Vacuum pump blade assembly quality detection method based on laser sensor

Abstract

The application is a laser sensor-based vacuum pump blade assembly quality detection method, and relates to the field of precise optical measurement technology, comprising: filtering physical noise and channel registration on the original polarization image, constructing Stokes parameters and analyzing linear polarization degree, generating dynamic polarization suppression weight based on the linear polarization degree, and multiplying the dynamic polarization suppression weight with the total light intensity image.In the application, linear polarization structured light is actively projected, and original images of multiple polarization directions are synchronously captured by combining a micro-polarization array, thereby constructing a physical separation model of diffuse reflection effective signal and specular reflection interference signal.In specific, by using the linear polarization degree as an intrinsic parameter, a high polarization component caused by single specular reflection and a depolarization component caused by multiple scattering, absorption and re-radiation in the reflected light are identified;accordingly, the dynamic polarization suppression weight function can retain the diffuse reflection topography information while adaptively and smoothly compressing the gray scale of the high light area.

Description

Technical Field

[0001] This invention relates to the field of precision optical measurement technology, and in particular to a method for detecting the assembly quality of vacuum pump blades based on laser sensors. Background Technology

[0002] Vacuum pump blades are typically made of precision-machined metals such as aluminum alloy, stainless steel, or ductile iron, resulting in a specific micro-roughness on their surface. During assembly and inspection, specular reflection on the blade surface can cause localized overexposure of the laser displacement sensor receiver, while the effective diffuse reflection signal is overwhelmed by noise. This leads to missing 3D point cloud data or flying point noise, directly affecting the measurement accuracy of key geometric parameters such as assembly gaps and surface profiles.

[0003] In existing technologies, various technical approaches have been developed to suppress or eliminate specular reflection interference. For example, multi-angle light source fusion or high dynamic range imaging can be used to extend the receiver's dynamic range; surface-spraying developers can be used to convert specular reflection into diffuse reflection; and polarization imaging techniques can be used to reduce the effect of specular highlights to some extent. These methods have their own applicability and effectiveness in different application scenarios. For instance, spraying methods are not suitable for precision finished parts, while conventional polarization imaging often relies on mechanically rotating polarizers, which presents engineering implementation considerations such as time asynchrony or system complexity.

[0004] For workpieces such as vacuum pump blades that have complex torsional surfaces and high light reflection characteristics, a detection scheme is needed that can actively control and sense the polarization properties of light waves from the physical optics level.

[0005] Starting from the hardware architecture, this invention adopts a multi-polarization state image acquisition method that simultaneously captures linearly polarized laser projection and micro-polarization array. Combined with a polarization optics calculation model, it physically separates the effective diffuse reflection signal from the specular reflection interference signal, thereby achieving high-precision three-dimensional point cloud reconstruction and assembly quality judgment. It aims to provide a stable and effective technical approach for the online inspection of highly reflective metal parts. Summary of the Invention

[0006] The purpose of this invention is to propose a method for detecting the assembly quality of vacuum pump blades based on a laser sensor in order to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for detecting the assembly quality of vacuum pump blades based on laser sensors includes: Linearly polarized structured light is projected onto the surface of a vacuum pump blade using a polarized laser emission source, and the original polarization images of the vacuum pump blade surface in four different polarization directions are captured at the same exposure time using a micro-polarization array. Physical noise filtering and channel registration are performed on the original polarization image. Stokes parameters are constructed and the linear polarization degree is resolved. Dynamic polarization suppression weights are generated based on the linear polarization degree. The dynamic polarization suppression weights are multiplied with the total light intensity image to output the effective light intensity distribution image against reflection. The sub-pixel coordinates of the laser stripe center are extracted from the anti-reflective effective light intensity distribution image. The sub-pixel coordinates are mapped to three-dimensional physical coordinates by combining the camera intrinsic parameter matrix, external rotation matrix, and translation vector. The three-dimensional point cloud model is generated by stitching together the encoder physical displacement values. The three-dimensional point cloud model is spatially registered with the theoretical model, the rate of change of the normal vector angle of the local micro-surface element is calculated to perform local surface segmentation, and the three-dimensional boundary curve is extracted to fit the key assembly reference surface. A local assembly coordinate system is established based on the key assembly reference surface. The physical assembly clearance between the vacuum pump blade and the rotor slot is calculated. The rotor rotation motion is simulated by combining the spatial three-dimensional boundary curve to calculate the dynamic running clearance of the blade tip. The form and position tolerance of the spatial torsional aerodynamic surface is evaluated, and the assembly quality judgment result is output.

[0008] Preferably, the micro-polarization array uses four adjacent photosensitive pixels as a basic polarization calculation unit, and micro-polarization gratings with light transmission axis directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees are respectively etched above the four adjacent photosensitive pixels. The optical axis of the polarized laser emission source, the principal optical axis of the imaging objective lens, and the photosensitive target surface of the micro-polarization array satisfy the constraints of Scham's law, which makes the laser plane, the principal plane of the lens, and the photosensitive target surface intersect on a common straight line in space.

[0009] Preferably, the process of constructing the Stokes parameter and resolving the degree of linear polarization includes: Suppose the grayscale values ​​of the original polarization images of a point on the surface of a vacuum pump blade in four different polarization directions are respectively , , and Construct Stokes vectors ; Among them, total light intensity The difference between the components of linearly polarized light in the horizontal and vertical directions The component difference between linearly polarized light in the +45 degree direction and the 135 degree direction ; In order to meet the total light intensity Under the premise of linear polarization degree The calculation formula is: ; linear polarization angle The calculation formula is: .

[0010] Preferably, the generation of dynamic polarization suppression weights based on linear polarization degree includes: A polarization degree threshold is set. When the linear polarization degree is lower than the polarization degree threshold, the dynamic polarization suppression weight approaches one. When the degree of linear polarization exceeds the polarization degree threshold and increases, the dynamic polarization suppression weight decreases according to the nonlinear decay curve. The dynamic polarization suppression weights are multiplied pixel by pixel with the total light intensity image to suppress the specular reflection component and retain the diffuse reflection component, thereby generating an effective light intensity distribution image for anti-reflection.

[0011] Preferably, extracting the sub-pixel coordinates of the laser stripe center from the anti-reflective effective light intensity distribution image includes: Based on the background noise statistics of a local window, a dynamic grayscale threshold is set to lock the effective physical width range of the laser stripe. Within the effective physical width range, a sub-pixel interpolation algorithm based on grayscale centroid is used to multiply the physical coordinates of each pixel within the effective physical width range with its corresponding grayscale value, sum the results, and then divide by the sum of the grayscale values ​​of all pixels within the effective physical width range to obtain the sub-pixel coordinates of the laser stripe center.

