A bearing groove roughness detection system and detection method

Abstract

This invention discloses a bearing raceway roughness detection system, including a frame, a bearing positioning mechanism mounted on the frame for fixing the position of the bearing under test, a multi-wavelength laser module mounted on the frame for generating at least two different wavelength laser beams to simultaneously irradiate the raceway surface of the bearing under test, an optical path adjustment mechanism connected to the multi-wavelength laser module and mounted on the frame for adjusting the emission angle of the multi-wavelength laser beams and combining them into an obliquely incident composite beam, a spectral analysis mechanism mounted on the frame and positioned on the reflected optical path of the obliquely incident composite beam for separating the multi-wavelength scattered light according to wavelength and collecting the scattered light information of each wavelength, and a polarization characteristic analysis module connected to the spectral analysis mechanism for analyzing the polarization characteristics of the scattered light of each wavelength and generating polarization characteristic parameters. The advantages are that it enables non-contact detection and has the advantages of high detection accuracy and high speed.

Description

Technical Field

[0001] This invention relates to the field of bearing testing technology, and in particular to a bearing raceway roughness testing system and method. Background Technology

[0002] As a core component of mechanical equipment, the surface roughness of the inner ring and outer raceway of bearings directly affects their service life, operational accuracy, and reliability. In high-precision mechanical equipment, even minute changes in the surface roughness of the bearing raceway can lead to decreased equipment performance, increased noise, and even serious mechanical failures. With the ever-increasing demands for product quality in modern manufacturing, establishing an efficient and accurate bearing raceway roughness detection system to achieve real-time quality monitoring during the production process is of great significance for improving bearing product quality, reducing production costs, and enhancing product competitiveness. Especially in automated production lines, online roughness detection systems have become a key technological requirement for modern bearing manufacturers to improve product quality and production efficiency.

[0003] Currently, bearing raceway roughness inspection primarily employs contact measurement methods, such as using profilometers and roughness meters to measure the surface through probe contact. These methods calculate roughness parameters by recording changes in surface height as the probe moves across the surface. Additionally, some companies utilize non-contact inspection techniques like optical interferometry and laser triangulation, inferring surface roughness by analyzing changes in optical signals. In recent years, with the development of machine vision technology, some research has begun exploring image processing-based surface quality inspection methods, evaluating roughness levels by analyzing surface texture image features.

[0004] However, existing technologies have many shortcomings: while contact measurement methods offer high accuracy, the contact between the probe and the workpiece surface can easily cause surface scratches, and the measurement speed is slow, failing to meet the online inspection requirements of production lines; traditional optical methods, such as interferometry, are subject to harsh environmental conditions and are easily affected by vibration and temperature, resulting in poor applicability in industrial settings; although laser triangulation is non-contact, it requires high surface reflectivity and has limited adaptability to different materials and surface conditions; existing image processing methods mostly rely on manually designed feature extraction algorithms, lacking adaptive learning capabilities and making it difficult to accurately identify subtle differences between different roughness levels. Furthermore, most existing inspection systems cannot be well integrated into automated production lines, lacking real-time performance and stability, making it difficult to meet the urgent needs of modern manufacturing for high-efficiency, high-precision online inspection. Summary of the Invention

[0005] The purpose of this invention is to provide a bearing raceway roughness detection system and method, which can achieve non-contact detection and has the advantages of high detection accuracy and high speed.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a bearing raceway roughness detection system, comprising a frame, and further comprising, A bearing positioning mechanism, mounted on the frame, is used to fix the position of the bearing being tested; A multi-wavelength laser module, mounted on the frame, is used to generate at least two different wavelength laser beams to simultaneously irradiate the surface of the bearing raceway under test. An optical path adjustment mechanism, connected to the multi-wavelength laser module and mounted on the frame, is used to adjust the emission angle of the multi-wavelength laser beam and combine it into an obliquely incident composite beam. A spectral analysis mechanism, mounted on the frame and positioned in the reflected light path of the obliquely incident composite beam, is used to separate multi-wavelength scattered light according to wavelength and collect scattered light information of each wavelength separately. A polarization characteristic analysis module, connected to the spectral analysis mechanism, is used to analyze the polarization characteristics of scattered light at each wavelength and generate polarization characteristic parameters; The frequency domain feature extraction module is connected to the spectral analysis mechanism and the polarization characteristic analysis module, and is used to perform frequency domain transformation on the scattered light information to generate frequency domain feature parameters. The multimodal fusion recognition module is connected to the frequency domain feature extraction module and is used to perform comprehensive analysis based on frequency domain feature parameters and polarization feature parameters to output bearing channel roughness detection results.

[0007] Preferably, the bearing positioning mechanism includes an inclined table, a micro-motion slide, and a stop. The inclined table is mounted on the frame and has a preset tilt angle. The micro-motion slide is disposed on the inclined table. The stop is fixed on the micro-motion slide and has a contact surface that mates with the outer circle of the bearing.

[0008] Preferably, the multi-wavelength laser module includes a first laser, a second laser, and a beam combiner. The first laser emits red light with a wavelength of 630-680nm, the second laser emits blue-violet light with a wavelength of 405-450nm, and the beam combiner combines the two laser beams into a coaxially propagating obliquely incident composite beam. The optical path adjustment mechanism includes a fixed base, a support frame, a clamping block, a rotating shaft, a mounting bracket, and a locking bolt. The fixed base is mounted on the table surface of the frame, the support frame is vertically mounted on the fixed base, the clamping block is fixed to the support frame by a first bolt, one end of the rotating shaft is fixed to the clamping block, the mounting bracket has a shaft hole and is fitted onto the other end of the rotating shaft to form a rotatable connection, and the locking bolt is set on the clamping block to lock the rotational position of the mounting bracket. The mounting bracket is used to mount a multi-wavelength laser module and adjust the incident angle of the laser beam by rotating the rotating shaft.

[0009] Preferably, the spectral analysis mechanism includes a spectrometer, a beam homogenizer, and an image acquisition device; The beam splitter is mounted on the frame and located in the reflection path of the multi-wavelength scattered light, and has an incident surface and an exit surface; The light-diffusing plate is disposed opposite to the exit surface of the beam-splitting element, and the light-diffusing plate includes an incident surface and an exit surface; The image acquisition device includes a photoelectric sensor array and a processing circuit. The photoelectric sensor array is arranged opposite to the light-emitting surface of the light-diffusing plate and maintains a distance between them. The photoelectric sensor array is divided into multiple acquisition areas, and each acquisition area corresponds to a different position on the light-emitting surface of the light-diffusing plate. The processing circuit is electrically connected to the photoelectric sensor array, and the output terminal of the processing circuit is connected to the input terminal of the polarization characteristic analysis module.

