Quality control method and system for small box girder

Abstract

The invention provides a quality control method and system for a small box girder, and the method comprises the following steps: constructing a small box girder design BIM model according to a bridge design drawing; converting the BIM design model of the small box girder into discrete point cloud; obtaining a prefabricated field template point cloud, and preprocessing the prefabricated field template point cloud; performing registration on the point cloud model and the BIM model; and selecting a control point on the registered point cloud model as a reference for external mold control lofting, and calculating and evaluating field lofting data. According to the quality control method and system for the small box girder provided by the invention, the point cloud is generated by adopting BIM model interpolation, high-precision registration is completed in combination with a point cloud-point cloud registration algorithm, and the registration effect is evaluated based on the root-mean-square error. And closed-loop quality control from design to construction reduces manual intervention and can realize full-process digitization. And through algorithm fusion and error registration, engineering requirements are met, and the precision is effectively improved. The problems that a traditional method is low in precision and poor in efficiency are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of precast box girders, and specifically relates to a quality control method and system for small box girders. Background Technique

[0002] Traditional bridge construction usually adopts the method of casting concrete on site, which has certain limitations in terms of construction period, cost, environmental impact and quality control. With the progress of technology, modern bridge construction increasingly adopts precast box girders and virtual installation technology to improve construction efficiency, reduce costs, minimize environmental impact, and ensure higher quality and safety.

[0003] Currently, among the number of highway bridges in China, bridges adopting precast prestressed concrete small box girder structures have accounted for 71.72%. Studying the precast control technology of precast small box girders is of great significance for the quality of modern bridges. In recent years, combined small box girder structure bridges have played an increasingly important role in bridge construction due to their excellent safety, convenience and durability. This structure is not only applicable to various complex terrain and environmental conditions, but also shows significant advantages during construction and maintenance, becoming one of the first choices for modern bridge engineering. The precast box girders are precast in the precast yard and then transported to the site for installation, greatly shortening the construction period. The precast yard for box girders on site needs to pass strict quality control to ensure that the precast box girders meet the design specifications and standards during the manufacturing process. The flatness, smoothness, non-deformation of the surface of the formwork in the precast yard and the control technology of the installation accuracy of the formwork determine whether the external shape of the precast box girder meets the design dimensional requirements. Currently, on the precast site, mainly the closing of the formwork is controlled by measuring tools such as a folding rule at key positions, lacking overall control. Therefore, it is very necessary to study a quality control method and system for small box girders to achieve digital quality control of box girder precast. Summary of the Invention

[0004] The purpose of the present invention is to provide a quality control method and system for small box girders, aiming to solve the technical problems of low precision and poor efficiency existing in the prior art.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is: to provide a quality control method for small box girders, including:

[0006] Construct a BIM design model of the small box girder according to the bridge design drawings; then convert the BIM design model of the small box girder into discrete point clouds;

[0007] Obtain the point cloud of the formwork in the precast yard and preprocess the point cloud of the formwork in the precast yard;

[0008] Register the point cloud model with the BIM model;

[0009] Select control points on the registered point cloud model as the benchmark for the external mold control lofting, and calculate and evaluate the on-site lofting data.

[0010] Preferably, according to the bridge design drawings, construct a BIM model for the design of small box girders; then convert the BIM design model of the small box girders into discrete point clouds, including:

[0011] Create a structure to store the model information of the BIM model for the design of small box girders;

[0012] Read the triangular patch data of the BIM model of the small box girder to determine the number of patches;

[0013] Obtain the point cloud corresponding to the BIM geometric model.

[0014] Preferably, the obtaining of the point cloud corresponding to the BIM geometric model includes:

[0015] Create a function to traverse the indices of all patches and perform equidistant interpolation on each triangular patch;

[0016] Output as a TXT file containing the virtual point cloud, which is merged with the measured point cloud for subsequent registration.

[0017] Preferably, the obtaining of the precast yard template point cloud and the preprocessing of the precast yard template point cloud include:

[0018] Collect on-site data;

[0019] Preprocess the 3D point cloud of the precast yard to obtain a 3D point cloud model; where the preprocessing includes one or more of denoising, downsampling, and stitching;

[0020] First perform plane fitting, and then calculate the surface flatness.

[0021] Preferably, the collecting of the on-site data includes:

[0022] Conduct exploration and point selection and control point layout around the precast yard;

[0023] Use a scanner to scan on-site to collect the data required for the 3D point cloud of the precast yard;

[0024] Perform supplementary scanning on the specular reflection area to reduce data loss.

[0025] Preferably, the registration of the point cloud model and the BIM model includes:

[0026] Achieve initial alignment through rough registration; where rough registration is performed by the SAC-IA algorithm method or / and the 4PCS algorithm;

[0027] Achieve refined alignment through fine registration; among them, fine registration is performed through the ICP algorithm or / and the key-point-based registration algorithm;

[0028] Use the root mean square error to evaluate the registration result.

