A method of additive manufacturing surface quality analysis

Abstract

The application discloses a kind of additive manufacturing surface quality analysis methods, it is related to 3D printing field, comprising: by layer-by-layer scanning additive manufacturing piece surface, obtain and process point cloud data generation regularized three-dimensional surface height matrix.Based on three-dimensional surface height matrix calculation two-dimensional and three-dimensional roughness parameter, correlation to each layer number, while using curvature analysis to identify and statistics surface defect type, quantity and size.Combining roughness variation trend and defect distribution weighted calculation each layer layering quality index, finally cumulative evaluation obtains the overall surface quality grade of piece.The application has the advantages that: along the additive manufacturing stacking direction layer-by-layer surface morphology is collected and is bound layer sequence, fusion multidimensional roughness calculation and curvature method defect classification identification, through roughness variation trend superposition defect correction factor layering weighted evaluation, realize layering traceable, multidimensional precision surface quality grading.

Description

Technical Field

[0001] This invention relates to the field of 3D printing, and in particular to a method for analyzing the surface quality of additive manufacturing. Background Technology

[0002] Additive manufacturing, with its advantages of personalized customization and integrated molding of complex structures, has been widely used in key fields such as aerospace, medical, and high-end equipment, and is one of the core directions for upgrading intelligent manufacturing technology. However, the layer-by-layer stacking forming characteristics of this technology, coupled with the complex temperature gradients, process parameter fluctuations, and material interactions during high-energy beam melting and solidification, result in surface defects such as roughness, obvious layering, spheroidization, and warping in the formed parts. These defects severely affect the dimensional accuracy, wear resistance, and fatigue performance of the parts, thus restricting their high-precision industrial applications.

[0003] Current mainstream additive manufacturing surface quality inspection methods mostly adopt overall morphology acquisition after part forming, lacking layer-by-layer scanning and layer sequence information binding mechanisms along the stacking direction. This makes it difficult to accurately locate the stacking layer corresponding to quality anomalies and trace the root cause of problems in the layering process of forming. Most solutions only calculate the roughness parameter in a single dimension, failing to distinguish the anisotropic morphological differences between the scanning direction and the stacking direction, and lacking defect feature extraction at the micro curvature level. They mostly rely on manual visual inspection or simple height thresholds to identify defects, and cannot automatically complete the geometric morphology classification and quantitative statistics of defects such as protrusions, scratches, and pits. The accuracy and efficiency of defect identification are low. Finally, the quality evaluation mostly relies on direct grading based on a single roughness index, without establishing a continuous trend of roughness change along the stacking direction, and without incorporating defect distribution density and geometric size as correction factors into the weighted evaluation system. This ignores the quality fluctuation law between stacking layers and the quality impact of local defects, resulting in a strong one-sided evaluation result that is difficult to provide layered and traceable quantitative support for process parameter optimization. Summary of the Invention

[0004] To improve existing methods, a surface quality analysis method for additive manufacturing is provided. This method collects surface morphology layer by layer along the additive manufacturing stacking direction and binds the layer sequence. It integrates multi-dimensional roughness calculation and curvature method defect classification and identification. By superimposing the roughness change trend and the defect correction factor, a layered weighted evaluation is achieved to realize layered traceability and multi-dimensional accurate surface quality grading.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for analyzing the surface quality of additive manufacturing, comprising:

[0007] The surface morphology of the additively manufactured part to be tested is scanned and collected layer by layer along the stacking direction to obtain initial three-dimensional point cloud data, and the initial spatial coordinate information and corresponding layer sequence number of each layer scan are recorded.

[0008] The initial 3D point cloud data is filtered and denoised to remove outlier noise caused by disturbances in the scanning environment. The denoised point cloud data is then registered and interpolated to generate regularized 3D point cloud data with uniform grid arrangement. The height values ​​of each grid node relative to the reference plane are extracted to form a surface height matrix.

[0009] The surface height matrix of the base is used to calculate the two-dimensional roughness parameters of the part surface along the scanning direction and the stacking direction, calculate the three-dimensional roughness parameters of the entire measurement area, and associate the two-dimensional roughness parameters and the three-dimensional roughness parameters with the recorded layer sequence number;

[0010] Based on the generated regular 3D point cloud data, the curvature analysis method is used to extract the principal curvature values ​​of each grid node. Continuous grid areas with principal curvature values ​​exceeding the preset normal fluctuation range are marked as surface abnormal areas. Each abnormal area is classified by geometric shape, and the distribution quantity and geometric size of various surface defects in the layer are identified and counted.

[0011] Based on the two-dimensional and three-dimensional roughness parameters associated with each layer sequence number, a roughness variation trend curve along the stacking direction is established. The distribution quantity and geometric size of various surface defects in each layer are statistically superimposed on the variation trend curve. With the roughness variation trend as the main framework and the defect distribution as the correction factor, the layer quality index of each layer is obtained by weighted calculation. The layer quality index is accumulated and evaluated according to the stacking direction to obtain the overall surface quality level of the part.

