Machine vision-based bearing rolling element surface quality automatic detection method and system

Abstract

The present application relates to the technical field of precision parts intelligent manufacturing and machine vision detection, specifically to a bearing rolling element surface quality automatic detection method and system based on machine vision, comprising: collecting multi-channel image data of the bearing rolling element surface by using a wave band-angle correlation imaging system; the multi-channel image data contains light field response information of different wave bands and different incident angles; performing gradient difference calculation on the multi-channel image data to generate a topological gradient index; performing coaxial reflectivity extraction on the multi-channel image data to generate a material reflectivity index; calculating a defect comprehensive confidence score; determining a defect classification result; counting the gray scale distribution characteristics of the multi-channel image data; calculating the gray scale deviation according to the gray scale distribution characteristics and adjusting the light source driving power; the present application effectively solves the technical problem that the prior art is difficult to simultaneously capture and distinguish three-dimensional topographic features and two-dimensional material features in a single imaging, and significantly improves the detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and machine vision inspection technology for precision parts, specifically to an automatic inspection method and system for the surface quality of bearing rolling elements based on machine vision. Background Technology

[0002] With the continuous advancement of precision manufacturing technology, the surface quality inspection standards for bearing rolling elements, as core components, are becoming increasingly stringent; this demand for high-precision inspection presents many challenges to the automated identification of complex surface defects.

[0003] Currently, although machine vision technology has been widely used in industrial inspection, it still has significant limitations when dealing with highly reflective metal surfaces. Existing technologies often struggle to distinguish between defects of different properties in a single imaging process. Specifically, three-dimensional morphological defects and two-dimensional material defects often exhibit similar characteristics under conventional imaging, making it difficult for the system to accurately identify them. However, traditional imaging methods are often limited to the acquisition of a single physical property, which cannot effectively decouple light field information. This makes it difficult to accurately capture minute changes in surface normals and differences in material absorption while suppressing imaging instability caused by high reflectivity of the metal surface. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic inspection method for the surface quality of bearing rolling elements based on machine vision. This method can solve the problem of imaging instability caused by high reflectivity of metal surfaces, and can accurately distinguish between three-dimensional morphological defects and two-dimensional material defects in a single imaging. Specifically, the technical solution of this invention is as follows:

[0005] An automatic inspection method for bearing rolling element surface quality based on machine vision includes the following steps: S1, acquiring multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; wherein, the multi-channel image data contains light field response information of different bands and different incident angles; S2, performing gradient difference calculation on the multi-channel image data to generate a topological gradient index; performing coaxial reflectivity extraction on the multi-channel image data to generate a material reflectivity index; S3, calculating a comprehensive confidence score for defects based on the topological gradient index and the material reflectivity index; determining the defect classification result based on the comprehensive confidence score for defects; S4, statistically analyzing the grayscale distribution characteristics of the multi-channel image data; calculating the grayscale deviation based on the grayscale distribution characteristics and adjusting the light source driving power.

[0006] Preferably, S1 specifically includes: S11, configuring the imaging optical path, establishing a physical mapping between the illumination band and the spatial incident angle; setting a spectral spatial filter at the aperture stop of the dual telecentric lens; S12, using the spectral spatial filter to modulate the incident light, forming a discrete band-angle pair distribution; S13, based on the band-angle pair distribution, controlling the image sensor to acquire the light field response; outputting red light channel image data, green light channel image data, and blue light channel image data, combining them to generate multi-channel image data; wherein, the red light channel image data responds to negative angle incident light; the green light channel image data responds to zero angle coaxial light; and the blue light channel image data responds to positive angle incident light.

[0007] Preferably, S2 specifically includes: S21, extracting the gray values ​​of the red light channel image data and the blue light channel image data from the multi-channel image data; S22, calculating the difference between the gray values ​​of the red light channel image data and the blue light channel image data; calculating the sum of the gray values ​​of the red light channel image data and the blue light channel image data; S23, calculating the ratio of the difference to the sum; S24, obtaining a preset gradient gain coefficient; and using the gradient gain coefficient to weight the ratio to generate a topological gradient index.

[0008] Preferably, S2 further includes: S25, extracting the grayscale value of the green light channel image data from the multi-channel image data; S26, obtaining a preset average grayscale reference value of the reference flat surface; S27, calculating the ratio of the grayscale value of the green light channel image data to the average grayscale reference value of the reference flat surface, and generating a material reflectivity index.

