Vacuum cup welding station online quality inspection system based on machine vision

Abstract

The invention relates to the technical field of online quality inspection, in particular to a vacuum cup welding station online quality inspection system based on machine vision, and the system comprises a positioning collection module which is used for generating a trigger signal when a cup body reaches a preset position; the multi-modal imaging module is used for carrying out image acquisition according to the trigger signal and acquiring a multi-angle two-dimensional image and a three-dimensional contour image of a welding area; according to the multi-angle two-dimensional image, when the device is used, after the multi-angle two-dimensional image and the three-dimensional contour image of the welding area are fused through the image processing module and the defect analysis module to restrain hot spots, a small gap is not covered by reflected light any more, false alarms are reduced, and meanwhile, the defect detection accuracy is improved. The amplification effect of the fine line light on the convex-concave part is overlaid, so that early risks can be found conveniently; and the air leakage risk which is exposed at the rear section originally is moved to the welding station, so that the reworking and scrapping of the rear section and the waste of invested working hours can be reduced.

Description

Technical Field

[0001] This invention relates to the field of online quality inspection technology, specifically to an online quality inspection system for the welding station of thermos cups based on machine vision. Background Technology

[0002] Machine vision is an interdisciplinary field that integrates technologies from multiple fields such as optics, computer science, image processing, and pattern recognition. Its core goal is to enable machines to "understand" image or video information like humans and to complete tasks such as measurement, detection, recognition, and positioning based on visual data. Online quality inspection at the welding station of thermos cups refers to a technical system that performs real-time, automatic, and comprehensive quality monitoring of the welding process on the thermos cup production line. It uses various detection methods to ensure that the weld quality meets the standard requirements and prevents defective products from flowing into subsequent processes.

[0003] Patent publication number CN201611129833.3 describes in its specification that "This invention provides an online production quality inspection system and method, the system comprising: a parameter acquisition unit for acquiring process parameters of each online production device during product production; a standard parameter storage unit for storing quality indicators of qualified products; a data analysis unit for comparing and analyzing each process parameter and quality indicator to determine whether the product is qualified, generating product qualification notification information, and searching for the corresponding online production device based on process parameters that do not meet quality indicators and generating alarm information; and a notification unit for sending product qualification notification information or alarm information to predetermined notification recipients. This invention achieves quality inspection of online production by acquiring each process parameter and comparing and analyzing it with pre-stored quality indicators, and by searching for the corresponding online production device based on process parameters that do not meet quality indicators, generating and sending alarm information, thus enabling timely identification of production equipment with problems."

[0004] While existing technologies offer the advantages mentioned above, they also have disadvantages: in current thermos production, welding points involved in sealing and preventing leaks only become apparent during subsequent leak or insulation verification stages, leading to delayed detection and significant material and labor costs. Furthermore, the cylindrical metal surface of the thermos has strong reflectivity and numerous bright spots, making it easy for traditional cameras to produce overly bright or shadowed areas when photographing after welding. This makes it difficult to reliably identify minor anomalies such as incomplete welds, misaligned welds, burn-throughs, and splatter adhesion, resulting in missed or misjudged inspections. This can lead to decreased insulation performance, returns, and brand risks.

[0005] In conclusion, developing an online quality inspection system for the welding station of thermos cups based on machine vision remains a key issue that urgently needs to be addressed in the field of online quality inspection technology. Summary of the Invention

[0006] The purpose of this invention is to address the existing problems in the production of insulated cups, where welding points related to sealing and preventing air leakage are only exposed during subsequent leakage or insulation verification stages. This leads to delayed detection and significant material and labor costs. Furthermore, the cylindrical metal surface of the cup is highly reflective and has many bright spots. Traditional cameras often produce overly bright or shadowed areas when photographing after welding, making it difficult to reliably identify minor anomalies such as incomplete welds, misaligned welds, burn-throughs, and spatter adhesion. This results in missed or misjudged inspections, leading to decreased insulation performance and potential returns, exchanges, and brand risks.

[0007] To achieve the above objectives, the present invention provides an online quality inspection system for the welding station of thermos cups based on machine vision, comprising: a positioning and acquisition module, used to generate a trigger signal when the cup body reaches a preset position;

[0008] The multimodal imaging module performs image acquisition based on the trigger signal to obtain multi-angle two-dimensional images and three-dimensional contour images of the solder area;

[0009] The image processing module generates a composite image based on the multi-angle two-dimensional images using a median-based fusion algorithm.

