Pipe belt machine pipe expansion and pipe twisting multi-sensor cooperative detection and 3D digital twin visualization system

Abstract

This invention relates to the field of digital twin technology and discloses a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a conveyor belt machine. The system includes: a 3D construction module for constructing full-segment 3D point cloud data of the conveyor belt machine and performing region growth segmentation to obtain candidate deformation region point clouds; an interior and exterior angle analysis module for performing interior and exterior angle type analysis and continuous exterior angle statistics to obtain the radial expansion amount of the tube expansion and the unit length twist angle of the twisting tube; and a surface bonding module for injecting the radial expansion amount and unit length twist angle into the full-segment 3D point cloud data for vertex offset and rotation deformation updates to obtain an initial 3D mesh model, and bonding synchronous images from a six-axis camera to the surface of the initial 3D mesh model to obtain a target 3D mesh model. This invention enables maintenance personnel to intuitively obtain the spatial position, deformation amplitude, and surface condition of the tube expansion and twisting tubes of the conveyor belt machine, improving the accuracy of fault location and the efficiency of maintenance decision-making.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor. Background Technology

[0002] As the core equipment for bulk material conveying in coal mines, the pipe conveyor is prone to expansion and twisting deformation of its pipe body during long-term operation due to factors such as material impact, uneven tension, and wear of idlers. If such deformation is not detected in time, it will lead to material spillage, belt damage, and even production stoppage, posing a serious threat to production safety.

[0003] Existing pipe conveyor deformation detection technology uses single-sensor point measurement. A single sensor can only cover a localized area of ​​the pipe cross-section, resulting in significant blind spots for eccentric expansion and circumferential torsional deformation, leading to a high rate of missed detections. Furthermore, the lack of a unified spatiotemporal alignment mechanism for sensor data results in insufficient accuracy in multi-source data fusion, making it difficult to reconstruct the complete contour information of the pipe cross-section. In addition, existing technologies use simple geometric shapes or two-dimensional numerical alarms to display fault information, failing to integrate detection results with the actual surface texture of the pipe. This makes it difficult for maintenance personnel to intuitively determine the spatial location, type, and severity of deformation, leading to low fault location efficiency. Summary of the Invention

[0004] The main objective of this invention is to provide a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor. This invention achieves complete point cloud acquisition of the tube cross-section with no blind spots at 360°, overcoming the problem of missed detection caused by insufficient cross-sectional coverage of a single sensor. It enables maintenance personnel to intuitively obtain the spatial position, deformation amplitude, and surface condition of the tube expansion and twisting in the tube conveyor, thereby improving the accuracy of fault location and the efficiency of maintenance decision-making.

[0005] To achieve the above objectives, the present invention provides a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube-belt machine, comprising: The 3D construction module is used to construct the full-segment 3D point cloud data of the conveyor belt machine, and to perform region growth segmentation on the full-segment 3D point cloud data to obtain candidate deformation region point clouds. The internal and external angle analysis module is used to perform internal and external angle type analysis and continuous external angle statistics on the point cloud of the candidate deformation region to obtain the radial expansion amount of the expansion tube and the twist angle per unit length of the twist tube. The surface bonding module is used to establish a parameterized initial mesh model based on the full-segment 3D point cloud data, inject the radial expansion amount of the expansion tube into the parameterized initial mesh model to perform vertex radial offset, inject the unit length torsion angle of the twist tube into the parameterized initial mesh model to perform axis rotation update, obtain the initial 3D mesh model, and bond the synchronously acquired surface texture to the surface of the initial 3D mesh model to obtain the target 3D mesh model.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the three-dimensional construction module further includes: The synchronous pulse triggering unit is used to arrange multiple line array laser sensors at uniform angular intervals around the tube body cross section of the conveyor belt machine. Each line array laser sensor scans the corresponding arc segment and is triggered by the same synchronous pulse to synchronously collect the radial distance value within each arc segment. The coordinate transformation unit is used to transform the radial distance value from polar coordinates to cross-sectional rectangular coordinates to construct the full-section three-dimensional point cloud data of the conveyor belt. The region growing segmentation unit is used to calculate the neighborhood curvature value of each three-dimensional point in the entire three-dimensional point cloud data, and to perform region growing segmentation on the entire three-dimensional point cloud data based on the neighborhood curvature value to obtain candidate deformable region point clouds.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the coordinate transformation unit is specifically used for: Based on the installation angle and scanning angle of each linear array laser sensor, the radial distance value is transformed from polar coordinates to cross-sectional rectangular coordinates to obtain the coordinates of the cross-sectional sampling point; The cumulative axial displacement recorded synchronously by the encoder is assigned as the axial coordinate of the sampling point coordinate of the cross section to generate initial three-dimensional point cloud data. Calculate the average distance between each three-dimensional point in the initial three-dimensional point cloud data and its nearest neighbor, and remove outliers from the initial three-dimensional point cloud data based on the average distance to obtain the full-segment three-dimensional point cloud data of the conveyor belt.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the region growing and segmenting unit further includes: The neighborhood curvature calculation subunit is used to construct a covariance matrix based on the neighborhood point set of each three-dimensional point in the full-segment three-dimensional point cloud data, and to calculate the neighborhood curvature value and neighborhood curvature distribution information entropy of each three-dimensional point based on the covariance matrix. The point cloud filtering subunit is used to take the point with the smallest neighborhood curvature value in the entire three-dimensional point cloud data as the initial seed point, and to allocate each three-dimensional point in the entire three-dimensional point cloud data according to the neighborhood curvature value and the neighborhood curvature distribution information entropy until all three-dimensional points are allocated, thereby determining the candidate deformation region point cloud.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the point cloud filtering subunit is specifically used for: From the full-segment 3D point cloud data, the point with the minimum neighborhood curvature value is taken as the initial seed point. The eigenvector corresponding to the minimum eigenvalue of the covariance matrix of the initial seed point is used as the initial region average normal vector to initialize the seed point queue and the region point set. Based on the normal vectors of each unassigned point in the neighborhood of each sub-point in the seed point queue and the average normal vector of the current region, the angle between the normal vectors is calculated, and the corresponding normal vector angle threshold is selected according to the neighborhood curvature distribution information entropy of the unassigned point. Points to be assigned whose normal vector angle does not exceed the normal vector angle threshold are included in the current region point set and added to the seed point queue. The region average normal vector is updated with the arithmetic mean of the normal vectors of all points in the current region point set. After iterating until the seed point queue is empty, the assignment is repeated for all unassigned points in the entire 3D point cloud data to obtain multiple segmented regions. Calculate the mean neighborhood curvature value of all points in each region point set within each segmented region, and select segmented regions with a mean neighborhood curvature value not lower than the curvature threshold as candidate deformable region point clouds.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the interior and exterior angle analysis module further includes: The internal and external angle calculation unit is used to extract the cross-sectional profile of the point cloud of the candidate deformation region along the tube axis, and perform internal and external angle type analysis based on the cross-sectional profile to calculate the radial bulge of the tube. The conversion unit is used to extract the circumferential centroid angle of the outer corner points of each adjacent section along the tube axis, and to calculate the torsional angle per unit length based on the relationship between the circumferential centroid angle and the axial position.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, the interior and exterior angle calculation unit is specifically used for: The candidate deformation region point cloud is cut into sections along the tube axis, and adjacent contour points are connected sequentially according to the contour point set in each section to obtain the contour of each section. Calculate the forward tangent vector and the backward tangent vector based on each contour point on each cross section contour, and calculate the angle between the contour tangent vectors and the axial component of the cross product based on the forward tangent vector and the backward tangent vector; Contour points whose cross product axial component is positive and whose tangent vector angle is lower than the angle threshold are designated as exterior corner points, and contour points whose cross product axial component is negative and whose tangent vector angle is lower than the angle threshold are designated as interior corner points. Based on the external corner points of each section, the number of consecutive external corner points and the corresponding arc length span within the same section are counted. For sections where the number of consecutive external corner points is not less than the point count threshold and the arc length span is not less than the arc length threshold, the consecutive external corner points are fitted with the minimum circumcircle, and the difference between the fitted radius and the reference radius of the standard tube outer contour is used as the radial expansion amount of the tube.

