Large-diameter 3D printing concrete morphology quality detection method

Abstract

The invention relates to 3D printing equipment and on-line quality detection thereof, in particular to a large-diameter 3D printing concrete morphology quality detection method which comprises the following steps: S1, acquiring point cloud and image data of 3D printing concrete strips; s2, carrying out coordinate alignment on the point cloud and the image data by adopting a generalized iterative closest point algorithm (GICP), and obtaining multi-modal fusion data by utilizing feature level fusion; s3, a global implicit curved surface fitting model is constructed for the multi-modal fusion data, and stripe appearance texture and shape features are restored; s4, combining to form a large-diameter 3D printing concrete strip three-dimensional shape reconstruction module; s5, a three-dimensional gray-level co-occurrence matrix (3D-GLCM) is adopted to carry out feature extraction; s6, carrying out feature extraction on the image by adopting a width point cloud extraction algorithm; and S7, the morphology quality of the concrete strip is evaluated, an evaluation result is fed back to a printing control system, and closed-loop optimization control is achieved. The forming precision and the material utilization efficiency of the large-diameter 3D printing concrete are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of 3D printing technology, and in particular to a method for reproducing and detecting the three-dimensional morphology of an initial strip of 3D printing. Background Art

[0002] With the rapid evolution of 3D printing technology, its applications in manufacturing, construction, and other fields are becoming increasingly widespread, and the demand for printing formats and precision has become more diverse. In the field of 3D printing of concrete materials, the use of large-diameter nozzles can significantly improve the deposition rate and printing efficiency of concrete materials, and enhance the overall load-bearing capacity of the printed structure, making it suitable for the one-time molding of large-scale components. However, the larger nozzle diameter also means that the concrete takes a long time to solidify after being sprayed, making wide layers more prone to collapse and deflection. Coarse-grained materials are also prone to clogging or uneven material flow at high flow rates, resulting in inconsistent printed line thickness. At the same time, the wide nozzle stripes significantly reduce surface smoothness, making it difficult to meet the requirements of fine texture and geometric accuracy. To this end, it is necessary to rely on manual online monitoring and experience-based parameter adjustment, which not only increases labor costs but also makes it difficult to achieve high consistency and automated production. Summary of the Invention

[0003] The present invention provides a method for detecting the morphology quality of large-caliber 3D printed concrete to solve the problems raised in the above background technology.

[0004] The method for detecting the morphology quality of large-caliber 3D printed concrete provided by the present invention comprises the following steps: S1, uses LiDAR and industrial cameras to establish a perception fusion module for radar visual data to collect point cloud and image data of 3D printed concrete strips; In the perception fusion module, S2 uses the generalized iterative closest point algorithm to align the point cloud and image data, and uses feature-level fusion to obtain multimodal fusion data. S3, using radial basis function interpolation algorithm to build a global implicit surface fitting model for multimodal fusion data to restore the texture and shape of the strip surface; S4, a module for reconstructing the three-dimensional morphology of strips of large-caliber 3D printed concrete formed based on the combination of texture and shape; S5, for the texture in the strip 3D shape reconstruction module, a 3D gray-level co-occurrence matrix is used to extract its features to obtain texture features; S6, for the shape in the strip 3D shape reconstruction module, a width point cloud extraction algorithm is used to extract features to obtain shape features; S7, a multi-feature comprehensive evaluation module is established based on the weight of quality indicators, and the quality of concrete strip morphology is evaluated by combining texture features and shape features. The evaluation results are fed back to the printing control system to achieve closed-loop optimization control.

