Metal vacuum cup product welding system based on visual monitoring
By extracting candidate contour feature points of the steps of the metal thermos cup base, calculating the reflection drift and return source residual, and generating the true weld occupancy and focus correction, the problem of false contour misleading caused by strong reflective materials and geometric interference in the visual monitoring system is solved, and the stability and accuracy of laser welding are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUYI HONGGUAN IND & TRADE CO LTD
- Filing Date
- 2026-05-19
- Publication Date
- 2026-07-14
AI Technical Summary
In the existing automated production of metal thermos cups, the visual monitoring system is interfered with by highly reflective materials and specific geometric structures at the base steps. This causes the continuous virtual weld contour to mislead the joint tracking algorithm, resulting in the laser focus deviating from the real joint trajectory and causing welding quality defects such as incomplete fusion, false welding, or sealing failure.
Candidate contour feature points are acquired through the contour acquisition module. The reflection drift is calculated by combining the radial normal of the preset step wall. The high reflection source area is confirmed by the return source module, and the actual weld occupancy is generated. The candidate contour feature points are fused into the actual circumferential weld bone line points by the bone line generation module. Finally, the laser focus correction amount is calculated by the focus correction module to achieve accurate positioning of the laser focus.
It effectively identifies and separates real and fake targets in complex optical environments, improves the system's recognition accuracy under extreme reflection conditions, ensures the stability and welding quality of the laser welding process, and reduces the equipment's sensitivity to ambient light and workpiece surface conditions in the production environment.
Smart Images
Figure CN122391186A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of laser welding automation and machine vision monitoring, specifically to a welding system for metal thermos cup products based on vision monitoring. Background Technology
[0002] Metal insulated cups are assembled from multiple thin-walled metal components, including an inner liner, outer shell, and base. At the bottom of the cup, the annular seam formed by the base step and the edge of the cup is crucial for ensuring the container's vacuum level and overall strength. Current automated production processes for metal insulated cups typically involve laser seam tracking using a vision monitoring system. Due to the extremely high surface reflectivity of stainless steel or titanium alloys, and the fact that the area to be welded is located in a complex transitional zone consisting of the base plane, the vertical wall, and rounded corners, the latent heat capacity is significantly affected by variations in material thickness. Therefore, the vision monitoring component needs to capture the seam position in real time and guide the laser focus to achieve sub-millimeter-level trajectory compensation in three-dimensional space during the continuous rotation of the cup by the rotating mechanism, thereby forming a dense and reliable fusion.
[0003] During the monitoring of the circumferential weld seam of the cup base step, the visual imaging environment exhibits strong mirror reflection properties due to the local spatial structure formed by the step wall and the base plane. When the measuring laser line is projected onto the base plane, it easily generates secondary reflection at the vertical step wall, forming a false contour within the imaging sensor that is highly similar to the actual joint geometry. Because this false contour exhibits a continuous topological structure and undergoes stable following motion with rotational displacement, the interference signal is highly deceptive in both morphology and dynamic behavior. Existing technical solutions mostly treat the reflection interference as randomly distributed isolated points and attempt to eliminate it through filtering or setting brightness thresholds. However, in this specific step structure environment, the false contour, which exhibits a continuous and traceable pattern, can disguise itself as the target joint, misleading existing recognition algorithms and causing the laser focus to deviate from the true fusion boundary, resulting in welding quality defects such as incomplete fusion, weak welds, or sealing failure. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a visual monitoring-based welding system for metal thermos cups. This system solves the problem that at the steps of the metal thermos cup base, interference from highly reflective materials and specific geometric structures creates a continuous virtual weld seam outline, misleading the joint tracking algorithm and causing the laser focus to deviate from the actual joint trajectory.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a visual monitoring-based welding system for metal insulated cups, comprising:
[0006] The contour acquisition module is used to obtain the circumferential sampling position, control the visual monitoring component to acquire images and map them to the physical coordinate system to extract inner and outer candidate contours, match the inner and outer candidate contours to generate candidate contour feature points, and calculate the reflection drift amount by combining the preset step wall radial normal.
[0007] The back-source module is used to extract physical contour point sets from the image to construct candidate reflective surfaces, identify high reflectivity source areas based on visual brightness values, and calculate the shortest Euclidean distance from the candidate contour's back-symmetric point about the candidate reflective surface to the high reflectivity source area to generate back-source residuals.
[0008] The bone line generation module is used to calculate the actual weld occupancy based on the return source residual and the reflection wander amount. The actual weld occupancy is negatively correlated with the reflection wander amount and positively correlated with the return source residual. The candidate contour feature points are fused with the preset clamping reference joint points according to the actual weld occupancy to generate the actual circumferential bone line points.
[0009] The focus correction module is used to map the real circumferential seam bone line points to the robot base coordinate system to obtain the execution system target point, map the candidate contour feature points to the base coordinate candidate contour points, combine the execution system target point, the base coordinate candidate contour points and the real weld seam occupancy to calculate the virtual traction vector, and generate the laser focus correction amount based on the obtained current laser focus point and the virtual traction vector.
[0010] Preferably, the contour acquisition module matches inner and outer candidate contours to generate candidate contour feature points, including:
[0011] Calculate the local curvature of each discrete point on the inner and outer candidate contours, and extract the cross-sectional inflection points corresponding to the local curvature maxima on the inner and outer candidate contours respectively, which are used as the inner contour feature points and the outer contour feature points respectively.