[0012] Preferably, the scanning and acquisition of the surface morphology of the additively manufactured part to be tested layer by layer along the stacking direction specifically includes:

[0013] The additive manufacturing part to be tested is fixed on the stage of the optical profilometer. Based on the total height of the part and the single-layer stacking thickness, the vertical scanning range of the optical profilometer is set to cover the height of all stacked layers of the part, and the vertical scanning step size is set to match the single-layer stacking thickness. The optical profilometer is then driven to perform scanning from the bottom of the part upwards along the stacking direction.

[0014] After each layer is scanned, the image acquisition card of the optical profilometer synchronously records the scanning interference signal of that layer. The interference signal is then converted from analog to digital to generate the initial 3D point cloud data corresponding to that layer. At the same time, the current layer sequence number and the current scanning depth coordinates are written into the data header to complete the binding of the scanning data and layer sequence information.

[0015] Preferably, the filtering and denoising process for the acquired initial 3D point cloud data specifically includes:

[0016] For the initial 3D point cloud data corresponding to each layer sequence number, the spatial search range is set layer by layer with each data point as the center, and the number of neighboring data points falling into each search range is counted.

[0017] Data points whose number of neighboring data points is less than the preset lower limit of the number of neighboring data points are identified as outliers and removed from the point cloud data of that layer, thus completing the denoising process of the initial 3D point cloud data of each layer.

[0018] The denoised 3D point cloud data of each layer are uniformly transformed into a spatial rectangular coordinate system with the center of the bottom surface of the part as the origin, the stacking direction as the vertical coordinate axis, and the scanning direction as the horizontal coordinate axis. Then, a spatial interpolation algorithm is used to perform grid interpolation on the registered point cloud data of each layer to generate regular 3D point cloud data with the same horizontal grid resolution for each layer.

[0019] Preferably, the calculation of the two-dimensional roughness parameters and the three-dimensional roughness parameters and their association with the layer sequence number specifically includes:

[0020] For the generated regularized 3D point cloud data of each layer, the height sequence is extracted row by row along the corresponding scanning direction of each layer, the arithmetic mean deviation value of the contour of each row is calculated, and then the arithmetic mean deviation value of the contour of each row is averaged to obtain the two-dimensional roughness parameters of each layer along the scanning direction.

[0021] The height sequence is extracted column by column along the stacking direction of each layer, the profile arithmetic mean deviation value of each column is calculated, and the profile arithmetic mean deviation value of each column is averaged to obtain the two-dimensional roughness parameters of each layer along the stacking direction.

[0022] Using the height values ​​of all nodes within the entire regularized mesh area as the whole, the root mean square deviation value of each layer is calculated as the three-dimensional roughness parameter corresponding to each layer. The two-dimensional roughness parameters along the scanning direction, the two-dimensional roughness parameters along the stacking direction, and the three-dimensional roughness parameters are stored layer by layer according to the layer sequence number to form the roughness parameter vector of each layer.

[0023] Preferably, the extraction of the principal curvature values ​​of each grid node using curvature analysis specifically includes:

[0024] For each layer of regularized 3D point cloud data generated, a local surface fitting window is constructed layer by layer on the regularized grid with each node as the center. The spatial range of the fitting window is set to a size related to the single-layer stacking thickness. The least squares method is used to fit the local surface equation to the node height value within the fitting window.

[0025] The first and second fundamental quantities of the surface at the center node are calculated based on the fitted local surface equation. The Gaussian curvature and mean curvature of the node are obtained by algebraic combination of the first and second fundamental quantities. The maximum and minimum principal curvature values ​​of the node are obtained by calculating the Gaussian curvature and mean curvature.

[0026] Traverse all grid nodes in each layer to obtain the maximum and minimum principal curvature values ​​of all nodes in each layer, and construct a layer-by-layer principal curvature distribution matrix using the layer sequence number as the primary index and the grid node coordinates as the secondary index.

[0027] Preferably, the step of marking continuous grid regions with principal curvature values ​​exceeding a preset normal fluctuation range as surface anomaly regions and classifying each anomaly region by geometric morphology specifically includes:

[0028] Based on the generated regularized 3D point cloud data of each layer, the central tendency and discrete tendency values ​​of the maximum and minimum principal curvature values ​​of all nodes in each layer are statistically analyzed, and the normal fluctuation range of the maximum and minimum principal curvature values ​​of that layer is set.

[0029] Each node is checked to determine whether the maximum and minimum principal curvature values ​​of the current node fall within their respective normal fluctuation ranges. If any principal curvature value exceeds the corresponding normal fluctuation range, the node is marked as an abnormal node, and adjacent abnormal nodes in spatial location are merged into a connected surface abnormal region.

[0030] For each connected surface anomalous region, calculate the ratio of the average maximum principal curvature to the average minimum principal curvature of the nodes in the region, and calculate the ratio of the maximum span value along the scanning direction to the maximum span value along the stacking direction. Combine the two ratios into a feature vector.

[0031] Based on the feature vector, the region is divided into one of three categories: spherical protrusion defects, stripe scratch defects, or pit defects, thus completing the classification and statistics of abnormal regions.