[0005] Preferably, in step S2, in the perception fusion module, a generalized iterative closest point algorithm is used to align the coordinates of the point cloud and the image data, and feature-level fusion is used to obtain multimodal fusion data, specifically: The three-dimensional point in the laser radar point cloud coordinate system Map to the camera coordinate system and get the point on the camera coordinate system X c :

[0006] Where, R c is the rotation matrix; t is the translation vector; Perform a normalized projection onto the camera plane:

[0007]

[0008] Where, is the normalized coordinate, u is the pixel coordinate, x c ,y c , z c for X c Coordinates on three axes; K is the camera intrinsic parameter matrix, f x , f y , c x , c y is the element of the camera intrinsic parameter matrix; The point cloud registration problem is formulated as minimizing the distance metric with covariance weights using generalized ICP:

[0009] Where, R represents the rigid transformation to be sought, d i ( R ) represents the source point after transformation a i With the target point b i The residual vector between and are the local covariance matrices estimated at the i-th corresponding point of the source point cloud and the target point cloud, T represents the transpose of the vector, and N is the total number of sampled point clouds; Then multimodal feature level fusion is performed at pixel coordinates u and 3D points X L After matching, a fusion vector is constructed, and complementary features are extracted through deep learning to obtain multimodal fusion data.

[0010] Preferably, in step S3, a radial basis function interpolation algorithm is used to construct a global implicit surface fitting model for the multimodal fusion data, comprising the following steps: Calculate the Euclidean distance matrix D :

[0011] In the formula , Respectively i and j The coordinates of multimodal fusion data; Select Gaussian kernel function to calculate radial basis function matrix :

[0012] In the formula is the shape parameter of the Gaussian kernel function; Solving for the weight vector :

[0013] In the formula is the point cloud data matrix, Get the surface function:

[0014] Where, w i For the i The weights of the basis functions.

[0015] Preferably, the texture features include surface continuity and surface smoothness; then in step S5, the texture in the strip three-dimensional morphology reconstruction module is subjected to feature extraction using a three-dimensional gray-level co-occurrence matrix, specifically: For a given gray level G Two-dimensional image of I ( x , y ), in the offset (Δ x ,Δ y ), the two-dimensional gray-level co-occurrence matrix C Δx,Δy ( i,j ) is defined as

[0016] Where, i , j =0,1,…, G -1 represents the gray level, n×m is the image size; Extend the above two-dimensional definition to three-dimensional volume data I( x,y,z ), and use the three-dimensional offset vector Δ=(Δ x ,Δ y ,Δ z ), then the three-dimensional gray-level co-occurrence matrix is recorded as

[0017] Where, X×Y×Z is the voxel scale of the volume data, i,j ∈{0,…, G- 1}, To convert the co-occurrence frequencies into a joint probability distribution, the matrix entries are normalized:

[0018] Where, In the normalized joint probability distribution matrix, the first ( i , j ) position, G is the grayscale level of the image, p is the grayscale value of the current pixel, q is the grayscale value of the adjacent pixels, is the sum of the co-occurrence frequencies of all grayscale pairs, used for normalization; Extracting surface continuity features S c and surface smoothness characteristics S s :

[0019]

[0020] Where, T G is the preset grayscale difference threshold; S c The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the better the surface continuity; S s The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the smoother the surface.

[0021] Preferably, the shape features include width, length, height, cross-sectional area and volume; then in step S6, the shape in the strip three-dimensional shape reconstruction module is subjected to feature extraction using a width point cloud extraction algorithm, comprising the following steps: Using the point cloud width extraction algorithm, the number of point clouds corresponding to sample lines of different widths is obtained. , then the mapping function between the number of laser points and the width is Expressed as:

[0022] Where, For the i The weight coefficient of the training samples, n is the total number of categories of width sample lines; Use Gaussian kernel function ; b w is the width bias term, which is used to adjust the baseline value predicted by the model; Extract the point cloud data corresponding to the object width , input the number of point clouds Substitute the mapping function of the number of laser points and width into Get the width information of the object B i ; Similarly, construct a mapping function between the number of laser points and the length Expressed as:

[0023] Where, b L is the length offset term, Will Substitute the mapping function of the number of laser points and width into , get the length information L of the object; Using surface functions Find all extreme points; Take the mean of the extreme points as the height of the detected object H , specifically as follows:

[0024]

[0025] In the formula, the Count function is used to count the number of points where the derivative A′(x,y) is equal to 0 in all (x,y) positions; the cross-sectional area For cross-sectional area Take the derivative and find the extreme point ; n is the total number of categories; Using surface functions Find the integral, the integral value is the cross-sectional area at that point :

[0026] Use the surface function to find the double integral, the double integral value is the volume of the point V :

[0027] In the formula U is the integration domain.