[0010] The defect analysis module extracts multiple feature indicators from the synthesized image and the three-dimensional contour image, automatically adjusts the judgment threshold, and calculates a comprehensive risk score using a preset risk assessment model.

[0011] The control and traceability module outputs production line control signals and records the entire process data based on the comprehensive risk score.

[0012] Furthermore, the operation procedure of the positioning and acquisition module includes:

[0013] The positioning and acquisition module includes a position sensor, an industrial camera, and a trigger circuit. The position sensor detects whether the cup has entered a preset imaging position and generates a trigger signal. The trigger signal needs to undergo preprocessing based on wavelet threshold denoising, expressed as:

[0014] ,

[0015] In the formula, This is the trigger signal generated after preprocessing. and These are wavelet transform and inverse transform, respectively. A soft threshold function set for high-frequency noise. The trigger signal is a raw binary signal, and it is synchronized with the exposure time of the industrial camera.

[0016] Furthermore, the operation procedure of the multimodal imaging module includes:

[0017] The multi-angle two-dimensional image includes: at least three independent LED light sources with adjustable brightness and angle, arranged around the imaging station, for illuminating the soldering area from three different directions, and controlling the three independent LED light sources according to the trigger signal. The lights are turned on in turn according to a preset time sequence, and an industrial camera is synchronously controlled to capture three images of the solder area at different illumination angles. The camera exposes synchronously when each light source is turned on; the direction vector of the light source is set as follows. The normal vector of a point on the surface of the weld area is Its brightness A simplified model conforming to the bidirectional reflection distribution function:

[0018] ,

[0019] In the formula, Indicates the first The image brightness received by the camera sensor at a point illuminated by an LED light source. Let be the reflectivity function of the solder joint material surface. To indicate the first The light intensity produced by the LED light source at this point For ambient light brightness, and then obtain .

[0020] Furthermore, the operation procedure of the multimodal imaging module also includes:

[0021] The three-dimensional contour image includes: after obtaining the multi-angle two-dimensional image, turning off all LEDs, projecting a laser line onto the weld bead using a line laser projector, and having the high-speed camera capture the three-dimensional contour image formed by the laser line on the weld bead surface, and then applying a pre-calibrated laser plane equation. and camera intrinsic parameter matrix ,in, The equation of the plane of light projected by the laser is given. It is the normal vector of the plane. It is a constant term, representing the coordinate variables of a point in three-dimensional space, and it represents the coordinates of the image points. When projected back onto the laser plane, where, Indicates transpose. , This represents the pixel coordinates of the 3D point to be solved on the 2D image plane, thus obtaining the 3D point. ,expression:

[0022] ,

[0023] In the formula, This represents a non-zero proportionality coefficient. This indicates that the camera's extrinsic parameter matrix is ​​a 3x3 rotation matrix. Given a 3x1 translation vector, solving this system of equations yields the 3D point cloud of the laser centerline. .

[0024] Furthermore, the operation flow of the image processing module includes:

[0025] Based on the multi-angle two-dimensional images, a median-based fusion algorithm is used for each pixel position. ,calculate Local contrast at this location and local information entropy As a quality evaluation indicator, the expression is:

[0026] ,

[0027] In the formula, Indicates the first Local contrast index in a multi-source image yes , and For The standard deviation and mean within the neighborhood of the center. Indicates the first Local information entropy in a multi-source image. Indicates the first The probability of a gray level appearing in the current local area.

[0028] Furthermore, the operation flow of the image processing module includes:

[0029] Assign a fusion weight to each pixel in each image. ,expression:

[0030] ,

[0031] In the formula, and It is a normalized indicator. , To adjust parameters, synthesize images. The pixel value is:

[0032] ,

[0033] In the formula, This indicates the pixel coordinates of the final synthesized image. The brightness value at that location, Indicates the first Location in the original input image The proportion of the pixel value at a given location to the final fusion result. Indicates the first The original input image in pixel coordinates The original brightness value at that location.

