Bearing part intelligent detection and management system based on machine vision

Abstract

The invention discloses a bearing part intelligent detection and management system based on machine vision, and relates to the technical field of bearing manufacturing and detection. The method is used for solving the problems of bearing surface foreign matter recognition and pollution source tracking optimization. The system comprises a phase order correction and reconstruction module which generates a regularization weight factor through a high-frequency phase gradient, dynamically inhibits an order jump error of a raceway steep region, and outputs a high-precision three-dimensional shape and a gradient field; the coupling analysis module synchronously calculates local curvature entropy and polarization scattering intensity to realize foreign matter type identification and accurate positioning; the stress mapping decision module maps the foreign matter coordinates to a Hertz contact stress field, calculates a service life attenuation value in combination with a stress amplification coefficient, and generates a sorting priority decision map; and the pollution source reverse optimization module is used for dynamically adjusting dust removal equipment parameters based on foreign matter coordinate space clustering and assembly equipment position association. According to the system, closed-loop intelligent management of foreign matter detection and pollution source control is realized, and the bearing product quality and the production efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of bearing manufacturing and inspection technology, specifically to an intelligent inspection and management system for bearing parts based on machine vision. Background Technology

[0002] As high-end manufacturing equipment continues to evolve towards higher speeds, higher loads, and longer lifespans, bearings, as key components for support and rotation, have become a core factor ensuring the operational safety of the entire system. In high-reliability industries such as aerospace, wind power, high-speed rail, and new energy, contaminants such as micron-sized metal shavings and fiber particles embedded in the bearing raceway surface can form stress concentration points, significantly reducing rolling contact fatigue life and inducing early pitting and spalling failure modes. Research shows that the cleanliness of bearings during assembly before leaving the factory directly determines their initial stable operating cycle, while the non-contact nature, high distribution, and diverse material properties of contaminants present significant technical challenges for high-precision, full-process testing.

[0003] Existing methods for detecting and assessing contamination in bearing components still face several technical limitations in engineering practice. First, 3D vision methods based on laser triangulation or structured light coding often encounter dead zones due to drastic changes in the incident angle when processing high-curvature areas such as bearing raceways. This results in incomplete reconstruction of surface information at critical locations, leading to potential missed detections. Second, commonly used grayscale thresholding algorithms struggle to accurately distinguish between foreign objects of different materials, especially in scenarios where highly reflective metal debris and low-contrast fibrous contaminants coexist, significantly reducing accuracy. Furthermore, current detection systems largely rely on foreign object size or brightness intensity as hazard assessment criteria, neglecting the correlation between the foreign object's location and the bearing's contact stress distribution. This can lead to misclassifying low-risk edge foreign objects as high-risk, resulting in wasted resources. Finally, existing detection results typically end with rejection or alarms, failing to integrate the spatial distribution of foreign objects with assembly conditions. This results in a lack of contamination source tracing pathways, hindering closed-loop optimization of production line dust removal strategies and parameters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent inspection and management system for bearing parts based on machine vision, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent inspection and management system for bearing parts based on machine vision, comprising the following modules: a phase order correction and reconstruction module, a coupling analysis module, a stress mapping decision module, and a pollution source reverse optimization module; the phase order correction and reconstruction module is used to generate a regularization weighting factor based on the high-frequency phase gradient amplitude when performing phase unwrapping on the interference fringe image, dynamically suppressing the order jump error in the steep region of the bearing raceway, and outputting three-dimensional morphology data and gradient field; the coupling analysis module is used to simultaneously calculate the local curvature entropy and polarization scattering intensity distribution of the three-dimensional morphology, identify the type of foreign object based on the coupling relationship between the curvature entropy extremum and the scattering intensity threshold, and locate the coordinates of the foreign object in the bearing contact ellipsoid area; the stress mapping decision module is used to map the coordinates of the foreign object to the Hertz contact stress distribution field, calculate the life decay value based on the stress amplification factor corresponding to the type of foreign object, and generate a sorting priority decision map; the pollution source reverse optimization module is used to perform spatial clustering analysis on the set of foreign object coordinates, and adjust the working parameters of the dust removal equipment in reverse based on the correlation between the cluster center coordinates and the assembly equipment position.

[0006] Furthermore, when performing phase unwrapping on the interference fringe image, the specific process of generating a regularized weighting factor based on the high-frequency phase gradient magnitude is as follows: Fourier frequency domain filtering is performed on the interference fringe image to separate the phase component with the highest spatial carrier frequency as the high-frequency phase field; the gradient magnitude of the high-frequency phase field in the x / y direction is calculated, and the gradient region exceeding the characteristic curvature threshold of the bearing raceway is marked as a steep sensitive region; the gradient magnitude is input into an exponential decay function to generate a regularized weighting factor that is inversely proportional to the gradient magnitude, wherein the weighting factor of the steep sensitive region approaches zero to forcibly suppress order jumps.

[0007] Furthermore, the specific process for dynamically suppressing the order jump error in the steep region of the bearing raceway and outputting three-dimensional topography data and gradient field is as follows: a regularization weighting factor is introduced as a penalty term into the least squares phase unwrapping algorithm to apply gradient continuity constraints in the steep sensitive region and reconstruct an absolute phase field without jumps; the height values ​​of each point on the bearing surface are calculated based on the absolute phase field to generate three-dimensional topography data with sub-micron resolution; the Sobel operator convolution operation is performed on the three-dimensional topography data to output a height gradient field representing the surface curvature change.