[0028] Preferably, in step S7, establishing a multi-index comprehensive evaluation model based on the weights of the quality indicators specifically includes: Set up a set containing the standard values of the features:

[0029] Its vector expression is:

[0030] Where, s c Represents surface continuity, s s Represents surface smoothness, s w Represents width, s L Represents length, s H Represents height, s A represents the cross-sectional area, s V Represents a standard reference value for volume; The vector expression of the characteristic measurement value is:

[0031] Where, f c Represents surface continuity, f s Represents surface smoothness, f w Represents width, f L Represents length, f H Represents height, f A represents the cross-sectional area, f V represents the actual measurement of volume; In order to obtain the comprehensive error value of the direct feedback controller, the overall error is defined as:

[0032] E totalis regarded as the overall performance deviation of the current printed strip in the morphology reconstruction module, T is the transpose of the vector, and M is the number of elements in the set; In order to reflect the difference in the amount of information of each feature, the entropy weight method is introduced to generate the initial weight; first construct the i The sample in j Normalized error distribution on features:

[0033] In the formula For the i The sample in j The original error on the features; k is 1 to 7, which means the 7 feature dimensions of the sample. For the i The sum of errors of samples in 7 feature dimensions; Calculate entropy:

[0034] Where n is the total number of categories, and the initial entropy weight is obtained based on it:

[0035] The initial weights and samples are j The sum of the features E j Combined, we get the comprehensive error evaluation:

[0036] Should E weight It is a real-time quality evaluation indicator of the 3D morphology reconstruction module and can be directly used as a feedback signal for the print controller to adjust the nozzle path and material flow rate online to achieve closed-loop optimization.

[0037] By adopting the above scheme, the present invention has the following advantages and beneficial effects: compared with the existing technology, the present invention constructs an end-to-end reconstruction and evaluation system from multimodal data acquisition to closed-loop optimization. First, multimodal data is acquired, and robust alignment is achieved through generalized ICP. Then, a continuous and smooth implicit surface model is constructed with the help of radial basis function interpolation; on this basis, texture features such as surface continuity and smoothness and shape features such as width, length, height, cross-sectional area, and volume are extracted to form a high-dimensional fusion feature vector; finally, based on a weighted comprehensive evaluation module, the real-time evaluation results are fed back to the printing control system to realize online path and material parameter adjustment, so that the printed products are improved in terms of geometric accuracy, surface quality and structural integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the overall processing flow of the present invention; Figure 2 It is a schematic diagram of the algorithm flow of the present invention; Figure 3 It is a schematic diagram of the structure of the device of the present invention; Figure 4 This is a diagram showing the three-dimensional shape restoration result of the printed strip according to the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention for which protection is sought, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0041] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0042] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integral connection; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0043] The following describes in detail the preferred embodiments of the present invention in conjunction with the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0044] Reference Manual Figure 1-4 A method for detecting the morphology quality of large-caliber 3D printed concrete comprises the following steps: S1 uses LiDAR and industrial cameras to establish a perception fusion module for radar vision data, collecting point cloud and image data for 3D printed concrete strips; In the S2 perception fusion module, the generalized iterative closest point (GICP) algorithm is used to align the coordinates of the point cloud and image data, and feature-level fusion is used to obtain multimodal fusion data; S3 uses the radial basis function (RBF) interpolation algorithm to construct a global implicit surface fitting model for multimodal fusion data to restore the surface texture and shape characteristics of the strip; S4, a module for reconstructing the three-dimensional morphology of strips of large-caliber 3D printed concrete formed based on the combination of texture and shape; S5, for the texture in the strip 3D shape reconstruction module, a 3D gray-level co-occurrence matrix is used to extract its features to obtain texture features; S6, for the shape in the strip 3D shape reconstruction module, a width point cloud extraction algorithm is used to extract features to obtain shape features; S7, a multi-feature comprehensive evaluation module is established based on the weight of quality indicators, and the quality of concrete strip morphology is evaluated by combining texture features and shape features. The evaluation results are fed back to the printing control system to achieve closed-loop optimization control.