[0012] Preferably, the step of mapping sub-pixel coordinates to three-dimensional spatial physical coordinates includes: Using camera intrinsic matrix In addition to radial and tangential distortion coefficients, the subpixel coordinates are restored to distortion-free physical projection coordinates; Combined with external rotation matrix Translation vector Construct the ray equation from the two-dimensional image plane to the three-dimensional camera local coordinate system; The ray equation and the laser plane equation are solved simultaneously to obtain the three-dimensional physical coordinates.

[0013] Preferably, the step of performing coarse spatial pose registration between the 3D point cloud model and the theoretical model includes: Calculate the physical centroid coordinates of the 3D point cloud model and make them coincide with the centroid of the theoretical model. Construct the covariance matrix of the point cloud coordinates, solve the eigenvalues ​​and eigenvectors of the covariance matrix to obtain three mutually perpendicular physical principal axis directions, and rotate and align the three mutually perpendicular physical principal axis directions with the preset reference coordinate axes of the theoretical model. The local surface segmentation process includes: Spatial normal vectors are obtained by fitting local small tangent planes, the rate of change of the spatial angle between the normal vectors of adjacent points is calculated, and a geometric topology algorithm based on region growing is used to split the 3D point cloud model into multiple feature point cloud subsets.

[0014] Preferably, the extraction of the three-dimensional boundary curve using a virtual rolling ball includes: Project the points within the feature point cloud subset onto the local fitting plane and set the radius of the virtual rolling ball. Virtual rolling ball radius The value of is strictly greater than the maximum value of the local average physical spacing of the point cloud; When the radius is When a virtual ball can pass through the gap between two points without touching other points, the two points are determined to be real physical boundary points. Spatial parametric spline curves are used to geometrically fit all discrete physical boundary points to generate spatial three-dimensional boundary curves.

[0015] Preferably, the step of simulating rotor rotational motion using spatial three-dimensional boundary curves to calculate blade tip dynamic operating clearance includes: Establish the theoretical spatial geometric equation of the inner wall of the stator pump cavity to determine the physical spatial position of the rotor rotation axis in the global coordinate system; The three-dimensional boundary curve of space is rotated around the rotor's rotation axis in small, predetermined angular steps. At each rotation angle position, calculate the shortest spatial normal distance from all discrete points on the three-dimensional boundary curve to the inner wall of the stator pump cavity, generate a continuous physical relationship curve between the dynamic operating clearance of the blade tip and the rotation angle, and find the global minimum distance value on the physical relationship curve.

[0016] Preferably, the form and position tolerances of the evaluated spatial torsional aerodynamic surface include: Spatial registration is performed between the point cloud corresponding to the spatial torsional aerodynamic surface and the theoretical surface model. Uniform sampling points are generated on the theoretical surface model, and the spatial normal vector of each uniform sampling point is calculated. Along the direction of the spatial normal vector, search for the nearest physical counterpart in the actual point cloud, calculate the directed spatial distance between the physical counterpart and the uniform sampling point, and construct a three-dimensional physical deviation field covering the entire torsional surface. The three-dimensional physical deviation field is compared with the preset three-dimensional profile tolerance envelope.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention constructs a physical separation model for the effective diffuse reflection signal and the interference signal from specular reflection by actively projecting linearly polarized structured light and simultaneously capturing original images with multiple polarization directions using a micro-polarization array. Specifically, by utilizing the intrinsic parameter of linear polarization, the high polarization component caused by a single specular reflection and the depolarization component caused by multiple scattering, absorption, and re-radiation are identified in the reflected light. Based on this, a dynamic polarization suppression weight function is established, which can adaptively and smoothly suppress the grayscale of the highlight region while preserving the diffuse reflection morphology information.

[0018] 2. This invention achieves surface segmentation by calculating the rate of change of the local normal vector of the point cloud, accurately extracts the spatial boundary curve of the blade using a virtual rolling ball algorithm, and reconstructs key assembly benchmarks such as the cylindrical or planar surface of the blade root with high precision through a robust fitting strategy. Based on these benchmarks, the system can automatically calculate the static assembly gap between the blade and the rotor slot, simulate the rotor rotation process to evaluate the dynamic running gap of the blade tip, and generate a three-dimensional deviation field covering the complex torsional surface. Attached Figure Description

[0019] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0021] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0022] Example 1 Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0023] Appendix Figure 1 The flowchart of the vacuum pump blade assembly quality inspection method based on laser sensor provided in the embodiments of the present invention shows the complete steps from projecting linearly polarized structured light onto the surface of the vacuum pump blade using a polarized laser emission source to evaluating the form and position tolerances of the spatial torsional aerodynamic surface.

[0024] In this embodiment, it includes: Step 1: Polarization Modulated Laser Scanning and Acquisition: Polarized structured light is projected onto the surface of the vacuum pump blade using a polarized laser emission source, and the original polarization images of the vacuum pump blade surface in four different polarization directions are captured synchronously at the same exposure time using a micro-polarization array. The core objective of this step is to build a laser scanning acquisition system that can actively control and sense the polarization properties of light waves, starting from the underlying hardware architecture of physical optics.

[0025] Vacuum pump blades are typically made of metal materials such as aluminum alloy, stainless steel or ductile iron through precision machining. Their surfaces have specific micro-roughness, which can produce strong specular reflection (i.e., high gloss or reflection) under the lighting conditions of the assembly site and when exposed to high-intensity laser irradiation.

[0026] This strong specular reflection can cause localized overexposure in traditional laser displacement sensor receivers, resulting in the diffuse reflection signal in that area being overwhelmed by noise. This ultimately manifests as missing data or severe flying point noise in the 3D point cloud. To address this issue at its physical source, this step employs polarization modulation technology to perform multi-dimensional physical constraints and acquisition of the incident and reflected light.

[0027] First, at the hardware transmitter, a laser transmitter with a specific wavelength and high polarization is configured. Considering the differences in the absorption and reflectivity of metal surfaces for different wavelengths of light, this scheme selects a short-wavelength blue-violet semiconductor laser (e.g., with a center wavelength of 405 nm or 450 nm).

[0028] Short-wavelength lasers penetrate metal surfaces very shallowly, allowing for a more accurate reflection of the physical contours of blade surfaces. Furthermore, their high single-photon energy helps improve the signal-to-noise ratio in complex workshop environments. To ensure wavelength stability and power consistency during long-term operation, a precision temperature control circuit based on a semiconductor thermoelectric cooler (TEC) is integrated within the laser source to maintain the laser diode's operating temperature within a predetermined physical range. At the laser's front end, a collimating lens group, a polarizer, and a Powell prism are sequentially arranged.

[0029] The collimating lens group is used to compress the diverging beam emitted by the semiconductor laser into a parallel beam. The polarizer uses a high extinction ratio nanowire grating polarizer, whose transmission axis is precisely mechanically adjusted so that the transmitted laser light is converted into linearly polarized light with a single vibration direction. This linearly polarized light then enters the Powell prism.

