Coating thickness measuring method and system based on multi-source sensor

Abstract

The invention relates to the technical field of three-dimensional measurement, and discloses a coating thickness measurement method and system based on a multi-source sensor, and the method comprises the steps: obtaining and fusing point cloud data of the multi-source sensor, and constructing a three-dimensional curved surface model of a component through curvature analysis and normal vector estimation; calculating an optimal incident angle of the sensor relative to each area of the curved surface based on the model, and generating a dynamic attitude adjustment parameter; according to the parameters, the three-axis attitude of the sensor is adjusted in real time for signal acquisition, and original measurement data are obtained; frequency domain analysis is carried out on the original data to diagnose reflection distortion, and correction is carried out in combination with time domain filtering and spatial domain smoothing; and finally, calculating a thickness value in combination with the three-dimensional model, detecting and repairing a data missing region by using Fourier transform and an interpolation algorithm, and generating a complete thickness distribution diagram. According to the invention, high-precision and automatic measurement of the thickness of the complex curved surface coating is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional measurement, in particular to a coating thickness measurement method and system based on multi-source sensors. BACKGROUND

[0002] In the modern high-end manufacturing field, such as aerospace, precision mold and shipbuilding, the coating thickness of the surface of the component is a key quality index to determine its corrosion resistance, wear resistance and overall service life. Therefore, accurate and efficient measurement of the coating thickness is of great significance to ensure product performance and optimize production processes.

[0003] At present, an existing technical solution for automatically measuring the coating thickness is to mount a single measurement probe (for example, a distance sensor based on laser triangulation) on a multi-axis robot, and to pre-plan a fixed motion trajectory according to the computer-aided design (CAD) model of the component to be measured. During the measurement process, the control system drives the robot to move strictly according to the preset trajectory, so that the distance sensor scans along the surface of the component, and the coating thickness is calculated by collecting the distance change between the sensor and the surface of the component.

[0004] However, the above-mentioned existing technical solution has inherent technical defects when applied to complex curved surface components. First of all, this solution highly depends on the theoretical CAD model, and the pre-set motion trajectory cannot adapt to the geometric tolerance or slight deformation of the actual component caused by processing, assembly or heat treatment, etc. When there is a deviation between the actual curved surface and the theoretical model, the robot still executes the rigid trajectory, which causes the sensor to be unable to always maintain the optimal measurement position and angle, especially in areas with dramatic changes in curvature, the deviation of the measurement angle will cause signal reflection distortion, which seriously affects the measurement accuracy. Secondly, this solution lacks real-time environmental perception and feedback adjustment capability. It only passively executes the preset instructions and cannot dynamically optimize the attitude of the sensor according to the actual surface profile of the component. Once the data is distorted or missing due to poor angle, the system cannot correct itself, resulting in insufficient reliability of the measurement results. Therefore, how to overcome the dependence on ideal models and realize accurate perception and adaptive measurement of the real appearance of complex curved surfaces is a technical problem to be solved in the current technical field. SUMMARY

[0005] The present application provides a coating thickness measurement method and system based on multi-source sensors to realize high-precision and automatic measurement of the coating thickness of complex curved surfaces.

[0006] In a first aspect, to solve the above technical problems, the present application provides a coating thickness measurement method based on multi-source sensors, comprising: obtaining initial three-dimensional point cloud data; According to the initial three-dimensional point cloud data, a three-dimensional curved surface model is obtained through curvature distribution analysis and normal vector estimation; According to the three-dimensional curved surface model, the optimal incident angle of the multi-source sensor relative to each region of the curved surface is calculated, and a dynamic attitude adjustment parameter is calculated according to the optimal incident angle; According to the dynamic attitude adjustment parameter, the three-axis attitude and the data acquisition frequency of the multi-source sensor are adjusted in real time, and a coating thickness related signal is collected to obtain original measurement data; The original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if there is, time domain filtering and spatial domain smoothing are performed to obtain corrected measurement data; Combined with the three-dimensional curved surface model and the corrected measurement data, the coating thickness value of each measurement point is calculated to obtain preliminary thickness distribution data; For the preliminary thickness distribution data, Fourier transform is used to detect whether there is a data missing area, and if a missing area is found, the preliminary thickness distribution data is interpolated to generate a complete thickness distribution map.

[0007] Preferably, the three-dimensional curved surface model is obtained according to the initial three-dimensional point cloud data through curvature distribution analysis and normal vector estimation, comprising: The initial three-dimensional point cloud data is subjected to denoising processing to obtain smoothed point cloud data; The smoothed point cloud data is subjected to region segmentation to obtain segmented point cloud regions; The covariance matrix corresponding to the segmented point cloud regions is extracted, principal component analysis is performed, and the curvature radius set and the normal vector direction set of each region are obtained; According to the curvature radius set and the normal vector direction set, the segmented point cloud regions are subjected to curved surface fitting to generate the three-dimensional curved surface model.

[0008] Preferably, the optimal incident angle of the multi-source sensor relative to each region of the curved surface is calculated according to the three-dimensional curved surface model to obtain a dynamic attitude adjustment parameter, comprising: The point cloud data of the three-dimensional curved surface model is extracted, and the point cloud data is subjected to local fitting to calculate the curvature radius value and the normal vector direction vector of each curved surface region, to obtain a curvature radius value set and a normal vector direction vector set; According to the curvature radius value set and the normal vector direction vector set, sub-region division is performed to obtain a divided sub-region set; The curvature radius value and the normal vector direction vector of each sub-region in the sub-region set are extracted, and the optimal incident angle of the multi-source sensor relative to the sub-region is iteratively calculated to obtain an optimal incident angle set; According to the optimal incidence angle set, the dynamic attitude adjustment parameters of the multi-source sensor are calculated to obtain the dynamic attitude adjustment parameters.

[0009] Preferably, according to the dynamic attitude adjustment parameters, the three-axis attitude and the data acquisition frequency of the multi-source sensor are adjusted in real time, and the coating thickness related signal is acquired to obtain the original measurement data, including: The preliminary measurement data acquired by the multi-source sensor is subjected to deviation compensation and weighted average fusion to obtain a comprehensive signal. The reflection signal intensity of the comprehensive signal is evaluated, and when the reflection signal intensity is lower than a preset intensity threshold, the dynamic attitude adjustment parameters are optimized to obtain optimized attitude control parameters. According to the optimized attitude control parameters, the three-axis attitude of the multi-source sensor is adjusted, and the data acquisition frequency of the multi-source sensor is dynamically adjusted according to the curvature radius value of each sub-region to acquire the coating thickness related signal and obtain the original measurement data.

[0010] Preferably, the original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, time domain filtering and spatial domain smoothing are performed to obtain corrected measurement data, including: The original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, the original measurement data is subjected to time domain filtering to obtain first corrected data. The first corrected data is subjected to spatial domain smoothing processing to obtain smoothed data. The smoothed data is subjected to cross-source consistency verification to determine an abnormal point set. The abnormal point set is subjected to mean clustering analysis, and according to the analysis result, outliers are removed to obtain corrected measurement data.

[0011] Preferably, the three-dimensional curved surface model and the corrected measurement data are combined to calculate the coating thickness value of each measurement point to obtain preliminary thickness distribution data, including: The three-dimensional spatial coordinates and surface normal vector directions of each measurement point are extracted from the corrected measurement data. According to the three-dimensional spatial coordinates and surface normal vector directions, and in combination with the base geometric information of the three-dimensional curved surface model, the vertical distance of each measurement point from the curved surface base is calculated to obtain the coating thickness value. It is determined whether the coating thickness value exceeds a preset thickness threshold, and values exceeding the thickness threshold are extracted to obtain an abnormal value set. The abnormal value set is subjected to filtering processing to obtain corrected coating thickness values. The corrected coating thickness values are spatially interpolated to generate a continuous distribution of thickness values, to obtain preliminary thickness distribution data.