[0045] In this embodiment, preferably, the operating environment is the Windows operating system, the algorithm is written using Python 3.8, and the result display is mainly achieved through the VScode tool.

[0046] This embodiment is implemented using a large-caliber 3D ceramic printer.

[0047] According to this embodiment, preferably, step S1 uses a laser radar and an industrial camera to establish a perception fusion module of radar visual data to collect point cloud and image data of the 3D printed concrete strip.

[0048] The perception fusion module is installed on the side of the 3D printing head and connected to the control system via Ethernet or CAN bus. The LiDAR and industrial camera each transmit raw data streams through their respective data interfaces to the fusion module's edge computing unit for parallel access and preprocessing.

[0049] According to this embodiment, preferably, in the perception fusion module in step S2, the generalized iterative closest point algorithm (GICP) is used to align the coordinates of the point cloud and the image data, and feature-level fusion is used to obtain multimodal fusion data, specifically: The three-dimensional point in the laser radar point cloud coordinate system Mapping to the camera coordinate system:

[0050] Where, R c is the rotation matrix; t is the translation vector; Normalized projection onto the camera plane:

[0051]

[0052] Where, is the normalized coordinate, u is the pixel coordinate, x c ,y c , z c for X c Coordinates on three axes; K is the camera intrinsic parameter matrix, f x , f y , c x , c y is the element of the camera intrinsic parameter matrix; Generalized ICP formulates the point cloud registration problem as minimizing a distance metric with covariance weights:

[0053] Where, R represents the rigid transformation to be sought, d i ( R ) represents the source point after transformation ai With the target point b i The residual vector between and are the local covariance matrices estimated at the i-th corresponding point of the source point cloud and the target point cloud respectively, T represents the transpose of the vector, and N is the total number of sampled point clouds.

[0054] Then multimodal feature level fusion is performed, and the image feature points u and point cloud features X L After matching, a fusion vector is constructed and complementary features are extracted through deep learning.

[0055] According to this embodiment, preferably, in step S3, a radial basis function (RBF) interpolation algorithm is used to construct a global implicit surface fitting model for the multimodal fusion data, specifically: Calculate the Euclidean distance matrix D :

[0056] In the formula , Respectively i and j The coordinates of multimodal fusion data; Select Gaussian kernel function to calculate radial basis function matrix :

[0057] In the formula is the shape parameter of the Gaussian kernel function; Solving for the weight vector :

[0058] In the formula is the point cloud data matrix, Get the surface function:

[0059] Where, w i For the i The weights of the basis functions.

[0060] According to this embodiment, preferably, in step S4, surface continuity and surface smoothness are used as texture features, and width, length, height, cross-sectional area, and volume are used as shape features, and the two are combined to form a large-caliber 3D printed concrete strip three-dimensional morphology reconstruction module.

[0061] The three-dimensional morphology reconstruction module jointly characterizes the surface and overall shape of the concrete strips through two types of features. The texture feature extracts surface continuity and surface smoothness from the reconstructed implicit surface, reflecting the integrity and roughness of the microstructure of the strip surface. The shape feature uses the reconstructed point cloud or mesh model to calculate the three-dimensional dimensions of the strip width, length, and height, as well as the cross-sectional area and overall volume on any section, to comprehensively quantify its macroscopic geometric shape. After combining the two, they can be assigned weights in the quality evaluation module and comprehensively analyzed.