[0034] Furthermore, the operation process of the defect analysis module also includes:

[0035] The multiple feature indicators include: extracting the weld bead contour through edge detection, calculating its perimeter to assess continuity, statistically analyzing whether weld rings / weld points are broken and the length of the break, obtaining information on whether the weld is broken, and setting a continuity index. For the front The proportion of the energy of each low-frequency descriptor to the total energy, expressed as:

[0036] ,

[0037] In the formula, For Fourier descriptors, The closer to 1, the better the continuity; sample the width along the normal direction of the weld bead contour, calculate its standard deviation to evaluate the width uniformity, and statistically analyze the fluctuation of the weld width to obtain information on whether it is sometimes wide and sometimes narrow; use morphological opening operation and spot analysis to detect micro-suspected leaks and spatter adhesion, and statistically analyze their area and number.

[0038] Furthermore, the operation process of the defect analysis module also includes:

[0039] The laser centerline is extracted from the 3D contour image and compared with an ideal straight line. The maximum bending distance and average offset of the laser centerline are calculated as quantitative indicators of flatness and weld misalignment. The 3D point cloud of the laser centerline is then processed. Projected onto the ideal plane fitted by the least squares method Calculate the directed distance from each point to the plane. The flatness index is quantified as the standard deviation of distance, expressed as:

[0040] ,

[0041] In the formula, It is a flatness index. Indicates the first Three-dimensional points to the fitting plane The directed distance, Represents directed distance The arithmetic mean of the welding deviation index is quantified as the average lateral offset between the three-dimensional point cloud of the laser centerline and the ideal straight line. .

[0042] Furthermore, the operation process of the defect analysis module also includes:

[0043] By calculating the reflectivity index, the expression is:

[0044] ,

[0045] In the formula, It is the reflectivity index. The average value of the top 10% of pixels in terms of brightness. The global average value is the defect determination threshold. Follow Dynamic adjustment, expression:

[0046] ,

[0047] In the formula, This represents the adaptively adjusted defect determination threshold. This represents the baseline value determined under standard lighting conditions for testing qualified products. It is the hyperbolic tangent function. It is a reflectance index calculated in real time. To adjust the coefficient, It is a preset constant, which obtains the reflectivity and automatically adjusts the judgment threshold; the preset risk assessment model calculates the comprehensive risk score, including: establishing a two-level fuzzy comprehensive evaluation model, the first level being the index layer, which transforms the six indicators extracted above—continuity, width uniformity, suspected leaks, spatter adhesion, flatness, and weld deviation—into a fuzzy comment set through a membership function. Membership vector on ,in, Indicates the first The first level is the membership vector of each indicator; the second level is the factor layer, which assigns a weight vector to each group of indicators. ,in, Representing the The importance of each indicator in the final evaluation is determined by calculating the comprehensive evaluation vector through fuzzy synthesis. ,in, It is a fuzzy relation matrix. It is a fuzzy synthesis operator. Then, the comprehensive evaluation vector Defuzzing was performed using a weighted average method, mapping the result to a comprehensive risk score of 0-100. The lower the score, the higher the risk.

[0048] Furthermore, the operation process of the control and traceability module also includes:

[0049] Based on the comprehensive risk score, output production line control signals and record full-process data, including: setting a preset release threshold. and minimum risk tolerance threshold ,exist When this happens, the output production line control signal is "release"; in When this happens, the output production line control signal is "rework," and it also outputs the defect type and the circumferential angle position calculated based on image coordinates. When this happens, the output production line control signal is "reject".

[0050] By generating a globally unique traceability code The multi-angle two-dimensional images, three-dimensional contour images, composite images, continuity, width uniformity, suspected leaks, spatter adhesion, flatness, weld deviation, comprehensive risk score, production line control signals, product ID, and timestamp are encapsulated in key-value pairs into a data packet and written to a time-series database and a relational database through asynchronous transactions.

