Laser scanning detection method for gear tooth error of gear motor

Abstract

The invention discloses a laser scanning detection method for gear tooth form errors of gear motors, and belongs to the technical field of industrial optical detection. The method comprises the following steps of: firstly, comprehensively acquiring high-density three-dimensional point cloud data of a pinwheel tooth profile by combining laser scanning and precise rotating motion; secondly, preprocessing the original point cloud, and performing high-precision automatic registration on the actually measured point cloud and the standard three-dimensional CAD theoretical model by using algorithms such as an iterative nearest point; and finally, generating a visual tooth profile error map by calculating the normal deviation between the measured data and the theoretical model point by point, and quantitatively evaluating error parameters. According to the scheme, high-efficiency, high-precision and all-shape nondestructive testing of the gear tooth error is realized, physical contact damage is avoided, and the quality control level and the production efficiency of the core part of the speed reducer are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial optical detection, and in particular relates to a laser scanning detection method for a pinwheel tooth profile error of a reduction motor. Background Art

[0002] Cycloidal pinwheel reducers are core transmission components in industrial automation and precision machinery. The tooth profile accuracy of their internal pinwheels is directly related to the transmission accuracy, load-bearing capacity, and service life of the entire machine. Therefore, high-precision and efficient detection of pinwheel tooth profile errors is a key production step in ensuring the performance and reliability of reducer products.

[0003] Currently, contact measurement methods, such as those using coordinate measuring machines (CMMs) or dedicated gear measuring instruments, are commonly used to detect pinwheel tooth profile errors. These methods involve driving a physical probe to scan the tooth surface point by point or perform contact motion along a predetermined trajectory, collecting data from a series of discrete coordinate points. The tooth profile error is then assessed through data fitting and analysis. However, these contact measurement methods have inherent technical limitations. Firstly, the point-by-point probing mechanism results in a lengthy measurement process, making it difficult to adapt to the full inspection requirements of high-volume production cycles. Furthermore, the limited sampling point density may not fully capture the full characteristics of the tooth profile, especially for microscopic, high-frequency profile errors. Secondly, the rigid physical contact between the measuring probe and the measured tooth surface not only risks scratching the high-quality precision surface, but also accumulates wear on the probe itself, introducing measurement errors that affect the long-term consistency and accuracy of the test results. Therefore, the industry is eager to develop a more efficient and non-destructive pinwheel tooth profile error detection solution. Summary of the Invention

[0004] To address the aforementioned technical problems, the present application is proposed. An embodiment of the present application provides a laser scanning method for detecting pinwheel tooth profile errors in a reduction motor. This method first uses high-precision, non-contact laser scanning to comprehensively acquire high-density, three-dimensional point cloud data of the pinwheel tooth profile surface. Combined with precise rotational motion control, this method enables rapid, comprehensive acquisition of the tooth surface without blind spots. Furthermore, the raw point cloud is preprocessed through filtering, denoising, and streamlined downsampling. An optimized iterative closest point (ICP) algorithm or its variant is then used to automatically and accurately align the measured point cloud data with a standard 3D CAD theoretical model of the pinwheel. Finally, the normal deviation between the measured data and the theoretical model is calculated point by point, generating an intuitive, visual tooth profile error distribution map. Key error parameters, such as tooth profile and pitch deviation, are then quantitatively assessed based on preset tolerance standards. This method achieves efficient, high-precision, and full-profile non-destructive testing of pinwheel tooth profile errors, significantly improving the quality control and production efficiency of core reducer components.

[0005] According to one aspect of the present application, a laser scanning detection method for a gear motor pinwheel tooth profile error is provided, which includes: S100: Execute system coordinate system calibration to obtain a unified transformation relationship between the laser scanner coordinate system and the rotary table coordinate system. Specifically, a standard part with known geometric features (for example, a high-precision ceramic calibration ball or a standard gauge block) is placed at the center of the rotary table, and the laser scanner is controlled to scan the standard part from at least three different azimuth angles to obtain multiple sets of local point clouds. Subsequently, the center of each set of point clouds is fitted with a sphere center or a plane to calculate the center point coordinates of the standard part at each scanning position. Based on the known precise rotation angle of the rotary table and combined with multiple sets of center point coordinates, the least squares method or singular value decomposition (SVD) method is used to accurately solve the rotation and translation transformation matrices from the laser scanner coordinate system to the rotary table coordinate system, thereby establishing a unified measurement coordinate system.