[0062] According to this embodiment, preferably, in step S5, a three-dimensional gray-level co-occurrence matrix (3D-GLCM) is used to extract the texture features in the strip three-dimensional shape reconstruction module, specifically: The given gray level is G Two-dimensional image of I ( x , y ), in the offset (Δ x ,Δ y ), the GLCM matrix C Δx,Δy ( I , j ) is defined as

[0063] Where, i , j =0,1,…, G -1 represents the grayscale level, and n×m is the image size.

[0064] Extend the above two-dimensional definition to three-dimensional volume data I( x,y,z ), and use the three-dimensional offset vector Δ=(Δ x ,Δ y ,Δ z ), then the three-dimensional gray-level co-occurrence matrix is recorded as

[0065] Where, X×Y×Z is the voxel scale of the volume data, i,j ∈{0,…, G- 1}.

[0066] To convert co-occurrence frequencies into a joint probability distribution, the matrix entries are usually normalized:

[0067] Where, In the normalized joint probability distribution matrix, the first ( i , j ) The element value at the position is the probability; G is the grayscale level of the image;p is the grayscale value of the current pixel; q is the gray value of the adjacent pixel; is the sum of the co-occurrence frequencies of all grayscale pairs, used for normalization.

[0068] Extracting surface continuity features S c and surface smoothness characteristics S s

[0069]

[0070] Where, T G is the preset grayscale difference threshold; S c The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the better the surface continuity; S s The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the smoother the surface.

[0071] According to this embodiment, preferably, in step S6, a width point cloud extraction algorithm is used to extract features of the shape features in the strip 3D shape reconstruction module, specifically: Using the point cloud width extraction algorithm, the number of point clouds corresponding to sample lines of different widths is obtained. , then the mapping function between the number of laser points and the width is Expressed as:

[0072] Where, For the i The weight coefficient of the training samples, n is the total number of categories of width sample lines; Use Gaussian kernel function ; b w is the width bias term, which is used to adjust the baseline value predicted by the model; Extract the point cloud data corresponding to the object width , input the number of point clouds Substitute the mapping function of the number of laser points and width into Get the width information of the object B i ; Similarly, construct a mapping function between the number of laser points and the length Expressed as:

[0073] Where,b L is the length offset term, Will Substitute the mapping function of the number of laser points and width into , get the length information L of the object; Using surface functions Find all extreme points; Take the mean of the extreme points as the height of the detected object H , specifically as follows:

[0074]

[0075] In the formula, the Count function is used to count the number of points where the derivative A′(x,y) is equal to 0 in all (x,y) positions; the cross-sectional area For cross-sectional area Take the derivative and find the extreme point ; n is the total number of categories; Using surface functions Find the integral, the integral value is the cross-sectional area at that point :

[0076] Use the surface function to find the double integral, the double integral value is the volume of the point V :

[0077] In the formula U is the integration domain.

[0078] According to this embodiment, preferably, in step S7, establishing a multi-index comprehensive evaluation model based on the weights of the quality indicators specifically includes: Set up a set containing the standard values of the features:

[0079] Its vector expression is:

[0080] Where, s c Represents surface continuity, s s Represents surface smoothness, s w Represents width, s L Represents length, s H Represents height,s A represents the cross-sectional area, s V Represents a standard reference value for volume.

[0081] The vector expression of the characteristic measurement value is:

[0082] Where, f c Represents surface continuity, f s Represents surface smoothness, f w Represents width, f L Represents length, f H Represents height, f A represents the cross-sectional area, f V Represents an actual measurement of volume.

[0083] Furthermore, in order to obtain a comprehensive error value that can directly feedback the controller, the overall error is defined as

[0084] M is the number of elements; E total It is regarded as the overall performance deviation of the current printed strip in the morphology reconstruction module.

[0085] In order to reflect the difference in the amount of information of each feature, the entropy weight method is introduced to generate the initial weight; first construct the i The sample in j Normalized error distribution on features:

[0086] In the formula For the i The sample in j The original error on the features; k is 1 to 7, which means the 7 feature dimensions of the sample. For the i The sum of errors of samples in 7 feature dimensions; Calculate entropy:

[0087] Where n is the total number of categories, and the initial entropy weight is obtained based on it:

[0088] The initial weights and samples arej The sum of the features E j Combined, we get the comprehensive error evaluation:

[0089] Should E weight It is a real-time quality evaluation indicator of the 3D morphology reconstruction module and can be directly used as a feedback signal for the print controller to adjust the nozzle path and material flow rate online to achieve closed-loop optimization.

