Steering wheel assembly size measurement method and system based on multi-data fusion

Abstract

The application discloses a kind of based on multi-data fusion's steering wheel assembly size measurement method and system, it is related to steering wheel assembly size measurement field, comprising: the standard three-dimensional point cloud model of steering wheel assembly is constructed, and the reference feature set for size and shape tolerance detection is calibrated;Standard three-dimensional point cloud model is segmented into K point cloud subsets, and digital reference comparison knowledge base is constructed;The three-dimensional point cloud data of the steering wheel assembly to be measured in static state is collected, and the static point cloud model to be measured is generated;The static point cloud model to be measured and standard three-dimensional point cloud model are based on the identification and space alignment of each physical component;Extract point, surface and line reference features, and calculate the size and shape tolerance of steering wheel assembly;Dynamic detection model is constructed, and its dynamic performance index is calculated.The intelligent identification and alignment of steering wheel assembly from unordered point cloud to engineering semantics are realized, so that the comprehensive detection of size, shape tolerance and dynamic performance is automatically completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of steering wheel assembly size measurement, in particular to a steering wheel assembly size measurement method and system based on multi-data fusion. BACKGROUND

[0002] With the improvement of the popularity of automobiles, driving experience has become one of the core decision-making factors for consumers to choose and purchase vehicles. The size precision of the steering wheel, as the core control component that the driver contacts most frequently, directly determines the holding comfort, the response sensitivity of the control, and the operation convenience of the multifunctional keys, and further deeply affects the driving safety and experience. For example, a slight deviation in the cross-sectional diameter of the grip ring may cause fatigue after long-time driving, inaccurate positioning of the feature hole may cause excessive assembly gap, and excessive displacement of the shaft sleeve may cause steering jam. These size defects will reduce the driving experience and even cause safety hazards. However, with the acceleration of the automobile industry automatic production line, the traditional steering wheel assembly size detection method (sampling detection and manual measurement) has been difficult to adapt to mass production requirements. On the one hand, manual measurement relies on the experience of the detector, is low in efficiency, and cannot achieve full detection, resulting in some defective products flowing into the market. On the other hand, manual measurement is highly subjective, and the measurement error of precise dimensions such as flatness and axis parallelism is large, which is difficult to meet the precision requirements of modern automobiles for parts.

[0003] The rise of three-dimensional scanning technology provides a technical possibility for full detection of steering wheel assemblies. Through high-precision three-dimensional scanning sensors, high-density three-dimensional point cloud data of the measured part can be quickly collected, and automatic and non-contact measurement of the size can be realized, which not only improves the detection efficiency but also ensures the objectivity of measurement. In theory, by scanning each steering wheel assembly before it is put into operation, full detection can be achieved, thereby reducing the defective rate of the steering wheel and improving the precision and reliability of the steering wheel measurement. Further, accurately converting the three-dimensional point cloud data obtained by three-dimensional scanning into steering wheel assembly size data is an important technical means to realize full detection of the steering wheel assembly.

[0004] But the prior art generally relies on strict matching of point cloud coordinates and three-dimensional digital model coordinates, and realizes feature division and data extraction through similarity algorithms (such as the iterative closest point ICP algorithm) or corresponding position deviation analysis, but in actual industrial detection scenarios, it is easy to be affected by multiple factors, which leads to the failure of the matching logic, causes the measurement result to deviate from the actual situation, and issues an incorrect maintenance warning, and the multiple factors that affect can include: 1, environmental light interference (such as strong light reflection, shadow shielding) and scanning angle limitation (such as blind area of the back of the steering wheel, the inside of the lever, etc.), which will cause the collected point cloud data to be partially missing or uneven in density, and the number of point clouds in some key feature areas (such as the inner wall of the feature hole and the shaft sleeve connection end) is insufficient; 2, the small displacement of the measured part during the scanning process and the system error of the scanner itself will cause the point cloud coordinates to deviate from the preset model coordinates, and even after correction through coordinate transformation, it is difficult to completely eliminate the deviation; 3, the existing technology uses interpolation completion for the missing area of the point cloud, but the interpolation process will introduce additional errors, causing the fitted geometric features (such as the center coordinates of the circle and the axis equation) to deviate from the actual situation. SUMMARY

[0005] To solve the above technical problems, a steering wheel assembly size measurement method and system based on multi-data fusion are provided, which solves the problem of strict matching of point cloud coordinates and three-dimensional digital model coordinates in the background art, which is easy to be affected by multiple factors in actual industrial detection scenarios, leading to the failure of the matching logic, causing the measurement result to deviate from the actual situation, and issuing an incorrect maintenance warning.

[0006] To achieve the above purpose, the technical scheme adopted by the present application is:

[0007] A steering wheel assembly size measurement method based on multi-data fusion, comprising:

[0008] Constructing a standard three-dimensional point cloud model of the steering wheel assembly, and defining and calibrating a set of reference features for size and geometric tolerance detection on the model, the reference features including: feature points, feature surfaces and feature lines;

[0009] According to the standard three-dimensional point cloud model, based on the K-Means clustering algorithm, the standard three-dimensional point cloud model is divided into K point cloud subsets, and a digital reference comparison knowledge base is constructed;

[0010] Collecting three-dimensional point cloud data of the measured steering wheel assembly in a static state through a three-dimensional scanning sensor to generate a measured static point cloud model;

[0011] Through the digital reference comparison knowledge base, the measured static point cloud model is compared with the standard three-dimensional point cloud model based on the identification and spatial alignment of each physical component;

[0012] According to the spatial alignment of the static point cloud model, based on the digital reference comparison knowledge base, the point, surface and line reference features are extracted, and the size and geometric tolerance of the steering wheel assembly are calculated;

[0013] Based on the size and geometric tolerance of the steering wheel assembly, a dynamic detection model is constructed, and the dynamic performance index is calculated.

[0014] Preferably, according to the standard three-dimensional point cloud model, based on the K-Means clustering algorithm, the standard three-dimensional point cloud model is divided into K point cloud subsets, and a digital reference comparison knowledge base is constructed, which specifically includes:

[0015] According to the type and number of components of the steering wheel assembly to be measured, the number K of clustering clusters is determined;

[0016] Based on the K-Means clustering algorithm, the data point set of the standard three-dimensional point cloud model is preliminarily aggregated to obtain K initial point cloud clusters;

[0017] According to the pre-defined reference feature set, the initial point cloud cluster is optimized and adjusted to obtain K component-level reference point cloud clusters, and each reference point cloud cluster accurately corresponds to a physical component;

[0018] For each reference point cloud cluster, the cluster centroid of each reference point cloud cluster is calculated;

[0019] Taking the cluster centroid as the origin, the covariance matrix of all points in the cluster relative to the origin is calculated, and the direction of the eigenvector with the largest eigenvalue is taken as the X axis, the direction of the eigenvector with the second largest eigenvalue is taken as the Y axis, and the cross product of the two is taken as the Z axis, to construct a right-handed orthogonal local coordinate system;

[0020] In this local coordinate system, a cuboid bounding box with edge length L is constructed around the origin along the X, Y and Z axes, and the cuboid space is uniformly divided into MxMxM three-dimensional voxel grids;

[0021] For each point in the cluster, its three-dimensional coordinates in the local coordinate system are calculated, and it is determined which specific voxel grid it falls into according to the three-dimensional coordinates;

[0022] The number of points falling into each voxel grid is counted to form an MxMxM dimensional histogram vector, and the L2 norm normalization is performed on the vector to obtain the reference descriptor vector;

[0023] A digital reference comparison knowledge base is constructed, and a separate storage record is created for each physical component;

[0024] Each record contains: component ID and name, reference point cloud cluster and centroid coordinates, reference descriptor vector, associated detection features and theoretical geometric parameters.