[0010] A method for detecting bearing raceway roughness, employing a bearing raceway roughness detection system, is characterized by comprising the following steps: S1: Bearing positioning steps: Fix the bearing to be tested in the preset position through the bearing positioning mechanism, and adjust the bearing position so that the target detection area of ​​the bearing groove is within the irradiation range of the multi-wavelength laser module. S2: Multi-wavelength laser irradiation steps: Activate the multi-wavelength laser module to generate at least two different wavelength laser beams, adjust the angle of the laser beams through the optical path adjustment mechanism, and irradiate the bearing raceway surface with the laser beams in an oblique incident manner. S3: Spectral Separation and Acquisition Steps: The spectral analysis unit receives the scattered light reflected from the bearing channel surface, separates the scattered light of different wavelengths through the beam splitter, acquires the intensity information of the scattered light of each wavelength through the image acquisition device, and transmits the acquired data to the polarization characteristic analysis module. S4: Polarization characteristic analysis steps: The polarization characteristic analysis module receives scattered light data of each wavelength transmitted by the spectral analysis mechanism, collects light intensity information at different polarization angles, calculates the degree of polarization and polarization angle parameters of scattered light of each wavelength, and transmits the polarization characteristic parameters to the frequency domain feature extraction module. S5: Frequency domain feature extraction step: The frequency domain feature extraction module receives the scattered light intensity information from the spectral analysis institution and the polarization feature parameters from the polarization characteristic analysis module, performs Fourier transform processing on the data, generates frequency domain feature parameters, and transmits the frequency domain feature parameters to the multimodal fusion recognition module; S6: Multimodal fusion identification steps: The multimodal fusion identification module receives frequency domain feature parameters and polarization feature parameters, performs fusion analysis on multiple feature parameters, and outputs the bearing channel roughness detection results.

[0011] Preferably, the specific process of step S2 is as follows: S21: Activate the multi-wavelength laser module, wherein the first laser outputs a laser beam with a wavelength of 630nm-680nm, and the second laser outputs a laser beam with a wavelength of 405nm-450nm. S22: Two laser beams are combined into a composite beam by a beam combiner; S23: Adjust the angle of the composite beam by means of the optical path adjustment mechanism so that the angle between the composite beam and the normal of the bearing groove surface is 50°-70°; S24: Adjust the output power of the first laser and the second laser to keep the power ratio of the two wavelength lasers within the range of 0.8:1 to 1.2:1; S25: The composite beam with adjusted angle and power continuously irradiates the bearing raceway surface in an oblique incidence manner.

[0012] Preferably, the specific process of step S3 is as follows: S31: The beam splitter receives the composite scattered light reflected from the bearing channel surface, the composite scattered light containing a first wavelength component from the first laser and a second wavelength component from the second laser. The beam splitter separates the composite scattered light by wavelength and outputs the spatially separated first wavelength beam and second wavelength beam from its exit surface. S32: The first wavelength beam and the second wavelength beam are emitted from the beam splitter at different angles and respectively irradiate different positions on the incident surface of the light homogenizer. The distance between the irradiation positions of the two beams on the incident surface of the light homogenizer is greater than 3mm. S33: The first wavelength beam passes through the homogenizing plate and forms a first light spot on its light-emitting surface; the second wavelength beam passes through the homogenizing plate and forms a second light spot on its light-emitting surface. S34: The first acquisition area of ​​the photoelectric sensor array of the image acquisition device acquires the light intensity signal of the first light spot, and the second acquisition area acquires the light intensity signal of the second light spot. S35: The processing circuit converts the light intensity signal collected by the photoelectric sensor array into a digital signal and transmits the digital signal containing the first wavelength and second wavelength information to the polarization characteristic analysis module at a frequency of not less than 15Hz.

[0013] Preferably, the specific process of step S4 is as follows: S41: The polarization characteristic analysis module receives digital signals from the image acquisition device processing circuit and resolves the digital signals into first wavelength light intensity data I1(x,y) and second wavelength light intensity data I2(x,y), where x and y are the pixel coordinates of the photoelectric sensor array; S42: Rotatable polarizers are respectively set in the optical path after the spectroscopic element of the spectral analysis mechanism and before the image acquisition device, so that the first wavelength beam and the second wavelength beam pass through independent polarizers respectively; starting from the initial position, each polarizer is rotated from 0° to 180° in 15° increments, and light intensity data at 13 polarization angles are collected to obtain the first wavelength light intensity sequence I1(θ) and the second wavelength light intensity sequence I2(θ). S43: Process the first wavelength light intensity sequence I1(θ), find the maximum value I1max and the minimum value I1min, and calculate the first degree of polarization:

[0014] The angle corresponding to I1max is determined as the first wavelength polarization angle φ1; S44: Process the second wavelength light intensity sequence I2(θ), find the maximum value I2max and the minimum value I2min, and calculate the second degree of polarization:

[0015] The angle corresponding to I2max is determined as the second wavelength polarization angle φ2; S45: For each pixel position (x, y) in the image, repeat steps S43 and S44 to obtain two-dimensional polarization distribution maps P1(x, y) and P2(x, y) respectively; S46: Calculate the average polarization degree of the entire image region ( , ) and standard deviation ( , Combining the first wavelength polarization angle φ1 and the second wavelength polarization angle φ2, the polarization characteristic parameter V is formed: The data is then transmitted to the frequency domain feature extraction module for fusion analysis.

[0016] Preferably, the specific process of step S5 is as follows: S51: The frequency domain feature extraction module receives the first wavelength light intensity data I1(x,y) and the second wavelength light intensity data I2(x,y), as well as the polarization feature parameter V output by the polarization characteristic analysis module; S52: Perform two-dimensional fast Fourier transform on I1(x,y) and I2(x,y) respectively to obtain the spectra F1(u,v) and F2(u,v), where u and v are spatial frequency coordinates; S53: Based on the spectrum F1(u,v), calculate the low-frequency energy E1, high-frequency energy E2, and the center position C1 of the spectrum energy of the first wavelength; S54: Based on the spectrum F2(u,v), calculate the low-frequency energy E3, high-frequency energy E4, and the center position C2 of the spectrum energy of the second wavelength; S55: Combine the frequency domain characteristic parameters [E1, E2, C1, E3, E4, C2] with the polarization characteristic parameter V to form a comprehensive feature vector Y, Y= The data is then transmitted to the multimodal fusion recognition module.

[0017] Preferably, the specific process of step S6 is as follows: S61: The multimodal fusion recognition module receives the comprehensive feature vector Y, standardizes each parameter in the vector, and maps each parameter to a value range of 0 to 1; S62: Access a pre-established roughness standard database, which stores the feature vectors of standard samples of different roughness grades and the corresponding measured roughness values, covering a roughness range from 0.05 micrometers to 3.2 micrometers; S63: Calculate the Euclidean distance between the current feature vector and the feature vector of each standard sample in the database, and select the 5 samples with the smallest distance as the nearest neighbor reference samples; S64: Determine the roughness value based on 5 nearest neighbor reference samples, extract the measured roughness values ​​corresponding to these 5 nearest neighbor samples and calculate their arithmetic mean, and use this mean as the predicted roughness value R for the current detection. a ; S65: Based on the roughness prediction value R a Determine the roughness grade: R a 0.2 micrometers or less is classified as precision grade. R a Sizes greater than 0.2 micrometers and less than or equal to 0.8 micrometers are classified as ordinary grade. R a If the value is greater than 0.8 micrometers, it is judged as rough.

