Method and system for detecting the quality of a connection strip bending based on machine vision

Abstract

The present application relates to the technical field of image data processing, in particular to a connecting row bending quality detection method and system based on machine vision, comprising: acquiring binocular images of a connecting row to be detected; performing edge detection on the binocular images to obtain a plurality of edge contours and acquire three-dimensional point clouds of each edge contour; classifying the three-dimensional point clouds based on a dynamic curvature threshold, determining that a dynamic curvature threshold greater than or equal to a set curvature threshold as an arc segment point cloud and less than the set curvature threshold as a straight line segment point cloud; fitting the straight line segment point cloud to obtain a bending angle of the connecting row; fitting the arc segment point cloud to obtain an initial bending radius of the connecting row; acquiring a final bending radius, and if the bending angle and the final bending radius are within their respective error ranges, determining that the connecting row to be detected is of normal quality. The present application solves the problem of low detection accuracy in the prior art.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to a machine vision-based method and system for detecting the bending quality of connected panels. Background Technology

[0002] Connecting busbars, as key components in electrical equipment, play a vital role in various fields such as automotive electronics, industrial automation control, home appliances, and power distribution systems. They not only transmit electrical signals and connect current, but also play a crucial role in ensuring assembly accuracy, electrical safety, and long-term stable operation of equipment throughout the system. During the production of connecting busbars, the quality of their bending and forming directly determines whether their geometry meets design requirements, thus affecting their positioning accuracy and fit reliability in subsequent assembly stages. Deviations in bending angle, bending radius, or bending transition area can lead to assembly difficulties, poor contact, degraded electrical performance, and even safety hazards. Therefore, achieving real-time, accurate, and non-contact inspection of bending and forming quality during the manufacturing process of connecting busbars has significant engineering importance and practical value. This significantly improves production efficiency, ensures product quality stability, and reduces production costs. Simultaneously, with the development of intelligent manufacturing and industrial automation, higher demands are placed on the accuracy, response speed, and intelligence level of inspection methods, urgently requiring the introduction of advanced inspection technologies and intelligent analysis methods to achieve comprehensive, accurate, and dynamic monitoring of connecting busbar bending quality.

[0003] Traditional inspection methods mostly rely on manual inspection or contact gauge measurement. The former suffers from low inspection efficiency, high subjectivity, and poor stability, while the latter struggles to achieve rapid and comprehensive inspection of complex geometries and is prone to damaging the surface of the workpiece. With the development of production automation and intelligence, machine vision-based inspection methods are gradually becoming an important means of bending quality inspection. By acquiring connected row images and 3D point cloud data through industrial cameras, binocular vision, or laser sensors, and combining edge detection, fitting calculation, and error analysis, high-precision measurement of bending angles and bending radii can be achieved without contact.

[0004] However, existing methods still have certain limitations in practical applications. Current machine vision-based methods for inspecting the bending quality of connected panels are easily affected by factors such as changes in the acquisition environment, material thickness differences, external forces, and temperature. Furthermore, they often rely on fixed thresholds and static detection models, lacking adaptive adjustment and intelligent optimization mechanisms. This results in the detection accuracy and robustness of bending angles and radii failing to meet the requirements of high-precision, high-consistency production, ultimately leading to low detection accuracy. Summary of the Invention

[0005] To address the problem of low detection accuracy mentioned in the background art, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a machine vision-based method for detecting the bending quality of a connector strip, comprising: acquiring a binocular image of the connector strip to be detected; performing edge detection on the binocular image to obtain multiple edge contours, and acquiring a three-dimensional point cloud of each edge contour; classifying the three-dimensional point cloud based on a dynamic curvature threshold, classifying those with a dynamic curvature threshold greater than or equal to a set curvature threshold as arc segment point clouds, and those with a curvature threshold less than the set curvature threshold as straight line segment point clouds; the dynamic curvature threshold is determined by the curvature and local geometric entropy of the three-dimensional point cloud of each edge contour. The Hausdorff distance between the line segment and the standard contour 3D point cloud is normalized and then weighted and summed according to preset weights. The point cloud of the line segment is fitted to obtain the bending angle of the connecting row. The point cloud of the arc segment is fitted to obtain the initial bending radius of the connecting row. The final bending radius is obtained. The final bending radius is positively correlated with the initial bending radius and deformation force of the connecting row, and negatively correlated with the bending angle and thickness of the connecting row. If the bending angle and the final bending radius are within their respective error ranges, the connecting row to be tested is determined to be of normal quality.

