Suspension spring cold roll forming precision online detection system based on image recognition

Abstract

The invention relates to the technical field of image data processing, in particular to a suspension spring cold roll forming precision online detection system based on image recognition, which comprises an image acquisition and preprocessing module, a spring morphology characterization module, a spring image reconstruction module, a three-dimensional image alignment module and a dynamic monitoring module. The image acquisition and preprocessing module dynamically acquires a multi-angle image of the spring and determines a target area; the spring morphology characterization module performs morphology analysis, establishes a first morphology graph and extracts two-dimensional features; the spring image reconstruction module adopts Faster R-CNN to generate a two-dimensional mask, reconstructs a three-dimensional point cloud through a multi-angle image and extracts key nodes; and the three-dimensional image alignment module aligns the three-dimensional data with the standard model, and calculates the precision deviation to carry out qualification judgment. And the dynamic monitoring module can perform periodic execution, and triggers a stop signal to realize closed-loop control when the precision reaches the standard. In addition, the system adopts a dynamic deviation threshold, and the detection efficiency is improved according to layered processing of spring types.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and particularly to an online detection system for the cold coiling forming accuracy of suspension springs based on image recognition. Background Art

[0002] In the manufacturing process of suspension springs, the cold coiling forming process is widely used due to its advantages such as high efficiency and low cost. However, the forming accuracy of the spring directly affects its mechanical properties and service life. Therefore, it is crucial to perform high-precision detection on the formed spring. Traditional detection systems mainly rely on manual measurement or contact measurement devices, which have problems such as low efficiency, strong subjectivity, and easy damage to the product surface. In recent years, with the development of machine vision technology, non-contact detection systems based on image processing have gradually become a research hotspot, but there are still many technical bottlenecks in practical applications.

[0003] Existing machine vision technologies mainly rely on traditional edge detection and feature matching systems, which have obvious environmental adaptability defects in practical applications. In dynamic shooting scenarios, they show significant instability when the lighting conditions change: low-contrast scenarios will cause edge features to break, while overexposed environments will result in the loss of effective features; at the same time, complex light sources such as highlights and reflections are prone to generating false edge features, significantly increasing the false matching rate. Background changes caused by a moving camera will lead to feature point drift, resulting in tracking failure. In addition, non-rigid deformation will destroy the effectiveness of feature descriptors, and scale changes will also render edge detection systems with fixed thresholds ineffective.

[0004] Therefore, an online detection system for the cold coiling forming accuracy of suspension springs based on image recognition is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide an online detection system for the cold coiling forming accuracy of suspension springs based on image recognition. Through multi-technology integration and algorithm innovation, it aims to solve the pain point of difficult dynamic pose capture in the traditional detection of the cold coiling forming accuracy of suspension springs. The present invention proposes an online detection system for the cold coiling forming accuracy of suspension springs based on image recognition. It uses an adaptive image enhancement module to process noise interference in complex industrial scenarios, relies on a cascaded convolutional neural network to achieve precise segmentation and reconstruction of spring contour features, uses a temporal pose estimation algorithm to establish a three-dimensional space coordinate mapping model, and forms a non-contact detection closed-loop integrating optical imaging, intelligent analysis, and real-time feedback through a dynamic weight distribution mechanism to autonomously match the detection parameters of different models of springs, effectively solving the technical problem of online monitoring of the forming quality of suspension springs and significantly improving the detection accuracy and efficiency.

[0006] To achieve the above object, the present invention provides the following technical solutions: An on-line detection system for the cold coiling forming accuracy of suspension springs based on image recognition, comprising the following steps: Dynamically collect multi-angle images of the suspension spring during the cold coiling forming process to obtain the spring target area; Preferably, perform median filtering and image enhancement processing on the image data within the spring target area; use gradient calculation, non-maximum suppression, double-threshold screening, and edge connection techniques to detect and generate the initial contour of the suspension spring from the preprocessed image data; Based on the spring target area, extract multi-scale features, perform contour correction and matrix decomposition processing, and establish a first morphology map representing the two-dimensional morphology of the suspension spring; further extract the two-dimensional structural features of the spring; Preferably, calculate the positions of two-dimensional key points by detecting the difference of Gaussians in the scale space of the spring target area image, and calculate the directions and feature descriptors of the two-dimensional key points; perform feature point extraction or matching processing on the initial contour to obtain multiple two-dimensional feature matching points; Combine the obtained feature matching points to perform correction, fitting, and skeletonization processing on the initial contour to form the final first morphology map, and extract the two-dimensional structural features therefrom.

[0007] Preferably, the steps of using the two-stage object detector Faster R-CNN to detect the suspension spring specifically include: Use the two-stage object detector to perform pixel-level segmentation on the suspension spring in the initial forming image information to generate an accurate two-dimensional mask; The accurate two-dimensional mask is used to guide the spring three-dimensional morphology reconstruction algorithm to generate the initial three-dimensional point cloud; Preferably, the spring three-dimensional morphology reconstruction algorithm specifically includes the following steps: Based on the accurate two-dimensional mask generated by the two-stage object detector, perform feature point matching on the suspension spring in the multi-angle images to establish the corresponding relationship between the images; Sparse point cloud reconstruction, calculate the camera pose through the structure from motion (SfM) algorithm, and combine the matching feature points to generate the initial sparse three-dimensional point cloud; Dense point cloud reconstruction, use the multi-view stereo vision (MVS) algorithm to optimize the sparse point cloud to generate a high-precision real-time three-dimensional point cloud; perform outlier filtering, smoothing processing, and normal vector estimation on the real-time three-dimensional point cloud for point cloud post-processing.