[0012] Linear fitting of the inner and outer candidate contours is performed based on the least squares method, and the unit direction vector of the fitted line is used as the inner and outer unit tangents.
[0013] Obtain the radial width of the region of interest, and calculate the Euclidean distance scalar between the inner and outer contour feature points within the two-dimensional physical section.
[0014] The matching cost is calculated by adding a constant to the ratio of the Euclidean distance scalar to the radial width of the region of interest, and subtracting the difference between the absolute values of the dot product of the inner and outer unit tangential directions.
[0015] Preferably, the contour acquisition module calculates the reflection shift of the candidate contour in conjunction with the radial normal of the preset step wall, including:
[0016] The position deviation vector is constructed by subtracting the feature points of the outer contour from those of the inner contour;
[0017] Perform a dot product operation between the position deviation vector and the radial normal of the preset step wall, and obtain the projection component of the dot product operation as the candidate contour reflection wander amount;
[0018] When the matching cost is greater than or equal to the preset dimensionless cost truncation limit or the candidate contour is only presented under unilateral illumination, the candidate contour is determined to be a single-state candidate contour.
[0019] The reflection wander amount of the single-state candidate contour is assigned to the extreme value of a constant equal to the radial width of the region of interest, and the feature points extracted under unilateral illumination are assigned as the candidate contour feature points corresponding to the single-state candidate contour.
[0020] Preferably, the reflective source module constructs candidate reflective surfaces and identifies high-reflectivity source regions, including:
[0021] Linear fitting is performed on local contour segments that exhibit a continuous distribution in the physical contour point set based on the least squares method.
[0022] Construct a two-dimensional spatial geometric equation that includes the unit normal vector of the candidate reflecting surface and the position constant of the candidate reflecting surface;
[0023] Calculate the median luminance and the absolute deviation of the median luminance within the spatial neighborhood of the candidate reflective surface;
[0024] The set of physical points whose visual brightness values are greater than or equal to the sum of the median brightness and three times the absolute deviation of the median brightness, and which have spatial continuous distribution characteristics, are identified as high reflectance source areas.
[0025] When there is no set of physical points in the spatial neighborhood of a candidate reflective surface whose visual brightness value is greater than or equal to the sum of the absolute deviations of the median brightness and three times the median brightness, the high reflectivity source region is marked as an empty set.
[0026] Preferably, the return-to-source module generates return-to-source residuals, including:
[0027] An inverse test model is constructed based on the unit normal vector of the candidate reflective surface and the position constant of the candidate reflective surface. A mirror coordinate transformation is performed on the visual measurement contour points contained in the candidate contour to obtain the folded symmetry points.
[0028] Calculate the arithmetic mean of the shortest Euclidean distances from all symmetrical points within the candidate profile to the high-reflectivity source region;
[0029] Iterate through and extract the minimum value from the arithmetic mean of the candidate reflective surfaces, and use the minimum value as the return source residual;
[0030] When the high-reflectivity source region is an empty set, the return source residual is assigned a constant equal to the size of the radial width scalar of the region of interest.
[0031] Preferably, the bone line generation module calculates the actual weld seam occupancy, including:
[0032] Determine the minimum resolvable length of the visual system;
[0033] Extract the absolute value scalar of the reflected shift;
[0034] Set the residual value of the return-to-source calculation as the numerator of the division operation;
[0035] The absolute values of the return-to-source residual and the reflection wander amount are added to the minimum resolvable length of the visual system to generate the division denominator.
[0036] The ratio of the numerator to the denominator of the division is used to generate the actual weld seam occupancy.
[0037] Preferably, the bone line generation module merges candidate contour feature points with preset clamping reference joint points according to the actual weld seam occupancy to generate actual circumferential seam bone line points, including:
[0038] Extract the radial and height coordinate components of the candidate contour feature points and the preset clamping reference joint points respectively;
[0039] The radial coordinate components of the candidate contour feature points are multiplied by the corresponding actual weld occupancy and summed. The product of the radial coordinate components of the preset clamping reference joint point and the preset dimensionless stability factor is added to generate a radially weighted numerator.
[0040] The product of the height coordinate components of the candidate contour feature points and the corresponding actual weld seam occupancy is summed, and the product of the height coordinate components of the preset clamping reference joint point and the preset dimensionless stability factor is added to generate a height-weighted numerator.
[0041] The actual weld occupancy is summed and a preset dimensionless stability factor is added to generate a weighted normalized denominator.
[0042] The ratios of the radially weighted numerator to the weighted normalized denominator and the ratios of the height-weighted numerator to the weighted normalized denominator are calculated separately, and the actual suture bone line points are reconstructed.
[0043] Preferably, the focus correction module maps the actual suture bone line points to the robot base coordinate system to obtain the execution system target points, including:
[0044] Extract the radial coordinate components and height coordinate components of the actual circumferential suture bone line points;
[0045] Based on the circumferential angle corresponding to the circumferential sampling position, the radial coordinate components of the real circumferential suture bone line points are decomposed into three-dimensional orthogonal coordinates using trigonometric function projection relationships.
[0046] By combining the height coordinate components of the real suture bone line point with the orthogonal coordinates of the three-dimensional plane, a three-dimensional spatial target point is constructed in the three-dimensional coordinate system of the metal thermos cup body, and the three-dimensional spatial target point is converted into a homogeneous coordinate vector containing the last constant one.