[0007] The above technical solution combines binocular vision to acquire image information with 3D point cloud reconstruction, and uses dynamic curvature threshold to intelligently classify the point cloud, achieving accurate differentiation between straight line segments and arc segments. Based on this, a fitting method is used to obtain the bending angle and initial bending radius, and then the final bending radius is dynamically calculated by combining material deformation force, bending angle and thickness factors. This forms an adaptive, multi-factor coupled bending quality detection mechanism. This mechanism not only significantly improves the accuracy and stability of bending parameter measurement, but also reliably evaluates the bending quality of connecting bars under different processing conditions, effectively ensuring the accuracy and practicality of the detection results.

[0008] Furthermore, the bending angle is specifically defined as follows: the point cloud of the straight line segment is fitted using the least squares fitting method to obtain two fitted straight lines, and the angle between the two fitted straight lines is the bending angle.

[0009] The above technical solution obtains two fitted straight lines by using the least squares fitting method on the point cloud of the straight line segment, and uses the angle between them as the bending angle, so that the bending angle can accurately reflect the actual bending geometry of the connecting row, thereby improving the accuracy and stability of the bending angle measurement.

[0010] Furthermore, the initial bending radius is specifically determined by fitting the point cloud of the arc segment using the least squares fitting method to determine the radius parameter, which is the initial bending radius.

[0011] The above technical solution calculates the bending radius by using the least squares fitting method on the point cloud of the arc segment, so that the initial bending radius can accurately reflect the actual geometry of the connecting row, thereby improving the accuracy and reliability of bending parameter measurement.

[0012] Furthermore, the final bending radius for, , The initial bending radius, For the deformation force of the connecting row, The maximum elastic strain of the connecting row. For the bending angle, For standard connection bar thickness, The thickness of the connecting strip to be tested. , These are dynamic hyperparameters, obtained through a reinforcement learning algorithm.

[0013] The above technical solution introduces a dynamic correction model that considers material deformation force, elastic properties, bending angle, and thickness differences based on the initial bending radius, and combines reinforcement learning algorithms to adaptively optimize the correction parameters. This enables intelligent adjustment of the bending radius under different processing conditions, thereby more accurately reflecting the actual bending state of the connecting strip and improving the accuracy and robustness of bending quality measurement.

[0014] Furthermore, it also includes correcting the bending angle, the corrected bending angle... for, , For the bending angle, To preset hyperparameters, The real-time ambient temperature is used to obtain binocular images of the connected row to be detected.

[0015] The above technical solution introduces an ambient temperature correction mechanism, which enables the measurement of bending angle to automatically compensate for deviations caused by changes in the ambient temperature. This ensures that accurate and consistent bending angle data can be obtained under different temperature conditions, significantly improving the stability and reliability of the test results. It effectively avoids measurement errors caused by temperature fluctuations and ensures the accuracy of bending quality assessment and subsequent assembly.

[0016] Furthermore, the bending angle error range is ±0.5, and the final bending radius error range is ±0.1.

[0017] Furthermore, the preset weights are obtained through a reinforcement learning algorithm, which employs a weight optimization method based on policy gradients.

[0018] Furthermore, the edge detection is Canny edge detection.

[0019] Furthermore, the binocular image of the connection row to be detected is subjected to grayscale conversion and median filtering.

[0020] In a second aspect, the present invention provides a machine vision-based connection row bending quality inspection system, including a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the machine vision-based connection row bending quality inspection method described above is implemented.

[0021] The beneficial effects of this invention are as follows: This invention combines machine vision with 3D point cloud analysis to achieve high-precision, non-contact inspection of the bending quality of connecting bars. It introduces a dynamic curvature threshold to intelligently classify point clouds of straight and circular segments, making the measurement of bending angles and radii more accurate and stable. Simultaneously, it obtains initial bending parameters through a least-squares fitting method and optimizes these parameters using a multi-factor dynamic correction model and reinforcement learning algorithms. This allows the final bending radius and angle to adapt to different materials, thicknesses, and environmental conditions, improving the robustness and intelligence of the inspection. Furthermore, a bending angle correction mechanism is introduced to address environmental temperature, and a reasonable error range is set, enabling reliable judgment of bending quality. This significantly improves the accuracy, stability, and application value of the inspection system in actual production processes, while reducing the risk of human intervention and misjudgment, demonstrating high automation and practicality. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a machine vision-based method for detecting the bending quality of connected rows according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the structural block diagram of a machine vision-based connected row bending quality inspection system according to an embodiment of the present invention. Detailed Implementation

[0023] Example of a machine vision-based method for inspecting the bending quality of connected rows.