[0030] Unlike traditional cylindrical lenses, the aspherical top curve of the Powell prism introduces nonlinear spherical aberration into a Gaussian-distributed laser beam, thus flattening the beam in space and forming a linearly structured beam with highly uniform energy distribution. This uniformly polarized structured beam, projected onto the surface of the vacuum pump blades, provides a stable initial physical field for subsequent polarization state analysis.

[0031] At the receiving end, to avoid the mechanical vibration interference, time asynchrony error, and large system size caused by traditional mechanically rotating polarizers, this solution employs a polarization image sensor based on a focal plane micro-polarization array. This sensor integrates a nanoscale metal wire grid polarization array between the microlens and photodiode of a conventional complementary metal-oxide-semiconductor (CMOS) photosensitive pixel. Specifically, this micro-polarization array uses four adjacent photosensitive pixels (i.e., a 2×2 pixel matrix) as a basic polarization calculation unit.

[0032] Above these four pixels, miniature polarization gratings with transmission axis directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees are etched, respectively. When the scattered light carrying complex polarization information reflected from the surface of the vacuum pump blades is converged by the imaging objective lens, it will pass through these four micro-polarization gratings in different directions simultaneously and be converted into electrical signals by the underlying photodiode.

[0033] This physical architecture enables the system to simultaneously capture the light intensity distribution in four different polarization directions on the same physical region of the vacuum pump blade surface at the same exposure time and from the same spatial perspective, fundamentally eliminating time phase errors caused by blade movement on the assembly line or low-frequency environmental vibrations. To improve photon collection efficiency, the curvature of the microlens array is specially designed to ensure that the converged light rays can pass perpendicularly through the nanowire grating, reducing optical crosstalk.

[0034] To ensure clear imaging of the laser line within the large depth of field of the vacuum pump blades, the optical axis of the transmitter, the principal optical axis of the imaging objective lens at the receiver, and the photosensitive target surface of the polarization image sensor must strictly satisfy Scherm's law. Through a precise mechanical fine-tuning mechanism, the tilt angle of the image sensor is adjusted so that the three physical planes—the laser plane, the lens principal plane, and the image sensor target surface—intersect in space on a common straight line.

[0035] This physical constraint ensures that, regardless of whether the laser beam is shone on the tip, root, or complex torsional surface of the blade, the reflected light can form a clear focused image on the sensor target surface, avoiding the interference of defocus blur on the polarized light intensity signal.

[0036] After the optical path is fixed, rigorous spatial coordinate system calibration is required. Using a diffuse reflection calibration plate with a high-precision circular array or checkerboard pattern (the surface of the calibration plate must be treated with a special matte coating to avoid glare interference during the calibration process), multi-pose translation and rotation are performed in the measurement space.

[0037] By extracting the pixel coordinates of the feature points of the calibration plate in the four polarization channel images, and combining them with the known physical dimensions of the calibration plate, the camera's internal parameters (including focal length, principal point coordinates, radial distortion and tangential distortion coefficients) and external parameters (rotation matrix and translation vector between the camera coordinate system and the world coordinate system) are calculated. At the same time, the three-dimensional spatial equation of the laser plane in the camera coordinate system is calibrated.

[0038] During the actual assembly and testing of vacuum pump blades, in order to suppress interference from ambient light (such as LED lighting and natural light from windows), the system employs a hardware-level timing synchronization triggering mechanism. The main control circuit board generates a high-frequency pulse width modulation (PWM) signal, which is divided into two paths: A constant current source directly drives the laser, controlling the laser to emit high-intensity linearly polarized light with extremely short pulse times (e.g., on the order of microseconds); The other path, after a fixed delay in the microsecond range, triggers the global shutter of the polarization image sensor for exposure.

[0039] The exposure time is strictly limited to the duration window of the laser pulse. Because the instantaneous power of the laser is much greater than the ambient background light power within the extremely short exposure time, this physical time gating mechanism can significantly improve the signal-to-noise ratio of the acquired polarization image.

[0040] After an exposure, the sensor’s internal analog-to-digital converter (ADC) quickly converts the analog electrical signals of the four polarization channels into a high bit depth (e.g., 12-bit or 16-bit) digital grayscale image, and transmits the raw data stream to the host computer’s memory in real time via Gigabit Ethernet or CoaXPress interface, completing the physical process of polarization-modulated laser scanning and acquisition.

[0041] Step 2: Calculation and separation of multi-polarization state light intensity signals: Physical noise filtering and channel registration are performed on the original polarization image, Stokes parameters are constructed and the linear polarization degree is analyzed, dynamic polarization suppression weights are generated based on the linear polarization degree, and the dynamic polarization suppression weights are multiplied with the total light intensity image to output the effective light intensity distribution image against reflection. The core logic of this step lies in using the original light intensity images of the same spatial location but four different polarization directions obtained in step one, and through the calculation rules of physical polarization optics, quantitatively separating the "diffuse reflection component carrying true morphological information" and the "spectral reflection (highlight) component causing sensor overexposure" mixed in the reflected light. This is the key signal processing step for achieving high-precision anti-reflective measurement.

[0042] Before performing polarization calculations, the original image transmitted to the host computer must undergo low-level physical correction.

[0043] First, dark current noise is subtracted: with the laser off and the environment completely darkened, multiple frames of background images from the sensor are acquired and averaged to obtain a dark field reference map. Considering that dark current drifts with changes in sensor operating temperature, the system dynamically compensates for the dark field reference map using data from the sensor's built-in temperature probe. During actual detection, this compensated dark field reference map is subtracted pixel-by-pixel from each frame of the original image to eliminate the effects of sensor thermal noise and fixed-mode noise. Second, instantaneous field-of-view error correction is performed on the micro-polarization array.

[0044] Since the four micro-polarizers at 0°, 45°, 90°, and 135° correspond to four different pixels in a 2×2 matrix in physical space, the microscopic regions on the blade surface observed by them exhibit sub-pixel-level spatial misalignment. Direct calculations would produce false polarization edge effects at high-frequency details such as blade edges or surface scratches.

[0045] Therefore, a spatial resampling algorithm based on bilinear interpolation or bicubic interpolation is adopted, and spatial phase compensation is performed on the gray values ​​of the four channels with the geometric center of the 2×2 matrix as the reference, so that the resampled images of the four polarization channels achieve absolute pixel-level alignment in physical space.

[0046] In physics, the polarization state of light waves can be fully macroscopically described using Stokes parameters. After spatial registration, let the gray values ​​of the four polarization channels corresponding to a point on the blade surface in the image be... , , and .

[0047] Based on the light intensity measurements in these four directions, the Stokes vector of this spatial point is constructed. Among them, parameters This represents the total light intensity received at that point, and its physical calculation method is as follows: .

[0048] Parameter This represents the component difference between linearly polarized light in the horizontal direction (0 degrees) and the vertical direction (90 degrees), calculated as follows: Parameters This represents the component difference between linearly polarized light in the +45° direction and the -45° (i.e., 135°) direction, calculated as follows: .