[0012] Preferably, for the preliminary thickness distribution data, it is detected whether there is a data missing area using a Fourier transform, and if a missing area is found, the preliminary thickness distribution data is interpolated to complete, to generate a complete thickness distribution map, including: The original signals in the preliminary thickness distribution data are subjected to Fourier transform to obtain frequency domain signals of the thickness distribution data, and the amplitude spectrum and phase spectrum of the frequency domain signals are calculated; If the value of a frequency point in the amplitude spectrum of the frequency domain signal is lower than a preset amplitude threshold, it is determined that it is a data missing or abnormal area, and the distribution position of the data missing area and abnormal area is obtained; According to the distribution position of the area, the preliminary thickness distribution data is denoised to obtain denoised thickness distribution data; The boundary point coordinates in the distribution position of the data missing and abnormal area are interpolated to obtain completed thickness distribution data; The boundary point coordinates of the data missing and abnormal area are extracted from the denoised thickness distribution data; According to the completed thickness distribution data, two-dimensional grid points are extracted, the two-dimensional grid points are mapped to pixel values, and a complete thickness distribution map is generated.

[0013] In a second aspect, the present application provides a coating thickness measurement system based on a multi-source sensor, comprising: A point cloud acquisition module is configured to acquire surface geometric information of a complex curved surface component through a multi-source sensor to obtain initial three-dimensional point cloud data; A curved surface modeling module is configured to obtain a three-dimensional curved surface model through curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data; An attitude planning module is configured to calculate the best incident angle of the multi-source sensor relative to each region of the curved surface according to the three-dimensional curved surface model to obtain dynamic attitude adjustment parameters; A signal acquisition module is configured to control the multi-source sensor to adjust the three-axis attitude in real time according to the dynamic attitude adjustment parameters, to acquire coating thickness related signals, and to obtain original measurement data; A data correction module is configured to perform time domain filtering and spatial domain smoothing on the original measurement data to generate a group of corrected measurement data; A thickness calculation module is configured to calculate a group of preliminary thickness distribution data in combination with the three-dimensional curved surface model and the corrected measurement data; A result generation module is configured to perform data completion processing on the preliminary thickness distribution data, and fuse data obtained through multiple measurements to obtain a final coating thickness measurement result.

[0014] In a third aspect, the present application also provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the picture color changing method according to any one of the above-mentioned embodiments when executing the computer program.

[0015] In a fourth aspect, the present application also provides a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the picture color changing method according to any one of the above-mentioned embodiments when the computer program runs.

[0016] Compared with the prior art, the present application has the following beneficial effects: The present application realizes adaptive measurement of the real appearance of a complex curved surface by first constructing a three-dimensional curved surface model of an actual component and then measuring based on the model, thereby significantly improving the accuracy and applicability of the measurement. First, real-time three-dimensional point cloud data of the component is collected by means of laser scanning and structured light scanning, and a three-dimensional curved surface model reflecting the real geometric characteristics of the component is constructed through curvature analysis and normal vector estimation. All subsequent measurement activities are based on this real model rather than a theoretical model. This approach solves the problem of measurement deviation caused by the difference between theory and reality, ensuring that the measurement is performed on the current real state of the component, and is particularly suitable for complex curved surface components that have undergone multiple processes and may have changed in appearance.

[0017] (2) Based on the constructed accurate three-dimensional curved surface model, the present application can use the curved surface normal vector and curvature radius information to actively calculate the best incident angle of the sensor for each region on the curved surface through gradient descent optimization algorithm, and generate dynamic posture adjustment parameters accordingly. This process enables the measurement system to have intelligent decision-making capability, and can control the multi-axis robot in real time to finely adjust the pitch, yaw and roll attitude of the sensor. By ensuring that the sensor always collects signals at the best angle, the present application maximizes the avoidance of signal reflection distortion at the source of data acquisition, ensuring that the original measurement data has a very high signal-to-noise ratio and effectiveness, and lays a solid foundation for obtaining high-precision and high-reliability coating thickness values.

[0018] (3) The present application establishes a set of accurate signal level correction mechanism for the collected raw measurement data. The mechanism first intelligently identifies the signal distortion area caused by poor curved surface reflection through frequency domain analysis and other means, and then uses time domain filtering and spatial domain smoothing and other targeted algorithms for data correction. This way of "purifying" the original signal before calculating the thickness value can effectively eliminate high-frequency noise and spatial distribution noise in the signal, prevent the pollution of bad data on the subsequent calculation results, and greatly improve the accuracy of the final thickness value calculation.

[0019] (4) After obtaining the preliminary thickness distribution data, the present application further designs a repair and reconstruction process to ensure the integrity of the final result. The process automatically detects and locates the missing or abnormal areas in the data using Fourier transform and other analysis methods, and uses interpolation algorithms to intelligently complete these areas with high fidelity. This step ensures that the final output of the present application is not a set of discrete measurement points that may have holes, but a spatially continuous and information complete coating thickness distribution map, providing users with comprehensive and highly reliable quality evaluation basis. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a coating thickness measurement method flowchart provided by the first embodiment of the present application based on multi-source sensors; Figure 2 is a coating thickness measurement system structure diagram provided by the second embodiment of the present application based on multi-source sensors. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] Referring to Figure 1 , the first embodiment of the present application provides a coating thickness measurement method based on multi-source sensors, including the following steps: S11, obtaining initial three-dimensional point cloud data; S12, obtaining a three-dimensional curved surface model through curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data; S13, calculating the best incident angle of the multi-source sensor relative to each area of the curved surface according to the three-dimensional curved surface model, and calculating the dynamic attitude adjustment parameter according to the best incident angle; S14, real-time adjust the three-axis attitude and data acquisition frequency of the multi-source sensor according to the dynamic attitude adjustment parameter, and acquire the coating thickness related signal to obtain original measurement data; S15, perform frequency domain analysis on the original measurement data to determine whether there is reflection distortion, and if so, perform time domain filtering and spatial domain smoothing to obtain corrected measurement data; S16, combine the three-dimensional curved surface model and the corrected measurement data to calculate the coating thickness value of each measurement point to obtain preliminary thickness distribution data; S17, for the preliminary thickness distribution data, use Fourier transform to detect whether there is a data missing area, and if a missing area is found, interpolate and complete the preliminary thickness distribution data to generate a complete thickness distribution map.

[0023] For the preliminary thickness distribution data, use Fourier transform to detect whether there is a data missing area, and if a missing area is found, interpolate and complete the preliminary thickness distribution data to generate a complete thickness distribution map.

[0024] In step S11, initial three-dimensional point cloud data is obtained.

[0025] Specifically, first, data is collected in a multi-source heterogeneous manner in parallel, for example, a line laser scanner is used to obtain the large-scale profile of a complex curved surface component to generate a first point cloud data set, and a structured light scanner is used to capture the detail area of the curved surface to generate a second point cloud data set with a resolution of up to 0.02 millimeters. Next, to unify the coordinate system, an iterative closest point (ICP) algorithm is used to perform high-precision spatial registration on the two groups of point clouds until the root mean square error of the point pairs is less than a predetermined threshold, for example, 0.01 millimeters, thereby generating a fused point cloud. Subsequently, to optimize the data and reduce noise, a voxel grid filtering method is applied, in which the voxel size is adaptively adjusted according to the local complexity of the curved surface (for example, 0.1 millimeters for smooth areas and 0.03 millimeters for high-curvature areas), to reduce data redundancy while preserving key geometric features. Finally, the filtered point cloud data is converted to a unified three-dimensional mesh format (such as STL format), and its spatial resolution is verified to meet the accuracy requirements of subsequent analysis, thereby obtaining a high-precision initial three-dimensional point cloud data.

