Method for detecting thickness increment of sprayed coating

Abstract

The invention provides a method for detecting the thickness increment of a sprayed coating, and relates to the technical field of industrial spraying. The method comprises the following steps: acquiring a depth image of a current sprayed coating, and determining first point cloud data of the current sprayed coating according to the depth image; registering the first point cloud data with second point cloud data, wherein the second point cloud data is depth point cloud data of an initial spraying coating corresponding to the current spraying coating; determining the offset of matching points in the first point cloud data and the second point cloud data according to a registration result; and determining the thickness increment of the current sprayed coating according to the offset of the matching point. According to the invention, the thickness increment of the sprayed coating can be accurately and rapidly detected.

Description

Technical Field

[0001] This disclosure relates to the field of industrial spraying technology, and more specifically, to a method for detecting the thickness increment of a sprayed coating. Background Technology

[0002] In the field of industrial spraying, coating thickness is a crucial indicator determining a material's corrosion resistance, electrical conductivity, and other properties. Detecting the thickness increment of the sprayed coating over a period of time is a core step in the spraying process. However, existing industrial spraying technologies suffer from slow thickness increment detection speed and low accuracy.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this disclosure is to provide a method for detecting the thickness increment of sprayed coatings, thereby overcoming, to at least some extent, the problems of slow speed and low efficiency in the process of detecting the thickness increment of sprayed coatings.

[0005] According to a first aspect of this disclosure, a method for detecting the thickness increment of a sprayed coating is provided, characterized by comprising: acquiring a depth image of the current sprayed coating; determining first point cloud data of the current sprayed coating based on the depth image; registering the first point cloud data with second point cloud data, wherein the second point cloud data is the depth point cloud data of an initial sprayed coating corresponding to the current sprayed coating; determining the offset of matching points in the first point cloud data and the second point cloud data based on the registration result; and determining the thickness increment of the current sprayed coating based on the offset of the matching points.

[0006] Optionally, registering the first point cloud data with the second point cloud data includes: performing a first registration process on the first point cloud data and the second point cloud data to obtain pose transformation parameters of the first point cloud data relative to the second point cloud data; and performing a second registration process on the first point cloud data and the second point cloud data based on the pose transformation parameters using a probability distribution matching algorithm.

[0007] Optionally, a first registration process is performed on the first point cloud data and the second point cloud data to obtain the pose transformation parameters of the first point cloud data relative to the second point cloud data. This includes: obtaining first local geometric feature descriptors corresponding to multiple points in the first point cloud data and the second point cloud data; performing multi-scale feature fusion and dynamic neighborhood adjustment on the first local geometric feature descriptors corresponding to each point to generate second local geometric feature descriptors corresponding to each point; performing nearest neighbor search on multiple points in the second point cloud data based on the second local geometric feature descriptors corresponding to multiple points in the first point cloud data to determine matching point pairs between the first point cloud data and the second point cloud data; and determining the pose transformation parameters of the first point cloud data relative to the second point cloud data based on the matching point pairs using a random sampling consensus algorithm.

[0008] Optionally, based on the pose transformation parameters, a second registration process is performed on the first point cloud data and the second point cloud data using a probability distribution matching algorithm. This includes: voxelizing the first point cloud data and the second point cloud data respectively to determine multiple voxels; constructing a Gaussian distribution model for each voxel to characterize each voxel; determining the sum of Mahalanobis distances between each voxel in the first point cloud data and the corresponding voxel in the second point cloud data using a probability distribution matching algorithm based on the Gaussian distribution model of each voxel; and optimizing the pose transformation parameters using the Gauss-Newton method to complete the second registration process between the first point cloud data and the second point cloud data. The optimization of the pose transformation parameters using the Gauss-Newton method is implemented using a parallel computing architecture.

[0009] Optionally, after determining the thickness increment of the current sprayed coating based on the offset of the matching point, the thickness increment detection method of the sprayed coating further includes: inputting the thickness increment into a pre-trained thickness error compensation model, the thickness error compensation model being determined based on a 3D convolutional neural network; processing the thickness increment based on the thickness error compensation model to output the thickness increment error; and adding the thickness increment and the thickness increment error to determine the corrected thickness increment.

[0010] Optionally, the thickness increment detection method further includes: mapping the thickness increment to a numerical distribution model in three-dimensional space; mapping the numerical distribution model to a color distribution model; and rendering the numerical distribution model in real time using real-time rendering technology to generate a color distribution map reflecting the thickness increment.