[0018] Compared with existing technologies, the advantages of this invention are as follows: the system is based on a frame platform and integrates multiple functional modules. Among them, the bearing positioning mechanism is responsible for firmly fixing the bearing to be tested in a predetermined position, providing a stable measurement reference for subsequent optical inspection; the multi-wavelength laser module generates at least two different wavelength laser beams. This multi-wavelength design can obtain richer surface scattering information because different wavelengths of light have different sensitivities to surface roughness; the optical path adjustment mechanism adjusts and combines the multi-wavelength laser beams into an obliquely incident composite beam. The oblique incident method can enhance the scattering effect of surface microstructure on light and improve detection sensitivity.

[0019] When the composite beam illuminates the bearing channel surface, the spectral analysis mechanism captures the reflected scattered light and separates the scattered light of different wavelengths for independent acquisition. This wavelength-separated processing method preserves the unique surface information carried by each wavelength. The polarization characteristic analysis module further mines the polarization characteristics of the scattered light, because surface roughness affects the polarization state of the scattered light. By analyzing the polarization parameters, additional surface quality information can be obtained. Subsequently, the frequency domain feature extraction module performs frequency domain analysis such as Fourier transform on the scattered light information, converting the roughness information in the spatial domain to a frequency domain representation, which can better reflect the periodic structure and random roughness components of the surface. Finally, the multi-modal fusion recognition module comprehensively utilizes frequency domain features and polarization features for intelligent analysis. Through cross-validation and fusion judgment of multi-dimensional information, it outputs high-precision bearing channel roughness detection results, achieving a breakthrough over traditional single detection methods. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a three-dimensional structural diagram of the present invention; Figure 2 This is the front view of the present invention; Figure 3 This is a schematic block diagram of the circuit part in this invention; Figure 4 This is a schematic diagram of the process of the present invention; In the diagram, 1. Frame; 2. Bearing positioning mechanism; 3. Multi-wavelength laser module; 4. Optical path adjustment mechanism; 5. Spectral analysis mechanism; 6. Polarization characteristic analysis module; 7. Frequency domain feature extraction module; 8. Multimodal fusion recognition module; 9. Tilt stage; 10. Micro-motion slide; 11. Stop block; 12. Fixed base; 13. Support frame; 14. Clamping block; 15. Rotating shaft; 16. Mounting frame; 17. Locking bolt; 18. Beam splitter; 19. Beam homogenizer; 20. Image acquisition device. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Example 1: As Figures 1-2 As shown, a bearing raceway roughness inspection system includes a frame 1, and further includes... The bearing positioning mechanism 2 is mounted on the frame 1 and is used to fix the position of the bearing being tested. A multi-wavelength laser module 3, mounted on the frame 1, is used to generate at least two different wavelength laser beams to simultaneously irradiate the surface of the bearing raceway under test. The optical path adjustment mechanism 4 is connected to the multi-wavelength laser module 3 and mounted on the frame 1. It is used to adjust the emission angle of the multi-wavelength laser beam and combine it into an obliquely incident composite beam. The spectral analysis unit 5 is mounted on the frame 1 and set in the reflected light path of the obliquely incident composite beam. It is used to separate the multi-wavelength scattered light according to wavelength and collect the scattered light information of each wavelength separately. The polarization characteristic analysis module 6 is connected to the spectral analysis mechanism 5 and is used to analyze the polarization characteristics of scattered light at each wavelength and generate polarization characteristic parameters. The frequency domain feature extraction module 7 is connected to the spectral analysis mechanism 5 and the polarization characteristic analysis module 6, and is used to perform frequency domain transformation on the scattered light information to generate frequency domain feature parameters. The multimodal fusion recognition module 8 is connected to the frequency domain feature extraction module 7 and is used to perform comprehensive analysis based on frequency domain feature parameters and polarization feature parameters to output the bearing channel roughness detection results.

[0024] Preferably, the bearing positioning mechanism 2 includes an inclined table 9, a micro-motion slide 10, and a stop 11. The inclined table 9 is mounted on the frame 1 and has a preset tilt angle. The micro-motion slide 10 is disposed on the inclined table 9. The stop 11 is fixed on the micro-motion slide 10 and has a contact surface that mates with the outer circle of the bearing.

[0025] The bearing positioning mechanism 2 achieves stable fixing and precise adjustment of the bearing. The tilting stage 9 is directly mounted on the frame 1 as the bottom support structure. Its preset tilt angle makes the bearing raceway form a specific tilt angle relative to the horizontal plane, which not only facilitates the laser to irradiate the raceway surface at the optimal oblique incident angle, but also helps to effectively collect scattered light. At the same time, the tilted placement can enhance the positioning stability of the bearing by utilizing the effect of gravity.

[0026] The micro-motion slide 10, mounted as an intermediate adjustment layer on the tilting stage 9, provides precise displacement adjustment. This design allows the operator to make micron-level fine adjustments to the bearing position, ensuring that the groove inspection area of ​​different bearing models can be accurately aligned with the laser irradiation point. The introduction of the micro-motion slide 10 greatly improves the system's versatility and inspection accuracy, enabling the same equipment to adapt to the inspection needs of various bearing specifications.

[0027] As a positioning element that directly contacts the bearing, the stop block 11 has an arc-shaped contact surface, which forms a stable surface contact with the outer circle of the bearing. This contact method has obvious advantages over traditional point contact or line contact: the contact stress distribution is more uniform, avoiding bearing deformation that may be caused by local stress concentration; it provides greater friction and support area, enhancing the reliability of positioning; at the same time, the arc design also has an automatic centering function, so that the bearing can naturally fit into the correct position under the action of gravity after placement.

[0028] In this embodiment, the multi-wavelength laser module 3 includes a first laser, a second laser, and a beam combiner. The first laser emits red light with a wavelength of 630-680nm, and the second laser emits blue-violet light with a wavelength of 405-450nm. The beam combiner combines the two laser beams into a coaxially propagating obliquely incident composite beam.

[0029] The multi-wavelength laser module 3 adopts a dual-wavelength laser detection technology. The first laser is selected in the 630-680nm red light band. The longer wavelength of the red laser, when it encounters the surface microstructure, mainly reflects the overall roughness level of the surface and can provide basic roughness information. In addition, the laser in this wavelength range has good penetration and stability, and has good detection capability for large-scale surface undulations and macro-roughness features.

[0030] The second laser uses the 405-450nm blue-violet light band. This short-wavelength laser has a higher resolution for surface micro-details. The wavelength of blue-violet light is close to the short-wavelength limit of the visible spectrum, and its interaction with surface microstructures is more sensitive, enabling it to capture subtle roughness changes that are difficult to distinguish with red light.

[0031] The beam combiner precisely combines two laser beams of different wavelengths into a coaxially propagating composite beam. This coaxial design ensures that both wavelengths illuminate the exact same location on the bearing raceway surface, eliminating measurement errors caused by spatial deviations. Coaxial propagation also guarantees that the two wavelengths have the same incident angle and spot size, allowing subsequent scattered light analysis to be compared under completely consistent geometric conditions. Utilizing the complementary wavelength characteristics of red and blue-violet light, combined with beam combining technology, this module can simultaneously acquire macroscopic and microscopic information on surface roughness, providing rich and reliable raw data for subsequent multimodal analysis.