[0024] like Figure 1 As shown in the flowchart of the machine vision-based connection row bending quality inspection method of the present invention, the method includes the following steps: S1: Obtain a binocular image of the connected row to be detected; perform edge detection on the binocular image to obtain multiple edge contours, and obtain the three-dimensional point cloud of each edge contour; In a preferred embodiment, binocular images of the connected row to be detected are acquired. Since binocular imaging can recover the depth information of the object's surface based on the parallax principle, it has higher reliability and accuracy in terms of stereoscopic perception and spatial geometry reconstruction compared to monocular imaging. By simultaneously acquiring images of the connected row from two perspectives using binocular cameras, the problem of missing image information or incomplete spatial features caused by a single perspective can be effectively avoided, thereby ensuring the integrity and accuracy of subsequent 3D point cloud generation. After acquiring the binocular images, preprocessing is preferably performed on them. Specifically, the binocular images are first converted to grayscale. Grayscale conversion transforms a color image into a single-channel image, reducing unnecessary color redundancy and helping to highlight the geometric features and edge information of the connected rows to be detected, thereby simplifying subsequent computational complexity and improving the stability of edge detection. Furthermore, a median filtering algorithm is used to denoise the grayscale images. Median filtering is a non-linear filtering method that effectively suppresses impulse noise and salt-and-pepper noise, while also having a significant advantage in preserving image edge details. This processing method not only improves the smoothness of the image but also ensures that the edge structure of the connected rows maintains high clarity after denoising, thereby improving the robustness and accuracy of subsequent edge detection.

[0025] After image preprocessing, edge detection is performed. This embodiment preferably uses the Canny edge detection algorithm to obtain multiple edge contours. After edge detection, 3D point cloud reconstruction is performed on the extracted edge contours based on the binocular images. Specifically, by calculating the disparity information of corresponding pixels in the binocular images and combining it with the imaging geometric parameters of the binocular camera, the 3D coordinates of the edge contours in space can be recovered, thus obtaining 3D point cloud data of multiple edge contours. This point cloud data can realistically reflect the geometric shape and spatial characteristics of the connecting strip surface, providing a reliable data foundation for subsequent defect identification and dimensional measurement.

[0026] S2: Classify the three-dimensional point cloud based on the dynamic curvature threshold, and determine that the point cloud with the dynamic curvature threshold greater than or equal to the set curvature threshold is the arc segment point cloud, and the point cloud with the dynamic curvature threshold less than the set curvature threshold is the straight line segment point cloud.

[0027] In a preferred embodiment, the dynamic curvature threshold is obtained by normalizing the curvature, local geometric entropy, and Hausdorff distance of each edge contour 3D point cloud and then summing them by weighted calculation according to a preset weight; the preset weight is obtained by a reinforcement learning algorithm, which adopts a weight optimization method based on policy gradient.

[0028] Specifically, the curvature characteristics of the acquired 3D point clouds of each edge contour are calculated first. Curvature reflects the degree of bending of the point cloud surface in a local area. High curvature values ​​usually correspond to the turning points or arc areas of the contour, while low curvature values ​​mostly correspond to straight lines. Therefore, the introduction of curvature can intuitively distinguish the geometric morphological features of the point cloud, providing a basic basis for subsequent point cloud classification.

[0029] Secondly, the local geometric entropy of each edge contour point cloud is calculated. Geometric entropy measures the complexity and uncertainty of point cloud distribution. A high geometric entropy value in a local area usually indicates that the point cloud distribution is relatively dispersed in space or has a complex structure, while a low geometric entropy value indicates that the point cloud distribution is relatively regular and uniform. By calculating geometric entropy, we can help judge the differences in local structure of the point cloud and avoid misjudgments caused by relying solely on curvature. For example, in cases where the local curvature is similar, geometric entropy can provide additional discriminative information, thereby improving the robustness and accuracy of point cloud classification.

[0030] Furthermore, this invention also calculates the Hausdorff distance between the edge contour point cloud and the standard contour 3D point cloud. The Hausdorff distance reflects the maximum deviation between two point sets in geometric space and is an important indicator for measuring point cloud similarity and contour deviation. By introducing the Hausdorff distance, not only can the overall deviation of the connected contour relative to the standard part be detected, but it can also provide a quantitative basis for the dynamic adjustment of the curvature threshold, making the threshold determination more consistent with the actual processing conditions and characteristics of the standard part.