[0011] Optionally, before acquiring the depth image of the current sprayed coating, the thickness increment detection method further includes: acquiring calibration plate pose data, the calibration plate pose data being acquired based on a depth image acquisition device, the depth image acquisition device also being used to acquire the depth image of the current sprayed coating; the depth image acquisition device being mounted on a robotic arm; establishing a device coordinate system and a robotic arm coordinate system based on the depth image acquisition device and the robotic arm respectively, and establishing an initial transformation relationship between the device coordinate system and the robotic arm coordinate system based on the calibration plate pose data; and optimizing the initial transformation relationship using a nonlinear optimization method to complete the hand-eye calibration process of the depth image acquisition device and the robotic arm.

[0012] Optionally, the thickness increment detection method further includes: real-time monitoring of the joint temperature of the robotic arm and establishing a thermal expansion coefficient model for the robotic arm; obtaining the deformation of the robotic arm based on the thermal expansion coefficient model; and dynamically compensating for the drift of the robotic arm coordinate system based on the deformation to correct the results of the hand-eye calibration process.

[0013] According to a second aspect of this disclosure, a method for controlling a spraying equipment is provided, characterized in that it includes: determining the thickness increment of the current sprayed coating using any of the above-mentioned thickness increment detection methods; obtaining the ideal thickness increment of the current sprayed coating; determining a thickness increment deviation based on the current thickness increment of the sprayed coating and the ideal thickness increment of the current sprayed coating; and controlling the spraying equipment based on the thickness increment deviation.

[0014] Optionally, the spraying equipment is controlled based on the thickness increment deviation, including: when the thickness increment deviation is greater than a first deviation threshold, increasing the spray gun moving speed of the spraying equipment and reducing the paint flow rate of the spraying equipment; when the thickness increment deviation is less than a second deviation threshold, reducing the spray gun moving speed of the spraying equipment and increasing the paint flow rate of the spraying equipment; and when the thickness increment deviation is greater than or equal to the second deviation threshold and less than or equal to the first deviation threshold, adjusting the process parameters of the spraying equipment in real time.

[0015] In some embodiments of this disclosure, the technical solutions involve acquiring a depth image of the current sprayed coating and generating point cloud data, then registering and comparing this data with a reference point cloud of the initial coating, and calculating the thickness increment based on the offset of the matching points. On one hand, this disclosure achieves accurate detection of the thickness increment of the sprayed coating; on the other hand, the matching process using voxel segmentation during the registration of the first and second point cloud data reduces the computational load.

[0016] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0018] Figure 1 A schematic diagram of an exemplary system architecture for a method for detecting the thickness increment of a sprayed coating and a method for controlling a spraying equipment according to an exemplary embodiment of the present disclosure is shown.

[0019] Figure 2 The diagram schematically illustrates the overall steps of a method for detecting the thickness increment of a sprayed coating and a method for controlling a spraying equipment according to an exemplary embodiment of the present disclosure.

[0020] Figure 3 A flowchart illustrating a method for detecting the thickness increment of a sprayed coating according to an exemplary embodiment of the present disclosure is shown.

[0021] Figure 4 A logic diagram illustrating the thickness increment detection process of a sprayed coating according to an exemplary embodiment of the present disclosure is shown.

[0022] Figure 5 The diagram schematically illustrates the steps of a method for detecting the thickness increment of a sprayed coating according to an exemplary embodiment of the present disclosure, including correcting the thickness increment based on a thickness increment error.

[0023] Figure 6 A flowchart illustrating a spraying equipment control method according to an exemplary embodiment of the present disclosure is shown. Detailed Implementation

[0024] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0025] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0026] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances. Furthermore, all terms such as "first," "second," etc., used below are for distinction purposes only and should not be construed as limiting the scope of this disclosure.

[0027] like Figure 1 As shown, the system architecture of the exemplary embodiments of this disclosure may include a depth image acquisition device 1001, a data processing device 1002, a spraying device 1003, and a display device 1004.

[0028] Image acquisition device 1001 is used to acquire depth images and can be a binocular fringe structured light camera. The camera has a resolution of 2448×2048, a frame rate of 60fps, and a blue light wavelength of 450±5nm; the Z-axis repeatability is 2μm, and the auxiliary subsystem uses DLP (Digital Light Processing) blue light projection. Image acquisition device 1001 can acquire first point cloud data and second point cloud data.

[0029] The data processing device 1002 can be used to perform a first registration process and a second registration process on the first point cloud data and the second point cloud data. It can also be used to preprocess the first point cloud data and the second point cloud data before registration, and finally calculate the corrected thickness increment. Based on the thickness increment of the current sprayed coating and the ideal thickness increment of the current process standard, the thickness increment deviation is determined to control the spraying device 1003.

