An online detection method and system for surface residual oil film thickness of a full-automatic oil throwing machine

Abstract

The present application relates to the technical field of industrial online detection, and specifically discloses a kind of full-automatic oil throwing machine surface residual oil film thickness online detection method and system, applied to full-automatic oil throwing machine, comprising: obtaining the multispectral reflectance image of workpiece in full-automatic oil throwing machine, three-dimensional point cloud data and preset workpiece CAD standard template, and according to the multispectral reflectance image, the measured reflectivity of each pixel point is obtained.The present application obtains the normal vector deviation angle of each pixel point on the surface of workpiece by registering three-dimensional point cloud data with preset workpiece CAD standard template, and carries out geometric normalization correction on measured reflectivity based on Fresnel equation, fundamentally solves the problem of reflectivity distortion caused by the posture change of curved surface.

Description

Technical Field

[0001] This invention relates to the field of industrial online detection technology, and in particular to an online detection method and system for the thickness of residual oil film on the surface of a fully automatic oil slinger. Background Technology

[0002] During the production of metal workpieces, such as cold heading and machining, a large amount of cutting oil or cooling oil adheres to the surface of the workpiece after plastic forming. Fully automatic oil-scraping machines use centrifugal force to remove the surface oil. However, in actual production, verifying the oil-scraping effect mainly relies on offline sampling, sending samples to a laboratory for weight measurement. This method suffers from problems such as susceptibility to interference during sampling and transportation, inability to reflect the distribution of the oil film across the entire surface, and inability to achieve online real-time feedback control.

[0003] To address the limitations of existing infrared spectroscopy thickness measurement techniques, which rely on flat surfaces and fixed measurement geometry and are therefore unsuitable for handling complex three-dimensional curved workpieces and dynamic oil mist environments within oil slingers, this paper proposes a fully automated online method and system for detecting residual oil film thickness on the surface of oil slingers. This system overcomes the challenges of reflectivity distortion caused by changes in surface orientation, uncompensated oil mist attenuation due to spatial separation between the beam splitter reference path and the measurement path, and isolated outliers caused by a lack of spatial constraints in thickness inversion in low-confidence regions. Furthermore, it constructs an online detection system architecture integrating multispectral imaging, three-dimensional topography measurement, and co-optical path reference compensation. Summary of the Invention

[0004] The purpose of this invention is to provide an online detection method and system for the thickness of residual oil film on the surface of a fully automatic oil slinger, so as to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An online detection method for the thickness of residual oil film on the surface of a fully automatic oil slinger, applicable to fully automatic oil slingers, includes: Acquire multispectral reflectance images, three-dimensional point cloud data, and preset workpiece CAD standard templates for the workpiece inside the fully automatic oil slinger, and obtain the measured reflectance of each pixel based on the multispectral reflectance images; The three-dimensional point cloud data is registered with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. Based on the normal vector deviation angle, the local roughness parameters of the workpiece surface are obtained from the three-dimensional point cloud data, and the equivalent normal incident reflectance of each pixel is obtained from the local roughness parameters of the workpiece surface and the normal vector deviation angle. Obtain the process setting parameters for this oil removal process, and establish an oil film residue distribution model after oil removal based on the process setting parameters; The real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece is obtained. The reversible scattering attenuation coefficient of the current optical path is obtained based on the real-time reflectivity signal. The compensation reflectivity is obtained based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. Using the residual distribution model as a regularization constraint, Lambert-Beer inversion is performed on the compensated reflectivity to obtain the preliminary value of the oil film thickness of each pixel. The final oil film thickness value of each pixel is obtained based on the registration confidence and the preliminary oil film thickness value, and an oil film thickness distribution map is generated based on the final oil film thickness value. The online detection results of residual oil film thickness are output based on the oil film thickness distribution map.

[0006] Preferably, the step of registering the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface includes: The three-dimensional point cloud data is iteratively registered with the CAD standard template to establish the spatial transformation relationship between each spatial point in the three-dimensional point cloud data and each corresponding point in the CAD standard template. Based on the spatial transformation relationship, the theoretical normal vector corresponding to each pixel is obtained from the CAD standard template; Based on the theoretical normal vector and the pre-calibrated sensor optical axis direction, the angle between the theoretical normal vector and the sensor optical axis direction of each pixel is calculated and used as the normal vector deviation angle; Obtain the registration residual of the iterative nearest point registration, and determine the registration confidence of each pixel based on the registration residual.

[0007] Preferably, the step of obtaining the local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and obtaining the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle, includes: The optical constants of the metal substrate material under the acquisition band of the multispectral reflectance image are obtained, and the optical constants include refractive index and extinction coefficient; Based on the normal vector deviation angle and the optical constant, the geometric correction factor corresponding to each pixel is calculated according to the Fresnel equation; The measured reflectance is corrected according to the geometric correction factor, and the measured reflectance is divided by the geometric correction factor to obtain the equivalent normal incident reflectance.

[0008] Preferably, the steps of acquiring the real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece, acquiring the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and acquiring the compensation reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient include: The real-time reflectance signal of a fixed reference plate within the same camera field of view as the workpiece is obtained, and the reference reflectance of the fixed reference plate at each wavelength is obtained synchronously at the time of the multispectral reflectance image acquisition based on the real-time reflectance signal. Obtain the initial calibrated reflectivity of the fixed reference plate; The comprehensive attenuation coefficient is obtained based on the initial calibrated reflectance and the reference reflectance; Acquire the signal from the oil mist concentration sensor inside the oil slinger and the cumulative usage time of the fixed reference plate; Based on the comprehensive attenuation coefficient, the oil mist concentration sensor signal, and the cumulative usage time, the reversible scattering attenuation coefficient of the current optical path is obtained: The compensated reflectivity is obtained based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient.

[0009] Preferably, the step of obtaining the process setting parameters for this oil-scraping process and establishing an oil film residue distribution model after oil-scraping based on the process setting parameters includes: The process setting parameters and oil parameters for this oil slinging process are obtained from the oil slinging machine control system. The process setting parameters include the rotational angular velocity and the oil slinging time. Based on the spatial position of the three-dimensional point cloud data after registration with the CAD standard template, the rotation radius of each pixel in the workpiece coordinate system is obtained. The oil density and dynamic viscosity are obtained based on the oil parameters. Based on the rotational angular velocity, oil ejection time, rotational radius, oil density, and oil dynamic viscosity described in the library, a theoretical residual distribution model of the oil film after oil ejection is established.

[0010] Preferably, the step of using the residual distribution model as a regularization constraint to perform Lambert-Beer inversion on the equivalent normal incident reflectance to obtain the preliminary value of the oil film thickness for each pixel includes: Obtain the reference reflectance of the workpiece at the characteristic absorption wavelength and the preset Lambert-Beer inversion parameters; Based on the equivalent normal incident reflectance, reference reflectance, and Lambert-Beer inversion parameters of each pixel, construct the Lambert-Beer reflectance function: A regularization constraint term is constructed based on the oil film residue distribution model after oil spillage; Based on the Lambert-Beer reflectivity function and regularization constraints, construct the coupled inversion objective function: The initial value of the oil film thickness at each pixel is obtained by iteratively optimizing the coupled inversion objective function.

[0011] Preferably, the step of obtaining the final value of the oil film thickness for each pixel based on the registration confidence level and the preliminary value of the oil film thickness includes: The registration reliability of each pixel is determined based on the registration confidence level. For pixels whose registration confidence is greater than or equal to a preset confidence threshold, the initial value of the oil film thickness is directly used as the final value of the oil film thickness of that pixel. For a pixel whose registration confidence is less than the preset confidence threshold, neighborhood smoothing interpolation is performed based on the preliminary value of the oil film thickness of the neighboring pixels to obtain the neighborhood interpolation thickness of the pixel. The initial value of the oil film thickness and the neighborhood interpolation thickness are weighted and fused according to the registration confidence. The registration confidence is used as the weight of the initial value of the oil film thickness, and the weight of the registration confidence as the neighborhood interpolation thickness is subtracted to obtain the final value of the oil film thickness of the pixel.

[0012] Preferably, the step of generating an oil film thickness distribution map based on the final oil film thickness value includes: Obtain the final value of the oil film thickness and its corresponding pixel coordinates for each pixel; A spatial mapping relationship is established based on the pixel coordinates of each pixel, and the final oil film thickness value of each pixel is filled into the corresponding pixel position of the blank map according to the spatial mapping relationship to generate an initial thickness distribution map; Unfilled pixels in the initial thickness distribution map are obtained, and hole interpolation is performed to fill the holes based on the final oil film thickness values ​​of the neighboring pixels of the unfilled pixels to generate a complete thickness distribution map. Spatial smoothing filtering is applied to the complete thickness distribution map to eliminate thickness jumps between pixels, thereby generating the oil film thickness distribution map.

[0013] Preferably, the step of outputting the online detection result of residual oil film thickness based on the oil film thickness distribution map includes: Obtain the final value of the oil film thickness and its corresponding pixel coordinates for each pixel in the oil film thickness distribution map; Based on the final value of the oil film thickness of each pixel, the thickness statistical characteristics of the oil film thickness distribution map are calculated. The thickness statistical characteristics include the mean thickness, peak thickness, minimum thickness, and standard deviation of thickness. The final value of the oil film thickness of each pixel is compared with a preset thickness threshold pixel by pixel, and a defect judgment label is generated for each pixel based on the comparison result. The total number of defective pixels in the oil film thickness distribution map is counted based on the defect judgment label, and the defect area ratio is calculated based on the total number of defective pixels and the total number of pixels in the oil film thickness distribution map. A detection result report is generated based on the thickness statistical characteristics, the defect judgment label, and the defect area ratio, and the detection result report is output as the online detection result of the residual oil film thickness.

[0014] This invention also discloses an online detection system for the thickness of residual oil film on the surface of a fully automatic oil slinger, comprising: The measured reflectance acquisition module is used to acquire multispectral reflectance images, three-dimensional point cloud data and preset workpiece CAD standard templates of the workpiece inside the fully automatic oil slinger, and to acquire the measured reflectance of each pixel based on the multispectral reflectance images. The registration module is used to register the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. The incident reflectance acquisition module is used to obtain the local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and to obtain the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle. The distribution model establishment module is used to obtain the process setting parameters for this oil-scraping process and establish an oil film residue distribution model after oil-scraping based on the process setting parameters. The compensation reflectivity acquisition module is used to acquire the real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece, acquire the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and acquire the compensation reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. The oil film thickness preliminary value acquisition module is used to perform Lambert-Beer inversion on the compensated reflectivity using the residual distribution model as a regularization constraint term to obtain the preliminary value of the oil film thickness of each pixel. The distribution map generation module is used to obtain the final value of the oil film thickness of each pixel based on the registration confidence and the preliminary value of the oil film thickness, and to generate an oil film thickness distribution map based on the final value of the oil film thickness. The result output module is used to output the online detection result of residual oil film thickness based on the oil film thickness distribution map.