[0032] Preferably, establishing the roughness variation trend curve along the stacking direction specifically includes:

[0033] Using the layer sequence number as the horizontal axis and the three-dimensional roughness parameter value of each layer as the vertical axis, data points are marked layer by layer in a two-dimensional coordinate system. The ratio of the two-dimensional roughness parameter along the scanning direction to the two-dimensional roughness parameter along the stacking direction of each layer is used as the directional weighting factor of each data point.

[0034] A local weighted regression algorithm is used to fit the labeled data points to a curve. During the fitting process, the direction weight factor of each data point is used as the sample weight of the local weighted regression algorithm, so that the direction of the fitted curve at each layer is simultaneously constrained by the three-dimensional roughness parameter value and the ratio of the two-dimensional roughness parameter value of that layer, thus obtaining the basic roughness change trend curve.

[0035] Calculate the difference between the actual value of the three-dimensional roughness parameter of each layer and the corresponding value of the roughness basic change trend curve in that layer, and arrange the differences of each layer in order of layer number to obtain the roughness residual sequence.

[0036] Preferably, the step of superimposing the distribution quantity and geometric dimensions of surface defects onto the trend curve specifically includes:

[0037] For the distribution of various types of surface defects in each layer, the number of each type of defect is multiplied by its corresponding preset defect severity weight coefficient, and then the weighted results of each type of defect are summed to obtain the defect density quantification value of that layer.

[0038] For the geometric dimensions of various surface defects in each layer, the maximum geometric dimension value of each type of defect in each layer is extracted. The maximum geometric dimension value of each type of defect is multiplied by its corresponding preset size influence coefficient. Then, the weighted results of each type of defect are summed to obtain the quantitative value of the defect size in that layer.

[0039] The defect density quantization value and the defect size quantization value are added together to obtain the defect comprehensive correction factor for this layer. The defect comprehensive correction factor is multiplied by the residual value corresponding to this layer in the roughness residual sequence to obtain the correction residual term for this layer. The correction residual term is superimposed on the vertical coordinate value corresponding to this layer in the roughness basic change trend curve to obtain the roughness change trend curve after superimposing the defect influence.

[0040] Preferably, the weighted calculation to obtain the layer quality index of each layer, and then to obtain the overall surface quality grade of the part, specifically includes:

[0041] Multiply the vertical coordinate value of each layer in the roughness change trend curve after superimposed defect influence by the preset roughness influence weight coefficient to obtain the roughness contribution value of that layer. Multiply the defect comprehensive correction factor corresponding to that layer by the preset defect influence weight coefficient to obtain the defect contribution value of that layer.

[0042] Add the roughness contribution value and the defect contribution value of each layer to obtain the layer quality index of that layer. Arrange all layer quality indices in order of layer sequence number to obtain the layer quality index sequence.

[0043] The central tendency statistic is calculated for the stratified quality index sequence to obtain the overall average quality index. Then, the dispersion statistic of the stratified quality index sequence is calculated. The overall average quality index is subtracted from the dispersion conversion value determined according to the preset rules to obtain the overall quality score of the part.

[0044] The overall surface quality grade of the part is determined based on the preset rating grade range into which the overall quality score falls. The preset rating grade range includes at least three grades: excellent, qualified, and unqualified.

[0045] Compared with the prior art, the advantages of the present invention are:

[0046] By collecting surface morphology layer by layer along the stacking direction and binding layer sequence information, the limitations of traditional overall inspection in locating layer quality differences are overcome, enabling precise tracing of the stacking layer corresponding to quality problems. After point cloud denoising, registration, and meshing, a standardized height matrix is ​​generated. Simultaneously, the two-dimensional roughness parameters along the scanning and stacking directions and the three-dimensional roughness parameters of the entire domain are calculated. Combined with curvature analysis, abnormal areas are automatically extracted and defect geometric morphology classification is completed, achieving multi-dimensional integration of micro-morphology quantification and defect identification, thus compensating for the one-sidedness of single-index evaluation. Finally, using the roughness change trend as the main framework and defect distribution and size as correction factors, a layer quality index is generated and accumulated to obtain the overall quality level. This system takes into account both the continuous change law of roughness along the stacking direction and the local quality influence of defects. The evaluation system is comprehensive and systematic, providing accurate, layered, and traceable quantitative support for additive manufacturing process optimization and part quality grading. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of an additive manufacturing surface quality analysis method proposed in this invention;

[0048] Figure 2 This is a schematic diagram illustrating the calculation of two-dimensional and three-dimensional roughness parameters and the association of layer sequence numbers proposed in this invention;

[0049] Figure 3 This is a schematic diagram illustrating the extraction of principal curvature values ​​for each grid node proposed in this invention. Detailed Implementation

[0050] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0051] See Figure 1 As shown, a method for analyzing the surface quality of additive manufacturing includes:

[0052] The surface morphology of the additively manufactured part to be tested is scanned and collected layer by layer along the stacking direction to obtain initial three-dimensional point cloud data, and the initial spatial coordinate information and corresponding layer sequence number of each layer scan are recorded.