[0008] Preferably, use the spring three-dimensional morphology reconstruction algorithm to generate the real-time three-dimensional point cloud of the suspension spring by using the located spring target area in the multi-angle real-time images; and based on the real-time three-dimensional point cloud, extract the key nodes characterizing its three-dimensional geometric shape; Preferably, the real-time three-dimensional point cloud or its key nodes are imported into the virtual space coordinate system and rigidly transformed and aligned with the preset suspension spring standard CAD model of the suspension spring; based on the aligned three-dimensional point cloud or key nodes and the suspension spring standard CAD model, calculate the cold coiling forming precision deviation of the suspension spring and perform a qualification determination, and the precision deviation includes at least one of length deviation, diameter deviation, number of turns deviation, pitch deviation, helix curvature deviation, and perpendicularity deviation; Preferably, the above steps are continuously and periodically executed to obtain the time series precision deviation data of the suspension spring during the cold coiling forming process. The online detection system proposed by the present invention can automatically call the preset three-dimensional reconstruction parameters, key node extraction strategies, or precision deviation calculation models for the specific type to perform precision detection according to the classified spring type or model.

[0009] Preferably, the specific steps of deviation threshold analysis and qualification determination include: The deviation threshold is not a fixed value, but can be dynamically adjusted or selected according to at least one of the following: The preset tolerance standard corresponding to the specific type or model of the suspension spring identified by the two-stage object detector or subsequent classifier; The dynamic threshold generated based on historical detection data and the statistical process control (SPC) principle to adapt to the normal fluctuations in the production process or detection process drift; The combined threshold set considering the mutual influence and correlation between multiple precision deviation parameters. Once the precision meets the standard and is determined to be qualified, a stop signal is triggered to control the cold coiling forming equipment to terminate the forming process; Preferably, an online detection system for the cold coiling forming precision of a suspension spring based on image recognition includes: Image acquisition and target positioning module: responsible for dynamically capturing multi-angle spring images and quickly locking the spring target area in the image, providing original visual data and accurate input range for subsequent analysis.

[0010] Spring morphology characterization module: based on the spring target area image, extract the multi-scale features of the spring and establish the first morphology map, and at the same time extract its two-dimensional structure parameters for preliminary evaluation or auxiliary three-dimensional analysis.

[0011] Spring image reconstruction module: use the spring target area in the multi-angle images to generate the real-time three-dimensional point cloud data of the spring, and extract the key geometric nodes from it to realize the conversion from two-dimensional to three-dimensional information.

[0012] Stereo Image Alignment Module: Based on the spring contour features extracted by the two-stage object detector, match the preset spring model database to determine the type of the current suspension spring; according to the identified spring type, automatically associate the preset detection parameters corresponding to this model, including the three-dimensional reconstruction accuracy threshold, the key node definition rule, and the tolerance standard; calculate the cold coiling forming accuracy deviation of the suspension spring, where the accuracy deviation includes at least one of length deviation, diameter deviation, number of turns deviation, pitch deviation, helix curvature deviation, and perpendicularity deviation. Dynamic Monitoring Module: Continuously and periodically execute the detection process to obtain real-time accuracy deviation data during the forming process, and receive the feedback data from the Qualified Judgment and Result Output Module in real time. Combine the statistical process control (SPC) algorithm to optimize the adaptability of the detection parameters. Configured to compare the calculated accuracy deviation with the preset tolerance standard. When all the accuracy deviations are within their corresponding tolerance standards, it is determined that the current forming accuracy of the suspension spring is qualified; otherwise, it is determined to be unqualified. After the determination, output the detection result. And after the spring accuracy meets the standard and is determined to be qualified, trigger a signal to control the cold coiling forming equipment to terminate the forming.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Perform median filtering and image enhancement processing on the collected multi-angle images, effectively suppressing noise and improving the clarity and contrast of the images, thereby enhancing the recognizability of the spring contour and details. On this basis, by extracting multi-scale features, the structural information of the spring at different scales can be comprehensively captured, including the overall shape, local bending, and subtle surface features. This hierarchical feature fusion strategy enhances the sensitivity of the system to spring defects at different scales, and effectively avoids false detection and missed detection problems at a single scale through the feature complementary mechanism, improving the reliability of the suspension spring accuracy detection.

[0014] 2. With the help of Faster R-CNN to achieve precise positioning and three-dimensional reconstruction optimization, the invention emphasizes using the two-stage object detector Faster R-CNN to detect the suspension spring, and further uses it for pixel-level segmentation to generate an accurate two-dimensional mask to further guide the spring stereo morphology reconstruction algorithm. This means that the reconstruction process can better focus on the spring itself, reducing the error caused by the spring attitude change, thereby generating more accurate key nodes, effectively eliminating the translation and rotation of the spring in space, and more accurately evaluating its true morphology deviation, providing a highly reliable data basis for subsequent accuracy calculation, and further improving the detection accuracy and efficiency of the system.