[0006] S200: Collect the three-dimensional point cloud data of the pinwheel to be measured. Install the reduction motor pinwheel to be measured firmly on the fixture of the rotary table, ensuring that its center of rotation is coaxial with the axis of rotation of the table. Set the scanning parameters, which may include the laser line scanning frequency, laser power, angular velocity of the rotary table, and trigger mode of data acquisition. Start the laser scanner and the rotary table so that the pinwheel rotates at a constant speed while the line laser emitted by the laser scanner continuously scans the tooth profile surface of the pinwheel. The camera installed inside the scanner synchronously captures the deformation of the laser stripes on the tooth surface, and solves a large number of (usually millions to tens of millions) three-dimensional coordinate points in real time through the principle of triangulation, forming the original three-dimensional point cloud data covering the entire pinwheel tooth profile.

[0007] Optionally, the laser scanner can be a scanner based on red laser, or in another preferred embodiment, in order to cope with highly reflective metal surfaces, a blue laser scanner can be used because it has a shorter wavelength and stronger anti-interference ability, and can obtain point cloud data with a higher signal-to-noise ratio.

[0008] S300: Preprocess the collected raw 3D point cloud data. This step aims to eliminate noise interference and reduce data redundancy, laying the foundation for subsequent accurate registration and analysis. The preprocessing process may include: a) Outlier Removal: The Statistical Outlier Removal (SOR) algorithm is used to perform statistical analysis on the neighborhood of each point and remove isolated noise points whose distance from their neighborhood exceeds a preset standard deviation.

[0009] b) Data smoothing and filtering: Optionally, apply Gaussian Laplace filtering or bilateral filtering algorithms to smooth the point cloud while preserving the edge features of the tooth profile to reduce the impact of random measurement noise.

[0010] c) Data Reduction: Voxel Grid Downsampling (Voxel Grid Downsampling) is used to divide the entire point cloud space into tiny three-dimensional grids (voxels). All points within each non-empty voxel are replaced by the centroid of all points within that voxel. This significantly reduces the number of point clouds and improves subsequent processing speed while effectively maintaining the macroscopic geometric shape characteristics of the tooth profile.

[0011] S400: Accurately align the pre-processed measured point cloud with the standard 3D CAD theoretical model of the pinwheel. This step is the core of the error analysis, and its purpose is to align the measured data spatially with the design benchmark.

[0012] First, load the standard 3D CAD theoretical model of the pinwheel, which can be in standard formats such as STEP, IGES or STL.

[0013] Next, perform a rough registration. This can be done by manually selecting at least three pairs of corresponding feature points on the measured point cloud and the theoretical model (for example, points on the addendum circle, root circle, or pin hole center), or by using a sample consistency-based initial registration algorithm (such as Sample Consensus Initial Alignment (SAC-IA)) to calculate an initial transformation matrix to roughly align the measured point cloud to the theoretical model.

[0014] Next, fine registration is performed. Based on the coarse registration, the Iterative Closest Point (ICP) algorithm is applied for precise alignment. The ICP algorithm iteratively performs the following two steps until convergence: 1) For each point in the measured point cloud, the closest corresponding point on the theoretical CAD model is found; 2) Based on these point pairs, a rigid transformation (rotation and translation) is calculated that minimizes the average distance between the point pairs, and this transformation is applied to update the position of the measured point cloud.

[0015] In another embodiment, in order to improve insensitivity to initial position and registration speed, the precise registration algorithm may also be a Normal-Distributions Transform (NDT) algorithm, which divides the point cloud into grid cells and uses normal distribution to describe the point distribution within each cell, and achieves registration by optimizing the matching degree of the probability distribution of the two point clouds.

[0016] S500: Based on the registration results, the pinwheel's tooth profile error is calculated and evaluated. The precisely aligned measured point cloud is compared and analyzed with the CAD theoretical model. Specifically, for each point in the measured point cloud, the shortest normal distance (i.e., signed distance) from the measured point to the CAD model surface is calculated. This distance represents the tooth profile error at that point. Positive values ​​indicate excess material, while negative values ​​indicate insufficient material. The collection of error values ​​for all points forms a high-density global error field.