[0031] To achieve effective fusion of the aforementioned multiple features, this invention normalizes curvature, geometric entropy, and Hausdorff distance to eliminate differences in numerical magnitude among different features and ensure comparability during the fusion process. Subsequently, each feature is weighted and summed based on preset weights to obtain a dynamic curvature threshold. This dynamic curvature threshold can adaptively adjust with changes in point cloud features, significantly improving the flexibility and accuracy of point cloud classification compared to traditional methods with fixed thresholds.

[0032] In determining the weights, this invention employs an adaptive optimization mechanism based on reinforcement learning. Specifically, a policy gradient-based reinforcement learning algorithm is used to dynamically adjust the weight parameters of each feature. The policy gradient method can continuously update the policy parameters to maximize the expected value of the reward function, thereby gradually approaching the optimal weight allocation. The reward function can be designed based on a comprehensive consideration of the accuracy of point cloud classification, the continuity of contour segmentation, and the accuracy of matching with standard parts. When the point cloud classification result is highly consistent with the expected judgment, the reward value is higher; when the result deviates significantly, the reward value is lowered. Through this adaptive learning mechanism, the system can continuously optimize the feature weight allocation during actual operation, gradually improving the calculation accuracy and judgment effect of the dynamic curvature threshold.

[0033] After calculating the dynamic curvature threshold, the point cloud is classified based on this threshold. Specifically, when the dynamic curvature threshold of a certain edge contour point cloud is greater than or equal to the set curvature threshold, it is classified as a circular arc segment point cloud; conversely, when the dynamic curvature threshold is less than the set threshold, it is classified as a straight line segment point cloud. This classification method can effectively distinguish between circular arc segments and straight line segments in the connecting rows, laying a data foundation for subsequent bending radius extraction, geometric error detection, and structural anomaly identification.

[0034] S3: Fit the point cloud of the straight line segment to obtain the bending angle of the connecting row; fit the point cloud of the arc segment to obtain the initial bending radius of the connecting row; obtain the final bending radius.

[0035] In a preferred embodiment, the bending angle is specifically achieved by fitting the point cloud of the straight line segment using the least squares fitting method to obtain two fitted straight lines, and the angle between the two fitted straight lines is the bending angle; the least squares fitting method is existing technology and will not be described in detail in this solution.

[0036] The initial bending radius is specifically determined by fitting the point cloud of the arc segment using the least squares fitting method to determine the radius parameter, which is the initial bending radius. The final bending radius... for, , The initial bending radius, For the deformation force of the connecting row, The maximum elastic strain of the connecting row. For the bending angle, For standard connection bar thickness, The thickness of the connecting strip to be tested. , These are dynamic hyperparameters, obtained through a reinforcement learning algorithm.

[0037] By combining fitting calculations with adaptive corrections on the point cloud of the arc segment, a more accurate initial bending radius can be obtained. Furthermore, considering multi-dimensional factors such as material deformation force, elastic properties, bending angle, and thickness differences, a final bending radius model that can dynamically adjust with changing working conditions is constructed, effectively improving the accuracy and robustness of bending detection. In particular, the introduction of a dynamic hyperparameter optimization mechanism through reinforcement learning algorithms allows the bending radius correction process to tend towards the optimal value through continuous learning, avoiding the insufficient adaptability caused by the reliance on fixed coefficients in traditional empirical formulas. This achieves accurate characterization and intelligent prediction of the bending characteristics of connecting bars under different processing conditions, significantly improving the reliability and practicality of the detection results.

[0038] It also includes correcting the bending angle, the corrected bending angle for, , For the bending angle, To preset hyperparameters, The real-time ambient temperature is used to obtain binocular images of the connected row to be detected.

[0039] By introducing a temperature correction mechanism into the calculation of bending angle, the deviation caused by changes in ambient temperature can be effectively compensated, so that high consistency and stability can be maintained under different testing environments. This improves the accuracy and reliability of bending angle measurement, avoids the accumulation of errors caused by temperature fluctuations, and ensures that the test results are closer to the actual bending state under real working conditions.

[0040] S4: If the bending angle and the final bending radius are within their respective error ranges, then the connection to be tested is determined to be of normal quality.

[0041] In a preferred embodiment, the bending angle error range is ±0.5, and the final bending radius error range is ±0.1. This setting enables high-precision control of the bending geometry. The bending angle error range ensures that the connecting strip maintains the same bending angle characteristics as the standard part during the forming process, thereby avoiding assembly difficulties or abnormal electrical performance caused by excessive angle deviation. On the other hand, the final bending radius error range ensures the smoothness of the bending transition and the reasonable distribution of structural strength, effectively preventing local stress concentration or fatigue damage caused by radius deviation.