[0032] The optical path adjustment mechanism 4 includes a fixed base 12, a support frame 13, a clamping block 14, a rotating shaft 15, a mounting bracket 16, and a locking bolt 17. The fixed base 12 is mounted on the table of the frame 1. The support frame 13 is vertically mounted on the fixed base 12. The clamping block 14 is fixed to the support frame 13 by a first bolt. One end of the rotating shaft 15 is fixed to the clamping block 14. The mounting bracket 16 is provided with a shaft hole and is fitted onto the other end of the rotating shaft 15 to form a rotatable connection. The locking bolt 17 is provided on the clamping block 14 to lock the rotation position of the mounting bracket 16. The mounting bracket 16 is used to mount the multi-wavelength laser module 3 and adjust the incident angle of the laser beam by rotating the rotating shaft 15.

[0033] The optical path adjustment mechanism 4 enables continuous adjustment of the laser incident angle. The fixed base 12 is installed on the table of the frame 1, forming the basic support for the entire adjustment system; the support frame 13 is vertically set on the base, forming a rigid column structure to ensure the stability and load-bearing capacity of the adjustment mechanism in the vertical direction; the clamping block 14 is fixed to the support frame 13 by the first bolt, and this detachable connection method facilitates the installation and maintenance of the mechanism.

[0034] The rotating shaft 15 serves as an angle adjustment element. One end is fixed inside the clamping block 14 to form a rotation fulcrum, and the other end engages with the precision shaft hole of the mounting bracket 16 to form a smooth rotating pair. The mounting bracket 16 carries the multi-wavelength laser module 3. The continuous adjustment of the incident angle of the laser beam is achieved by the rotational motion around the rotating shaft 15. This design converts the angle change of the laser module into the rotational motion of the mounting bracket 16, ensuring the stability and controllability of the adjustment process.

[0035] Locking bolt 17 is mounted on clamping block 14. After angle adjustment, tightening action fixes mounting bracket 16 at the predetermined angle position. This locking mechanism effectively prevents angle deviation caused by vibration or external force during detection, ensuring consistency of measurement conditions. The entire optical path adjustment mechanism 4 achieves precise positioning and reliable fixation of the laser oblique incidence angle within a continuous range through the rotational freedom of rotating shaft 15 and the constraint of locking bolt 17.

[0036] In this embodiment, the spectral analysis mechanism 5 includes a spectrometer 18, a light homogenizer 19, and an image acquisition device 20; The beam splitter 18 is mounted on the frame 1 and is located in the reflection path of the multi-wavelength scattered light. It has an incident surface and an exit surface. The light homogenizing plate 19 is disposed opposite to the exit surface of the beam splitting element 18, and the light homogenizing plate 19 includes an incident surface and an exit surface. Image acquisition device 20 includes photoelectric sensor array and processing circuit. The photoelectric sensor array is arranged opposite to the light-emitting surface of light-diffusing plate 19 and maintains a distance. The photoelectric sensor array is divided into multiple acquisition areas, each of which corresponds to a different position on the light-emitting surface of the light-diffusing plate 19. The processing circuit is electrically connected to the photoelectric sensor array, and the output of the processing circuit is connected to the input of the polarization characteristic analysis module 6.

[0037] The spectral analysis unit 5 employs a three-stage optical processing architecture of spectral dispersion, homogenization, and acquisition, achieving effective separation and quantitative detection of multi-wavelength scattered light. The spectral dispersion element 18, as the core device for spectral separation, utilizes the principle of dispersion to spatially separate the composite scattered light containing red and blue-violet components according to wavelength. When the composite scattered light is incident on the incident surface of the spectral dispersion element 18, different wavelengths of light are deflected at different angles due to differences in refractive index, exiting from the exit surface at different angles, thus achieving spatial unfolding of the spectrum.

[0038] The introduction of the homogenizing plate 19 solves the technical problem of uneven intensity distribution of scattered light. Due to the microstructure of the bearing channel surface, the scattered light has spatial intensity fluctuations, and direct collection will cause measurement errors. The homogenizing plate 19 makes the originally uneven light intensity distribution form a relatively uniform light spot on the light-emitting surface, which improves the accuracy and repeatability of subsequent light intensity measurement. Furthermore, the relative arrangement of the homogenizing plate 19 and the exit surface of the beam splitter 18 ensures that the separated wavelength beams can be effectively incident and fully homogenized.

[0039] The image acquisition device 20 employs a photoelectric sensor array for parallel acquisition, a design that fully utilizes the spatial dispersion characteristics of the beam splitter 18. The photoelectric sensor array is divided into multiple independent acquisition areas, each corresponding to a specific wavelength of light spot, enabling simultaneous acquisition of multi-wavelength signals. The sensor array maintains an appropriate distance from the light-emitting surface of the homogenizing plate 19, avoiding potential optical interference from direct contact while ensuring the light spot completely covers the corresponding acquisition area. The processing circuit performs analog-to-digital conversion, noise filtering, and data preprocessing of the photoelectric signal, converting the original analog light intensity signal into a digital signal and outputting it to the polarization characteristic analysis module 6. This modular spectral analysis architecture achieves a complete conversion from composite scattered light to wavelength-division digital signals, laying the foundation for subsequent multi-dimensional feature extraction.

[0040] Example 2: Figure 3 As shown, a bearing raceway roughness detection method, using the bearing raceway roughness detection system in Embodiment 1, includes the following steps: S1: Bearing positioning steps: Fix the bearing to be tested in the preset position by the bearing positioning mechanism 2, and adjust the bearing position so that the target detection area of ​​the bearing groove is within the irradiation range of the multi-wavelength laser module 3. S2: Multi-wavelength laser irradiation steps: Activate the multi-wavelength laser module 3 to generate at least two different wavelength laser beams, adjust the angle of the laser beams through the optical path adjustment mechanism 4, and irradiate the bearing raceway surface with the laser beams in an oblique incident manner. S3: Spectral separation and acquisition steps: The spectral analysis unit 5 receives the scattered light reflected from the bearing channel surface, separates the scattered light of different wavelengths through the beam splitter 18, and acquires the intensity information of the scattered light of each wavelength through the image acquisition device 20. The acquired data is then transmitted to the polarization characteristic analysis module 6. S4: Polarization characteristic analysis steps: The polarization characteristic analysis module 6 receives the scattered light data of each wavelength transmitted by the spectral analysis mechanism 5, collects light intensity information at different polarization angles, calculates the degree of polarization and polarization angle parameters of the scattered light of each wavelength, and transmits the polarization characteristic parameters to the frequency domain feature extraction module 7. S5: Frequency domain feature extraction step: The frequency domain feature extraction module 7 receives the scattered light intensity information from the spectral analysis unit 5 and the polarization feature parameters from the polarization characteristic analysis module 6, performs Fourier transform processing on the data, generates frequency domain feature parameters, and transmits the frequency domain feature parameters to the multimodal fusion recognition module 8; S6: Multimodal fusion identification steps: The multimodal fusion identification module 8 receives frequency domain feature parameters and polarization feature parameters, performs fusion analysis on multiple feature parameters, and outputs the bearing channel roughness detection results.

[0041] In the above steps, the S1 bearing positioning step lays the physical foundation for the detection accuracy. This step not only achieves mechanical fixation of the bearing, but more importantly, through the preset angle of the tilting stage 9 and the precise adjustment of the micro-motion slide 10, it ensures the optimal geometric relationship between the test groove area and the laser beam path. This positioning method takes into account the cylindrical structure characteristics of the bearing, so that the groove surface can receive laser irradiation at an appropriate angle, while facilitating the effective collection of scattered light.