[0030] Display device 1004 can be used to visualize data by displaying a color distribution map. The process that can be displayed by display device 1004 in the thickness increment detection of the sprayed coating according to an exemplary embodiment of this disclosure includes: mapping the thickness increment to a numerical distribution model in three-dimensional space; mapping the numerical distribution model to a color distribution model; and rendering the numerical distribution model in real time using real-time rendering technology to generate a color distribution map reflecting the thickness increment. In addition, both the first registration process and the second registration process can be displayed by display device 1004.

[0031] like Figure 2 As shown in the exemplary embodiments of this disclosure, the front-end portion may include phase extraction, image acquisition and processing, and system calibration. The specific processes of system calibration and phase extraction are described later. The front-end portion can be used for data acquisition and preprocessing, providing high-quality input data for thickness increment calculation. In the phase extraction stage, the system acquires the phase information of the sprayed surface through optical methods such as structured light projection or laser interferometry, and uses phase shift algorithms or Fourier transforms to demodulate a high-precision phase distribution, thereby characterizing the microstructure of the coating surface. The image acquisition and processing module uses a high-resolution industrial camera to simultaneously capture depth images of the sprayed area, eliminates interference from splash particles, environmental dust, etc. through statistical filtering and radius filtering, and combines multi-view image fusion technology to improve the integrity of the three-dimensional point cloud. System calibration is the core link of the front-end portion. Through a checkerboard calibration plate or a precision displacement platform, the mapping relationship between the camera-projector coordinate system and the workpiece coordinate system is established, lens distortion is corrected, and the conversion parameters between pixel scale and actual physical size are determined to ensure the geometric accuracy of subsequent point cloud data. The accuracy of the front-end portion directly affects the calculation results of the thickness increment in the back-end.

[0032] The backend is used for post-processing of thickness increments. It focuses on the calculation and optimization of thickness increments. Based on the registered point cloud data provided by the frontend, the normal distance between matching point pairs is first calculated using a point cloud differencing algorithm to generate an initial thickness increment distribution map. Then, statistical filtering (such as Gaussian smoothing or median filtering) is used to eliminate local outliers, and a region growing algorithm is combined to segment abnormal regions (such as spray splatter or bubbles). For workpieces with large curvatures in industrial scenarios, surface fitting compensation is introduced to map the thickness increment onto the theoretical model, eliminating systematic errors introduced by the workpiece shape. Finally, a time-series analysis module performs trend fitting on the incremental data from multiple spraying operations, outputting a thickness increment uniformity report, maximum / minimum increment values, and standard deviation, among other key indicators, and displays a 3D thickness increment cloud map through a visualization interface. The backend uses a closed-loop feedback mechanism to transmit the location and deviation of out-of-tolerance areas to the spraying robot, enabling dynamic adjustment of process parameters.

[0033] Figure 3 A flowchart illustrating an exemplary embodiment of a method for detecting the thickness increment of a sprayed coating according to the present disclosure is shown schematically. Reference Figure 3 The method for detecting the thickness increment of a sprayed coating may include the following steps:

[0034] S30. Obtain the depth image of the current sprayed coating, and determine the first point cloud data of the current sprayed coating based on the depth image.

[0035] According to an exemplary embodiment of this disclosure, the current sprayed coating is the target sprayed coating to be detected, that is, the current sprayed coating is the end point of the thickness increment of the sprayed coating. Theoretically, the thickness increment detection method of the sprayed coating according to an exemplary embodiment of this disclosure can be used to obtain any two depth images of the sprayed coating and calculate the thickness increment of the corresponding sprayed plane.

[0036] Depth images are a special type of image that needs to be acquired using devices such as structured light cameras and sensors. Taking a structured light camera as an example, it can project a coded grating onto the surface of a workpiece, capture deformed fringes using a multi-view camera, and generate a high-precision depth map using a phase-shifting algorithm.

[0037] According to an exemplary embodiment of this disclosure, the system can be calibrated before acquiring depth images. The calibration process includes: placing a calibration board in the field of view, continuously changing its position and orientation to cover as many positions as possible within the field of view, and covering the height of the measurement space. A set of images is acquired for each pose of the calibration board, with at least four sets acquired. The world coordinate system is established by taking the center of the first circle at the top left corner of the calibration board as the origin, defining the x-axis horizontally to the right and the y-axis vertically downwards, and determining the z-axis direction using the right-hand rule. Based on the relationship between the world coordinates of key points on the calibration board and their corresponding points in the pixel coordinate system, the homography matrix between the two coordinate systems is calculated. The camera's intrinsic and extrinsic parameters and distortion parameters are solved using the homography matrix calculated from multiple sets of data. The camera's intrinsic and extrinsic parameters and distortion parameters are updated and optimized using the Levenberg-Marquardt optimization algorithm.