[0026] In step S12, a three-dimensional curved surface model is obtained from the initial three-dimensional point cloud data through curvature distribution analysis and normal vector estimation. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S121 to S124: S121, denoising the initial three-dimensional point cloud data to obtain smoothed point cloud data; S122, region segmentation is performed on the smoothed point cloud data to obtain segmented point cloud regions; S123, extract the covariance matrix corresponding to the segmented point cloud region, perform principal component analysis to obtain the curvature radius set and the normal vector direction set of each region; S124, according to the curvature radius set and the normal vector direction set, surface fitting is performed on the segmented point cloud region to generate a three-dimensional curved surface model.

[0027] In step S121, the initial three-dimensional point cloud data is denoised to obtain smoothed point cloud data. Specifically, this step aims to eliminate random noise points that may exist in the initial point cloud to improve the stability of subsequent geometric analysis. In a preferred embodiment, a mean filtering algorithm can be used. For each point in the point cloud, a neighborhood space centered on the point is defined, for example, a spherical space with a radius of 2 mm. Then, the algorithm finds all neighborhood points falling within the spherical space, and calculates the arithmetic mean of the three-dimensional coordinates of these neighborhood points (including the center point itself). Finally, the original coordinates of the center point are updated and replaced by the calculated mean coordinates. By performing this operation on all points in the entire point cloud, high-frequency random noise introduced by sensor micro-vibration or environmental light reflection can be effectively suppressed, generating point cloud data with smoother coordinate distribution and closer to the true surface morphology.

[0028] In step S122, the smoothed point cloud data is regionally segmented to obtain the segmented point cloud region. Specifically, the purpose of this step is to organize point clouds with similar geometric features together for subsequent localized analysis. In a preferred embodiment, a clustering segmentation method based on Euclidean distance is used. The basic principle is to calculate the spatial straight-line distance between points and divide points with similar distances into the same cluster (i.e., a segmented region).

[0029] The execution process of the algorithm is as follows: first, set a distance threshold d (e.g., 3 mm) and a minimum number of points N in a region (e.g., 50 points). Among them, the setting of the distance threshold d is based on the average point distance of the smoothed point cloud, which can be set to 2 to 5 times the average point distance to ensure that points belonging to the same continuous surface can be connected. The setting of the minimum number of points N in the region is mainly used to filter out sparse outlier noise points, and the value should be greater than the size of the expected noise cluster and less than the number of points contained in the smallest effective feature on the component.

[0030] ​The algorithm iterates through the entire smoothed point cloud dataset, and for any point that has not been assigned to any region, it selects this point as the initial seed point of a new region. Then, starting from this seed point, the algorithm looks for all neighboring points within a distance d, and if the number of these points is greater than N, a new region is formed and these neighboring points are also added to the region. Then, taking these newly added points as the new basis, the algorithm iterates outward and absorbs unassigned neighboring points that satisfy the distance threshold d. This expansion process continues until no new points can be added to the current region. When a region is grown, the algorithm continues to look for the next unassigned point in the outer iteration and takes it as the initial seed point of the next new region, repeating the above process.

[0031] By this method, the entire point cloud data can be segmented into several independent point cloud regions until all points are successfully assigned, for example, the relatively flat side surface of a car part can be segmented into a region, while the complex connection part can be segmented into another region.

[0032] In step S123, the covariance matrix corresponding to the segmented point cloud region is extracted, and principal component analysis is performed to obtain the curvature radius set and the normal vector direction set of each region. Specifically, this step assigns accurate geometric properties to each segmented region. This step uses the principal component analysis (PCA) technique. For each segmented point cloud region (for example, a region containing 1000 points), the algorithm first calculates the covariance matrix of the three-dimensional coordinates of all points in the region. This is a 3x3 symmetric matrix that describes the distribution of the point set in each direction. Then, by performing eigenvalue decomposition on the covariance matrix, three orthogonal eigenvectors and three non-negative eigenvalues are obtained. Among them, the direction of the eigenvector corresponding to the smallest eigenvalue is the average normal vector direction of the local tangent plane fitted by the point cloud region. At the same time, the distribution relationship between the three eigenvalues reflects the bending degree of the region, and the principal curvature and the curvature radius can be calculated accordingly, for example, the curvature of a region can be approximated and quantified by .

[0033] In step S124, a surface fitting is performed on the segmented point cloud region according to the set of curvature radius and the set of normal vector direction, to generate a three-dimensional surface model. Specifically, this step is to convert the discrete point cloud with direction and curvature information into a continuous mesh model that can be operated on geometrically. In a preferred embodiment, a Poisson Surface Reconstruction algorithm is adopted. The principle is to convert the surface reconstruction problem into solving a Poisson equation. The algorithm takes the normal vector information of all point clouds as a three-dimensional vector field, and finds a scalar function (implicit function) such that the gradient field of the function is as consistent as possible with the input normal vector field. This process is usually efficiently performed on an Octree data structure, and the depth of the tree (e.g., can be set to 8 to 12) determines the level of detail of the reconstructed model. After solving the scalar field, an isosurface, for example, a surface with a scalar value of 0, is extracted from the field by the Marching Cubes algorithm. This extracted surface is the final three-dimensional surface model, which has the excellent characteristics of smoothness, continuity and water tightness, and can accurately express the complex geometric shape of the original component.

[0034] In step S13, according to the three-dimensional surface model, the optimal incident angle of the multi-source sensor relative to each region of the surface is calculated, and the dynamic pose adjustment parameter is calculated according to the optimal incident angle. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S131 to S134: S131, extracting the point cloud data of the three-dimensional surface model, locally fitting the point cloud data, calculating the curvature radius value and the normal vector direction vector of each surface region, to obtain the set of curvature radius values and the set of normal vector direction vectors; S132, according to the set of curvature radius values and the set of normal vector direction vectors, sub-region division is performed to obtain a set of divided sub-regions; S133, extracting the curvature radius value and the normal vector direction vector of each sub-region in the set of sub-regions, iteratively calculating the optimal incident angle of the multi-source sensor relative to the sub-region, to obtain a set of optimal incident angles; S134, according to the set of optimal incident angles, calculating the dynamic pose adjustment parameter of the multi-source sensor, to obtain the dynamic pose adjustment parameter.

[0035] In step S131, the point cloud data of the three-dimensional surface model is extracted, the point cloud data is locally fitted, the curvature radius value and the normal vector direction vector of each surface region are calculated, and the set of curvature radius values and the set of normal vector direction vectors are obtained.

[0036] In particular, this step directly operates on the three-dimensional surface model generated in step S12, i.e. a triangular mesh model, to prepare the basic geometry data for the subsequent segmentation and optimization. In a preferred embodiment, for each vertex on the model, its normal vector and curvature are calculated by the following way: First, the normal vector calculation is performed. For each vertex, first find all the triangular facets that share this vertex. Then, calculate the unit face normal of each triangular facet. Finally, use a method based on the weighted average of the adjacent face angles to sum all these face normals to obtain the accurate normal vector of the vertex. This method can ensure that the normal vector field is smooth even at sharp edges.

[0037] Then, the curvature calculation is performed. For each vertex, use a method based on local quadratic surface fitting to estimate its curvature. The core principle is that for the target vertex and its "1-ring" neighborhood (i.e. all the adjacent vertices directly connected by an edge), fit the three-dimensional coordinates of these points to a best quadratic surface. By performing differential geometry analysis on the fitted quadratic surface equation, the principal curvature, Gaussian curvature and mean curvature at the vertex can be accurately calculated, and the curvature radius is obtained accordingly.