[0047] Retrieve the fourth-order homogeneous coordinate transformation matrix from the rotation center axis of the cup to the robot base;
[0048] Perform matrix multiplication on the fourth-order homogeneous coordinate transformation matrix and the homogeneous coordinate vector to obtain the target point of the execution system.
[0049] Preferably, the focus correction module calculates the virtual weld traction vector, including:
[0050] The candidate contour feature points are mapped to base coordinate candidate contour points in the robot's base coordinate system using a homogeneous coordinate transformation path.
[0051] Subtract the actual weld seam occupancy from the constant to obtain the confidence inverse weight;
[0052] Extract the first three-dimensional physical coordinate components of the candidate contour points of the base coordinate system and the target point of the execution system respectively, and perform a subtraction operation to obtain the position deviation vector;
[0053] Multiply the confidence score inverse weight by the position deviation vector and perform a traversal summation calculation to generate the traction weighted numerator;
[0054] The confidence score inverse weights are iterated and summed, and a preset dimensionless stabilization factor is added to generate the traction normalized denominator.
[0055] Calculate the ratio of the traction weighted numerator to the traction normalized denominator to generate a virtual weld traction vector.
[0056] Preferably, the focus correction module generates the laser focus correction amount, including:
[0057] The minimum resolvable length of the vision system, the preset laser spot radius, the preset traction suppression coefficient, and the preset follow correction coefficient are retrieved.
[0058] Extract the magnitude of the virtual weld traction vector;
[0059] The virtual weld traction vector magnitude is added to the minimum resolvable length of the vision system to generate a limiting denominator.
[0060] Calculate the ratio of the preset laser spot radius to the limiting denominator to generate a dimensionless spot limiting ratio;
[0061] Compare the preset traction suppression coefficient with the dimensionless spot limiting ratio, and extract the minimum value as the actual traction suppression coefficient;
[0062] Subtract the current laser focus point from the target point of the execution system to generate a positive position deviation vector;
[0063] Multiply the positive position deviation vector by the preset following correction coefficient to generate a positive proportional control vector;
[0064] Multiply the virtual weld traction vector by the actual traction suppression coefficient to generate a three-dimensional correction vector;
[0065] Subtract the three-dimensional correction vector from the positive proportional control vector to generate the laser focus correction amount.
[0066] Compared with existing technologies, it has the following advantages:
[0067] This proposed visual monitoring-based welding system for metal thermos cups addresses the problem of laser trajectory yaw caused by false contours resulting from reflections from stepped walls by establishing dynamic verification logic. Utilizing the radial offset attribute generated by optical path switching and geometric mirror tracing logic, this system can identify pseudo-feature signals with continuous topological morphology. While the actual seam position does not change with the incident angle, the reflected ghost image exhibits significant radial drift with optical axis switching. By quantifying this drift and combining it with the return-to-source residual, the system assigns a differentiated confidence level to each identified trajectory. This process effectively separates true and false targets in complex optical environments, ensuring that the seam tracking algorithm can accurately extract the actual fusion boundary from high-brightness reflection interference, effectively improving the system's recognition accuracy under extreme reflection conditions.
[0068] At the control execution level, this solution further enhances the stability of the laser welding process by constructing an active compensation mechanism for virtual traction vectors. This solution quantifies the reflection interference that originally caused deviations into spatial traction components and performs a reverse repulsion operation when generating the laser focus correction, thereby offsetting the erroneous induction force of optical ghosting on the processing path. Compared to conventional noise filtering methods, this logic transforms interference factors into quantifiable control parameters. Combined with preset clamping references and amplitude limiting mechanisms, it ensures that the laser focus can still be safely anchored in the actual joint area even when the visual signal is contaminated or subjected to strong reflective contamination. This control strategy effectively solves the problems of incomplete fusion and weld misalignment caused by step image misleading, significantly reducing the equipment's sensitivity to ambient light and workpiece surface conditions while ensuring welding quality. Attached Figure Description
[0069] Figure 1 This is a schematic diagram of the system framework of the present invention.
[0070] Figure 2 This is a schematic diagram of the system execution flow of the present invention. Detailed Implementation
[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0072] Please see Figures 1 to 2 This application provides a welding system for metal thermos cup products based on visual monitoring, including a contour acquisition module, a return and source homing module, a bone line generation module and a focus correction module;
[0073] The contour acquisition module monitors the current circumferential sampling position in real time through the pulse signal output by the servo motor encoder as the rotating mechanism drives the metal thermos cup to rotate continuously. When the pulse value reaches the preset angle pulse threshold, the control unit will issue a hardware synchronization trigger command to drive the vision monitoring component within the defined region of interest. The inner-opening measurement uses a line structured light to perform the first image acquisition, extracting at least one inner candidate contour. After a fixed system switching delay, the visual monitoring component activates the outer illumination at the same sampling location, performs a second image acquisition, and extracts at least one outer candidate contour. .
[0074] Furthermore, to address the phase time difference caused by the continuous rotation of the cup between the two acquisitions, the control unit constructs a coordinate rotation compensation matrix based on the real-time angular velocity feedback of the rotating mechanism and the time interval between the two acquisition actions, and then applies this matrix to the outer candidate contour. Perform inverse displacement registration. The system maps the 2D pixel set acquired by the camera to the radial-height physical coordinate system of the current sampling section using pre-calibrated light plane parameters. Under this physical plane, the control unit determines the contour connection length. Area of interest The discrete measurement points within the range are connected in series to form a continuous geometric trajectory.