[0025] Preferably, the identification and spatial alignment of the to-be-tested static point cloud model with the standard three-dimensional point cloud model based on each physical component through the digital reference comparison knowledge base specifically comprises:

[0026] According to the to-be-tested static point cloud model, a plurality of to-be-tested point cloud clusters are obtained by using the Euclidean clustering algorithm;

[0027] For each to-be-tested point cloud cluster, a to-be-tested descriptor vector is calculated by using the same method as that for constructing the reference descriptor;

[0028] For each to-be-tested descriptor vector, the cosine similarity of the to-be-tested descriptor vector with all reference descriptor vectors in the digital reference comparison knowledge base is calculated;

[0029] Based on historical measurement data or experimental data, an optimal similarity threshold is determined by maximizing the Youden index;

[0030] It is judged whether the maximum similarity value of each to-be-tested descriptor vector is greater than the similarity threshold. If yes, the to-be-tested point cloud cluster is identified as the component corresponding to the reference descriptor with the highest similarity. Otherwise, the to-be-tested point cloud cluster is marked as unknown or noise;

[0031] For each successfully matched component pair, a vector obtained by subtracting the centroid coordinates of the reference point cloud cluster from the centroid coordinates of the to-be-tested point cloud cluster is taken as a translation vector for preliminarily aligning the to-be-tested point cloud cluster to the reference position;

[0032] Based on the correspondence between the to-be-tested point cloud cluster and the reference point cloud cluster in the local coordinate system, the ICP algorithm is used to accurately align the to-be-tested point cloud cluster preliminarily translated with the reference point cloud cluster.

[0033] Preferably, the extraction of the point, surface and line reference features and the calculation of the size and geometric and position tolerances of the steering wheel assembly based on the spatially aligned static point cloud model and the digital reference comparison knowledge base specifically comprises:

[0034] Based on the successfully identified and aligned component point cloud cluster, the theoretical associated detection features and their theoretical geometric parameters corresponding to the ID of the component are retrieved from the digital reference comparison knowledge base;

[0035] The local coordinate system of the aligned to-be-tested point cloud cluster is taken as a reference, and the retrieved theoretical associated detection features are mapped to the to-be-tested static point cloud model to determine the target fitting regions of the features;

[0036] The range of the target fitting region is set based on the theoretical predicted position and size of the feature and is expanded outward by a preset three-dimensional tolerance, and the tolerance value is determined by experiment: the known reference features are repeatedly measured under fixed conditions, the standard deviation σ of the feature position measurement results is calculated, and the tolerance is set as several times of σ;

[0037] According to the target fitting region of each to-be-measured point cloud cluster, actual correlation detection features of each to-be-measured point cloud cluster are fitted based on a least square method, and corresponding measured feature parameters are obtained;

[0038] Based on the measured feature parameters, measured values of each key size are calculated;

[0039] The measured values of each size are compared with corresponding theoretical size values retrieved from a digital reference comparison knowledge base, a difference value is calculated, and an actual size deviation is obtained;

[0040] The key sizes at least include linear sizes, curved surface sizes and gap sizes, the linear sizes at least include straight line distances between feature points and component lengths, the curved surface sizes at least include cylindrical radii and curved radii, and the gap sizes at least include vertical distances between adjacent component feature surfaces;

[0041] Based on spatial geometric relations represented by the measured feature parameters and the retrieved theoretical geometric parameters, form and position tolerances are calculated, and the form and position tolerances at least include parallelism, perpendicularity, roundness / cylindricity and coaxiality;

[0042] Results of all sizes and form and position tolerances are compared with preset qualified tolerance thresholds, a detection report is generated, qualified items or unqualified items are marked, and deviation values, exceeding component and corresponding reference feature information are output.

[0043] Preferably, based on the static measurement method of the steering wheel assembly sizes and form and position tolerances, a dynamic detection model is constructed, and dynamic performance indexes are calculated, and the dynamic performance indexes specifically include:

[0044] The to-be-measured steering wheel assembly is controlled to step rotate around a designed rotation axis at preset angle intervals, at each rotation angle position, a static three-dimensional point cloud in this posture is collected, and a series of time sequence point cloud frames are formed;

[0045] For each collected point cloud frame, the static measurement method based on the steering wheel assembly sizes and form and position tolerances is independently executed, and measured feature parameters of all associated detection features in this frame are obtained;

[0046] For the same associated detection feature, measured feature parameters of the feature in all frames are extracted, arranged in an angle order, and a motion trajectory of the feature in a three-dimensional space is reconstructed;

[0047] Based on a theoretical motion law of the feature in design, an ideal motion trajectory of the feature is determined;

[0048] Based on the reconstructed motion trajectory and the ideal motion trajectory, dynamic performance indexes of the feature are calculated, and the dynamic performance indexes specifically include:

[0049] Rotation run-out: the maximum radial deviation of the actual motion trajectory of the feature point from its ideal rotation cylindrical surface;

[0050] Axial run-out: the displacement fluctuation of the feature point along the ideal rotation axis direction;

[0051] Angular positioning error: the deviation between the actual rotation angle of the steering wheel obtained by the high-precision angle sensor and the command angle;

[0052] Gap dynamic change: the change range and periodicity of the gap size between specific components when the steering wheel is turned.

[0053] Further, the scheme proposes a steering wheel assembly size measurement system based on multi-data fusion, which is used to realize the steering wheel assembly size measurement method based on multi-data fusion as described above, and includes:

[0054] A reference construction module is configured to construct a standard three-dimensional point cloud model of the steering wheel assembly, define and calibrate a set of reference features for size and geometric tolerance detection on the model, segment the standard three-dimensional point cloud model into K point cloud subsets based on the K-Means clustering algorithm according to the standard three-dimensional point cloud model, and construct a digital reference comparison knowledge base.

[0055] A data acquisition module is configured to acquire three-dimensional point cloud data of the to-be-measured steering wheel assembly in a static state through a three-dimensional scanning sensor, and generate a to-be-measured static point cloud model.

[0056] An alignment and measurement module is configured to perform identification and spatial alignment of the to-be-measured static point cloud model and the standard three-dimensional point cloud model based on each physical component through the digital reference comparison knowledge base, extract point, surface and line reference features based on the digital reference comparison knowledge base according to the spatially aligned static point cloud model, and calculate the size and geometric tolerance of the steering wheel assembly. Based on the static measurement method of the size and geometric tolerance of the steering wheel assembly, a dynamic detection model is constructed, and its dynamic performance indicators are calculated.

[0057] Preferably, the reference construction module includes:

[0058] A reference feature unit is configured to construct a standard three-dimensional point cloud model of the steering wheel assembly, and define and calibrate a set of reference features for size and geometric tolerance detection on the model.

[0059] A knowledge base unit is configured to segment the standard three-dimensional point cloud model into K point cloud subsets based on the K-Means clustering algorithm according to the standard three-dimensional point cloud model, and construct a digital reference comparison knowledge base.

[0060] Preferably, the alignment and measurement module includes:

[0061] a coordinate alignment unit configured to recognize and spatially align the to-be-tested static point cloud model and the standard three-dimensional point cloud model based on each physical component by means of the digital reference comparison knowledge base;

[0062] a static measurement unit configured to extract point, surface and line reference features and calculate the size and geometric and position tolerance of the steering wheel assembly based on the digital reference comparison knowledge base according to the spatially aligned static point cloud model;

[0063] a dynamic measurement unit configured to construct a dynamic detection model and calculate the dynamic performance index thereof based on the static measurement mode of the size and geometric and position tolerance of the steering wheel assembly.