[0017] Based on this error field, we can: a) Generate a visual error map: Calculated error values ​​at each point are intuitively displayed on the 3D model using pseudo-color rendering, forming a color-coded error cloud map. Operators can clearly observe the distribution, area, and magnitude of errors, quickly locating areas with out-of-tolerance conditions.

[0018] b) Quantitative evaluation of error parameters: According to national standards or internal enterprise standards (such as GB / T 10095 or similar standards), key tooth profile error indicators such as total tooth profile deviation (Fα), tooth profile shape deviation (ffα) and tooth profile slope deviation (fHα) are extracted and calculated from the error field data.

[0019] c) Test report output: Automatically generate a test report containing an error cloud diagram, quantitative values ​​of key error parameters, and pass / fail judgment results based on preset tolerances.

[0020] Compared with the existing technology, the laser scanning detection method for the tooth profile error of the pinwheel of a reduction motor provided in this application first obtains high-density three-dimensional point cloud data of the pinwheel tooth profile surface through high-precision non-contact laser scanning, and combines it with precise rotational motion control to achieve rapid acquisition of the entire tooth surface without blind spots. Furthermore, by pre-processing the original point cloud through filtering and denoising, streamlining and downsampling, and using an optimized iterative closest point (ICP) algorithm or its variant, the measured point cloud data is automatically aligned with the standard three-dimensional CAD theoretical model of the pinwheel with high precision. Finally, by calculating the normal deviation between the measured data and the theoretical model point by point, an intuitive visual tooth profile error distribution map is generated, and according to the preset tolerance standards, key error parameters such as tooth profile and pitch deviation are quantitatively evaluated. In this way, not only the potential damage to the precision tooth surface caused by physical contact is avoided, but also the detection cycle is greatly shortened by virtue of the extremely high data acquisition density and automated analysis process, and the comprehensive capture of the micro-geometric features of the tooth profile can be achieved, thereby realizing high-efficiency, high-precision, full-morphology non-destructive testing of the pinwheel tooth profile error, and significantly improving the quality control level and production efficiency of the core components of the reducer. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1The present invention is a schematic diagram of an application environment of a laser scanning detection system for a reduction motor pinwheel tooth profile error provided by an embodiment of the present invention.

[0022] Figure 2 The present invention is a flow chart of an embodiment of a laser scanning detection method for a pinwheel tooth profile error of a reduction motor.

[0023] Figure 3 3D point cloud data preprocessing steps according to an embodiment of the present invention.

[0024] Figure 4 3 is a flow chart of the steps for registering a measured point cloud with a standard CAD model according to an embodiment of the present invention.

[0025] Figure 5 Schematic diagram of a pinwheel tooth profile error visualization map generated according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0027] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0028] In addition, numerous specific details are provided in the following detailed description to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0029] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0030] Conventional technology often uses contact-type coordinate measuring machines (CMMs) to measure tooth profile errors on pinwheels, a core component of geared motors. However, this method suffers from the following drawbacks: First, low detection efficiency and severely insufficient data coverage. CMMs employ single-point trigger measurement, requiring dozens or even hundreds of points to capture the profile of a single tooth surface. This makes full pinwheel tooth inspection extremely time-consuming and unsuitable for mass production cycles. Furthermore, this sparse sampling method fails to construct a complete three-dimensional tooth surface topography and easily misses minor local defects (such as machining burrs, tiny pits, or worn areas), leading to inaccuracies in the overall assessment of tooth profile accuracy. Second, there is a risk of damaging the precision surface and compromising measurement accuracy. The CMM's measuring probe requires physical contact with the tooth surface being measured. For hardened, high-finish, and hardened tooth surfaces, contact forces can cause scratches or indentations, impacting the pinwheel's meshing performance and service life. Furthermore, probe wear, measurement force instability, and complex probe radius compensation algorithms all introduce additional measurement uncertainty.

[0031] In response to the above technical problems, the present application proposes a laser scanning detection method for the tooth profile error of the pinwheel of a reduction motor. Figure 1 A schematic diagram of the application environment of a laser scanning detection system for the pinwheel tooth profile error of a reduction motor provided by an embodiment of the present invention. Figure 2 This is a flow chart of an embodiment of a laser scanning detection method for a gear motor pinwheel tooth profile error according to the present invention. Figure 1 and Figure 2 As shown, the method according to the embodiment of the present application first uses a detection system integrating a high-precision laser scanner and a precision rotary table to perform non-contact scanning of the pinwheel to be measured, acquiring complete 3D point cloud data of its surface. Subsequently, through a series of data preprocessing, precise alignment with a standard CAD model, and final deviation calculation and analysis processes, a comprehensive, rapid, and highly accurate assessment of tooth profile errors is achieved.