[0038] After the system is calibrated, the real-time pose data of the robotic arm can be received via the EtherCAT bus (Ethernet for Control Automation Technology). The trigger signal is linked with the spray gun switch valve to ensure that the density of paint atomized particles is <1000ppm at the time of image acquisition. Finally, the phase information is extracted from the stripe pattern through image processing algorithm, and the phase principal value is calculated using the phase shift method.

[0039] The absolute phase value is extracted from the modulated fringes, followed by stereo matching. Finally, the depth (Z-axis coordinate) of each point on the object's surface is calculated based on the principle of triangulation. The calculation formula is:

[0040]

[0041] Where B is the baseline distance, f is the focal length, and Δd is the view phase difference. After the calculation is completed, the depth data can be converted into a 3D point cloud or mesh model. For the sake of consistency, the results obtained in this step are collectively referred to as the first point cloud data and the second point cloud data in this scheme.

[0042] S32. Register the first point cloud data with the second point cloud data, where the second point cloud data is the depth point cloud data of the initial sprayed coating corresponding to the current sprayed coating.

[0043] According to an exemplary embodiment of this disclosure, the process of acquiring the first point cloud data has been mentioned above and will not be repeated here. The second point cloud data is the depth point cloud data of the initial sprayed coating relative to the first point cloud data. The initial sprayed coating corresponding to the second point cloud data can be a plane that has not undergone any spraying or a plane that has undergone spraying.

[0044] It is important to note that in order to achieve high-precision spatiotemporal synchronization of multi-source data, hand-eye calibration of the depth image acquisition device and the robotic arm is required before registering the first point cloud data with the second point cloud data.

[0045] First, a reference for hand-eye calibration is established, and the origin of the workpiece coordinate system is constructed. A phase-type laser tracker is used to measure the spatial pose of the calibration plate, establishing a global coordinate system reference (accuracy ±0.01mm). Then, the robotic arm carrying a structured light camera moves along a preset trajectory, collecting 20 sets of point cloud data of the calibration plate in different poses. The initial value of the camera-robotic arm transformation matrix is ​​solved using the Tsai-Lenz algorithm, and then the parameters are iteratively corrected using the Levenberg-Marquardt nonlinear optimization algorithm. The objective function is to minimize the back projection error.

[0046]

[0047] Where π is the camera projection model, and the calibration residual after optimization is ≤0.02mm.

[0048] According to exemplary embodiments of this disclosure, a dynamic compensation mechanism can be added simultaneously during the hand-eye calibration process to monitor the joint temperature of the robotic arm in real time (using a PT100 sensor, ±0.1℃) and compensate for coordinate system drift caused by robotic arm deformation using a thermal expansion coefficient model. In addition, a lightweight calibration can be automatically performed every 8 hours to ensure spatial alignment stability during long-term operation.

[0049] According to an exemplary embodiment of this disclosure, the process of registering the first point cloud data with the second point cloud data may include: performing a first registration process on the first point cloud data and the second point cloud data to obtain pose transformation parameters of the first point cloud data relative to the second point cloud data; and performing a second registration process on the first point cloud data and the second point cloud data based on the pose transformation parameters using a probability distribution matching algorithm.

[0050] According to an exemplary embodiment of this disclosure, the first registration process may include: acquiring first local geometric feature descriptors corresponding to multiple points in first point cloud data and second point cloud data; performing multi-scale feature fusion and dynamic neighborhood adjustment on the first local geometric feature descriptors corresponding to each point to generate second local geometric feature descriptors corresponding to each point; performing nearest neighbor search on multiple points in second point cloud data based on the second local geometric feature descriptors corresponding to multiple points in first point cloud data to determine matching point pairs between first point cloud data and second point cloud data; and determining pose transformation parameters of first point cloud data relative to second point cloud data based on the matching point pairs using a random sampling consensus algorithm.

[0051] The first local geometric feature descriptor can be an FPFH descriptor (Fast Point Feature Histogram). Multi-scale feature fusion and dynamic neighborhood adjustment are performed on the first local geometric feature descriptors corresponding to each point to generate a second local geometric feature descriptor corresponding to each point, which can be an improved weighted SPFH descriptor (Weighted Simplified Point Feature Histogram). The specific process is as follows:

[0052] Traditional FPFH constructs a histogram using three geometric angles, where each angle is defined (for point p and its neighboring point q):

[0053] α = arctan(n) q ·(u×v),u·v)

[0054]

[0055] θ = arctan(n1·(u×w),u·w)

[0056] Where u = qp, v = n1, w = n q n1 and n q It is the normal vector.

[0057] The improved weighted SPFH calculation introduces distance weights. The corrected angle statistics are as follows:

[0058]

[0059] Where d = ||qp||, and σ is a scaling parameter that controls the rate of weight decay.