[0015] The beneficial effects of this application are as follows: This invention obtains the normal vector deviation angle of each pixel on the workpiece surface by registering three-dimensional point cloud data with a preset workpiece CAD standard template, and performs geometric normalization correction on the measured reflectivity based on the Fresnel equation, fundamentally solving the problem of reflectivity distortion caused by surface attitude changes; by setting a fixed reference plate in the camera field of view with the same optical path as the workpiece to obtain a real-time reflectivity signal reflecting the true attenuation state of the current optical path, and calculating the comprehensive attenuation coefficient to perform common-path differential compensation on the equivalent normal incident reflectivity, it eliminates the two interferences of multiplicative fluctuation of the light source and multiplicative scattering of oil mist in one go, overcoming the structural defect of the known beam splitter compensation scheme where the spatial separation of the reference optical path and the measurement optical path leads to the inability to compensate for oil mist attenuation; by performing preliminary value of oil film thickness based on registration confidence... Weighted constraints are applied, directly using the inversion results for high-confidence pixels and filling low-confidence pixels with neighborhood smoothing interpolation. This effectively suppresses isolated outliers in low-confidence regions, ensuring spatial continuity and consistency in the thickness distribution map. Finally, through the collaborative work of a multispectral camera, structured light scanner, fixed reference plate, rotary encoder, and processor, a complete online inspection system architecture is constructed. This system achieves fully automated inspection from multi-source data acquisition, 3D registration, geometric normalization correction, common-path differential compensation, thickness inversion, weighted constraints to thickness distribution map generation. It can output pixel-by-pixel oil film thickness distribution maps of the entire workpiece surface, providing a complete spatial distribution data foundation for oil removal process quality assessment and closed-loop control. It has significant advantages such as online real-time operation, non-contact operation, and adaptability to oil mist environments. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] like Figure 1 As shown, this application provides an online detection method for the thickness of residual oil film on the surface of a fully automatic oil slinger, applicable to a fully automatic oil slinger, comprising: S1. Obtain the multispectral reflectance image, three-dimensional point cloud data and preset workpiece CAD standard template of the workpiece in the fully automatic oil slinger, and obtain the measured reflectance of each pixel point according to the multispectral reflectance image; Exemplarily, this invention triggers an online detection process after the oil-slinging process of the fully automatic oil-slinging machine is completed but before the workpiece is removed from the oil-slinging machine cavity. A multispectral imaging device deployed inside the oil-slinging machine simultaneously acquires reflected light intensity images of the workpiece surface under multiple characteristic wavelengths, obtaining a multispectral reflectance image. Simultaneously, a structured light 3D scanning device acquires 3D point cloud data of the workpiece surface. Before acquisition, the multispectral reflectance image and the 3D point cloud data have been calibrated by the system to establish a precise mapping relationship between pixels and 3D spatial points, ensuring that every pixel in the image can find a corresponding spatial point in the 3D point cloud, and vice versa. The preset workpiece CAD standard template is derived from the workpiece design file, which contains the workpiece's ideal 3D geometric model and the theoretical normal vector information for each surface region. For the multispectral reflectance image, based on pre-performed radiometric calibration, the image grayscale values ​​under each wavelength are converted into measured reflectance, thereby obtaining the measured reflectance of each pixel.

[0021] S2. Register the three-dimensional point cloud data with the preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. For example, after acquiring the 3D point cloud data of the workpiece, it is spatially registered with a preset workpiece CAD standard template to obtain the pose deviation of each spatial point on the actual workpiece surface relative to the theoretical model. Since a mapping relationship has been established between the pixels of the multispectral reflectance image and the 3D point cloud, after registration, the normal vector deviation angle and registration confidence score in the registration result can be assigned pixel by pixel to each pixel of the multispectral reflectance image, thereby providing a spatial information basis for subsequent reflectivity geometric correction and thickness-weighted fusion. The normal vector deviation angle refers to the angle between the actual normal vector of the surface position corresponding to each pixel and the optical axis direction of the multispectral imaging device, which reflects the degree of deviation of the incident angle of light at that point from the perpendicular incident condition; the registration confidence score characterizes the reliability of the registration result at that spatial point.

[0022] S3. Based on the normal vector deviation angle, obtain the local roughness parameters of the workpiece surface according to the three-dimensional point cloud data, and obtain the equivalent normal incident reflectivity of each pixel point according to the local roughness parameters of the workpiece surface and the normal vector deviation angle. For example, since the workpiece surface is usually a non-planar freeform surface with micro-roughness, the measured reflectance at each pixel is not only modulated by the oil film thickness, but also affected by the angle between the local surface normal vector and the imaging optical axis at that point, as well as the surface micro-roughness, thus introducing geometric attenuation and diffuse reflection components independent of the oil film thickness. This invention utilizes the normal vector deviation angle of each pixel obtained in step S2, combined with the local surface roughness parameters of the workpiece extracted from the three-dimensional point cloud data, to calculate the geometric correction factor for each pixel based on the Fresnel-Gaussian mixed reflection model of the rough surface. Then, the measured reflectance is divided by this geometric correction factor to obtain the reflectance value corrected to the equivalent normal incident condition, i.e., the equivalent normal incident reflectance.

[0023] S4. Obtain the process setting parameters for this oil-scraping process, and establish an oil film residue distribution model after oil-scraping based on the process setting parameters. For example, this invention obtains the process setting parameters for the current oil-slinging process from the oil-slinging machine control system. These parameters include the rotational angular velocity and the oil-slinging duration. Since the detection is triggered after the oil-slinging process is completed but before the workpiece is removed from the oil-slinging machine cavity, the workpiece is stationary at this time, and centrifugal force no longer acts on the oil film. However, the centrifugal force applied during high-speed rotation during the oil-slinging process determines the final residual distribution pattern of the oil film at the end of the oil-slinging process—centrifugal force drives the oil to migrate radially outward, resulting in a thicker oil film in the central region and a thinner oil film in the edge region. This distribution pattern is "frozen" after the oil-slinging process is completed. Therefore, this invention establishes an oil film residual distribution model after oil-slinging based on the process setting parameters (including rotational angular velocity and oil-slinging duration), combined with the rotation radius of each pixel obtained from the three-dimensional point cloud data and the oil parameters (including oil density and oil dynamic viscosity). This model describes the theoretical oil film residual thickness distribution law at various locations on the entire surface of the workpiece after the oil-slinging process.

[0024] S5. Obtain the real-time reflectivity signal of the fixed reference plate in the same optical path as the workpiece, obtain the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and obtain the compensation reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. By way of example, the present invention sets a fixed reference plate at a fixed stationary position within the camera's field of view. The fixed reference plate and the workpiece are within the same camera's field of view, and the fixed reference plate's surface is not covered by an oil film. The fixed reference plate and the workpiece share the same optical path environment—that is, light emitted from the light source passes through the same optical elements and the same oil mist medium before reaching the reference plate surface, and the reflected light from the reference plate surface then passes through the same oil mist medium and the same optical elements before reaching the sensor. Therefore, the reflectivity signal of the fixed reference plate contains all the attenuation information of the current optical path, including the multiplicative fluctuations of the light source and the multiplicative scattering attenuation of the oil mist.

[0025] During the acquisition of multispectral reflectance images, this invention simultaneously acquires the reflectance of the fixed reference plate in each band as a real-time reflectance signal. Simultaneously, it obtains the initial calibration reflectance of the fixed reference plate, which is the reference reflectance of the fixed reference plate pre-acquired under conditions of no oil mist and no oil slicker. Based on the initial calibration reflectance and the real-time reflectance signal, a comprehensive attenuation coefficient is obtained. Since the fixed reference plate is in an oil mist environment for a long time, its surface may experience irreversible contamination attenuation due to the slow deposition of oil mist particles. This contamination attenuation is time-cumulative and should not be compensated; however, the scattering attenuation caused by real-time fluctuations in oil mist concentration is reversible and should be compensated. This invention acquires the oil mist concentration sensor signal inside the oil slicker and the cumulative usage time of the fixed reference plate. Based on the comprehensive attenuation coefficient, the oil mist concentration sensor signal, and the cumulative usage time, the reversible scattering attenuation coefficient of the current optical path (i.e., the reversible multiplicative attenuation caused by oil mist scattering) is separated from the comprehensive attenuation coefficient.

[0026] Finally, the compensated reflectance is obtained based on the equivalent normal incident reflectance and the reversible scattering attenuation coefficient. Through this multiplicative compensation, the portion of the equivalent normal incident reflectance lost due to the reversible multiplicative attenuation of the optical path is accurately recovered, thereby eliminating the interference of oil mist scattering on reflectance measurement. This compensation method is fundamentally different from the known beam splitter reference compensation scheme: the beam splitter takes the reference light at the light source end, and the reference light does not pass through the oil mist region, making it impossible to sense the oil mist attenuation in the latter part of the optical path; the fixed reference plate of this invention is set at the workpiece end, sharing the complete optical path with the workpiece, and can comprehensively sense the attenuation of the entire optical path, including oil mist.

[0027] S6. Using the residual distribution model as a regularization constraint, perform Lambert-Beer inversion on the compensated reflectivity to obtain the preliminary value of the oil film thickness of each pixel. For example, after obtaining the compensated reflectance, the present invention acquires preset Lambert-Beer inversion parameters, which include the molar absorption coefficient of the oil film at the characteristic absorption wavelength and the oil film density correlation constant. Based on the compensated reflectance of each pixel, the reference reflectance of the substrate material at the characteristic absorption wavelength, and the Lambert-Beer inversion parameters, a Lambert-Beer reflectance function is constructed. Simultaneously, a regularization constraint term is constructed based on the residual oil film distribution model after oil removal. Using the residual distribution model as a regularization constraint term embeds the physical laws of the oil removal process (centrifugal force-driven radial migration effect) into the optical measurement inversion process, ensuring that the thickness inversion result has physical consistency across the entire surface. Even if there is noise in the local reflectance measurement, the inversion result is still constrained by the residual distribution and will not produce physically unreasonable isolated extrema.

[0028] This invention constructs a coupled inversion objective function based on the squared difference between the Lambert-Beer reflectivity function and the compensated reflectivity, as well as the regularization constraint term. The coupled inversion objective function is iteratively optimized to obtain the oil film thickness at each pixel that minimizes the objective function, which serves as the initial value of the oil film thickness at each pixel. S7. Obtain the final value of the oil film thickness for each pixel based on the registration confidence and the preliminary value of the oil film thickness, and generate an oil film thickness distribution map based on the final value of the oil film thickness. For example, for pixels with high registration confidence, the registration result is reliable and the normal vector deviation angle is accurate. Therefore, the preliminary oil film thickness value obtained from this inversion has high reliability and can be directly used as the final oil film thickness value for that pixel. For pixels with low registration confidence, registration may have deviations. If the preliminary oil film thickness value is directly used, it may introduce thickness anomalies caused by registration errors. Therefore, for low-confidence pixels, this invention performs neighborhood smoothing interpolation based on the preliminary oil film thickness values ​​of high-confidence pixels in the pixel's neighborhood to obtain the neighborhood interpolated thickness of that pixel. Then, using the registration confidence as a weight, the preliminary oil film thickness value and the neighborhood interpolated thickness are weighted and fused to obtain the final oil film thickness value for each pixel. This step incorporates spatial continuity constraints into the pixel-by-pixel thickness information, effectively suppressing isolated anomalies caused by local registration failures. Subsequently, a spatial mapping relationship is established based on the pixel coordinates corresponding to each pixel. The final value of the oil film thickness of each pixel is filled into the corresponding pixel position of the blank map according to the spatial mapping relationship. Hole interpolation is performed to fill the unfilled pixels, and spatial smoothing filtering is performed on the complete thickness distribution map to generate an oil film thickness distribution map that characterizes the thickness of the residual oil film on the entire surface.

[0029] S8. Output the online detection result of residual oil film thickness based on the oil film thickness distribution map; For example, after generating the oil film thickness distribution map, the present invention outputs the oil film thickness distribution map as the online detection result of residual oil film thickness. The online detection result of residual oil film thickness includes at least the oil film thickness distribution map itself, visually presenting the quantitative value of residual oil film thickness at each pixel on the entire surface of the workpiece. This detection result can be output to the oil-scraping machine control system or the host computer quality management system to guide the adjustment of oil-scraping process parameters or the determination of product quality. Through this step, a complete online detection process from multispectral image acquisition to thickness distribution map output is completed.