[0032] Reference Figure 2 , the method specifically comprises the following steps: Step S100: Perform system coordinate system calibration. It should be understood that since the laser scanner acquires data in its own coordinate system, while the pinwheel to be measured rotates in the coordinate system of the rotary table, in order to accurately fuse the local point cloud data from multiple scans (or continuous acquisition during rotation) into the same global coordinate system, a rigid transformation relationship must first be precisely established between the two coordinate systems. This calibration process is fundamental to ensuring ultimate measurement accuracy. In particular, in a specific example of the present application, a standard ceramic calibration sphere with precisely known geometric dimensions can be fixed to the center of the rotary table. The laser scanner is then controlled to scan the calibration sphere from at least three different angular positions, generating three independent sets of local point clouds. By applying a high-precision spherical fitting algorithm to each set of point clouds, the three center coordinates of the calibration sphere in the laser scanner coordinate system can be calculated. Combined with the precise rotation angle read by the rotary table's own encoder, an overdetermined set of equations can be constructed, and the least squares method or the more robust singular value decomposition (SVD) method can be used to solve the rotation matrix and translation vector from the laser scanner coordinate system to the rotary table coordinate system, thereby completing the establishment of a unified measurement coordinate system.

[0033] Step S200: Collect three-dimensional point cloud data of the pinwheel to be measured. It should be understood that in order to achieve complete, blind-angle measurement of the pinwheel tooth profile, line scanning must be closely integrated with rotational motion. The pinwheel to be measured is securely mounted on a rotary table using a dedicated fixture and precisely adjusted to ensure that the pinwheel's centerline of rotation is strictly coaxial with the table's axis of rotation to avoid measurement errors introduced by eccentricity. Subsequently, optimized scanning parameters, such as laser power, scanning frequency, and the angular velocity of the rotary table, are set based on the pinwheel's material, surface finish, and dimensions. For example, for a pinwheel with a mirror-polished surface, a blue laser scanner with enhanced anti-reflective interference capabilities can be selected, and the scanning speed can be appropriately reduced to obtain data with a higher signal-to-noise ratio. After the scan is initiated, the rotary table drives the pinwheel to rotate at a set constant speed. Simultaneously, the laser scanner projects a line of laser light onto the tooth profile surface. The high-speed camera within the scanner captures the deformation profile of the laser line on the complex tooth surface in real time. Based on the principle of optical triangulation, the system can solve the three-dimensional spatial coordinates of thousands of points on the laser line in real time. As the needle wheel continues to rotate, it will eventually be compiled into millions or even tens of millions of high-density original point cloud data covering all tooth surfaces, tooth roots and transition fillets.

[0034] Step S300: pre-processing the collected original three-dimensional point cloud data. Figure 3The figure is a flow chart illustrating the steps for preprocessing 3D point cloud data according to one embodiment of the present invention. It should be understood that the raw point cloud data inevitably contains isolated noise points (outliers) caused by ambient light interference, surface impurity reflections, or electrical noise, as well as information redundancy due to massive data collection. Performing subsequent analysis directly on the raw data will not only affect the accuracy of registration and calculations, but will also significantly increase the computational burden and reduce detection efficiency. Therefore, effective preprocessing is a key step in ensuring the quality and speed of subsequent analysis. Specifically, the preprocessing process may include: a) Outlier Removal: Statistical Outlier Removal (SOR) is used. For each point in the point cloud, the algorithm calculates the average distance to a specified number of neighboring points (e.g., 50). If this average distance for a point exceeds a threshold determined by the average distance and standard deviation of all points (e.g., the average distance plus twice the standard deviation), the point is considered an outlier and removed.

[0035] b) Data smoothing and filtering: Optionally, to reduce the impact of measurement noise on tooth surface smoothness, a Laplacian of Gaussian filter or bilateral filtering algorithm can be applied. Bilateral filtering not only performs Gaussian smoothing in the spatial domain but also introduces Gaussian weighting in the pixel value domain. This effectively filters out random noise while preserving the geometric features of key edges, such as the tooth tip and tooth root, to the greatest extent possible, avoiding feature blurring.