[0038] By traversing all the vertices of the three-dimensional surface model, a set of curvature radius values and a set of normal vector directions corresponding to each vertex are finally obtained, which contain accurate geometric information.

[0039] In step S132, according to the set of curvature radius values and the set of normal vector direction vectors, sub-region division is performed to obtain a set of divided sub-regions. In particular, this step aims to classify the measurement regions with similar geometric characteristics into a class for unified pose optimization. In a preferred embodiment, a region growing algorithm based on dual constraints of curvature and normal vector is used. The core working principle is as follows: First, the algorithm traverses all the vertices on the three-dimensional surface model. For any vertex that has not been assigned to any sub-region, it is selected as the initial seed point of a new sub-region.

[0040] Then, starting from the seed point, the algorithm iteratively examines its adjacent vertices that have not been assigned by the edge connection relationship of the mesh. An adjacent vertex can be absorbed into the current sub-region if it meets the following two conditions simultaneously: Curvature similarity condition: the difference between the curvature radius value of the vertex and the average curvature radius value of all existing vertices within the current sub-region is no more than a preset curvature difference threshold C. For example, C can be set to 0.1 meters. The threshold C is set according to the requirement of the subsequent measurement task for the fineness of path planning. For components that require high-precision measurement and have complex curved surface morphologies, a smaller C value is selected to generate more subdivided measurement regions.

[0041] Normal vector consistency condition: the included angle between the normal vector direction of the vertex and the average normal vector direction of the current sub-region is less than a preset angle consistency threshold. For example, 15 degrees can be set. The threshold is directly derived from the technical specifications of the sensor used, and is usually set to half of the effective angle range in which the sensor can receive high-quality reflection signals, to ensure that the surface orientations of all points in the sub-region are within the effective working cone angle of the sensor.

[0042] Only when both conditions are met, the adjacent vertex is officially added to the current sub-region, and itself becomes the new basis for the next iteration of outward expansion. If there is no longer a vertex that meets the condition in one direction, the growth in that direction stops. When a region can no longer be expanded, the algorithm continues to traverse the model to find the next unassigned vertex as a new seed point, repeating the above process until all vertices are assigned.

[0043] By this method, the entire three-dimensional curved surface model can be divided into a set of sub-regions that are highly consistent in internal geometric characteristics, including curvature and surface orientation, and are composed of connected triangular facets.

[0044] In step S133, the curvature radius value and the normal vector direction vector of each sub-region in the sub-region set are extracted, and the best incidence angle of the multi-source sensor relative to the sub-region is iteratively calculated to obtain a set of best incidence angles. Specifically, this step is the core calculation link for implementing pose optimization. It should be noted that the "best incidence angle" is defined in this embodiment as a three-dimensional best incidence direction vector, which clearly indicates the best orientation of the sensor axis in the global coordinate system. Gradient descent method is used for iterative optimization in this step. The working principle is as follows: Define the objective function: establish an objective function for quantifying the quality of the measurement signal , wherein is the incidence direction vector of the sensor. The form of the objective function depends on the physical principle of the sensor used. For example, for laser or optical sensors, the function can be constructed based on a simplified Phong reflection model. The core idea is that when the receiving direction of the sensor is close to the ideal specular reflection direction The closer the object is, the higher the received signal strength is. The function can be specifically expressed as: where "·" represents the vector dot product. The incident direction and the average normal vector n of the sub-region can be calculated by the reflection law: n is the glossiness coefficient of the coating material, which is obtained by a pre-calibration experiment on a coating sample (for example, n can be calibrated to 50 for a certain high-gloss coating).

[0045] Initialize the incident direction: for each sub-region, the initial incident direction is set by considering the light source direction and the average normal vector n of the sub-region. In a preferred embodiment, the initial incident direction is set as the direction of the angle bisector of the plane formed by the light source direction and the normal vector n. This initialization method takes into account the light source position and provides a more reasonable starting point for subsequent optimization.

[0046] Iterative update: in each iteration, the gradient of the objective function J(d_in) under the current incident direction is calculated, and the incident direction vector is fine-tuned by updating along the direction of gradient ascent (since the goal is to maximize J): where t represents the current iteration number, for example, starting from t=0. is the incident direction vector at the tth iteration. is the gradient of the objective function J at this point. The gradient itself is also a vector, which points to the direction of the fastest growth of the function J value, and a is the learning rate (for example, it can be set to 0.01). This process is repeated (for example, 10 to 20 iterations) until the change in the incident direction vector (for example, the vector angle is less than 0.1 degree) is less than the convergence threshold, and the incident direction vector at this time is considered to be the optimal incident direction of the sub-region.

[0047] In step S134, the dynamic pose adjustment parameters of the multi-source sensor are calculated according to the optimal incident angle set, and the dynamic pose adjustment parameters are obtained.

[0048] Specifically, this step is to convert the calculated abstract optimal incidence angle vector into specific motion instructions that can be understood and executed by the actuators such as multi-axis robot arms. In a preferred embodiment, this process is a coordinate system solving process. For each sub-region, its optimal incidence angle is a three-dimensional directional vector. This step needs to calculate the amount of rotation and translation that the end of the robot arm needs to perform in order to align the central axis of the sensor with the optimal incidence angle vector. The amount of rotation can be represented as a set of Euler angles, for example, rotate 5.2 degrees around the Z axis, and then rotate 3.1 degrees around the Y axis. The amount of translation corresponds to the target measurement point position to which the sensor needs to move. The combination of these rotations and translations together constitutes the dynamic pose adjustment parameters for the sub-region. Collecting the parameters of all sub-regions together forms a complete and dynamically changing measurement path planning.

[0049] In step S14, the three-axis pose of the multi-source sensor and the data acquisition frequency are adjusted in real time according to the dynamic pose adjustment parameters, and the coating thickness related signal is collected to obtain the original measurement data. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S141 to S143: S141, deviation compensation and weighted average fusion are performed on the preliminary measurement data collected by the multi-source sensor to obtain a comprehensive signal; S142, the reflection signal intensity of the comprehensive signal is evaluated, and when the reflection signal intensity is lower than a preset intensity threshold, the dynamic pose adjustment parameters are optimized to obtain optimized pose control parameters; S143, the three-axis pose of the multi-source sensor is adjusted according to the optimized pose control parameters, and the data acquisition frequency of the multi-source sensor is dynamically adjusted according to the curvature radius value of each sub-region to collect the coating thickness related signal and obtain the original measurement data.

[0050] In step S141, deviation compensation and weighted average fusion are performed on the preliminary measurement data collected by the multi-source sensor to obtain a comprehensive signal. Specifically, this step aims to integrate measurement information from different sensors and eliminate their inherent deviations. In a preferred embodiment, deviation compensation is achieved by applying a preset calibration matrix. The calibration matrix is generated in advance by least squares fitting method according to the historical test data of each sensor under standard conditions and the measurement results of known reference points, and it can quantify and compensate for systematic deviations caused by sensor hardware aging or environmental temperature changes (for example, a temperature increase of 10 degrees Celsius may cause a drift of 0.05 millimeters). After compensation, weighted average method is used for data fusion. This method assigns different weights to each sensor according to its prior accuracy or reliability For example, if the measurement accuracy of three sensors are 95%, 90% and 85% respectively, then they can be assigned weights of 0.4, 0.3 and 0.3 respectively (the sum of the weights is 1). The final integrated signal value S is calculated by the formula The result is a more stable and reliable signal than any single source.

[0051] In step S142, the reflection signal strength of the integrated signal is evaluated, and when the reflection signal strength is lower than a preset strength threshold, the dynamic posture adjustment parameter is calculated to obtain the optimized posture control parameter.