[0042] In step S2, the multi-wavelength laser irradiation step embodies the core principle of optical detection. It uses a combination of red light (630-680nm) and blue-violet light (405-450nm). The longer wavelength of red light primarily reflects larger-scale roughness features when interacting with the surface, while the shorter wavelength of blue-violet light is more sensitive to microstructures and can distinguish finer surface undulations. The oblique incidence method enhances the scattering effect of the surface microstructure, amplifying the originally weak roughness information. Furthermore, the angle optimization (50°-70° incident angle) achieved through the optical path adjustment mechanism 4 ensures sufficient scattering intensity while avoiding interference from specular reflection.

[0043] In step S3, the spectral separation and acquisition step solves the problem of parallel acquisition of multi-wavelength information. The beam splitter 18 uses the dispersion principle to spatially expand the composite scattered light, with different wavelengths forming spatially separated light spots on the exit surface. The application of the homogenizing plate 19 eliminates the intensity inhomogeneity within the scattered light spots; this homogenization process is crucial for accurately measuring the average scattered intensity. The array design of the image acquisition device 20 enables synchronous acquisition of multi-wavelength signals, avoiding measurement errors that may be caused by sequential scanning and improving detection efficiency.

[0044] In step S4, when light is scattered on a rough surface, different roughnesses will cause different changes in the polarization state of the scattered light. By collecting light intensity at multiple polarization angles and calculating the degree of polarization and polarization angle parameters, surface information that cannot be obtained by intensity measurement alone can be obtained.

[0045] In step S5, the Fourier transform converts the roughness distribution in the spatial domain into a frequency domain representation, which can clearly separate the contributions of different spatial frequency components. Low-frequency components correspond to the slow undulations and waviness of the surface, while high-frequency components reflect micro-roughness. The position of the spectral energy center comprehensively characterizes the main scale of roughness. This frequency domain analysis method is particularly suitable for identifying the mixture of periodic processing textures and random roughness.

[0046] In step S6, the multimodal fusion identification step constructs a multi-dimensional feature space by comprehensively utilizing spectral information (scattering intensity at different wavelengths), polarization information (degree of polarization and polarization angle), and frequency domain information (spectral characteristics). This fusion strategy leverages the complementarity between different physical mechanisms: spectral information reflects the overall roughness level, polarization information reveals surface anisotropy, and frequency domain information characterizes spatial distribution properties. Multimodal fusion not only improves detection accuracy but also enhances the system's ability to identify different types of surface defects, achieving a technological upgrade from simple roughness numerical measurement to comprehensive surface quality assessment.

[0047] Preferably, the specific process of step S2 is as follows: S21: Activate the multi-wavelength laser module 3, wherein the first laser outputs a laser beam with a wavelength of 630nm-680nm, and the second laser outputs a laser beam with a wavelength of 405nm-450nm; S22: Two laser beams are combined into a composite beam by a beam combiner; S23: Adjust the angle of the composite beam by means of the optical path adjustment mechanism 4 so that the angle between the composite beam and the normal of the bearing groove surface is 50°-70°. S24: Adjust the output power of the first laser and the second laser to keep the power ratio of the two wavelength lasers within the range of 0.8:1 to 1.2:1; S25: The composite beam with adjusted angle and power continuously irradiates the bearing raceway surface in an oblique incidence manner.

[0048] In step S21, the 630-680nm red light band is located at the long end of the visible spectrum. Laser technology in this band is mature, has stable output, and exhibits moderate sensitivity to macroscopic surface roughness features. The 405-450nm blue-violet light band is close to the ultraviolet region, with a wavelength approximately two-thirds that of red light. According to Rayleigh scattering theory, its scattering intensity is inversely proportional to the fourth power of the wavelength, thus making it more sensitive to minute surface undulations. This wavelength combination achieves scale complementarity in roughness detection: red light primarily detects roughness features with large Ra values, while blue-violet light can resolve submicron-level fine structures.

[0049] In step S22, the beam combiner uses a dichroic mirror or polarization beam combining principle to precisely combine two wavelengths of laser beams into a coaxially propagating composite beam. This spatial overlap ensures that the two wavelengths illuminate the exact same position in the bearing groove, eliminating spatial sampling differences. This allows subsequent multi-wavelength analysis to be based on the same surface area, improving the comparability and accuracy of measurements.

[0050] In step S23, the oblique incidence angle range of 50°-70° is the optimal range verified by theoretical calculations and experiments. Within this angle range, the intensity of surface scattered light is moderate, which avoids the strong interference of specular reflection when near normal incidence and prevents the problem of too weak scattered light when grazing incidence. According to optical scattering theory, the diffuse scattering component of rough surfaces dominates within this angle range, and the intensity of scattered light shows a good correlation with the surface roughness parameter.

[0051] In step S24, power ratio control ensures the balance of the detection signal. The power ratio range of 0.8:1 to 1.2:1 considers two factors: the difference in quantum efficiency of lasers of different wavelengths and the difference in response sensitivity of photoelectric sensors to different wavelengths. By adjusting the power ratio, the scattered light signals of the two wavelengths are kept at similar intensity levels, avoiding the problem of sensor saturation due to an excessively strong signal of one wavelength or a decrease in signal-to-noise ratio due to an excessively weak signal of one wavelength. This power balance also facilitates subsequent signal processing and feature extraction, giving the two wavelength data considerable weight.

[0052] In step S25, continuous illumination avoids timing synchronization issues that may arise from pulse measurements, ensuring a stable scattered light signal and facilitating subsequent multi-angle polarization analysis and time averaging. Stable illumination also reduces the impact of laser power fluctuations and mechanical vibrations on the measurement results, improving the repeatability of the detection. The entire S2 step, through precise control of laser parameters, lays a solid foundation for obtaining high-quality scattered light signals.

[0053] Preferably, the specific process of step S3 is as follows: S31: The beam splitter 18 receives the composite scattered light reflected from the bearing channel surface. The composite scattered light contains a first wavelength component from the first laser and a second wavelength component from the second laser. The beam splitter 18 separates the composite scattered light by wavelength and outputs the spatially separated first wavelength beam and second wavelength beam from its exit surface. S32: The first wavelength beam and the second wavelength beam are emitted from the beam splitter 18 at different angles and respectively irradiate different positions on the light-incident surface of the light-diffusing plate 19. The distance between the irradiation positions of the two beams on the light-incident surface of the light-diffusing plate 19 is greater than 3mm. S33: The first wavelength beam passes through the light homogenizing plate 19 and forms a first light spot on its light-emitting surface; the second wavelength beam passes through the light homogenizing plate 19 and forms a second light spot on its light-emitting surface. S34: The first acquisition area of ​​the photoelectric sensor array of the image acquisition device 20 acquires the light intensity signal of the first light spot, and the second acquisition area acquires the light intensity signal of the second light spot. S35: The processing circuit converts the light intensity signal collected by the photoelectric sensor array into a digital signal and transmits the digital signal containing the first wavelength and second wavelength information to the polarization characteristic analysis module 6 at a frequency of not less than 15Hz.

