rules from data. The core idea is to allow machines to extract knowledge from observed data and generalize that knowledge to unknown scenarios, solving complex problems that are difficult to handle with traditional programming.

[0066] S3. Construct a hyperelastic constitutive model of the flexible workpiece. Based on the deformation distribution data, generate three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle, and roll compensation angle.

[0067] By establishing a hyperelastic constitutive model of the flexible workpiece through finite element analysis, three-dimensional deformation compensation parameters are generated. Compared with the traditional 2D scheme that only compensates for translation, it can correct pitch, yaw, and roll angle deviations caused by uneven clamping force, reduce fitting angle errors, and improve fitting accuracy.

[0068] In an optional embodiment, in S3, a hyperelastic constitutive model of the flexible workpiece is constructed, and three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process are generated based on the deformation distribution data, as follows:

[0069] A hyperelastic constitutive model of the flexible workpiece is established through finite element analysis. A geometric mesh model of the flexible workpiece is constructed based on the preset hyperelastic constitutive model type. The hyperelastic constitutive model is an existing nonlinear elastic theory model that describes the stress-strain relationship of materials under large deformation. The material parameters and boundary conditions of the flexible workpiece are obtained as inputs, and the deformation field distribution function of the flexible workpiece under different clamping forces is calculated through the finite element solver.

[0070] Finite element analysis (FEM) is an existing numerical computation method that discretizes a continuum into a finite number of elements and solves engineering and physics problems based on variational principles. Its core lies in transforming complex problems in mechanics, thermodynamics, and electromagnetism into systems of algebraic equations, which are then solved efficiently by computers. It is widely used in engineering design, simulation, and optimization.

[0071] The clamping force data of the robotic arm when grasping the flexible workpiece is obtained, and combined with the deformation distribution data, it is input into the hyperelastic constitutive model to solve the three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle and roll compensation angle.

[0072] In an optional embodiment, when acquiring the clamping force data when the robotic arm grasps the flexible workpiece, a six-dimensional force sensor is used to collect the clamping force data when the robotic arm grasps the flexible workpiece. The six-dimensional force sensor should be installed on the end effector of the robotic arm.

[0073] In an optional embodiment, S3 further includes:

[0074] Acquire historical flexible workpiece bonding data, which includes three-dimensional deformation compensation parameters, actual bonding error data, and working condition parameters;

[0075] Based on historical flexible workpiece bonding data, a spatiotemporal sequence model is constructed, which is used to predict bonding errors.

[0076] Obtain the current three-dimensional deformation compensation parameters and working condition parameters, input the current three-dimensional deformation compensation parameters and working condition parameters into the spatiotemporal sequence model, and the spatiotemporal sequence model outputs the fitting error prediction data;

[0077] Based on the current three-dimensional deformation compensation parameters and bonding error prediction data, composite compensation parameters are generated. These composite compensation parameters are used as the final three-dimensional deformation compensation parameters to guide the alignment adjustment of the robotic arm or bonding platform before bonding.

[0078] This solution needs to be implemented after using the technical solution of this application for a certain period of time, because without using this solution, it is impossible to obtain the three-dimensional deformation compensation parameters from historical data; this solution is a secondary calibration of the technical solution of this application, and the calibration targets the X-axis compensation displacement, Y-axis compensation displacement, and Z-axis compensation displacement to further improve the fitting accuracy;

[0079] In an optional embodiment, a spatiotemporal sequence model is constructed based on historical flexible workpiece bonding data, as follows:

[0080] Let the three-dimensional deformation compensation parameters be... ,in, This represents the X-axis compensation displacement. This is the Y-axis compensation displacement. This is the Z-axis compensation displacement. For pitch compensation angle, For yaw compensation angle, For roll compensation angle;

[0081] Let the operating parameters be... ,in, For clamping force, For ambient temperature, For workpiece thickness, The elastic modulus of the material;

[0082] Let the bonding error prediction data be ,in, This is the predicted X-axis fitting error vector. This is the predicted Y-axis fitting error vector. This is the predicted Z-axis fitting error vector;

[0083] The formula for calculating E is: ;

[0084] In the formula, , , , These are the weighting coefficients. , , , The value is obtained using the least squares method;

[0085] When generating composite compensation parameters, let the composite compensation parameters be: ;

[0086] Obtain current 3D deformation compensation parameters and operating parameters Fitting error prediction data are obtained through the formula for calculating E. ;

[0087] Calculate current operating parameters using multidimensional Euclidean distance Similarity S with historical operating parameters;

[0088] Get the current 3D deformation compensation parameters Formation of X-axis, Y-axis, and Z-axis compensation displacements ; Retrieve current 3D deformation compensation parameters Formation of pitch compensation angle, yaw compensation angle, and roll compensation angle ;

[0089] Let the corrected displacement be ,but:

[0090] but In the formula, ;

[0091] ;

[0092] By acquiring historical flexible workpiece bonding data containing three-dimensional deformation compensation parameters, actual bonding error data, and working condition parameters, a spatiotemporal sequence model is constructed. Based on the current parameters, the bonding error can be predicted and composite compensation parameters can be generated. The X-axis compensation displacement, Y-axis compensation displacement, and Z-axis compensation displacement can be calibrated to further improve the bonding accuracy. This enables dynamic calibration of the three-dimensional deformation compensation parameters, solves the problem that traditional static compensation cannot adapt to changes in working conditions, and further reduces the bonding displacement error.