[0052] Specifically, this step is a closed-loop feedback optimization link for online correction of posture deviation caused by mechanical arm execution error or model deviation. In a preferred embodiment, a Kalman filter is used to smooth and optimize the posture control parameter. In order to apply the Kalman filter, this step establishes an indirect and local linear mathematical relationship between the posture angle (state variable) and the signal strength (observation value). The core working principle is as follows: First, define the state and observation model: State variable: defined as the three-axis posture angle of the sensor (pitch, yaw, roll). An intermediate variable is introduced: define a "posture deviation angle" as an intermediate variable. This step assumes that, within the small neighborhood of the current posture, the change in signal strength is linearly related to the change in posture deviation angle. This linear relationship coefficient can be approximately obtained by performing a small perturbation around the optimal posture and measuring the signal strength change during the calibration phase. Observation model: Therefore, the "observation value" of the system is defined as the difference between the measured signal strength and the target signal strength (i.e. the expected value of the objective function in S133 at the optimal posture). This difference indirectly reflects the size of the "posture deviation angle" through the above-mentioned local linear relationship. Kalman filtering process: Prediction: based on the posture angle at the last time and the kinematics model of the mechanical arm, the posture angle at the current time is predicted (prior estimate). Update: when the measured integrated signal strength is lower than the preset strength threshold (e.g. 50 decibels), the difference between it and the target signal strength (observation value) is calculated. The Kalman filter will use this difference (which indirectly reflects the posture deviation) to generate a more accurate posterior estimate of the current posture angle, combining the predicted posture and the Kalman gain.

[0053] In this way, the Kalman filter does not directly take the signal strength as an observation value, but takes the difference between the signal strength and the target value as an indirect measurement of the "attitude deviation", thereby effectively smoothing and correcting the attitude control parameters. For example, an initial attitude angle instruction is (10.0, 5.0°, 0.0°), and after smoothing and correction by the Kalman filter, the output optimized attitude control parameters can be (9.8°, 4.7°, 0.2°), thereby effectively compensating for high-frequency jitter and execution delay errors in the system.

[0054] In step S143, the three-axis attitude of the multi-source sensor is adjusted according to the optimized attitude control parameters, and the data acquisition frequency of the multi-source sensor is dynamically adjusted according to the curvature radius value of each sub-region, and the coating thickness related signal is acquired to obtain the original measurement data.

[0055] Specifically, this step is the link of performing the final measurement. The multi-axis robot adjusts the pitch angle, yaw angle and roll angle of its end effector (i.e. the sensor) in real time according to the received optimized attitude control parameters, to ensure that the measurement axis of the sensor is accurately aligned with the target. At the same time, the data acquisition frequency is also dynamically adjusted.

[0056] In a preferred embodiment, the acquisition frequency f is inversely proportional to the curvature radius p of the current measurement region, for example, it can be determined by a pre-set lookup table or function , where k is a proportional coefficient. The proportional coefficient k is an engineering parameter pre-calibrated according to the target sampling density and scanning speed. Its setting is based on the following assumption: it is required that the spatial distance between measurement points (i.e. the spatial resolution) on the scanning path is not greater than a pre-set value (e.g. 0.5 mm), and the nominal scanning speed of the sensor is (e.g. 50 mm / s). In order to achieve this goal on a circular arc with a curvature radius of p, the required acquisition frequency f is approximately equal to / . At the same time, considering that the same linear speed is scanned on a curved surface, the angle swept per unit time is larger where the curvature radius is smaller, in order to maintain the sampling point density, the frequency needs to be increased accordingly. Therefore, by calibration experiment, a best fitting coefficient k can be determined which can balance the speed, curvature and target sampling density. For example, by testing on a plurality of standard curvature samples at a speed of , it can be calibrated that when k is set to 3.0 m·Hz, it can better maintain a spatial resolution of about 0.5 mm under various curvatures. In actual measurement, this step will calculate the required data acquisition frequency in real time according to the curvature radius p value of the current measurement region by the function (where k = 3.0). For example, in a flat area with a large curvature radius ( =1 meter), the acquisition frequency is calculated as 3 Hz; while in the high curvature area with small radius of curvature (R = 0.1 meter), the acquisition frequency is automatically increased to 30 Hz. Through this dual dynamic adjustment of posture and frequency, a set of original measurement data with more uniform spatial distribution and more reasonable information density is finally acquired. =0.1 meter), the acquisition frequency is automatically increased to 30 Hz. Through this dual dynamic adjustment of posture and frequency, a set of original measurement data with more uniform spatial distribution and more reasonable information density is finally acquired.

[0057] In step S15, the original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, time domain filtering and spatial domain smoothing are performed to obtain corrected measurement data. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S151 to S154: S151, the original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, the original measurement data is subjected to time domain filtering to obtain first corrected data; S152, the first corrected data is subjected to spatial domain smoothing processing to obtain smoothed data; S153, the smoothed data is subjected to cross-source consistency verification to determine an abnormal point set; S154, the abnormal point set is subjected to mean clustering analysis, and outliers are removed according to the analysis results to obtain corrected measurement data.

[0058] In step S151, the original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, the original measurement data is subjected to time domain filtering to obtain first corrected data. Specifically, this step aims to diagnose and repair signal high-frequency noise caused by poor reflection. In a preferred embodiment, first, the original measurement data is converted from time domain to frequency domain using Fast Fourier Transform (FFT) to generate a frequency spectrum. By analyzing the spectral characteristics, if it is found that the energy is concentrated in a certain high frequency band (for example, the frequency is higher than 100 Hz, and this threshold is pre-set according to the expected frequency band of the coating thickness signal), it is determined that this segment of data has reflection distortion. For the data determined to be distorted, Discrete Wavelet Transform (DWT) is immediately used for time domain filtering. The core principle is to use a wavelet basis function with time-frequency localization characteristics (for example, Daubechies wavelet db4) to perform multi-scale decomposition of the signal, for example, to 4 layers, so as to effectively separate the signal into low-frequency components representing trends and high-frequency components representing noise. By performing hard thresholding or soft thresholding on the high-frequency component coefficients (for example, setting the high-frequency components to zero), and then reconstructing the signal, the high-frequency noise can be effectively removed (for example, reducing the signal amplitude fluctuation from 0.05 mm to 0.01 mm), and the first corrected data is obtained.

[0059] In step S152, the first corrected data is spatially smoothed to obtain smoothed data. Specifically, this step aims to eliminate the noise in the three-dimensional spatial distribution of the point cloud data, which can be caused by the accumulation of minor measurement errors. In a preferred embodiment, Gaussian convolution is used to perform smoothing in the three-dimensional spatial domain. For each data point, the algorithm calculates the weighted average of other points in the neighborhood as its new value, and the size of the weight is determined by a Gaussian kernel function according to the Euclidean distance between the point and the neighborhood points. The closer the point is to the center point, the higher the weight it is given, and vice versa. The standard deviation σ of the Gaussian kernel function is a key parameter, which can be set to 0.5 meters, for example, and it determines the degree of smoothing. Compared with simple mean filtering, Gaussian convolution can better preserve the edges and geometric features of the point cloud, because it performs smooth, nonlinear weighting on the contributions of the neighborhood, thereby effectively filtering out spatial distribution noise while avoiding excessive blurring, resulting in a more continuous and regular smoothed data in space.

[0060] In step S153, the smoothed data is checked for cross-source consistency to determine the set of abnormal points. Specifically, this step aims to identify and locate outlier data caused by systematic bias between different sensors or single measurement faults. In a preferred embodiment, by comparing the sets of smoothed data points from multiple sensor sources (e.g., two laser radars and a depth camera) in the same coordinate system, the spatial distribution difference is checked. For example, for a theoretical position point in space, if sensor A measures its coordinates as (10.0, 10.0, 5.0) and sensor B measures it as (10.0, 10.2, 5.0), there is a 0.2-meter deviation in the Y-axis between the two. This step calculates the spatial deviation between all corresponding points or clusters of adjacent area points, and marks and includes those data points with deviation values exceeding a predetermined consistency threshold (e.g., 0.1 meters, which is set according to the nominal accuracy of the sensor and the expected measurement error range) in a set of abnormal points.