[0030] As described in steps S1-S8 above, the present invention first acquires multispectral reflectance images, three-dimensional point cloud data, and a preset workpiece CAD standard template of the workpiece inside the fully automatic oil slinger, and obtains the measured reflectance of each pixel based on the multispectral reflectance image, providing a complete data foundation including spectral features, spatial morphology, and theoretical geometric model for subsequent processing; then, the three-dimensional point cloud data is registered with the preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface, thereby establishing a spatial correspondence between the actual workpiece surface and the theoretical model and providing a quantitative basis for surface attitude correction; then, the measured reflectance is geometrically normalized based on the normal vector deviation angle to obtain the equivalent normal incident reflectance of each pixel, thereby... The method eliminates reflectivity distortion caused by the curved surface posture of the workpiece, unifying the reflectivity of surfaces with different postures to an equivalent normal incident reference. Then, based on the equivalent normal incident reflectivity, the preliminary oil film thickness value of each pixel is inverted according to the Lambert-Beer law, realizing the quantitative conversion from corrected reflectivity to thickness value. Next, based on the registration confidence and the preliminary oil film thickness value, the final oil film thickness value of each pixel is obtained, and an oil film thickness distribution map is generated based on the final oil film thickness value. Through a weighted fusion mechanism, the thickness anomaly value in the low confidence region is effectively suppressed, and a thickness spectrum with continuous distribution across the entire surface is generated. Finally, based on the oil film thickness distribution map, the online detection result of residual oil film thickness is output, presenting the pixel-by-pixel quantitative thickness information in a visual way as a traceable and evaluable detection conclusion. Through the synergistic effect of the above steps, this invention realizes a complete online detection chain from multi-source data acquisition, surface posture correction, reflectivity normalization, quantitative thickness inversion to spatial weighted fusion and distribution map output, fundamentally solving the technical problem that existing infrared spectroscopy thickness measurement technology is only suitable for flat surfaces and cannot be adapted to the online detection of three-dimensional curved workpieces inside the oil slinger.

[0031] In one embodiment, the step of registering the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface includes: S201. Iteratively register the three-dimensional point cloud data with the CAD standard template to establish the spatial transformation relationship between each spatial point in the three-dimensional point cloud data and each corresponding point in the CAD standard template. For example, after acquiring three-dimensional point cloud data and a preset workpiece CAD standard template, the present invention performs iterative nearest point registration between the two to establish the spatial transformation relationship between each spatial point in the three-dimensional point cloud data and each corresponding point in the CAD standard template.

[0032] Since the 3D point cloud data is the spatial position information of the workpiece surface collected in real time within the oil slinger, its coordinate system is the sensor measurement coordinate system, while the coordinate system of the preset workpiece CAD standard template is the workpiece design coordinate system. There is a difference in spatial transformation between the two due to rotation and translation. Simultaneously, a correspondence needs to be established between the spatial points in the 3D point cloud data and the pixels in the multispectral reflectance image to map the normal vector information to each pixel. To solve the above coordinate alignment and mapping problems, this invention achieves accurate registration between the 3D point cloud data and the CAD standard template through an iterative nearest-point registration algorithm. Using the preset workpiece CAD standard template as the target point cloud and the 3D point cloud data as the source point cloud, the optimal rigid body transformation parameters between the source and target point clouds are solved iteratively to minimize the average distance between each spatial point in the source point cloud and each corresponding point in the target point cloud. The rigid body transformation parameters include rotation matrices and translation vectors, which together constitute the spatial transformation relationship.

[0033] This invention first preprocesses the 3D point cloud data, including voxel filtering downsampling and statistical outlier removal, to reduce the amount of data and remove noise points. Voxel filtering downsampling divides the 3D space into a regular voxel grid, retaining the point closest to the voxel center within each voxel as a representative point, thereby significantly reducing the number of points while maintaining the overall shape characteristics of the point cloud and lowering the computational cost of subsequent registration. Statistical outlier removal calculates the average distance between each spatial point and its neighborhood points, identifying spatial points whose average distance exceeds a preset multiple of the standard deviation as outliers and removing them to avoid interference from measurement noise on registration accuracy.

[0034] Then, using a preset workpiece CAD standard template as the target point cloud and preprocessed 3D point cloud data as the source point cloud, iterative nearest-neighbor registration is performed. In each iteration, the algorithm finds the nearest neighbor point in the target point cloud for each spatial point in the source point cloud. Then, based on all corresponding point pairs, it solves for the optimal rigid body transformation parameters and updates the source point cloud according to the solved transformation parameters. This process is repeated until the registration residual converges or the preset number of iterations is reached. When the iteration terminates, the source point cloud and the target point cloud reach optimal alignment. The rigid body transformation parameters at this point represent the spatial transformation relationship between each spatial point in the 3D point cloud data and its corresponding point in the CAD standard template. This spatial transformation relationship characterizes the rotational and displacement offsets between the actual pose of the workpiece and the theoretical model.

[0035] S202. Based on the spatial transformation relationship, obtain the theoretical normal vector corresponding to each pixel from the CAD standard template; For example, after obtaining the spatial transformation relationship, the present invention maps each spatial point in the 3D point cloud data to the corresponding theoretical point in the CAD standard template according to the spatial transformation relationship, thereby determining the corresponding position of each spatial point in the 3D point cloud data on the CAD standard template. Since each point in the CAD standard template carries the theoretical normal vector information of the surface at that point (i.e., the normal direction of the surface where the point is located in the ideal geometric model), the corresponding theoretical normal vector can be obtained from the CAD standard template for each spatial point in the 3D point cloud data through the correspondence relationship. Then, according to the mapping relationship between the multispectral reflectance image pixels and the 3D point cloud spatial points pre-calibrated by the system, the theoretical normal vector of each spatial point is assigned to the corresponding pixel, thereby obtaining the theoretical normal vector corresponding to each pixel.

[0036] The theoretical normal vector is the theoretical orientation of the workpiece surface at the corresponding position of the pixel, which reflects the ideal surface normal direction of the workpiece under conditions of no manufacturing error and no pose deviation.

[0037] S203. Based on the theoretical normal vector and the pre-calibrated sensor optical axis direction, calculate the angle between the theoretical normal vector and the sensor optical axis direction for each pixel, and use it as the normal vector deviation angle. For example, after obtaining the theoretical normal vector of each pixel, the present invention further obtains the pre-calibrated sensor optical axis direction. The sensor optical axis direction is the direction vector of the optical system optical axis of the multispectral imaging device in the sensor coordinate system, and this direction vector is obtained in advance through a camera calibration program. During the camera calibration process, based on the pinhole imaging model and the lens distortion model, calibration board images are acquired under different poses, and the camera intrinsic parameters (including focal length, principal point coordinates, and distortion coefficients) and extrinsic parameters (camera pose in the world coordinate system) are solved using the Zhang Zhengyou calibration method or a nonlinear optimization method. Then, the sensor optical axis direction is determined based on the rotation matrix in the extrinsic parameters.

[0038] This invention calculates the angle between the theoretical normal vector and the sensor's optical axis for each pixel, based on the theoretical normal vector and the sensor's optical axis direction, and uses this angle as the normal vector deviation angle for that pixel. This deviation angle characterizes the degree to which the theoretical normal vector of the workpiece surface at that pixel position deviates from the sensor's optical axis direction, reflecting the geometric relationship between the orientation of the surface at that position in the ideal geometric model and the imaging optical path. When the theoretical normal vector coincides with the sensor's optical axis direction, the normal vector deviation angle is zero, indicating that the surface at that position is facing directly towards the sensor, and the incident light is perpendicular to the surface. When an angle exists between them, it indicates that the surface at that position is tilted relative to the sensor.

[0039] S204. Obtain the registration residual of the iterative nearest point registration, and determine the registration confidence of each pixel based on the registration residual; For example, during the iterative nearest-point registration process, the present invention obtains the registration residual after each iteration. The registration residual is the average distance or root mean square distance between all corresponding point pairs between the source point cloud and the target point cloud after the current iteration. When the iterative nearest-point registration converges, the final registration residual reflects the degree of matching between the 3D point cloud data and the CAD standard template.

[0040] In actual operation of the oil slinger, for flat surfaces of the workpiece, the 3D point cloud data is relatively complete, and iterative nearest-point registration can achieve high matching accuracy with small registration residuals. However, for deep hole edges, steep slope transition areas, or areas with occlusions, the 3D point cloud data is sparse or has scanning blind spots, limiting the matching accuracy of iterative nearest-point registration and resulting in larger registration residuals. Therefore, the magnitude of the registration residual directly reflects the reliability of the registration results at each spatial point.

[0041] This invention determines the registration confidence of each pixel based on the registration residual of iterative nearest-point registration. Specifically, the smaller the registration residual, the more reliable the registration result at that spatial point, and the higher the registration confidence; conversely, the larger the registration residual, the less reliable the registration result at that spatial point, and the lower the registration confidence. This invention obtains the registration confidence of each pixel by normalizing the registration residual and mapping it to the [0, 1] interval (the specific mapping method is as follows:). (a) Obtain the registration residuals corresponding to each pixel after the iterative nearest point registration converges. (i represents the i-th pixel), the registration residual is defined as the Euclidean distance between the spatial point corresponding to the pixel and its nearest neighbor on the CAD standard template, in mm.

[0042] (b) Calculate the statistical characteristics of the registration residuals for all pixels in the entire image: , ; Where N is the total number of valid pixels.

[0043] (c) Use a Sigmoid mapping function to convert the registration residuals into registration confidence scores. : in, Let be the registration confidence score of the i-th pixel, which is dimensionless and takes a value in the range (0,1). This represents the registration residual for that pixel, in mm. Let be the steepness coefficient of the Sigmoid function, taken as... =3 / (That is, for every standard deviation increase in the registration residual, the confidence level decreases by approximately 95%); τ is the confidence threshold parameter, taken as... (That is, when the registration residuals exceed one standard deviation of the mean, the confidence level begins to decrease significantly.)

[0044] The physical meaning of the above mapping function is: the registration residual is relatively small ( When the confidence level is close to 1, it indicates that the registration of that pixel is reliable; a large registration residual indicates that the registration is reliable. When the confidence level approaches 0, it indicates that the registration of that pixel is unreliable; the intermediate region exhibits a smooth transition, avoiding abrupt edge changes caused by hard thresholding. The values ​​of λ and τ can be adjusted according to the actual registration accuracy requirements: for detection scenarios with high accuracy requirements (thickness error <0.1g / m²), a value of λ can be used. For scenarios with lower accuracy requirements, the following can be adopted: .

[0045] (d) For cases where the registration residual distribution has a heavy-tailed distribution (i.e., there are a large number of outlier residuals), the following steps can be taken first: Perform truncation: Take To truncate the upper limit, The residual values ​​of the pixels are uniformly assigned to Then, the aforementioned Sigmoid mapping is performed to avoid compressing the dynamic range of low-confidence regions by a few outliers. The larger the registration confidence value, the more reliable the registration, which is subsequently used to apply weighted constraints to the initial oil film thickness value to ensure the accuracy of thickness inversion.

[0046] As described in steps S201-S204 above, the present invention first performs iterative nearest-point registration between the three-dimensional point cloud data and the CAD standard template to establish a spatial transformation relationship between each spatial point in the three-dimensional point cloud data and each corresponding point in the CAD standard template, thereby achieving precise spatial alignment between the actual measurement data and the theoretical geometric model; then, based on the spatial transformation relationship, the theoretical normal vector corresponding to each pixel is obtained from the CAD standard template, thereby assigning each pixel theoretical surface orientation information from the high-precision design model; then, based on the theoretical normal vector and the pre-calibrated sensor optical axis direction, the angle between the theoretical normal vector of each pixel and the sensor optical axis direction is calculated as the normal vector deviation angle, thereby quantifying the degree of deviation of the actual orientation of the surface position corresponding to each pixel relative to the imaging optical path; finally, the registration residual of the iterative nearest-point registration is obtained and the registration confidence of each pixel is determined based on the registration residual, thereby providing a quantitative assessment of the registration reliability of each pixel. Through the synergistic effect of the above steps, the present invention provides a pixel-by-pixel attitude correction basis, namely the normal vector deviation angle, for subsequent geometric normalization correction, and provides a pixel-by-pixel reliability weight, namely the registration confidence, for subsequent weighted constraints. Together, they ensure the accuracy of reflectivity correction and thickness inversion of each pixel of the three-dimensional curved workpiece.