[0053] The initial 3D point cloud data is filtered and denoised to remove outlier noise caused by disturbances in the scanning environment. The denoised point cloud data is then registered and interpolated to generate regularized 3D point cloud data with uniform grid arrangement. The height values ​​of each grid node relative to the reference plane are extracted to form a surface height matrix.

[0054] The surface height matrix of the base is used to calculate the two-dimensional roughness parameters of the part surface along the scanning direction and the stacking direction, calculate the three-dimensional roughness parameters of the entire measurement area, and associate the two-dimensional roughness parameters and the three-dimensional roughness parameters with the recorded layer sequence number;

[0055] Based on the generated regular 3D point cloud data, the curvature analysis method is used to extract the principal curvature values ​​of each grid node. Continuous grid areas with principal curvature values ​​exceeding the preset normal fluctuation range are marked as surface abnormal areas. Each abnormal area is classified by geometric shape, and the distribution quantity and geometric size of various surface defects in the layer are identified and counted.

[0056] Based on the two-dimensional and three-dimensional roughness parameters associated with each layer sequence number, a roughness variation trend curve along the stacking direction is established. The distribution quantity and geometric size of various surface defects in each layer are statistically superimposed on the variation trend curve. With the roughness variation trend as the main framework and the defect distribution as the correction factor, the layer quality index of each layer is obtained by weighted calculation. The layer quality index is accumulated and evaluated according to the stacking direction to obtain the overall surface quality level of the part.

[0057] Specifically, the additively manufactured part to be tested is fixed on the stage of the optical profilometer. Based on the total height of the part and the single-layer stacking thickness set by the additive manufacturing process, the vertical scanning range of the optical profilometer is set, and the vertical scanning step size is set. The optical profilometer is driven to scan upwards layer by layer with a fixed step size, starting from the bottom of the part, according to the above parameters.

[0058] After each scan at a step position, the white-light interferometer built into the optical profilometer generates interference fringe signals at that cross-section. These signals are converted into digital interferograms by an analog-to-digital converter, and then the surface height information of that cross-section is extracted using a phase-shifting algorithm to generate the initial 3D point cloud data for that layer. During this process, the scanning depth coordinates of each layer are calculated from the displacement feedback voltage of the piezoelectric ceramic actuator. To ensure the traceability of each layer's data, the current layer sequence number and the corresponding Z-axis scanning depth coordinates are written into the header information of each layer's point cloud data, binding the spatial location information with the process sequence information.

[0059] Suppose that the phase value is obtained by phase calculation of the interference light intensity signal of the i-th pixel in the j-th layer acquired by the optical profilometer. Then the height value of that point relative to the reference plane Calculate using the following formula:

[0060]

[0061] Where λ is the center wavelength of the light source used in the optical profilometer. The absolute phase value is obtained by phase expansion after solving the wrapped phase.

[0062] Due to scanning step size Matching the single-layer stack thickness t, when the scan depth coordinate of the j-th layer satisfy The height values ​​of all pixels in this layer Together with its in-plane coordinates Together with the layer sequence number j, they are stored as the initial 3D point cloud dataset. This is the initial scan depth. Simultaneously, the header information of each layer of point cloud data records the scan depth coordinates for that layer. The entire scanning process continues until the objective lens moves to the preset scanning termination depth, covering all stacked layers of the part, thus obtaining a sequence of initial 3D point cloud data layer by layer with layer sequence identifiers.

[0063] For the initial 3D point cloud data corresponding to each obtained layer sequence number, filtering and denoising processing is performed layer by layer. A spatial search sphere with a preset search radius is defined as the center of each data point in the layer, and the number of neighboring data points falling within this sphere is counted. For a given point to be judged, if the number of neighboring points contained within its search sphere is less than a preset lower limit for the number of neighboring points, it indicates that the point cloud density around that point is too low and does not conform to the point cloud distribution characteristics that a continuous area on the surface of the part should have. Therefore, this point is judged as an outlier and deleted from the point cloud data of that layer. This judgment condition can be expressed as:

[0064]

[0065] in, Let be the spatial coordinate vector of the point to be determined. Let j be the spatial coordinate vector of the j-th other data point in this layer. Let R represent the Euclidean distance, R be the preset search radius, 1 (⋅) be an indicator function that takes the value 1 when the condition is met and 0 otherwise, and M be the total number of data points in this layer of the point cloud. This is the number of neighboring points that fall within the search range. If If the value is below the preset threshold, delete the point. After traversing all data points of the layer point by point, the denoising process for that layer is completed. Each layer performs the above operation independently.

[0066] After denoising, coordinate registration is performed on the point cloud of each layer. A spatial rectangular coordinate system is established with the center of the bottom surface of the part as the origin, the stacking direction as the positive Z-axis, the scanning direction as the X-axis, and the horizontal direction perpendicular to the scanning direction as the Y-axis. The three-dimensional coordinates of each layer of point cloud are registered to this coordinate system through rigid body transformation, so that all layers of point cloud have a consistent coordinate reference.