[0060] For each point p at K scales {r1, r2, ..., r... K} Calculate SPFH and perform weighted fusion:

[0061]

[0062] Where, γ k The scale weights can be set according to task requirements (e.g., γ). k =1 / K).

[0063] The radius is dynamically adjusted based on the local point density ρ(p) (number of points per unit volume):

[0064]

[0065] Where, r base Based on the radius, ρ max This represents the maximum density of the scene.

[0066] According to an exemplary embodiment of this disclosure, the second registration process may include: voxelizing the first point cloud data and the second point cloud data respectively to determine multiple voxels; constructing a Gaussian distribution model for each voxel to characterize each voxel; determining the sum of Mahalanobis distances between each voxel in the first point cloud data and the corresponding voxel in the second point cloud data using a probability distribution matching algorithm based on the Gaussian distribution model of each voxel; and optimizing the pose transformation parameters using the Gauss-Newton method to complete the second registration process between the first point cloud data and the second point cloud data; wherein the process of optimizing the pose transformation parameters using the Gauss-Newton method is implemented based on a parallel computing architecture.

[0067] The probability distribution matching algorithm can be VGICP (Voxelized Generalized Iterative Closest Point), an improved version of ICP (Iterative Closest Point) algorithm. VGICP is a point cloud registration algorithm that integrates voxelization strategy and probability distribution matching, aiming to solve the problems of low computational efficiency in traditional ICP methods and sensitivity to voxel parameters in NDT (Normal Distributions Transform) methods. Its core idea is to achieve efficient and robust distribution-to-distribution matching through multi-point distribution aggregation within voxels. The specific process is as follows:

[0068] The probabilistic model of a point cloud is represented using a single-point Gaussian distribution model:

[0069] Let p be any point in the point cloud. i The geometric uncertainty is determined by the local surface properties, and its location follows a Gaussian distribution:

[0070] p i ~N(μ) i , ∑ i )

[0071] Where μ i Let ∑ be the coordinates of the point (observation value). i The covariance matrix is ​​obtained using nearest neighbor PCA:

[0072]

[0073] K is the number of nearest neighbors. It is the neighborhood centroid.

[0074] Intravoxel distribution aggregation theory:

[0075] For all points {p1, p2, ..., p} within voxel V n Regenerative polymerization utilizing Gaussian distribution:

[0076]

[0077] Even if n=1 (single-point voxel), the distribution is still valid: N V ~N(μ) i , ∑ i )

[0078] Objective function construction: Maximum likelihood estimation:

[0079] After transformation T(·), the source voxel distribution N is defined. s With the target voxel distribution N t The matching probability is a negative exponent of the Mahalanobis distance:

[0080]

[0081] Where the Mahalanobis distance is:

[0082]

[0083] Maximize the overall log-likelihood:

[0084]

[0085] This is equivalent to minimizing the sum of Mahalanobis distances.

[0086] Optimization solution using the Gauss-Newton method:

[0087] Let the transformation T be represented by the Lie algebra ξ∈se(3):

[0088] T(μ)=exp(ξ ∧ )·μ

[0089] Where exp(·) is the exponential mapping, ξ ∧ It is the cross product operator.

[0090] Perform iterative optimization: Let the current transformation parameter be ξ, and solve for the increment Δξ:

[0091] Define residual:

[0092]

[0093] Jacobian matrix:

[0094]

[0095] Information matrix:

[0096]

[0097] Incremental equations:

[0098]

[0099] Parameter update:

[0100] ξ←ξ+Δξ

[0101] The objective function can be decomposed into:

[0102]

[0103] The Hessian block and gradient block are calculated separately. The summation of the Hessian block and gradient block is accelerated by GPU (Graphics Processing Unit) based on a parallel computing architecture.

[0104] S34. Determine the offset of the matching points in the first point cloud data and the second point cloud data based on the registration results.

[0105] According to an exemplary embodiment of this disclosure, after completing the first registration process and the second registration process according to the above steps, the thickness increment of the sprayed coating can be calculated, and the specific process is as follows:

[0106] By establishing a KD-tree (k-dimensional tree) index, the nearest neighbor relationship is built point by point. Finally, the thickness increment is calculated and the thickness increment matrix is ​​generated using the following mathematical expression:

[0107] ΔT i =||P i -Q j||2

[0108] Δn=Δd·N

[0109] Where N is the original surface normal vector, Q j =arg min ··CAD ||P i -Q‖2.

[0110] S36. Determine the thickness increment of the current sprayed coating based on the offset of the matching point.