[0051] Beneficial effects

[0052] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects:

[0053] In use, this invention uses an image processing module and a defect analysis module to fuse multi-angle two-dimensional and three-dimensional contour images of the weld area to suppress bright spots. Small gaps are no longer covered by reflections, reducing false alarms and allowing for the early detection of small but potentially leaky hazards. The superimposed fine line light amplifies the unevenness, facilitating the detection of early risks. By moving leak risks that would otherwise only be exposed later to the welding station, this invention helps reduce rework, scrap, and wasted time. Through the control and traceability module, after a leak risk is detected, it can indicate which section of the cup body needs to be re-welded, reducing blind re-welding and repeated trial welding, and lowering the probability of line stoppage. Attached Figure Description

[0054] Figure 1 This is a system diagram of an online quality inspection system for the welding station of thermos cups based on machine vision, according to the present invention. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0056] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0057] The present invention will now be described in further detail with reference to the accompanying drawings:

[0058] Example 1:

[0059] like Figure 1 As shown, the present invention provides an online quality inspection system for the welding station of thermos cups based on machine vision, including: a positioning and acquisition module, used to generate a trigger signal when the cup body reaches a preset position;

[0060] The multimodal imaging module performs image acquisition based on the trigger signal to obtain multi-angle two-dimensional images and three-dimensional contour images of the solder area;

[0061] The image processing module generates a composite image based on the multi-angle two-dimensional images using a median-based fusion algorithm.

[0062] The defect analysis module extracts multiple feature indicators from the synthesized image and the three-dimensional contour image, automatically adjusts the judgment threshold, and calculates a comprehensive risk score using a preset risk assessment model.

[0063] The control and traceability module outputs production line control signals and records full-process data based on the comprehensive risk score.

[0064] Specifically, this invention is integrated into an automatic welding production line for thermos cups, located after the sealing welding station. After welding, the thermos cup is conveyed by a conveyor belt to a rotating fixture and clamped. The rotating table is started, driving the cup body to rotate at a constant speed. When a specific mark on the cup body passes a preset position, the position acquisition module detects the signal change through the positioning sensor, generating a raw binary transition signal. First, the signal is preprocessed based on wavelet threshold denoising to filter out electromagnetic interference. Then, its rising edge slope and high-level duration are verified. After confirming that it is a valid trigger, a standard synchronization pulse is generated and sent to a 5-megapixel color area array industrial camera equipped with an anti-glare lens, controlling it to expose at a stable moment when the cup body vibration is minimal.

[0065] The multimodal imaging module sequentially illuminates three ring-shaped LED light sources according to a preset timing sequence. When each light source is lit, the area array industrial camera simultaneously exposes once, acquiring three images of the welding area under different lighting conditions. Then, all LED light sources are turned off, the line laser projector is turned on, and a laser line is projected onto the weld bead. The high-speed camera captures the three-dimensional contour image of the laser line on the weld bead surface at a high frame rate.

[0066] The three two-dimensional images are fused using an image processing module, and for each pixel... The local contrast and local information entropy of the pixel are calculated in each of the three images; based on this, the fusion weight of the pixel in each image is calculated. The final pixel value of the synthesized image effectively suppressed overexposed bright spots in a single image;

[0067] The defect analysis module is used in the synthesized image. Above, the weld contour is extracted through edge detection, its Fourier descriptor is calculated, and then... Low-frequency energy proportion Assess continuity; sample width along the contour normal and calculate its coefficient of variation to assess uniformity; detect and statistically analyze the number and area of ​​suspected leaks and splashes through morphological operations and spot analysis;

[0068] The 3D contour image is processed by the defect analysis module, based on the calibrated laser plane equation. and camera intrinsic parameter matrix Image points Back-projecting to three-dimensional space, solve the system of equations. A three-dimensional point cloud characterizing the weld cross-section is obtained. ;

[0069] The point cloud is projected onto the fitting plane using the defect analysis module, and the directed distance of each point is calculated. Standard deviation as a flatness index And calculate the average lateral offset between the point cloud centerline and the theoretical centerline. As an indicator of weld deviation;

[0070] The reflection intensity index of the current fused image is calculated using the defect analysis module. And adjust the defect judgment threshold dynamically accordingly. Finally, the six extracted indicators—continuity, uniformity, leaks, spatter, flatness, and weld misalignment—are used to generate a comprehensive risk score of 0-100 using a risk assessment model. ;

[0071] Through the control and traceability module Compared with a preset threshold: If the risk level is low, a release signal is sent to the PLC; If the risk level is medium, a repair signal and the location and angle of the defect will be sent. ;exist If the risk level is high, a rejection signal is triggered; at the same time, a globally unique traceability code (UUID) is generated, binding all images, data, and results within this detection cycle and storing them in the database.