[0067] After registration, each layer of the point cloud is subjected to gridded interpolation. For each layer, a uniform regular grid is set within its XY plane projection area. The resolution of the grid in the X and Y directions is equal and consistent with the preset value, ensuring that the grid nodes of all layers completely overlap in the horizontal direction. For each grid node, the nearest original data points in the registered point cloud of that layer are searched for using its horizontal coordinates as the center, and the height value at that node is calculated using a spatial interpolation algorithm. During the interpolation process, the reciprocal of the horizontal distance from the neighboring points to the grid node is used as a weight to sum the height values ​​of the neighboring points, resulting in the regularized height value of the grid node. The interpolation expression is as follows:

[0068]

[0069] in, The horizontal coordinates of the current grid node. Let x be the horizontal coordinate of the k-th neighboring original data point found. Here, K represents the height of the original data point, and K represents the total number of neighboring points found. The weight is the inverse of the distance. This is the height value of the grid node obtained through interpolation. After traversing all grid nodes in this layer, a regularized 3D point cloud data with a uniform grid arrangement is generated for this layer. All layers use the same grid resolution. Finally, the height value of each grid node relative to the reference plane is extracted from the regularized point cloud, and these height values ​​are arranged into a two-dimensional array according to the row and column order of the grid, which is the surface height matrix of this layer.

[0070] See Figure 2 As shown, based on the generated regularized 3D point cloud data of each layer, the data of each layer is arranged in a uniform grid, with the row direction of the grid consistent with the scanning direction and the column direction consistent with the projection of the stacking direction onto the horizontal plane. For each layer, the two-dimensional roughness parameters along the scanning direction and along the stacking direction of the layer are first calculated, then the three-dimensional roughness parameters of the entire measurement area of ​​the layer are calculated, and finally the three are associated with the corresponding layer sequence number.

[0071] For a specific layer, suppose the mesh of that layer has N rows and M columns, and the coordinates of the mesh nodes are... The corresponding height value is Where i=1,…,N represents the row number (along the scan direction), and j=1,…,M represents the column number (along the stacking direction). The arithmetic mean of the heights of all nodes in this layer is denoted as . The calculation method is to sum the heights of all nodes and then divide by the total number of nodes.

[0072] The two-dimensional roughness parameters along the scanning direction are calculated as follows: extract the height sequence row by row, that is, fix the row number i and extract the height values ​​of all columns in that row. Calculate the arithmetic mean deviation value of the profile for each row, which is the average of the absolute deviations of the height of each point in that row relative to the average height of that row. After traversing all N rows, obtain the arithmetic mean deviation values ​​of the profile for N rows. Then, average these results to obtain the two-dimensional roughness parameter of that layer along the scanning direction. The formula is:

[0073]

[0074] in, For the two-dimensional roughness parameter in the scanning direction, the inner layer summation and absolute value represent the average value of the absolute deviation of the height of each point in the i-th row from the average height of that row, and the outer layer summation and division by N represents the average of this average value over all rows.

[0075] The two-dimensional roughness parameters along the stacking direction are calculated as follows: extract the height sequence column by column, that is, fix the column index j and extract the height values ​​of all rows in that column. Calculate the arithmetic mean deviation of the profile in each column, which is the average of the absolute deviations of the height of each point in the column relative to the average height of the column. After traversing all M columns, obtain the arithmetic mean deviation values ​​of the profile in M ​​columns. Then, take the average of these column results. The resulting value is the two-dimensional roughness parameter of the layer along the stacking direction, and the formula is:

[0076]

[0077] in, For two-dimensional roughness parameters along the stacking direction, the inner layer represents the average of the absolute deviations of the height of each point in the j-th column from the average height of that column, and the outer layer represents the average of all columns.

[0078] The calculation of three-dimensional roughness parameters takes the height of all nodes in the entire layer as the whole, and calculates the height of all nodes in the layer relative to the overall average height of the layer. The root mean square deviation value. This parameter comprehensively reflects the overall magnitude of the surface height deviation from the reference plane within the entire measurement area of ​​the layer, and is not affected by texture in a single direction. The formula is:

[0079]

[0080] in, This represents the average height of all nodes in this layer. This is the root mean square value of all height deviations of that layer.

[0081] After completing the above calculations, , and The three parameters are stored layer by layer according to their sequence number, using the layer number as the primary key, forming a roughness parameter vector for that layer. After all layers have been processed, each layer has a vector containing three roughness feature values.

[0082] See Figure 3 As shown, curvature analysis is performed layer by layer on the generated regularized 3D point cloud data. Taking a certain layer as an example, the nodes in this layer are arranged in a uniform grid, and each node has horizontal coordinates and height values. On this layer's grid, a local surface fitting window is constructed with each node as the center. The horizontal side length of the window is set to an integer multiple of the single-layer stacking thickness to ensure that the window contains a sufficient number of nodes reflecting local geometric features, while avoiding the window being too large and smoothing out minor defects. The least squares method is used to perform local quadratic surface fitting on the height values ​​of all nodes within the window to obtain the surface equation at the central node. The first and second fundamental quantities of the surface are calculated from the first and second partial derivatives of the fitting equation, respectively. Then, the average curvature H and Gaussian curvature K of the node are obtained through the algebraic combination of the two. Maximum principal curvature and minimum principal curvature We obtain it from the following formula:

[0083]

[0084] Where H and K are the average curvature and Gaussian curvature at the node, respectively, obtained by combining the first and second fundamental quantities. This formula establishes a direct conversion relationship between curvature invariants and principal curvatures. After traversing all grid nodes in this layer, the layer-by-layer principal curvature distribution matrix is ​​constructed using the layer sequence number as the primary index and the node coordinates as the secondary index.