[0049] Because this system uses linearly polarized laser illumination, and the reflection process on the metal surface produces very little circular polarization component, the circular polarization parameter... In this physical model, these parameters are negligible. Based on the Stokes parameters mentioned above, two key physical characteristic quantities are further analyzed: linear polarization degree. and linear polarization angle .

[0050] To ensure the rigor of the physical calculations, the system first performs a physical premise determination before analyzing the characteristic quantities: Total light intensity must be met This is to eliminate the extreme physical state where the sensor is in absolute darkness and no effective photons fall into the pixel, thus preventing mathematical errors such as a denominator of zero in subsequent calculations.

[0051] In satisfying Under the premise of linear polarization degree The physical meaning of is the proportion of linearly polarized light intensity in the total light intensity, and its calculation formula is clearly defined as: ; linear polarization angle The physical meaning of is the azimuth angle of the principal axis of the polarization ellipse, and its calculation formula is clearly defined as: ; These two characteristic quantities provide a solid physical basis for the subsequent separation of reflection components.

[0052] The reflected light from the surface of a vacuum pump blade is a superposition of diffuse reflection and specular reflection. From a physical optics perspective, when an incident linearly polarized laser irradiates the blade surface, the diffusely reflected light undergoes multiple scattering and internal absorption and re-radiation, severely disrupting its polarization state. This manifests as depolarization, i.e., the degree of linear polarization of the diffusely reflected light (…). Extremely low, close to natural light.

[0053] Conversely, specularly reflected light that undergoes a single reflection directly on the metal surface (i.e., light that causes overexposure of highlights) largely retains the polarization characteristics of the incident light, and its degree of linear polarization ( () is relatively high.

[0054] Based on this physical difference, this step establishes a dynamic separation model based on polarization degree constraints. For each pixel in the image, the system evaluates its polarization degree in real time. Value. In an ideal diffuse reflection region with uniform surface roughness on the blade, The value is in a lower baseline range; while in areas of strong specular reflection caused by blade machining marks, chamfers, or specific curvature areas, The value will increase significantly.

[0055] By setting a pre-calibrated polarization threshold based on the physical properties of the blade material, the system can accurately identify which pixels in the image are affected by specular interference. This threshold is not fixed but dynamically and adaptively adjusted based on surface roughness inspection reports of different batches of blades at the assembly site.

[0056] To extract the pure diffuse reflection signal, the system does not use a simple threshold truncation method, but instead constructs a continuous dynamic polarization suppression weight function. This weight function is based on the polarization suppression of each pixel. The value is the independent variable. When When the value is below the set safety threshold, it indicates that the area is dominated by diffuse reflection, the weighting coefficient approaches 1, and its original total light intensity is preserved. ;when When the value exceeds the safety threshold and gradually increases, it indicates that the specular reflection component is gradually becoming dominant, and the weighting coefficient decreases rapidly according to a non-linear decay curve (such as Gaussian decay or inverse proportional decay function). This smooth transition weighting function design can effectively avoid physical abrupt changes or artifacts in grayscale values ​​in the reconstructed image.

[0057] The calculated dynamic polarization suppression weights are then pixel-by-pixel compared with the total light intensity image. Perform a multiplication operation. In this physical reconstruction process, the highlight areas that were originally saturated in grayscale (e.g., reaching 255 or 4095) due to specular reflection are now... The value is extremely high, and it is assigned a very small weighting coefficient, thus its grayscale value is greatly suppressed; while the weak diffuse reflection signal, which was originally covered by the highlight or located at the edge of the highlight, is due to its The value was low, so it was preserved intact.

[0058] After this reconstruction process, the system outputs an "effective light intensity distribution image for anti-reflection," which eliminates the interference of specular highlights on the metal surface and retains only the true diffuse reflection morphology. The energy distribution of the laser stripes in this image is more in line with the ideal Gaussian distribution, laying a high-quality data foundation for subsequent high-precision 3D point cloud computing.

[0059] Step 3: Reconstruction of point cloud on anti-reflective blade surface: Extract the sub-pixel coordinates of the laser stripe center from the effective light intensity distribution image of anti-reflective blade, and map the sub-pixel coordinates into three-dimensional physical coordinates by combining the camera intrinsic matrix, external rotation matrix and translation vector, and then combine the encoder physical displacement values ​​to generate a three-dimensional point cloud model. The core objective of this step is to convert the "anti-reflective effective light intensity distribution image" obtained in step two, which has been separated by polarization calculations and has eliminated the interference of overexposure of the metal surface highlights, into a high-precision three-dimensional spatial physical coordinate set (i.e., three-dimensional point cloud) of the vacuum pump blade surface through a rigorous physical optical mapping model and mechanical kinematic model.

[0060] Since the preceding steps have purified the light intensity signal from the physical source, the geometric reconstruction in this step can achieve extremely high spatial resolution and physical realism.

[0061] In the two-dimensional grayscale image reconstructed with polarization suppression weights, the cross-sectional energy distribution of the laser stripes exhibits an ideal Gaussian distribution that highly conforms to physical laws. In traditional unprocessed images, overexposure caused by metallic mirror reflection flattens the peaks of the Gaussian distribution (i.e., grayscale saturation truncation), resulting in severe physical shifts in center extraction. However, the effective light intensity image in this scheme retains the complete peak shape.

[0062] To extract the precise physical center of the laser stripes, the system scans pixel grayscale values ​​row by row along the column direction of the image (or the normal direction perpendicular to the laser stripe direction). First, instead of using a globally fixed grayscale threshold, the system sets a dynamic grayscale threshold based on background noise statistics within a local window. This dynamic threshold adapts to the reflectivity variations in different areas of the blade surface, thereby accurately eliminating background diffuse scattering noise and locking in the effective physical width range of the laser stripes.

[0063] Subsequently, a sub-pixel interpolation algorithm based on grayscale centroid is employed within this interval. Specifically, the physical coordinates of each pixel within the interval are multiplied by its corresponding grayscale value (i.e., light intensity energy), summed, and then divided by the sum of all pixel grayscale values ​​within the interval. This physical calculation process is similar to solving for the centroid of an object's mass distribution, enabling the center coordinates of the laser stripe to be accurate to the sub-pixel level (e.g., 0.01 pixel width). This sub-pixel level extraction accuracy directly determines the micrometer-level measurement resolution of subsequent three-dimensional spatial coordinates.

[0064] After obtaining the sub-pixel coordinates of the laser stripes on the target surface of the two-dimensional image sensor, it is necessary to map them onto the real three-dimensional physical space using the principle of optical triangulation.

[0065] This process relies on the Schamer's Law constraint model and spatial calibration parameters established in step one. From a geometric optics perspective, each sub-pixel center point on the image sensor corresponds to a unique ray of light in space. This ray originates from the sensor target surface, passes through the physical optical center of the camera lens, and extends into three-dimensional space.