Claims

1. A method for detecting the morphology quality of large-caliber 3D printed concrete, characterized in that: The method comprises the following steps: S1, uses LiDAR and industrial cameras to establish a perception fusion module for radar visual data to collect point cloud and image data of 3D printed concrete strips; In the perception fusion module, S2 uses the generalized iterative closest point algorithm to align the point cloud and image data, and uses feature-level fusion to obtain multimodal fusion data. S3, using radial basis function interpolation algorithm to build a global implicit surface fitting model for multimodal fusion data to restore the texture and shape of the strip surface; S4, a module for reconstructing the three-dimensional morphology of strips of large-caliber 3D printed concrete formed based on the combination of texture and shape; S5, for the texture in the strip 3D shape reconstruction module, a 3D gray-level co-occurrence matrix is used to extract its features to obtain texture features; S6, for the shape in the strip 3D shape reconstruction module, a width point cloud extraction algorithm is used to extract features to obtain shape features; S7, a multi-feature comprehensive evaluation module is established based on the weight of quality indicators, and the quality of concrete strip morphology is evaluated by combining texture features and shape features. The evaluation results are fed back to the printing control system to achieve closed-loop optimization control.

2. A large-caliber 3D printed concrete morphology quality detection method according to claim 1, characterized in that: In step S2, in the perception fusion module, the generalized iterative closest point algorithm is used to align the coordinates of the point cloud and the image data, and feature-level fusion is used to obtain multimodal fusion data, specifically: The three-dimensional point in the laser radar point cloud coordinate system Map to the camera coordinate system and get the point on the camera coordinate system X c : Where, R c is the rotation matrix; t is the translation vector; Perform a normalized projection onto the camera plane: Where, is the normalized coordinate, u is the pixel coordinate, x c ,y c , z c for X c Coordinates on three axes; K is the camera intrinsic parameter matrix, f x , f y , c x , c y is the element of the camera intrinsic parameter matrix; The point cloud registration problem is formulated as minimizing the distance metric with covariance weights using generalized ICP: Where, R represents the rigid transformation to be sought, d i ( R ) represents the source point after transformation a i With the target point b i The residual vector between and are the local covariance matrices estimated at the i-th corresponding point of the source point cloud and the target point cloud, T represents the transpose of the vector, and N is the total number of sampled point clouds; Then multimodal feature level fusion is performed at pixel coordinates u and 3D points X L After matching, a fusion vector is constructed, and complementary features are extracted through deep learning to obtain multimodal fusion data.

3. A large-caliber 3D printed concrete morphology quality detection method according to claim 2, characterized in that: In step S3, a radial basis function interpolation algorithm is used to construct a global implicit surface fitting model for the multimodal fusion data, which includes the following steps: Calculate the Euclidean distance matrix D : In the formula , Respectively i and j The coordinates of multimodal fusion data; Select Gaussian kernel function to calculate radial basis function matrix : In the formula is the shape parameter of the Gaussian kernel function; Solving for the weight vector : In the formula is the point cloud data matrix, Get the surface function: Where, w i For the i The weights of the basis functions.