[0085] After obtaining the maximum and minimum principal curvature values ​​of all nodes in a certain layer, surface anomaly detection is performed on that layer. First, the average and standard deviation of the maximum principal curvature values ​​of all nodes in the layer are calculated, and their normal fluctuation range is defined. Similarly, the average and standard deviation of the minimum principal curvature values ​​are calculated, and their normal fluctuation range is defined. For each node in the layer, it is determined whether its maximum and minimum principal curvature values ​​fall within their respective normal ranges. If any principal curvature value exceeds the range, the node is marked as an anomaly. After all nodes are marked, spatially adjacent anomaly nodes are merged into a connected surface anomaly region using a connected component labeling algorithm, thereby grouping scattered anomaly points into physically meaningful defect regions.

[0086] For each connected region, calculate the average maximum principal curvature of all nodes within that region. With average minimum principal curvature The ratio of the two values ​​is used to calculate the maximum span of the region along the scanning direction. Maximum span of horizontal projection along the stacking direction The ratio of the two ratios is combined to form the feature vector of the region:

[0087]

[0088] The first ratio reflects the region's concavity / convexity and curvature, while the second ratio reflects the region's horizontal extension. This feature vector is compared with feature templates for three pre-defined defect types, and the defect category is determined based on the Euclidean distance principle. After classifying all regions in this layer, the distribution quantity and geometric dimensions of each defect type within that layer are statistically analyzed. Each layer independently performs the above identification and classification, forming a layer-by-layer defect distribution dataset.

[0089] Using the layer sequence number as the x-axis and the stored three-dimensional roughness parameters of each layer as the y-axis. The value is the ordinate, and data points are labeled layer by layer in a two-dimensional coordinate system, resulting in the same number of discrete points as the total number of layers. For the i-th layer, the two-dimensional roughness parameters along the scanning direction of that layer are also obtained. and two-dimensional roughness parameters along the stacking direction Calculate the ratio of the two. This ratio is then used as the directional weighting factor for the i-th data point. This weighting factor reflects the dominance of the surface texture of this layer in each direction. When the value is large, it indicates that the roughness contribution is dominant in the scanning direction, and the fitted curve should be more biased towards the trend in that direction.

[0090] A locally weighted regression algorithm is used to fit the curves of the discrete points. For each target point, this algorithm assigns different weights to its neighboring data points. The weights of data points within the neighborhood radius decrease as distance increases, and the final weight of each data point is multiplied by its orientation weight factor. This causes the direction of the fitted curve at each layer to be affected. Value and The system employs a dual constraint. After performing local weighted regression, a smooth basic roughness variation trend curve is obtained, which reflects the basic evolution law of roughness along the stacking direction under the influence of anisotropy. Subsequently, the actual values ​​of the three-dimensional roughness of each layer are calculated. Fitted value corresponding to the basic trend curve at the i-th layer The difference between them yields the residual. :

[0091]

[0092] The residuals of each layer are arranged according to the layer sequence number to form a roughness residual sequence.

[0093] For each layer, the number of defects of each type is multiplied by the preset defect severity weighting coefficient corresponding to that type, and then the weighted results of the three types are summed to obtain the defect density quantification value of that layer. Simultaneously, the maximum geometric dimension value of each type of defect in each layer is extracted. The maximum dimension of each type of defect is multiplied by the corresponding preset size influence coefficient. The three types are then weighted and summed to obtain the quantified defect size value of that layer. Adding the two together yields the defect comprehensive correction factor for this layer. This factor combines the effects of both the quantity and size of defects.

[0094] Next, the defect comprehensive correction factor will be... The residual value of this layer Multiplying them together yields the corrected residual term. .Will The fitted value superimposed on the basic trend curve at the i-th layer The vertical coordinate value after superimposed defect effects is obtained. This operation amplifies or reduces the original residual by the defect factor, thus reflecting the contribution of defects to roughness deviation in the trend curve.

[0095] Obtain the ordinate of the curve after each layer is stacked. Defect Comprehensive Correction Factor Then, multiply by the preset roughness influence weighting coefficient α and defect influence weighting coefficient β respectively, and add the two to obtain the layer quality index of this layer. :

[0096]

[0097] Where α and β are non-negative coefficients predetermined according to the usage requirements of the part, representing the relative importance of roughness level and defect severity in quality evaluation, respectively. All are arranged in hierarchical order. This yields a hierarchical quality index sequence.