[0054] In the above steps, the spectral separation process in S31 utilizes the principle of dispersive optics. The spectral separation element 18 (such as a grating or prism) receives composite scattered light containing red and blue-violet light components. Due to the difference in refractive index of different wavelengths in the spectral separation material, different deflection angles are generated. The wavelength difference of approximately 250 nm between red light (630-680 nm) and blue-violet light (405-450 nm) ensures sufficient angular separation. This physical spectral separation method avoids the light energy loss and bandwidth limitations that may be caused by filter schemes, achieving highly efficient spectral separation.

[0055] In step S32, a spatial spacing of more than 3mm ensures that the two wavelength light spots will not overlap or crosstalk in the subsequent optical path, thus guaranteeing the independence of each wavelength signal. The larger spatial separation also facilitates the insertion of independent polarization analysis elements in the optical path, creating conditions for polarization characteristic analysis in step S4. At the same time, the appropriate spacing design takes into account the effective working area of ​​the light homogenizing plate 19 and the size of the photoelectric sensor array, realizing a compact layout of the optical system.

[0056] The homogenization process in step S33 solves the key problem of uneven intensity distribution of the scattered light spot. Due to the microscopic irregularities on the bearing channel surface, random intensity fluctuations exist within the scattered light spot, which introduces significant errors when measured directly. The homogenizing plate 19, through its internal diffuse structure or microlens array, transforms the uneven incident light spot into a light spot with a relatively uniform intensity distribution on the output surface. This optical homogenization not only improves the accuracy of the measurement but also enhances the system's tolerance to local surface defects, making the measurement results more representative of the average roughness level of the detection area.

[0057] S34's partitioned acquisition demonstrates the advantages of parallel detection. The partitioned design of the photoelectric sensor array allows for the simultaneous acquisition of two wavelength signals, avoiding time delays and synchronization errors that may arise from time-division multiplexing. The number and arrangement of pixels in each acquisition area are optimized to ensure complete coverage of the corresponding spot range. This spatially separated acquisition method also facilitates the implementation of different signal processing strategies, such as using different gain settings or noise filtering parameters for different wavelengths.

[0058] The S35's high-speed digital transmission ensures the system's real-time performance. A sampling frequency above 15Hz meets the needs of dynamic measurement, capturing transient changes that may occur during the detection process. This frequency setting considers multiple factors: the mechanical settling time of the bearing positioning, the fluctuation frequency of the laser power, and the data update rate required for subsequent polarization analysis. While performing analog-to-digital conversion, the processing circuit also performs necessary signal preprocessing, such as dark current compensation and nonlinear correction, ensuring that the polarization characteristic analysis module 6 receives a high-quality digital signal.

[0059] Preferably, the specific process of step S4 is as follows: S41: The polarization characteristic analysis module 6 receives the digital signal from the processing circuit in the image acquisition device 20 and parses the digital signal into first wavelength light intensity data I1(x,y) and second wavelength light intensity data I2(x,y), where x and y are the pixel coordinates of the photoelectric sensor array; S42: Rotatable polarizers are respectively set in the optical path after the spectroscopic element 18 of the spectral analysis unit 5 and before the image acquisition device 20, so that the first wavelength beam and the second wavelength beam pass through independent polarizers respectively; starting from the initial position, each polarizer is rotated from 0° to 180° in 15° increments, and light intensity data at 13 polarization angles are collected to obtain the first wavelength light intensity sequence I1(θ) and the second wavelength light intensity sequence I2(θ). S43: Process the first wavelength light intensity sequence I1(θ), find the maximum value I1max and the minimum value I1min, and calculate the first degree of polarization:

[0060] The angle corresponding to I1max is determined as the first wavelength polarization angle φ1; S44: Process the second wavelength light intensity sequence I2(θ), find the maximum value I2max and the minimum value I2min, and calculate the second degree of polarization:

[0061] The angle corresponding to I2max is determined as the second wavelength polarization angle φ2; S45: For each pixel position (x, y) in the image, repeat steps S43 and S44 to obtain two-dimensional polarization distribution maps P1(x, y) and P2(x, y) respectively; S46: Calculate the average polarization degree of the entire image region ( , ) and standard deviation ( , Combining the first wavelength polarization angle φ1 and the second wavelength polarization angle φ2, the polarization characteristic parameter V is formed: The data is then transmitted to the frequency domain feature extraction module for fusion analysis.

[0062] The polarization characteristic analysis process in step S4 utilizes a systematic polarization measurement technique. First, the received digital signal is accurately resolved into first-wavelength light intensity data I1(x,y) and second-wavelength light intensity data I2(x,y) according to a preset sensor partition mapping relationship. This two-dimensional data structure preserves the complete spatial distribution information of the scattered light spot. Then, independent rotatable polarizers are set in the two wavelength optical paths, rotating from 0° to 180° in 15° increments, collecting light intensity data at 13 polarization angles. This equidistant sampling ensures accurate reconstruction of the polarization curve while avoiding the time consumption of oversampling. By analyzing the light intensity changes at each polarization angle, the maximum value Imax and the minimum value Imin are extracted, and the degree of polarization p is calculated. This parameter quantitatively describes the degree of polarization of the scattered light, and its magnitude directly reflects the orderliness of the surface microstructure. Simultaneously, the angle corresponding to the maximum light intensity is determined as the polarization angle φ, characterizing the dominant vibration direction of the scattered photoelectric vector.

[0063] Furthermore, polarization parameters are calculated for each pixel in the image, generating two-dimensional polarization degree distribution maps P1(x,y) and P2(x,y). This high spatial resolution polarization mapping can identify local features and defect distributions on the surface. Finally, the mean and standard deviation of the polarization degree are obtained through statistical analysis, and combined with the polarization angle information, a polarization feature vector V containing six key parameters is constructed. These parameters quantify the polarization scattering characteristics of the surface from different angles. The mean reflects the overall roughness level, the standard deviation characterizes the spatial uniformity of roughness, and the parameter differences between the two wavelengths reveal multi-scale surface features, providing a unique and rich physical feature dimension for subsequent multimodal fusion analysis.

[0064] Preferably, the specific process of step S5 is as follows: S51: The frequency domain feature extraction module 7 receives the first wavelength light intensity data I1(x,y) and the second wavelength light intensity data I2(x,y), as well as the polarization feature parameter V output by the polarization characteristic analysis module; S52: Perform two-dimensional fast Fourier transform on I1(x,y) and I2(x,y) respectively to obtain the spectra F1(u,v) and F2(u,v), where u and v are spatial frequency coordinates; S53: Based on the spectrum F1(u,v), calculate the low-frequency energy E1, high-frequency energy E2, and the center position C1 of the spectrum energy of the first wavelength; S54: Based on the spectrum F2(u,v), calculate the low-frequency energy E3, high-frequency energy E4, and the center position C2 of the spectrum energy of the second wavelength; S55: Combine the frequency domain characteristic parameters [E1, E2, C1, E3, E4, C2] with the polarization characteristic parameter V to form a comprehensive feature vector Y, Y= The data is then transmitted to the multimodal fusion recognition module.