[0064] Compared with the prior art, the present application has the following beneficial effects:

[0065] The present application provides a steering wheel assembly size measurement scheme based on multi-data fusion, which solves the intelligent mapping problem from unordered point cloud to engineering semantics by constructing a digital reference comparison knowledge base of fused component feature descriptors. Firstly, the present scheme has invariance for overall translation and rotation based on the descriptors of main direction and spatial distribution, fundamentally overcoming the global matching failure problem caused by environmental interference, point cloud loss or initial pose deviation in traditional methods, and realizing robust recognition and coarse positioning at the component level; secondly, after successfully recognizing and aligning the components, the present scheme performs accurate geometric fitting (such as least squares fitting of circles and surfaces) only in the reliable data neighborhood within a preset tolerance range defined by the point cloud cluster of the component, avoiding the interpolation error introduced by traditional methods to complete the point cloud, and ensuring high precision of size and geometric and position tolerance extraction; finally, the static detection framework is naturally extended to dynamic performance sequence analysis, and through reconstruction of the motion trajectory from multiple frames of static parameters and comparison with the standard trajectory, the full index, automatic and high-precision comprehensive detection of the size, geometric and position tolerance and dynamic rotation performance of the steering wheel assembly is realized. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 A flowchart of a steering wheel assembly size measurement method based on multi-data fusion according to the present application;

[0067] Figure 2 A flowchart of dividing the standard three-dimensional point cloud model into K point cloud subsets and constructing a digital reference comparison knowledge base according to the present application;

[0068] Figure 3 A flowchart of recognizing and spatially aligning the to-be-tested static point cloud model and the standard three-dimensional point cloud model based on each physical component according to the present application. DETAILED DESCRIPTION

[0069] The following description is presented to enable any person skilled in the art to practice the application as claimed. Preferred embodiments are presented in the following description only as examples and modifications will be readily apparent to those of ordinary skill in the art having the benefit of this disclosure.

[0070] Referring to Figure 1 As shown in the drawings, a steering wheel assembly size measurement method based on multi-data fusion includes:

[0071] A standard three-dimensional point cloud model of the steering wheel assembly is constructed, and a set of reference features for size and geometric tolerance detection is defined and calibrated on the model, including feature points, feature surfaces, and feature lines;

[0072] According to the standard three-dimensional point cloud model, the standard three-dimensional point cloud model is segmented into K point cloud subsets based on the K-Means clustering algorithm, and a digital reference comparison knowledge base is constructed;

[0073] Three-dimensional point cloud data of the static steering wheel assembly to be measured is collected by a three-dimensional scanning sensor to generate a static point cloud model to be measured;

[0074] Through the digital reference comparison knowledge base, the static point cloud model to be measured and the standard three-dimensional point cloud model are recognized and spatially aligned based on each physical component;

[0075] According to the spatially aligned static point cloud model, point, surface, and line reference features are extracted based on the digital reference comparison knowledge base, and the size and geometric tolerance of the steering wheel assembly are calculated;

[0076] Based on the static measurement method of the size and geometric tolerance of the steering wheel assembly, a dynamic detection model is constructed, and its dynamic performance indicators are calculated.

[0077] It can be explained that the scheme is based on the automatic measurement process of steering wheel assembly size based on multi-data fusion, which is divided into two core stages of benchmark preparation and online detection, and a full closed-loop detection system from digital model to physical entity is constructed. The first stage is the benchmark preparation stage, which aims to establish accurate and complete digital benchmark for automatic detection. A standard three-dimensional point cloud model is constructed by high-precision scanning or CAD model sampling, and all measured engineering features (feature points, feature surfaces, feature lines) and their theoretical geometric parameters are defined manually on the model to form a "task list" for detection. Based on K-Means clustering and feature-guided optimization, the standard three-dimensional point cloud model is segmented into point cloud subsets corresponding to physical components one by one, and the "fingerprint" of the space distribution structure of each component point cloud subset is calculated, i.e. the descriptor vector of the benchmark. Based on this, a digital benchmark comparison knowledge base is constructed, which is structured to store component identification, benchmark point cloud, descriptor vector and associated benchmark feature definition. This knowledge base is not only a "feature dictionary" for subsequent component identification, but also a "standard answer library" for all size and tolerance comparison, which lays a unique digital benchmark for online detection; The second stage is the online detection stage. This stage uses the digital benchmark comparison knowledge base to automatically measure the measured physical object. The complete measured static point cloud model of the steering wheel is generated by multi-view three-dimensional scanning, and the descriptor of the measured point cloud cluster is calculated and compared with the knowledge base to realize the component-level rapid identification and coarse positioning without pre-setting the positioning point. Then, the accurate spatial alignment is completed through rigid transformation, and the measured feature parameters are accurately extracted in the local reliable data neighborhood guided by the theoretical feature position in the aligned component point cloud cluster. Then, each size and geometric tolerance is calculated. On this basis, it is extended to dynamic performance evaluation. By controlling the step-by-step rotation of the steering wheel and repeating the above static measurement process, the multi-frame static parameter sequence is reconstructed into a motion trajectory, which is compared with the standard motion trajectory to calculate the dynamic performance indicators such as rotation runout and axial runout. Thus, the full-dimensional automatic precision detection from static size to dynamic performance is realized in a single system, and the full-index, automatic, high-precision comprehensive detection of steering wheel assembly size, geometric tolerance and dynamic rotation performance is realized.

[0078] The method for constructing a standard three-dimensional point cloud model of a steering wheel assembly and defining and calibrating a set of benchmark features for size and geometric tolerance detection on the model specifically includes:

[0079] A standard three-dimensional point cloud model of a steering wheel assembly is constructed by high-precision three-dimensional scanning of a standard physical steering wheel assembly, or discrete sampling of an existing high-precision steering wheel assembly CAD model.

[0080] According to the measurement requirements of the steering wheel assembly, a set of benchmark features of feature points, feature surfaces and feature lines are defined manually on the standard three-dimensional point cloud model.

[0081] The feature points include but are not limited to: geometric center points, assembly reference points, boundary extreme points, and position deviation of each component of the steering wheel is calculated by a spatial distance formula;

[0082] The feature surfaces include but are not limited to: mounting reference planes, spoke side surfaces, and rim cylindrical surfaces, and angle deviation of each component surface of the steering wheel is calculated by a normal vector angle formula;

[0083] The feature lines include but are not limited to: outer ring curve, shaft sleeve axis, and feature hole edge line, and size deviation of each component of the steering wheel is calculated by a straight line or curvature.