[0111] According to exemplary embodiments of this disclosure, after generating the thickness increment matrix, it can be based on single-point thickness increment, global thickness increment, average thickness increment, or thickness increment distribution. For example, Δd i Let P be the point i The local thickness increment at that location.

[0112] The logic diagram of the thickness increment detection process of the sprayed coating according to an exemplary embodiment of the present disclosure is as follows: Figure 4 As shown, after inputting the first and second point cloud data, the acquired point cloud data can be preprocessed. The preprocessing process may include: using the Open3D library (an open-source 3D data processing library) for statistical filtering (removing outliers that deviate from the standard deviation of the point cloud mean by more than twice) and radius filtering (removing sparse noise points with a radius of 0.5 mm) to eliminate interference from spray splatter particles, environmental dust, etc., thereby improving data quality.

[0113] After preprocessing is completed, the processed first and second point cloud data undergo a first registration process and a second registration process. The specific processes of the first registration process, the second registration process, and the thickness increment calculation have been explained above, so they will not be repeated here.

[0114] According to an exemplary embodiment of this disclosure, in order to meet the real-time visualization requirements of online monitoring of coating thickness increment, a dedicated plug-in module can be built based on CloudCompare (an open-source 3D point cloud processing platform) to achieve multi-dimensional fusion visualization of coating thickness increment distribution, process parameters and defect characteristics.

[0115] After the first registration process, the second registration process, the thickness increment calculation process, and the deep learning error correction process, the corrected thickness increment is output.

[0116] The first registration process, the second registration process, and the thickness increment calculation process can all be visualized using the display device 1004. The specific steps include: mapping the thickness increment to a numerical distribution model in three-dimensional space; mapping the numerical distribution model to a color distribution model; and rendering the numerical distribution model in real time using real-time rendering technology to generate a color distribution map that reflects the thickness increment.

[0117] According to an exemplary embodiment of this disclosure, the process of visualizing the thickness increment method of a sprayed coating may include constructing a thickness increment-hue mapping model based on the HSV (Hue-Saturation-Value) color space. The construction process of this model may include:

[0118] The numerical range of the scalar field representing the thickness increment is linearly mapped to the hue value. The minimum thickness increment corresponds to blue (H = 240°), and the maximum thickness increment corresponds to red (H = 0°), forming a gradient color spectrum of blue-green-yellow-red, which visually reflects the spatial distribution characteristics of the coating thickness increment. OpenGL (Open Graphics Library) shader technology is used to dynamically update the RGB (Red-Green-Blue) attributes of the point cloud, ensuring that the thickness increment changes are displayed in real time during the spraying process.

[0119] For defect markers, abnormal areas are manually selected. The system automatically extracts local point clouds and constructs a NURBS (Non-Uniform Rational B-Spline) surface. The system then optimizes the fit to an ideal surface morphology using control points. A defect analysis report is generated, including parameters such as maximum indentation depth, curvature deviation, and defect area ratio. Simultaneously, a heatmap visually displays the deviation between the fitted surface and the measured point cloud.

[0120] The steps of correcting the thickness increment based on the thickness increment error in the spray coating thickness increment detection method according to an exemplary embodiment of the present disclosure are shown in the figure below. Figure 5 As shown, the specific steps include: inputting the thickness increment into a pre-trained thickness error compensation model, which is determined based on a 3D convolutional neural network; processing the thickness increment based on the thickness error compensation model to output the thickness increment error; and adding the thickness increment and the thickness increment error to determine the corrected thickness increment.

[0121] The input registered point cloud data includes local point cloud patches and additional features (coordinates of corresponding points in the reference point cloud, spraying parameters). A 3D convolutional neural network (ResNet-3D) based on residual learning is used to solve the gradient vanishing problem and improve the deep feature extraction capability, resulting in a residual learning prediction Δ. Finally, the corrected thickness increment is obtained by adding the initial deformation variable obtained in S4 to the predicted error deformation variable Δ. The design process of the loss function is as follows:

[0122] Joint optimization of geometric error and surface smoothness constraints:

[0123] L=α·MAE(Δ pred ,Δ true )+β·TV(S)

[0124] MAE (Mean Absolute Error) is used to directly constrain the error between the predicted correction Δ and the true value, while TV (Total Variation Regularization) is used to force the predicted thickness increment field to be spatially smooth, avoiding abrupt changes.

[0125]

[0126] Where S is the thickness field matrix, α = 1.0, β = 0.1, and it can be trained using the AdamW optimizer (an adaptive moment estimation weight decay optimizer) with a learning rate of 10. -3 The decay weight is 10. -4 The learning cycle is 50 epochs (1 epoch means that all training samples are learned once).