[0047] In one embodiment, the step of obtaining local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and obtaining the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle, includes: S301. Obtain the optical constants of the metal substrate material in the acquisition band of the multispectral reflectance image, wherein the optical constants include refractive index and extinction coefficient; For example, this invention obtains the optical constants of a metal substrate material in the acquisition wavelength band of a multispectral reflectance image. The optical constants include the refractive index and the extinction coefficient, both of which are functions of wavelength and together constitute the complex refractive index of the metal substrate material, as shown in the following formula: ; in, The complex refractive index of the substrate, For refractive index, κ represents the extinction coefficient, and i is the imaginary unit. The metal substrate material is the base material of the workpiece; for example, when the workpiece is made of steel, the substrate material is steel; when the workpiece is made of copper, the substrate material is copper.

[0048] S302. Based on the three-dimensional point cloud data, obtain the local root mean square height of the surface of the local area corresponding to each pixel. For example, the present invention extracts surface roughness parameters of the local region corresponding to each pixel from the three-dimensional point cloud data. The surface roughness parameters include the local root-mean-square height of the surface. Taking the pixel as the center, a neighborhood point cloud within a preset radius is taken. Least-square plane fitting is performed on this neighborhood point cloud, and the height deviation of each point relative to the fitted plane is calculated. The root-mean-square value of this deviation is then used as the local root-mean-square height of the surface. Regions where the local root-mean-square height exceeds a preset upper limit are marked as high-roughness regions to distinguish them from smooth regions.

[0049] S303. Based on the normal vector deviation angle, optical constant, and local root mean square height, calculate the geometric correction factor corresponding to each pixel point based on the Fresnel-Gaussian mixed reflection model of rough surface. For example, the present invention calculates the geometric correction factor corresponding to each pixel based on the normal vector deviation angle of each pixel obtained in step S2, the optical constant obtained in step S301, and the local root mean square height of the surface obtained in step S302, using a Fresnel-Gaussian mixed reflection model for rough surfaces. This model comprehensively considers the attenuation effect of surface micro-roughness on specular reflection and the diffuse reflection component generated by the rough surface.

[0050] For unpolarized light, the standard Fresnel reflectance The reflectance of s-polarized and p-polarized light is the average value, depending only on the normal vector deviation angle θ and the complex refractive index of the substrate. The Fresnel reflectance coefficient of s-polarized light is calculated as follows: ; in, This represents the Fresnel reflection coefficient of s-polarized light. The complex refractive index of the substrate, The deviation angle; The formula for calculating the Fresnel reflection coefficient of p-polarized light is: ; in, This represents the Fresnel reflection coefficient of p-polarized light. The complex refractive index of the substrate, The deviation angle; Standard Fresnel reflectance The average value of the modulus square of reflectance for s-polarized and p-polarized light is calculated using the following formula: ; Based on the Fresnel-Gaussian mixed reflection model for rough surfaces, the formula for calculating the geometric correction factor is as follows: ; in, This represents the geometric correction factor, where θ is the normal vector deviation angle. The wavelength of the current acquisition band. For diffuse reflection components. Exponential decay term. Characterizing the attenuation of specular reflection due to roughness—local root mean square of the surface The greater the height or the shorter the wavelength, the more significant the attenuation; diffuse reflection component The relationship between surface roughness and diffuse reflection of workpieces in the same batch was obtained through a prior calibration experiment (specifically, standard samples of different roughness grades were prepared, and their diffuse reflection intensity was measured under the same incident conditions to establish a standard sample). and (The mapping relationship between them).

[0051] When θ = 0° and When =0, =1 indicates no correction is needed; when θ>0° or When >0, for metal surfaces, >1 indicates that combined geometric and roughness attenuation compensation is required.

[0052] The above diffuse reflection components The relationship between surface roughness and diffuse reflection of workpieces in the same batch was obtained through a prior calibration experiment. The calibration experiment was performed according to the following steps: (a) Preparation of standard samples: Take a metal plate sample of the same material as the workpiece to be tested (such as 45# steel, Q235 steel or copper alloy, depending on the specific workpiece material), with dimensions of 50mm×50mm×10mm. Prepare samples with no less than 6 roughness grades through machining and sandblasting / grinding processes. The surface roughness Ra values ​​are 0.1μm, 0.4μm, 1.0μm, 2.5μm, 5.0μm and 10.0μm (corresponding to root mean square height). (Approximately 1.1 to 1.25 times the Ra value), with no fewer than 3 samples for each grade, and the average value is taken to eliminate random errors.

[0053] (b) Measurement apparatus and conditions: The same multispectral camera and optical lens as the online detection system were used. In a dark room environment (ambient illuminance <0.1 lux), the sample surface was irradiated with the same light source incident angle (45°±1°) and the same working distance (300mm±5mm) as the online detection system. The reflected light intensity images of the sample at each roughness grade at each characteristic wavelength (λ1, λ2, ... λ_n) were collected, and the signal of the standard reference plate was measured simultaneously to eliminate light source fluctuations.

[0054] (c) Diffuse reflection separation method: For each sample block, its total reflectance Among them, the specular reflection component By setting up a blocking screen along the specular reflection direction (i.e., the direction symmetrical to the incident angle) to eliminate the obstruction, the integrated signal of the detector in the range of 10° to 30° away from the specular reflection direction is taken as the diffuse reflection component. The measured values ​​were obtained. Five measurements were taken from three samples for each roughness grade, and the average value was taken as the roughness value for that grade. value.

[0055] (d) Mapping relationship fitting: using the local root mean square height of the surface As the independent variable, with diffuse reflection component Using the least squares method as the dependent variable, the mapping relationship is obtained by fitting: in, , and To the angle of incidence The relevant fitting coefficients were determined by least-squares curve fitting of the above 6 calibration points. For 45# steel, the typical fitting values ​​under the conditions of wavelength 3.4μm and incident angle 45° are: α≈0.025, β≈0.15μm⁻², γ≈0.003μm⁻¹ (these values ​​vary depending on the material batch and surface condition, and need to be recalibrated for actual application).

[0056] (e) Scope of Application and Update Rules: The above calibration relationship applies to workpieces of the same material and with the same batch of surface treatment processes. When the workpiece material is changed or the surface pretreatment process is altered, the above calibration experiment must be repeated. It is recommended that the calibration be performed once before the first inspection of each batch of workpieces.

[0057] S304. Obtain the equivalent normal incident reflectance based on the geometric correction factor and the measured reflectance.

[0058] For example, the present invention calculates the geometric correction factor for each pixel. Then, based on the geometric correction factor and the measured reflectivity Obtain the equivalent normal incident reflectivity. The calculation formula is: ; The measured reflectance is a reflectance value obtained under actual measurement conditions, which is simultaneously affected by oil film thickness, surface attitude (i.e., normal vector deviation angle), and surface micro-roughness. According to the theory of electromagnetic wave reflection on a rough, oil-film-containing metal surface, the measured reflectance is equal to the spectral response of the oil film itself multiplied by a geometric correction factor (which includes the combined effects of Fresnel reflection, roughness attenuation, and diffuse reflection). Therefore, in order to extract the reflectance component determined solely by oil film thickness from the measured reflectance, it is necessary to divide the measured reflectance by the geometric correction factor to isolate the contributions of surface attitude and surface roughness to the reflectance.

[0059] The physical meaning of the equivalent normal incident reflectance is: if the rough curved surface is "flattened" into an ideal smooth plane with the normal vector coinciding with the optical axis, the reflectance value that should be measured under the same oil film thickness. This equivalent normal incident reflectance depends only on the oil film thickness and is independent of the surface orientation and surface roughness, thus satisfying the applicability premise of the Lambert-Beer law. Through the correction in this step, curved surfaces with different orientations and roughnesses are uniformly mapped to the reflectance value under the equivalent normal incident reference, enabling the subsequent Lambert-Beer law inversion to be accurately applied to three-dimensional curved surface workpieces with arbitrary orientations and surface roughness within the oil slinger.

[0060] As described in steps S301-S304 above, the present invention first obtains the optical constants (including refractive index and extinction coefficient) of the metal substrate material in the acquisition band of the multispectral reflection image, providing material optical property parameters for the calculation of the geometric correction factor; then, based on the three-dimensional point cloud data, it obtains the local root mean square height of the surface of each pixel, thereby quantifying the micro-roughness distribution of the workpiece surface; then, based on the normal vector deviation angle, optical constants, and local root mean square height, it calculates the geometric correction factor corresponding to each pixel based on the Fresnel-Gaussian mixed reflection model of rough surface, thereby simultaneously quantifying the Fresnel reflection gain caused by surface tilt and the specular reflection attenuation and diffuse reflection components caused by surface roughness of each pixel; finally, based on the geometric correction factor and the measured reflectance, it obtains the equivalent normal incident reflectance, thereby removing the component jointly contributed by surface attitude and surface roughness from the measured reflectance. Through the synergistic effect of the above steps, this invention uniformly corrects the reflectivity measured under arbitrary curved surface posture and arbitrary surface roughness to the standard reflectivity under equivalent normal incident conditions. This eliminates the reflectivity distortion caused by the combined effects of changes in the three-dimensional curved surface posture and surface micro-roughness of the workpiece, enabling the subsequent Lambert-Beer law thickness inversion to be accurately applied to three-dimensional curved surface workpieces with arbitrary roughness inside the oil slinger. This fundamentally solves the technical limitation of existing infrared spectroscopy thickness measurement technology, which is only applicable to ideal smooth and flat surfaces with perpendicular incident light.

[0061] In one embodiment, the steps of obtaining the process setting parameters for this oil-scraping process, establishing an oil film residue distribution model after oil scraping based on the process setting parameters, using the residue distribution model as a regularization constraint term, performing Lambert-Beer inversion on the equivalent normal incident reflectance, and obtaining preliminary values ​​of the oil film thickness for each pixel include: S401. Obtain the real-time reflectance signal of the fixed reference plate within the same camera field of view as the workpiece, and based on the real-time reflectance signal, synchronously obtain the reference reflectance of the fixed reference plate at each wavelength at the time of multispectral reflectance image acquisition. Exemplarily, the present invention sets a fixed reference plate at a fixed stationary position within the camera's field of view. The fixed reference plate and the workpiece are located within the same camera field of view and at a fixed stationary position within the camera's field of view. The surface of the fixed reference plate is not covered by an oil film. Because the fixed reference plate is located at a fixed stationary position within the camera's field of view, it shares the same optical path environment with the workpiece—that is, light emitted from the light source passes through the same optical elements and the same oil mist medium before reaching the reference plate surface, and reflected light from the reference plate surface then passes through the same oil mist medium and the same optical elements before reaching the sensor. Therefore, the real-time reflectivity signal of the fixed reference plate naturally contains all the attenuation information of the current optical path, including the multiplicative fluctuations of the light source and the multiplicative scattering attenuation of the oil mist.

[0062] During the acquisition of the multispectral reflectance image, this invention simultaneously acquires the reference reflectance of a fixed reference plate at each wavelength, serving as a real-time reflectance signal. This synchronous acquisition is achieved through a synchronous trigger signal from a rotary encoder, ensuring strict temporal alignment between the reference plate signal and the workpiece's multispectral reflectance image. Since the fixed reference plate surface is uncovered by an oil film, any changes in its real-time reflectance signal are entirely caused by optical path interference (light source fluctuations and oil mist attenuation), and do not contain any oil film thickness information.

[0063] S402. Obtain the initial calibrated reflectivity of the fixed reference plate; For example, in the present invention, when the oil slinger is unloaded and free of oil mist, the reference reflectance of the fixed reference plate is pre-collected as the initial calibration reflectance. The unloaded and oil-mist-free state refers to the initial state where there are no workpieces inside the oil slinger and no oil mist has been generated. At this time, there is no oil mist scattering attenuation in the optical path, and the light source is in its initial working state. The reflectance of the fixed reference plate collected in this state is the initial calibration reflectance, representing the reference plate reflectance benchmark value under interference-free conditions. This initial calibration reflectance is collected during the system initialization phase or each time the system is powered on and stored in the system for subsequent calculations.