[0075] Based on this, the control unit calculates the local curvature of each discrete point on the continuous geometric trajectory of the inner and outer sides, and extracts the inflection points of the cross sections corresponding to the local curvature maxima (or extracts the geometric intersection points of the trajectory fitting line segments), which are used as the inner contour feature points. and outer contour feature points The corresponding trajectory point set is linearly fitted using the least squares method, and the unit direction vector of the fitted line is taken as the inner unit tangent. and outer unit tangential To determine the homology of contours under different lighting conditions, the control unit traverses the region of interest. The matching cost for all combinations of inner and outer contours is calculated using the following formula. :
[0076]
[0077] In the formula, the matching cost It is a dimensionless evaluation parameter; is the Euclidean distance scalar between the inner and outer contour feature points within the two-dimensional physical cross section; Region of Interest The radial width; and The unit tangent of the two contours is given. This formula transforms static image comparison into a dynamic optical path consistency check by weighted summation of the positional deviation ratio and orientation consistency weights, in order to quantitatively identify visual features belonging to the same physical entity or the same reflection source.
[0078] To achieve matching determination, the system employs a greedy matching or bipartite graph matching algorithm, selecting the matching cost sequentially while ensuring the radial arrangement order remains consistent. The smallest multiple contour pairs are used as matching targets. The matching cost of the matching target... When the cost is less than the preset dimensionless cost cutoff limit, the system identifies it as the Kth candidate profile under the current section. and the inner contour feature points With outer contour feature points The geometric midpoint within the two-dimensional physical cross-section is used as the candidate contour feature point for the fused contour. Subsequently, the control unit retrieves the pre-calibrated reference data of the cup bottom step to determine the radial normal of the preset step vertical wall. And calculate the Kth candidate contour according to the following formula. Reflection drift :
[0079]
[0080] In the formula, A physical quantity characterizing the degree of radial drift of the profile as the incident light changes; Let be the unit direction vector indicating the orientation of the normal to the solid wall of the step at the bottom of the cup, and its magnitude is strictly equal to one. This formula effectively separates the radial wandering component guided by the mirror reflection path of the step through the projection properties of the dot product operation, while effectively isolating mechanical vibration interference in irrelevant dimensions.
[0081] When the minimum matching cost If the cost is greater than or equal to the upper limit of the cutoff, or if the contour is only visible under unilateral illumination, the system marks it as a singlet candidate contour. And let it reflect the amount of free movement. The output is a constant extreme value, the value of which is directly equal to the radial width. The scalar size, and the feature points extracted from the singlet candidate contour under unilateral illumination ( or Directly assign values as candidate contour feature points .
[0082] It should be noted that the energy density of the structured light used for measurement must be set below the thermal damage threshold of the target material to ensure that no heat-affected zone is generated during the measurement process. The setting of the angle pulse threshold is related to the encoder resolution and the spatial sampling step size to ensure the uniformity of the physical distribution of sampling points. Since the spatial position of the solid seam is stable under different incident angles, the reflection migration of the real weld tends to be zero, while the virtual bright band will slide as the optical path switches.
[0083] It should be noted that the above-mentioned cost cutoff limit is determined by the system based on the laser spot radius. With radial width The ratio correlation setting stipulates that the normal value of the upper limit of cost truncation is no greater than [value missing]. This limit is used to block logical associations between irrelevant noise points in highly reflective environments. The Kth candidate contour after extraction is complete. and its representative points The reflected drift will be transmitted as an entity target to the return-to-origin module for geometric mirroring and tracing, while the reflected drift will be transmitted as a result. Then it is considered as the candidate contour. The amount of anti-counterfeiting labels is directly transmitted to the bone line generation module to participate in the determination of the actual weld seam occupancy.
[0084] Among them, the backtracking module receives the Kth candidate contour transmitted by the contour acquisition module. and its candidate contour feature points The system extracts the physical contour point set of the cup bottom step region within the two-dimensional physical coordinate system of the current sampling section. The control unit performs line fitting on the continuously distributed local contour segments within the physical contour point set using the least squares method to construct at least one candidate reflecting surface. Candidate reflective surface The two-dimensional spatial geometric equation is expressed as:
[0085]
[0086] In the formula, The unit normal vector of the candidate reflective surface represents the orientation of the normal to the corresponding local contour segment within the physical cross section, and its magnitude is equal to one. These are visually measured contour points within a two-dimensional physical cross-section; Let be the position constant of the candidate reflecting surface. This formula uses a linear equation to abstract discrete metal surface data into an ideal reflection boundary that assists in optical path tracing.
[0087] Furthermore, the control unit acquires the reflection intensity attributes of each point coupled in the physical contour point set, which are used as visual brightness values. To identify potential reflective light sources, the control unit operates on each candidate reflective surface. Calculate visual brightness value within spatial neighborhood Median brightness and its median absolute deviation Then, the control unit extracts the data according to the extraction criteria. The physical contour point set is screened, and the set of physical points that meet the extraction criteria and have spatially continuous distribution characteristics is identified as the corresponding candidate reflecting surface. High reflectivity source area High reflectivity source area This is the set of points in the two-dimensional physical coordinate system of the current sampling section, whose coordinate units are the same as those of the visually measured contour points. Maintain consistency. Use the median absolute deviation of brightness. Determining the extraction threshold is to filter out random arc pulse interference from the production site through statistical robustness, ensuring that the extraction results accurately correspond to the physically strong reflective areas on the cup bottom structure. If a candidate reflective surface... If the above extraction criteria are not met within the neighborhood, then the corresponding high-reflectivity source region will be... Marked as an empty set.