[0015] 3. Implement real-time dynamic online detection and closed-loop control. The system can continuously and periodically execute the detection process to obtain the time-series precision deviation data of the suspension spring during the cold coiling forming process. This online monitoring method can promptly detect abnormalities during the forming process and avoid the production of defective products. More importantly, the system is equipped with a cold coiling forming dynamic monitoring and stop control module. When the spring precision meets the standard, the system can trigger a stop signal to control the cold coiling forming equipment to terminate the forming. At the same time, the system can automatically call the preset 3D reconstruction parameters, key node extraction strategies, or precision deviation calculation models for this specific type according to the classification result. This indicates that the detection system has the ability to support the precision measurement of multiple types of springs, with higher flexibility and applicability, and can meet different production requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of an online detection system for the cold coiling forming precision of a suspension spring based on image recognition proposed in an embodiment of the present invention application; Figure 2 It is a diagram of the system anomaly detection and adjustment steps of the present invention application; Figure 3 It is a schematic structural diagram of the detection system proposed in an embodiment of the present invention application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] The present invention provides an online detection system for the cold coiling forming precision of a suspension spring based on image recognition and designs a set of online detection systems for the cold coiling forming precision of a suspension spring. As an implementation manner of the present invention, refer to Figure 1 It is a flowchart of an online detection system for the cold coiling forming precision of a suspension spring based on image recognition proposed in an embodiment of the present invention application.

[0019] As a key functional component in the automotive suspension system, the forming quality of the spring directly affects the safety and comfort of the whole vehicle. Traditional detection systems have problems such as low detection accuracy, long detection time, and inability to reflect the actual deformation of the spring after being stressed online, making it difficult to meet the high standards of modern automotive intelligent manufacturing. In view of these deficiencies, the present invention proposes an online detection system for the cold coiling forming accuracy of springs based on high-resolution visual imaging and deep learning. This system combines multi-scale feature extraction and spring three-dimensional morphology reconstruction algorithms to achieve high-precision online real-time measurement of suspension springs, greatly improving the detection accuracy and efficiency. At the same time, it has strong adaptability and can meet the online measurement requirements of springs of different specifications.

[0020] A certain automotive parts manufacturer has successfully deployed and operated an online detection system for cold coiling forming accuracy based on image recognition on its suspension spring production line. To illustrate the system of the present invention and the effectiveness of the system, it will be specifically described in combination with the accompanying drawings of this embodiment and the following two embodiments.

[0021] Embodiment 1 A certain manufacturer has applied an online detection system for cold coiling forming accuracy of suspension springs based on image recognition on the production line. Refer to Figure 1 Referring to the flowchart, the specific implementation steps are as follows: Collect images of the initial state of the spring. Use 3 industrial cameras with a resolution of 1920x1080 pixels to synchronously capture the initial forming state of the suspension spring from 0° / 60° / 120° respectively. The current spring being detected is numbered "SPG-20230728-001".

[0022] Furthermore, perform grayscale processing to enhance the contrast: Specifically, for the high-gray-scale area (spring body), set the gray-scale threshold and use linear stretching to enhance the contrast: for pixels exceeding the threshold, increase the gray-scale value proportionally, and the upper limit of the increase is its maximum value; for the low-gray-scale area (background), use neighborhood local mean replacement to smooth the noise. At the same time, apply a median filter to remove salt-and-pepper noise and optimize the clarity of the spring contour.

[0023] In the image preprocessing stage, first perform grayscale conversion on the input color image to generate a single-channel grayscale image. Furthermore, perform histogram stretching on the high-gray-scale area to enhance the contrast of the main body; perform local mean integration on the low-gray-scale area to suppress background interference. Through segmented processing, while enhancing the significance of spring features, the influence of noise is effectively reduced.

[0024] Apply a 3x3 median filter to remove noise. For each pixel point, take the median of the pixel gray-scale values within the neighborhood 3x3 to eliminate noise and make the spring contour smoother.

[0025] Further, using gradient calculation, non-maximum suppression, double-threshold screening, and edge connection techniques, detect and generate the initial contour of the suspension spring from the preprocessed image data; Apply the Canny algorithm to the recognized spring morphology region image for edge detection, and use the convolution kernel operator to perform convolution operations on the image. For the image, calculate the gradient magnitude and gradient direction in the x and y directions, and use non-maximum suppression to refine the edge to ensure that each edge is only one pixel wide. Then apply double thresholds: set the high threshold to 100 and the low threshold to 50 for screening and edge connection to obtain 3 clear initial contour images of the spring.