[0098] Calculate the central tendency statistic for this series and dispersion statistics The overall quality score of the part is obtained by subtracting the standard deviation of the preset multiple k from the overall average quality index. Wherein, k is a preset coefficient based on the rigor of the quality evaluation. This score reflects both the average quality level of each layer and penalizes inter-layer quality fluctuations through standard deviation. Finally, the surface quality grade of the part is determined based on the preset score grade range into which the overall quality score falls. The range includes at least three grades: excellent, qualified, and unqualified. The boundary values ​​of each range are pre-defined by the process standard.

[0099] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0100] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing the surface quality of additive manufacturing, characterized in that, include: The surface morphology of the additively manufactured part to be tested is scanned and collected layer by layer along the stacking direction to obtain initial three-dimensional point cloud data, and the initial spatial coordinate information and corresponding layer sequence number of each layer scan are recorded. The initial 3D point cloud data is filtered and denoised to remove outlier noise caused by disturbances in the scanning environment. The denoised point cloud data is then registered and interpolated to generate regularized 3D point cloud data with uniform grid arrangement. The height values ​​of each grid node relative to the reference plane are extracted to form a surface height matrix. The surface height matrix of the base is used to calculate the two-dimensional roughness parameters of the part surface along the scanning direction and the stacking direction, calculate the three-dimensional roughness parameters of the entire measurement area, and associate the two-dimensional roughness parameters and the three-dimensional roughness parameters with the recorded layer sequence number; Based on the generated regular 3D point cloud data, the curvature analysis method is used to extract the principal curvature values ​​of each grid node. Continuous grid areas with principal curvature values ​​exceeding the preset normal fluctuation range are marked as surface abnormal areas. Each abnormal area is classified by geometric shape, and the distribution quantity and geometric size of various surface defects in the layer are identified and counted. Based on the two-dimensional and three-dimensional roughness parameters associated with each layer sequence number, a roughness variation trend curve along the stacking direction is established. The distribution quantity and geometric size of various surface defects in each layer are statistically superimposed on the variation trend curve. With the roughness variation trend as the main framework and the defect distribution as the correction factor, the layer quality index of each layer is obtained by weighted calculation. The layer quality index is accumulated and evaluated according to the stacking direction to obtain the overall surface quality level of the part.

2. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The scanning and acquisition of the surface morphology of the additively manufactured part under test along the stacking direction layer by layer specifically includes: The additive manufacturing part to be tested is fixed on the stage of the optical profilometer. Based on the total height of the part and the single-layer stacking thickness, the vertical scanning range of the optical profilometer is set to cover the height of all stacked layers of the part, and the vertical scanning step size is set to match the single-layer stacking thickness. The optical profilometer is then driven to perform scanning from the bottom of the part upwards along the stacking direction. After each layer is scanned, the image acquisition card of the optical profilometer synchronously records the scanning interference signal of that layer. The interference signal is then converted from analog to digital to generate the initial 3D point cloud data corresponding to that layer. At the same time, the current layer sequence number and the current scanning depth coordinates are written into the data header to complete the binding of the scanning data and layer sequence information.

3. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The filtering and denoising process for the acquired initial 3D point cloud data specifically includes: For the initial 3D point cloud data corresponding to each layer sequence number, the spatial search range is set layer by layer with each data point as the center, and the number of neighboring data points falling into each search range is counted. Data points whose number of neighboring data points is less than the preset lower limit of the number of neighboring data points are identified as outliers and removed from the point cloud data of that layer, thus completing the denoising process of the initial 3D point cloud data of each layer. The denoised 3D point cloud data of each layer are uniformly transformed into a spatial rectangular coordinate system with the center of the bottom surface of the part as the origin, the stacking direction as the vertical coordinate axis, and the scanning direction as the horizontal coordinate axis. Then, a spatial interpolation algorithm is used to perform grid interpolation on the registered point cloud data of each layer to generate regular 3D point cloud data with the same horizontal grid resolution for each layer.

4. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The calculation of the two-dimensional roughness parameters and the three-dimensional roughness parameters, and their association with the layer sequence number, specifically include: For the generated regularized 3D point cloud data of each layer, the height sequence is extracted row by row along the corresponding scanning direction of each layer, the arithmetic mean deviation value of the contour of each row is calculated, and then the arithmetic mean deviation value of the contour of each row is averaged to obtain the two-dimensional roughness parameters of each layer along the scanning direction. The height sequence is extracted column by column along the stacking direction of each layer, the profile arithmetic mean deviation value of each column is calculated, and the profile arithmetic mean deviation value of each column is averaged to obtain the two-dimensional roughness parameters of each layer along the stacking direction. Using the height values ​​of all nodes within the entire regularized mesh area as the whole, the root mean square deviation value of each layer is calculated as the three-dimensional roughness parameter corresponding to each layer. The two-dimensional roughness parameters along the scanning direction, the two-dimensional roughness parameters along the stacking direction, and the three-dimensional roughness parameters are stored layer by layer according to the layer sequence number to form the roughness parameter vector of each layer.

5. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The extraction of principal curvature values ​​for each grid node using curvature analysis specifically includes: For each layer of regularized 3D point cloud data generated, a local surface fitting window is constructed layer by layer on the regularized grid with each node as the center. The spatial range of the fitting window is set to a size related to the single-layer stacking thickness. The least squares method is used to fit the local surface equation to the node height value within the fitting window. The first and second fundamental quantities of the surface at the center node are calculated based on the fitted local surface equation. The Gaussian curvature and mean curvature of the node are obtained by algebraic combination of the first and second fundamental quantities. The maximum and minimum principal curvature values ​​of the node are obtained by calculating the Gaussian curvature and mean curvature. Traverse all grid nodes in each layer to obtain the maximum and minimum principal curvature values ​​of all nodes in each layer, and construct a layer-by-layer principal curvature distribution matrix using the layer sequence number as the primary index and the grid node coordinates as the secondary index.

6. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The step of marking continuous grid regions whose principal curvature values ​​exceed the preset normal fluctuation range as surface anomaly regions and classifying each anomaly region by geometric morphology specifically includes: Based on the generated regularized 3D point cloud data of each layer, the central tendency and discrete tendency values ​​of the maximum and minimum principal curvature values ​​of all nodes in each layer are statistically analyzed, and the normal fluctuation range of the maximum and minimum principal curvature values ​​of that layer is set. Each node is checked to determine whether the maximum and minimum principal curvature values ​​of the current node fall within their respective normal fluctuation ranges. If any principal curvature value exceeds the corresponding normal fluctuation range, the node is marked as an abnormal node, and adjacent abnormal nodes in spatial location are merged into a connected surface abnormal region. For each connected surface anomalous region, calculate the ratio of the average maximum principal curvature to the average minimum principal curvature of the nodes in the region, and calculate the ratio of the maximum span value along the scanning direction to the maximum span value along the stacking direction. Combine the two ratios into a feature vector. Based on the feature vector, the region is divided into one of three categories: spherical protrusion defects, stripe scratch defects, or pit defects, thus completing the classification and statistics of abnormal regions.

7. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The establishment of the roughness variation trend curve along the stacking direction specifically includes: Using the layer sequence number as the horizontal axis and the three-dimensional roughness parameter value of each layer as the vertical axis, data points are marked layer by layer in a two-dimensional coordinate system. The ratio of the two-dimensional roughness parameter along the scanning direction to the two-dimensional roughness parameter along the stacking direction of each layer is used as the directional weighting factor of each data point. The local weighted regression algorithm is used to fit the labeled data points to curves. During the fitting process, the direction weight factor of each data point is used as the sample weight of the local weighted regression algorithm, so that the direction of the fitting curve at each layer is simultaneously constrained by the three-dimensional roughness parameter value and the ratio of the two-dimensional roughness parameter value of that layer, thus obtaining the basic roughness change trend curve. Calculate the difference between the actual value of the three-dimensional roughness parameter of each layer and the corresponding value of the roughness basic change trend curve in that layer, and arrange the differences of each layer in order of layer number to obtain the roughness residual sequence.

8. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The specific steps of superimposing the distribution quantity and geometric dimensions of surface defects onto the trend curve include: For the distribution of various types of surface defects in each layer, the number of each type of defect is multiplied by its corresponding preset defect severity weight coefficient, and then the weighted results of each type of defect are summed to obtain the defect density quantification value of that layer. For the geometric dimensions of various surface defects in each layer, the maximum geometric dimension value of each type of defect in each layer is extracted. The maximum geometric dimension value of each type of defect is multiplied by its corresponding preset size influence coefficient. Then, the weighted results of each type of defect are summed to obtain the quantitative value of the defect size in that layer. The defect density quantization value and the defect size quantization value are added together to obtain the defect comprehensive correction factor for this layer. The defect comprehensive correction factor is multiplied by the residual value corresponding to this layer in the roughness residual sequence to obtain the correction residual term for this layer. The correction residual term is superimposed on the vertical coordinate value corresponding to this layer in the roughness basic change trend curve to obtain the roughness change trend curve after superimposing the defect influence.

9. The method for analyzing the surface quality of additive manufacturing according to claim 1, characterized in that, The weighted calculation to obtain the layer quality index of each layer, and then to obtain the overall surface quality grade of the part, specifically includes: Multiply the vertical coordinate value of each layer in the roughness change trend curve after superimposed defect influence by the preset roughness influence weight coefficient to obtain the roughness contribution value of that layer. Multiply the defect comprehensive correction factor corresponding to that layer by the preset defect influence weight coefficient to obtain the defect contribution value of that layer. Add the roughness contribution value and the defect contribution value of each layer to obtain the layer quality index of that layer. Arrange all layer quality indices in order of layer sequence number to obtain the layer quality index sequence. The central tendency statistic is calculated for the stratified quality index sequence to obtain the overall average quality index. Then, the dispersion statistic of the stratified quality index sequence is calculated. The overall average quality index is subtracted from the dispersion conversion value determined according to the preset rules to obtain the overall quality score of the part. The overall surface quality grade of the part is determined based on the preset rating grade range into which the overall quality score falls. The preset rating grade range includes at least three grades: excellent, qualified, and unqualified.

Images