[0008] Furthermore, the specific process of simultaneously calculating the local curvature entropy and polarization scattering intensity distribution of the three-dimensional morphology is as follows: Based on the height gradient field of the three-dimensional morphology, an analysis window matching the curvature radius of the bearing raceway is dynamically generated with each pixel as the center, and the information entropy value of the height gradient within the window is calculated as the local curvature entropy; the Stokes vector in the interference optical path is analyzed, and the polarization scattering intensity component related to the surface micromorphology is separated to construct the scattering intensity distribution field; the curvature entropy distribution map and the scattering intensity distribution map are registered to the same bearing surface coordinate system through coordinate transformation.

[0009] Furthermore, the specific process of identifying foreign object types and locating the coordinates of foreign objects in the bearing contact ellipsoid region based on the coupling relationship between curvature entropy extrema and scattering intensity threshold is as follows: A joint probability distribution function is constructed for the coordinates of curvature entropy extrema points and their corresponding scattering intensity values; a nonlinear boundary equation for judging metal debris and fibrous contaminants is defined. When the curvature entropy value at a certain coordinate point exceeds the curvature entropy threshold and the corresponding scattering intensity value exceeds the scattering intensity threshold, the point is judged to be metal debris. When the curvature entropy value exceeds the curvature entropy threshold but the scattering intensity value is lower than the scattering intensity threshold, the point is judged to be fibrous contaminants. For areas where the curvature entropy value does not exceed the curvature entropy threshold, they are marked as background surface features and do not trigger foreign object judgment. Based on the real-time bearing operating parameters, including load and speed, the three-dimensional spatial boundary of the contact ellipsoid region is dynamically calculated. The coordinates of the identified foreign objects are transformed to the relative position coordinate system of the contact ellipsoid region, and the hazard level is marked according to the type of foreign object and its position within the contact ellipsoid region.

[0010] Furthermore, the specific process of mapping the foreign object coordinates to the Hertz contact stress distribution field and calculating the life attenuation value based on the stress amplification factor corresponding to the foreign object type is as follows: Map the foreign object coordinates to the Hertz contact stress distribution field, calculate the continuous stress distribution in the contact ellipsoid region based on the bearing geometric parameters and real-time operating load, and extract the maximum stress value at the center coordinates of the foreign object; match the pre-calibrated stress amplification factor according to the foreign object type, with metal debris using a coefficient that increases nonlinearly with embedding depth and fiber contaminants using a linear coefficient that increases proportionally with the contact area; input the amplified stress value into the bearing material fatigue characteristic curve, and calculate the life attenuation value in combination with the cumulative damage.

[0011] Furthermore, the specific process for generating the sorting priority decision map is as follows: A judgment space consisting of the foreign object hazard level and lifespan decay value is established. The foreign object characteristics are mapped to the two-dimensional judgment space, and the bearing surface sorting priority coding map is output. The coding rules are: Immediate sorting priority: when the lifespan decay value exceeds the first preset threshold, or the hazard level is high; Batch sampling priority: when the lifespan decay value exceeds the second preset threshold but does not exceed the first preset threshold, or the hazard level is medium; Release priority: when the lifespan decay value does not exceed the second preset threshold and the hazard level is low.

[0012] Furthermore, the pollution source reverse optimization module includes the following steps: analyzing the set of foreign object coordinates using an adaptive neighborhood radius density clustering algorithm to identify high-density pollution core areas; calculating the spatial distance between the cluster center point and the coordinates of the assembly line equipment, and marking the equipment corresponding to the minimum distance as a potential pollution source; generating a dust removal intensity control coefficient based on the physical area of ​​the clustered area and the number of foreign objects, and increasing the working air volume and negative pressure value of the dust removal equipment at the potential pollution source in proportion to the coefficient.

[0013] The present invention has the following beneficial effects:

[0014] (1) A machine vision-based intelligent inspection and management system for bearing parts, through a phase order correction and reconstruction module, generates a regularized weighting factor based on the high-frequency phase gradient amplitude during the phase unwrapping process of the interference fringe image. This effectively suppresses the order jump error in the curvature change region of the bearing raceway and significantly improves the restoration accuracy of the three-dimensional topography data in key areas such as the groove edge. At the same time, the coupling analysis module introduces a multi-modal feature fusion mechanism of curvature entropy and polarization scattering intensity, which can accurately identify the type of foreign object in the background of complex material mixing and achieve sub-pixel-level positioning of foreign objects in the bearing contact ellipsoid area, overcoming the limitations of traditional image processing methods in terms of foreign object recognition accuracy and positioning accuracy.

[0015] (2) A machine vision-based intelligent inspection and management system for bearing parts combines the spatial distribution of foreign objects with the Hertz contact stress field through a stress mapping decision module. It can calculate the degree of influence on fatigue life based on the differences in material and location of foreign objects, and construct a sorting priority decision map to realize the risk classification and optimal resource allocation of foreign object handling strategies. At the same time, the pollution source reverse optimization module performs spatial clustering analysis on the coordinate set of foreign objects and realizes adaptive adjustment of dust removal equipment parameters in combination with the assembly unit layout. This constructs a closed-loop control chain of inspection-sorting-dust removal, which solves the problem that the existing inspection system cannot link process adjustment and causes pollution recurrence, and significantly improves the process control capability of assembly cleanliness.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of an intelligent inspection and management system for bearing parts based on machine vision, according to the present invention. Detailed Implementation

[0018] This application provides an intelligent inspection and management system for bearing parts based on machine vision, which solves the problems of low accuracy in identifying contaminants on bearing raceways, inaccurate hazard level assessment, unreasonable sorting strategies, and weak ability to trace contaminant sources in existing technologies.