[0026] Further, calculate the positions of the two-dimensional key points by detecting the difference of Gaussian in the scale space of the spring target region image, and calculate the directions and feature descriptors of the two-dimensional key points; perform feature point extraction or matching processing on the initial contour to obtain multiple two-dimensional feature matching points; Apply the difference of Gaussian detection to find the local extreme points of the difference of Gaussian response at different scales and spatial positions to determine the positions of the key points. The calculation method is as follows: ; Where, are the pixel coordinates of the image, is the scale parameter representing the degree of Gaussian blur, is the scale factor, which defines the proportional relationship of the blur degree between adjacent scales. In this embodiment, the scale parameter is set to start from 1.6, and the scale factor k is set to , detect 3 octaves, and each octave contains 4 scale levels.

[0027] Further, for each detected key point, calculate the gradient direction distribution in its surrounding neighborhood, determine the main direction of the key point and generate a rotation-invariant feature descriptor. Use the corner detection algorithm to extract feature points on the initial contour of the spring, and perform feature matching between the initial contour and the preset two-dimensional model contour of the spring to obtain multiple corresponding two-dimensional feature matching points.

[0028] Specifically, assume that 150 - 200 key points are detected per image on average, calculate the descriptors of the key points, and match them with 180 feature points in the pre-stored standard two-dimensional model contour of the spring. Finally, an average of 120 - 160 feature point pairs are successfully matched per image.

[0029] Using the feature matching points as strong constraints, perform contour correction, fitting, and skeletonization processing on the initial contour of the spring to generate an accurate two-dimensional shape representation - "the first morphology map".

[0030] By dynamically collecting suspension springs from multiple angles and accurately identifying target areas of the springs, combined with median filtering and image enhancement processing, the system can fully capture spring status information, significantly improve image quality, suppress noise interference, and provide high-quality geometric basis for subsequent contour extraction.

[0031] Extract key 2D structural parameters from the first topography image to assist in initial estimation and verification of 3D reconstruction and quantify the geometric properties of the spring, including: (1) Length: Calculate the length of the skeleton, or the distance between key points on the contour.

[0032] (2) Diameter: Analyze the local width of the contour, or measure the length of the section perpendicular to the skeleton.

[0033] (3) Spring pitch: Analyze the number of bending times of the frame or the number of periodic fluctuations of the profile in the axial direction to estimate the effective number of turns of the spring in the two-dimensional projection.

[0034] (4) Number of turns: Analyze the number of bends of the skeleton or the number of periodic fluctuations of the profile in the axial direction to estimate the effective number of turns of the spring in the two-dimensional projection.

[0035] (5) Perpendicularity: The maximum allowable angle or linear offset between the center axis and the theoretical installation reference plane, which is used to measure the axial alignment accuracy of the spring after assembly.

[0036] (6) Curvature: Calculate the local curvature of each point on the contour or skeleton.

[0037] The initial contour is generated by combining gradient calculation, non-maximum suppression and other technologies, which improves the stability of suspension spring edge detection. Through feature point matching and contour correction, a more accurate second topography image is formed, laying the foundation for two-dimensional structural feature extraction and improving detection reliability.

[0038] The detection branch of Faster R-CNN is after RPN, and performs more accurate classification and bounding box regression for each candidate region. The integrated Mask R-CNN branch is used for pixel-level segmentation, that is, a small fully convolutional network for predicting segmentation masks is added on the basis of Faster R-CNN, and the cross-entropy loss function is used for training and optimization. An accurate two-dimensional mask is generated, and the pixel value in the mask indicates whether the corresponding image pixel belongs to the suspension spring.

[0039] Through pixel-level segmentation prediction, a mask is generated to accurately locate the two-dimensional area of the suspension spring to avoid false detection. At the same time, the two-dimensional mask matches the size of the original image to ensure the accuracy of spatial alignment during three-dimensional reconstruction and improve model consistency.

[0040] The key steps of SfM include feature matching, by matching feature points in different images; Epipolar geometry constraint, using the fundamental matrix F or the essential matrix E to constrain the matching points: or ; where and are the matching image points. The fundamental matrix F contains the relative rotation and translation information of the two cameras. In normalized image coordinates, the essential matrix E contains the relative rotation and translation information.

[0041] Apply the MVS algorithm to multiple images with known camera poses. By pixel-level matching and depth estimation, find corresponding and consistent pixels in images from different perspectives, and calculate their three-dimensional coordinates, and then generate a dense point cloud.

[0042] The MVS algorithm selects the sum of the squared differences of the corresponding pixels between two image patches as the cost function for photometric consistency. The smaller its value, the higher the similarity.

[0043] Multi-view feature matching and the MVS algorithm generate a high-precision sparse point cloud, providing a reliable data basis for 3D reconstruction; using outlier filtering and normal vector estimation to optimize the point cloud quality, enhancing the detail restoration ability of the 3D model, and ensuring the credibility of accuracy detection.

[0044] Furthermore, perform the pixel back-projection step, which is the process of converting the pixel points on the two-dimensional image back to the three-dimensional camera coordinate system. For each reference image, estimate the depth value of each pixel. For a pixel in the reference image, assuming a depth, the pixel can be back-projected into the three-dimensional space to obtain a three-dimensional point P. Then, project P onto the images of other known camera poses to obtain the corresponding pixels. Calculate the photometric consistency cost between the reference image patch and the target image patch. By searching within the depth range, find the depth value that minimizes the cost function.