[0064] S403. Obtain the comprehensive attenuation coefficient based on the initial calibrated reflectivity and the reference reflectivity; For example, the present invention obtains the comprehensive attenuation coefficient of the current optical path based on the initial calibrated reflectance and the reference reflectance. The comprehensive attenuation coefficient is the ratio of the initial calibrated reflectance to the reference reflectance, characterizing all multiplicative attenuation changes that occur in the optical path from the initial calibration time to the current acquisition time, including changes in incident light intensity caused by light source fluctuations and attenuation of reflected light caused by oil mist scattering. When there is no attenuation change in the optical path, the comprehensive attenuation coefficient is 1; when there is attenuation in the optical path, the comprehensive attenuation coefficient is greater than 1.

[0065] S404. Obtain the signal from the oil mist concentration sensor inside the oil slinger and the cumulative usage time of the fixed reference plate; Exemplarily, the present invention acquires the oil mist concentration sensor signal inside the oil slinger and the cumulative usage time of the fixed reference plate. The oil mist concentration sensor signal is acquired in real time by an industrial-grade dust concentration sensor installed near the camera's optical path, and is used to characterize the real-time concentration level of oil mist in the current optical path. The cumulative usage time of the fixed reference plate refers to the cumulative operating time since the last cleaning of the reference plate, and is used to characterize the degree of contamination on the surface of the reference plate caused by the slow deposition of oil mist particles.

[0066] S405. Based on the comprehensive attenuation coefficient, the oil mist concentration sensor signal, and the cumulative usage time, obtain the reversible scattering attenuation coefficient of the current optical path; For example, the present invention obtains the reversible scattering attenuation coefficient of the current optical path based on the comprehensive attenuation coefficient, the oil mist concentration sensor signal, and the cumulative usage time. Specifically, the comprehensive attenuation coefficient includes two multiplicative components: one is the reversible multiplicative attenuation caused by oil mist scattering, which changes dynamically in real time with the oil mist concentration—the reversible scattering attenuation coefficient increases when the concentration increases and decreases when the concentration decreases; the other is the irreversible multiplicative attenuation caused by contamination of the reference plate surface, which monotonically increases with the cumulative usage time and does not recover with fluctuations in oil mist concentration.

[0067] By using a pre-calibrated oil mist concentration-scattering attenuation curve, the reversible scattering attenuation coefficient corresponding to the current oil mist concentration is separated from the comprehensive attenuation coefficient. The calibration method for the oil mist concentration-scattering attenuation curve is as follows: with the oil slinger unloaded, oil mist of a known concentration is injected into the cavity through an artificial atomizer, and the comprehensive attenuation coefficient at different oil mist concentrations is measured to establish a mapping relationship between oil mist concentration and scattering attenuation coefficient. After calibration, for any moment in actual detection, the reversible scattering attenuation coefficient of the current optical path can be obtained by querying the relationship curve based on the real-time oil mist concentration sensor signal.

[0068] The calibration of the above oil mist concentration-scattering attenuation relationship curve shall be performed according to the following steps: (a) Calibration Environment Preparation: With the oil slinger unloaded (no workpiece inside), close the oil slinger chamber door, turn on the lighting source and preheat for 30 minutes until the light output is stable (light source fluctuation <±0.5%). The oil mist concentration sensor (using an industrial-grade laser scattering dust concentration sensor, range 0~500mg / m³, accuracy ±5%FS) is installed approximately 50mm from the camera's optical path, at the same horizontal height as the camera's field of view center.

[0069] (b) Oil mist generation and concentration control: The actual oil used in this oil-sanding process (i.e., the cutting oil / cooling oil used on the workpiece to be tested; the oil type is based on the production batch, and new oil of the same type must be used during calibration) is atomized using an ultrasonic atomizer and injected into the oil-sanding machine cavity. By adjusting the power and injection time of the ultrasonic atomizer, six concentration gradients are generated: 0 mg / m³ (background), 10 mg / m³, 25 mg / m³, 50 mg / m³, 100 mg / m³, and 200 mg / m³. Each concentration gradient is monitored and confirmed in real time by an oil mist concentration sensor. Subsequent measurements can only be performed after the sensor reading stabilizes (fluctuation does not exceed ±5% for 30 consecutive seconds).

[0070] (c) Measurement of comprehensive attenuation coefficient: Under each concentration gradient, the reflected light intensity signal of the fixed reference plate is collected synchronously. and reference board initial calibration signal (i.e., the signal when the oil mist concentration is 0), calculate the comprehensive attenuation coefficient at each concentration. At least 10 images were continuously acquired for each concentration gradient, and the average value was taken.

[0071] (d) Reversible scattering attenuation coefficient separation: irreversible contamination attenuation on the reference plate surface The exponential decay model based on the cumulative usage time t Estimate (η is the contamination rate constant of the reference plate, calibrated through a cleanliness comparison experiment before and after the reference plate is used). Then the reversible scattering attenuation coefficient at the current concentration is: .

[0072] (e) Calibration curve fitting: using oil mist concentration C as the independent variable and reversible scattering attenuation coefficient as the inverse scattering attenuation coefficient. For the dependent variable, piecewise fitting is used: -When C≤50mg / m³: (Linear segment, k1 is the linear coefficient); -When C>50mg / m³: (Logarithmic saturation segment, k2 is the logarithmic coefficient); The values ​​of k1 and k2 are determined using the least squares method. For typical mineral oil-based cutting oils, the calibration results at a wavelength of 3.4 μm are k1≈0.0015 (mg / m³)⁻¹, k2≈0.12 (these values ​​vary depending on the oil type and optical path geometry, and must be recalibrated for practical applications). The coefficient of determination R² of the calibration curve should not be less than 0.98; otherwise, recalibration is required.

[0073] (f) Periodic calibration: The calibration curve is calibrated once a month. During calibration, the measured values ​​at two concentration points, 50 mg / m³ and 100 mg / m³, are compared with the predicted values ​​of the curve. If the deviation exceeds ±10%, the entire calibration process must be repeated.

[0074] S406. Obtain the compensated reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient.

[0075] For example, after obtaining the reversible scattering attenuation coefficient of the current optical path, the present invention obtains the compensated reflectance based on the equivalent normal incident reflectance and the reversible scattering attenuation coefficient. Specifically, the compensated reflectance is obtained by multiplying the equivalent normal incident reflectance by the reversible scattering attenuation coefficient. Since the reversible scattering attenuation coefficient is the reciprocal of the multiplicative attenuation of the optical path, this multiplication operation accurately recovers the portion of the equivalent normal incident reflectance lost due to reversible scattering of oil mist, thereby eliminating the interference of real-time oil mist concentration fluctuations on reflectance measurement.

[0076] The compensated reflectivity has simultaneously eliminated the effects of surface attitude, surface roughness, and reversible scattering attenuation caused by oil mist in the optical path, retaining only the reflectivity information determined by the oil film thickness as input for subsequent Lambert-Beer inversion.

[0077] Through the synergistic effect of steps S401-S406 above, this invention utilizes a fixed reference plate within the same camera field of view as the workpiece to achieve compensation for the multiplicative attenuation ratio between the workpiece measurement optical path and the reference optical path. This compensation method differs fundamentally from the known beam splitter reference compensation scheme: the beam splitter takes the reference light at the light source end, and the reference light does not pass through the oil mist area, making it impossible to sense the oil mist attenuation in the latter part of the optical path; the fixed reference plate of this invention is set at the workpiece end, sharing the complete optical path with the workpiece, and can comprehensively sense the attenuation of the entire optical path, including oil mist. Simultaneously, by introducing the oil mist concentration sensor signal and cumulative usage time, this invention further distinguishes between reversible oil mist scattering attenuation and irreversible reference plate contamination attenuation, avoiding measurement errors introduced by mistakenly compensating for optical path attenuation due to reference plate contamination itself. This fundamentally solves the technical problem of inaccurate thickness measurement caused by reflectivity signal drift under dynamically changing oil mist environments.

[0078] S501. Obtain the process setting parameters and oil parameters for this oil-slinging process from the oil-slinging machine control system, wherein the process setting parameters include the rotational angular velocity and the oil-slinging duration. For example, the present invention obtains the process setting parameters and oil parameters for this oil-slinging process from the oil-slinging machine control system. The oil-slinging machine control system is the main control unit of the oil-slinging machine, which records and stores all process setting values ​​for this oil-slinging process. The process setting parameters include rotational angular velocity and oil-slinging duration, where the rotational angular velocity is the target rotational angular velocity set for the oil-slinging machine spindle during the oil-slinging stage, in rad / s; and the oil-slinging duration is the duration for which the oil-slinging machine performs the oil-slinging action, in seconds. The oil parameters are the oil property parameters used in this oil-slinging process, including oil type, oil density, and oil dynamic viscosity, which can be obtained from the oil management database of the oil-slinging machine control system.

[0079] S502. Based on the spatial position of the three-dimensional point cloud data after registration with the CAD standard template, obtain the rotation radius of each pixel in the workpiece coordinate system. For example, the present invention obtains the rotation radius of each pixel in the workpiece coordinate system based on the spatial position of the 3D point cloud data registered with the CAD standard template in step S2. Specifically, after the 3D point cloud data and the CAD standard template complete iterative nearest-point registration, each spatial point in the 3D point cloud data has been mapped to its corresponding position in the workpiece coordinate system, and a pixel-level one-to-one correspondence has been established between each pixel in the multispectral reflectance image and each spatial point in the 3D point cloud. For any pixel (x, y), its spatial position in the workpiece coordinate system is (X, Y, Z), then the rotation radius r(x, y) corresponding to this pixel is calculated by the following formula: ; Where X and Y are the two coordinate components of the spatial position of the pixel in the workpiece coordinate system, perpendicular to the rotation axis. Through the above calculations, the distribution of rotation radii corresponding to each pixel on the entire surface of the workpiece is obtained.

[0080] S503. Obtain the oil density and dynamic viscosity based on the oil parameters; For example, the present invention obtains the oil density ρ and the oil dynamic viscosity μ based on the oil parameters obtained in step S501. The oil density ρ represents the mass per unit volume of the oil, with units of kg / m³, and can be obtained by measuring the oil sample with a densitometer or by consulting an oil property handbook. The oil dynamic viscosity μ represents the viscosity of the oil, with units of Pa·s, and can be obtained by measuring with a rotational viscometer or capillary viscometer, or by consulting an oil property handbook. For the same batch of oil, the oil density and oil dynamic viscosity can be pre-calibrated and stored in the system during system initialization, and can be directly read and retrieved from the oil parameters during each test.

[0081] S504. Based on the rotational angular velocity, oil swing duration, rotation radius, oil density, and oil dynamic viscosity, establish a theoretical residual distribution model of the oil film after oil swing. For example, the present invention is based on the said rotational angular velocity The duration of oil slinging Based on the rotation radius r(x,y), the oil density ρ, and the oil dynamic viscosity μ, a theoretical residual distribution model of the oil film after oil ejection is established. .

[0082] Since the detection is triggered after the oil-slinging process is completed but before the workpiece has been removed from the oil-slinging machine cavity, the workpiece is stationary at this time, and centrifugal force no longer acts on the oil film. However, the centrifugal force applied by the high-speed rotation during the oil-slinging process determines the final residual distribution of the oil film at the end of the oil-slinging process—during the oil-slinging stage, the oil film migrates radially outward under the action of centrifugal force, with a thicker oil film in the central area and a thinner oil film in the edge area. This distribution is "frozen" after the oil-slinging process is completed. Therefore, the residual distribution is modeled a priori using the process setting parameters of this oil-slinging process.