[0012] Optionally, in a seventh implementation of the first aspect of the present invention, the surface bonding module further includes: A rotation transformation unit is used to inject the radial expansion amount of the expansion tube and the unit length torsion angle of the twist tube into the parameterized initial mesh model, and perform radial offset and rotation update around the axis respectively to obtain the initial three-dimensional mesh model; The texture bonding unit is used to reproject the synchronized images from the six-axis camera onto the cylindrical unfolded texture plane according to the geometric relationship of the central axis of the external participating tube, and stitch them together to obtain the target cylindrical unfolded texture map. Then, the sampled texture is bonded to the surface of the initial three-dimensional mesh model according to the target cylindrical unfolded texture map to obtain the target three-dimensional mesh model.

[0013] Optionally, in an eighth implementation of the first aspect of the present invention, the rotation transformation unit is specifically used for: In the parameterized initial mesh model, locate the first mesh vertex corresponding to the expansion tube, and offset the vertex coordinates of the first mesh vertex according to the radial expansion amount of the expansion tube to obtain the first mesh vertex coordinates; Based on the twist angle per unit length of the twist tube, the second grid vertex corresponding to the twist tube is located in the parameterized initial grid model. The rotation angle of each second grid vertex is calculated based on the axial distance of each second grid vertex from the starting section of the twist tube and the twist angle per unit length of the twist tube. Based on the rotation angle, the second grid vertex is rotated around the axis to obtain the coordinates of the second grid vertex. The coordinates of the first and second mesh vertices are merged and updated to the vertex set of the parameterized initial mesh model to obtain the initial 3D mesh model.

[0014] Optionally, in a ninth implementation of the first aspect of the present invention, the texture bonding unit is specifically used for: Acquire synchronized images from six cameras. Based on the relationship between the external and internal parameters of each camera and the axis of the tube, reproject the synchronized images from each camera onto a unified cylindrical unfolded texture plane, and stitch the synchronized images from each camera into a target cylindrical unfolded texture map. The U coordinates are calculated based on the ratio of the axial arc length of each vertex along the tube axis to the total detection length, and the V coordinates are calculated based on the ratio of the circumferential angle of each vertex to 360°, thus obtaining the UV coordinates of each vertex of the initial three-dimensional mesh model. The target cylinder is unfolded and its texture map is sampled by bilinear interpolation according to the UV coordinates of each vertex to obtain the texture color value of each vertex. The texture color value is then attached to the surface of the corresponding vertex of the initial three-dimensional mesh model to obtain the target three-dimensional mesh model.

[0015] In summary, this invention achieves complete point cloud acquisition of the pipe cross section with 360° blind zone-free operation by arranging multiple line array laser sensors at uniform angular intervals and triggering them with the same synchronous pulse, combined with encoder axial displacement assignment. This overcomes the problem of missed detection caused by insufficient cross-sectional coverage of single-channel sensors. In the point cloud segmentation stage, the curvature value of each point's neighborhood is calculated based on the eigenvalues ​​of the neighborhood covariance matrix. The information entropy of the neighborhood curvature distribution is introduced to dynamically adjust the threshold of the angle between the region growth normal vectors. The threshold is automatically tightened at high curvature deformation boundaries and automatically widened at flat pipe walls, effectively solving the problem of blurred segmentation boundaries in the vibration and noise environment of the conveyor belt under the fixed threshold method, and accurately separating the point cloud of candidate deformation regions. In the deformation recognition stage, by extracting contour polylines section by section and calculating the angle between the front and rear tangent vectors and the axial component of the cross product, the outward convexity feature of the expanded pipe and the circumferential deflection feature of the twisted pipe are quantitatively distinguished from the same deformation region. The radial expansion of the expanded pipe and the unit length torsion angle of the twisted pipe are extracted respectively, achieving accurate decoupling and quantitative characterization of the two types of fault deformation mechanisms. In the 3D visualization stage, deformation parameters directly drive the vertex offset and rotation deformation update of the entire 3D point cloud data. Then, combined with the synchronous images from the six-axis camera, the real tube surface texture is seamlessly attached to the deformed mesh model through cube mapping and UV texture mapping, generating a digital twin model with realistic texture details. This allows maintenance personnel to intuitively obtain the spatial position, deformation amplitude, and surface condition of the tube expansion and twisting of the tube conveyor, improving the accuracy of fault location and the efficiency of maintenance decision-making. Attached Figure Description

[0016] Figure 1 This is a structural block diagram of a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Reference Figure 1 This embodiment provides a multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube-belt machine, including: The 3D construction module 100 is used to construct the full-segment 3D point cloud data of the conveyor belt machine, and to perform region growth segmentation on the full-segment 3D point cloud data to obtain candidate deformation region point clouds. The internal and external angle analysis module 200 is used to perform internal and external angle type analysis and continuous external angle statistics on the point cloud of candidate deformation regions to obtain the radial expansion amount of the expansion tube and the twist angle per unit length of the twist tube. The surface bonding module 300 is used to inject the radial expansion amount of the expansion tube and the unit length torsion angle of the twist tube into the full-segment three-dimensional point cloud data for vertex offset and rotation deformation update to obtain the initial three-dimensional mesh model, and to bond the synchronous image of the six-axis camera to the surface of the initial three-dimensional mesh model to obtain the target three-dimensional mesh model.