[0045] The point obtained by back-projecting the pixel into the three-dimensional space, the pixel obtained by projecting the three-dimensional point P onto the image plane of another camera. For a pixel in a reference image, the algorithm searches within a certain depth range. For each assumed depth, the pixel is back-projected into the three-dimensional space and re-projected into other known camera views. Then, calculate the photometric consistency cost between the image patch around the pixel in the reference image and the corresponding image patch projected into other views. Select the depth value that minimizes the cost, that is, the depth value with the highest similarity, as the estimated depth of the pixel.

[0046] When the depth value of each pixel is estimated, the internal and external parameters of the camera can be used to back-project each pixel into the three-dimensional space to obtain the three-dimensional point coordinates corresponding to the pixel.

[0047] Finally, by rotating and translating according to the external parameters of the reference camera, the three-dimensional points in the camera coordinate system can be converted into coordinates in the world coordinate system; by performing depth estimation and back-projection on the pixels of all reference images and carrying out coordinate transformation, a dense three-dimensional point cloud of the scene can be obtained.

[0048] Verify and correct the first topography map through projection to reduce the cumulative error between multi-view images. Subsequently, perform a rigid transformation on the three-dimensional point cloud in the virtual space coordinate system to align it with the standard model, directly quantify the cold roll forming deviation, simplify the determination process and improve the efficiency.

[0049] Furthermore, based on multi-angle images and accurate two-dimensional masks, use the spring stereo topography reconstruction algorithm to generate a real-time three-dimensional point cloud of the spring. Apply the Faster R-CNN two-stage object detector again to extract the key nodes from the three-dimensional point cloud that can represent the three-dimensional geometric shape of the spring. These key nodes represent the main geometric features of the spring and are used for subsequent error calculation.

[0050] Furthermore, use the Iterative Closest Point (ICP) algorithm to achieve precise alignment between the three-dimensional point cloud and the standard model. The algorithm iteratively finds the closest point pairs between the two point clouds and calculates the rotation and translation transformation that minimizes the distance between these point pairs. The objective function is set to minimize the sum of the squared distances between the point pairs.

[0051] It can also automatically identify the type or model of the spring, load the corresponding detection parameters, standard model and tolerances, and assist in dynamically adjusting the threshold. Continuously perform the detection periodically, obtain the real-time precision deviation data, and trigger a stop signal to control the forming equipment when the precision meets the standard. Compare the calculated precision deviation with the tolerance standard, make a pass / fail determination, and summarize and output the detection results and reports.

[0052] Specifically, the calculation of the spring precision deviation includes: Length deviation: The difference between the distance between the actual key points of the spring and the distance between the corresponding points of the standard model can be calculated; Diameter deviation: Fit the helix of the actual spring and calculate the difference between its average diameter and the diameter of the standard model; Number of turns deviation: Analyze the structural characteristics of the three-dimensional point cloud to identify the actual number of turns of the spring and compare it with the number of turns of the standard model; Pitch deviation: Calculate the difference between the average axial distance between adjacent turns of the actual spring and the corresponding distance of the standard model; Verticality deviation: Analyze the angle between the spring axis and the reference plane and calculate the deviation from the vertical angle of the standard model; Helix curvature deviation: Fit the helix of the actual spring and calculate the difference between its curvature and the curvature of the helix of the standard model.

[0053] Table 1: Three-dimensional precision calculation and deviation analysis

[0054] Furthermore, to establish a unified error measurement standard, the deviation threshold between two topography maps needs to be calculated. Specifically, the topography deviation is measured by the corresponding distances between two sets of points, and the deviation index is measured by the root mean square error.

[0055] Furthermore, for the topography comparison and deviation calculation in the virtual space, the deviation threshold standard is set by the maximum deviation. If it exceeds this threshold standard, it is determined that the forming accuracy is unqualified; otherwise, it is qualified.

[0056] Specifically, the corresponding point distances between the aligned measured point cloud and the standard model point cloud are calculated, and the topography deviation is ; the maximum deviation is calculated as ; Accurately calculating the spring curvature is the key to accurately evaluating the internal stress and predicting the fatigue life, ensuring the safety and reliability of the spring design. If the geometric mutation points of the spring can be identified based on this, the weakest links with the most concentrated stress can be located, guiding the optimization design and preventing early failure. The abnormal condition for curvature detection is set as when the curvature exceeds the standard by 5% and the curvature gradient of adjacent points , it is marked as a mutation point. When the curvature deviation exceeds the process tolerance, it is marked as a forming defect area and fed back to the dynamic monitoring module for early warning and process parameter adjustment.

[0057] Table 2 Topography Comparison and Qualification Judgment

[0058] Based on this, it is determined that the spring "SPG-20230728-001" is qualified. From image acquisition to result output, the whole process takes about 3.5 seconds. The online system performs real-time detection and result output. The system collects the spring status in real time and automatically executes the above steps; the detection results are immediately feedback, including the graphic deviation display and qualification judgment information; it provides a data report for the production line management personnel, supporting traceability and quality control. When the system is continuously abnormal, it will adjust the parameters and import them, referring to Figure 2 which is the system abnormal detection and adjustment step diagram of this invention application; Result Interface: Displays the 3D reconstruction model of the spring and the deviation heat map from the standard model. Lists the measured values, standard values, and deviation values of various geometric parameters. Clearly gives the judgment of "qualified" or "unqualified". When it is unqualified, it is highlighted.