4. A large-caliber 3D printed concrete morphology quality detection method according to claim 1, characterized in that: The texture features include surface continuity and surface smoothness. In step S5, the texture in the strip three-dimensional morphology reconstruction module is subjected to feature extraction using a three-dimensional gray-level co-occurrence matrix, specifically: For a given gray level G Two-dimensional image of I ( x , y ), in the offset (Δ x ,Δ y ), the two-dimensional gray-level co-occurrence matrix C Δx,Δy ( i,j ) is defined as Where, i , j =0,1,…, G -1 represents the gray level, n×m is the image size; Extend the above two-dimensional definition to three-dimensional volume data I( x,y,z ), and use the three-dimensional offset vector Δ=(Δ x ,Δ y ,Δ z ), then the three-dimensional gray-level co-occurrence matrix is recorded as Where, X×Y×Z is the voxel scale of the volume data, i,j ∈{0,…, G- 1}, To convert the co-occurrence frequencies into a joint probability distribution, the matrix entries are normalized: Where, In the normalized joint probability distribution matrix, the first ( i , j ) position, G is the grayscale level of the image, p is the grayscale value of the current pixel, q is the grayscale value of the adjacent pixels, is the sum of the co-occurrence frequencies of all grayscale pairs, used for normalization; Extracting surface continuity features S c and surface smoothness characteristics S s : Where, T G is the preset grayscale difference threshold; S c The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the better the surface continuity; S s The larger the sum value, the higher the probability that the grayscale remains unchanged under the offset, and the smoother the surface.

5. A large-caliber 3D printed concrete morphology quality detection method according to claim 3, characterized in that: The shape features include width, length, height, cross-sectional area and volume. In step S6, the shape in the strip 3D shape reconstruction module is subjected to feature extraction using a width point cloud extraction algorithm, including the following steps: Using the point cloud width extraction algorithm, the number of point clouds corresponding to sample lines of different widths is obtained. , then the mapping function between the number of laser points and the width is Expressed as: Where, For the i The weight coefficient of the training samples, n is the total number of categories of width sample lines; Use Gaussian kernel function ; b w is the width bias term, which is used to adjust the baseline value predicted by the model; Extract the point cloud data corresponding to the object width , input the number of point clouds Substitute the mapping function of the number of laser points and width into Get the width information of the object B i ; Similarly, construct a mapping function between the number of laser points and the length Expressed as: Where, b L is the length offset term, Will Substitute the mapping function of the number of laser points and width into , get the length information L of the object; Using surface functions Find all extreme points; Take the mean of the extreme points as the height of the detected object H , as follows: In the formula, the Count function is used to count the number of points where the derivative A′(x,y) is equal to 0 in all (x,y) positions; the cross-sectional area For cross-sectional area Take the derivative and find the extreme point ; n is the total number of categories; Using surface functions Find the integral, the integral value is the cross-sectional area at that point : Use the surface function to find the double integral, the double integral value is the volume of the point V : In the formula U is the integration domain.

6. A large-caliber 3D printed concrete morphology quality detection method according to claim 1, characterized in that: In step S7, establishing a multi-index comprehensive evaluation model based on the weights of the quality indicators specifically includes: Set up a set containing the standard values of the features: Its vector expression is: Where, s c Represents surface continuity, s s Represents surface smoothness, s w Represents width, s L Represents length, s H Represents height, s A represents the cross-sectional area, s V Represents a standard reference value for volume; The vector expression of the characteristic measurement value is: Where, f c Represents surface continuity, f s Represents surface smoothness, f w Represents width, f L Represents length, f H Represents height, f A represents the cross-sectional area, f V represents the actual measurement of volume; In order to obtain the comprehensive error value of the direct feedback controller, the overall error is defined as: E total is regarded as the overall performance deviation of the current printed strip in the morphology reconstruction module, T is the transpose of the vector, and M is the number of elements in the set; In order to reflect the difference in the amount of information of each feature, the entropy weight method is introduced to generate the initial weight; first construct the i The sample in j Normalized error distribution on features: Where, For the i The sample in j The original error on the features; k is 1 to 7, which means the 7 feature dimensions of the sample. For the i The sum of errors of samples in 7 feature dimensions; Calculate entropy: Where n is the total number of categories, and the initial entropy weight is obtained based on it: The initial weights and samples are j The sum of the features E j Combined, we get the comprehensive error evaluation: Should E weight It is a real-time quality evaluation indicator of the 3D morphology reconstruction module and can be directly used as a feedback signal for the print controller to adjust the nozzle path and material flow rate online to achieve closed-loop optimization.

Images