[0020] Optionally, the 3D building module 100 also includes: The synchronous pulse triggering unit is used to arrange multiple line array laser sensors at uniform angular intervals around the tube body cross section of the conveyor belt machine. Each line array laser sensor scans the corresponding arc segment and is triggered by the same synchronous pulse to synchronously collect the radial distance value within each arc segment. The coordinate transformation unit is used to transform the radial distance value from polar coordinates to cross-sectional rectangular coordinates to construct the full-section three-dimensional point cloud data of the conveyor belt. The region growing segmentation unit is used to calculate the neighborhood curvature value of each 3D point in the entire 3D point cloud data, and to perform region growing segmentation on the entire 3D point cloud data based on the neighborhood curvature value to obtain candidate deformable region point clouds.

[0021] In this embodiment, multi-channel laser displacement sensors are uniformly installed along the periphery of the tube cross-section, and the number of sensors can be set to three. Each frame of cross-sectional distance data corresponds to a specific cumulative axial displacement during stitching, thus avoiding cross-sectional misalignment caused by asynchronous multi-source sampling. The synchronization pulse triggering unit simultaneously acquires the radial distance values ​​of each arc segment at each synchronization moment and binds the circumferential contour data and axial displacement data at the same moment into a unified data frame.

[0022] After receiving the radial distance value acquired synchronously, the coordinate transformation unit converts the ranging results within the arc segment controlled by different sensors into the same cross-sectional coordinate system. During processing, it reads the installation angle of each linear array laser displacement sensor, the scanning angle position inside the linear array, and the synchronously recorded axial cumulative displacement, thereby converting the circumferential polar coordinate ranging result into cross-sectional rectangular coordinate points. Then, it writes the corresponding axial cumulative displacement into the axial component of the spatial coordinates, forming a single-frame cross-sectional point cloud. As the conveyor belt continues to operate, the single-frame cross-sectional point cloud accumulates frame by frame along the axial direction, forming three-dimensional point cloud data covering the entire detection length.

[0023] The region growth segmentation unit first evaluates curvature changes within a local neighborhood, then dynamically controls the region expansion range based on the geometric complexity of the neighborhood. A local neighborhood point set is established around each 3D point, and a covariance matrix is ​​constructed from this set. The neighborhood curvature value and its distribution information entropy are then calculated based on the characteristic distribution of the covariance matrix, thereby distinguishing between smooth pipe wall regions, deformation transition regions, and high-curvature anomaly regions. Region growth begins at the point with the minimum neighborhood curvature value, thus prioritizing the establishment of initial regions from relatively flat and stable locations. The angle between the candidate point normal vector and the current region's average normal vector determines whether to continue absorbing neighboring points. After all 3D points are allocated, candidate deformation regions are further filtered based on the number of region points and the region's average curvature.

[0024] Optionally, the coordinate transformation unit is specifically used for: Based on the installation angle and scanning angle of each linear array laser sensor, the radial distance value is transformed from polar coordinates to cross-sectional rectangular coordinates to obtain the coordinates of the cross-sectional sampling point; The axial cumulative displacement recorded synchronously by the encoder is assigned as the axial coordinate of the cross-section sampling point to generate initial three-dimensional point cloud data. Calculate the average distance between each 3D point and its nearest neighbor in the initial 3D point cloud data, and remove outliers from the initial 3D point cloud data based on the average distance to obtain the full 3D point cloud data of the conveyor belt.

[0025] In this embodiment, after synchronous sampling, the coordinate transformation unit directly performs unified spatial reconstruction on the radial distance values ​​output by each line array laser displacement sensor, so that the ranging results originally scattered in each scanning arc segment are mapped to the same tube cross-section coordinate system. (Reading the first...) The linear array laser displacement sensor in the first The output at the next synchronous sampling time is the first Radial distance value Combined with the installation angle of the corresponding linear array laser displacement sensor The radial distance value is converted to cross-sectional plane coordinates based on the scanning angular distribution position within the linear array. The conversion relationship is as follows: in, and These represent the two coordinate components of the corresponding sampling point in the Cartesian coordinate system of the cross section. After conversion, the radial ranging values ​​output by each line array laser displacement sensor are no longer separate arc segment data, but are transformed into a set of discrete contour points on the same cross section.

[0026] The cumulative axial displacement recorded synchronously by the encoder is directly written into the axial coordinates of the corresponding sampling point, thereby expanding the two-dimensional contour of a single cross-section into a cross-sectional point cloud frame in three-dimensional space. The encoder resolution can be set to 0.1 mm to ensure that the axial position markings have fine resolution, so that adjacent cross-sections can maintain a clear spatial order when spliced ​​along the pipeline. As the conveyor belt continues to run, the cross-sectional point cloud frames formed at different times are sequentially superimposed along the axial direction to generate initial three-dimensional point cloud data covering the detection section.

[0027] Considering the presence of dust obstruction, localized glare, and occasional missed measurements in the field, the initial 3D point cloud data will contain a small number of outliers deviating from the main outline. Therefore, statistical outlier removal is performed after stitching. For each 3D point in the initial 3D point cloud data, a nearest neighbor set is searched, and the average distance from that 3D point to the nearest neighbor set is calculated. The global mean is then calculated for all average distances. and standard deviation Then according to Identify outliers, among which, Indicates the first The average neighborhood distance corresponding to each 3D point. After outlier removal, the remaining point cloud is more stable in terms of spatial continuity and local geometric consistency, thus obtaining the full-segment 3D point cloud data of the conveyor belt.

[0028] Optionally, the region growing segmentation unit also includes: The neighborhood curvature calculation subunit is used to construct a covariance matrix based on the neighborhood point set of each 3D point in the full 3D point cloud data, and to calculate the neighborhood curvature value and neighborhood curvature distribution information entropy of each 3D point based on the covariance matrix. The point cloud filtering sub-unit is used to take the point with the smallest neighborhood curvature value in the entire 3D point cloud data as the initial seed point, and to allocate each 3D point in the entire 3D point cloud data according to the neighborhood curvature value and the neighborhood curvature distribution information entropy until all 3D points are allocated, thereby determining the candidate deformable region point cloud.