[0019] The overall concept of the solution in this application embodiment is as follows:

[0020] High-precision three-dimensional topography of bearing raceways is obtained based on the principle of interferometry. The phase order correction and reconstruction module dynamically constrains the unwrapping error, thereby improving the topography restoration capability of curvature change zones.

[0021] Subsequently, the coupling analysis module is used to jointly analyze the local curvature entropy and polarization scattering intensity to achieve accurate classification, identification and spatial positioning of foreign objects of various materials; then, the stress mapping decision module maps the position of the foreign object to the bearing contact stress field, and calculates its impact on fatigue life in combination with the type of foreign object, generating a sorting priority map.

[0022] Finally, by using the pollution source reverse optimization module to cluster and model the spatial distribution of foreign objects, and combining the association relationship of assembly stations, the working parameters of the dust removal equipment are dynamically adjusted to achieve closed-loop collaborative control from detection to dust removal, ensuring the stability and consistency of the cleanliness of bearing parts.

[0023] Please see Figure 1 This invention provides a technical solution: an intelligent inspection and management system for bearing parts based on machine vision, comprising the following modules: a phase order correction and reconstruction module, a coupling analysis module, a stress mapping decision module, and a pollution source reverse optimization module. The phase order correction and reconstruction module generates a regularization weighting factor based on the high-frequency phase gradient amplitude when performing phase unwrapping on the interference fringe image, dynamically suppressing the order jump error in the steep region of the bearing raceway, and outputting three-dimensional morphology data and gradient field. The coupling analysis module synchronously calculates the local curvature entropy and polarization scattering intensity distribution of the three-dimensional morphology, identifies the type of foreign object based on the coupling relationship between the curvature entropy extremum and the scattering intensity threshold, and locates the coordinates of the foreign object in the bearing contact ellipsoidal region. The stress mapping decision module maps the coordinates of the foreign object to the Hertz contact stress distribution field, calculates the life decay value based on the stress amplification factor corresponding to the foreign object type, and generates a sorting priority decision map. The pollution source reverse optimization module performs spatial clustering analysis on the foreign object coordinate set, and adjusts the working parameters of the dust removal equipment in reverse based on the correlation between the cluster center coordinates and the assembly equipment location.

[0024] In this implementation scheme, the phase order correction and reconstruction module processes the fringe image obtained from interferometry to extract high-precision three-dimensional topographic information of the bearing raceway. Its main function is to dynamically correct the "order jump error" caused by abrupt curvature changes (such as at the edge of the groove) during phase unwrapping, ensuring the continuity and accuracy of the measurement data in steep regions. Phase unwrapping refers to the process of restoring the periodic phase information in the interferogram to the actual continuous height values. The original interferogram only provides phase values ​​from 0 to 2π, requiring an unwrapping algorithm to restore the true surface contour of the object. Order jump error: When there are abrupt changes in height on the object's surface (such as sharp edges), the phase unwrapping algorithm is prone to errors, causing misjudgment of the phase order, manifesting as local distortion. High-frequency phase gradient amplitude: Represents the phase change rate between adjacent pixels, reflecting the detailed changes on the object's surface, and is used to assist in identifying areas prone to errors. Regularization weighting factor: Used in mathematical optimization to balance the smoothness and fidelity of the solution. In this module, the error penalty weight for high-gradient regions is dynamically adjusted through this factor to achieve more robust morphology restoration. Coupled Analysis Module: This module, based on a multimodal image data fusion strategy, simultaneously analyzes the geometric features (curvature entropy) and optical reflection features (polarization scattering intensity) of the 3D morphology to identify contaminant types (such as metal debris, fibers, etc.) and accurately determine their location in the bearing raceway contact area. Local Curvature Entropy: An indicator describing the geometric complexity of a surface, quantifying disturbed areas such as foreign object edges and protrusions. Regions with large curvature and rapid changes are usually where contaminants or defects are located. Polarization-Resolved Scattering Intensity: By analyzing the scattering intensity distribution after changes in the polarization state of light waves, the material differences of contaminants are extracted; for example, metals have high reflectivity, and fibers have strong anisotropy. Coupled Analysis: Refers to the joint processing of data from multiple modes (morphology + optics), identifying foreign object types through rules or learning models, and improving identification accuracy. Contact Ellipsoidal Region: The area where the rollers and raceways contact in a rolling bearing is typically approximately ellipsoidal. Contaminants in this region are more likely to cause stress concentration and should be prioritized for detection and location. Stress Mapping Decision Module: This module projects the coordinates of foreign objects onto the bearing's contact stress distribution model, assesses their impact on fatigue life based on their type and location, and generates a sorting priority map to guide subsequent processing (such as rejection or reprocessing). Hertzian Contact Stress: A classic theory describing the stress distribution within the contact area of ​​two elastic bodies (such as rollers and raceways), used to model local contact load distribution. Stress Amplification Factor: When foreign objects are present in the contact area, they cause local amplification of the contact load. This factor is determined based on the shape, material, and location of the foreign object and is used to estimate the actual stress level.Lifespan Decay Value: By substituting the local stress changes caused by foreign objects into the bearing fatigue life model, the potential lifespan loss due to contamination is calculated. Sorting Priority Decision Map: All detected foreign objects are sorted according to their lifespan impact, forming a heatmap or priority list to guide resource allocation in the automated sorting system. Contamination Source Reverse Optimization Module: This module identifies potential contamination source locations by performing cluster analysis on the spatial distribution of multiple foreign objects, combined with the layout of equipment such as assembly stations and tool tables, and adjusts the operating parameters of dust removal equipment accordingly to achieve dynamic contamination control. Spatial Cluster Analysis: Using statistical or machine learning methods (such as DBSCAN or K-Means), the coordinates of foreign objects are clustered to extract concentrated areas of contaminants, thus reflecting their source direction. Cluster Center Coordinates: The statistical center point of multiple contaminants in space, which may correspond to actual contamination sources (such as operating holes, lubrication stations, etc.). Assembly Equipment Location Correlation: Refers to the spatial mapping relationship between the foreign object cluster center and the workshop equipment layout, used to infer the likelihood of equipment causing contamination. Dust removal equipment operating parameters, such as air volume, frequency, and cleaning cycle, are adjusted through feedback to improve the efficiency and targeting of cleanliness control.