[0009] Preferably, S3 specifically includes: S31, obtaining a preset deformation threshold and a preset material threshold; S32, activating the first weight coefficient when the absolute value of the topological gradient index is greater than or equal to the deformation threshold; setting the first weight coefficient to zero when the absolute value of the topological gradient index is less than the deformation threshold; S33, activating the second weight coefficient when the material reflectivity index is less than the material threshold; setting the second weight coefficient to zero when the material reflectivity index is greater than or equal to the material threshold; S34, linearly summing the activated first weight coefficient and the activated second weight coefficient to generate a comprehensive defect confidence score.

[0010] Preferably, S3 further includes: S35, obtaining a preset gradient saturation constant; S36, when the absolute value of the topological gradient exponent is greater than the gradient saturation constant, forcing the scoring item corresponding to the second weight coefficient to zero; S37, when the absolute value of the topological gradient exponent is less than or equal to the gradient saturation constant, retaining the scoring item corresponding to the second weight coefficient; wherein, the scoring item is used to characterize the confidence level of corrosion-type defects.

[0011] Preferably, S4 specifically includes: S41, statistically analyzing the maximum grayscale value of multi-channel image data within the region of interest; S42, obtaining a preset target grayscale peak value; calculating the normalized grayscale deviation between the maximum grayscale value and the target grayscale peak value; S43, obtaining a preset adjustment step gain and a preset maximum adjustable power range of the light source; calculating the light source driving power correction amount using the normalized grayscale deviation, adjustment step gain, and maximum adjustable power range of the light source; S44, updating the current light source driving power based on the light source driving power correction amount.

[0012] The machine vision-based automatic inspection system for the surface quality of bearing rolling elements includes the following modules: a data acquisition module for acquiring multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; a feature decoupling module for performing gradient difference calculation on the multi-channel image data to generate a topological gradient index; the feature decoupling module is also used to extract coaxial reflectance from the multi-channel image data to generate a material reflectance index; a defect rating module for calculating a comprehensive confidence score for defects based on the topological gradient index and the material reflectance index; the defect rating module is also used to determine the defect classification result based on the comprehensive confidence score; and a light field control module for statistically analyzing the grayscale distribution characteristics of the multi-channel image data; the light field control module is also used to calculate the grayscale deviation based on the grayscale distribution characteristics and adjust the light source driving power.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] 1. This invention constructs a band-angle-correlated imaging system, establishing a physical mapping between the illumination band and the spatial incident angle in the imaging optical path, thereby achieving the synchronous acquisition of multi-dimensional light field information in a single exposure. This design utilizes a spectral spatial filter to modulate light of different colors to different incident angles, effectively solving the technical problem that existing technologies cannot simultaneously capture and distinguish three-dimensional morphological features and two-dimensional material features in a single imaging, and significantly improving detection efficiency.

[0015] 2. This invention employs a feature decoupling algorithm to separate mixed light field data into independent physical properties. By performing gradient difference calculation on the red and blue channels to generate a topological gradient index, and using the green light channel to extract coaxial reflectivity to generate a material reflectivity index, the system can accurately distinguish deformation features with surface normal changes from material features with light absorption differences. This decoupling mechanism eliminates mutual interference between different types of defect features and improves the accuracy of detection.

[0016] 3. This invention introduces a gradient saturation constant and a multidimensional confidence scoring model, which effectively prevents misjudgment of complex defects. To address the problem that deep scratches are easily misjudged as rust spots due to shadows caused by light occlusion, the system dynamically masks material scoring items by determining whether the topological gradient is saturated. This logic ensures that only deformation features are responded to in high gradient regions, thereby achieving accurate classification of deformation defects such as deep pits and scratches and material defects such as rust and stains.