[0088] Based on this, the control unit processes the received Kth candidate contour. Each visual measurement contour point in Based on the symmetry of mirror reflection, an inverse test model is constructed, and its relationship with each candidate reflecting surface is calculated. The point of reflection and symmetry :
[0089]
[0090] In the formula, For visual measurement of contour points Regarding candidate reflective surfaces The symmetrical coordinate vector within the two-dimensional physical section after virtual optical path tracing. This formula uses mirror transformation to map points on the candidate contour back to possible luminous points, in order to verify whether the contour is generated by reflection from the stepped surface.
[0091] To quantify the reflection correlation of candidate contours, the control unit calculates the Kth candidate contour. All internal turning points symmetrical To the corresponding high reflectivity source area The arithmetic mean of the shortest Euclidean distances. The control unit traverses all candidate reflective surfaces. The minimum value among the above averages is taken as the candidate contour. Return to source residual quantity :
[0092]
[0093] In the formula, The residual measure, in millimeters, characterizes the candidate profile as a product of secondary reflection from a stepped specular surface. This represents the total number of points contained in the contour. Indicates the point of reflection and symmetry With high reflectivity source area The shortest physical distance between them; This serves as the traversal index for candidate reflective surfaces. When a candidate profile is indeed formed by secondary reflections from a step, its reflection trajectory will point vertically towards the region of high reflection. This results in a return-to-source residual that approaches zero. Due to the physical reality of their location, solid metal seams, when reflected back, cannot hit the area of high reflectivity. This results in a significantly larger distance residual.
[0094] It should be noted that, for a given candidate contour If all candidate reflective surfaces at the current circumferential position Corresponding high reflectivity source area All are empty sets; the control unit directly returns the source residual. The output is equal to the radial width of the region of interest. A constant of scalar magnitude. After completing the above calculations, the backtracking module outputs the Kth candidate contour. Corresponding return-to-source residual amount This data is used as the core data label for determining optical artifacts and is directly transmitted to the bone line generation module to participate in the subsequent calculation of the actual weld seam occupancy.
[0095] Among them, the bone line generation module receives the Kth candidate contour transmitted by the contour acquisition module. and its candidate contour feature points With reflection shift And the corresponding return-to-source residual amount passed by the return-to-source module. To reduce the geometric contribution of the virtual weld profile to the calculation of the actual joint location, the control unit performs a specific calculation for each candidate profile under the current section. The actual weld seam occupancy is calculated using the following formula. :
[0096]
[0097] In the formula, A dimensionless weighting parameter with a value range greater than or equal to zero and less than one; The residual amount returned to the source is measured in millimeters. Reflection shift The absolute value scalar, with the unit being millimeters; This is the minimum resolvable length of the vision system, measured in millimeters. In the low-level calculations, the control unit will return the homing residual. Set as the numerator of the division, and return the residual to the source. scalar of absolute value of reflected free displacement With the minimum resolvable length of the visual system The addition operation generates the denominator for division, and the actual weld occupancy is generated by calculating the ratio of the numerator to the denominator. This formula couples and quantizes the two isolated optical features of the secondary reflection of the step. When a candidate profile has significant reflection shift and is very likely to reflect back to a highly reflective light source, its denominator increases and its numerator decreases, resulting in an output with extremely low actual weld seam occupancy. Conversely, due to the weak migration and difficulty in returning to the source, physical seams will output higher weights, thereby achieving dimensionality reduction and separation of physical authenticity features at the mathematical level.
[0098] It should be noted that the minimum resolvable length Determined by the actual physical size corresponding to a single pixel of the camera or the width of the line laser stripe calibration, its role here is not as a conventional defect judgment threshold, but as a small physical quantity to eliminate mathematical singularities and prevent the denominator from being zero, ensuring the robustness of the underlying industrial calculation when both the drift and the residual approach zero.
[0099] Based on this, the control unit retrieves the clamping reference joint point determined by the offline calibration of the tooling or the product design parameters. The control unit will extract all candidate contour feature points within the current two-dimensional physical section. According to its corresponding actual weld seam occupancy Spatial weighted fusion is performed to calculate and generate the actual circumferential suture bone line points under the current cross-section. :
[0100]
[0101] In the formula, This is the final target bone line point coordinate vector calculated in the two-dimensional physical coordinate system of the current sampling section; For rigid reference coordinate vectors in the same physical coordinate system; To prevent the dimensionless stabilizing factor from reaching zero; This represents the mathematical summation of the corresponding parameters of all candidate contours extracted under the current cross-section. This formula constructs a dynamic gravitational field driven by physical optical characteristics within the two-dimensional cross-section, representing the actual weld seam occupancy. The larger the representative point is relative to the bone line point The stronger the traction force, the more likely the final tracking target will be anchored in the actual physical seam area.