[0036] c) Data Reduction: Voxel Grid Downsampling (Voxel Grid Downsampling) is used. This method divides the 3D space containing the entire point cloud into a grid of tiny cubes (voxels). Then, all points within each non-empty voxel are traversed and replaced with their centroids (i.e., average coordinates). This significantly reduces the number of point clouds (for example, from tens of millions to millions), significantly improving subsequent processing speed, while effectively preserving the macroscopic geometry of the tooth profile, laying the foundation for accurate registration.

[0037] Step S400: accurately aligning the pre-processed measured point cloud with the standard 3D CAD theoretical model of the pinwheel. Figure 4 This is a flow chart illustrating the steps for registering a measured point cloud with a standard CAD model, according to one embodiment of the present invention. It should be understood that the essence of error analysis is to compare the difference between "actual" and "ideal." Therefore, the measured point cloud representing "actual" and the CAD model representing "ideal" must be precisely aligned within the same coordinate system. This step is the core of the entire inspection process.

[0038] First, load the standard 3D CAD theoretical model of the pinwheel representing the design benchmark. The model is usually in a format such as STEP or IGES with precise surface definition.

[0039] Secondly, perform a coarse registration. Since the spatial position and posture of the measured point cloud and the CAD model are arbitrary when initially loaded, direct fine registration may converge slowly or fall into a local optimum. Therefore, an initial alignment is required. This can be achieved by manually selecting at least three pairs of obvious corresponding feature points (such as the center of the pin hole and the point on the top circle of the gear) on the two models, or by using a more automated sampling consistency initial alignment algorithm (SAC-IA). The SAC-IA algorithm randomly samples points on the measured point cloud and searches for corresponding points with similar geometric features (such as normal vectors and curvature) on the CAD model, thereby quickly calculating an initial transformation matrix to move the measured point cloud roughly to the vicinity of the CAD model.

[0040] Then, fine registration is performed. Based on the coarse registration, the iterative closest point (ICP) algorithm is applied for the final precise alignment. The ICP algorithm is an iterative optimization process: in each iteration, 1) for each point in the measured point cloud, the point with the closest geometric distance to it on the CAD model surface is searched as the corresponding point; 2) based on all the corresponding point pairs found, a rigid transformation (a rotation matrix and a translation vector) is calculated that minimizes the average squared distance between them; 3) this transformation is applied to the entire measured point cloud, updating its position. This process is repeated until the average distance between the point pairs is less than a preset minimum threshold or the number of iterations reaches an upper limit, thus achieving high-precision alignment of the two models.

[0041] In another preferred embodiment, to improve registration speed and robustness, a Normal Distribution Transform (NDT) algorithm can also be employed. The NDT algorithm does not process points directly. Instead, it first divides the space into a grid and uses a normal distribution (i.e., a mean and covariance matrix) to probabilistically describe the distribution of the point cloud within each grid cell. The registration process optimizes an objective function to find a transformation that maximizes the probability of points in one point cloud being represented by the NDT model of the other point cloud, thereby achieving alignment.

[0042] Step S500: Calculate and evaluate the tooth profile error of the pinwheel based on the registration result. It should be understood that after the measured point cloud and the theoretical model are accurately aligned, the geometric deviation between them intuitively reflects the manufacturing error. Figure 5 The figure below is a schematic diagram of a pinwheel tooth profile error visualization map generated according to one embodiment of the present invention. Specifically, for each point in the aligned measured point cloud, the shortest normal distance from that point to the surface of the CAD theoretical model is calculated. This distance is a signed value; a positive value typically indicates excess material (overcut) compared to the theoretical model, while a negative value indicates insufficient material (undercut). This distance represents the tooth profile error at that point. The collection of error values ​​for all points constitutes a high-density global error field.

[0043] Based on this error field data, multi-dimensional analysis and evaluation can be performed: a) Generate error visualization map: Figure 5 As shown, the error value calculated for each point is mapped to a preset color spectrum (for example, from blue for maximum negative deviation, through green for zero deviation, to red for maximum positive deviation). These colors are then rendered onto the 3D model, forming an intuitive error cloud map. This allows inspectors to clearly see the distribution of errors, the location of out-of-tolerance areas, and their severity.