[0084] It can be explained that the size measurement data source of the steering wheel assembly in the scheme mainly depends on the point cloud data generated by the three-dimensional scanning sensor scanning, and the steering wheel assembly size is fitted through the point cloud data, so as to realize a fast and automatic measurement mode. The implementation basis of this mode is to need to set a contrast model in advance, that is, the standard three-dimensional point cloud model of the steering wheel assembly, and define and label the reference feature set of the feature points, feature surfaces and feature lines on the basis of the model, so as to cut and identify the huge three-dimensional point cloud data, thereby providing a basis for subsequent size fitting and geometric tolerance. This is the reference preparation stage of the automatic measurement of the steering wheel assembly size. Secondly, the point cloud data generated by the three-dimensional scanning sensor scanning is essentially a huge, disordered and semantic-free three-dimensional coordinate set. In order to realize automatic measurement, a set of translation rules from geometric point cloud to engineering semantics must be established for the computer. The core of this rule is to define and label a complete reference feature set (point, surface, line). It is further explained that the feature point is an abstraction and anchoring of the key position. For example, the mounting hole center feature point anchors the assembly position of the steering wheel and the steering column. In detection, the spatial Euclidean distance between the corresponding feature points on the measured part and the standard part can be directly and quantitatively judged to determine whether the position tolerance, concentricity and other position tolerances are qualified. The feature surface is the definition and reference of the surface orientation and shape of the part. For example, the mounting reference plane defines the ideal orientation of the end surface of the steering wheel. In detection, the angle between the normal vector of the measured surface and the standard feature surface can be calculated to evaluate the parallelism and perpendicularity, and the distance distribution of the point group on the measured surface to the standard plane can be analyzed to evaluate the flatness. The feature line is the description and quantification of the contour boundary and center axis. For example, the grip ring outer curve completely describes the contour of the steering wheel. By calculating the curve (such as fitting the minimum circumscribed circle diameter), the key contour dimension of the steering wheel can be obtained. The analysis of the smoothness of the curve and the deviation of the standard curve can evaluate the profile tolerance. The feature line converts the discrete boundary points into continuous and geometrically operable entities, which is the direct source of size and macro shape tolerance. Therefore, the predefinition of this set of point, surface and line reference feature set on the standard three-dimensional point cloud model is essentially the conversion of the size and geometric tolerance requirements on the engineering drawing into computer recognizable and calculable three-dimensional geometric elements, thereby completing the reference preparation stage of automatic detection, so that the subsequent purposeful cutting (finding the corresponding area) and identification (extracting the corresponding features) of the huge measured point cloud data can provide accurate and reliable basis for the final size fitting and geometric tolerance determination. The acquisition method of the three-dimensional digital model includes but is not limited to the original CAD model exported from the product data management (PDM) system, the digital model generated by high-precision three-dimensional scanning of the standard part after denoising and splicing processing, or the official design model provided by the cooperation party according to the confidentiality agreement.

[0085] Reference Figure 2As shown, the dividing the standard three-dimensional point cloud model into K point cloud subsets and constructing a digital reference matching knowledge base specifically includes:

[0086] According to the type and quantity of the components of the steering wheel assembly to be tested, the number K of clustering clusters is determined;

[0087] Based on the K-Means clustering algorithm, the data point set of the standard three-dimensional point cloud model is preliminarily aggregated to obtain K initial point cloud clusters;

[0088] According to the pre-defined reference feature set, the initial point cloud clusters are optimized and adjusted to obtain K component-level reference point cloud clusters, and each reference point cloud cluster accurately corresponds to a physical component;

[0089] For each reference point cloud cluster, the cluster center of each reference point cloud cluster is calculated;

[0090] Taking the cluster center as the origin, the covariance matrix of all points in the cluster relative to the origin is calculated, and the direction of the eigenvector with the largest eigenvalue is taken as the X axis, the direction of the eigenvector with the second largest eigenvalue is taken as the Y axis, and the cross product of the two is taken as the Z axis. A right-handed orthogonal local coordinate system is constructed;

[0091] In this local coordinate system, a cube bounding box with a side length of L is constructed with the origin as the center along the X, Y, and Z axes, and the cube space is uniformly divided into MxMxM three-dimensional voxel grids;

[0092] For each point in the cluster, its three-dimensional coordinates in the local coordinate system are calculated, and it is determined which specific voxel grid it falls into according to the three-dimensional coordinates;

[0093] The number of point clouds falling into each voxel grid is counted to form an MxMxM-dimensional histogram vector, and the reference descriptor vector is obtained by L2 norm normalization of the vector;

[0094] A digital reference matching knowledge base is constructed, and a separate storage record is created for each physical component;

[0095] Each record contains: component ID and name, reference point cloud cluster and its center coordinates, reference descriptor vector, associated detection features and their theoretical geometric parameters.

[0096] It can be explained that computers cannot directly identify different components of a steering wheel assembly from disordered 3D point cloud data. To solve this problem, this solution constructs a highly discriminative "fingerprint," or benchmark descriptor vector, for each component's point cloud. This vector uses the centroid of the component's point cloud as a spatial reference and encodes the local spatial structure features of the component by statistically analyzing the distribution density of its surrounding point clouds within a regular 3D grid with its principal direction as the axis. Since the calculation of the descriptor is entirely based on the component's own geometric properties (centroid and principal direction), it is invariant to the component's overall spatial translation and rotation. In subsequent online detection, the system only needs to calculate a similar descriptor for each suspected component cluster in the test point cloud. Then, by calculating the similarity (such as cosine similarity) with all descriptors in the benchmark knowledge base, it can achieve fast, coarse-grained identification and matching of components. The purpose of this step is not high-precision alignment, but to solve the semantic recognition problem of "which point cloud block corresponds to which component," laying the foundation for subsequent precise feature extraction for each component.

[0097] The expression for the centroid of each reference point cloud cluster is as follows:

[0098]

[0099] In the formula, For the first The three-dimensional centroid coordinates of a reference cloud cluster. For clusters Total number of points contained within For clusters The 3D coordinate vector of the point cloud at any point within it;

[0100] The three-dimensional coordinate expressions of each point within the cluster in the local coordinate system are as follows:

[0101]

[0102] In the formula, For point Three-dimensional coordinates in a local coordinate system Let X, Y, and Z be the three orthogonal unit vectors representing the main spatial distribution directions of the point cloud of the component, obtained through principal component analysis. The unit vectors of the local coordinate system's X, Y, and Z axes are denoted as follows: Right now The eigenvector corresponding to the largest eigenvalue, The eigenvector corresponding to the second largest eigenvalue, The cross product corresponds to the local coordinate system;

[0103] It should be noted that, in the initialization of the K-Means clustering algorithm, the initial cluster center can be roughly set by combining the three-dimensional size prior knowledge of the steering wheel to speed up the convergence of the K-Means clustering algorithm; secondly, the initial point cloud cluster is adjusted according to the feature points, feature surfaces and feature lines predefined according to the standard three-dimensional point cloud model, which can be realized through the following collaborative strategy:

[0104] For components with regular geometric features such as rims and spokes, the corresponding initial point cloud cluster is fitted with a parametric geometric model (for example, a cylindrical model is used to fit the steering wheel rim, and a plane model is used to fit the spoke side), and a distance tolerance threshold related to the model is set (the threshold can be set according to the point cloud measurement accuracy, model fitting residual or statistical analysis, for example, it can be set as several times of the standard deviation of the distance distribution of the points in the initial cluster to the model), and the optimization process takes the initial cluster point as the growth seed. In the entire standard point cloud, search and absorb all data points with a shortest spatial distance less than the tolerance threshold to the fitted model, this process constitutes a region growing with model consistency as the criterion, which can effectively correct the classification deviation of the initial clustering on the geometric boundary, and ensure the shape integrity and boundary accuracy of the component point cloud cluster;

[0105] The part attribution semantics contained in the predefined reference feature set (feature points, lines and surfaces) is used to logically verify the segmentation result, and each reference feature is traversed to check whether the initial cluster where the spatial position is located is consistent with the expected component to which it belongs. If not, perform feature-guided redivision: take the feature as the core, define a spatial neighborhood range (such as a sphere with the feature point as the center and a specified radius R), remove all point clouds in the neighborhood from the current misattributed cluster, and reassign them to the correct target component cluster. This strategy uses prior engineering knowledge as a strong constraint to directly correct semantic assignment errors that may occur based on pure geometric clustering;

[0106] The above automatically optimized segmentation result is presented through a visual interface, and with the reference of the standard three-dimensional point cloud model (i.e. known and certain geometric truth), the operator can perform final verification on the result. For the few abnormal points or fuzzy boundaries that are difficult to handle by the algorithm, manual fine-tuning can be performed through human-computer interaction tools.