[0029] In this embodiment, after the formation of the entire 3D point cloud data, the neighborhood curvature calculation subunit establishes a local neighborhood point set for each three-point repair point and performs joint calculation of geometric undulation and curvature dispersion within the local neighborhood. A neighborhood search is performed around the current three-point repair point to obtain the neighborhood point set. The mean coordinates of the neighborhood points are calculated, and a third-order covariance matrix is ​​constructed using the offsets of each neighborhood point relative to the mean point. Eigenvalue decomposition is performed on the covariance matrix, extracting three eigenvalues ​​sorted by size, and then... Calculate the neighborhood curvature values ​​of the current three maintenance points, where, Indicates the first The neighborhood curvature values ​​of three maintenance points , , These represent the three eigenvalues ​​of the covariance matrix. The minimum eigenvalue is calculated. The computational relationship can compress the local spatial distribution morphology into a single curvature index. When a neighboring point is closer to a smooth pipe wall, the neighborhood curvature value is relatively small; when a neighboring point is at the edge of a bulge, in a torsional transition zone, or in a location with significant local undulations, the neighborhood curvature value will increase accordingly. To distinguish between the overall high curvature region and the boundary transition region, the distribution of neighborhood curvature values ​​for all three maintenance points within the current neighborhood is statistically analyzed. The neighborhood curvature values ​​can be divided into eight equally spaced intervals, and then the neighborhood curvature distribution information entropy is calculated based on the proportion of each interval. ,in, Indicates the first The curvature distribution information entropy of the neighborhood corresponding to each of the three maintenance points. Indicates the first The percentage of points within a curvature interval. The lower the information entropy, the more concentrated the curvature distribution in the current neighborhood, and the closer it is to a flat and structurally simple pipe wall region; the higher the information entropy, the more dispersed the curvature distribution in the current neighborhood, corresponding to the transition boundary between normal pipe wall and abnormal deformation.

[0030] The point cloud filtering subunit selects the 3D point with the smallest neighborhood curvature value from the entire 3D point cloud data as the initial seed point. This is because such points are located in relatively flat and stable pipe wall positions, making them more conducive to constructing a geometrically consistent initial region. The eigenvector corresponding to the smallest eigenvalue of the initial seed point's covariance matrix is ​​read and used as the average normal vector of the current region, and the seed point queue and region point set are initialized. For each seed point in the seed point queue, unassigned points in the neighborhood are searched, and the angle between the normal vector of the unassigned point and the average normal vector of the current region is calculated. Then, the normal vector angle threshold is adaptively selected based on the neighborhood curvature distribution information entropy of the unassigned point. When the information entropy is higher than 1.8, the normal vector angle threshold is tightened to suppress region growth beyond the deformation boundary; when the information entropy is not higher than 1.8, the normal vector angle threshold is relaxed to ensure good connectivity integrity of the smooth pipe wall region during growth. All points to be assigned whose included normal vector angle does not exceed the corresponding threshold are added to the current region point set and simultaneously added to the seed point queue. Then, the region's average normal vector is updated with the arithmetic mean of the normal vectors of all points in the current region. This process is repeated iteratively until the current seed point queue is empty. Afterward, the same region generation process is repeated from the remaining unassigned points until all 3D points in the entire 3D point cloud data have been assigned.

[0031] Optionally, the point cloud filtering sub-unit is specifically used for: From the full-segment 3D point cloud data, the point with the minimum neighborhood curvature value is taken as the initial seed point. The eigenvector corresponding to the minimum eigenvalue of the covariance matrix of the initial seed point is used as the initial region average normal vector to initialize the seed point queue and the region point set. Based on the normal vectors of each unassigned point in the neighborhood of each sub-point in the seed point queue and the average normal vector of the current region, the angle between the normal vectors is calculated, and the corresponding normal vector angle threshold is selected according to the curvature distribution information entropy of the neighborhood of the unassigned point. Points whose normal vector angle does not exceed the threshold of normal vector angle are included in the current region point set and added to the seed point queue. The region average normal vector is updated with the arithmetic mean of the normal vectors of all points in the current region point set. After iterating until the seed point queue is empty, the allocation is repeated for all unassigned points in the entire 3D point cloud data to obtain multiple segmented regions. Calculate the mean neighborhood curvature value of all points in each region point set within each segmented region, and select segmented regions with a mean neighborhood curvature value not lower than the curvature threshold as candidate deformable region point clouds.

[0032] In this embodiment, the point cloud filtering subunit performs region growth allocation after all 3D points have completed the calculation of neighborhood curvature values, neighborhood curvature distribution information entropy, and local normal vectors. The starting point is set to the 3D point with the smallest neighborhood curvature value in the entire 3D point cloud data. The point with the smallest neighborhood curvature value is located at the position with the weakest geometric undulation and the most stable local surface, thus initiating region expansion, which is beneficial for first establishing a basic region with consistent direction and stable boundaries. The eigenvector corresponding to the smallest eigenvalue of the neighborhood covariance matrix of the initial seed point is read and written into the region average normal vector register. At the same time, a seed point queue and a region point set are created, so that the initial seed point enters both the seed point queue and the region point set simultaneously, thereby completing the first round of region initialization. Seed points are sequentially extracted from the seed point queue. For each seed point, all unassigned points within its neighborhood are traversed. The normal vector of the unassigned point is read, and its angle with the average normal vector of the current region is calculated. The curvature distribution entropy of the neighborhood corresponding to the unassigned point is then read to determine whether the current location belongs to a region with concentrated or dispersed curvature. When the unassigned point is located in a region with high curvature dispersion, a pre-set tightening angle threshold is applied to prevent region expansion from crossing deformation boundaries. When the unassigned point is located in a region with concentrated curvature distribution, a pre-set widening angle threshold is applied to ensure continuous inclusion within the same pipe wall region. After threshold selection, unassigned points with normal vector angles not exceeding the corresponding thresholds are included in the current region point set and simultaneously pushed into the seed point queue, allowing newly included points to continue participating in subsequent neighborhood expansion. As the region point set continues to grow, the arithmetic mean of the normal vectors of all points within the region point set is updated, ensuring that the average normal vector of the region reflects the overall directional characteristics of the currently segmented region in real time. This avoids local offsets caused by relying on a single seed point direction during region expansion. The above process iterates until the current seed point queue is empty, indicating that all neighboring points with the same directional features as the current region and satisfying curvature constraints have been absorbed, forming an independent segmented region. Then, the point with the smallest neighborhood curvature value is selected from the remaining unassigned points, and the same initialization and expansion process is repeated until all 3D points in the entire 3D point cloud data have been assigned to a region, resulting in multiple segmented regions. After all segmented regions are generated, the average neighborhood curvature value of all points within each segmented region is calculated, and the average neighborhood curvature value is compared with a predetermined curvature threshold. Segmented regions that reach the curvature threshold are retained as candidate deformable region point clouds, while segmented regions that do not reach the curvature threshold are assigned to normal pipe wall regions or background regions. This ensures that the output results retain regions with more obvious curvature anomalies and more prominent spatial clustering characteristics.