[0025] Specifically, when performing phase unwrapping on the interference fringe image, the process of generating a regularized weighting factor based on the high-frequency phase gradient magnitude is as follows: Fourier frequency domain filtering is performed on the interference fringe image to separate the phase component with the highest spatial carrier frequency as the high-frequency phase field; the gradient magnitude of the high-frequency phase field in the x / y direction is calculated, and gradient regions exceeding the characteristic curvature threshold of the bearing raceway are marked as steep sensitive regions; the gradient magnitude is input into an exponential decay function to generate a regularized weighting factor that is inversely proportional to the gradient magnitude, wherein the weighting factor of the steep sensitive region approaches zero to forcibly suppress order jumps.

[0026] In this implementation scheme, a two-dimensional Fourier transform is performed on the acquired interference fringe image I(x,y): I(x,y): Interference fringe image acquired in the spatial domain, a two-dimensional function; (x,y): Spatial coordinates of the image; F{}: Two-dimensional Fourier transform operator; Frequency domain image representation; (u,v): frequency domain coordinates. The phase component with the highest spatial carrier frequency is extracted using a bandpass filter H(u,v). H(u,v): Bandpass filter, designed to retain the high-frequency components corresponding to the interference fringe frequencies; The filtered frequency domain image contains only high-frequency phase components. The high-frequency phase field is obtained through inverse Fourier transform. F -1 {}: Two-dimensional inverse Fourier transform operator; arg{}: Complex phase angle function; φHF(x,y): Recovered high-frequency phase field, used for subsequent gradient calculation. Calculate the approximate partial derivative of the gradient magnitude of the high-frequency phase field in the x-direction: The phase gradient along the horizontal direction; Φ HF (x+1,y),φ HF (x, y): High-frequency phase values ​​of adjacent pixels. Approximate partial derivative in the y-direction: The phase gradient along the vertical direction; φ HF (x,y+1),φ HF (x, y): High-frequency phase values ​​of adjacent pixels. Calculate gradient magnitude: G(x,y): Gradient magnitude, representing the intensity of local phase change; The square root operation synthesizes the magnitudes of the gradients in two directions. Steepness sensitivity regions are determined using the feature curvature threshold Tc. S(x,y): Binary sensitive region identifier, 1 indicates that the point is a steep sensitive region; T c The gradient threshold corresponding to the bearing raceway curvature is used to distinguish between normal and steep regions. The gradient magnitude G(x,y) is input into an exponential decay function to generate a regularization weight factor. The weights are calculated using the exponential decay function: W(x,y)=exp(-qαG(x,y)); W(x,y): basic regularization weight, the larger the gradient, the smaller the weight; exp{·}: natural exponential function, ensuring the weights are non-negative and monotonically decreasing; qα>0: coefficient adjusting the gradient decay intensity, controlling the rate of weight change; G(x,y): local gradient magnitude. The weights are adjusted using a sensitive region mask: W'(x,y)=W(x,y)×(1-S(x,y))+ε×S(x,y); W'(x,y): final regularization weight factor, used in the phase unwrapping algorithm; S(x,y): sensitive region mask, 1 represents a steep region, 0 represents a non-steep region. 0: Minimal positive value, forces the weight of steep regions to be close to zero, and suppresses order jumps.

[0027] Specifically, the process of dynamically suppressing the order jump error in the steep region of the bearing raceway and outputting three-dimensional topography data and gradient field is as follows: a regularization weighting factor is introduced as a penalty term into the least squares phase unwrapping algorithm to apply gradient continuity constraints in the steep sensitive region and reconstruct the absolute phase field without jumps; the height value of each point on the bearing surface is calculated based on the absolute phase field to generate three-dimensional topography data with sub-micron resolution; the Sobel operator convolution operation is performed on the three-dimensional topography data to output the height gradient field representing the surface curvature change.