[0072] Example 2:

[0073] like Figure 1 As shown, the present invention provides an online quality inspection system for the welding station of thermos cups based on machine vision, including: a positioning and acquisition module, used to generate a trigger signal when the cup body reaches a preset position;

[0074] The multimodal imaging module performs image acquisition based on the trigger signal to obtain multi-angle two-dimensional images and three-dimensional contour images of the solder area;

[0075] The image processing module generates a composite image based on the multi-angle two-dimensional images using a median-based fusion algorithm.

[0076] The defect analysis module extracts multiple feature indicators from the synthesized image and the three-dimensional contour image, automatically adjusts the judgment threshold, and calculates a comprehensive risk score using a preset risk assessment model.

[0077] The control and traceability module outputs production line control signals and records full-process data based on the comprehensive risk score.

[0078] Specifically, based on Example 1, optimizations are made for high-speed production lines with a production cycle time of less than 3 seconds, with a focus on improving the real-time performance of image acquisition and processing;

[0079] Replace the area scan camera with a high-resolution line scan camera and synchronize it with the pulse signal of the high-precision servo motor encoder. When the cup rotates at a constant speed on the turntable, the line scan camera will capture a line of images for each pulse emitted by the encoder. By continuously scanning, a panoramic unfolded image of the weld is synthesized, which helps to avoid the global motion blur that exists at the moment of shooting by the area scan camera and is conducive to obtaining higher circumferential resolution.

[0080] The software processing steps adopt a multi-threaded parallel pipeline architecture; Thread 1 is responsible for controlling triggering and image acquisition; Thread 2 is responsible for real-time fusion calculation of the acquired multi-angle 2D images; Thread 3 is responsible for 3D reconstruction of 3D contour images; Thread 4 is responsible for extraction of multiple feature indicators and comprehensive scoring; the threads exchange data through shared memory, which is conducive to completing all processing within a single production cycle.

[0081] The complex weight calculation based on local information entropy and contrast is replaced with a more efficient median algorithm. That is, for each pixel, the three images are directly compared. The brightness value at that point is discarded, and the median value is used as the fused image. The method has low computational cost and can still effectively suppress extreme bright or dark spots;

[0082] By deploying the core matrix operations in 3D point cloud reconstruction to the graphics processor of an industrial control computer for parallel computing, the 3D reconstruction time can be reduced by more than 60%. While ensuring detection accuracy, the processing speed of this invention is further improved, making it suitable for high-volume, high-rate-of-production scenarios.

[0083] Example 3:

[0084] like Figure 1 As shown, the present invention provides an online quality inspection system for the welding station of thermos cups based on machine vision, including: a positioning and acquisition module, used to generate a trigger signal when the cup body reaches a preset position;

[0085] The multimodal imaging module performs image acquisition based on the trigger signal to obtain multi-angle two-dimensional images and three-dimensional contour images of the solder area;

[0086] The image processing module generates a composite image based on the multi-angle two-dimensional images using a median-based fusion algorithm.

[0087] The defect analysis module extracts multiple feature indicators from the synthesized image and the three-dimensional contour image, automatically adjusts the judgment threshold, and calculates a comprehensive risk score using a preset risk assessment model.

[0088] The control and traceability module outputs production line control signals and records full-process data based on the comprehensive risk score.

[0089] Specifically, based on Embodiment 1, a machine learning model is introduced, enabling the invention to learn and optimize from historical data, further improving the accuracy and adaptability of detection. By recording the original image, fusion weight distribution, final defect determination, and its true category for each detection (the true category is confirmed through subsequent manual review or destructive testing), a lightweight neural network model is periodically trained using this data. This neural network model uses three currently captured original images... Using local image patches as input, the system directly outputs the optimal fusion weights for the current scene. , replacing fixed based and The weight calculation formula is beneficial to improving the adaptability of the fusion strategy to complex and ever-changing reflection patterns;

[0090] By analyzing long-term production data and the airtightness test results of the final product, a correlation model is established, and the weight vector in the risk assessment model is automatically adjusted periodically. And the parameters in the adaptive threshold adjustment formula for reflectivity , No longer a fixed value; conducive to enhancing the overall risk score. The accuracy of predicting the actual risk of product leakage;

[0091] The detection of suspected leaks and splashes is upgraded from a rule-based traditional image processing algorithm to a micro-defect classification model based on a convolutional neural network. This micro-defect classification model is trained on a large number of labeled samples, which makes it easier to accurately distinguish real leaks and splashes from harmless image noise or texture, and helps to significantly reduce the false alarm rate.