[0066] Simultaneously, the laser emitter projects a physical light plane with a defined spatial equation. According to the principles of spatial analytical geometry, the spatial intersection of this ray and the laser plane is the actual physical point on the surface of the vacuum pump blades where the laser light is reflected.

[0067] To provide a mathematical and physical foundation for this geometric mapping process, the system introduces the core parameter matrix obtained from the calibration in step one: Define the camera intrinsic parameter matrix as (Including focal length and principal point coordinates), define the external rotation matrix between the camera coordinate system and the world coordinate system as follows: Define the translation vector as The system first utilizes the intrinsic parameter matrix. The radial and tangential distortion coefficients obtained from the calibration are used to eliminate the physical distortion caused by the lens group and restore the extracted image coordinates to ideal distortion-free physical projection coordinates.

[0068] Combined with external rotation matrix Translation vector A ray equation is constructed from the 2D image plane to the 3D camera local coordinate system. This ray equation is then solved simultaneously with the laser plane equation, which is also expressed in the camera local coordinate system. By solving this system of linear equations, the precise 3D spatial coordinates of the intersection point in the camera local coordinate system can be obtained. .

[0069] Thus, the two-dimensional laser stripes in a single frame image have been completely converted into a three-dimensional physical contour line that fits the surface contour of the blade.

[0070] A single frame image can only capture the three-dimensional contour of a vacuum pump blade on a specific cross-section. To obtain the three-dimensional morphology of the entire blade, relative mechanical motion must be introduced. In assembly and inspection, laser sensors are typically mounted on high-precision linear servo slides, or the vacuum pump blades are fixed to a precision rotating indexing plate. Taking linear scanning as an example, the servo slide integrates a high-resolution optical scale or rotary encoder. As the slide drives the sensor to move at a constant speed along a set physical axis, the encoder outputs physical displacement pulse signals in real time. Upon receiving these pulse signals, the system hardware triggers laser emission and polarization image acquisition at preset physical intervals (e.g., every 0.05 mm). At the data processing end, the calculated three-dimensional contour of each frame lies in the sensor's current local coordinate system. The system reads the encoder's physical displacement value, which is strictly synchronized with the image frame, and combines it with a pre-calibrated straightness error compensation table for the slide guide to construct a high-precision spatial translation vector that dynamically changes over time.

[0071] Using a homogeneous coordinate transformation matrix, the 3D coordinate points in the local coordinate system are uniformly transformed into a fixed global physical coordinate system (i.e., the world coordinate system). As the scan continues, tens of thousands of 3D contour lines are closely arranged in the global coordinate system according to the actual physical spacing, and finally stitched together to form a high-density 3D point cloud model covering the entire surface of the vacuum pump blades.

[0072] Although polarization modulation technology has eliminated most of the flying point noise caused by specular highlights, in actual industrial workshop environments, tiny dust particles suspended in the air, tiny chips that have not been cleaned off the blade surface, or a very few multiple diffuse reflection dead angles may still generate a small amount of free noise in three-dimensional space.

[0073] These noise points physically appear as isolated coordinate points offset from the surface of the blade body. To ensure the accuracy of assembly feature extraction, the system performs statistical filtering on the stitched global point cloud based on spatial physical distance.

[0074] For any spatial point in the point cloud, the system sets a search radius with a clear physical meaning (e.g., 0.5 mm, slightly larger than the physical row spacing of the scan) with that point as the center. It then counts the number of neighboring points falling within this spherical space and calculates the average physical distance from that point to all its neighbors. On a normal continuous surface of a blade, the physical distribution between points is highly dense, resulting in a small average distance; however, for dust noise points in the air, their surroundings are often lacking, leading to a significantly larger average distance. The system calculates the global expected value and standard deviation of the average distance for all points in the entire point cloud.

[0075] When the local average distance of a point exceeds the global expected value plus a certain multiple of the standard deviation, the system physically and logically determines that the point is a spatially isolated noise point detached from the blade surface and removes it from the point cloud set. After this physical morphological denoising process, the final output 3D point cloud model has a smooth surface and compact structure, highly restoring the true physical processing morphology of the vacuum pump blade.

[0076] Step 4: Adaptive extraction of assembly feature regions: Perform coarse spatial orientation registration between the 3D point cloud model and the theoretical model, calculate the rate of change of the normal vector angle of local micro-surface elements to perform local surface segmentation, and use a virtual rolling ball to extract the spatial 3D boundary curves to fit the key assembly reference surface. The core objective of this step is to automatically identify and separate key geometric regions (such as blade root assembly surfaces, blade tip clearance boundaries, complex torsional aerodynamic surfaces, etc.) that are closely related to the assembly quality of the vacuum pump from the massive unordered point cloud containing millions of coordinate points obtained in step three, using spatial geometric topology and surface physical features.

[0077] This step is the key geometric bridge connecting "full surface physical scanning" and "local assembly dimension measurement".

[0078] The point cloud of the blade acquired through scanning resides in the global physical coordinate system of the measurement system. Its spatial attitude (position and orientation) is random, depending on the initial physical state of the blade placed on the scanning stage. To accurately extract specific regions, the coordinate system of this point cloud must first be spatially and physically aligned with the coordinate system of the theoretical computer-aided design (CAD) model of the vacuum pump blade. The system first calculates the physical centroid coordinates of the actual scanned point cloud, i.e., the arithmetic mean of all points in the X, Y, and Z directions, and then translates the point cloud as a whole so that its centroid coincides with the centroid of the CAD model in space.

[0079] Subsequently, the system analyzes the physical mass distribution characteristics of the point cloud in three-dimensional space and constructs the covariance matrix of the point cloud coordinates. By solving the eigenvalues ​​and eigenvectors of this covariance matrix, the three mutually perpendicular physical principal axes (usually corresponding to the length, width, and thickness directions of the blade) of the point cloud in space can be obtained, representing the most dispersed, second most dispersed, and most concentrated distribution of the point cloud in space.

[0080] The three physical principal axes are rotated and aligned with the three preset reference coordinate axes of the CAD model. This coarse registration process based on physical mass distribution can quickly adjust the overall spatial pose of the actual point cloud to a state that is basically consistent with the theoretical model without relying on any manual intervention or preset marker points, thus establishing a correct spatial orientation reference for subsequent local fine feature extraction.

[0081] Vacuum pump blades typically contain complex spatial torsional surfaces (for gas compression or suction) and regular machined surfaces (for mating with rotor slots).

[0082] In order to assess the assembly quality of these different functional areas separately, the point cloud needs to be segmented according to the physical geometry of the surface.

[0083] The system iterates through each point in the point cloud and uses the set of nearest neighbors within a certain physical radius to fit a locally small tangent plane using the least squares method. The normal direction of this tangent plane is the spatial normal vector of the center point. The normal vector represents the surface orientation of this small physical region.