[0028] In this implementation scheme, the least squares phase unwrapping algorithm with regularized weights introduces a weight matrix Q = [q mn As a penalty term, the constrained phase unwrapping problem is transformed into the following minimization problem: Φ m,n : The absolute phase value after unwrapping, a variable on a two-dimensional grid; q mn Regularization weight factor: the penalty weight at the corresponding point. The larger the weight, the stricter the suppression of jumps. The phase difference between adjacent pixels in the horizontal direction, and the wrapped phase difference after modulo 2π processing; The input is the phase difference between adjacent pixels in the vertical direction, processed modulo 2π to enclose the phase difference; m, n: indices of the two-dimensional grid points in the image. Gradient continuity constraints and iterative solutions: To ensure gradient continuity, strong weights are applied, especially in steep sensitive regions. The weights corresponding to sensitive regions in the weight matrix Q are close to zero, while the weights in other regions are larger, dynamically adjusting the jump penalty. This minimization problem is solved iteratively by solving a system of linear equations: AΦ = b; A: coefficient matrix, composed of the weight matrix and difference operators, ensuring the smoothness of the solution; Φ: the absolute phase vector to be solved, containing the phase values ​​of all pixels; b: a constant vector composed of enclosed phase differences. Mapping from absolute phase to 3D height: Based on the solved absolute phase Φ... m,n Calculate the bearing surface height value H m,n : H m,n : The three-dimensional height value at the corresponding point; λ: The wavelength constant of the laser or interference light; Φ m,n : Absolute phase value. This mapping utilizes the linear relationship between interference fringes and optical path difference to convert phase into physical height, achieving sub-micron resolution measurement. Gradient field calculation of 3D topography data: The Sobel operator is applied to the 3D height data H to calculate the gradient field and estimate the local curvature change of the surface: G x =S x *H,G y =S y *H, H: 3D height matrix; S x ,S y : Sobel operator kernel, used for edge detection in the horizontal and vertical directions respectively; *: convolution operator; G x G y : Gradient components of height in two directions; G H Gradient amplitude field, characterizing the intensity of local curvature changes.

[0029] Specifically, the process of simultaneously calculating the local curvature entropy and polarization scattering intensity distribution of the three-dimensional morphology is as follows: Based on the height gradient field of the three-dimensional morphology, an analysis window matching the curvature radius of the bearing raceway is dynamically generated with each pixel as the center, and the information entropy value of the height gradient within the window is calculated as the local curvature entropy; the Stokes vector in the interference optical path is analyzed, and the polarization scattering intensity component related to the surface micromorphology is separated to construct the scattering intensity distribution field; the curvature entropy distribution map and the scattering intensity distribution map are registered to the same bearing surface coordinate system through coordinate transformation.

[0030] In this implementation scheme, the calculation of local curvature entropy is based on the three-dimensional topographic height gradient field G = [g pq An analysis window W is dynamically generated, centered on pixel (p, q). pq The window size is matched to the local radius of curvature of the bearing raceway. The probability distribution P of the height gradient within the window is calculated. pq (k), defining the local curvature entropy E pq for: E pq : Local curvature entropy at position (p,q); P pq (k): Window W pq The probability of the k-th height gradient magnitude interval; K: the total number of discrete intervals divided by the height gradient magnitude; p, q: two-dimensional pixel coordinate indices. In the separation interference optical path of polarization scattering intensity components, the measured polarization state is represented by the Stokes vector S = [S0, S1, S2, S3]. T The scattering intensity component I, which is related to the microstructure of the bearing surface, is extracted using a polarization analyzer. scat : I scat S0: Polarization scattering intensity distribution field; S1, S2, S3: represent the linear and circular polarization components of the polarization state, respectively. Registration of curvature entropy with scattering intensity involves mapping the curvature entropy map E = [E...]. pq ] and scattering intensity diagram I=[I pq The coordinate transformation function T:(p,q)→(x,y) is used to map the data to a unified bearing surface coordinate system: (x,y)=T(p,q); this step achieves spatial registration of the data, ensuring that the two distribution images correspond in space.

[0031] Specifically, the process of identifying foreign object types and locating their coordinates in the bearing contact ellipsoid region based on the coupling relationship between curvature entropy extrema and scattering intensity threshold is as follows: A joint probability distribution function is constructed for the coordinates of curvature entropy extrema points and their corresponding scattering intensity values; a nonlinear boundary equation for determining metal debris and fibrous contaminants is defined. When the curvature entropy value at a certain coordinate point exceeds the curvature entropy threshold and the corresponding scattering intensity value also exceeds the scattering intensity threshold, the point is determined to be metal debris. When the curvature entropy value exceeds the curvature entropy threshold but the scattering intensity value is lower than the scattering intensity threshold, the point is determined to be fibrous contaminants. For areas where the curvature entropy value does not exceed the curvature entropy threshold, these areas are marked as background surface features and do not trigger foreign object determination. Based on the bearing's real-time operating parameters, including load and speed, the three-dimensional spatial boundary of the contact ellipsoid region is dynamically calculated. The identified foreign object coordinates are then converted to the relative position coordinate system of the contact ellipsoid region, and the hazard level is labeled according to the foreign object type and its location within the contact ellipsoid region.