[0044] b) Quantitatively evaluate error parameters: Key tooth profile error indicators are extracted and calculated from high-density error field data according to relevant national or industry standards (e.g., GB / T 10095). For example, an error curve can be extracted along the normal section of the tooth profile, and a series of quantitative evaluation parameters such as total tooth profile deviation (Fα), tooth form deviation (ffα), and tooth profile slope deviation (fHα) can be calculated based on this curve.

[0045] c) Test report output: The system automatically generates a detailed test report containing basic information about the pinwheel, a visual error map, quantitative values ​​of all key error parameters, and the final pass / fail judgment result based on the preset tolerance standard (for example, profile tolerance ±0.02mm).

[0046] In summary, a laser scanning detection method for the tooth profile error of the pinwheel of a reduction motor based on an embodiment of the present application realizes high-density data collection of the entire pinwheel tooth surface through the combination of high-precision non-contact scanning and a precision rotary table, fundamentally avoiding the surface damage that may be caused by contact measurement. Through systematic preprocessing of point cloud data, precise alignment with the CAD model, and final global error field analysis, not only can an intuitive error visualization map be generated, but also various tooth profile error parameters that meet the standards can be accurately calculated. This method shortens the workload of traditional detection methods from several hours to several minutes, greatly improving the detection efficiency. At the same time, with its comprehensive morphology capture capability, it significantly improves the accuracy and reliability of detection, providing strong technical support for improving the quality control level of core components of the reducer.

[0047] As described above, the laser scanning detection method for gear motor pinwheel tooth profile errors according to embodiments of the present application can be implemented in a variety of testing equipment. In one example, the method can be integrated into a set of automated gear testing equipment as a core software module and / or hardware control module. For example, the method can be run as software on an industrial computer, controlling hardware such as a laser scanner and a rotary stage to perform measurement tasks and process data. Of course, the method's specific algorithms (such as ICP registration) can also be accelerated using hardware such as an FPGA and integrated into the detection system as a dedicated data processing unit.

[0048] Example 1 This example illustrates the application of the method described herein in a typical high-precision reducer manufacturing production line, performing fully automated quality inspection on a batch of hardened steel pinwheels with an outer diameter of 120 mm and 11 teeth. Specifically, the inspection system environment in this example is deployed on an industrial control computer equipped with a multi-core central processing unit (CPU) with 16 physical cores, a base clock frequency of 2.8 GHz, and 128 GB of DDR4 main memory. The system integrates a laser scanning device, including a red laser line scanner with an emission wavelength of 660 nm and a precision rotary stage with an angular positioning accuracy of 0.001 degrees. The system operates in a Linux operating system with a version 5.4 kernel. The data processing software is implemented using the programming language version 3.9 and the point cloud processing library version 1.12. The inspection process strictly adheres to the method described herein: First, the system coordinate system is calibrated using a G5-grade standard ceramic ball with a diameter of 25 mm. The transformation matrix between the scanner and the rotary stage is accurately solved using singular value decomposition (SVD). The pinwheel to be measured was then mounted on a workbench and rotated at a constant angular velocity of 10 rpm. A laser scanner was used at 500 Hz to collect raw 3D point cloud data consisting of approximately 20 million points covering the entire tooth profile. The raw point cloud was then preprocessed using the statistical outlier removal (SOR) algorithm (with a neighborhood of k=50 points and a standard deviation threshold of 2.0) to remove noise. Data was then reduced using a voxelized grid downsampling method with a voxel size of 0.1 mm, resulting in a reduced point cloud consisting of approximately 1.5 million points. A standard 3D CAD model in STEP format was then loaded and coarsely aligned using the sampled consistency initial registration algorithm (SAC-IA). Fine registration was then performed using the iterative closest point (ICP) algorithm until the average point distance between pairs of points was less than 0.001 mm. Finally, the normal deviation of each point in the reduced point cloud to the CAD model surface was calculated to generate a pseudo-color error contour map, and key parameters such as the total tooth profile deviation (Fα) were extracted. After testing, this embodiment demonstrates high efficiency and high precision detection capabilities, and its specific performance data is shown in Table 1 below.