[0033] Optionally, the interior and exterior angle analysis module 200 also includes: The internal and external angle calculation unit is used to extract the cross-sectional profile of the point cloud of the candidate deformation region along the tube axis, and perform internal and external angle type analysis based on the cross-sectional profile to calculate the radial bulge of the tube. The conversion unit is used to extract the circumferential centroid angle of the outer corner points of each adjacent section along the tube axis, and to calculate the torsional angle per unit length based on the relationship between the circumferential centroid angle and the axial position.

[0034] In this embodiment, the point cloud of the candidate deformation region is cut section by section along the pipe axis. After each section is cut, the corresponding two-dimensional point set is extracted, and then the point set is reordered according to the circumferential angle. Adjacent points are then connected sequentially to form a cross-sectional profile polyline, thereby reorganizing the originally discrete cross-sectional point cloud into a continuous profile that can be used for geometric analysis. To avoid local noise points directly participating in deformation determination, during profile analysis, the preceding and following profile points are taken for each profile point, and forward and backward tangent vectors are constructed. Based on these, the profile turning direction and turning intensity are calculated. When the turning direction is outward and the turning amplitude reaches the set condition, the corresponding profile point is recorded as an outer corner point; when the turning direction is inward and the turning amplitude reaches the set condition, the corresponding profile point is recorded as an inner corner point. The lower bound of the angle between the profile tangent vectors can distinguish slight disturbances on a normal smooth pipe wall from real convex and concave turning points.

[0035] After identifying the type of internal and external angles of a single cross-section, the internal and external angle calculation unit identifies the expansion position and degree of expansion based on the continuous distribution characteristics of the external angle points. During processing, the number of consecutive external angle points and their corresponding arc length spans within the same cross-section are counted. When the number of consecutive external angle points reaches 3 and the arc length covered by the consecutive external angles exceeds 10mm, it is determined that the current cross-section has effective convex deformation. The number of external angle points is set to 3 to avoid the influence of local reflections and distance jumps on single or two isolated inflection points. The arc length span is set to 10mm to distinguish between minor local surface defects and expansion bulges with practical engineering significance. For the set of consecutive external angle points that meet the conditions, minimum circumcircle fitting is performed. The difference between the radius of the fitted circle and the reference radius of the standard pipe outer contour is used to characterize the radial expansion amount of the current cross-section. This transforms the convex deformation from discrete contour inflection point information into a stable radial dimension index, and simultaneously records the axial position of the corresponding cross-section, forming a sequence of expansion parameters distributed along the pipe axis. The reference radius of the standard tube's outer contour is determined by the nominal outer contour dimension of the tube conveyor design or the average fitted radius of the deformation-free calibration section.

[0036] The conversion unit extracts torsion tube features based on the external corner point identification results. For the external corner point group of adjacent sections along the axial direction within the same candidate deformation region, the circumferential centroid angle is calculated, and then an axial torsion sequence is constructed based on the change in centroid angle between the preceding and following sections. Since the torsion tube essentially exhibits the same outward convex feature continuously rotating along the axial direction, as long as the circumferential position of the external corner point group on continuous sections undergoes a systematic shift, the torsional slope can be established accordingly. To reduce the influence of local disturbances at a single section on the results, linear least squares fitting is performed on the circumferential shifts of all adjacent sections to obtain the torsional slope.

[0037] Optionally, the interior and exterior angle calculation unit is specifically used for: The point cloud of the candidate deformation region is cut into sections along the tube axis, and adjacent contour points are connected sequentially according to the contour point set in each section to obtain the contour of each section. Calculate the forward tangent vector and the backward tangent vector based on each contour point on each cross section contour, and calculate the angle between the contour tangent vectors and the axial component of the cross product based on the forward tangent vector and the backward tangent vector. Contour points with positive cross product axial components and an angle between their tangent vectors below the angle threshold are designated as exterior corner points, while contour points with negative cross product axial components and an angle between their tangent vectors below the angle threshold are designated as interior corner points. Based on the number of consecutive outer corner points and the corresponding arc length span within the same cross section, for cross sections where the number of consecutive outer corner points is not less than the point count threshold and the arc length span is not less than the arc length threshold, the consecutive outer corner points are fitted with the minimum circumcircle, and the difference between the fitted radius and the reference radius of the standard tube outer contour is used as the radial expansion amount of the tube.

[0038] In this embodiment, the point cloud of the candidate deformation region is continuously sectioned along the tube axis, transforming the spatial anomaly region into a set of two-dimensional cross-sectional points arranged in axial order. After each cross-sectional point set is formed, it is reordered according to the circumferential angle of each contour point relative to the center of the cross section, and then adjacent contour points are connected sequentially to form a closed or nearly closed cross-sectional contour polyline, restoring the originally discrete point cloud sampling results to a geometrically continuous contour curve. For any contour point on the cross-sectional contour... Take the adjacent previous point and the adjacent point Construct the forward tangent vector With the backward tangent vector And after normalization, the angle between the contour tangent vectors is calculated. and the axial component of the cross product .in, Represents contour points Local inflection intensity at the point, The sign characteristics indicating the direction of a local turning point along the tube axis. Indicates the component of the tube axis.

[0039] Based on the direction and magnitude of the turning point, contour points are categorized. Contour points with a positive cross-product axial component and an angle between the contour tangent vectors below a pre-defined angle threshold are marked as external corner points, while those with a negative cross-product axial component and an angle between the contour tangent vectors below the same angle threshold are marked as internal corner points. The set of external corner points corresponds to an outward convexity of the tube, while the set of internal corner points corresponds to a local concaveness. Tube expansion identification focuses on the continuous distribution of internal and external corner points within the same cross-section. Therefore, it is necessary to count the number of consecutive external corner points and their corresponding arc length spans along the contour connection sequence. Only when the number of consecutive external corner points reaches a pre-defined threshold and the corresponding arc length span reaches a pre-defined arc length threshold is the current cross-section considered to have valid tube expansion characteristics, thus avoiding misjudgments caused by local measurement jumps, dust obstruction, or scattered noise. For the set of consecutive external corner points that meets the conditions, minimum circumcircle fitting is performed to obtain the radius of the fitted circle. Then according to Calculate the radial expansion amount of the expansion tube, where, This indicates the radial bulge of the current cross-section. This represents the minimum circumradius of the set of consecutive exterior angles. This indicates the reference radius of the outer contour of the standard pipe.