[0127] Figure 6 A flowchart illustrating a spraying equipment control method according to an exemplary embodiment of the present disclosure is shown schematically. Referring to 6, the spraying equipment control method for spraying a coating may include the following steps:

[0128] S60. Determine the thickness increment of the current sprayed coating.

[0129] According to exemplary embodiments of this disclosure, the method for determining the thickness increment of the current sprayed coating may include Figure 3 The specific steps of the method for detecting the thickness increment of the sprayed coating have been mentioned above and will not be repeated here.

[0130] S62. Obtain the ideal thickness increment for the current sprayed coating.

[0131] According to an exemplary embodiment of this disclosure, the ideal thickness increment of the current sprayed coating can be a process parameter, a fixed value, which is the ideal thickness increment of the current sprayed coating compared to the initial sprayed coating. Therefore, the ideal thickness increment can be used as a standard to determine the thickness increment deviation.

[0132] S64. Determine the thickness increment deviation based on the current thickness increment of the sprayed coating and the ideal thickness increment of the current sprayed coating.

[0133] According to an exemplary embodiment of this disclosure, the thickness increment deviation is the result of subtracting the current thickness increment of the sprayed coating from the ideal thickness increment. Therefore, if the current thickness increment of the sprayed coating is higher than the ideal thickness increment, the thickness increment deviation is positive; if the current thickness increment of the sprayed coating is equal to the ideal thickness increment, the thickness increment deviation is zero; and if the current thickness increment of the sprayed coating is lower than the ideal thickness increment, the thickness increment deviation is negative.

[0134] S66. Control the spraying equipment based on the thickness increment deviation.

[0135] According to an exemplary embodiment of this disclosure, the specific process of controlling the spraying equipment based on the thickness increment deviation may include: when the thickness increment deviation is greater than a first deviation threshold, increasing the spray gun moving speed of the spraying equipment and reducing the paint flow rate of the spraying equipment; when the thickness increment deviation is less than a second deviation threshold, reducing the spray gun moving speed of the spraying equipment and increasing the paint flow rate of the spraying equipment; and when the thickness increment deviation is greater than or equal to the second deviation threshold and less than or equal to the first deviation threshold, adjusting the process parameters of the spraying equipment in real time.

[0136] According to an exemplary embodiment of this disclosure, a hierarchical control strategy is established based on the thickness increment detection results and the process knowledge base, and the control logic based on the thickness increment deviation Δh is as follows:

[0137] When Δh>+20μm, increasing the spray gun movement speed by 10% (maximum speed limit 5m / s) and decreasing the paint flow rate by 15% can suppress sagging defects caused by excessive coating thickness by reducing the amount of paint deposited per unit area.

[0138] When Δh < -15μm, the spray gun moving speed is reduced by 8% (minimum speed limit 0.3m / s), and the paint flow rate is increased by 12%, thereby increasing the paint coverage density, avoiding the defect of exposed substrate, and preventing uneven paint atomization caused by excessive speed.

[0139] When +20μm ≥ Δh ≥ -15μm, the output spray gun parameters are adjusted based on a PID (Proportional-Integral-Derivative) controller:

[0140]

[0141] Among them, the proportionality coefficient K p Take 0.8, coefficient K of the integral term i Take 0.2, proportionality coefficient K d A value of 0.1 is used, and the integral term is limited by ±5%. The derivative term uses a four-point center difference method to reduce noise sensitivity. In the PID controller, e(t) represents the thickness increment deviation, and the proportional coefficient K... p The setting is based on the sensitivity to thickness increment deviation. For example, the spray gun speed is adjusted by 0.1% for every 1μm increase in Δh. Integral coefficient K i A small value (e.g., 0.001) should be chosen to avoid integral saturation, while ensuring a ±5% limit to prevent sudden flow changes. Differential coefficient K d For smooth adjustment, K is usually taken. p It is 1 / 10 to 1 / 5 of the original value, suppressing high-frequency noise.

[0142] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0143] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for detecting the thickness increment of a sprayed coating, characterized in that, include: Acquire a depth image of the current sprayed coating, and determine the first point cloud data of the current sprayed coating based on the depth image; The first point cloud data and the second point cloud data are registered together, where the second point cloud data is the depth point cloud data of the initial sprayed coating corresponding to the current sprayed coating. The offset of the matching points in the first point cloud data and the second point cloud data is determined based on the registration result. The thickness increment of the current sprayed coating is determined based on the offset of the matching point.

2. The method for detecting the thickness increment of a sprayed coating according to claim 1, characterized in that, The registration of the first point cloud data with the second point cloud data includes: A first registration process is performed on the first point cloud data and the second point cloud data to obtain the pose transformation parameters of the first point cloud data relative to the second point cloud data. Based on the pose transformation parameters, a second registration process is performed on the first point cloud data and the second point cloud data using a probability distribution matching algorithm.