[0061] In step S154, mean clustering analysis is performed on the set of abnormal points, and outliers are removed according to the analysis results to obtain corrected measurement data. Specifically, this step is an intelligent screening of the set of abnormal points determined in the previous step to distinguish between true noise points and useful edge points. In a preferred embodiment, the K-means clustering algorithm is used. First, a cluster number K (for example, K = 3) needs to be preset. The algorithm randomly initializes K cluster centers, and then iteratively performs two steps: each point in the set of abnormal points is assigned to the nearest cluster center; the geometric center of each cluster (i.e., the mean of the coordinates of all points in the cluster) is recalculated as the new cluster center. This process is repeated until the cluster centers no longer change. After the iteration is completed, the set of abnormal points is divided into K clusters. By analyzing the characteristics of each cluster, such as the number of points in the cluster and the compactness of the spatial distribution, clusters that are too small or too dispersed can be automatically identified and removed, as these clusters are likely to be composed of false points (i.e., outliers) caused by random reflections or occlusions. The remaining clusters that are larger in size and more concentrated in distribution are likely to be useful data and will be retained. Through this step, a set of corrected measurement data with high reliability is obtained after double filtering and outlier removal.

[0062] In step S16, the coating thickness values of each measurement point are calculated based on the three-dimensional curved surface model and the corrected measurement data to obtain preliminary thickness distribution data. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S161 to S165: S161, extracting the three-dimensional spatial coordinates and surface normal vector directions of each measurement point from the corrected measurement data; S162, calculating the perpendicular distance of each measurement point from the curved surface base based on the three-dimensional spatial coordinates and surface normal vector directions, and combining the base geometric information of the three-dimensional curved surface model to obtain the coating thickness value; S163, determining whether the coating thickness value exceeds the preset thickness threshold, and extracting the values that exceed the thickness threshold to obtain a set of abnormal values; S164, filtering the set of abnormal values to obtain corrected coating thickness values; S165, performing spatial interpolation on the corrected coating thickness values to generate a continuous distribution of thickness values, and obtaining preliminary thickness distribution data.

[0063] In step S161, the three-dimensional spatial coordinates and surface normal vector directions of each measurement point are extracted from the corrected measurement data. Specifically, this step aims to match each measurement point with an accurate direction vector for calculating the thickness. The execution process is as follows: First, this step will iterate through the corrected measurement data and extract only the 3D spatial coordinates of each measurement point, e.g. P(10.5, 20.3, 15.2).

[0064] Next, in order to obtain the surface normal vector corresponding to the measurement point, this step will correlate the coordinate P with the 3D surface model representing the component substrate that was constructed in step S12. In a preferred embodiment, this is achieved through a Nearest Neighbor Query. Specifically, for each measurement point P, the algorithm will perform an efficient search on the 3D surface model (which already has normal vector information stored for each vertex) to find the model vertex or triangle patch that is closest in space to the point P.

[0065] Once the nearest neighbor location is found, the algorithm extracts the pre-computed surface normal vector direction from the location, stored in the model data structure, e.g. n(0.7, 0.2, 0.6). This normal vector obtained from the substrate model is then designated as the projection direction to be followed when calculating the thickness for the measurement point P.

[0066] In this way, each measurement point is given a 3D spatial coordinate from the actual measurement, as well as an accurate surface normal vector direction from the high-precision substrate model, providing complete and logically correct geometric inputs for the subsequent perpendicular distance calculation.

[0067] In step S162, the perpendicular distance between each measuring point and the surface base is calculated according to the three-dimensional spatial coordinates and the surface normal vector direction, combined with the base geometry information of the three-dimensional curved surface model, to obtain the coating thickness value. Specifically, this step is the core link of thickness calculation. In a preferred embodiment, to ensure the uniqueness and accuracy of the calculation on complex non-convex surfaces, a ray projection method based on nearest neighbor search is used. The core working principle is: for each measuring point P and its normal vector n, first define a ray originating from point P and along the normal vector n direction. Then, calculate all the intersection points of the ray and the base geometry model of the three-dimensional curved surface model. When dealing with non-convex surfaces, the ray can have multiple intersection points. To solve this ambiguity, this step calculates the Euclidean distance between each intersection point and the original measuring point P. Finally, the algorithm selects the intersection point with the shortest distance as the only effective vertical projection point S. This "select the nearest" strategy ensures that the calculated distance is the most physically meaningful and real distance from the measuring point to the base directly below it. Finally, the coating thickness value T is the shortest Euclidean distance between the measuring point P and the uniquely determined projection point S. For example, a measuring point has coordinates (5.0, 10.0, 3.0), and two intersection points are calculated by ray projection, with distances of 0.02 mm and 5.80 mm respectively, then the algorithm selects 0.02 mm as the final coating thickness value of the point.

[0068] In step S163, it is determined whether the coating thickness value exceeds the preset thickness threshold value, and the values exceeding the thickness threshold value are extracted to obtain an abnormal value set. Specifically, this step is a data verification process aimed at identifying extreme measurement errors caused by accidental factors. In a preferred embodiment, a preset thickness threshold range is set, for example, according to the quality control standard (SPC) of the spraying process, the range can be set to [0.01 mm, 0.05 mm]. This step will traverse all the calculated coating thickness values, and identify any values outside this range (for example, a point with a calculated value of 0.08 mm) and classify them into a separate abnormal value set for subsequent special processing.

[0069] In step S164, the set of outliers is filtered to obtain corrected coating thickness values. Specifically, this step aims to correct the outliers identified in the previous step. In a preferred embodiment, a median filtering algorithm is employed. For each outlier thickness value in the set of outliers, the algorithm defines a local window containing the point and its spatially nearest N points (e.g. N = 4, then the window size is 5 points). Then, the algorithm reads the thickness values of the 5 points, sorts them, and selects the value in the middle (i.e. the median) to replace the original outlier. For example, for an outlier of 0.08 mm, the thickness values of its neighborhood points are 0.03, 0.02, 0.04, 0.03 mm, and the sorted sequence is (0.02, 0.03, 0.03, 0.04, 0.08), then the median 0.03 mm will be the corrected coating thickness value. This algorithm is particularly effective for handling pulse-like, isolated extreme noise points.

[0070] In step S165, the corrected coating thickness values are spatially interpolated to generate a continuous distribution of thickness values, obtaining preliminary thickness distribution data. Specifically, this step aims to convert the discrete measurement point data into a continuous, information-complete thickness distribution map. In a preferred embodiment, a Kriging interpolation algorithm is employed. This algorithm first analyzes the spatial autocorrelation of the data by calculating the spatial distances and numerical differences between all corrected thickness points, and then fits a theoretical variogram model (e.g. spherical model) that quantifies the law of thickness value similarity changing with distance. When interpolating, for each grid position whose thickness value needs to be predicted, the algorithm uses the variogram model to construct and solve a Kriging equation system, thereby calculating an optimal weight coefficient for each known measurement point around it. Finally, the predicted value of the grid position is obtained by weighted sum of the thickness values of these known points. By traversing all grid positions, a smooth and statistically unbiased optimal continuous thickness distribution is generated, obtaining the final preliminary thickness distribution data.