[0032] In this implementation scheme, a joint probability distribution function and a nonlinear decision boundary are constructed for the set of extreme points of curvature entropy {(α m ,β m )} and the corresponding scattering intensity value {I m Construct a joint probability distribution function P(α,I), where: α m I is the local curvature entropy value at the m-th extreme point; m Let be the polarization scattering intensity value corresponding to this point; m = 1, 2, ..., M, where M is the total number of extreme points. Define a boundary function F(α,I) to distinguish the type of foreign object: F(α,I) = a·α 2 +be·I 2 +ceαI+dα+e·I+f=0;F(α,I)>0 is determined to be metal debris;F(α,I)≤0 and α>α th Determined as a fiber contaminant; α th This represents the curvature entropy threshold. Parameter descriptions: ae, be, ce, d, e, f: coefficients of the decision boundary function, determined by fitting training samples and experimental data; α: local curvature entropy; I: polarization scattering intensity; α th Curvature entropy threshold. Foreign object detection rules, the specific detection logic is described as follows: If α m >α th And I m >I th If α m >α th And I m ≤I th If α m ≤α th It was determined to be a background surface feature, with no foreign objects. Among them, I th Scattering intensity threshold. Calculation and coordinate transformation of the three-dimensional spatial boundary of the bearing contact ellipsoid region: The spatial boundary of the bearing contact ellipsoid region is dynamically calculated based on real-time operating parameters load L and rotational speed N. (X c ,Y c Z c ): ellipsoid center coordinates; a, b, c: lengths of the semi-axes of the ellipsoid's length, width, and height; f a ,f b ,f c The load L and rotational speed N are functions describing the dynamic changes in the ellipsoidal boundary dimensions. The coordinates (x, y) of the identified foreign object are then used. m ,y m ,z m Transform from the original coordinate system to the ellipsoidal relative coordinate system (u m ,v m ,w m ), satisfying the definition of ellipsoidal coordinates: R: Rotation matrix, used to convert the global coordinate system to an ellipsoidal coordinate system. Hazard level labeling is based on the type of foreign object and its relative position (u). m ,v m ,w m Within the contact ellipsoid region, based on empirical models, corresponding hazard level labels are assigned to the locations for subsequent lifetime decay and priority sorting decisions.

[0033] Specifically, the process of mapping the foreign object coordinates to the Hertz contact stress distribution field and calculating the life attenuation value based on the stress amplification factor corresponding to the foreign object type is as follows: Map the foreign object coordinates to the Hertz contact stress distribution field, calculate the continuous stress distribution in the contact ellipsoid region based on the bearing geometry parameters and real-time operating load, and extract the maximum stress value at the center coordinates of the foreign object; match the pre-calibrated stress amplification factor according to the foreign object type, with metal debris using a coefficient that increases nonlinearly with the embedding depth and fiber contaminants using a linear coefficient that increases proportionally with the contact area; input the amplified stress value into the bearing material fatigue characteristic curve, and calculate the life attenuation value in combination with the cumulative damage.

[0034] In this implementation scheme, the foreign object coordinates are mapped to the Hertz contact stress distribution field, and the stress distribution in the ellipsoidal contact area is calculated based on the bearing geometry parameters and real-time load. σ(ξ,η): Contact stress at point (ξ,η); σ0: Maximum contact stress at the center of the contact ellipsoid; ξ,η: Coordinates within the ellipsoidal contact surface, corresponding to the coordinates of the foreign object's center; ga,gb: Semi-major and semi-minor axes of the contact ellipsoid. The maximum stress value at the center of the foreign object is extracted from the coordinates (u,v) of the foreign object's center within the ellipsoidal plane, corresponding to the maximum contact stress: σ max =σ(u,v); Stress amplification factor matching: Select the corresponding stress amplification factor κ according to the foreign object type τ. τ Where: metal debris: the stress amplification factor increases nonlinearly with the embedding depth d, defined as: κ metal =κ0+κ1(1-e -λd ); Fiber contaminants: stress amplification factor and contact area A c Linear proportion, defined as: κ fiber =μA c +ν; Parameter explanation: κ0: Initial value of stress amplification for metal debris; κ1: Maximum amplification of metal debris; λ: Attenuation factor for metal debris embedding depth; d: Foreign object embedding depth; μ: Stress amplification ratio coefficient for fiber contaminants; ν: Reference amplification coefficient for fiber contaminants; A c : Contact area with foreign object. Calculation of magnified stress value: Combining the stress amplification factor, calculate the magnified stress value at the foreign object location: σ amp =κ τ ·σ maxThe life decay value is calculated using the bearing material fatigue characteristic curve SN curve (stress-life curve) combined with a cumulative damage model. Parameter explanation: D: Cumulative damage value; σ amp,m : The amplified stress value of the m-th load cycle; σ f : Material fatigue limit stress; γ: Material fatigue index; Δn m The m-th load cycle number. When the cumulative damage D reaches 1, it indicates the end of the bearing's life, and the life decay value is calculated from this.

[0035] Specifically, the process of generating the sorting priority decision map is as follows: A judgment space consisting of the foreign object hazard level and lifespan decay value is established. The foreign object characteristics are mapped to the two-dimensional judgment space, and the bearing surface sorting priority coding map is output. The coding rules are: Immediate sorting priority: when the lifespan decay value exceeds the first preset threshold, or the hazard level is high; Batch sampling priority: when the lifespan decay value exceeds the second preset threshold but does not exceed the first preset threshold, or the hazard level is medium; Release priority: when the lifespan decay value does not exceed the second preset threshold and the hazard level is low.

[0036] In this implementation scheme, the system establishes a judgment space composed of two core parameters: the hazard level of the foreign object and its lifespan decay value. The hazard level reflects the potential impact of the foreign object on bearing performance and safety, typically categorized into three levels: high, medium, and low. The lifespan decay value quantifies the specific impact of the foreign object on the bearing's lifespan. These two characteristics of the detected foreign object are mapped onto two-dimensional coordinate points in the judgment space, with one axis representing the lifespan decay value and the other representing the hazard level. By dividing the judgment space into regions, priority classification and coding of foreign objects at different locations on the bearing surface are achieved. The sorting priority coding rules are as follows: Immediate sorting priority: When the lifespan decay value corresponding to the foreign object exceeds the first preset threshold, or the hazard level of the foreign object is high, the foreign object is marked for immediate sorting, indicating that it needs to be prioritized to prevent serious damage to bearing performance. Batch sampling priority: When the lifespan decay value exceeds the second preset threshold but does not reach the first threshold, or the hazard level is medium, the foreign object is classified as batch sampling priority, indicating that it needs to be sampled and closely monitored, but does not need to be immediately removed. Release Priority: When the lifespan degradation value does not exceed the second preset threshold and the hazard level is low, the foreign object is marked for release, meaning that the foreign object has little impact on the bearing and can be allowed to pass without special treatment. Finally, the system outputs the above priority information in coded form to generate a sorting priority decision map covering the bearing surface, providing an intuitive and actionable guide for subsequent automated sorting and quality control.