[0040] Optionally, the surface bonding module 300 also includes: The rotation transformation unit is used to inject the radial expansion amount of the expansion tube and the unit length torsion angle of the twisted tube into the parameterized initial mesh model, and perform radial offset and rotation update around the axis respectively to obtain the initial three-dimensional mesh model; The texture bonding unit is used to reproject the synchronized images from the six-axis camera onto the cylindrical unfolded texture plane according to the geometric relationship of the central axis of the external part of the tube, and stitch them together to obtain the target cylindrical unfolded texture map. Then, the sampled texture is bonded to the surface of the initial three-dimensional mesh model according to the target cylindrical unfolded texture map to obtain the target three-dimensional mesh model.

[0041] In this embodiment, the rotation transformation unit constructs a parameterized mesh representation based on the full-segment 3D point cloud data, and reads the position of the expansion tube section, the corresponding radial bulge, the start and end sections of the twist tube, and the corresponding twist angle per unit length. Then, it locates the first mesh vertex corresponding to the expansion tube and the second mesh vertex corresponding to the twist tube in the full-segment 3D point cloud data. For the expansion tube region, the key processing point is to inject the cross-section-level convexity into the radial coordinate update process of the mesh vertex. Therefore, it is necessary to select the first mesh vertex within the corresponding range according to the axial position of the expansion tube, and perform vertex coordinate offset along the radial direction of the cross-section where each vertex is located to make the local mesh shape consistent with the detected bulge amplitude. For the twist tube region, the key processing point is to inject the circumferential deflection accumulated axially into the mesh rotation process. Therefore, it is necessary to calculate the corresponding rotation amount point by point according to the axial distance of each second mesh vertex from the starting cross-section of the twist tube, and then perform rotation transformation on the second mesh vertex around the tube axis to form a continuous circumferential rotation effect in the same abnormal area during axial extension. After completing the two types of vertex updates, the coordinates of the first and second grid vertices are merged and written back to the full grid vertex set to obtain an initial three-dimensional mesh model that synchronously reflects the convex features of the expansion tube and the torsional features of the twisting tube.

[0042] The texture bonding unit performs cylindrical unfolded texture construction and surface sampling mapping based on synchronized images from six-axis cameras. It acquires image frames from the six cameras at the same synchronous moment and projects each of the six images onto its corresponding cube face according to the cube face orientation of each camera, thus merging them to form the target cylindrical unfolded texture map. The six-axis cameras use a shared synchronization trigger relationship with the point cloud acquisition link, ensuring that the texture image and the geometric model are at the same time section, so the surface texture after mapping can maintain synchronous correspondence with the current deformation state. After generating the target cylindrical unfolded texture map, it calculates the surface sampling coordinates for each vertex in the initial 3D mesh model. During processing, it reads the axial arc length and circumferential angle information of each vertex along the tube axis and generates the corresponding UV coordinates accordingly, ensuring that each mesh vertex can find a unique sampling position in the target cylindrical unfolded texture map. The texture bonding unit performs bilinear interpolation sampling on the unfolded texture map of the target cylinder according to the UV coordinates of each vertex to obtain the corresponding texture color value. Then, the texture color value is written into the surface properties of the corresponding vertex of the initial 3D mesh model. In this way, the color, dirt, scratches and local abnormal appearance of the real tube surface are synchronously attached to the deformed geometric model.

[0043] Optionally, the rotation transformation unit is specifically used for: In the parametric initial mesh model, locate the first mesh vertex corresponding to the expansion tube, and offset the vertex coordinates of the first mesh vertex according to the radial expansion amount of the expansion tube to obtain the first mesh vertex coordinates; Based on the twist angle per unit length of the twist tube, the second grid vertex corresponding to the twist tube is located in the parameterized initial mesh model. The rotation angle of each second grid vertex is calculated based on the axial distance of each second grid vertex from the starting section of the twist tube and the twist angle per unit length of the twist tube. Based on the rotation angle, the second grid vertex is rotated around the axis to obtain the coordinates of the second grid vertex. The vertex coordinates of the first mesh and the vertex coordinates of the second mesh are merged and updated to the vertex set of the parameterized initial mesh model to obtain the initial 3D mesh model.

[0044] In this embodiment, the rotation transformation unit performs a partitioned geometric update on the parameterized mesh corresponding to the entire 3D point cloud data. A full-segment mesh topology is established using a standard tubular shape. Then, the axial position and corresponding radial bulge of each bulging section are read, and a vertex ring matching the bulging section position is searched within the mesh coordinate set. To ensure that the bulge of a single section acts on the corresponding local area, the axial filtering window can be set to... This value, in conjunction with the section cutting spacing, can concentrate the section-level deformation into adjacent mesh rings, preventing the disordered diffusion of deformation effects along the axial direction. After positioning is completed, the first mesh vertex within the window is offset along the radial direction of its corresponding section, with the offset relationship being... ,in, This indicates the coordinates of the vertex before the offset. Indicates the coordinates of the vertex after offset. This indicates the radial bulge amount corresponding to the expansion tube cross-section. This represents the radial unit direction vector at the vertex location. Using this update method, the cross-sectional convexity obtained from the tube expansion identification process is directly written into the grid's radial coordinates, allowing the local bulge morphology to form a geometric appearance on the 3D model consistent with the detection results.

[0045] For the twisted tube region, the processing focus shifts to continuous circumferential rotation updates along the axial direction. Therefore, it is necessary to first select the second grid vertices whose axial coordinates fall within the twisted tube segment from the entire grid coordinate set, based on the starting and ending section positions of the twisted tube. Then, the corresponding rotation angle is calculated based on the axial distance of each second grid vertex from the starting section of the twisted tube. The twist angle per unit length has already been converted by the identification process. Therefore, the rotation angle is:

[0046] in, Indicates axial position The rotation angle at the vertex, Indicates the starting section position of the twisted tube. This represents the angle of twist per unit length. After obtaining the rotation angle, a rotation transformation is performed on the second mesh vertex around the tube axis, causing the mesh vertices within the same twisted tube segment to gradually produce a continuous circumferential offset along the axial direction. This transforms the "angle of twist per unit length" into a "vertex-level rotation coordinate update," thereby forming a twisted shape on the model that continuously unfolds from the starting section to the ending section.

[0047] After the coordinates of the first grid vertex corresponding to the expansion tube and the second grid vertex corresponding to the torsion tube are updated, they are written back to the vertex set of the full-segment parametric initial mesh model. If a local area is simultaneously within the range of action of both the expansion and torsion tubes, radial offset is performed first, followed by rotation around the axis, so that the bulge amplitude and circumferential deflection can be superimposed and reflected at the same vertex. After all vertices are written back, the full-segment mesh coordinates have synchronously included the local outward convexity of the expansion tube and the axial torsion of the torsion tube, and the original standard tube shape is thus updated to an initial three-dimensional mesh model consistent with the measured deformation parameters.