[0071] In step S17, for the preliminary thickness distribution data, Fourier transform is used to detect whether there is a missing area of data, and if a missing area is found, the preliminary thickness distribution data is interpolated to complete the generation of a complete thickness distribution map. It should be noted that in the first embodiment of the present application, this processing step specifically includes steps S171 to S175: S171, Fourier transform is performed on the original signal in the preliminary thickness distribution data to obtain the frequency domain signal of the thickness distribution data, and the amplitude spectrum and phase spectrum of the frequency domain signal are calculated; S172, if the value of the frequency point in the amplitude spectrum of the frequency domain signal is lower than the preset amplitude threshold, it is determined that the data is missing or the abnormal region, and the distribution position of the data missing region and the abnormal region is obtained; S173, according to the distribution position of the region, the preliminary thickness distribution data is denoised to obtain the denoised thickness distribution data; S174, the boundary point coordinates in the distribution position of the data missing and abnormal region are calculated by interpolation to obtain the completed thickness distribution data.

[0072] S175, according to the completed thickness distribution data, a two-dimensional grid point is extracted, the two-dimensional grid point is mapped to a pixel value, and a complete thickness distribution map is generated.

[0073] In step S171, the original signal in the preliminary thickness distribution data is subjected to Fourier transform to obtain the frequency domain signal of the thickness distribution data, and the amplitude spectrum and the phase spectrum of the frequency domain signal are calculated. Specifically, this step is to convert the preliminary thickness distribution data from the spatial domain or the time domain to the frequency domain for analysis. In a preferred embodiment, the Fast Fourier Transform (FFT) algorithm is used. Specifically, the input thickness value sequence (original signal) is decomposed into a series of superimposed sine and cosine waves of different frequencies. The result of the transformation is a complex number sequence, from which two key parts can be calculated: the amplitude spectrum, which reflects the intensity or energy of each frequency component in the original signal; and the phase spectrum, which describes the phase shift of each frequency component. For example, a thickness value sequence with a sampling frequency of 1000 Hz can clearly show the frequency range where the main energy is concentrated after FFT transformation.

[0074] In step S172, if the value of the frequency point in the amplitude spectrum of the frequency domain signal is lower than the preset amplitude threshold, it is determined that the data is missing or the abnormal region, and the distribution position of the data missing region and the abnormal region is obtained. Specifically, this step is to locate the problem area based on the results of the FFT transformation in the previous step. In a preferred embodiment, a preset amplitude threshold is set. The threshold is set based on the fact that the energy of the effective coating thickness signal is usually concentrated in the low frequency part, and data missing or serious abnormality will cause signal interruption, which is reflected in the spectrum as a significant decrease in energy. For example, by statistically analyzing the frequency spectrum of a large number of normal measurement signals, the threshold can be set to 0.1. When analyzing the amplitude spectrum, if the amplitude value of a certain frequency point or a certain frequency range is lower than the threshold, it is considered that the corresponding original signal part has data missing or serious abnormality, and the start and end positions of these regions in the original data sequence are recorded.

[0075] In step S173, the preliminary thickness distribution data is denoised according to the distribution position of the region to obtain denoised thickness distribution data. Specifically, this step is to perform a whole smoothing denoising on the original data before data interpolation, so as to improve the accuracy of interpolation calculation. In a preferred embodiment, discrete wavelet transform (DWT) is used for denoising. The core principle is to use wavelet basis functions with time-frequency localization characteristics (for example, Daubechies wavelet db4) to perform multi-scale decomposition on the signal, for example, to decompose to 3 levels. This decomposition can effectively separate the signal into low-frequency approximation components representing macro trends and high-frequency detail components representing local details and noise. By applying threshold processing (for example, soft threshold or hard threshold method) to the high-frequency detail component coefficients of each layer, the noise component is removed, and then the signal is reconstructed. In this way, the high-frequency noise can be effectively filtered without damaging the main trend of the signal, and a denoised thickness distribution data with better continuity and smoothness is obtained.

[0076] In step S174, the boundary point coordinates in the distribution position of the data missing and abnormal region are interpolated to obtain the completed thickness distribution data. Specifically, this step aims to repair the identified data holes. In a preferred embodiment, cubic spline interpolation algorithm is used. For each identified data missing region, the algorithm first obtains the boundary point coordinates of the left and right ends of the region. Then, a series of piecewise cubic polynomial functions are constructed by solving a set of equations between these boundary points, and it is ensured that these functions are not only equal in value at the connection points (i.e. boundary points), but also continuous in the first and second derivatives. The interpolation curve generated in this way has very good smoothness. For example, for a missing region with boundary points (10, 0.5) and (15, 0.7), cubic spline interpolation will generate a very natural curve passing through these two points to complete the missing thickness values between them.

[0077] In step S175, two-dimensional grid points are extracted according to the completed thickness distribution data, the two-dimensional grid points are mapped to pixel values, and a complete thickness distribution map is generated. Specifically, this step is to convert one-dimensional or three-dimensional numerical data into a two-dimensional visual image. In a preferred embodiment, the completed thickness distribution data is first rearranged or interpolated onto a regular two-dimensional grid, such as a 100x100 matrix. Then, pixel value mapping is performed. This step determines the maximum and minimum thickness values in the data and linearly maps this range to the grayscale range of the image, for example, 0 to 255. For example, the minimum thickness value is mapped to grayscale 0 (black), the maximum thickness value is mapped to 255 (white), and the intermediate values are proportionally distributed. In this way, areas with higher thickness will appear brighter on the image, and vice versa. Finally, a complete thickness distribution map is generated, which can intuitively show the thickness distribution of the entire component surface coating without data gaps.

[0078] In summary, the present application starts with precise three-dimensional real scene modeling of complex curved surface components through multi-source data fusion. Based on this, adaptive optimization and active control of the measurement posture are further realized, thereby ensuring signal quality at the source of data acquisition. On this basis, the method further integrates a multi-level, closed-loop data processing procedure from signal level distortion correction, spatial domain noise filtering, to final thickness map missing data repair. The above steps are closely linked and logically rigorous, and together constitute a complete technical solution from accurate perception, intelligent measurement to data refinement, effectively solving the precision and automation problems in complex curved surface measurement of the prior art.

[0079] With reference to Figure 2 The second embodiment of the present application provides a coating thickness measurement system based on a multi-source sensor, characterized in that it comprises: a point cloud acquisition module, configured to acquire surface geometric information of a complex curved surface component through a multi-source sensor, and obtain initial three-dimensional point cloud data; a curved surface modeling module, configured to obtain a three-dimensional curved surface model through curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data; a posture planning module, configured to calculate the best incident angle of the multi-source sensor relative to each region of the curved surface according to the three-dimensional curved surface model, and obtain dynamic posture adjustment parameters; a signal acquisition module, configured to control the multi-source sensor to adjust the three-axis posture in real time according to the dynamic posture adjustment parameters, acquire coating thickness related signals, and obtain original measurement data; a data correction module, configured to perform time domain filtering and spatial domain smoothing on the original measurement data, and generate a set of corrected measurement data; a thickness calculation module, configured to combine the three-dimensional curved surface model and the corrected measurement data to calculate a set of preliminary thickness distribution data; a result generation module, configured to perform data completion processing on the preliminary thickness distribution data, and fuse data obtained through multiple measurements to obtain a final coating thickness measurement result.

[0080] It should be noted that the coating thickness measurement system based on multiple source sensors provided by the embodiments of the present application is used to perform all process steps of the coating thickness measurement method based on multiple source sensors of the above-mentioned embodiments, and the working principles and beneficial effects of the two are one-to-one correspondence, thus not being repeated.

[0081] The embodiments of the present application further provide an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, a point cloud acquisition program. The processor implements the steps in the above various coating thickness measurement methods based on multiple source sensors when executing the computer program, for example Figure 1 the step S11 shown. Alternatively, the processor implements the functions of each module / unit in the above various device embodiments when executing the computer program, for example, the data acquisition module.