[0107] Therefore, through the collaborative strategy, a final determined reference point cloud cluster set corresponding to each physical component can be output, and it is stored in association with the corresponding component identifier;

[0108] It needs to be further explained that the correlation detection feature refers to the point, surface, line reference feature of the component, and the theoretical geometric parameter refers to the key mathematical attribute derived from the geometric definition of each reference feature for subsequent quantitative calculation. The specific implementation case includes: the geometric parameter of the feature point is its three-dimensional space coordinate, the geometric parameter of the feature surface is the parameter describing its orientation and shape, such as the unit normal vector of the plane, the axis direction and radius of the cylindrical surface, and the geometric parameter of the feature line is the parameter describing its trend and contour, such as the direction vector of the straight line, the curvature distribution or control point set of the spatial curve. These parameters constitute the direct input and reference for size and position tolerance calculation.

[0109] The three-dimensional point cloud data of the static steering wheel assembly to be measured is collected by the three-dimensional scanning sensor, and a static point cloud model to be measured is generated, which specifically includes:

[0110] A plurality of optical positioning marker points are arranged on the surface of the steering wheel assembly to be measured.

[0111] Among them, the optical positioning marker point is usually a high-contrast circular light return target point, which serves as a reference for automatic multi-view data splicing by the scanner.

[0112] The three-dimensional scanning sensor is used to scan the steering wheel assembly to be measured with optical positioning marker points from multiple angles to obtain high-density three-dimensional point cloud data.

[0113] According to the optical positioning marker points, the point cloud data obtained by multiple scans is automatically spliced and fused based on the function of the three-dimensional scanning sensor to obtain a complete and unified static point cloud model to be measured.

[0114] It can be explained that after the digital reference comparison knowledge base is constructed, the scheme enters the online detection stage. The primary task of this stage is to obtain the actual three-dimensional geometric data of the product to be measured, i.e. the static point cloud model to be measured. This data is the only source for all subsequent calculations. The scheme uses three-dimensional scanning technology to realize data acquisition, i.e. by arranging marker points on the steering wheel surface and using a scanner to scan from multiple angles (usually with a built-in multi-view automatic splicing algorithm), the complete surface point cloud can be efficiently and accurately obtained and automatically combined into a seamless three-dimensional model, thereby ensuring the integrity of the input data. Since the model alignment is based on the matching and recognition of the geometric features of the component itself, the scheme does not need to accurately process complex installation backgrounds on the premise of ensuring the completeness of the steering wheel point cloud, thereby providing qualified input for subsequent intelligent comparison.

[0115] Referring to Figure 3 It is shown that the identification and spatial alignment of the static point cloud model to be measured and the standard three-dimensional point cloud model based on each physical component specifically includes:

[0116] According to the to-be-measured static point cloud model, a plurality of to-be-measured point cloud clusters are obtained by using the Euclidean clustering algorithm;

[0117] For each to-be-measured point cloud cluster, a to-be-measured descriptor vector is calculated by using the same method as that for constructing the reference descriptor;

[0118] For each to-be-measured descriptor vector, the cosine similarity between the to-be-measured descriptor vector and all reference descriptor vectors in the digital reference comparison knowledge base is calculated;

[0119] Based on historical measurement data or experimental data, an optimal similarity threshold is determined by maximizing the Youden index;

[0120] It is judged whether the maximum similarity value of each to-be-measured descriptor vector is greater than the similarity threshold, if yes, the to-be-measured point cloud cluster is identified as a component corresponding to the reference descriptor with the highest similarity, otherwise, the to-be-measured point cloud cluster is marked as unknown or noise;

[0121] For each successfully matched component pair, a vector obtained by subtracting the centroid coordinates of the reference point cloud cluster from the centroid coordinates of the to-be-measured point cloud cluster is calculated as a translation vector for preliminarily aligning the to-be-measured point cloud cluster to the reference position;

[0122] Based on the correspondence between the to-be-measured point cloud cluster and the reference point cloud cluster in the local coordinate system, the ICP algorithm is used to accurately align the to-be-measured point cloud cluster after preliminary translation with the reference point cloud cluster.

[0123] It can be explained that traditional size measurement often relies on pre-set accurate positioning points (such as feature hole center, bolt mounting surface) to realize component matching, but in the actual scanning scene, factors such as collection angle offset, light change, local occlusion (such as dust in steering wheel gap, occlusion of surrounding components) can easily lead to distortion of accurate positioning point extraction, and then introduce systematic measurement error. The scheme follows the construction logic of the reference descriptor, that is, the centroid, local coordinate system and spatial distribution histogram, generates the same descriptor for each to-be-measured point cloud cluster, calculates the cosine similarity of the reference descriptor in the digital reference comparison knowledge base without relying on any pre-set accurate positioning point, and quickly completes the semantic recognition of "which to-be-measured point cloud cluster corresponds to which reference component". At the same time, based on the reference component with the highest similarity, coarse positioning is realized. This process only focuses on the local spatial structure features of the component itself, and is independent of the scanning environment and installation posture, thereby avoiding the error caused by inaccurate setting of the accurate positioning point. Secondly, the coarse positioning stage only solves the "component correspondence relationship" problem, and does not eliminate the translation and rotation deviation of the to-be-measured component and the reference component, so it cannot be directly used for size calculation. Therefore, based on coarse positioning, the three-dimensional spatial centroid coordinates of the successfully matched to-be-measured point cloud cluster and the reference point cloud cluster are used to calculate the preliminary translation vector of the to-be-measured point cloud cluster relative to the reference point cloud cluster, quickly eliminating the global translation deviation between the two clusters, laying a foundation for subsequent fine alignment, and based on the rigid body transformation algorithm (only containing rotation and translation, not changing the geometric shape and size of the component, completely adapting to the characteristics of the steering wheel rigid component), the fine alignment of the to-be-measured point cloud cluster and the reference point cloud cluster is realized. Avoid the dependence of traditional methods on strict matching of to-be-measured point cloud coordinates and pre-set three-dimensional digital model coordinates, and avoid matching failure due to excessive global deviation. The scheme can still be corrected through rigid body transformation even if the to-be-measured steering wheel installation posture has a slight deviation, greatly reducing the constraint on the installation environment. It needs to be further explained that when standard point cloud segmentation, K value can be determined by the number of physical components of the steering wheel, but due to factors such as scanning noise (discrete points generated by scanning device precision fluctuation), component fracture (point cloud discontinuity caused by local occlusion of the grip ring), and small attached debris (steering wheel surface stickers and stains), additional "pseudo clusters" (noise clusters) or "fracture clusters" (the same component is segmented into multiple sub-clusters) may be generated, resulting in the actual number of clusters being greater than or equal to the number of standard components K. Using the Euclidean clustering does not need to specify the number of clusters, but only by setting a distance threshold (if the distance between adjacent point clouds is less than the threshold, they are considered to belong to the same cluster), all continuous point cloud blocks can be adaptively segmented, thereby adapting to the uncertainty of the to-be-measured point cloud. The distance threshold is usually set to 2 to 5 times the average point distance of the three-dimensional point cloud data and the surface features of the to-be-measured object.