[0093] S4. Based on the three-dimensional deformation compensation parameters, control the robotic arm to perform alignment adjustments and complete the bonding.

[0094] This application uses a TOF camera or structured light sensor to scan a flexible workpiece, acquire three-dimensional point cloud data of the workpiece surface, and after noise reduction, smoothing, and feature extraction, constructs a hyperelastic constitutive model of the flexible workpiece to generate three-dimensional deformation compensation parameters. Based on the three-dimensional deformation compensation parameters, the robotic arm is controlled to perform alignment adjustment to complete the bonding. This solves the problem that existing 2D vision cannot handle the Z-axis warping, stretching deformation, and angular deviation of flexible workpieces, reduces the error of flexible workpiece bonding on curved surfaces, improves the accuracy of visual alignment bonding, reduces the occurrence of hollow areas in the bonding of multi-layer flexible workpieces, and thus improves the bonding yield.

[0095] It can be applied to high-precision bonding scenarios for curved surfaces such as automotive center console screens.

[0096] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0097] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0098] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0100] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0101] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-precision visual alignment and bonding method, characterized in that, Includes the following steps: S1. Use a TOF camera or structured light sensor to scan the flexible workpiece and obtain three-dimensional point cloud data of the surface of the flexible workpiece. S2. Perform noise reduction and smoothing processing on the three-dimensional point cloud data, extract the three-dimensional contour features of the flexible workpiece, and generate deformation distribution data. S3. Construct a hyperelastic constitutive model of the flexible workpiece. Based on the deformation distribution data, generate three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle, and roll compensation angle. S4. Based on the three-dimensional deformation compensation parameters, control the robotic arm to perform alignment adjustments and complete the bonding; S3 also includes: Acquire historical flexible workpiece bonding data, which includes three-dimensional deformation compensation parameters, actual bonding error data, and working condition parameters; Based on historical flexible workpiece bonding data, a spatiotemporal sequence model is constructed, which is used to predict bonding errors. Obtain the current three-dimensional deformation compensation parameters and working condition parameters, input the current three-dimensional deformation compensation parameters and working condition parameters into the spatiotemporal sequence model, and the spatiotemporal sequence model outputs the fitting error prediction data; Based on the current 3D deformation compensation parameters and fitting error prediction data, composite compensation parameters are generated, and these composite compensation parameters are used as the final 3D deformation compensation parameters.

2. The high-precision visual alignment and bonding method according to claim 1, characterized in that, In S1, when scanning the flexible workpiece using a TOF camera or structured light sensor, an initial image of the flexible workpiece is acquired. A lightweight convolutional neural network is then used to perform semantic segmentation on the initial image of the flexible workpiece to identify the planar and curved regions of the flexible workpiece.

3. The high-precision visual alignment and bonding method according to claim 2, characterized in that, S1 also includes: using a combination of binocular vision and laser line scanning to perform layered scanning of the curved surface area of ​​the flexible workpiece, thereby obtaining three-dimensional point cloud data of the curved surface area of ​​the flexible workpiece.

4. The high-precision visual alignment and bonding method according to claim 1 or 3, characterized in that, In step S2, the 3D point cloud data is denoised and smoothed to extract the 3D contour features of the flexible workpiece and generate deformation distribution data, as follows: Voxel mesh filtering is used to downsample the 3D point cloud data of the surface of the flexible workpiece and the 3D point cloud data of the curved area of ​​the flexible workpiece to generate noise-reduced 3D point cloud data. Outliers in the denoised 3D point cloud data are removed using the RANSAC algorithm to generate smoothed 3D point cloud data. Based on the smoothed 3D point cloud data, the edge contour points and warping feature points of the flexible workpiece are extracted to generate deformation distribution data.

5. The high-precision visual alignment and bonding method according to claim 4, characterized in that, In S2, based on the smoothed 3D point cloud data, the edge contour points and warping feature points of the flexible workpiece are extracted, as follows: Based on the smoothed 3D point cloud data, a deep learning algorithm is used to extract the edge contour points of the flexible workpiece. Based on the smoothed 3D point cloud data, machine learning algorithms are used to extract the warping feature points of the flexible workpiece.

6. The high-precision visual alignment and bonding method according to claim 1, characterized in that, In S3, a hyperelastic constitutive model of the flexible workpiece is constructed. Based on the deformation distribution data, three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process are generated, as follows: A hyperelastic constitutive model of a flexible workpiece is established through finite element analysis. A geometric mesh model of the flexible workpiece is constructed based on a preset hyperelastic constitutive model type. The material parameters and boundary conditions of the flexible workpiece are obtained as inputs. The deformation field distribution function of the flexible workpiece under different clamping forces is calculated through a finite element solver. The clamping force data of the robotic arm when grasping the flexible workpiece is obtained, and combined with the deformation distribution data, it is input into the hyperelastic constitutive model to solve for the three-dimensional deformation compensation parameters of the flexible workpiece during the bonding process. The three-dimensional deformation compensation parameters include X-axis compensation displacement, Y-axis compensation displacement, Z-axis compensation displacement, pitch compensation angle, yaw compensation angle, and roll compensation angle.

7. The high-precision visual alignment and bonding method according to claim 6, characterized in that, When acquiring the clamping force data of the robotic arm when grasping a flexible workpiece, a six-dimensional force sensor is used to collect the clamping force data of the robotic arm when grasping the flexible workpiece. The six-dimensional force sensor should be installed on the end effector of the robotic arm.

Images