[0017] 4. This invention designs an adaptive closed-loop control mechanism for light source power based on grayscale distribution characteristics, which solves the problem of imaging instability caused by high reflectivity of metal surfaces. The system dynamically adjusts the light source driving power by statistically analyzing the grayscale peak deviation of the region of interest in the image in real time, ensuring that the imaging is always maintained in the optimal linear response range of the sensor. This effectively avoids overexposure or underexposure caused by changes in reflectivity, and ensures the stability of feature extraction under different surface finishes. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention;

[0019] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0021] Example 1:

[0022] Please see Figure 1 An automatic inspection method for bearing rolling element surface quality based on machine vision includes the following steps: S1, acquiring multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; wherein, the multi-channel image data contains light field response information of different bands and different incident angles; S2, performing gradient difference calculation on the multi-channel image data to generate a topological gradient index; performing coaxial reflectivity extraction on the multi-channel image data to generate a material reflectivity index; S3, calculating a comprehensive confidence score for defects based on the topological gradient index and the material reflectivity index; determining the defect classification result based on the comprehensive confidence score for defects; S4, statistically analyzing the grayscale distribution characteristics of the multi-channel image data; calculating the grayscale deviation based on the grayscale distribution characteristics and adjusting the light source driving power.

[0023] This embodiment provides an automatic detection method for the surface quality of bearing rolling elements based on machine vision. Addressing the technical challenge of simultaneously distinguishing between three-dimensional morphological defects and two-dimensional material defects in a single image, and the imaging instability caused by high reflectivity of metal surfaces, this embodiment constructs a complete detection logic. The specific steps are as follows: Step S1: Multi-channel image data of the bearing rolling element surface is acquired using a band-angle correlation imaging system. In this step, the multi-channel image data not only contains traditional color spectral information but is also endowed with spatial angular attributes, i.e., it contains light field response information strongly correlated with different bands and different incident angles, thus realizing the physical encoding of the light field data. Step S2: Feature decoupling operation is performed on the acquired multi-channel image data. This step aims to separate the mixed light field information into independent physical attributes. On one hand, gradient difference calculation is performed on the multi-channel image data to generate a feature decoupling operation for the surface. The system calculates the topological gradient index, which represents the degree of inclination of the surface normal vector along the principal axis of illumination. On the other hand, it extracts coaxial reflectance using specific coaxial components from the multi-channel image data to generate a material reflectance index that characterizes the surface's light absorption and reflection capabilities. Step S3 is executed, where a comprehensive defect confidence score is calculated based on the decoupled topological gradient index and material reflectance index. This score quantifies the probability of defects existing in the current detection area. Based on this comprehensive defect confidence score, the system can determine the final defect classification result, thereby automatically distinguishing between deformation defects such as scratches and pits and material defects such as rust and stains. Simultaneously, to ensure the stability of the imaging system, step S4 is executed to statistically analyze the grayscale distribution characteristics of the multi-channel image data. The system calculates the grayscale deviation based on these characteristics and uses this as feedback to dynamically adjust the light source driving power, ensuring that image acquisition is always within the sensor's optimal linear response range.

[0024] S1 specifically includes: S11, configuring the imaging optical path, establishing a physical mapping between the illumination band and the spatial incident angle; setting a spectral spatial filter at the aperture stop of the dual telecentric lens; S12, using the spectral spatial filter to modulate the incident light, forming a discrete band-angle pair distribution; S13, based on the band-angle pair distribution, controlling the image sensor to acquire the light field response; outputting red light channel image data, green light channel image data, and blue light channel image data, combining them to generate multi-channel image data; wherein, the red light channel image data responds to negative angle incident light; the green light channel image data responds to zero angle coaxial light; and the blue light channel image data responds to positive angle incident light.

[0025] This embodiment further specifies step S1 above, and elaborates in detail the physical implementation mechanism of band-angle correlation imaging.

[0026] In step S11, the imaging optical path is configured to establish a physical mapping between the illumination band and the spatial incident angle. Specifically, a spectral spatial filter is set at the aperture stop of the dual telecentric lens. The aperture stop, as the frequency domain transformation plane of the optical system, is a key position for controlling the incident angle of light. The spectral spatial filter consists of three parallel strip-shaped filter areas, and the arrangement direction of the filter areas is perpendicular to the axial direction of the bearing rolling element. The specific structure is as follows: the left side is the red light filter area that transmits the red light band, corresponding to negative angle incident light; the middle side is the green light filter area that transmits the green light band, corresponding to zero angle coaxial light; and the right side is the blue light filter area that transmits the blue light band, corresponding to positive angle incident light.