[0059] Data Report: Automatically generates a detection report containing the above information and stores it in the database for quality traceability. It can adapt to the production requirements of different models of springs, realizes flexible detection through parameter self-adaptation, and expands the application scenarios of the system.

[0060] Furthermore, excessive cold coiling of suspension springs can lead to serious consequences such as out-of-tolerance dimensions, abnormal stiffness, and a sharp drop in fatigue life, and even trigger safety accidents. Therefore, the system is designed with full-process quality monitoring to ensure forming accuracy, continuously monitoring the time-series data of spring deviations, avoiding excessive intervention while being able to correct deviations in a timely manner. The adjustment results are fed back to the equipment control system to form a closed loop of "detection-verification-termination of forming", continuously optimizing the production process.

[0061] Table 3 Record of Time-Series Precision Data during the Forming Process of "SPG-DYNAMIC-001"

[0062] Note that: At T0 + 2.3s, although a single parameter may be on the verge of being qualified, the system may be designed such that all key parameters need to stably meet the qualified standard within 2 consecutive detection cycles, or the comprehensive profile deviation meets the standard before being judged as finally qualified to avoid misjudgment caused by critical fluctuations. Here, it is simplified to judge that all parameters are qualified at the current time.

[0063] Specifically, when the detection system determines at T0 + 2.5s that all key geometric parameters of the spring "SPG-DYNAMIC-001" are within the preset deviation threshold range, that is, when it is comprehensively judged as qualified, the system will immediately perform the following operations: Generate a qualified judgment signal: The internal state of the detection system is updated to "forming qualified".

[0064] Send a stop instruction: Send a stop signal to the PLC of the cold coiling forming equipment through a preset communication interface.

[0065] Equipment response: After the PLC of the cold coiling forming equipment receives the stop signal, it immediately executes its preset program to safely stop the material feeding and winding actions and complete the forming process of the current spring. Finally, the result output and data recording are carried out: The real-time interface displays the 3D model, deviation heat map, parameter list, and qualified judgment of the finally qualified spring.

[0066] Data report: Automatically generate a detection report containing time-series data and the finally qualified status and store it in the database. This report not only includes the accuracy of the final product but also records its production process, providing richer data for process analysis and optimization.

[0067] Reduce manual intervention: The fully automated adjustment process reduces the dependence on operators, improving production efficiency while reducing the risk of human errors.

[0068] By real-time monitoring the change rates of multi-dimensional parameters and dynamically adjusting the system thresholds, it can effectively cope with the process fluctuations in the production process and avoid misjudgments caused by fixed thresholds. Automatically adjust the sampling density and feature extraction parameters of the spring three-dimensional topography reconstruction algorithm based on the equipment status, significantly reducing vibration interference and ensuring the accuracy of point cloud data. The real-time calibration mechanism verifies the adjustment effect, avoids introducing new errors due to parameter changes, and ensures the consistency and reliability of the detection results.

[0069] This on-line detection system for the cold coiling forming accuracy of suspension springs realizes the precise control of key quality parameters through an intelligent detection process. Combining multiple indicators such as the length deviation, elastic coefficient, and surface defect score of the spring, a threshold combination is formed by dynamically adjusting the segment thresholds of each indicator in real time, simplifying the judgment process. Set the weight distribution: the length is 1.2, the diameter is 0.8, the pitch is 0.5, the perpendicularity is 1.0, and the curvature is 0.7; calculate the comprehensive score: 0.35×1.2 + 0.12×0.8 + 0.09×0.5 + 0.25×1.0 + 2×0.7 = 2.231 points. The dynamic passing line is set at 2.5, and based on this, it is determined that the suspension spring passes the detection. At the same time, the veto power of key parameters can also be set. If the curvature deviation rises to 4%, the passing line may be tightened to 2.0, then it is determined as unqualified.

[0070] Embodiment 2 A certain automobile brand has a variable-diameter and variable-pitch suspension spring with a special design (the coil diameters at both ends are smaller, the coil diameter in the middle is larger, and the pitch is uneven to achieve specific non-linear stiffness characteristics), which has extremely high forming accuracy requirements. This on-line detection system described in the present invention is deployed on this production line.

[0071] The differences from Embodiment 1 are: the potential improvement in the number and resolution of images, the increase in the number of feature points, the improvement in the three-dimensional point cloud density, and more complex geometric analysis algorithms.

[0072] Specifically, in the face of the more complex contour and potential self-occlusion of the variable-diameter and variable-pitch spring, a more diverse and comprehensive combination of shooting angles (0°, 45°, 90°, 135°, 180°, 225°) is adopted. This circumferential camera layout, which may be strengthened for specific complex areas, ensures that all key geometric transition areas on the spring, especially the diameter change points and pitch mutation points, can be clearly and unobstructedly presented in at least several images, providing a more complete and redundant data source for subsequent high-precision three-dimensional reconstruction. This is in contrast to the relatively uniform viewing angle requirements of conventional springs.