[0049] Example 2 This example illustrates the adaptability and superiority of the method described herein for handling special surface materials. The specific application scenario involves inspecting a batch of special alloy pinwheels with mirror-polished, highly reflective surfaces. Aside from the differences in the surface characteristics of the workpieces being tested, the computing hardware, operating system, software environment, and basic inspection process employed in this example remain consistent with those in Example 1. To mitigate interference from highly reflective surfaces with laser measurement, this example utilizes a line scanner based on blue laser technology (emission wavelength of 450 nm). Due to its shorter wavelength, it offers higher energy absorption on high-gloss metal surfaces and enhanced interference resistance, thereby generating raw point cloud data with a higher signal-to-noise ratio. In the point cloud preprocessing step, in addition to the statistical outlier removal and voxelized grid downsampling performed in Example 1, an additional bilateral filtering step is added to effectively smooth out minor measurement noise while maximally preserving the edge features of key contours, such as tooth tips and roots, to avoid feature blurring. The subsequent registration and error analysis steps are identical to those in Example 1. After testing, this embodiment successfully overcomes the detection difficulty of highly reflective surfaces by adopting blue laser and optimized pre-processing process, and obtains extremely high measurement accuracy. Its specific performance data can also be found in Table 1 below.

[0050] Example 3 This embodiment is intended to illustrate another preferred technical solution of the present invention in scenarios where extreme inspection efficiency is sought. The application scenario of this embodiment is high-volume, high-frequency online full inspection of pinwheels for automotive automatic transmissions, which places extremely stringent requirements on inspection speed. The hardware configuration, software environment, workpiece to be tested, and all processing steps except the fine registration algorithm employed in this embodiment are consistent with those of Example 1. The core difference lies in the use of the Normal Distribution Transform (NDT) algorithm in the fine registration phase of step S400, replacing the Iterative Closest Point (ICP) algorithm used in Example 1. In the specific implementation, both the CAD theoretical model and the measured point cloud after coarse registration are divided into a 1.0 mm cubic grid, and the normal distribution parameters (mean and covariance) of the internal point cloud are calculated for each grid cell. The registration process uses an optimization algorithm based on Newton's method to rapidly search for a rigid transformation that maximizes the probability of the measured point cloud being represented in the NDT representation of the CAD model. Because the NDT algorithm doesn't rely on point-to-point nearest neighbor searches, its computational complexity is less dependent on point cloud size. Especially when point cloud density is high, its convergence speed is significantly faster than ICP. Testing has shown that this embodiment significantly reduces data processing time while maintaining extremely high registration accuracy, demonstrating exceptional efficiency. Specific performance data is also provided in Table 1 below.

[0051] Comparative Example 1 This comparative example is used for comparison with Example 1 to highlight the necessity and importance of the point cloud preprocessing step in the present invention. The only difference between this comparative example and Example 1 is that after acquiring the raw 3D point cloud data, no preprocessing steps are performed (i.e., step S300 is omitted). Instead, the raw point cloud, containing approximately 20 million points and containing a significant amount of noise and outliers, is directly used for subsequent registration with the CAD model. Apart from this difference, the hardware system, software environment, workpiece to be measured, scanning parameters, and registration algorithm (SAC-IA + ICP) employed in this comparative example remain identical to those in Example 1. Directly processing massive amounts of noisy data significantly increases the computational burden of the ICP algorithm, and it is highly susceptible to interference from outliers during the iterative process, resulting in slow convergence and ultimately falling into an erroneous local optimum, making precise alignment impossible. Testing under identical conditions revealed that this comparative example significantly increased detection time and significantly reduced registration accuracy, resulting in a completely distorted calculated tooth profile error, making the detection objective of the present invention infeasible. Specific performance data is also provided in Table 1 below.

[0052] Comparative Example 2 This comparative example is used for comparison with Example 1, aiming to highlight the overall technical advantages of the present invention by comparing it with the prior art. This comparative example employed a standard contact coordinate measuring machine (CMM) from the prior art to inspect a hardened steel pinwheel identical to that used in Example 1. The inspection scheme involved planning a measurement path along three sections at different heights along each pinwheel tooth, collecting 30 measurement points along each section path. Additionally, 44 points were collected on both the root and addendum circles. A total of approximately 3,500 discrete 3D coordinate points were collected for the entire pinwheel. The measurement process was manually programmed and executed by a skilled inspection engineer. Aside from fundamental differences in the inspection method and equipment, the inspection objectives and error parameters to be assessed were the same as those in Example 1. Testing revealed that this contact measurement method was extremely time-consuming and unable to meet the cycle time requirements of mass production. Furthermore, its data acquisition density was extremely low, capturing only profile information along a limited path and failing to construct a complete 3D tooth surface topography. This made it easy to miss minor local machining defects, leading to systemic risks and uncertainties in the assessment of total tooth profile deviation. Its specific performance data can also be found in Table 1 below.