[0027] In step S12, the incident light is modulated using a spectral spatial filter to form a discrete distribution of band-angle pairs; through the gating effect of the filter, the system constructs a band distribution. With the angle of incidence The strong constraint relationship: the red light band is restricted to the negative angle region, for example... The green light band is restricted to the zero-angle region, for example... The blue light band is restricted to a positive angle region, for example... .

[0028] In step S13, based on the above band-angle pair distribution, the image sensor is controlled to acquire the light field response; the sensor outputs red light channel image data, green light channel image data, and blue light channel image data, and combines them to generate multi-channel image data; among them, the red light channel image data responds to negative angle incident light and mainly carries the shadow and highlight information under left side illumination; the green light channel image data responds to zero angle coaxial light and mainly carries the reflectivity information of the surface itself; the blue light channel image data responds to positive angle incident light and mainly carries the shadow and highlight information under right side illumination; this physical mapping ensures that the subsequent algorithm can infer the microscopic three-dimensional normal changes of the object surface through the differences in color channels.

[0029] S2 specifically includes: S21, extracting the grayscale values ​​of the red light channel image data and the blue light channel image data from the multi-channel image data; S22, calculating the difference between the grayscale values ​​of the red light channel image data and the blue light channel image data; calculating the sum of the grayscale values ​​of the red light channel image data and the blue light channel image data; S23, calculating the ratio of the difference to the sum; S24, obtaining a preset gradient gain coefficient; and using the gradient gain coefficient to weight the ratio to generate a topological gradient index.

[0030] This embodiment is a detailed description of the process of generating the topological gradient exponent in step S2 above.

[0031] In step S21, the grayscale value of the red light channel image data is extracted from the multi-channel image data. Gray values ​​of blue channel image data These two channels correspond to symmetrical negative and positive angle incident light, respectively.

[0032] In steps S22 and S23, a differential normalization model is constructed; since the change in the normal vector of the defect surface with a three-dimensional morphology will lead to and The two exhibit inverse changes, while changes in the material's reflectivity will cause both to change proportionally; therefore, this embodiment calculates the difference between the grayscale values ​​of the red channel and the blue channel. and the sum of the two. And calculate the ratio of the difference to the sum.

[0033] In step S24, the ratio is weighted using a preset gradient gain coefficient to generate the topological gradient exponent. The specific calculation formula is as follows: ;in, Representing coordinates The topological gradient exponent at a given location is a dimensionless quantity. This represents the gradient gain coefficient, which originates from the system calibration phase. The specific calibration method is as follows: select a system with a known depth... ,For example The standard step sample was imaged, and the unweighted original ratio of the red and blue channels was calculated. Set the target gradient exponent For example, 0.8, then calculate This is how it was determined. Used as a fixed parameter for subsequent detection, it maps the ratio to a suitable numerical range to enhance the contrast of small deformations; This represents the numerical stability constant, and its value is usually set to an extremely small positive number (e.g., ...). This formula aims to prevent computational overflow caused by a zero denominator; through this formula, the gray dimensions in the numerator and denominator cancel each other out, and the material reflectivity as a common-mode signal is effectively suppressed, thereby... Purely characterizes the three-dimensional deformation features of the surface.

[0034] S2 also includes: S25, extracting the grayscale value of the green light channel image data from the multi-channel image data; S26, obtaining the preset average grayscale reference value of the reference flat surface; S27, calculating the ratio of the grayscale value of the green light channel image data to the average grayscale reference value of the reference flat surface, and generating the material reflectivity index.

[0035] This embodiment is a detailed description of the process of generating the material reflectivity index in step S2 above.

[0036] In step S25, the grayscale value of the green channel image data is extracted from the multi-channel image data. Because the green light channel is configured for zero-angle coaxial incidence, its imaging characteristics are not sensitive to minute changes in surface normal, but are extremely sensitive to the light absorption rate of the surface material.

[0037] In steps S26 and S27, the material reflectivity index is calculated. The system acquires the preset average grayscale reference value of a flat reference surface. This benchmark value During the system initialization phase, a constant is obtained by acquiring images of a standard, defect-free sample of the same material and calculating its average grayscale value. The ratio of the green channel grayscale value to the average grayscale reference value of the flat surface is calculated using the following formula: ;in, It is a dimensionless material reflectivity index; this index directly reflects the reflectivity of the test point relative to a standard surface; when the surface has rust, burns, or light-absorbing stains, It will deviate significantly from 1.0, thus enabling independent extraction of material features.