[0073] Furthermore, for the more drastic light and shadow changes that may exist in complex springs or the stray highlights caused by complex surfaces, the preprocessing parameters have been adjusted more meticulously. In this embodiment, the gray threshold for contrast enhancement is adjusted to 170, and the stretching coefficient α is increased to 1.6 to better distinguish the spring body from the background; the neighborhood for local mean replacement is expanded to 7x7 to more effectively smooth the background noise, especially when dealing with complex backgrounds or slight oil stain reflections; the window for median filtering is also increased to 5x5 to more powerfully remove salt-and-pepper noise and the tiny highlight points that may be caused by complex surfaces, ensuring the continuity and smoothness of the contour. These adjustments reflect a stronger adaptability to complex image features.

[0074] Furthermore, considering the multi-scale features of variable-diameter spring wires due to the large variation in apparent diameter and uneven pitch density, the Difference of Gaussian (DoG) detection is extended to 4 octaves, with each octave containing 5 scale levels. This increases the scale range and density of feature detection, enabling better capture of various features from coarse to fine, especially in the regions of diameter and pitch changes. Correspondingly, the pre-stored standard two-dimensional model contour contains approximately 250 feature points distributed for variable-diameter and variable-pitch features to cover more complex geometric details, and the finally successfully matched feature point pairs also increase to approximately 240 accordingly. This reflects a stronger ability to capture and match complex geometric features.

[0075] Specifically, for the characteristics of variable-diameter and variable-pitch springs, the parameter definitions are more complex and localized. Two-dimensional parameter extraction adds the projected minimum outer diameter (ends 1 and 2), projected pitch at end 1, projected pitch in the middle, etc. The three-dimensional accuracy calculation goes even further, including the maximum outer diameter (in the middle), minimum outer diameter (ends 1 and 2), average pitch at end 1 (first two coils), average pitch in the middle (middle three coils), and the curvature radius of the key cross-section. This parameter refinement from the overall to the local is the key to meeting the functional requirements of complex springs and can more comprehensively evaluate their forming quality.

[0076] Due to the parameter refinement, the deviation threshold standards are also set correspondingly for different parts and different types of parameters, and the passing judgment is also based on these more refined parameters. In this embodiment, the overall failure is caused by the out-of-tolerance of the "average pitch deviation at end 1", even though other macroscopic parameters may be qualified. This reflects the key quality control of the functional critical areas of complex springs.

[0077] Table 4 Morphology Comparison and Passing Judgment (SPG-VARIO-20230915-077)

[0078] It can be seen from the test results that the qualified items are as follows: all 9 error parameters are within the allowable deviation range and meet the standards. Only the pitch deviation of the end part is +0.51 mm, exceeding the threshold of ±0.5 mm, so it is judged as a non-qualified item and needs to be focused on. Therefore, except for the end pitch, the other indicators are qualified, and the overall quality is controllable, but the end pitch process needs to be adjusted.

[0079] The online detection system for the variable-diameter and variable-pitch suspension spring production line of this automobile brand has successfully overcome the detection difficulties of complex spring geometric features through an upgraded multi-dimensional data acquisition and intelligent analysis architecture. The system adopts a six-view (0° - 225°, with an interval of 45°) high-resolution synchronous imaging scheme, achieving a feature point coverage rate of more than 200% in key areas such as the diameter mutation area and the pitch transition section. Through actual measurement, the three-dimensional point cloud density reaches 3 times that of conventional springs. Combining with an improved NURBS surface fitting algorithm, the contour reconstruction error of the variable-diameter area is controlled within ±0.08 mm, and the measurement accuracy of the non-linear section of the pitch reaches ±0.12 mm, fully meeting the strict tolerance requirements of this special spring.

[0080] Compared with the uniform spring detection in Embodiment 1, this system effectively solves the problem of self-occlusion detection of variable-diameter and variable-pitch springs through an innovative adaptive view fusion technology and an intelligent segmented detection algorithm, and realizes the accurate capture of complex geometric features. At the same time, the integrated non-linear stiffness verification function upgrades the traditional single-size detection to a comprehensive quality assessment including functional parameters, significantly improving the process control ability of special spring products.

[0081] Refer to Figure 3 , which is the structural schematic diagram of the detection system proposed in the embodiment of this invention application.

[0082] An online detection system for the cold coiling forming accuracy of suspension springs based on image recognition is formed. According to the two embodiments in this article, the detection accuracy of the system is fully verified. This system innovatively integrates multi-scale feature extraction and Faster R-CNN object detection technology, and realizes the synchronous measurement of key parameters such as spring length, diameter, number of turns, pitch, and perpendicularity through dynamically capturing multi-angle spring images, extracting multi-scale features, three-dimensional reconstruction, and virtual space transformation analysis. It can adapt to the continuous operation requirements of industrial production lines, and the efficiency is more than 10 times higher than that of traditional manual detection, demonstrating the remarkable effectiveness of this system in improving product quality and reducing labor costs.