3. The method for detecting the thickness increment of a sprayed coating according to claim 2, characterized in that, A first registration process is performed on the first point cloud data and the second point cloud data to obtain the pose transformation parameters of the first point cloud data relative to the second point cloud data, including: Obtain the first local geometric feature descriptor corresponding to multiple points in the first point cloud data and the second point cloud data; Multi-scale feature fusion and dynamic neighborhood adjustment are performed on the first local geometric feature descriptors corresponding to each point to generate a second local geometric feature descriptor corresponding to each point. Based on the second local geometric feature descriptors corresponding to multiple points in the first point cloud data, a nearest neighbor search is performed on multiple points in the second point cloud data to determine the matching point pairs between the first point cloud data and the second point cloud data. Based on the matching point pairs, the pose transformation parameters of the first point cloud data relative to the second point cloud data are determined by a random sampling consensus algorithm.

4. The method for detecting the thickness increment of a sprayed coating according to claim 2, characterized in that, Based on the pose transformation parameters, a second registration process is performed on the first point cloud data and the second point cloud data using a probability distribution matching algorithm, including: The first point cloud data and the second point cloud data are voxelized to determine multiple voxels; Gaussian distribution models are constructed for each voxel to characterize each voxel; Based on the Gaussian distribution model of each voxel, the sum of Mahalanobis distances between each voxel in the first point cloud data and the corresponding voxel in the second point cloud data is determined by a probability distribution matching algorithm. The pose transformation parameters are optimized using the Gauss-Newton method to complete the second registration process between the first point cloud data and the second point cloud data; wherein, the process of optimizing the pose transformation parameters using the Gauss-Newton method is implemented based on a parallel computing architecture.

5. The method for detecting the thickness increment of a sprayed coating according to claim 1, characterized in that, After determining the thickness increment of the current sprayed coating based on the offset of the matching point, the thickness increment detection method of the sprayed coating further includes: The thickness increment is input into a pre-trained thickness error compensation model, which is determined based on a 3D convolutional neural network; The thickness increment is processed based on the thickness error compensation model to output the thickness increment error; The thickness increment and the thickness increment error are added together to determine the corrected thickness increment.

6. The method for detecting the thickness increment of a sprayed coating according to claim 1, characterized in that, The thickness increment detection method further includes: The thickness increment is mapped to a numerical distribution model in three-dimensional space; Map the numerical distribution model to the color distribution model; The numerical distribution model is rendered in real time using real-time rendering technology to generate a color distribution map that reflects the thickness increment.

7. The thickness increment detection method according to claim 1, characterized in that, Before acquiring a depth image of the current sprayed coating, the thickness increment detection method further includes: The calibration plate pose data is acquired based on a depth image acquisition device, which is also used to acquire a depth image of the current sprayed coating; the depth image acquisition device is mounted on a robotic arm. Based on the depth image acquisition device and the robotic arm, a device coordinate system and a robotic arm coordinate system are established respectively. An initial transformation relationship between the device coordinate system and the robotic arm coordinate system is established according to the pose data of the calibration plate. The initial transformation relationship is optimized using a nonlinear optimization method to complete the hand-eye calibration process for the depth image acquisition device and the robotic arm.

8. The thickness increment detection method according to claim 7, characterized in that, The thickness increment detection method further includes: The joint temperature of the robotic arm is monitored in real time, and a thermal expansion coefficient model of the robotic arm is established. The deformation of the robotic arm is obtained based on the thermal expansion coefficient model; Dynamic compensation is performed on the drift of the robotic arm coordinate system based on the deformation to correct the results of the hand-eye calibration process.

9. A method for controlling a spraying equipment, characterized in that, include: The thickness increment of the sprayed coating is determined by the thickness increment detection method of any one of claims 1 to 8; Obtain the ideal thickness increment of the current sprayed coating; The thickness increment deviation is determined based on the current thickness increment of the sprayed coating and the ideal thickness increment of the current sprayed coating; The spraying equipment is controlled based on the thickness increment deviation.

10. The spraying equipment control method according to claim 9, characterized in that, Controlling the spraying equipment based on the thickness increment deviation includes: When the thickness increment deviation is greater than the first deviation threshold, the spray gun moving speed of the spraying equipment is increased and the paint flow rate of the spraying equipment is reduced; When the thickness increment deviation is less than the second deviation threshold, the spray gun moving speed of the spraying equipment is reduced and the paint flow rate of the spraying equipment is increased; When the thickness increment deviation is greater than or equal to the second deviation threshold and less than or equal to the first deviation threshold, the process parameters of the spraying equipment are adjusted in real time.

Images