[0065] The frequency domain feature extraction step transforms the scattered light intensity information from the spatial domain to the frequency domain for analysis, revealing the frequency characteristics of surface roughness. By performing two-dimensional fast Fourier transforms on the two wavelength light intensity data I1(x,y) and I2(x,y), the spectra F1(u,v) and F2(u,v) are obtained, where the spatial frequency coordinates (u,v) correspond to different spatial periodic components. Low-frequency energies E1 and E3 reflect the large-scale undulations and waviness characteristics of the surface, which typically originate from systematic errors during processing. High-frequency energies E2 and E4 correspond to micro-roughness and random surface texture, reflecting the randomness of the material removal process. The spectral energy centers C1 and C2 characterize the dominant spatial scale of roughness; their position biased towards low frequencies indicates that the surface is dominated by large-scale undulations, while their bias towards high frequencies indicates that micro-roughness is dominant. Combining these frequency domain features with the polarization feature parameter V forms a comprehensive feature vector Y containing 12 elements, achieving information integration across the three physical dimensions of spectrum, polarization, and frequency domain.

[0066] Preferably, the specific process of step S6 is as follows: S61: The multimodal fusion recognition module 8 receives the comprehensive feature vector Y, performs standardization processing on each parameter in the vector, and maps each parameter to a numerical range of 0 to 1; S62: Access a pre-established roughness standard database, which stores the feature vectors of standard samples of different roughness grades and the corresponding measured roughness values, covering a roughness range from 0.05 micrometers to 3.2 micrometers; S63: Calculate the Euclidean distance between the current feature vector and the feature vector of each standard sample in the database, and select the 5 samples with the smallest distance as the nearest neighbor reference samples; S64: Determine the roughness value based on 5 nearest neighbor reference samples, extract the measured roughness values ​​corresponding to these 5 nearest neighbor samples and calculate their arithmetic mean, and use this mean as the predicted roughness value R for the current detection. a ; S65: Based on the roughness prediction value R a Determine the roughness grade: R a 0.2 micrometers or less is classified as precision grade. R a Sizes greater than 0.2 micrometers and less than or equal to 0.8 micrometers are classified as ordinary grade. R a If the value is greater than 0.8 micrometers, it is judged as rough.

[0067] The multimodal fusion identification step employs a machine learning-based intelligent identification method. Standardization eliminates differences in the dimensions and numerical ranges of different feature parameters, ensuring that each feature has equal weight in distance calculation. The pre-established standard database covers a wide roughness range of 0.05-3.2 micrometers, including various typical surfaces from precision machining to rough machining, providing a comprehensive reference benchmark for accurate identification. The nearest neighbor algorithm (k=5) is used for roughness prediction, finding the most similar standard samples in the feature space by calculating Euclidean distance. This method fully utilizes the discriminative power of multidimensional features and avoids the limitations that may exist with single features. Selecting 5 nearest neighbor samples for averaging ensures the stability of the prediction and avoids noise interference that may be introduced by too many samples. The final three-level classification standard (precision grade ≤0.2μm, ordinary grade 0.2-0.8μm, rough grade >0.8μm) meets the actual application needs of the machining industry and provides a clear judgment basis for bearing quality control.

[0068] The above are merely embodiments of this application and do not limit the scope of this patent application. Any equivalent structural or procedural changes made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A bearing raceway roughness testing system, comprising a frame, characterized in that: It also includes, A bearing positioning mechanism, mounted on the frame, is used to fix the position of the bearing being tested; A multi-wavelength laser module, mounted on the frame, is used to generate at least two different wavelength laser beams to simultaneously irradiate the surface of the bearing raceway under test. An optical path adjustment mechanism, connected to the multi-wavelength laser module and mounted on the frame, is used to adjust the emission angle of the multi-wavelength laser beam and combine it into an obliquely incident composite beam. A spectral analysis mechanism, mounted on the frame and positioned in the reflected light path of the obliquely incident composite beam, is used to separate multi-wavelength scattered light according to wavelength and collect scattered light information of each wavelength separately. A polarization characteristic analysis module, connected to the spectral analysis mechanism, is used to analyze the polarization characteristics of scattered light at each wavelength and generate polarization characteristic parameters; The frequency domain feature extraction module is connected to the spectral analysis mechanism and the polarization characteristic analysis module, and is used to perform frequency domain transformation on the scattered light information to generate frequency domain feature parameters. A multimodal fusion recognition module, connected to the frequency domain feature extraction module, is used to perform comprehensive analysis based on frequency domain feature parameters and polarization feature parameters, and output the bearing channel roughness detection results. The spectral analysis mechanism includes a spectrometer, a beam homogenizer, and an image acquisition device; The beam splitter is mounted on the frame and located in the reflection path of the multi-wavelength scattered light, and has an incident surface and an exit surface; The light-diffusing plate is disposed opposite to the exit surface of the beam-splitting element. The light-diffusing plate includes an incident surface and an exit surface. The light-diffusing plate is used to convert the incident non-uniform light spot into a light spot with a relatively uniform intensity distribution on the exit surface. The image acquisition device includes a photoelectric sensor array and a processing circuit. The photoelectric sensor array is arranged opposite to the light-emitting surface of the light-diffusing plate and maintains a distance between them. The photoelectric sensor array is divided into multiple acquisition areas, and each acquisition area corresponds to a different position on the light-emitting surface of the light-diffusing plate. The processing circuit is electrically connected to the photoelectric sensor array, and the output terminal of the processing circuit is connected to the input terminal of the polarization characteristic analysis module.

2. The bearing raceway roughness detection system according to claim 1, characterized in that: The bearing positioning mechanism includes an inclined table, a micro-motion slide, and a stop. The inclined table is mounted on the frame and has a preset tilt angle. The micro-motion slide is set on the inclined table. The stop is fixed on the micro-motion slide and has a contact surface that mates with the outer circle of the bearing.

3. The bearing raceway roughness detection system according to claim 1, characterized in that: The multi-wavelength laser module includes a first laser, a second laser, and a beam combiner. The first laser emits red light with a wavelength of 630-680nm, and the second laser emits blue-violet light with a wavelength of 405-450nm. The beam combiner combines the two laser beams into a coaxially propagating obliquely incident composite beam. The optical path adjustment mechanism includes a fixed base, a support frame, a clamping block, a rotating shaft, a mounting bracket, and a locking bolt. The fixed base is mounted on the table surface of the frame, the support frame is vertically mounted on the fixed base, the clamping block is fixed to the support frame by a first bolt, one end of the rotating shaft is fixed to the clamping block, the mounting bracket has a shaft hole and is fitted onto the other end of the rotating shaft to form a rotatable connection, and the locking bolt is set on the clamping block to lock the rotational position of the mounting bracket. The mounting bracket is used to mount a multi-wavelength laser module and adjust the incident angle of the laser beam by rotating the rotating shaft.