[0083] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An on-line detection system for the cold coiling forming accuracy of suspension springs based on image recognition, characterized in that, Including: Image acquisition and preprocessing module: Dynamically acquire multi-angle images of the suspension spring to obtain the spring target area; Spring morphology characterization module: Extract multi-scale features of the spring target area, including the overall shape of the suspension spring, local bending, and fine surface features. Perform contour correction and matrix decomposition on the multi-scale features to obtain the first morphology map, and then extract the two-dimensional structure features of the spring; Spring image reconstruction module: Segment the spring target area, use the two-stage object detector Faster R-CNN to locate the spring position, frame the spring projection shape to generate a two-dimensional mask; Use the spring three-dimensional morphology reconstruction algorithm to generate real-time three-dimensional point clouds and extract key nodes from them; Stereo image alignment module: After rigid transformation alignment, compare with the standard CAD model of the suspension spring, calculate the forming accuracy deviation of the suspension spring and perform a qualification determination, and summarize the detection results; Dynamic monitoring module: Obtain the time series data of the accuracy deviation of the suspension spring, and terminate the forming when it is determined to be qualified.

2. The on-line detection system for the cold coiling forming accuracy of a suspension spring based on image recognition according to claim 1, characterized in that The image acquisition and preprocessing module includes: Dynamically acquire multi-angles of the suspension spring during the cold coiling forming process to obtain spring image information, and obtain the spring target area according to the spring image information; Perform median filtering and image enhancement processing on the image data in the spring target area to obtain the first morphology map; Use gradient calculation, non-maximum suppression, double-threshold screening, and edge connection techniques to detect and generate the initial contour of the suspension spring from the first morphology map.

3. An online detection system for the cold coiling forming precision of a suspension spring based on image recognition according to claim 2, characterized in that, The spring morphology characterization module includes: Extract and match feature points of the initial contour to obtain multiple two-dimensional feature matching points; Combine the obtained two-dimensional feature matching points to perform correction, fitting, and skeletonization processing on the initial contour to form the second morphology map, and extract the two-dimensional structure features from the second morphology map.

4. An on-line detection system for the cold coiling forming accuracy of a suspension spring based on image recognition according to claim 1, characterized in that, The spring image reconstruction module includes: Use the two-stage object detector Faster R-CNN to segment the suspension spring in the first morphology map to generate a two-dimensional mask; Based on the two-dimensional mask, perform feature point matching on the suspension spring in multi-angle images to establish the corresponding relationship between spring images; Combine the feature matching points to generate sparse point clouds; Use the multi-view stereo vision (MVS) algorithm to optimize the sparse point clouds, generate real-time three-dimensional point clouds and further enhance the features.

5. An on-line detection system for the cold coiling forming precision of a suspension spring based on image recognition according to claim 1, characterized in that The stereo image alignment module includes: Project the key nodes back into the two-dimensional coordinate system of the corresponding multi-angle images; Use the projected two-dimensional key node positions and their neighborhood image information in the corresponding images to verify, correct, and enhance the first morphology map; Align the key nodes of the real-time point cloud with the standard CAD model through rigid transformation, calculate the cold coiling forming deviation, and output the qualification determination result.

6. An on-line detection system for the cold coiling forming accuracy of a suspension spring based on image recognition according to claim 4, characterized in that, Classifying the suspension spring models includes: The system automatically calls the preset three-dimensional reconstruction parameters, key node extraction strategies, and deviation calculation models for the model to perform accuracy detection according to the classified spring model; 7. An on-line detection system for the cold coiling forming accuracy of a suspension spring based on image recognition according to claim 1, characterized in that, The dynamic monitoring module includes: Collect multi-parameter time-series data during the cold roll forming process in real time. When all key parameters reach the preset thresholds simultaneously for the first time and all parameters remain qualified within the subsequent N consecutive detection cycles, it is determined to be qualified. When making the final qualified determination, send a hierarchical stop instruction to the PLC, first cut off the material feeding, and then completely stop the machine after the winding kinetic energy decays.

8. An on-line detection system for the cold coiling forming precision of suspension springs based on image recognition according to claim 1, characterized in that The dynamic threshold adjustment of the dynamic monitoring module is specifically as follows: Calculate the change rate of each parameter in real time according to the time series data, and trigger an alarm when the change rate exceeds the process tolerance. Take the equipment operation status as a boundary condition and integrate it into the spring three-dimensional topography reconstruction algorithm to establish a mapping relationship. When the parameter change rate continuously exceeds the tolerance range, automatically optimize the determination threshold and sampling frequency of the subsequent detection cycles, and perform parameter adjustment according to the degree of exceeding the standard.

9. The on-line detection system for the cold coiling forming precision of a suspension spring based on image recognition according to claim 1, characterized in that The three-dimensional image alignment module also includes curvature analysis: Based on the real-time three-dimensional point cloud data, calculate the curvature distribution of the spring center axis through the differential geometry algorithm to identify local bending mutation points of the spring. Compare and analyze the spring curvature extreme points with the theoretical curvature of the standard CAD model. When the curvature deviation exceeds the process tolerance, mark it as a forming defect area and feedback it to the dynamic monitoring module for early warning and process parameter adjustment.

Images