[0082] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0083] The electronic device can be a desktop computer, a notebook, a palm computer, and a smart tablet, etc. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0084] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0085] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0086] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0087] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0088] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A multi-source sensor based coating thickness measurement method, characterized in that, The method comprises the following steps: acquiring initial three-dimensional point cloud data; obtaining a three-dimensional curved surface model through curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data; calculating the optimal incident angle of a multi-source sensor relative to each region of the curved surface according to the three-dimensional curved surface model, and calculating a dynamic attitude adjustment parameter according to the optimal incident angle; adjusting the three-axis attitude and data acquisition frequency of the multi-source sensor in real time according to the dynamic attitude adjustment parameter, and collecting coating thickness related signals to obtain original measurement data; performing frequency domain analysis on the original measurement data to determine whether there is reflection distortion, and if there is, performing time domain filtering and spatial domain smoothing to obtain corrected measurement data; combining the three-dimensional curved surface model and the corrected measurement data to calculate the coating thickness value of each measurement point and obtain preliminary thickness distribution data; detecting whether there is a data missing area by using Fourier transform for the preliminary thickness distribution data, and if a missing area is found, interpolating and completing the preliminary thickness distribution data to generate a complete thickness distribution map.

2. The multi-source sensor based coating thickness measurement method of claim 1, wherein, The method of obtaining a three-dimensional curved surface model through curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data comprises the following steps: performing denoising processing on the initial three-dimensional point cloud data to obtain smoothed point cloud data; performing region segmentation on the smoothed point cloud data to obtain segmented point cloud regions; extracting the covariance matrix corresponding to the segmented point cloud regions, performing principal component analysis to obtain a curvature radius set and a normal vector direction set of each region; performing surface fitting on the segmented point cloud regions according to the curvature radius set and the normal vector direction set to generate the three-dimensional curved surface model.

3. The multi-source sensor based coating thickness measurement method of claim 1, wherein, The method of calculating the optimal incident angle of the multi-source sensor relative to each region of the curved surface according to the three-dimensional curved surface model to obtain a dynamic attitude adjustment parameter comprises the following steps: extracting point cloud data of the three-dimensional curved surface model, performing local fitting on the point cloud data to calculate the curvature radius value and the normal vector direction vector of each curved surface region, and obtaining a curvature radius value set and a normal vector direction vector set; performing sub-region division according to the curvature radius value set and the normal vector direction vector set to obtain a segmented sub-region set; extracting the curvature radius value and the normal vector direction vector of each sub-region in the segmented sub-region set, and iteratively calculating the optimal incident angle of the multi-source sensor relative to the sub-region to obtain an optimal incident angle set; calculating the dynamic attitude adjustment parameter of the multi-source sensor according to the optimal incident angle set to obtain a dynamic attitude adjustment parameter.

4. The multi-source sensor based coating thickness measurement method of claim 3, wherein, The method of adjusting the three-axis attitude and data acquisition frequency of the multi-source sensor in real time according to the dynamic attitude adjustment parameter, and collecting coating thickness related signals to obtain original measurement data comprises the following steps: performing deviation compensation and weighted average fusion on the preliminary measurement data collected by the multi-source sensor to obtain a comprehensive signal; evaluating the reflection signal strength of the comprehensive signal, and when the reflection signal strength is lower than a preset strength threshold, optimizing and calculating the dynamic attitude adjustment parameter to obtain an optimized attitude control parameter; According to the optimized attitude control parameter, the three-axis attitude of the multi-source sensor is adjusted, and the data acquisition frequency of the multi-source sensor is dynamically adjusted according to the curvature radius value of each sub-region to collect coating thickness related signals and obtain original measurement data.

5. The multi-source sensor based coating thickness measurement method of claim 1, wherein, The original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, time domain filtering and spatial domain smoothing are performed to obtain corrected measurement data, including: The original measurement data is subjected to frequency domain analysis to determine whether there is reflection distortion, and if so, time domain filtering is performed on the original measurement data to obtain first corrected data; The first corrected data is subjected to spatial domain smoothing processing to obtain smoothed data; The smoothed data is subjected to cross-source consistency verification to determine an abnormal point set; The abnormal point set is subjected to mean clustering analysis, and outliers are removed according to the analysis result to obtain corrected measurement data.

6. The multi-source sensor based coating thickness measurement method of claim 1, wherein, The three-dimensional curved surface model and the corrected measurement data are combined to calculate the coating thickness value of each measurement point to obtain preliminary thickness distribution data, including: The three-dimensional space coordinates and surface normal vector directions of each measurement point are extracted from the corrected measurement data; According to the three-dimensional space coordinates and surface normal vector directions, and in combination with the base geometric information of the three-dimensional curved surface model, the vertical distance of each measurement point from the curved surface base is calculated to obtain the coating thickness value; It is judged whether the coating thickness value exceeds a preset thickness threshold, and values exceeding the thickness threshold are extracted to obtain an abnormal value set; The abnormal value set is subjected to filtering processing to obtain corrected coating thickness values; The corrected coating thickness values are subjected to spatial interpolation to generate a continuous distribution of thickness values to obtain preliminary thickness distribution data.

7. The multi-source sensor based coating thickness measurement method of claim 1, wherein, For the preliminary thickness distribution data, Fourier transform is used to detect whether there is a data missing area, and if a missing area is found, the preliminary thickness distribution data is interpolated to complete the generation of a complete thickness distribution map, including: The original signal in the preliminary thickness distribution data is subjected to Fourier transform to obtain the frequency domain signal of the thickness distribution data, and the amplitude spectrum and phase spectrum of the frequency domain signal are calculated; If the value of the frequency point in the amplitude spectrum of the frequency domain signal is lower than a preset amplitude threshold, it is determined that there is a data missing or abnormal area, and the distribution position of the data missing area and the abnormal area is obtained; According to the distribution position of the area, the preliminary thickness distribution data is subjected to denoising processing to obtain denoised thickness distribution data; From the denoised thickness distribution data, the boundary point coordinates of the data missing and abnormal area are extracted; The boundary point coordinates in the distribution position of the data missing and abnormal area are subjected to interpolation calculation to obtain the completed thickness distribution data; According to the completed thickness distribution data, two-dimensional grid points are extracted, the two-dimensional grid points are mapped to pixel values, and a complete thickness distribution map is generated.

8. A multi-source sensor based coating thickness measurement system, characterized by, It includes: A point cloud acquisition module is configured to acquire surface geometric information of a complex curved surface component through a multi-source sensor to obtain initial three-dimensional point cloud data; a curved surface modeling module, configured to obtain a three-dimensional curved surface model by curvature distribution analysis and normal vector estimation according to the initial three-dimensional point cloud data; a pose planning module, configured to calculate optimal incident angles of the multi-source sensor relative to regions of the curved surface according to the three-dimensional curved surface model, and obtain dynamic pose adjustment parameters; a signal acquisition module, configured to control the multi-source sensor to adjust a three-axis pose in real time according to the dynamic pose adjustment parameters, acquire a coating thickness related signal, and obtain original measurement data; a data correction module, configured to perform time domain filtering and spatial domain smoothing on the original measurement data, and generate a set of corrected measurement data; a thickness calculation module, configured to calculate a set of preliminary thickness distribution data in combination with the three-dimensional curved surface model and the corrected measurement data; a result generation module, configured to perform data completion processing on the preliminary thickness distribution data, and fuse data obtained through multiple measurements, to obtain a final coating thickness measurement result.

9. An electronic device, comprising: A computer readable storage medium includes a computer program stored therein, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the multi-source sensor based coating thickness measurement method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a computer program stored therein, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the multi-source sensor based coating thickness measurement method according to any one of claims 1 to 7.

Images