[0038] S3 specifically includes: S31, obtaining a preset deformation threshold and a preset material threshold; S32, activating the first weight coefficient when the absolute value of the topological gradient index is greater than or equal to the deformation threshold; setting the first weight coefficient to zero when the absolute value of the topological gradient index is less than the deformation threshold; S33, activating the second weight coefficient when the material reflectivity index is less than the material threshold; setting the second weight coefficient to zero when the material reflectivity index is greater than or equal to the material threshold; S34, linearly summing the activated first weight coefficient and the activated second weight coefficient to generate a comprehensive confidence score for the defect; S3 also includes: S35, obtaining a preset gradient saturation constant; S36, forcing the scoring item corresponding to the second weight coefficient to zero when the absolute value of the topological gradient index is greater than the gradient saturation constant; S37, retaining the scoring item corresponding to the second weight coefficient when the absolute value of the topological gradient index is less than or equal to the gradient saturation constant; wherein, the scoring item is used to characterize the confidence of corrosion-type defects.

[0039] This embodiment is a detailed description of the defect classification and confidence assessment in step S3 above.

[0040] In step S31, a preset deformation threshold is obtained. With preset material threshold ;in, The value is set based on the statistical characteristics of the background noise, for example, three times the standard deviation of the background noise. This is used to determine whether there is significant three-dimensional deformation; The value is set based on the reflectivity distribution of the standard sample, for example, taking the reference brightness. 80% is used to determine whether there are significant light-absorbing material defects.

[0041] In steps S32 to S34, and in conjunction with steps S35 to S37, the system calculates the comprehensive confidence score of the defect. To accurately distinguish between deformation and material defects, and to prevent deep scratches from being misjudged as material defects due to the shadow effect, the gradient saturation constant... The calibration method is as follows: Select a sample with a standard depth scratch defect, such as a V-groove with a depth of 50 μm, for imaging, and calculate the maximum topological gradient exponent within the defect region. ,set up Where α is a preset safety factor, ranging from 0.8 to 0.9, to ensure that all light intensity attenuation at the center of the deep scratch is correctly classified as deformation features rather than material black spots; based on this, this embodiment introduces a gradient saturation constant. The following scoring model is constructed: ;in, This indicates the overall confidence score of the defect; and These represent the first weighting coefficient and the second weighting coefficient, respectively, and are typically set as follows: This reflects the priority for detecting deformation-related defects; This is a unit step function that outputs 1 when the independent variable is greater than or equal to 0, and 0 otherwise.

[0042] When the absolute value of the topological gradient exponent Greater than or equal to the deformation threshold When the first unit step function in the first term outputs 1, the first weighting coefficient is activated. This indicates the presence of deformation characteristics.

[0043] When the material reflectivity index Less than the material threshold At this point, the unit step function in the second term outputs 1, indicating the presence of a suspected material feature; at this time, the system further introduces a gradient saturation constant. Correction is made: if the absolute value of the topological gradient exponent... Greater than the gradient saturation constant Correction item It will approach zero or become zero, thus forcibly masking the second weighting coefficient. The corresponding scoring items; this mechanism effectively prevents deep scratches from being misjudged as rust due to dark areas caused by light obstruction; if the absolute value of the topological gradient exponent Less than or equal to the gradient saturation constant Then, the scoring item corresponding to the second weight coefficient is retained to accurately reflect material defects; the weighted sum of each part is used to generate the score. It enables accurate classification of complex defect types in a single calculation.

[0044] To achieve automatic defect classification, the system further executes the following logical judgment steps: defining deformation confidence components. Define material confidence components. ;like and The current area is determined to be a deformation defect, such as scratches or pits; if and The current area is determined to have a material defect, such as rust or stains; if and It was determined to be a composite defect.

[0045] The formula (3) is used to calculate... It is used only as a score to indicate the severity of the defect, and is used for subsequent comparison of good / defective product sorting thresholds.

[0046] S4 specifically includes:

[0047] S41. Calculate the maximum grayscale value of the multi-channel image data within the region of interest; S42. Obtain the preset target grayscale peak value; calculate the normalized grayscale deviation between the maximum grayscale value and the target grayscale peak value; S43. Obtain the preset adjustment step gain and the preset maximum adjustable power range of the light source; calculate the light source driving power correction amount using the normalized grayscale deviation, adjustment step gain, and maximum adjustable power range of the light source; S44. Update the current light source driving power based on the light source driving power correction amount.