[0084] The system calculates the rate of change of the spatial angle between the normal vectors of adjacent points, which is physically equivalent to the local curvature of a surface. On the gently twisted surface of a blade, the angle between the normal vectors of adjacent points changes very little, and the curvature tends to be continuous and smooth. However, at the blade edge, chamfer, or the boundary between different processed surfaces, the normal vectors undergo drastic spatial deflection, and the curvature exhibits a physical abrupt change. The system employs a geometric topology algorithm based on region growing, starting from a seed point with minimal curvature and continuously expanding outwards. If the angle between the normal vectors and the curvature difference between adjacent points are both less than a set physical threshold, they are considered to belong to the same continuous physical processed surface and are grouped into the same region set; growth stops when the expansion encounters a boundary with a curvature abrupt change. Through this physical segmentation based on underlying geometric topology, the system can adaptively divide the entire blade point cloud into multiple feature point cloud subsets with independent physical meaning, such as the blade back surface, blade base surface, blade root cylindrical surface, and blade tip surface.

[0085] In the actual assembly of vacuum pumps, the physical clearance between the blade edge and the inner wall of the pump chamber (i.e., tip clearance and side clearance) is a core indicator determining the pumping efficiency and ultimate vacuum. Therefore, accurately extracting the physical boundary lines of the blade point cloud is crucial. Based on the surface segmentation results from step four, the system performs boundary tracking for specific feature point cloud subsets (e.g., blade tip surfaces).

[0086] In physical geometry, a point inside a surface has neighboring points evenly distributed in a 360-degree direction; however, a point on the physical boundary of a surface has neighboring points only distributed on one side of the boundary, with a clear physical spatial discontinuity on the other side. The system projects points within a feature subset onto its local fitting plane and uses convex or concave hull algorithms from computational geometry (such as the physical implementation logic of the AlphaShape algorithm) to create a virtual physical rolling ball that rolls over the projected point set. To ensure the rigor of boundary extraction, the system assigns a specific sign to the radius of this rolling ball. .

[0087] about The system sets the value retrieval logic as follows: The value of must be strictly greater than the maximum value of the local average physical spacing of the point cloud. This physical premise is crucial; it effectively prevents the virtual rolling sphere from leaking out of the gaps within the normal continuous surface, thus avoiding misclassifying normal points within the surface as boundary points. When the virtual physical rolling sphere (radius ) When a point can pass through the gap between two points without touching other points, those two points are considered true physical boundary points. After extracting all discrete boundary points, the system uses spatially parameterized spline curves (such as B-spline curves) to geometrically fit these discrete points.

[0088] During the fitting process, a physical smoothness constraint is introduced to eliminate the serrated edges caused by minor machining burrs, ultimately generating a continuous, smooth, and strictly conforming spatial three-dimensional boundary curve that fits the actual physical edge of the blade. This curve will serve as the direct geometric reference for subsequent calculations of assembly clearance.

[0089] When vacuum pump blades are assembled into rotor slots, the blade root portion typically serves as the primary mechanical positioning reference. The physical shape of the blade root may be a precise plane, a cylindrical surface, or a specific wedge-shaped surface. To assess positional deviations during assembly, the ideal geometric model of these reference surfaces must be reconstructed with high precision from the point cloud. Taking the cylindrical surface of the blade root as an example, the system extracts the point cloud subset segmented into the blade root region in step four.

[0090] Due to the unavoidable presence of tool marks, micro-scratches, or material deformation during actual processing, these point clouds are not situated on an absolutely ideal mathematical cylindrical surface. The system employs a robust geometric fitting strategy (such as iterative least squares based on physical distance weighting). First, preliminary cylindrical surface geometric fitting is performed using all leaf root point clouds to obtain the initial spatial equation of the cylindrical axis and the physical radius.

[0091] Then, the vertical physical distance from each actual point cloud to the initial cylindrical surface is calculated. If the distance deviation of a point exceeds the preset machining tolerance zone (e.g., determined to be a scratch, pit, or burr), the system will significantly reduce the weight of that point in the next iteration of fitting, or even physically remove it. After multiple iterations, the fitting process gradually converges, and the final determined cylindrical axis and radius can represent the true contact state of the leaf root in macroscopic physical assembly to the greatest extent, without being affected by microscopic surface defects. For the planar reference, a robust planar fitting algorithm that removes physical outliers is also used to obtain high-precision reference surface normal vectors and spatial intercepts. These high-precision fitted geometric primitives provide a solid physical coordinate reference for the final assembly quality judgment.

[0092] Step 5: Assembly quality geometric deviation judgment: Establish a local assembly coordinate system based on the key assembly reference surface, calculate the physical assembly clearance between the vacuum pump blade and the rotor slot, combine the three-dimensional boundary curve of space to simulate the rotor rotation motion to calculate the dynamic running clearance of the blade tip, evaluate the form and position tolerance of the spatial torsional aerodynamic surface, and output the assembly quality judgment result. The core objective of this step is to perform in-depth spatial geometric comparison and physical quantitative calculation between the high-precision assembly feature regions (including physical boundary lines, datum planes, and local feature point clouds) adaptively extracted in step four and the theoretical mechanical design tolerance model of the vacuum pump. By constructing a multi-dimensional geometric deviation evaluation system, the physical state of the blades in the actual assembly environment is accurately reflected, thereby providing a decisive basis for judging the overall aerodynamic performance and mechanical reliability of the vacuum pump.

[0093] In the mechanical structure of rotary vane or slide valve vacuum pumps, the vanes need to be inserted into the radial or chordal slots of the rotor and slide back and forth along the slot walls at high frequency while rotating at high speed in the pump chamber. The physical assembly clearance between the vane root and both sides and the rotor slots is a key mechanical indicator that determines whether the vanes can slide smoothly without jamming or serious leakage.

[0094] To accurately quantify this assembly gap, the system first establishes a local "blade assembly coordinate system" in three-dimensional space based on the blade root reference plane (such as the blade root central symmetry plane or the positioning cylindrical surface) obtained through high-precision fitting in step four. The establishment of this coordinate system strictly adheres to the principle of datum coincidence in mechanical drawing, ensuring that subsequent geometric calculations have a clear physical reference frame. Subsequently, the system imports the theoretical computer-aided design (CAD) model of the rotor slots and extracts the equations of the two parallel sidewall planes of the rotor slots.

[0095] The feature point clouds of the two sides of the blade in the local assembly coordinate system are spatially projected along the physical normal direction perpendicular to the rotor slot sidewall. For each actual physical coordinate point on the blade side, the vertical spatial distance from the projection ray to the theoretical sidewall plane of the rotor slot is calculated.

[0096] Because blades exhibit microscopic waviness or macroscopic bending deformation during actual manufacturing, these projected distances are not fixed values ​​but rather a distribution set. The system performs statistical analysis on this distance set, extracting the maximum physical clearance value, the minimum physical clearance value, and the average physical clearance value.