[0037] Specifically, the pollution source reverse optimization module includes the following steps: analyzing the set of foreign object coordinates using an adaptive neighborhood radius density clustering algorithm to identify high-density pollution core areas; calculating the spatial distance between the cluster center point and the coordinates of the assembly line equipment, and marking the equipment corresponding to the minimum distance as a potential pollution source; generating a dust removal intensity control coefficient based on the physical area of ​​the clustered area and the number of foreign objects, and increasing the working air volume and negative pressure value of the dust removal equipment at the potential pollution source in proportion to the coefficient.

[0038] In this implementation scheme, the density clustering analysis of the foreign object coordinate set employs an adaptive neighborhood radius density clustering algorithm (such as the improved DBSCAN) to cluster the foreign object spatial coordinate set, thereby identifying high-density distribution areas of foreign objects, i.e., the core pollution areas. This method can automatically adjust the neighborhood radius to adapt to different densities of foreign objects, avoiding mis-clustering or missed clustering. The specific steps are as follows: for each foreign object point q... m Define the neighborhood radius r m The density of its neighboring points is dynamically adjusted based on the density of the point q. m The number of points N contained in the neighborhood of . m If the density threshold η is exceeded, it is considered a core point. Core points are connected through their neighborhoods to form clusters. The spatial distance between the cluster center and the assembly line equipment coordinates is calculated. The equipment with the smallest distance is identified as a potential pollution source. The formula is expressed as: Where, d uv : Spatial distance between cluster center point u and device coordinate point v; x u ,y u ,z u : The three-dimensional coordinates of the cluster center point u; x v ,y v ,z v The three-dimensional coordinates of equipment coordinate point v. The generation and adjustment of the dust removal intensity control coefficient are based on the physical area A covered by the cluster. w Number of foreign objects N w Calculate the dust removal intensity control coefficient C w This quantifies the degree of pollution. The control coefficient proportionally adjusts the airflow and negative pressure value of the dust removal equipment at potential pollution sources. The calculation formula is: Among them, C w : Dust removal intensity control coefficient for the w-th cluster; N w : Number of foreign points within the w-th cluster; A w : Physical area of ​​the w-th cluster; α: Foreign matter density weighting coefficient, reflecting the impact of foreign matter quantity on control intensity; β: Baseline control coefficient, ensuring that the dust removal equipment maintains the lowest operating parameters. According to C wBased on the given values, the system automatically adjusts the airflow and negative pressure of the corresponding dust removal equipment to improve the dust removal efficiency in polluted areas.

[0039] In summary, this application has at least the following effects:

[0040] A machine vision-based intelligent inspection and management system for bearing parts effectively suppresses phase order jump errors in steep bearing raceway regions by introducing a regularization weighting factor based on high-frequency phase gradients, thus improving the accuracy and stability of 3D topography reconstruction. Combined with the coupled analysis of local curvature entropy and polarization scattering intensity, it achieves accurate identification of foreign object types and precise positioning within the bearing contact ellipsoidal region. Based on the Hertz contact stress distribution field and stress amplification factor derived from foreign object coordinate mapping, it accurately calculates bearing life attenuation values, providing scientific support for sorting priority decisions. Furthermore, by associating spatial clustering of foreign object coordinates with assembly equipment positions, it dynamically adjusts dust removal equipment parameters, achieving precise reverse optimization and intelligent control of pollution sources. This enhances the intelligence and automation level of bearing part inspection.

[0041] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0042] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0043] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0044] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0045] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0046] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A machine vision-based intelligent inspection and management system for bearing parts, characterized in that, It includes the following modules: phase order correction and reconstruction module, coupling analysis module, stress mapping decision module, and pollution source reverse optimization module; The phase order correction and reconstruction module is used to generate a regularized weighting factor based on the high-frequency phase gradient amplitude when performing phase unwrapping on the interference fringe image, dynamically suppress the order jump error in the steep region of the bearing raceway, and output three-dimensional topography data and gradient field. The coupling analysis module is used to simultaneously calculate the local curvature entropy and polarization scattering intensity distribution of the three-dimensional topography, identify the type of foreign object based on the coupling relationship between the curvature entropy extreme value and the scattering intensity threshold, and locate the coordinates of the foreign object in the bearing contact ellipsoid region. The stress mapping decision module is used to map the coordinates of foreign objects to the Hertz contact stress distribution field, calculate the lifetime decay value according to the stress amplification factor corresponding to the type of foreign object, and generate a sorting priority decision map. The pollution source reverse optimization module is used to perform spatial cluster analysis on the set of foreign object coordinates, and adjust the working parameters of the dust removal equipment in reverse according to the correlation between the coordinates of the cluster center and the location of the assembly equipment.