[0083] Based on the radial flow theory of rotating thin liquid films under centrifugal force in fluid mechanics, the expression for the theoretical residual distribution model of the oil film after oil spillage is as follows: ; in The theoretical residual oil film thickness reference value for the central region of the workpiece (r=0) is in g / m², and is an unknown parameter to be estimated. It will be solved together with the thickness of each pixel in the subsequent inversion. R is the maximum rotation radius of the workpiece in m, which is obtained from the CAD standard template. r is the rotation radius corresponding to the pixel in m, which is obtained from step S502.

[0084] The physical meaning of this model is: after oil spillage, the residual oil film thickness exhibits an exponentially decreasing distribution along the radial direction—the central region (r=0) has the largest residual amount, and the edge region (r→R) has the smallest residual amount. The attenuation coefficient and... Positive correlation, i.e., rotational angular velocity The higher the viscosity (ρ), the greater the centrifugal force, the more significant the radial distribution gradient of the oil film, and the greater the thickness difference between the center and the edge. The higher the oil density (ρ) or the lower the dynamic viscosity (μ), the stronger the fluidity of the oil film under centrifugal force, the more complete the radial migration, and the more significant the distribution gradient. Oil removal time. In this model, through The value of indirectly reflects — The longer the film, the greater the total amount of oil film ejected, and the higher the residual baseline value in the central area. The smaller.

[0085] Through the above steps S501-S504, this invention establishes a theoretical residual distribution model of the oil film after oil removal based on the physical laws of the oil removal process. This model serves as a physical prior for subsequent regularization constraint inversion, ensuring that the thickness inversion results conform to the natural distribution law of the oil film after oil removal, which is "more residual in the central region and less residual in the edge region," effectively suppressing thickness anomalies caused by local measurement noise or registration errors.

[0086] As described in steps S501-S504 above, this invention first obtains the process setting parameters and oil parameters for this oil-slinging process from the oil-slinging machine control system. The process setting parameters include the rotational angular velocity and the oil-slinging duration, thus obtaining the process input required to establish the residual distribution model. Then, based on the spatial position of each pixel in the workpiece coordinate system after registration with the 3D point cloud data and the CAD standard template, the rotation radius of each pixel is obtained, thus establishing the correspondence between the pixel position and the radius of centrifugal force. Next, the oil density and dynamic viscosity are obtained based on the oil parameters, thus obtaining the physical property parameters of the oil itself. Finally, a theoretical residual distribution model of the oil film after oil slinging is established based on the rotational angular velocity, oil-slinging duration, rotation radius, oil density, and dynamic viscosity, thereby quantifying the physical laws of the oil-slinging process into a priori theoretical residual thickness distribution of each pixel on the entire surface of the workpiece. Through the synergistic effect of the above steps, this invention provides a physically consistent thickness distribution prior model for subsequent regularized constraint inversion, making the inversion results interpretable at the level of physical laws.

[0087] In one embodiment, the step of using the residual distribution model as a regularization constraint to perform Lambert-Beer inversion on the equivalent normal incident reflectance to obtain the preliminary value of the oil film thickness for each pixel includes: S601. Obtain the reference reflectance of the workpiece at the characteristic absorption wavelength and the preset Lambert-Beer inversion parameters; For example, the present invention obtains the reference reflectance of a workpiece at a characteristic absorption wavelength and preset Lambert-Beer inversion parameters. The characteristic absorption wavelength... The wavelength at which oil film molecules exhibit significant absorption characteristics in the infrared band is referred to as the reference reflectance. For example, for mineral oil-based cutting oils, the characteristic absorption wavelength is typically located in the CH stretching vibration absorption band around 3.4 μm. For the substrate material (i.e., a clean metal surface without an oil film covering) at the characteristic absorption wavelength The reflectance is a dimensionless quantity, obtained through prior reflectance measurement and calibration of clean workpiece samples. The preset Lambert-Beer inversion parameters include the characteristic absorption wavelength of the oil film. molar absorption coefficient at the location The constant c is related to the oil film density. The molar absorption coefficient... The density coefficient (c) is an inherent optical property of the oil, measured in m² / mol, characterizing the oil film's absorption capacity for infrared light of characteristic absorption wavelengths. It can be obtained by preparing a standard oil film sample of known thickness, measuring its transmission spectrum using an infrared spectrometer, and then calculating it using the transmission form of Lambert-Beer's law. The oil film density correlation constant (c) is the ratio of the oil's density to its molecular weight, measured in mol / m³. It can be obtained by measuring the oil's density with a hydrometer and calculating its average molecular weight based on the oil's main chemical components. All of the above parameters are inherent properties of the oil itself. Since the oil used in the same batch of production is identical, they can be pre-calibrated and stored in the system during system initialization for subsequent calculations.

[0088] S602. Construct the Lambert-Beer reflectance function based on the equivalent normal incident reflectance, reference reflectance and Lambert-Beer inversion parameters of each pixel. For example, the present invention obtains the equivalent normal incident reflectance of each pixel. The reference reflectivity and the Lambert-Beer inversion parameters (including molar absorption coefficient) After determining the constant c related to the oil film density, the Lambert-Beer reflectivity function is constructed. The Lambert-Beer law describes the physical law that the intensity of transmitted light decreases exponentially with the thickness of the medium when light passes through an absorbing medium. Its reflection form is expressed as follows: when incident light passes through the oil film layer, reaches the substrate surface, is reflected, and then passes through the oil film layer again, there is an exponential decay relationship between the intensity of reflected light and the thickness of the oil film.

[0089] The oil film thickness at pixel (x, y) is to be determined. Lambert-Beer reflectivity function The expression is: ; in Let be the oil film thickness at pixel (x, y), expressed in g / m². The physical meaning of this function is: [equation missing - likely referring to a reference reflectance value]. Based on the oil film thickness As the concentration of infrared light increases, the oil film absorbs infrared light more strongly, and the reflectivity decreases exponentially, with the rate of decrease decreasing from [previous value]. The material properties of the oil film (absorption capacity and density) together determine the intensity of the modulation of reflectivity by the thickness.

[0090] S603. Construct regularization constraint terms based on the oil film residue distribution model after oil spillage; For example, the present invention uses the oil film residue distribution model established in step S4 after oil spillage. A regularization constraint term is constructed. The regularization constraint term is used to embed the physical law of the oil-spinning process (radial migration effect driven by centrifugal force) into the optical measurement inversion process, so that the oil film thickness distribution obtained by inversion conforms to the theoretical residual law of the oil film after oil-spinning on the whole surface, that is, more residue in the central region and less residue in the edge region.

[0091] Specifically, regularization constraint terms The expression is: ; in, For pixels The oil film thickness to be determined is... This is the theoretical residual oil film distribution model value corresponding to this pixel after oil spillage. This term quantifies the sum of squares of the deviations between the thickness distribution to be solved and the theoretical residual distribution—the greater the deviation of the inverted thickness from the theoretical distribution, the larger the value of the regularization constraint term, and the stronger the penalty it receives in subsequent objective function optimization.

[0092] S604. Construct a coupled inversion objective function based on the Lambert-Beer reflectivity function and regularization constraint term; For example, the present invention is based on the Lambert-Beer reflectivity function and the regularization constraint term A coupled inversion objective function J(d) is constructed. This coupled inversion objective function comprises two terms: the first term is a data fidelity term, requiring the Lambert-Beer reflectance corresponding to the inverted thickness to approximate the measured equivalent normal incident reflectance; the second term is a physical constraint term, requiring the inverted thickness to conform to the theoretical residual distribution law of the oil film after oil spillage across the entire surface. The expression for the coupled inversion objective function is: ; in, For pixels Located at the characteristic absorption wavelength The equivalent normal incident reflectance is a dimensionless quantity, obtained by the geometric normalization correction in step S3. The Lambert-Beer reflectivity function is constructed by step S602; The theoretical residual distribution model value of the oil film after oil spillage is established by step S4; , is the regularization weighting coefficient, a dimensionless quantity used to balance the relative importance between data fidelity terms and physical constraint terms. It is determined from historical test data using the L-curve method, and typically takes a value of 0.3 to 0.7. When a larger value is taken, the weight of the physical constraints is enhanced, and the inversion results are more likely to conform to the theoretical residual distribution; when When a smaller value is taken, the weight of the data fidelity term is increased, and the inversion results are more likely to match the measured reflectance data.

[0093] S605. Iteratively optimize the coupled inversion objective function to obtain the preliminary value of the oil film thickness of each pixel.

[0094] For example, the present invention iteratively optimizes the coupled inversion objective function J(d) to obtain the oil film thickness of each pixel that minimizes the objective function, which is used as the preliminary value of the oil film thickness of each pixel. Specifically, the objective function is solved iteratively using the gradient descent method: the oil film thickness at each pixel is initialized as follows. That is, the initial value of the thickness distribution to be solved is set as the theoretical residual distribution model value; the objective function is calculated. right The gradient direction is updated along the gradient descent direction. Repeat the above process iteratively until the change in d between two adjacent iterations is less than a preset convergence threshold (e.g., 1e-6g / m²), or until a preset maximum number of iterations (e.g., 1000) is reached. After convergence... This is the preliminary value of the oil film thickness at each pixel.

[0095] The physical meaning of this iterative optimization process is as follows: while keeping the inverted thickness as close as possible to the theoretical residual distribution (the second constraint), the Lambert-Beer reflectivity corresponding to the inverted thickness should be as close as possible to the measured equivalent normal incident reflectivity (the first constraint). Finally, by balancing these two objectives across the entire surface, a physically consistent and data-matched preliminary thickness estimate is obtained. Through this coupled inversion step, even if local regions experience deviations in the data fidelity term due to reflectivity measurement noise or registration errors, the physical constraints can still confine the inversion results within a reasonable physical distribution range, thereby effectively suppressing the generation of isolated outliers.

[0096] As described in steps S601-S605 above, the present invention first obtains the reference reflectance of the workpiece at the characteristic absorption wavelength and the preset Lambert-Beer inversion parameters, providing material property parameters for the quantitative application of the Lambert-Beer law; then, based on the equivalent normal incident reflectance, reference reflectance, and Lambert-Beer inversion parameters of each pixel, a Lambert-Beer reflectance function is constructed, thereby establishing a physical quantitative relationship between oil film thickness and reflectance; then, based on the oil film residue distribution model after oil removal, a regularization constraint term is constructed, thereby quantifying the physical laws of the oil removal process into constraint components of the inversion objective function; then, based on the Lambert-Beer reflectance function and the regularization constraint term, a coupled inversion objective function is constructed, thereby realizing the joint optimization of optical measurement data and physical prior constraints; finally, the coupled inversion objective function is iteratively optimized to obtain the preliminary value of the oil film thickness of each pixel, thereby achieving the optimal balance between data fidelity and physical constraints. Through the synergistic effect of the above steps, this invention embeds the physical laws of the oil-splashing process into the optical measurement inversion process, so that the thickness inversion results have physical consistency across the entire surface. Even if there is noise in the local reflectivity measurement, the inversion results are still constrained by the residual distribution and will not produce physically unreasonable isolated extreme values, providing a high-quality initial thickness estimate for subsequent confidence-weighted fusion.

[0097] In one embodiment, the step of generating an oil film thickness distribution map based on the final oil film thickness value includes: S505. Obtain the final value of the oil film thickness of each pixel and its corresponding pixel coordinates; For example, after obtaining the final value of the oil film thickness of each pixel, the present invention obtains the final value of the oil film thickness of each pixel. and its corresponding pixel coordinates (x, y). The final value of the oil film thickness. Obtained in step S504, the unit is g / m²; the pixel coordinates (x, y) are the spatial position indices of each pixel in the multispectral reflectance image, where x is the horizontal coordinate and y is the vertical coordinate, with units of pixels. The final oil film thickness of each pixel and its corresponding pixel coordinates together constitute the basic data pair for generating the thickness distribution map, providing a data source for subsequent spatial mapping.