[0124] The digital reference comparison knowledge base based on the extraction of point, surface and line reference features, and the calculation of the size and geometric and positional tolerance of the steering wheel assembly specifically includes:

[0125] Based on the successfully identified and aligned part point cloud clusters, according to the part ID matched thereto, the theoretical associated detection features corresponding to the ID and the theoretical geometric parameters thereof are retrieved from the digital reference comparison knowledge base;

[0126] With the local coordinate system of the aligned to-be-measured point cloud cluster as a reference, the retrieved theoretical associated detection features are mapped into the to-be-measured static point cloud model to determine the target fitting region of each feature;

[0127] The target fitting region range is set based on the theoretical predicted position and size of the feature and is expanded outward by a preset three-dimensional tolerance, and the tolerance value is determined through experiments: the known reference features are repeatedly measured under fixed conditions, the standard deviation σ of the feature position measurement results is counted, and the tolerance is set as several times of σ;

[0128] According to the target fitting region of each to-be-measured point cloud cluster, the actual associated detection features of each to-be-measured point cloud cluster are fitted based on the least square method, and the corresponding measured feature parameters are obtained;

[0129] Based on the measured feature parameters, the measured values of each key dimension are calculated;

[0130] The measured values of each dimension are compared with the corresponding theoretical size values retrieved from the digital reference comparison knowledge base, the difference values are calculated, and the actual size deviation is obtained;

[0131] The key dimensions at least include linear dimensions, curved surface dimensions and gap dimensions, the linear dimensions at least include straight line distances between feature points and part lengths, the curved surface dimensions at least include cylindrical radii and curve radii, and the gap dimensions at least include perpendicular distances between adjacent part feature surfaces;

[0132] Based on the spatial geometric relationship represented by the measured feature parameters and the retrieved theoretical geometric parameters, the form and position tolerances are calculated, and the form and position tolerances at least include parallelism, perpendicularity, roundness / cylindricity and coaxiality;

[0133] The results of all dimensions and form and position tolerances are compared with the preset qualified tolerance threshold, a detection report is generated, qualified items or unqualified items are marked, and deviation values, exceeding parts and corresponding reference feature information are output.

[0134] It can be explained that after the components of the steering wheel assembly are precisely aligned through the ICP algorithm, the spatial poses of the to-be-measured point cloud clusters and the reference point cloud clusters are highly consistent, but the point cloud data itself is still an unordered coordinate set, and the conversion of geometric data to engineering semantics needs to be realized through pre-defined associated detection features (points, surfaces, lines) to complete the quantitative calculation of size and geometric tolerance. The scheme retrieves the corresponding theoretical associated detection features and theoretical geometric parameters from the digital reference comparison knowledge base through the component ID, and completes feature mapping with the help of the aligned local coordinate system to ensure accurate matching of the theoretical features and the spatial position of the to-be-measured point cloud clusters, providing clear targets for subsequent fitting. After obtaining the measured feature parameters, the size of the steering wheel assembly is quantified through the measured feature parameters, that is, the linear size measured value can be obtained by calculating the Euclidean distance between two measured feature point coordinates; the curved surface size measured value can be obtained by extracting the radius parameter from the fitting equation of the measured feature surface (such as the cylindrical surface equation) or by performing circle fitting on the measured feature line (such as the contour line) to obtain the radius or diameter; the gap size measured value can be obtained by calculating the shortest perpendicular distance between the measured feature surfaces of two different components. The qualified tolerance threshold is set according to the size and geometric tolerance marked on the design drawing of the steering wheel assembly. It needs to be further explained that the core principle and logic of fitting the actual associated detection features based on the least squares method are as follows:

[0135] Point feature fitting: The "geometric center point" (such as the center of the mounting hole) in the theoretical associated detection features corresponds to the cluster of points in the target fitting region of the to-be-measured point cloud cluster. For such features, the "spatial circle / sphere fitting" algorithm based on least squares is usually used. With the spatial coordinates of the theoretical feature points as the initial reference, the least squares method is used to minimize the sum of the squared geometric distances of all points in the region to the fitted circle (or sphere), and the solved center (or sphere) coordinates are the high-precision actual feature points. This process directly relies on the initial value and constraint provided by the theoretical parameters to accurately extract the measured point position through error minimization logic;

[0136] Surface feature fitting: For the "plane / cylinder" (such as the spoke side surface, rim cylinder) in the theoretical associated detection features, for a plane, the standard "least squares plane fitting" algorithm is used to directly solve the optimal plane equation. For a cylindrical surface, the "nonlinear least squares fitting" algorithm (such as the Gauss-Newton method) is used. The initial value of the fitting model is set based on the theoretical geometric parameters (such as the plane normal vector, cylinder axis direction) of the theoretical feature surface, and then the squared distances of each point in the to-be-measured point cloud to the model surface are minimized through iterative optimization to solve the accurate parameters of the model (such as the plane equation, cylinder radius and axis equation).

[0137] Line feature fitting: for detecting "straight line / curve" in theoretical correlation features (such as shaft sleeve axis, grip ring outer edge curve), for straight line, "least square space straight line fitting" is adopted, for complex space curve, "least square B-spline curve fitting" is adopted, taking the theoretical geometric parameters (such as direction vector or control point) of the theoretical feature line as the initial constraint, the least square method is used to minimize the sum of squares of the distance from each point in the measured point cloud to the target line, and the accurate equation of the straight line or the control point parameters of the B-spline curve is fitted;

[0138] Specific implementation cases are listed to further illustrate how to calculate the corresponding dimensional tolerance and geometric tolerance through theoretical and actual correlation detection features:

[0139] Dimensional tolerance calculation: difference value operation is performed on the core parameters of the measured and theoretical features through basic geometric formulas, for example: the difference between the measured center distance of the mounting hole (calculated based on the coordinates of two actual feature points) and the theoretical center distance is the linear dimensional deviation; the difference between the measured radius of the rim cylinder (based on the actual surface feature fitting parameters) and the theoretical radius is the curved surface dimensional deviation; the difference between the measured vertical distance of the center cover plate and the feature surface of the grip ring (calculated based on the normal vectors and equations of two actual surface features) and the theoretical gap is the gap dimensional deviation;

[0140] Geometric tolerance calculation: the difference between the actual and theoretical features is quantified by analyzing the spatial attitude or shape difference, for example: the angle between the normal vector of the actual spoke side surface (surface feature) and the theoretical normal vector is the parallelism deviation; the angle between the actual mounting hole axis (line feature) and the normal vector of the theoretical reference surface is the perpendicularity deviation; the maximum distance from each point on the actual rim inner circle (surface feature) to the theoretical cylindrical surface is the roundness deviation; the distance from multiple actual mounting hole axes to the theoretical reference axis is the coaxiality deviation;

[0141] The theoretical dimensional value and the theoretical geometric parameter are both derived from the theoretical correlation detection feature definition stored in the digital reference comparison knowledge base.

[0142] The static measurement method based on the dimensional and geometric tolerance of the steering wheel assembly constructs a dynamic detection model, and calculates its dynamic performance indicators, which specifically include:

[0143] The measured steering wheel assembly is controlled to rotate around its designed rotation axis at a preset angle interval, and at each rotation angle position, a static three-dimensional point cloud at that attitude is collected to form a series of time sequence point cloud frames;

[0144] For each collected point cloud frame, the static measurement method based on the dimensional and geometric tolerance of the steering wheel assembly is independently executed to obtain the measured feature parameters of all the correlation detection features in that frame;

[0145] For the same correlation detection feature, the measured feature parameters in all frames are extracted, arranged in angle order, and the motion trajectory of the feature in three-dimensional space is reconstructed;

[0146] Based on the theoretical motion law of the feature in design, the ideal motion trajectory is determined;

[0147] Based on the reconstructed motion trajectory and the ideal motion trajectory, the dynamic performance indicators are calculated, including:

[0148] Rotary runout: the maximum radial deviation of the actual motion trajectory of the feature point to the ideal rotary cylindrical surface;

[0149] Axial movement: the displacement fluctuation of the feature point along the ideal rotary axis;

[0150] Angular positioning error: the deviation between the actual rotation angle of the steering wheel obtained by the high-precision angle sensor and the command angle;

[0151] Gap dynamic change: the change range and periodicity of the gap size between the specific components when the steering wheel rotates.