4. A method for detecting bearing raceway roughness, using the bearing raceway roughness detection system described in claim 1 or 3, characterized in that: Includes the following steps, S1: Bearing positioning steps: Fix the bearing to be tested in the preset position through the bearing positioning mechanism, and adjust the bearing position so that the target detection area of ​​the bearing groove is within the irradiation range of the multi-wavelength laser module. S2: Multi-wavelength laser irradiation steps: Activate the multi-wavelength laser module to generate at least two different wavelength laser beams, adjust the angle of the laser beams through the optical path adjustment mechanism, and irradiate the bearing raceway surface with the laser beams in an oblique incident manner. S3: Spectral Separation and Acquisition Steps: The spectral analysis unit receives the scattered light reflected from the bearing channel surface, separates the scattered light of different wavelengths through the beam splitter, acquires the intensity information of the scattered light of each wavelength through the image acquisition device, and transmits the acquired data to the polarization characteristic analysis module. S4: Polarization characteristic analysis steps: The polarization characteristic analysis module receives scattered light data of each wavelength transmitted by the spectral analysis mechanism, collects light intensity information at different polarization angles, calculates the degree of polarization and polarization angle parameters of scattered light of each wavelength, and transmits the polarization characteristic parameters to the frequency domain feature extraction module. S5: Frequency domain feature extraction step: The frequency domain feature extraction module receives the scattered light intensity information from the spectral analysis institution and the polarization feature parameters from the polarization characteristic analysis module, performs Fourier transform processing on the data, generates frequency domain feature parameters, and transmits the frequency domain feature parameters to the multimodal fusion recognition module; S6: Multimodal fusion identification steps: The multimodal fusion identification module receives frequency domain feature parameters and polarization feature parameters, performs fusion analysis on multiple feature parameters, and outputs the bearing channel roughness detection results.

5. The bearing raceway roughness detection method according to claim 4, characterized in that: The specific process of step S2 is as follows: S21: Activate the multi-wavelength laser module, wherein the first laser outputs a laser beam with a wavelength of 630nm-680nm, and the second laser outputs a laser beam with a wavelength of 405nm-450nm. S22: Two laser beams are combined into a composite beam by a beam combiner; S23: Adjust the angle of the composite beam by means of the optical path adjustment mechanism so that the angle between the composite beam and the normal of the bearing groove surface is 50°-70°; S24: Adjust the output power of the first laser and the second laser to keep the power ratio of the two wavelength lasers within the range of 0.8:1 to 1.2:1; S25: The composite beam with adjusted angle and power continuously irradiates the bearing raceway surface in an oblique incidence manner.

6. The bearing raceway roughness detection method according to claim 5, characterized in that: The specific process of step S3 is as follows: S31: The beam splitter receives the composite scattered light reflected from the bearing channel surface, the composite scattered light containing a first wavelength component from the first laser and a second wavelength component from the second laser. The beam splitter separates the composite scattered light by wavelength and outputs the spatially separated first wavelength beam and second wavelength beam from its exit surface. S32: The first wavelength beam and the second wavelength beam are emitted from the beam splitter at different angles and respectively irradiate different positions on the incident surface of the light homogenizer. The distance between the irradiation positions of the two beams on the incident surface of the light homogenizer is greater than 3mm. S33: The first wavelength beam passes through the homogenizing plate and forms a first light spot on its light-emitting surface; the second wavelength beam passes through the homogenizing plate and forms a second light spot on its light-emitting surface. S34: The first acquisition area of ​​the photoelectric sensor array of the image acquisition device acquires the light intensity signal of the first light spot, and the second acquisition area acquires the light intensity signal of the second light spot. S35: The processing circuit converts the light intensity signal collected by the photoelectric sensor array into a digital signal and transmits the digital signal containing the first wavelength and second wavelength information to the polarization characteristic analysis module at a frequency of not less than 15Hz.

7. The bearing raceway roughness detection method according to claim 6, characterized in that: The specific process of step S4 is as follows: S41: The polarization characteristic analysis module receives digital signals from the image acquisition device processing circuit and resolves the digital signals into first wavelength light intensity data I1(x,y) and second wavelength light intensity data I2(x,y), where x and y are the pixel coordinates of the photoelectric sensor array; S42: Rotatable polarizers are respectively set in the optical path after the spectroscopic element of the spectral analysis mechanism and before the image acquisition device, so that the first wavelength beam and the second wavelength beam pass through independent polarizers respectively; starting from the initial position, each polarizer is rotated from 0° to 180° in 15° increments, and light intensity data at 13 polarization angles are collected to obtain the first wavelength light intensity sequence I1(θ) and the second wavelength light intensity sequence I2(θ), where θ is the polarization angle; S43: Process the first wavelength light intensity sequence I1(θ), find the maximum value I1max and the minimum value I1min, and calculate the first degree of polarization: And determine the angle corresponding to I1max as the first wavelength polarization angle φ1; S44: Process the second wavelength light intensity sequence I2(θ), find the maximum value I2max and the minimum value I2min, and calculate the second degree of polarization: And determine the angle corresponding to I2max as the second wavelength polarization angle φ2; S45: For each pixel position (x, y) in the image, repeat steps S43 and S44 to obtain two-dimensional polarization distribution maps P1(x, y) and P2(x, y) respectively; S46: Calculate the average polarization degree of the entire image region. and standard deviation Combining the first wavelength polarization angle φ1 and the second wavelength polarization angle φ2, the polarization characteristic parameter V is formed: The data is then transmitted to the frequency domain feature extraction module for fusion analysis.

8. The bearing raceway roughness detection method according to claim 7, characterized in that: The specific process of step S5 is as follows: S51: The frequency domain feature extraction module receives the first wavelength light intensity data I1(x,y) and the second wavelength light intensity data I2(x,y), as well as the polarization feature parameter V output by the polarization characteristic analysis module; S52: Perform two-dimensional fast Fourier transform on I1(x,y) and I2(x,y) respectively to obtain the spectra F1(u,v) and F2(u,v), where u and v are spatial frequency coordinates; S53: Based on the spectrum F1(u,v), calculate the low-frequency energy E1, high-frequency energy E2, and the center position C1 of the spectrum energy of the first wavelength; S54: Based on the spectrum F2(u,v), calculate the low-frequency energy E3, high-frequency energy E4, and the center position C2 of the spectrum energy of the second wavelength; S55: Combine the frequency domain characteristic parameters [E1, E2, C1, E3, E4, C2] with the polarization characteristic parameter V to form a comprehensive feature vector Y, Y= The data is then transmitted to the multimodal fusion recognition module.

9. The bearing raceway roughness detection method according to claim 8, characterized in that: The specific process of step S6 is as follows: S61: The multimodal fusion recognition module receives the comprehensive feature vector Y, standardizes each parameter in the vector, and maps each parameter to a value range of 0 to 1; S62: Access a pre-established roughness standard database, which stores the feature vectors of standard samples of different roughness grades and the corresponding measured roughness values, covering a roughness range from 0.05 micrometers to 3.2 micrometers; S63: Calculate the Euclidean distance between the current feature vector and the feature vector of each standard sample in the database, and select the 5 samples with the smallest distance as the nearest neighbor reference samples; S64: Determine the roughness value based on 5 nearest neighbor reference samples, extract the measured roughness values ​​corresponding to these 5 nearest neighbor samples and calculate their arithmetic mean, and use this mean as the predicted roughness value R for the current detection. a ; S65: Based on the roughness prediction value R a Determine the roughness grade: R a 0.2 micrometers or less is classified as precision grade. R a Sizes greater than 0.2 micrometers and less than or equal to 0.8 micrometers are classified as ordinary grade. R a If the value is greater than 0.8 micrometers, it is judged as rough.

Images