[0042] Furthermore, the error range set in this embodiment is not a fixed value, but can be adjusted according to different process requirements and application scenarios. For example, in the assembly of high-precision electronic devices, the error range can be appropriately reduced to ensure strict consistency of the geometric parameters of the connector; while in some application scenarios with larger tolerances, the error range can be widened to improve production efficiency and yield. This adjustability makes the detection method of the present invention highly adaptable and versatile.

[0043] This embodiment improves the accuracy and rigor of inspection by constraining both the bending angle and the final bending radius, while avoiding the risk of misjudgment that may arise from judging based on a single parameter. Furthermore, by reasonably setting the error range, this invention can ensure both inspection accuracy and production efficiency, achieving standardization and intelligentization of the inspection process, ultimately enhancing the quality control level and stability of the connecting strip production process.

[0044] The present invention combines binocular vision image acquisition with 3D point cloud reconstruction, uses dynamic curvature thresholds to intelligently classify point clouds, and calculates bending angles and initial bending radii for straight and circular segments respectively. Simultaneously, it dynamically calculates the final bending radius by incorporating multiple factors such as material deformation force, bending angle, and thickness, and introduces environmental temperature correction and reinforcement learning optimization mechanisms. This achieves adaptive, intelligent, and high-precision control of bending parameter measurement, enabling real-time, accurate, and reliable evaluation of the bending quality of connecting bars under different working conditions. This effectively improves the stability, robustness, and practicality of the detection, providing a scientific and reliable technical means for quality control in the production process.

[0045] Example of a machine vision-based connection row bending quality inspection system: like Figure 2 As shown in the figure, the structural block diagram of the machine vision-based connected row bending quality inspection system of the present invention includes a processor and a memory.

[0046] This invention also provides a machine vision-based system for inspecting the bending quality of connected panels. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the machine vision-based connection row bending quality detection method according to the present invention.

[0047] The machine vision-based connection row bending quality inspection system also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0048] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0049] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0050] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A machine vision-based method for inspecting the bending quality of connected rows, characterized in that, include: Acquire binocular images of the connected rows to be detected; Edge detection is performed on the binocular image to obtain multiple edge contours, and the three-dimensional point cloud of each edge contour is obtained. The three-dimensional point cloud is classified based on a dynamic curvature threshold. Points with a dynamic curvature threshold greater than or equal to a set curvature threshold are classified as arc segment point clouds, while those with a dynamic curvature threshold less than the set curvature threshold are classified as straight line segment point clouds. The dynamic curvature threshold is obtained by normalizing the curvature, local geometric entropy, and Hausdorff distance of each edge contour three-dimensional point cloud and then summing them according to a preset weight. The point cloud of the straight line segment is fitted to obtain the bending angle of the connecting row; the point cloud of the arc segment is fitted to obtain the initial bending radius of the connecting row. The final bending radius is obtained, which is positively correlated with the initial bending radius and deformation force of the connecting strip, and negatively correlated with the bending angle and thickness of the connecting strip. If the bending angle and the final bending radius are within their respective error ranges, then the connection to be tested is determined to be of normal quality.

2. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The bending angle is specifically defined as follows: the point cloud of the straight line segment is fitted using the least squares fitting method to obtain two fitted straight lines, and the angle between the two fitted straight lines is the bending angle.

3. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The initial bending radius is specifically determined by fitting the point cloud of the arc segment using the least squares fitting method to determine the radius parameter, which is the initial bending radius.

4. The machine vision-based method for detecting the bending quality of connected panels according to claim 1, characterized in that, It also includes correcting the bending angle, the corrected bending angle for, , For the bending angle, To preset hyperparameters, The real-time ambient temperature is used to obtain binocular images of the connected row to be detected.

5. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The bending angle error range is ±0.5, and the final bending radius error range is ±0.

1.

6. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The preset weights are obtained through a reinforcement learning algorithm, which employs a weight optimization method based on policy gradients.

7. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The edge detection method is Canny edge detection.

8. The machine vision-based method for detecting the bending quality of connecting strips according to claim 1, characterized in that, The binocular image of the connection row to be detected is subjected to grayscale conversion and median filtering.

9. A machine vision-based connection row bending quality inspection system, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the machine vision-based connection row bending quality inspection method according to any one of claims 1 to 8 is implemented.

Images