[0097] The minimum physical clearance value determines whether the blades will interfere with the groove wall or even seize up under conditions of long-term operation of the vacuum pump and thermal expansion of the material due to increased temperature; the maximum physical clearance value determines the sealing effect of the lubricating oil film and the leakage of gas backflow. The system strictly compares the calculated actual physical clearance range with the preset mechanical assembly tolerance zone (e.g., nominal clearance of 0.03 mm to 0.05 mm).

[0098] If the calculated minimum clearance is lower than the lower limit of the tolerance zone, or the maximum clearance is higher than the upper limit of the tolerance zone, the system will determine that the root assembly quality of the blade is unqualified in terms of physical logic, and record the specific out-of-tolerance space location for subsequent rework and repair.

[0099] In actual operation, the rotor of a vacuum pump is usually eccentrically mounted inside the stator (pump chamber). Under the action of centrifugal force or bottom spring force, the blade tip faces slide against the inner wall of the pump chamber, thus dividing the pump chamber into an intake chamber and an exhaust chamber. The dynamic operating clearance between the blade tip and the inner wall of the pump chamber directly determines the ultimate vacuum degree and pumping speed of the vacuum pump.

[0100] Since the clearance during high-speed rotation cannot be directly measured on a static assembly testing table, this system adopts a dynamic simulation calculation method based on spatial kinematics. First, based on the complete assembly drawings of the vacuum pump, the system establishes the theoretical spatial geometric equation of the inner wall of the stator pump cavity (usually a standard inner cylindrical surface equation or a specific profile equation), and determines the physical spatial position of the rotor rotation axis in the global coordinate system and the eccentricity parameter.

[0101] The system extracts the continuous and smooth three-dimensional boundary curve of the blade tip surface generated in step four. In a virtual physical space, the system simulates the rotational motion of the rotor, causing the blade tip boundary curve to undergo spatial rotation transformation around the rotor's rotation axis in small angular steps (e.g., 0.1 degrees per step). At each rotation angle position, the system calculates the shortest spatial normal distance from all discrete points on the blade tip boundary curve to the theoretical cylindrical surface of the stator pump cavity.

[0102] By traversing the complete rotation cycle from 0 to 360 degrees, the system can generate a continuous physical relationship curve of "dynamic operating gap at the blade tip - rotation angle".

[0103] The system searches for the global minimum distance value on this curve. If the global minimum distance value is less than the limit physical thickness required to maintain a safe oil film, it indicates that the blade tip is very likely to have a rigid collision with the pump chamber during actual operation, resulting in rubbing damage and seriously affecting the equipment life. If the distance value is too large, it indicates that the blade tip is severely worn or the machining size is too small, which will lead to serious internal gas leakage and failure to reach the rated vacuum level.

[0104] The system compares the extreme values ​​calculated by this dynamic simulation with the design specifications, thereby enabling a predictive judgment of the blade tip assembly quality.

[0105] For vacuum pump blades with complex spatial torsion angles (such as certain types of Roots pump blades or helical surface blades), the three-dimensional morphology of their surfaces directly affects the gas compressibility and flow field distribution. Traditional two-dimensional cross-sectional projection detection methods can only obtain the contour lines of discrete cross-sections and cannot truly reflect the spatial torsion deviations of such complex surfaces.

[0106] This system employs a geometric tolerance evaluation logic based on three-dimensional spatial normal projection. The system performs high-precision spatial registration between the point clouds of the blade underside and blade base twisted surfaces segmented in step four and the theoretical CAD surface model. After registration, the system generates high-density uniform sampling points on the theoretical CAD surface and calculates the precise spatial normal vector of each sampling point on the theoretical surface. Subsequently, the system searches for the nearest physical counterpart in the actual blade point cloud along the directions of these theoretical normal vectors.

[0107] Calculate the directed spatial distance between the actual point and the theoretical sampling point. If the actual point is located outside the theoretical surface (i.e., material is redundant), the distance is recorded as a positive value; if the actual point is located inside the theoretical surface (i.e., material is missing or excessively cut), the distance is recorded as a negative value.

[0108] By traversing all theoretical sampling points, the system constructs a three-dimensional physical deviation field covering the entire torsional surface. The system not only calculates the overall maximum positive deviation, maximum negative deviation, and root mean square error, but also further analyzes the spatial gradient variation of the deviation. If the deviation gradient in a certain local area is too large, it indicates the presence of severe local distortion or machining depressions. In fluid dynamics, this can lead to airflow separation and local eddies, significantly reducing pump efficiency and increasing operating noise.

[0109] The system compares the calculated three-dimensional deviation field with the preset three-dimensional profile tolerance envelope. Any physical point that exceeds the envelope is judged to be non-compliant in terms of form and position tolerance.

[0110] After completing the geometric deviation calculations for each of the aforementioned local features, the system enters the final comprehensive judgment and data output stage.

[0111] The assembly quality of vacuum pump blades is a multi-dimensional physical system engineering project, and any deviation in any single indicator may lead to a decrease in the overall performance of the machine. The system's main control program establishes a multi-dimensional quality judgment matrix, which summarizes the physical calculation results of various factors such as blade root assembly clearance, blade tip dynamic running clearance, three-dimensional profile deviation of curved surfaces, and spatial torsion angle deviation of the entire blade.

[0112] The system checks each of these indicators to see if they fall within their respective independent mechanical tolerance zones. Only when all physical indicators in all dimensions meet the design requirements will the system output a final judgment signal of "assembly quality qualified"; if any indicator exceeds the tolerance, the system will output an alarm signal of "assembly quality unqualified" and clearly indicate the specific physical cause of the unqualified result (e.g., "the dynamic clearance of the blade tip is too small when rotated to 145 degrees, which poses a risk of rotor rubbing").

[0113] To help on-site assembly engineers intuitively understand the complex spatial deviation distribution, the system does not use dry data tables, but instead generates intuitive three-dimensional physical deviation chromatograms.

[0114] The system maps the calculated geometric deviation value of each spatial point to different colors in the visible spectrum. For example, the area at the center of the tolerance zone (with deviation close to zero) is rendered as green, the area with gradually increasing positive deviation is rendered as a gradient from yellow to red, and the area with gradually increasing negative deviation is rendered as a gradient from light blue to dark blue.

[0115] This visualization output of physical deviations based on real 3D point clouds allows engineers to rotate and zoom in all directions on the screen to view the actual processing and assembly status of the blades. The red areas intuitively indicate interference areas that need grinding or adjustment, while the blue areas indicate areas with insufficient material.

[0116] Meanwhile, the system packages and stores all raw point cloud data, polarization calculation parameters, geometric fitting equations, and final deviation judgment results in a unified data structure to a local industrial database or uploads them to the Manufacturing Execution System (MES), providing detailed and reliable underlying physical data support for subsequent production process traceability, machining parameter adjustment, and assembly process optimization.