2. The intelligent inspection and management system for bearing parts based on machine vision according to claim 1, characterized in that: The specific process of generating regularization weighting factors based on high-frequency phase gradient magnitude when performing phase unwrapping on interference fringe images is as follows: Fourier frequency domain filtering is performed on the interference fringe image to separate the phase component with the highest spatial carrier frequency as the high-frequency phase field; Calculate the gradient magnitude of the high-frequency phase field in the x / y direction, and mark the gradient region that exceeds the characteristic curvature threshold of the bearing raceway as a steep sensitive region; The gradient magnitude is input into the exponential decay function to generate a regularization weight factor that is inversely proportional to the gradient magnitude. The weight factor in the steep sensitive region approaches zero to forcibly suppress order jumps.

3. The intelligent inspection and management system for bearing parts based on machine vision according to claim 2, characterized in that: The specific process for dynamically suppressing the order jump error in the steep region of the bearing raceway and outputting three-dimensional topography data and gradient field is as follows: The regularization weight factor is introduced as a penalty term into the least squares phase unwrapping algorithm to apply gradient continuity constraints in steep sensitive regions and reconstruct the absolute phase field without jumps. The height values ​​of each point on the bearing surface are calculated based on the absolute phase field, generating three-dimensional topographic data with sub-micron resolution; Perform Sobel operator convolution on the 3D topography data to output a height gradient field that represents the surface curvature variation.

4. The intelligent inspection and management system for bearing parts based on machine vision according to claim 1, characterized in that: The specific process for simultaneously calculating the local curvature entropy and polarization scattering intensity distribution of the three-dimensional topography is as follows: Based on the height gradient field of the three-dimensional topography, an analysis window matching the curvature radius of the bearing raceway is dynamically generated with each pixel as the center, and the information entropy value of the height gradient within the window is calculated as the local curvature entropy. By analyzing the Stokes vector in the interference optical path, the polarization scattering intensity components related to the surface micromorphology are separated, and the scattering intensity distribution field is constructed. The curvature entropy distribution map and the scattering intensity distribution map are registered to the same bearing surface coordinate system through coordinate transformation.

5. The intelligent inspection and management system for bearing parts based on machine vision according to claim 4, characterized in that: The specific process of identifying foreign object types and locating the coordinates of foreign objects in the bearing contact ellipsoid region based on the coupling relationship between the curvature entropy extremum and the scattering intensity threshold is as follows: Construct a joint probability distribution function of the coordinates of the curvature entropy extreme point and its corresponding scattering intensity value, and define a nonlinear decision boundary equation for metal debris and fiber contaminants; When the curvature entropy value at a certain coordinate point exceeds the curvature entropy threshold and the scattering intensity value corresponding to that point exceeds the scattering intensity threshold, the point is determined to be metal debris. When the curvature entropy value exceeds the curvature entropy threshold but the scattering intensity value is lower than the scattering intensity threshold, the point is determined to be fiber contaminant. For areas where the curvature entropy value does not exceed the curvature entropy threshold, they are marked as background surface features and foreign object determination is not triggered. Based on the real-time operating parameters of the bearing, including load and speed, the three-dimensional spatial boundary of the contact ellipsoid area is dynamically calculated. The coordinates of the identified foreign objects are converted to the relative position coordinate system of the contact ellipsoid area, and the hazard level is marked according to the type of foreign object and its position in the contact ellipsoid area.

6. The intelligent inspection and management system for bearing parts based on machine vision according to claim 5, characterized in that: The specific process of mapping the foreign object coordinates to the Hertz contact stress distribution field and calculating the lifetime decay value based on the stress amplification factor corresponding to the foreign object type is as follows: The coordinates of the foreign object are mapped to the Hertz contact stress distribution field. Based on the bearing geometric parameters and real-time operating load, the continuous stress distribution in the contact ellipsoid region is calculated, and the maximum stress value at the center coordinates of the foreign object is extracted. The pre-calibrated stress amplification factor is matched according to the type of foreign object. Metal debris is obtained through a coefficient that increases non-linearly with the embedding depth, while fibrous contaminants are obtained through a linear coefficient that increases proportionally with the contact area. The amplified stress value is input into the fatigue characteristic curve of the bearing material, and the life decay value is calculated in combination with the cumulative damage.

7. The intelligent inspection and management system for bearing parts based on machine vision according to claim 6, characterized in that: The specific process of generating a sorting priority decision map is as follows: Establish a judgment space consisting of foreign object hazard level and life decay value, map foreign object characteristics to the two-dimensional judgment space, and output a bearing surface sorting priority coding map. The coding rule is: immediate sorting priority: when the life decay value exceeds the first preset threshold, or the hazard level is high. Batch sampling priority: When the lifespan degradation value exceeds the second preset threshold but does not exceed the first preset threshold, or the hazard level is medium; Release priority: when the lifetime decay value does not exceed the second preset threshold and the hazard level is low.

8. The intelligent inspection and management system for bearing parts based on machine vision according to claim 1, characterized in that: The pollution source reverse optimization module includes the following steps: The set of foreign object coordinates is analyzed by an adaptive neighborhood radius density clustering algorithm to identify high-density pollution core areas; Calculate the spatial distance between the cluster center point and the coordinates of the assembly line equipment, and mark the equipment with the smallest distance as a potential pollution source; Based on the physical area of ​​the clustered region and the number of foreign objects, a dust removal intensity control coefficient is generated, and the working air volume and negative pressure value of the dust removal equipment at the potential pollution source are increased proportionally according to the coefficient.

Images