[0098] S506. Establish a spatial mapping relationship based on the pixel coordinates of each pixel, and fill the final value of the oil film thickness of each pixel into the corresponding pixel position of the blank map according to the spatial mapping relationship to generate an initial thickness distribution map; For example, this invention establishes a spatial mapping relationship based on the pixel coordinates (x, y) of each pixel, wherein the spatial mapping relationship is a one-to-one correspondence between pixel coordinates and pixel positions in the blank atlas. Specifically, the blank atlas is a blank image matrix with the same width and height (same number of pixel rows and columns) as the multispectral reflectance image, and each pixel position is initially empty. This invention sets the final value of the oil film thickness at each pixel. Fill the blank map with the corresponding pixel coordinates (x, y) according to the spatial mapping relationship, that is, let ,in This represents the thickness value at pixel (x, y) in the initial thickness distribution map, in g / m². After filling, the initial thickness distribution map is obtained. .

[0099] S507. Obtain the unfilled pixels in the initial thickness distribution map, and perform hole interpolation filling based on the final value of the oil film thickness of the neighboring pixels of the unfilled pixels to generate a complete thickness distribution map. For example, the present invention generates an initial thickness distribution map. Then, all pixel positions in the initial thickness distribution map are traversed to detect whether there are any unfilled pixels. Unfilled pixels refer to pixel positions that failed to be assigned a thickness value in step S506 because the final oil film thickness value of the corresponding pixel failed to be obtained or was invalid. In actual operation, due to potential inconsistencies in coverage between the multispectral reflectance image and the 3D point cloud data at the edge of the field of view, or because the registration confidence of some pixels is extremely low, causing their final oil film thickness value to be determined as invalid in step S504, a small number of unfilled pixels (holes) may exist in the initial thickness distribution map.

[0100] For each unfilled pixel detected The present invention performs hole interpolation filling based on the final value of the oil film thickness of the neighboring pixels of the unfilled pixel to obtain the interpolated thickness value of the unfilled pixel. The calculation formula is: ; in, Unfilled pixels The interpolated thickness value at the location, in g / m², is the quantity to be solved; N is the number of filled pixels in the neighborhood, a positive integer, determined by the preset neighborhood search radius. Determined by the actual distribution of filled pixels; For the i-th filled pixel in the neighborhood to unfilled pixels The Euclidean distance, in pixels, is calculated using the following formula: ; in, The thickness value of the i-th filled pixel in the neighborhood, in units of g / m², read from the initial thickness distribution map.

[0101] The calculated interpolated thickness value Fill to unfilled pixels Positioning is used to fill holes. After traversing all unfilled pixels, thickness values ​​are assigned to all pixel positions, generating a complete thickness distribution map. .

[0102] S508. Perform spatial smoothing filtering on the complete thickness distribution map to eliminate thickness jumps between pixels and generate the oil film thickness distribution map. For example, the present invention obtains a complete thickness distribution map. Then, spatial smoothing filtering is applied to the complete thickness distribution map to eliminate thickness jumps between pixels, generating the final oil film thickness distribution map. In a complete thickness distribution map, different regions may use different thickness sources (high-confidence regions use inversion results directly, low-confidence regions use neighborhood interpolation results, and hole regions use hole interpolation results), leading to abrupt changes in thickness values ​​between adjacent pixels. This results in unnatural edges or stair-step effects in the thickness distribution map. To eliminate these thickness abrupt changes and improve the spatial continuity of the thickness distribution map, this invention performs spatial smoothing filtering on the complete thickness distribution map.

[0103] The spatial smoothing filter can be implemented using Gaussian filtering or median filtering. Taking Gaussian filtering as an example, for each pixel in the complete thickness distribution map, a filter window of a preset size is taken centered on that pixel. Within the filter window, a Gaussian kernel function is used to perform a weighted average of the thickness values ​​of each pixel. The weighted average result is then used as the filtered thickness value for that pixel. Gaussian filtering can smooth the thickness distribution while effectively preserving the overall trend of thickness variation. Taking median filtering as an example, for each pixel in the complete thickness distribution map, a filter window of a preset size is taken centered on that pixel. The thickness values ​​of all pixels within the filter window are sorted by size, and the median value is used as the filtered thickness value for that pixel. Median filtering can effectively suppress isolated impulse noise while smoothing the thickness. After spatial smoothing filtering, thickness jumps between pixels are eliminated, generating the final oil film thickness distribution map. Through the above steps S505 to S508, the present invention realizes the generation of a complete and smooth oil film thickness distribution map from the final value of the oil film thickness of each pixel.

[0104] As described in steps S505-S508 above, the present invention first obtains the final value of the oil film thickness of each pixel and its corresponding pixel coordinates, providing a data foundation for the generation of the thickness distribution map; then, a spatial mapping relationship is established based on the pixel coordinates of each pixel, and the final value of the oil film thickness of each pixel is filled into the corresponding pixel position of the blank map according to the spatial mapping relationship to generate an initial thickness distribution map, thereby completing the basic mapping of thickness values ​​from each pixel to the spatial map; then, the unfilled pixels in the initial thickness distribution map are obtained, and hole interpolation is performed based on the final value of the oil film thickness of the neighboring pixels of the unfilled pixels to generate a complete thickness distribution map, thereby eliminating holes in the distribution map caused by inconsistent data coverage or invalid values; finally, spatial smoothing filtering is performed on the complete thickness distribution map to eliminate thickness jumps between pixels and generate the oil film thickness distribution map, thereby improving the spatial continuity of the thickness distribution map.

[0105] Through the synergistic effect of the above steps, this invention transforms the discrete oil film thickness values ​​of each pixel into a continuous, smooth, and complete full-surface thickness distribution map, providing a structured and visualized data foundation for the subsequent online detection results of residual oil film thickness.

[0106] In one embodiment, the step of outputting the online detection result of residual oil film thickness based on the oil film thickness distribution map includes: S601. Obtain the final value of the oil film thickness and its corresponding pixel coordinates for each pixel in the oil film thickness distribution map; For example, the present invention generates an oil film thickness distribution map. Then, the final value of the oil film thickness at each pixel in the oil film thickness distribution map is obtained. and its corresponding pixel coordinates (x, y). The final value of the oil film thickness. The thickness values ​​of each pixel obtained after weighted constraints in step S504 are the pixel coordinates (x, y), which are the spatial position indices of each pixel in the oil film thickness distribution map. The final oil film thickness values ​​of each pixel in the oil film thickness distribution map and their corresponding pixel coordinates together constitute the data basis for subsequent statistical analysis, defect judgment, and detection result output.

[0107] S602. Based on the final value of the oil film thickness of each pixel, calculate the thickness statistical characteristics of the oil film thickness distribution map, wherein the thickness statistical characteristics include the mean thickness, peak thickness, minimum thickness, and standard deviation of thickness. For example, after obtaining the final oil film thickness value of each pixel in the oil film thickness distribution map, the present invention performs statistical analysis on the final oil film thickness values ​​of all pixels in the entire map to calculate the thickness statistical characteristics of the oil film thickness distribution map. The thickness statistical characteristics include the average thickness. Peak thickness Minimum thickness and thickness standard deviation .

[0108] average thickness The formula used to characterize the overall level of residual oil film thickness on the workpiece surface is as follows: ; in, The average thickness is expressed in g / m²; N represents the total number of pixels in the oil film thickness distribution map. This represents the final oil film thickness at the i-th pixel, expressed in g / m².

[0109] peak thickness The maximum value of the final oil film thickness at all pixels in the oil film thickness distribution map is used to characterize the most severe local oil film exceedance. The calculation formula is: ; Minimum thickness The minimum value of the final oil film thickness at all pixels in the oil film thickness distribution map is used to characterize the residual oil film level in the thinnest region of the workpiece surface. The calculation formula is: ; Thickness standard deviation The formula used to characterize the dispersion and non-uniformity of oil film thickness distribution across the entire surface is as follows: ; in, The value represents the standard deviation of the oil film thickness, expressed in g / m². A larger value indicates a more uneven distribution of the oil film thickness across the entire surface, resulting in a less stable oil-removing effect.

[0110] By calculating the above-mentioned thickness statistical characteristics, this invention quantifies the overall level, extreme values, and dispersion of residual oil film on the workpiece surface from a macroscopic perspective, providing multi-dimensional quantitative indicators for process evaluation.

[0111] S603. Compare the final value of the oil film thickness of each pixel with a preset thickness threshold pixel by pixel, and generate a defect judgment label for each pixel based on the comparison result. For example, the present invention obtains the final value of the oil film thickness for each pixel. Then, obtain the preset thickness threshold. The preset thickness threshold The maximum allowable residual oil film thickness is expressed in g / m². It is preset by the process requirements and stored in the system. When the final oil film thickness of a pixel exceeds this threshold, the pixel is considered a defective pixel.

[0112] This invention calculates the final value of the oil film thickness for each pixel. With preset thickness threshold Perform pixel-by-pixel comparison and generate defect identification labels for each pixel based on the comparison results. Defect identification label For binary tags, the value is either 0 or 1. The decision logic is: if... ,but (Defective pixels, excessive oil film thickness); If ,but (Qualified pixels, oil film thickness meets standards). By comparing pixels one by one and generating defect judgment tags, this invention converts continuously distributed thickness values ​​into discretely distributed defect judgment results, providing basic data for subsequent defect statistics. S604. Count the total number of defect pixels in the oil film thickness distribution map according to the defect judgment label, and calculate the defect area ratio based on the total number of defect pixels and the total number of pixels in the oil film thickness distribution map. For example, the present invention generates defect determination labels for each pixel. Then, the total number of defective pixels in the oil film thickness distribution map was counted. The calculation formula is: ; in, is the total number of defective pixels, in units of pixels; N is the total number of pixels in the oil film thickness distribution map; This is the defect label for the i-th pixel, with a value of 0 or 1.

[0113] Further calculation of the defect area ratio This indicator is used to characterize the proportion of the area with excessive residual oil film thickness on the workpiece surface to the total surface area. The calculation formula is: ; in, The percentage represents the area of ​​defects; it is dimensionless and expressed as a percentage. denoted as the total number of defective pixels; N represents the total number of pixels in the oil film thickness distribution map. By calculating the percentage of defective area, this invention quantifies the severity of excessive oil film from the perspective of area coverage.

[0114] S605. Generate a detection result report based on the thickness statistical characteristics, the defect judgment label, and the defect area ratio, and output the detection result report as the online detection result of the residual oil film thickness; For example, after completing the calculation of thickness statistical features, the generation of defect judgment labels, and the calculation of defect area ratio, the present invention, based on the thickness statistical features (including the average thickness), Peak thickness Minimum thickness and thickness standard deviation The defect determination label (Including the defect determination results and spatial distribution information of each pixel) and the percentage of the defect area. Generate a test result report.

[0115] The detection result report is a structured data file or message, which includes at least: pixel coordinates of each pixel, final oil film thickness of each pixel, defect determination label of each pixel, thickness statistical characteristics (mean thickness, peak thickness, minimum thickness, standard deviation of thickness), defect area ratio, and an overview of defect region distribution generated based on the defect determination labels. The detection result report may also include the oil film thickness distribution map itself as a visualization attachment.

[0116] This invention outputs the detection result report as the online detection result of residual oil film thickness. The online detection result of residual oil film thickness can be output to the oil slinger control system to guide the real-time adjustment of oil slinger process parameters. For example, when the defect area ratio exceeds a preset upper limit, the oil slinger control system automatically increases the oil slinger time or increases the oil slinger speed. The online detection result of residual oil film thickness can also be output to the host computer quality management system for product quality traceability, batch qualification judgment, and process trend analysis. Through the above steps S601 to S605, this invention completes the output of a structured, traceable, and verifiable online detection result of residual oil film thickness, from an oil film thickness distribution map.