[0048] Optionally, the texture bonding unit is specifically used for: Acquire synchronized images from six cameras. Based on the relationship between the external and internal parameters of each camera and the axis of the tube, reproject the synchronized images from each camera onto a unified cylindrical unfolded texture plane, and stitch the synchronized images from each camera into a target cylindrical unfolded texture map. The U coordinates of each vertex of the initial 3D mesh model are calculated based on the ratio of the axial arc length of each vertex along the tube axis to the total detection length, and the V coordinates are calculated based on the ratio of the circumferential angle of each vertex to 360°, thus obtaining the UV coordinates of each vertex of the initial 3D mesh model. The target cylinder is unfolded and its texture map is sampled by bilinear interpolation according to the UV coordinates of each vertex to obtain the texture color value of each vertex. The texture color value is then attached to the surface of the corresponding vertex of the initial 3D mesh model to obtain the target 3D mesh model.

[0049] In this embodiment, the six-axis camera images corresponding to the point cloud acquisition time are read, and the six monocular industrial cameras are respectively arranged facing the six standard directions of the cube, namely +X, ... X, +Y, Y, +Z and Z represents six directions; the field of view of each camera can be set to 90°, determined by the coverage relationship of a single cube face, so that the output image of each camera naturally corresponds to a cube face; after image reading is completed, the texture coordinates of the pixels on the corresponding cube face are calculated for each image of the corresponding direction. For example, for the image pixels of the +Z face, when the spatial direction vector satisfies and When the pixel's normalized coordinates in the cube map are given, they can be written as: ;the remaining Z, +X, X, +Y, The five Y-axis surfaces are normalized according to their respective principal direction components, thereby projecting the six synchronized images onto the six cube faces. The images are then stitched together in the order of face pixels to generate the unfolded texture map of the target cylinder covering the surface of the tube.

[0050] Texture sampling coordinates are established for each vertex of the initial 3D mesh model. The axial arc length and corresponding circumferential angle of each mesh vertex along the tube axis are read, and vertex UV coordinates are generated based on the axial unfolding relationship and the circumferential normalization relationship. The U and V coordinates are determined according to... , Calculate, where, This represents the axial texture coordinates of the vertex. Represents the circumferential texture coordinates of the vertex. This represents the cumulative axial arc length obtained from the vertex along the tube axis starting from the detection start point. Indicates the total detection length. This represents the circumferential angle of a vertex relative to the +X direction within the cross-sectional plane. The axial and circumferential positions of the mesh vertices in 3D space are mapped one-to-one onto the 2D texture plane, ensuring that each vertex can be uniquely sampled in the unfolded texture map of the target cylinder. The texture bonding unit performs bilinear interpolation sampling in the unfolded texture map of the target cylinder based on the UV coordinates of each vertex, obtaining the texture color value of the corresponding vertex. This texture color value is then written into the surface properties of the corresponding vertex in the initial 3D mesh model to generate the target 3D mesh model. Bilinear interpolation sampling utilizes the combined color smoothing of adjacent pixels to reduce jagged edges and jumps caused by texture resolution discrepancies during vertex sampling, allowing dirt marks, wear boundaries, and local anomalies to adhere more naturally to the deformed mesh surface.

[0051] In this embodiment of the invention, during the point cloud segmentation stage, the neighborhood curvature value of each point is calculated based on the eigenvalues ​​of the neighborhood covariance matrix. The threshold of the angle between the region growth normal vectors is dynamically adjusted by introducing the neighborhood curvature distribution information entropy. The threshold is automatically tightened at high curvature deformation boundaries and automatically widened at flat pipe walls, effectively solving the problem of blurred segmentation boundaries in the case of vibration and noise from the conveyor belt under fixed threshold conditions, and accurately separating the point cloud of candidate deformation regions. In the deformation recognition stage, by extracting contour polylines section by section and calculating the angle between the front and rear tangent vectors and the axial component of the cross product, the outward convexity feature of the expanding pipe and the circumferential deflection feature of the twisted pipe are quantitatively distinguished from the same deformation region. The radial bulge of the expanding pipe and the unit length torsion angle of the twisted pipe are extracted respectively, achieving accurate decoupling and quantitative characterization of the two types of fault deformation mechanisms. In the 3D visualization stage, deformation parameters are directly used to drive the vertex offset and rotation deformation update of the entire 3D point cloud data. Then, combined with the synchronous images from the six-axis camera, cube mapping and UV texture mapping are used to seamlessly attach the real pipe surface texture to the deformed mesh model, generating a digital twin model with real texture details, which improves the accuracy of fault location and the efficiency of operation and maintenance decision-making.

[0052] In this embodiment, the specific implementation of each unit in the above device embodiment is described in the above system embodiment, and will not be repeated here.

[0053] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or system that includes that element.

[0054] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A pipe belt machine pipe expanding and twisting multi-sensor collaborative detection and 3D digital twin visualization system, characterized in that, include: The 3D construction module is used to construct the full-segment 3D point cloud data of the conveyor belt machine, and to perform region growth segmentation on the full-segment 3D point cloud data to obtain candidate deformation region point clouds. The internal and external angle analysis module is used to perform internal and external angle type analysis and continuous external angle statistics on the point cloud of the candidate deformation region to obtain the radial expansion amount of the expansion tube and the twist angle per unit length of the twist tube. The surface bonding module is used to establish a parameterized initial mesh model based on the full-segment 3D point cloud data, inject the radial expansion amount of the expansion tube into the parameterized initial mesh model to perform vertex radial offset, inject the unit length torsion angle of the twist tube into the parameterized initial mesh model to perform axis rotation update, obtain the initial 3D mesh model, and bond the synchronously acquired surface texture to the surface of the initial 3D mesh model to obtain the target 3D mesh model.

2. The pipe-bending and torsioning multi-sensor coordinated detection and 3D digital twin visualized system of the pipe-belt machine of claim 1, wherein, The three-dimensional construction module also includes: The synchronous pulse triggering unit is used to arrange multiple line array laser sensors at uniform angular intervals around the tube body cross section of the conveyor belt machine. Each line array laser sensor scans the corresponding arc segment and is triggered by the same synchronous pulse to synchronously collect the radial distance value within each arc segment. The coordinate transformation unit is used to transform the radial distance value from polar coordinates to cross-sectional rectangular coordinates to construct the full-section three-dimensional point cloud data of the conveyor belt. The region growing segmentation unit is used to calculate the neighborhood curvature value of each three-dimensional point in the entire three-dimensional point cloud data, and to perform region growing segmentation on the entire three-dimensional point cloud data based on the neighborhood curvature value to obtain candidate deformable region point clouds.