[0117] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0118] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0119] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0120] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0121] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting the assembly quality of vacuum pump blades based on laser sensors, characterized in that, include: Linearly polarized structured light is projected onto the surface of a vacuum pump blade using a polarized laser emission source, and the original polarization images of the vacuum pump blade surface in four different polarization directions are captured at the same exposure time using a micro-polarization array. Physical noise filtering and channel registration are performed on the original polarization image. Stokes parameters are constructed and the linear polarization degree is resolved. Dynamic polarization suppression weights are generated based on the linear polarization degree. The dynamic polarization suppression weights are multiplied with the total light intensity image to output the effective light intensity distribution image against reflection. The sub-pixel coordinates of the laser stripe center are extracted from the anti-reflective effective light intensity distribution image. The sub-pixel coordinates are mapped to three-dimensional physical coordinates by combining the camera intrinsic parameter matrix, external rotation matrix, and translation vector. The three-dimensional point cloud model is generated by stitching together the encoder physical displacement values. The three-dimensional point cloud model is spatially registered with the theoretical model, the rate of change of the normal vector angle of the local micro-surface element is calculated to perform local surface segmentation, and the three-dimensional boundary curve is extracted to fit the key assembly reference surface. A local assembly coordinate system is established based on the key assembly reference surface. The physical assembly clearance between the vacuum pump blade and the rotor slot is calculated. The rotor rotation motion is simulated by combining the spatial three-dimensional boundary curve to calculate the dynamic running clearance of the blade tip. The form and position tolerance of the spatial torsional aerodynamic surface is evaluated, and the assembly quality judgment result is output.

2. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, The micro-polarization array uses four adjacent photosensitive pixels as a basic polarization calculation unit. Above the four adjacent photosensitive pixels, micro-polarization gratings with light transmission axis directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees are respectively etched. The optical axis of the polarized laser emission source, the principal optical axis of the imaging objective lens, and the photosensitive target surface of the micro-polarization array satisfy the constraints of Scham's law, which makes the laser plane, the principal plane of the lens, and the photosensitive target surface intersect on a common straight line in space.

3. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, Constructing the Stokes parameters and resolving the degree of linear polarization includes: Suppose the grayscale values ​​of the original polarization images of a point on the surface of a vacuum pump blade in four different polarization directions are respectively , , and Construct Stokes vectors ; Among them, total light intensity The difference between the components of linearly polarized light in the horizontal and vertical directions The component difference between linearly polarized light in the +45 degree direction and the 135 degree direction ; In order to meet the total light intensity Under the premise of linear polarization degree The calculation formula is: ; linear polarization angle The calculation formula is: .

4. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, The dynamic polarization suppression weights generated based on linear polarization degree include: A polarization degree threshold is set. When the linear polarization degree is lower than the polarization degree threshold, the dynamic polarization suppression weight approaches one. When the degree of linear polarization exceeds the polarization degree threshold and increases, the dynamic polarization suppression weight decreases according to the nonlinear decay curve. The dynamic polarization suppression weights are multiplied pixel by pixel with the total light intensity image to suppress the specular reflection component and retain the diffuse reflection component, thereby generating an effective light intensity distribution image for anti-reflection.

5. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, Extracting the sub-pixel coordinates of the laser stripe center from the effective intensity distribution image for anti-reflection includes: Based on the background noise statistics of a local window, a dynamic grayscale threshold is set to lock the effective physical width range of the laser stripe. Within the effective physical width range, a sub-pixel interpolation algorithm based on grayscale centroid is used to multiply the physical coordinates of each pixel within the effective physical width range with its corresponding grayscale value, sum the results, and then divide by the sum of the grayscale values ​​of all pixels within the effective physical width range to obtain the sub-pixel coordinates of the laser stripe center.

6. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, Mapping sub-pixel coordinates to three-dimensional physical coordinates includes: Using camera intrinsic matrix In addition to radial and tangential distortion coefficients, the subpixel coordinates are restored to distortion-free physical projection coordinates; Combined with external rotation matrix Translation vector Construct the ray equation from the two-dimensional image plane to the three-dimensional camera local coordinate system; The ray equation and the laser plane equation are solved simultaneously to obtain the three-dimensional physical coordinates.

7. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, The coarse spatial pose registration between the 3D point cloud model and the theoretical model includes: Calculate the physical centroid coordinates of the 3D point cloud model and make them coincide with the centroid of the theoretical model. Construct the covariance matrix of the point cloud coordinates, solve the eigenvalues ​​and eigenvectors of the covariance matrix to obtain three mutually perpendicular physical principal axis directions, and rotate and align the three mutually perpendicular physical principal axis directions with the preset reference coordinate axes of the theoretical model. The local surface segmentation process includes: Spatial normal vectors are obtained by fitting local small tangent planes, the rate of change of the spatial angle between the normal vectors of adjacent points is calculated, and a geometric topology algorithm based on region growing is used to split the 3D point cloud model into multiple feature point cloud subsets.

8. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, Extracting spatial 3D boundary curves using a virtual rolling ball includes: Project the points within the feature point cloud subset onto the local fitting plane and set the virtual rolling ball radius. Virtual rolling ball radius The value of is strictly greater than the maximum value of the local average physical spacing of the point cloud; When the radius is When a virtual ball can pass through the gap between two points without touching other points, the two points are determined to be real physical boundary points. Spatial parametric spline curves are used to geometrically fit all discrete physical boundary points to generate spatial three-dimensional boundary curves.

9. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, The calculation of the dynamic operating clearance at the blade tip combines the simulation of rotor rotational motion using three-dimensional boundary curves in space. Establish the theoretical spatial geometric equation of the inner wall of the stator pump cavity to determine the physical spatial position of the rotor rotation axis in the global coordinate system; The three-dimensional boundary curve in space is rotated around the rotor's axis of rotation in small, predetermined angular steps. At each rotation angle position, calculate the shortest spatial normal distance from all discrete points on the three-dimensional boundary curve to the inner wall of the stator pump cavity, generate a continuous physical relationship curve between the dynamic operating clearance of the blade tip and the rotation angle, and find the global minimum distance value on the physical relationship curve.

10. The method for detecting the assembly quality of vacuum pump blades based on a laser sensor according to claim 1, characterized in that, The evaluation of the form and position tolerances of a spatial torsional aerodynamic surface includes: Spatial registration is performed between the point cloud corresponding to the spatial torsional aerodynamic surface and the theoretical surface model. Uniform sampling points are generated on the theoretical surface model, and the spatial normal vector of each uniform sampling point is calculated. Along the direction of the spatial normal vector, search for the nearest physical counterpart in the actual point cloud, calculate the directed spatial distance between the physical counterpart and the uniform sampling point, and construct a three-dimensional physical deviation field covering the entire torsional surface. The three-dimensional physical deviation field is compared with the preset three-dimensional profile tolerance envelope.

Images