[0102] Furthermore, the aforementioned weighted fusion formula involves algebraic operations on two-dimensional vectors. In the actual execution of the underlying control program, the control unit will select candidate contour feature points... Joint point with clamping reference The radial and height coordinate components are extracted separately along the two-dimensional physical cross-section. For the radial dimension, the product of the radial coordinate component of the candidate contour feature point and the corresponding actual weld occupancy is summed, and the product of the radial coordinate component of the preset clamping reference joint point and the preset dimensionless stability factor is added to generate a radially weighted numerator. Similarly, the same logic is performed for the height dimension to generate a height-weighted numerator. At the same time, the actual weld occupancy is summed and the preset dimensionless stability factor is added to generate a weighted normalized denominator. The control unit calculates the ratio of the radially weighted numerator to the weighted normalized denominator and the ratio of the height-weighted numerator to the weighted normalized denominator, respectively. Finally, the two output scalar components are recombined to form the two-dimensional actual circumferential seam bone line points. .
[0103] It should be noted that the clamping reference joint point With stable factors The introduction of this constitutes a fallback mechanism against system collapse under harsh industrial conditions. Dimensionless stability factor. The normal value is set to a small positive number (ranging from 0.01 to 0.1) to avoid significant interference with the weighting of the actual weld occupancy during normal measurements. When the field of view is subject to extremely high reflectivity, causing all actual weld occupancy to be affected... When all values approach zero, or when the summation term is an empty set due to smoke and dust obscuring the current section, the true circumferential suture bone line points are... The coordinates will be stabilized by the stabilization factor. Smooth transition and forced anchoring to the clamping reference joint point This effectively prevents the laser focus from being drawn to dangerous processing areas by optical ghosting. After completing the calculation of the current section, the bone line generation module will output the actual circumferential bone line points. The spatial tracking target is directly transmitted to the focus correction module, while the actual weld occupancy of each candidate contour is also transmitted. It is passed to the focus correction module to participate in subsequent virtual traction and repulsion calculations.
[0104] The focus correction module receives the actual circumferential suture bone line points transmitted by the bone line generation module. and each candidate contour feature point and the actual weld seam occupancy of the bonded area The control unit is based on the current circumferential sampling position. The corresponding circumferential angles are used to project the actual circumferential suture bone line points in the radial and height two-dimensional sections using trigonometric functions. The radial coordinate components are decomposed into three-dimensional planar orthogonal coordinates, thereby constructing a three-dimensional spatial target point in the three-dimensional coordinate system of the metal thermos cup body (whose origin is located at the reference point of the central axis of the rotating mechanism). Subsequently, the control unit retrieves the calibration transformation matrix from the cup's three-dimensional coordinate system to the robot's base coordinate system. The execution system target point is calculated using the following matrix multiplication formula. :
[0105]
[0106] In the formula, This refers to the physical coordinate vector of the laser welding head in the three-dimensional robot base coordinate system. This is the fourth-order homogeneous coordinate transformation matrix from the rotation center axis of the cup to the robot base; Let be a homogeneous coordinate vector containing the last constant 1. This formula establishes a rigid mathematical connection between the physical space of the tooling rotation and the mechanical motion space of the robot.
[0107] Furthermore, the control unit uses the same homogeneous coordinate transformation path as described above to transform the current circumferential sampling position. Each candidate contour feature point Synchronous mapping unfolds into candidate contour points of base coordinates in the robot's base coordinate system. Each candidate contour point of the base coordinate system The actual weld occupancy corresponding to its original candidate profile. To quantify the tendency of highly reflective ghost images to mislead the laser focus, the control unit traverses all candidate contour points of the base coordinates within the current cross-section. The virtual weld traction vector is calculated using the following formula. :
[0108]
[0109] In the formula, A traction vector with three-dimensional spatial directionality, measured in millimeters; and All three-dimensional physical coordinate components are extracted for spatial vector operations. Characterize the corresponding candidate contour points in the base coordinate system As an inverse weight of the confidence level of spurious welds; To prevent the dimensionless stabilizing factor from reaching zero; This involves iterating and summing the corresponding parameters of all candidate contour points within the current cross-section. In the specific calculation process, the system extracts the first three-dimensional physical coordinate components of the candidate contour points and the target point, subtracts them to obtain the position deviation vector, multiplies the confidence inverse weight by the position deviation vector, and performs a summation calculation to generate the traction weighted numerator. The confidence inverse weight is then summed again and a preset dimensionless stability factor is added to generate the traction normalized denominator. Finally, the ratio of the traction weighted numerator to the traction normalized denominator is calculated to generate the virtual weld traction vector. The core of this formula lies in the actual weld seam occupancy. The lower the point, the greater the probability of it being a reflected virtual image. By applying reverse weights and obtaining the spatial geometric centroid, the erroneous guidance direction component caused by the virtual image of the step reflection can be effectively extracted.
[0110] Based on this, the control unit obtains the current laser focus point by reading the real-time position feedback of the tool center point of the mechanical actuator. And combined with the laser spot radius With the minimum resolvable length of the visual system Dynamically calculate the actual traction suppression coefficient The specific calculation logic is as follows: The magnitude of the virtual weld seam traction vector is... With the minimum resolvable length of the visual system Perform addition to generate the amplitude-limiting denominator and calculate the preset laser spot radius. The ratio of the limiting denominator to the value of the limiting denominator generates a dimensionless spot limiting ratio. This is then compared to the preset traction suppression coefficient. The minimum value of the ratio between the dimensionless beam limiter and the actual traction suppression coefficient is extracted. .