3. The pipe-bending and torsioning multi-sensor coordinated detection and 3D digital twin visualized system of the pipe-belt machine of claim 2, wherein, The coordinate transformation unit is specifically used for: Based on the installation angle and scanning angle of each linear array laser sensor, the radial distance value is transformed from polar coordinates to cross-sectional rectangular coordinates to obtain the coordinates of the cross-sectional sampling point; The cumulative axial displacement recorded synchronously by the encoder is assigned as the axial coordinate of the sampling point coordinate of the cross section to generate initial three-dimensional point cloud data. Calculate the average distance between each three-dimensional point in the initial three-dimensional point cloud data and its nearest neighbor, and remove outliers from the initial three-dimensional point cloud data based on the average distance to obtain the full-segment three-dimensional point cloud data of the conveyor belt.

4. The pipe-bending and torsioning multi-sensor coordinated detection and 3D digital twin visualized system of the pipe-belt machine of claim 2, wherein, The region growth segmentation unit further includes: The neighborhood curvature calculation subunit is used to construct a covariance matrix based on the neighborhood point set of each three-dimensional point in the full-segment three-dimensional point cloud data, and to calculate the neighborhood curvature value and neighborhood curvature distribution information entropy of each three-dimensional point based on the covariance matrix. The point cloud filtering subunit is used to take the point with the smallest neighborhood curvature value in the entire three-dimensional point cloud data as the initial seed point, and to allocate each three-dimensional point in the entire three-dimensional point cloud data according to the neighborhood curvature value and the neighborhood curvature distribution information entropy until all three-dimensional points are allocated, thereby determining the candidate deformation region point cloud.

5. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor as described in claim 4, is characterized in that... The point cloud filtering subunit is specifically used for: From the full-segment 3D point cloud data, the point with the minimum neighborhood curvature value is taken as the initial seed point. The eigenvector corresponding to the minimum eigenvalue of the covariance matrix of the initial seed point is used as the initial region average normal vector to initialize the seed point queue and the region point set. Based on the normal vectors of each unassigned point in the neighborhood of each sub-point in the seed point queue and the average normal vector of the current region, the angle between the normal vectors is calculated, and the corresponding normal vector angle threshold is selected according to the neighborhood curvature distribution information entropy of the unassigned point. Points to be assigned whose normal vector angle does not exceed the normal vector angle threshold are included in the current region point set and added to the seed point queue. The region average normal vector is updated with the arithmetic mean of the normal vectors of all points in the current region point set. After iterating until the seed point queue is empty, the assignment is repeated for all unassigned points in the entire 3D point cloud data to obtain multiple segmented regions. Calculate the mean neighborhood curvature value of all points in each region point set within each segmented region, and select segmented regions with a mean neighborhood curvature value not lower than the curvature threshold as candidate deformable region point clouds.

6. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor as described in claim 5, is characterized in that, The interior and exterior angle analysis module also includes: The internal and external angle calculation unit is used to extract the cross-sectional profile of the point cloud of the candidate deformation region along the tube axis, and perform internal and external angle type analysis based on the cross-sectional profile to calculate the radial bulge of the tube. The conversion unit is used to extract the circumferential centroid angle of the outer corner points of each adjacent section along the tube axis, and to calculate the torsional angle per unit length based on the relationship between the circumferential centroid angle and the axial position.

7. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor as described in claim 6, is characterized in that, The interior and exterior angle calculation unit is specifically used for: The candidate deformation region point cloud is cut into sections along the tube axis, and adjacent contour points are connected sequentially according to the contour point set in each section to obtain the contour of each section. Calculate the forward tangent vector and the backward tangent vector based on each contour point on each cross section contour, and calculate the angle between the contour tangent vectors and the axial component of the cross product based on the forward tangent vector and the backward tangent vector; Contour points whose cross product axial component is positive and whose tangent vector angle is lower than the angle threshold are designated as exterior corner points, and contour points whose cross product axial component is negative and whose tangent vector angle is lower than the angle threshold are designated as interior corner points. Based on the external corner points of each section, the number of consecutive external corner points and the corresponding arc length span within the same section are counted. For sections where the number of consecutive external corner points is not less than the point count threshold and the arc length span is not less than the arc length threshold, the consecutive external corner points are fitted with the minimum circumcircle, and the difference between the fitted radius and the reference radius of the standard tube outer contour is used as the radial expansion amount of the tube.

8. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor as described in claim 7, is characterized in that, The surface bonding module further includes: A rotation transformation unit is used to inject the radial expansion amount of the expansion tube and the unit length torsion angle of the twist tube into the parameterized initial mesh model, and perform radial offset and rotation update around the axis respectively to obtain the initial three-dimensional mesh model; The texture bonding unit is used to reproject the synchronized images from the six-axis camera onto the cylindrical unfolded texture plane according to the geometric relationship of the central axis of the external participating tube, and stitch them together to obtain the target cylindrical unfolded texture map. Then, the sampled texture is bonded to the surface of the initial three-dimensional mesh model according to the target cylindrical unfolded texture map to obtain the target three-dimensional mesh model.

9. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor as described in claim 8, is characterized in that, The rotation transformation unit is specifically used for: In the parameterized initial mesh model, locate the first mesh vertex corresponding to the expansion tube, and offset the vertex coordinates of the first mesh vertex according to the radial expansion amount of the expansion tube to obtain the first mesh vertex coordinates; Based on the twist angle per unit length of the twist tube, the second grid vertex corresponding to the twist tube is located in the parameterized initial grid model. The rotation angle of each second grid vertex is calculated based on the axial distance of each second grid vertex from the starting section of the twist tube and the twist angle per unit length of the twist tube. Based on the rotation angle, the second grid vertex is rotated around the axis to obtain the coordinates of the second grid vertex. The coordinates of the first and second mesh vertices are merged and updated to the vertex set of the parameterized initial mesh model to obtain the initial 3D mesh model.

10. The multi-sensor collaborative detection and 3D digital twin visualization system for tube expansion and twisting in a tube conveyor according to claim 8, characterized in that, The texture bonding unit is specifically used for: Acquire synchronized images from six cameras. Based on the relationship between the external and internal parameters of each camera and the axis of the tube, reproject the synchronized images from each camera onto a unified cylindrical unfolded texture plane, and stitch the synchronized images from each camera into a target cylindrical unfolded texture map. The U coordinates are calculated based on the ratio of the axial arc length of each vertex along the tube axis to the total detection length, and the V coordinates are calculated based on the ratio of the circumferential angle of each vertex to 360°, thus obtaining the UV coordinates of each vertex of the initial three-dimensional mesh model. The target cylinder is unfolded and its texture map is sampled by bilinear interpolation according to the UV coordinates of each vertex to obtain the texture color value of each vertex. The texture color value is then attached to the surface of the corresponding vertex of the initial three-dimensional mesh model to obtain the target three-dimensional mesh model.

Images