[0152] It can be explained that through the static measurement method based on the size and geometric tolerance of the steering wheel assembly, the static size and static geometric tolerance of the steering wheel assembly can be effectively obtained, but the dynamic performance indicators in the actual rotation process of the steering wheel are lacking, and the dynamic performance indicators can be quantified by analyzing several frames of static measured feature parameters, which is reflected in: accurate component recognition and alignment ability, that is, to ensure that the same component and feature can be stably found and positioned in each frame, which is the premise of sequence comparison; high-precision feature parameterization ability, that is, to convert complex point cloud into accurate and mathematically comparable geometric parameters (coordinates, normal vectors, radii), providing data basis for trajectory calculation; digital reference comparison knowledge base, that is, to provide theoretical rotary axis, ideal motion trajectory and other dynamic evaluation references, wherein the theoretical rotary axis and ideal motion trajectory are obtained by rotating the standard steering wheel assembly around the shaft and collecting its standard point cloud sequence, and the motion trajectory of the reference feature points in the sequence is fitted with a spatial cylinder.

[0153] Further, based on the same inventive concept as the above-mentioned steering wheel assembly size measurement method based on multi-data fusion, the present scheme proposes a steering wheel assembly size measurement system based on multi-data fusion, comprising:

[0154] The reference construction module is used to construct a standard three-dimensional point cloud model of the steering wheel assembly, and define and calibrate a set of reference features for size and geometric tolerance detection on the model; according to the standard three-dimensional point cloud model, based on the K-Means clustering algorithm, the standard three-dimensional point cloud model is segmented into K point cloud subsets, and a digital reference comparison knowledge base is constructed;

[0155] a data acquisition module, configured to acquire three-dimensional point cloud data of a to-be-tested steering wheel assembly in a static state through a three-dimensional scanning sensor, and generate a to-be-tested static point cloud model;

[0156] an alignment and measurement module, configured to perform identification and spatial alignment of the to-be-tested static point cloud model and a standard three-dimensional point cloud model based on respective physical components through a digital reference comparison knowledge base, extract point, surface and line reference features based on the digital reference comparison knowledge base according to the spatially aligned static point cloud model, and calculate the size and geometric and position tolerances of the steering wheel assembly, and construct a dynamic detection model based on a static measurement mode of the size and geometric and position tolerances of the steering wheel assembly, and calculate dynamic performance indexes thereof;

[0157] the reference construction module comprises:

[0158] a reference feature unit, configured to construct a standard three-dimensional point cloud model of the steering wheel assembly, and define and calibrate a set of reference features for size and geometric and position tolerance detection on the model;

[0159] a knowledge base unit, configured to segment the standard three-dimensional point cloud model into K point cloud subsets based on a K-Means clustering algorithm according to the standard three-dimensional point cloud model, and construct a digital reference comparison knowledge base;

[0160] the alignment and measurement module comprises:

[0161] a coordinate alignment unit, configured to perform identification and spatial alignment of the to-be-tested static point cloud model and the standard three-dimensional point cloud model based on respective physical components through the digital reference comparison knowledge base;

[0162] a static measurement unit, configured to extract point, surface and line reference features based on the digital reference comparison knowledge base according to the spatially aligned static point cloud model, and calculate the size and geometric and position tolerances of the steering wheel assembly;

[0163] a dynamic measurement unit, configured to construct a dynamic detection model based on a static measurement mode of the size and geometric and position tolerances of the steering wheel assembly, and calculate dynamic performance indexes thereof.

[0164] In summary, the present application has the advantage that by constructing a digital reference comparison knowledge base that fuses component-level feature descriptors, intelligent identification and high-precision alignment of a steering wheel assembly from unordered point clouds to engineering semantics are achieved, so that comprehensive precision detection of full size, geometric and position tolerances and dynamic performance is automatically and robustly completed in a single system.

[0165] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A steering wheel assembly dimensional measurement method based on multi-data fusion, characterized in that, The method comprises the following steps: Construct a standard three-dimensional point cloud model of the steering wheel assembly, and define and calibrate a set of reference features for size and geometric tolerance detection on the model, including feature points, feature surfaces and feature lines; According to the standard three-dimensional point cloud model, the standard three-dimensional point cloud model is segmented into K point cloud subsets based on the K-Means clustering algorithm, and a digital reference comparison knowledge base is constructed; Collect three-dimensional point cloud data of the static steering wheel assembly to be measured by a three-dimensional scanning sensor, and generate a static point cloud model to be measured; Through the digital reference comparison knowledge base, the static point cloud model to be measured is compared with the standard three-dimensional point cloud model based on the identification and spatial alignment of each physical component; According to the spatially aligned static point cloud model, the digital reference comparison knowledge base is used to extract point, surface and line reference features, and calculate the size and geometric tolerance of the steering wheel assembly; Based on the static measurement method of the size and geometric tolerance of the steering wheel assembly, a dynamic detection model is constructed, and the dynamic performance index is calculated.

2. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 1, characterized in that, The method of constructing a standard three-dimensional point cloud model of a steering wheel assembly, and defining and calibrating a set of reference features for size and geometric tolerance detection on the model comprises the following steps: Construct a standard three-dimensional point cloud model of the steering wheel assembly by high-precision three-dimensional scanning of the standard physical steering wheel assembly, or discretize and sample the existing high-precision steering wheel assembly CAD model to construct a standard three-dimensional point cloud model of the steering wheel assembly; According to the measurement requirements of the steering wheel assembly, define a set of reference features of feature points, feature surfaces and feature lines on the standard three-dimensional point cloud model with the help of artificial assistance; The feature points include but are not limited to geometric center points, assembly reference points and boundary extreme points, and the position deviation of each component of the steering wheel is calculated by a spatial distance formula; The feature surfaces include but are not limited to mounting reference planes, spoke side surfaces and rim cylindrical surfaces, and the angle deviation of each component of the steering wheel is calculated by a normal vector angle formula; The feature lines include but are not limited to grip ring outer curve, shaft sleeve axis and feature hole edge line, and the size deviation of each component of the steering wheel is calculated by a straight line or curvature.

3. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 2, characterized in that, The method of constructing a standard three-dimensional point cloud model of a steering wheel assembly, and defining and calibrating a set of reference features for size and geometric tolerance detection on the model comprises the following steps: Determine the number K of clustering clusters according to the type and number of components of the steering wheel assembly to be measured; Based on the K-Means clustering algorithm, the data point set of the standard three-dimensional point cloud model is preliminarily aggregated to obtain K initial point cloud clusters; According to the pre-defined set of reference features, the initial point cloud clusters are optimized and adjusted to obtain K component-level reference point cloud clusters, and each reference point cloud cluster accurately corresponds to a physical component; For each reference point cloud cluster, calculate the cluster centroid of each reference point cloud cluster; Take the cluster centroid as the origin, calculate the covariance matrix of all points in the cluster relative to the origin, and take the direction of the largest eigenvalue as the X axis, the second largest eigenvalue as the Y axis, and the cross product of the two as the Z axis to construct a right-handed orthogonal local coordinate system; In this local coordinate system, a cube bounding box with a side length of L is constructed with the origin as the center along the X, Y, Z axis directions, and the cube space is evenly divided into MxMxM three-dimensional voxel grids; For each point in the cluster, calculate its three-dimensional coordinates in the local coordinate system, and determine which specific voxel grid it falls into according to the three-dimensional coordinates; Statistical point cloud quantity falling into each voxel grid, form a MxMxM dimension histogram vector, and obtain the reference descriptor vector by L2 norm normalization of the vector; A digital reference comparison knowledge base is constructed, and a separate storage record is created for each physical component; Each record contains: component ID and name, reference point cloud cluster and its centroid coordinates, reference descriptor vector, associated detection features and their theoretical geometric parameters.

4. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 3, characterized in that, The three-dimensional point cloud data of the static steering wheel assembly to be measured is collected by the three-dimensional scanning sensor, and a static point cloud model to be measured is generated, which specifically includes: A plurality of optical positioning marker points are arranged on the surface of the steering wheel assembly to be measured; The optical positioning marker points are usually high-contrast circular light-reflecting target points, which serve as reference bases for automatic multi-view data splicing by the scanner; The three-dimensional scanning sensor is used to scan the steering wheel assembly to be measured with optical positioning marker points from multiple angles to obtain high-density three-dimensional point cloud data; Based on the three-dimensional scanning sensor function, the point cloud data obtained by multiple scans is automatically spliced and fused based on the optical positioning marker points to obtain a complete and unified static point cloud model to be measured.

5. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 4, characterized in that, The digital reference comparison knowledge base is used to identify and spatially align the static point cloud model to be measured with the standard three-dimensional point cloud model based on each physical component, which specifically includes: A plurality of point cloud clusters to be measured are obtained using the Euclidean clustering algorithm based on the static point cloud model to be measured; For each point cloud cluster to be measured, a measured descriptor vector is calculated in the same way as the reference descriptor; For each measured descriptor vector, the cosine similarity between it and all reference descriptor vectors in the digital reference comparison knowledge base is calculated; Based on historical measurement data or experimental data, an optimal similarity threshold is determined by maximizing the Youden index; Determine whether the maximum similarity value of each measured descriptor vector is greater than the similarity threshold, if yes, identify the point cloud cluster to be measured as the component corresponding to the reference descriptor with the highest similarity, otherwise, mark the point cloud cluster to be measured as unknown or noise; For each successfully matched component pair, the vector obtained by subtracting the centroid coordinates of the reference point cloud cluster from the centroid coordinates of the point cloud cluster to be measured is used as the translation vector for preliminary alignment of the point cloud cluster to be measured to the reference position. Based on the correspondence between the point cloud cluster to be measured and the reference point cloud cluster in the local coordinate system, the ICP algorithm is used to accurately align the point cloud cluster to be measured after preliminary translation with the reference point cloud cluster.

6. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 5, characterized in that, Based on the digital reference comparison knowledge base, the point, surface and line reference features are extracted from the spatially aligned static point cloud model, and the size and geometric tolerance of the steering wheel assembly are calculated. Based on the successfully identified and aligned part point cloud clusters, according to the part ID matched, the theoretical associated detection features and their theoretical geometric parameters corresponding to the ID are retrieved from the digital reference comparison knowledge base; Taking the local coordinate system of the aligned to-be-measured point cloud cluster as a reference, the retrieved theoretical associated detection features are mapped into the to-be-measured static point cloud model to determine the target fitting region of each feature; The range of the target fitting region is set based on the theoretical predicted position and size of the feature and is extended outward by a preset three-dimensional tolerance, and the tolerance value is determined through experiments: under fixed conditions, the known reference features are repeatedly measured, the standard deviation σ of the feature position measurement results is calculated, and the tolerance is set to several times of σ; According to the target fitting region of each to-be-measured point cloud cluster, the actual associated detection features of each to-be-measured point cloud cluster are fitted based on the least square method, and the corresponding measured feature parameters are obtained; Based on the measured feature parameters, the measured values of each key dimension are calculated; The measured values of each dimension are compared with the corresponding theoretical size values retrieved from the digital reference comparison knowledge base, the difference values are calculated, and the actual size deviation is obtained; The key dimensions at least include linear dimensions, curved surface dimensions and gap dimensions, the linear dimensions at least include straight line distances between feature points and part lengths, the curved surface dimensions at least include cylindrical radii and curve radii, and the gap dimensions at least include perpendicular distances between adjacent part feature surfaces; Based on the spatial geometric relationship represented by the measured feature parameters and the retrieved theoretical geometric parameters, the geometric tolerance is calculated, and the geometric tolerance at least includes parallelism, perpendicularity, roundness / cylindricity and coaxiality; The results of all dimensions and geometric tolerances are compared with the preset qualified tolerance threshold, a detection report is generated, qualified items or unqualified items are marked, and the deviation values, exceeding parts and corresponding reference feature information are output.

7. A steering wheel assembly dimensional measurement method based on multi-data fusion according to claim 6, characterized in that, The static measurement method based on the size and geometric tolerance of the steering wheel assembly is used to construct a dynamic detection model, and the dynamic performance indicators are calculated, which specifically include: The to-be-measured steering wheel assembly is controlled to step rotate around its design rotation axis at preset angle intervals, and at each rotation angle position, a static three-dimensional point cloud in this posture is collected to form a series of time sequence point cloud frames; For each collected point cloud frame, the static measurement method based on the size and geometric tolerance of the steering wheel assembly is independently executed to obtain the measured feature parameters of all associated detection features in this frame; For the same associated detection feature, the measured feature parameters of the feature in all frames are extracted, arranged in order of angle, and the motion trajectory of the feature in the three-dimensional space is reconstructed; Based on the theoretical motion law of the feature in design, the ideal motion trajectory of the feature is determined; Based on the reconstructed motion trajectory and the ideal motion trajectory, the dynamic performance indicators of the feature are calculated, which specifically include: Rotation runout: the radial maximum deviation of the actual motion trajectory of the feature point to the ideal rotation cylindrical surface; Axial runout: the displacement fluctuation of the feature point along the ideal rotation axis direction; Angle positioning error: the deviation between the actual rotation angle of the steering wheel obtained by a high-precision angle sensor and the command angle; Gap dynamic change: the change range and periodicity of the gap dimension between specific parts when the steering wheel rotates.

8. A steering wheel assembly dimensional measurement system based on multi-data fusion, characterized by, A method for realizing a steering wheel assembly size measurement based on multi-data fusion according to any one of claims 1-7, comprising: a reference construction module for constructing a standard three-dimensional point cloud model of the steering wheel assembly, and defining and calibrating a set of reference features for size and geometric tolerance detection on the model; according to the standard three-dimensional point cloud model, the standard three-dimensional point cloud model is segmented into K point cloud subsets based on a K-Means clustering algorithm, and a digital reference comparison knowledge base is constructed; a data acquisition module for acquiring three-dimensional point cloud data of a static steering wheel assembly to be measured by a three-dimensional scanning sensor, and generating a static point cloud model to be measured; an alignment and measurement module for identifying and spatially aligning the static point cloud model to be measured with the standard three-dimensional point cloud model based on each physical component through the digital reference comparison knowledge base; according to the spatially aligned static point cloud model, the point, surface and line reference features are extracted based on the digital reference comparison knowledge base, and the size and geometric tolerance of the steering wheel assembly are calculated; based on the static measurement method of the size and geometric tolerance of the steering wheel assembly, a dynamic detection model is constructed, and its dynamic performance indicators are calculated.

9. A steering wheel assembly dimensional measurement system based on multi-data fusion according to claim 8, wherein, The reference construction module comprises: a reference feature unit for constructing a standard three-dimensional point cloud model of the steering wheel assembly, and defining and calibrating a set of reference features for size and geometric tolerance detection on the model; a knowledge base unit for segmenting the standard three-dimensional point cloud model into K point cloud subsets based on a K-Means clustering algorithm according to the standard three-dimensional point cloud model, and constructing a digital reference comparison knowledge base.

10. A steering wheel assembly dimensional measurement system based on multi-data fusion according to claim 9, wherein, The alignment and measurement module comprises: a coordinate alignment unit for identifying and spatially aligning the static point cloud model to be measured with the standard three-dimensional point cloud model based on each physical component through the digital reference comparison knowledge base; a static measurement unit for extracting point, surface and line reference features based on the digital reference comparison knowledge base according to the spatially aligned static point cloud model, and calculating the size and geometric tolerance of the steering wheel assembly; a dynamic measurement unit for constructing a dynamic detection model based on the static measurement method of the size and geometric tolerance of the steering wheel assembly, and calculating its dynamic performance indicators.

Images