[0111] The control unit then calculates the final laser focus correction using the following formula. :
[0112]
[0113] In the formula, The vector correction amount used to drive the actuator to move in three-dimensional space, in millimeters; The laser focal point represents the approach of the actuator target point. The positive position deviation vector. It should be noted that the following correction coefficient... With preset traction suppression coefficient All values are dimensionless proportional gain constants with a range greater than zero and less than or equal to one, preset by the system, to ensure the stability of the control closed loop. This formula actively superimposes a vector with the virtual weld traction vector into the positive proportional control vector that drives the laser focus to approach the actual weld position. A three-dimensional correction vector exhibiting an inverse repulsion relationship. The actual traction suppression coefficient is calculated using the beam limiting ratio. The aim is to achieve physical-level compensation cutoff, effectively preventing the laser focus from deviating from the effective coverage range of the actual physical size of the molten pool due to excessive reverse repulsion.
[0114] Finally, the control unit will include laser focus correction values in three dimensions, representing both the direction and the magnitude of displacement. The command is sent to the servo driver; the servo driver drives the laser welding head from the current point of the laser focus. Perform the laser focus correction along three-dimensional space The vector displacement is determined, and the laser fusion action is performed while maintaining the set process parameters after reaching the target position; the current circumferential sampling position is completed. After processing, the control unit drives the rotating mechanism to rotate in a circumferential direction, triggering the data flow of the next module until the three-dimensional trajectory obstacle avoidance tracking and continuous welding closed loop of the entire metal thermos cup bottom step annular weld is completed.
[0115] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A welding system for metal insulated cup products based on visual monitoring, characterized in that, include: The contour acquisition module is used to obtain the circumferential sampling position, control the visual monitoring component to acquire images and map them to the physical coordinate system to extract inner and outer candidate contours, match the inner and outer candidate contours to generate candidate contour feature points, and calculate the reflection drift amount by combining the preset step wall radial normal. The back-source module is used to extract physical contour point sets from the image to construct candidate reflective surfaces, identify high reflectivity source areas based on visual brightness values, and calculate the shortest Euclidean distance from the candidate contour's back-symmetric point about the candidate reflective surface to the high reflectivity source area to generate back-source residuals. The bone line generation module is used to calculate the actual weld occupancy based on the return source residual and the reflection wander amount. The actual weld occupancy is negatively correlated with the reflection wander amount and positively correlated with the return source residual. The candidate contour feature points are fused with the preset clamping reference joint points according to the actual weld occupancy to generate the actual circumferential bone line points. The focus correction module is used to map the real circumferential seam bone line points to the robot base coordinate system to obtain the execution system target point, map the candidate contour feature points to the base coordinate candidate contour points, combine the execution system target point, the base coordinate candidate contour points and the real weld seam occupancy to calculate the virtual traction vector, and generate the laser focus correction amount based on the obtained current laser focus point and the virtual traction vector.
2. The metal thermos cup welding system based on visual monitoring according to claim 1, characterized in that, The contour acquisition module matches inner and outer candidate contours to generate candidate contour feature points, including: Calculate the local curvature of each discrete point on the inner and outer candidate contours, and extract the cross-sectional inflection points corresponding to the local curvature maxima on the inner and outer candidate contours respectively, which are used as the inner contour feature points and the outer contour feature points respectively. Linear fitting of the inner and outer candidate contours is performed based on the least squares method, and the unit direction vector of the fitted line is used as the inner and outer unit tangents. Obtain the radial width of the region of interest, and calculate the Euclidean distance scalar between the inner and outer contour feature points within the two-dimensional physical section. The matching cost is calculated by adding a constant to the ratio of the Euclidean distance scalar to the radial width of the region of interest, and subtracting the difference between the absolute values of the dot product of the inner and outer unit tangential directions.
3. The metal thermos cup welding system based on visual monitoring according to claim 2, characterized in that, The contour acquisition module calculates the reflection shift of candidate contours by combining the radial normal of the preset step wall, including: The position deviation vector is constructed by subtracting the feature points of the outer contour from those of the inner contour; Perform a dot product operation between the position deviation vector and the radial normal of the preset step wall, and obtain the projection component of the dot product operation as the candidate contour reflection wander amount; When the matching cost is greater than or equal to the preset dimensionless cost truncation limit or the candidate contour is only presented under unilateral illumination, the candidate contour is determined to be a single-state candidate contour. The reflection wander amount of the single-state candidate contour is assigned to the extreme value of a constant equal to the radial width of the region of interest, and the feature points extracted under unilateral illumination are assigned as the candidate contour feature points corresponding to the single-state candidate contour.
4. The metal thermos cup welding system based on visual monitoring according to claim 1, characterized in that, The reflective source module constructs candidate reflective surfaces and identifies high-reflectivity source regions, including: Linear fitting is performed on local contour segments that exhibit a continuous distribution in the physical contour point set based on the least squares method. Construct a two-dimensional spatial geometric equation that includes the unit normal vector of the candidate reflecting surface and the position constant of the candidate reflecting surface; Calculate the median luminance and the absolute deviation of the median luminance within the spatial neighborhood of the candidate reflective surface; The set of physical points whose visual brightness values are greater than or equal to the sum of the median brightness and three times the absolute deviation of the median brightness, and which have spatial continuous distribution characteristics, are identified as high reflectance source areas. When there is no set of physical points in the spatial neighborhood of a candidate reflective surface whose visual brightness value is greater than or equal to the sum of the absolute deviations of the median brightness and three times the median brightness, the high reflectivity source region is marked as an empty set.