[0092] The traceability and recording module not only stores the test results, but also continuously monitors the long-term changes of various intermediate indicators. When a certain indicator is found to show a slow drift trend, it issues an early warning to the maintenance personnel, indicating possible wear of the welding head or positioning deviation of the fixture, thus realizing predictive maintenance.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online quality inspection system for the welding station of thermos cups based on machine vision, characterized in that, include: The positioning and acquisition module is used to generate a trigger signal when the cup reaches the preset position; The multimodal imaging module performs image acquisition based on the trigger signal to obtain multi-angle two-dimensional images and three-dimensional contour images of the solder area; The image processing module generates a composite image based on the multi-angle two-dimensional images using a median-based fusion algorithm. The defect analysis module extracts multiple feature indicators from the synthesized image and the three-dimensional contour image, automatically adjusts the judgment threshold, and calculates a comprehensive risk score using a preset risk assessment model. The control and traceability module outputs production line control signals and records the entire process data based on the comprehensive risk score.

2. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 1, characterized in that, The operation process of the positioning and acquisition module includes: The positioning and acquisition module includes a position sensor, an industrial camera, and a trigger circuit. The position sensor detects whether the cup has entered a preset imaging position and generates a trigger signal. The trigger signal needs to undergo preprocessing based on wavelet threshold denoising, expressed as: ， In the formula, This is the trigger signal generated after preprocessing. and These are wavelet transform and inverse transform, respectively. A soft threshold function set for high-frequency noise. The trigger signal is a raw binary signal, and it is synchronized with the exposure time of the industrial camera.

3. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 2, characterized in that, The operation procedure of the multimodal imaging module includes: The multi-angle two-dimensional image includes: at least three independent LED light sources with adjustable brightness and angle, arranged around the imaging station, for illuminating the soldering area from three different directions, and controlling the three independent LED light sources according to the trigger signal. The lights are turned on in turn according to a preset time sequence, and an industrial camera is synchronously controlled to capture three images of the solder area at different illumination angles. The camera exposes synchronously when each light source is turned on; the direction vector of the light source is set as follows. The normal vector of a point on the surface of the weld area is Its brightness A simplified model conforming to the bidirectional reflection distribution function: ， In the formula, Indicates the first The image brightness received by the camera sensor at a point illuminated by an LED light source. Let be the reflectivity function of the solder joint material surface. To indicate the first The light intensity produced by the LED light source at this point For ambient light brightness, and then obtain .

4. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 3, characterized in that, The operation procedure of the multimodal imaging module also includes: The three-dimensional contour image includes: after obtaining the multi-angle two-dimensional image, turning off all LEDs, projecting a laser line onto the weld bead using a line laser projector, and having the high-speed camera capture the three-dimensional contour image formed by the laser line on the weld bead surface, and then applying a pre-calibrated laser plane equation. and camera intrinsic parameter matrix ,in, The equation of the plane of light projected by the laser is given. It is the normal vector of the plane. It is a constant term, representing the coordinate variables of a point in three-dimensional space, and it represents the coordinates of the image points. When projected back onto the laser plane, where, Indicates transpose. , This represents the pixel coordinates of the 3D point to be solved on the 2D image plane, thus obtaining the 3D point. ,expression: In the formula, This represents a non-zero proportionality coefficient. This indicates that the extrinsic parameter matrix of the camera is a 3x3 rotation matrix. Given a 3x1 translation vector, solving this system of equations yields the 3D point cloud of the laser centerline. .

5. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 4, characterized in that, The operation flow of the image processing module includes: Based on the multi-angle two-dimensional images, a median-based fusion algorithm is used for each pixel position. ,calculate Local contrast at this location and local information entropy As a quality evaluation indicator, the expression is: ， In the formula, Indicates the first Local contrast index in a multi-source image yes , and For The standard deviation and mean within the neighborhood of the center. Indicates the first Local information entropy in a multi-source image. Indicates the first The probability of a gray level appearing in the current local area.

6. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 5, characterized in that, The operation flow of the image processing module includes: Assign a fusion weight to each pixel in each image. ,expression: ， In the formula, and It is a normalized indicator. , To adjust parameters, synthesize images. The pixel value is: ， In the formula, This indicates the pixel coordinates of the final synthesized image. The brightness value at that location, Indicates the first Location in the original input image The proportion of the pixel value at a given location to the final fusion result. Indicates the first The original input image in pixel coordinates The original brightness value at that location.

7. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 6, characterized in that, The operation process of the defect analysis module also includes: The multiple feature indicators include: extracting the weld bead contour through edge detection, calculating its perimeter to assess continuity, statistically analyzing whether weld rings / weld points are broken and the length of the break, obtaining information on whether the weld is broken, and setting a continuity index. For the front The proportion of the energy of each low-frequency descriptor to the total energy, expressed as: ， In the formula, For Fourier descriptors, The closer to 1, the better the continuity; sample the width along the normal direction of the weld bead contour, calculate its standard deviation to evaluate the width uniformity, and statistically analyze the fluctuation of the weld width to obtain information on whether it is sometimes wide and sometimes narrow; use morphological opening operation and spot analysis to detect micro-suspected leaks and spatter adhesion, and statistically analyze their area and number.

8. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 7, characterized in that, The operation process of the defect analysis module also includes: The laser centerline is extracted from the 3D contour image and compared with an ideal straight line. The maximum bending distance and average offset of the laser centerline are calculated as quantitative indicators of flatness and weld misalignment. The 3D point cloud of the laser centerline is then processed. Projected onto the ideal plane fitted by the least squares method Calculate the directed distance from each point to the plane. The flatness index is quantified as the standard deviation of distance, expressed as: ， In the formula, It is a flatness index. Indicates the first Three-dimensional points to the fitting plane The directed distance, Represents directed distance The arithmetic mean of the welding deviation index is quantified as the average lateral offset between the three-dimensional point cloud of the laser centerline and the ideal straight line. .

9. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 8, characterized in that, The operation process of the defect analysis module also includes: By calculating the reflectivity index, the expression is: ， In the formula, It is the reflectivity index. The average value of the top 10% of pixels in terms of brightness. The global average value is the defect determination threshold. Follow Dynamic adjustment, expression: ， In the formula, This represents the adaptively adjusted defect determination threshold. This represents the baseline value determined under standard lighting conditions for testing qualified products. It is the hyperbolic tangent function. It is a reflectance index calculated in real time. To adjust the coefficient, It is a preset constant, which obtains the reflectivity and automatically adjusts the judgment threshold; the preset risk assessment model calculates the comprehensive risk score, including: establishing a two-level fuzzy comprehensive evaluation model, the first level being the index layer, which transforms the six indicators extracted above—continuity, width uniformity, suspected leaks, spatter adhesion, flatness, and weld deviation—into a fuzzy comment set through a membership function. Membership vector on ,in, Indicates the first The first level is the membership vector of each indicator; the second level is the factor layer, which assigns a weight vector to each group of indicators. ,in, Representing the The importance of each indicator in the final evaluation is determined by calculating the comprehensive evaluation vector through fuzzy synthesis. ,in, It is a fuzzy relation matrix. It is a fuzzy synthesis operator. Then, the comprehensive evaluation vector Defuzzing was performed using a weighted average method, mapping the result to a comprehensive risk score of 0-100. The lower the score, the higher the risk.

10. The online quality inspection system for the welding station of thermos cups based on machine vision according to claim 9, characterized in that, The operation process of the control and traceability module also includes: Based on the comprehensive risk score, output production line control signals and record full-process data, including: setting a preset release threshold. and minimum risk tolerance threshold ,exist When this happens, the output production line control signal is "release"; in When this happens, the output production line control signal is "rework," and it also outputs the defect type and the circumferential angle position calculated based on image coordinates. When this happens, the output production line control signal is "reject". By generating a globally unique traceability code The multi-angle two-dimensional images, three-dimensional contour images, composite images, continuity, width uniformity, suspected leaks, spatter adhesion, flatness, weld deviation, comprehensive risk score, production line control signals, product ID, and timestamp are encapsulated in key-value pairs into a data packet and written to a time-series database and a relational database through asynchronous transactions.

Images