[0053] Effect verification In order to more intuitively demonstrate the beneficial effects of the present invention, the key performance test data of the above embodiments and comparative examples are summarized in Table 1. Table 1 Performance comparison data table; The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A laser scanning detection method for the pinwheel tooth profile error of a reduction motor, characterized in that: include: Perform system coordinate system calibration to obtain the unified transformation relationship between the laser scanner coordinate system and the rotary table coordinate system; Collecting three-dimensional point cloud data of the pinwheel to be measured, the collecting step comprising mounting the pinwheel to be measured on the rotary table and using the laser scanner to obtain original three-dimensional point cloud data covering the tooth profile surface of the pinwheel to be measured when the pinwheel to be measured rotates; Preprocessing the original three-dimensional point cloud data to eliminate noise and simplify the data; Accurately registering the pre-processed measured point cloud with the standard three-dimensional CAD theoretical model of the pinwheel to be measured, so as to spatially align the measured point cloud with the standard three-dimensional CAD theoretical model; Based on the registration result, the tooth profile error of the pin wheel to be measured is calculated and evaluated.

2. The method according to claim 1, characterized in that The execution system coordinate system calibration includes: Placing the standard part on the rotating workbench, and controlling the laser scanner to scan the standard part from at least three azimuth angles to obtain multiple sets of local point clouds; Performing geometric feature fitting on the multiple groups of local point clouds to calculate the center point coordinates of the standard component at each scanning position; Based on the known rotation angle of the rotary table and multiple sets of center point coordinates, the least square method or singular value decomposition method is used to solve the rotation and translation transformation matrices from the laser scanner coordinate system to the rotary table coordinate system.

3. The method according to claim 1, characterized in that The laser scanner is a blue laser scanner.

4. The method according to claim 1, wherein The preprocessing of the original three-dimensional point cloud data includes: Using a statistical outlier removal algorithm to remove isolated noise points from the original three-dimensional point cloud data; The voxelized grid downsampling method is used to simplify the 3D point cloud data after removing isolated noise points.

5. The method according to claim 1, characterized in that The precise registration of the pre-processed measured point cloud with the standard three-dimensional CAD theoretical model of the pinwheel to be measured comprises: Performing coarse registration to calculate an initial transformation matrix to roughly align the measured point cloud to the position of the standard 3D CAD theoretical model; Based on the coarse registration, fine registration is performed to accurately align the measured point cloud with the standard 3D CAD theoretical model.

6. The method according to claim 5, characterized in that The coarse registration is performed by adopting an initial registration algorithm based on sampling consistency.

7. The method according to claim 5, characterized in that The precise registration is performed by applying an iterative closest point algorithm, which iteratively searches for corresponding points of points in the measured point cloud on the standard three-dimensional CAD theoretical model and calculates a rigid transformation that minimizes the average distance between point pairs to update the position of the measured point cloud until convergence.

8. The method according to claim 5, characterized in that The precise registration is performed by applying a normal distribution transformation algorithm, which divides the point cloud into grid cells and describes the point distribution within the cell using a normal distribution, and achieves registration by optimizing the matching degree of probability distributions of the two point clouds.

9. The method according to claim 1, characterized in that The calculating and evaluating the tooth profile error of the pin wheel to be measured based on the alignment result includes: For each point in the measured point cloud, the shortest normal distance from the point to the surface of the standard 3D CAD theoretical model is calculated to form a global error field; The calculated error values ​​of each point are rendered in pseudo-color to generate a color-coded error visualization map.

10. The method according to claim 9, characterized in that The method further comprises: Extracting and calculating the total tooth profile deviation, tooth profile shape deviation and tooth profile slope deviation from the global error field data; A test report is generated, which includes the error visualization map, the quantitative values ​​of the total tooth profile deviation, the tooth profile shape deviation and the tooth profile slope deviation, and a qualified or unqualified judgment result based on a preset tolerance.

Images