[0117] As described in steps S601-S605 above, the present invention first obtains the final value of the oil film thickness and its corresponding pixel coordinates for each pixel in the oil film thickness distribution map, providing basic data for subsequent statistical analysis and defect judgment; then, based on the final value of the oil film thickness for each pixel, the thickness statistical characteristics of the oil film thickness distribution map are calculated, including the thickness mean, thickness peak, thickness minimum, and thickness standard deviation, thereby quantifying the overall level, extreme values, and dispersion of the residual oil film on the workpiece surface from a macroscopic perspective; then, the final value of the oil film thickness for each pixel is compared pixel by pixel with a preset thickness threshold, and a defect judgment label for each pixel is generated based on the comparison results, thereby converting the continuously distributed thickness values ​​into discretely distributed defect judgment results.

[0118] Next, based on the defect judgment label, the total number of defect pixels in the oil film thickness distribution map is counted, and the defect area ratio is calculated based on the total number of defect pixels and the total number of pixels in the oil film thickness distribution map, thereby quantifying the severity of oil film exceeding the standard from the perspective of area coverage. Finally, based on the thickness statistical characteristics, the defect judgment label, and the defect area ratio, a detection result report is generated, and the detection result report is output as the online detection result of the residual oil film thickness, thereby integrating the quantitative information of the thickness of the entire surface and the defect judgment information into a structured, traceable, and verifiable online detection conclusion.

[0119] Through the synergistic effect of the above steps, the present invention achieves a complete closed loop from oil film thickness distribution map to process-usable detection conclusions, providing a quantitative decision-making basis for oil-scraping process quality assessment and closed-loop control.

[0120] like Figure 2 As shown, the present invention also provides an online detection system for the thickness of residual oil film on the surface of a fully automatic oil slinger, comprising: The measured reflectance acquisition module is used to acquire multispectral reflectance images, three-dimensional point cloud data and preset workpiece CAD standard templates of the workpiece inside the fully automatic oil slinger, and to acquire the measured reflectance of each pixel based on the multispectral reflectance images. The registration module is used to register the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. The incident reflectance acquisition module is used to obtain the local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and to obtain the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle. The distribution model establishment module is used to obtain the process setting parameters for this oil-scraping process and establish an oil film residue distribution model after oil-scraping based on the process setting parameters. The compensation reflectivity acquisition module is used to acquire the real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece, acquire the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and acquire the compensation reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. The oil film thickness preliminary value acquisition module is used to perform Lambert-Beer inversion on the compensated reflectivity using the residual distribution model as a regularization constraint term to obtain the preliminary value of the oil film thickness of each pixel. The distribution map generation module is used to obtain the final value of the oil film thickness of each pixel based on the registration confidence and the preliminary value of the oil film thickness, and to generate an oil film thickness distribution map based on the final value of the oil film thickness. The result output module is used to output the online detection result of residual oil film thickness based on the oil film thickness distribution map.

[0121] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent results or equivalent process transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for online detection of residual oil film thickness on the surface of a fully automatic oil slinger, applied to a fully automatic oil slinger, characterized in that, include: Acquire multispectral reflectance images, three-dimensional point cloud data, and preset workpiece CAD standard templates for the workpiece inside the fully automatic oil slinger, and obtain the measured reflectance of each pixel based on the multispectral reflectance images; The three-dimensional point cloud data is registered with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. Based on the normal vector deviation angle, the local roughness parameters of the workpiece surface are obtained from the three-dimensional point cloud data, and the equivalent normal incident reflectance of each pixel is obtained from the local roughness parameters of the workpiece surface and the normal vector deviation angle. Obtain the process setting parameters for this oil removal process, and establish an oil film residue distribution model after oil removal based on the process setting parameters; The real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece is obtained. The reversible scattering attenuation coefficient of the current optical path is obtained based on the real-time reflectivity signal. The compensation reflectivity is obtained based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. Using the residual distribution model as a regularization constraint, Lambert-Beer inversion is performed on the compensated reflectivity to obtain the preliminary value of the oil film thickness of each pixel. The final oil film thickness value of each pixel is obtained based on the registration confidence and the preliminary oil film thickness value, and an oil film thickness distribution map is generated based on the final oil film thickness value. The online detection results of residual oil film thickness are output based on the oil film thickness distribution map.

2. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The step of registering the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface includes: The three-dimensional point cloud data is iteratively registered with the CAD standard template to establish the spatial transformation relationship between each spatial point in the three-dimensional point cloud data and each corresponding point in the CAD standard template. Based on the spatial transformation relationship, the theoretical normal vector corresponding to each pixel is obtained from the CAD standard template; Based on the theoretical normal vector and the pre-calibrated sensor optical axis direction, the angle between the theoretical normal vector and the sensor optical axis direction of each pixel is calculated and used as the normal vector deviation angle; Obtain the registration residual of the iterative nearest point registration, and determine the registration confidence of each pixel based on the registration residual.

3. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The steps of obtaining local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and obtaining the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle, include: The optical constants of the metal substrate material under the acquisition band of the multispectral reflectance image are obtained, and the optical constants include refractive index and extinction coefficient; Based on the three-dimensional point cloud data, obtain the local root mean square height of the surface of the local area corresponding to each pixel; Based on the normal vector deviation angle, optical constant, and local root mean square height, the geometric correction factor corresponding to each pixel is calculated using the Fresnel-Gaussian mixed reflection model for rough surfaces. The equivalent normal incident reflectance is obtained based on the geometric correction factor and the measured reflectance.

4. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The steps of acquiring the real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece, acquiring the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and acquiring the compensated reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient include: The real-time reflectance signal of a fixed reference plate within the same camera field of view as the workpiece is obtained, and the reference reflectance of the fixed reference plate at each wavelength is obtained synchronously at the time of the multispectral reflectance image acquisition based on the real-time reflectance signal. Obtain the initial calibrated reflectivity of the fixed reference plate; The comprehensive attenuation coefficient is obtained based on the initial calibrated reflectance and the reference reflectance; Acquire the signal from the oil mist concentration sensor inside the oil slinger and the cumulative usage time of the fixed reference plate; Based on the comprehensive attenuation coefficient, the oil mist concentration sensor signal, and the cumulative usage time, the reversible scattering attenuation coefficient of the current optical path is obtained: The compensated reflectivity is obtained based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient.

5. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 4, characterized in that, The steps of obtaining the process setting parameters for this oil-scraping process and establishing an oil film residue distribution model after oil-scraping based on the process setting parameters include: The process setting parameters and oil parameters for this oil slinging process are obtained from the oil slinging machine control system. The process setting parameters include the rotational angular velocity and the oil slinging time. Based on the spatial position of the three-dimensional point cloud data after registration with the CAD standard template, the rotation radius of each pixel in the workpiece coordinate system is obtained. The oil density and dynamic viscosity are obtained based on the oil parameters. Based on the rotational angular velocity, oil ejection time, rotational radius, oil density, and oil dynamic viscosity described in the library, a theoretical residual distribution model of the oil film after oil ejection is established.

6. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 4, characterized in that, The step of using the residual distribution model as a regularization constraint to perform Lambert-Beer inversion on the equivalent normal incident reflectance to obtain the preliminary value of the oil film thickness for each pixel includes: Obtain the reference reflectance of the workpiece at the characteristic absorption wavelength and the preset Lambert-Beer inversion parameters; Based on the equivalent normal incident reflectance, reference reflectance, and Lambert-Beer inversion parameters of each pixel, construct the Lambert-Beer reflectance function: A regularization constraint term is constructed based on the oil film residue distribution model after oil spillage; Based on the Lambert-Beer reflectivity function and regularization constraints, construct the coupled inversion objective function: The initial value of the oil film thickness at each pixel is obtained by iteratively optimizing the coupled inversion objective function.

7. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The step of obtaining the final oil film thickness value for each pixel based on the registration confidence level and the preliminary oil film thickness value includes: The registration reliability of each pixel is determined based on the registration confidence level. For pixels whose registration confidence is greater than or equal to a preset confidence threshold, the initial value of the oil film thickness is directly used as the final value of the oil film thickness of that pixel. For a pixel whose registration confidence is less than the preset confidence threshold, neighborhood smoothing interpolation is performed based on the preliminary value of the oil film thickness of the neighboring pixels to obtain the neighborhood interpolation thickness of the pixel. The initial value of the oil film thickness and the neighborhood interpolation thickness are weighted and fused according to the registration confidence. The registration confidence is used as the weight of the initial value of the oil film thickness, and the weight of the registration confidence as the neighborhood interpolation thickness is subtracted to obtain the final value of the oil film thickness of the pixel.

8. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The step of generating an oil film thickness distribution map based on the final oil film thickness value includes: Obtain the final value of the oil film thickness and its corresponding pixel coordinates for each pixel; A spatial mapping relationship is established based on the pixel coordinates of each pixel, and the final oil film thickness value of each pixel is filled into the corresponding pixel position of the blank map according to the spatial mapping relationship to generate an initial thickness distribution map; Unfilled pixels in the initial thickness distribution map are obtained, and hole interpolation is performed to fill the holes based on the final oil film thickness values ​​of the neighboring pixels of the unfilled pixels to generate a complete thickness distribution map. Spatial smoothing filtering is applied to the complete thickness distribution map to eliminate thickness jumps between pixels, thereby generating the oil film thickness distribution map.

9. The method for online detection of residual oil film thickness on the surface of a fully automatic oil-slinging machine according to claim 1, characterized in that, The step of outputting the online detection result of residual oil film thickness based on the oil film thickness distribution map includes: Obtain the final value of the oil film thickness and its corresponding pixel coordinates for each pixel in the oil film thickness distribution map; Based on the final value of the oil film thickness of each pixel, the thickness statistical characteristics of the oil film thickness distribution map are calculated. The thickness statistical characteristics include the mean thickness, peak thickness, minimum thickness, and standard deviation of thickness. The final value of the oil film thickness of each pixel is compared with a preset thickness threshold pixel by pixel, and a defect judgment label is generated for each pixel based on the comparison result. The total number of defective pixels in the oil film thickness distribution map is counted based on the defect judgment label, and the defect area ratio is calculated based on the total number of defective pixels and the total number of pixels in the oil film thickness distribution map. A detection result report is generated based on the thickness statistical characteristics, the defect judgment label, and the defect area ratio, and the detection result report is output as the online detection result of the residual oil film thickness.

10. An online detection system for the thickness of residual oil film on the surface of a fully automatic oil slinger, characterized in that, include: The measured reflectance acquisition module is used to acquire multispectral reflectance images, three-dimensional point cloud data and preset workpiece CAD standard templates of the workpiece inside the fully automatic oil slinger, and to acquire the measured reflectance of each pixel based on the multispectral reflectance images. The registration module is used to register the three-dimensional point cloud data with a preset workpiece CAD standard template to obtain the normal vector deviation angle and registration confidence of each pixel on the workpiece surface. The incident reflectance acquisition module is used to obtain the local roughness parameters of the workpiece surface based on the normal vector deviation angle and the three-dimensional point cloud data, and to obtain the equivalent normal incident reflectance of each pixel based on the local roughness parameters of the workpiece surface and the normal vector deviation angle. The distribution model establishment module is used to obtain the process setting parameters for this oil-scraping process and establish an oil film residue distribution model after oil-scraping based on the process setting parameters. The compensation reflectivity acquisition module is used to acquire the real-time reflectivity signal of a fixed reference plate on the same optical path as the workpiece, acquire the reversible scattering attenuation coefficient of the current optical path based on the real-time reflectivity signal, and acquire the compensation reflectivity based on the equivalent normal incident reflectivity and the reversible scattering attenuation coefficient. The oil film thickness preliminary value acquisition module is used to perform Lambert-Beer inversion on the compensated reflectivity using the residual distribution model as a regularization constraint term to obtain the preliminary value of the oil film thickness of each pixel. The distribution map generation module is used to obtain the final value of the oil film thickness of each pixel based on the registration confidence and the preliminary value of the oil film thickness, and to generate an oil film thickness distribution map based on the final value of the oil film thickness. The result output module is used to output the online detection result of residual oil film thickness based on the oil film thickness distribution map.

Images