5. The metal thermos cup welding system based on visual monitoring according to claim 4, characterized in that, The backtracking module generates backtracking residuals, including: An inverse test model is constructed based on the unit normal vector of the candidate reflective surface and the position constant of the candidate reflective surface. A mirror coordinate transformation is performed on the visual measurement contour points contained in the candidate contour to obtain the folded symmetry points. Calculate the arithmetic mean of the shortest Euclidean distances from all symmetrical points within the candidate profile to the high-reflectivity source region; Iterate through and extract the minimum value from the arithmetic mean of the candidate reflective surfaces, and use the minimum value as the return source residual; When the high-reflectivity source region is an empty set, the return source residual is assigned a constant equal to the size of the radial width scalar of the region of interest.
6. The metal thermos cup welding system based on visual monitoring according to claim 1, characterized in that, The bone line generation module calculates the actual weld seam occupancy, including: Determine the minimum resolvable length of the visual system; Extract the absolute value scalar of the reflected shift; Set the residual value of the return-to-source calculation as the numerator of the division operation; The absolute values of the return-to-source residual and the reflection wander amount are added to the minimum resolvable length of the visual system to generate the division denominator. The ratio of the numerator to the denominator of the division is used to generate the actual weld seam occupancy.
7. The metal thermos cup welding system based on visual monitoring according to claim 1, characterized in that, The bone line generation module merges candidate contour feature points with preset clamping reference joint points according to the actual weld seam occupancy to generate actual circumferential seam bone line points, including: Extract the radial and height coordinate components of the candidate contour feature points and the preset clamping reference joint points respectively; The radial coordinate components of the candidate contour feature points are multiplied by the corresponding actual weld seam occupancy and summed. The product of the radial coordinate components of the preset clamping reference joint point and the preset dimensionless stability factor is added to generate a radially weighted numerator. The product of the height coordinate components of the candidate contour feature points and the corresponding actual weld seam occupancy is summed, and the product of the height coordinate components of the preset clamping reference joint point and the preset dimensionless stability factor is added to generate a height-weighted numerator. The actual weld occupancy is summed and a preset dimensionless stability factor is added to generate a weighted normalized denominator. The ratios of the radially weighted numerator to the weighted normalized denominator and the ratios of the height-weighted numerator to the weighted normalized denominator are calculated separately, and the actual suture bone line points are reconstructed.
8. The metal thermos cup welding system based on visual monitoring according to claim 7, characterized in that, The focus correction module maps the actual suture bone line points to the robot's base coordinate system to obtain the execution system target points, including: Extract the radial coordinate components and height coordinate components of the actual circumferential suture bone line points; Based on the circumferential angle corresponding to the circumferential sampling position, the radial coordinate components of the real circumferential suture bone line points are decomposed into three-dimensional orthogonal coordinates using trigonometric function projection relationships. By combining the height coordinate components of the real suture bone line point with the orthogonal coordinates of the three-dimensional plane, a three-dimensional spatial target point is constructed in the three-dimensional coordinate system of the metal thermos cup body, and the three-dimensional spatial target point is converted into a homogeneous coordinate vector containing the last constant one. Retrieve the fourth-order homogeneous coordinate transformation matrix from the rotation center axis of the cup to the robot base; Perform matrix multiplication on the fourth-order homogeneous coordinate transformation matrix and the homogeneous coordinate vector to obtain the target point of the execution system.
9. The metal thermos cup welding system based on visual monitoring according to claim 8, characterized in that, The focus correction module calculates the virtual weld pull vector, including: The candidate contour feature points are mapped to base coordinate candidate contour points in the robot's base coordinate system using a homogeneous coordinate transformation path. Subtract the actual weld seam occupancy from the constant to obtain the confidence inverse weight; Extract the first three-dimensional physical coordinate components of the candidate contour points of the base coordinate system and the target point of the execution system respectively, and perform a subtraction operation to obtain the position deviation vector; Multiply the confidence score inverse weight by the position deviation vector and perform a traversal summation calculation to generate the traction weighted numerator; The confidence score inverse weights are iterated and summed, and a preset dimensionless stabilization factor is added to generate the traction normalized denominator. Calculate the ratio of the traction weighted numerator to the traction normalized denominator to generate a virtual weld traction vector.
10. The metal thermos cup welding system based on visual monitoring according to claim 9, characterized in that, The focus correction module generates laser focus correction values, including: The minimum resolvable length of the vision system, the preset laser spot radius, the preset traction suppression coefficient, and the preset follow correction coefficient are retrieved. Extract the magnitude of the virtual weld traction vector; The virtual weld traction vector magnitude is added to the minimum resolvable length of the vision system to generate a limiting denominator. Calculate the ratio of the preset laser spot radius to the limiting denominator to generate a dimensionless spot limiting ratio; Compare the preset traction suppression coefficient with the dimensionless spot limiting ratio, and extract the minimum value as the actual traction suppression coefficient; Subtract the current laser focus point from the target point of the execution system to generate a positive position deviation vector; Multiply the positive position deviation vector by the preset following correction coefficient to generate a positive proportional control vector; Multiply the virtual weld traction vector by the actual traction suppression coefficient to generate a three-dimensional correction vector; Subtract the three-dimensional correction vector from the positive proportional control vector to generate the laser focus correction amount.