[0048] This embodiment is a detailed description of the adaptive adjustment of the light source driving power in step S4 above.

[0049] In steps S41 and S42, the maximum grayscale value of the multi-channel image data within the region of interest is counted. Specifically, iterate through the coordinates of every pixel within the region of interest. Obtain the grayscale of its red channel respectively. Green channel grayscale With blue channel grayscale Calculate the maximum value among the three. And find the global peak of this maximum value among all coordinate points, as And obtain the preset target grayscale peak value. For example, it can be set to 90% of the sensor's saturation value; the system calculates the normalized grayscale deviation between the two. : ;in, The bit depth of the image sensor, for example, 8 bits. This represents the theoretical maximum gray level; the normalization process eliminates the influence of sensor range differences.

[0050] In steps S43 and S44, the light source is adjusted based on the discrete-time proportional control principle; and the preset adjustment step size gain is obtained. The value typically ranges from 0.1 to 0.5, which is related to the preset maximum adjustable power range of the light source. Using normalized grayscale deviation, adjusted step gain, and the maximum adjustable power range of the light source, the light source driving power correction is calculated, and the current time is updated. Light source driving power For the next moment power At system startup The system initializes the light source drive power to the preset nominal power. For example, when the maximum rated power of the light source reaches 50%, the system enters the cyclic detection and adjustment phase. ; for the calculated Perform boundary limiting processing: if Greater than the maximum adjustable power range of the light source The upper limit is then set to It equals the upper limit value; if If it is less than the minimum starting power of the light source, then let This is equal to the minimum starting power; through the above closed-loop control, when the minimum starting power is detected... Exceed When the system is in a certain state, it will automatically reduce the driving power; otherwise, it will increase the power. This mechanism ensures that the imaging grayscale is always maintained within the optimal dynamic range when detecting the rolling elements of highly reflective bearings, avoiding interference from overexposure or underexposure on feature extraction.

[0051] Please see Figure 2 An automatic inspection system for the surface quality of bearing rolling elements based on machine vision includes the following modules: a data acquisition module for acquiring multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; a feature decoupling module for performing gradient difference calculation on the multi-channel image data to generate a topological gradient index; the feature decoupling module is also used to perform coaxial reflectance extraction on the multi-channel image data to generate a material reflectance index; a defect rating module for calculating a comprehensive confidence score for defects based on the topological gradient index and the material reflectance index; the defect rating module is also used to determine the defect classification result based on the comprehensive confidence score; and a light field control module for statistically analyzing the grayscale distribution characteristics of the multi-channel image data; the light field control module is also used to calculate the grayscale deviation based on the grayscale distribution characteristics and adjust the light source driving power.

[0052] This embodiment provides an automatic inspection system for the surface quality of bearing rolling elements based on machine vision. The system is used to execute the aforementioned method. The system mainly includes the following modules: a data acquisition module, which integrates hardware components of a band-angle correlation imaging system, used to execute step S1 and its sub-steps, acquiring multi-channel image data containing light field information; a feature decoupling module, configured with an image processing unit, used to execute step S2 and its sub-steps, generating a topological gradient index through gradient difference calculation and generating a material reflectivity index through coaxial reflectivity extraction; a defect rating module, used to execute step S3 and its sub-steps, calculating a comprehensive confidence score for defects and determining the defect classification result based on preset thresholds and weight parameters; and a light field control module, used to execute step S4 and its sub-steps, achieving adaptive control of imaging quality by real-time monitoring of grayscale distribution characteristics and feedback adjustment of the light source driving power.

[0053] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automatic inspection method for the surface quality of bearing rolling elements based on machine vision, characterized in that, Includes the following steps: S1. Acquire multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; Among them, the multi-channel image data contains light field response information for different bands and different incident angles; S2. Perform gradient difference calculation on multi-channel image data to generate topological gradient exponent; Perform coaxial reflectance extraction on multi-channel image data to generate material reflectance index; S3. Calculate the comprehensive confidence score of defects based on the topological gradient index and the material reflectivity index; The defect classification results are determined based on the overall confidence score of the defects; S4. Statistical analysis of grayscale distribution characteristics of multi-channel image data; The grayscale deviation is calculated based on the grayscale distribution characteristics, and the driving power of the light source is adjusted accordingly.

2. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 1, characterized in that, S1 specifically includes: S11. Configure the imaging optical path. The imaging optical path establishes a physical mapping between the illumination band and the spatial incident angle. A spectral spatial filter is set at the aperture stop of the double telecentric lens; S12. Modulate the incident light using a spectral spatial filter to form a discrete distribution of band-angle pairs; S13. Based on the band-angle pair distribution, control the image sensor to acquire the light field response; Output red light channel image data, green light channel image data and blue light channel image data, and combine them to generate multi-channel image data; Among them, the red light channel image data responds to negative angle incident light; Green channel image data response to zero-angle coaxial light; The blue channel image data responds to incident light at a positive angle.

3. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 1, characterized in that, S2 specifically includes: S21. Extract the grayscale values ​​of the red light channel image data and the blue light channel image data from the multi-channel image data; S22. Calculate the difference between the gray values ​​of the red channel image data and the gray values ​​of the blue channel image data; Calculate the sum of the gray values ​​of the red channel image data and the blue channel image data; S23. Calculate the ratio of the difference to the sum; S24. Obtain the preset gradient gain coefficient; The ratio is weighted using gradient gain coefficients to generate the topological gradient exponent.

4. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 3, characterized in that, S2 also includes: S25. Extract the grayscale value of the green light channel image data from the multi-channel image data; S26. Obtain the preset average grayscale reference value of a flat reference surface; S27. Calculate the ratio of the gray value of the green light channel image data to the average gray value of the reference flat surface, and generate the material reflectivity index.

5. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 1, characterized in that, S3 specifically includes: S31. Obtain the preset deformation threshold and the preset material threshold; S32. When the absolute value of the topological gradient exponent is greater than or equal to the deformation threshold, the first weight coefficient is activated. When the absolute value of the topological gradient exponent is less than the deformation threshold, the first weight coefficient is set to zero. S33. When the material reflectivity index is less than the material threshold, activate the second weighting coefficient. When the material reflectivity index is greater than or equal to the material threshold, the second weighting coefficient is set to zero. S34. Linearly sum the activated first weight coefficient and the activated second weight coefficient to generate the comprehensive confidence score of the defect.

6. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 5, characterized in that, S3 also includes: S35. Obtain the preset gradient saturation constant; S36. When the absolute value of the topological gradient exponent is greater than the gradient saturation constant, the scoring item corresponding to the second weight coefficient is forced to zero. S37. When the absolute value of the topological gradient exponent is less than or equal to the gradient saturation constant, retain the scoring item corresponding to the second weight coefficient. Among them, the scoring item is used to characterize the confidence level of corrosion-related defects.

7. The automatic inspection method for bearing rolling element surface quality based on machine vision according to claim 1, characterized in that, S4 specifically includes: S41. Calculate the maximum gray value of multi-channel image data within the region of interest; S42. Obtain the preset target grayscale peak value; Calculate the normalized grayscale deviation between the maximum grayscale value and the target grayscale peak value; S43. Obtain the preset adjustment step gain and the preset maximum adjustable power range of the light source; The correction amount of the light source driving power is calculated by using the normalized gray level deviation, the adjustment step gain and the maximum adjustable power range of the light source; S44. Update the current light source driving power based on the light source driving power correction amount.

8. A machine vision-based automatic inspection system for the surface quality of bearing rolling elements, applied to any one of the machine vision-based automatic inspection methods for the surface quality of bearing rolling elements as described in claims 1 to 7, characterized in that, Includes the following modules: The data acquisition module is used to acquire multi-channel image data of the bearing rolling element surface using a band-angle correlation imaging system; The feature decoupling module is used to perform gradient difference calculation on multi-channel image data and generate a topological gradient exponent. The feature decoupling module is also used to perform coaxial reflectance extraction on multi-channel image data and generate material reflectance index; The defect rating module is used to calculate the overall confidence score of defects based on the topological gradient index and the material reflectivity index. The defect rating module is also used to determine the defect classification result based on the overall confidence score of the defect; The light field control module is used to statistically analyze the grayscale distribution characteristics of multi-channel image data. The light field control module is also used to calculate grayscale deviation based on grayscale distribution characteristics and adjust the light source driving power.

Images