[0029] Preferably, the initial alignment through coarse registration includes:

[0030] Input two site point clouds: the point cloud P to be registered, the reference point cloud Q; calculate the normal information of their respective point clouds: N P 、N Q ;

[0031] Calculate the Fast Point Feature Histogram FPFH of their respective point clouds: F P 、F Q ;

[0032] Randomly sample m points P1, P2, P3,..., P m in the point cloud P to be registered; and find m points Q1, Q2, Q3,..., Q m in the reference site point cloud Q that are closest to the Fast Point Feature Histogram of the above m points; then use these m pairs of points to calculate the transformation matrix T, and apply this matrix T to the point cloud P to be registered: P' = P * T; find corresponding points;

[0033] Calculate the registration error ∈ between P and Q', and store the corresponding transformation matrix T;

[0034] Judge whether the iteration count i_iter is less than the set count MaxIterCounts, and jump out of the loop if it is not less than;

[0035] Compare all sets of registration errors in the loop process, select a set of the smallest registration error, output the corresponding transformation matrix T, and apply it to the original point cloud P to be registered to complete the initial registration.

[0036] Preferably, the refined alignment through fine registration includes:

[0037] Initialization: P0 = P,

[0038] the point cloud P to be registered k Use the neighborhood search method in the reference point cloud Q to search for the nearest neighbor points corresponding to each point in the point cloud P' to form adjacent point pairs

[0039] Calculate the parameters corresponding to the three-dimensional transformation matrix using all the nearest neighbor point pairs

[0040] Let R k 、Tk Applied to the point cloud P to be registered, a new point cloud of the station to be registered is obtained

[0041] Compare whether the difference between the error value this time and the error value last time is less than the given threshold ε: d k -d k+1 < ε, if satisfied, stop the iteration.

[0042] It should be noted that this algorithm requires the two-site point clouds to be in a good initial superposition state at the beginning, and it is required that the point cloud to be registered is a subset of the reference site point cloud.

[0043] Preferably, control points are selected on the registered point cloud model as the reference for external mold control lofting, and the on-site lofting data is calculated and evaluated, including:

[0044] Select the control points for template assembly and extract the coordinates;

[0045] Output suggestions for adjusting the external template lofting.

[0046] The present invention also provides a quality control system for small box girders, including:

[0047] The BIM model module is used to construct the BIM design model of the small box girder according to the bridge design drawings; then convert the BIM design model of the small box girder into a discrete point cloud;

[0048] The three-dimensional point cloud module is used to obtain the point cloud of the precast yard template and preprocess the point cloud of the precast yard template;

[0049] The registration module is used to register the point cloud model with the BIM model;

[0050] The evaluation module is used to select control points on the registered point cloud model as the reference for external mold control lofting, and calculate and evaluate the on-site lofting data.

[0051] The beneficial effects of a quality control method and system for small box girders provided by the present invention are as follows: Compared with the prior art, the quality control method and system for small box girders of the present invention use BIM model interpolation to generate point clouds, complete high-precision registration by combining point cloud-point cloud registration algorithms, and evaluate the registration effect based on the root mean square error. The closed-loop quality control from design to construction reduces manual intervention and can achieve full-process digitization. Through algorithm fusion (4PCS + ICP + FPFH), the registration error ≤ 4 cm, meeting the engineering requirements and effectively improving the accuracy. Avoiding traditional measurement rework, shortening the template adjustment time by more than 50%, and improving the efficiency. Solving the pain points of low accuracy and poor efficiency of traditional methods. Description of the Drawings

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 Schematic diagram of the small box girder design BIM model constructed for a quality control method for small box girders provided by an embodiment of the present invention;

[0054] Figure 2 Effect schematic diagram of model interpolation and completion of the small box girder design BIM model in a quality control method for small box girders provided by an embodiment of the present invention Figure 1 ;

[0055] Figure 3 Effect schematic diagram of model interpolation and completion of the small box girder design BIM model in a quality control method for small box girders provided by an embodiment of the present invention Figure 2 ;

[0056] Figure 4 Registration effect in a quality control method for small box girders provided by an embodiment of the present invention Figure 1 ;

[0057] Figure 5 Registration effect in a quality control method for small box girders provided by an embodiment of the present invention Figure 2 ;

[0058] Figure 6 Registration effect in a quality control method for small box girders provided by an embodiment of the present invention Figure 3 ;

[0059] Figure 7 Registration effect in a quality control method for small box girders provided by an embodiment of the present invention Figure 4 ;

[0060] Figure 8 Structure block diagram of a quality control system for small box girders provided by an embodiment of the present invention.

[0061] In the figure: 1. BIM model module; 2. Three-dimensional point cloud module; 3. Registration module; 4. Evaluation module. Detailed implementation manners

[0062] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0063] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same category, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the associated objects before and after.

[0064] The following will, with reference to the accompanying drawings, through specific embodiments and their application scenarios, elaborate in detail on a quality control method for small box girders provided by the embodiments of the present application.

[0065] Please refer to Figures 1 to 8 to illustrate a quality control method for small box girders provided by the present invention. The quality control method for small box girders obtains the internal point cloud of the outer formwork of the small box girder by scanning and constructing the outer formwork of the small box girder, analyzes the flatness of the bottom formwork and the side formwork, compares the laser point cloud with the designed BIM model, finds the accurate positions of the control points for the assembly of the outer formwork, and proposes adjustment suggestions for the closure of the jig, so as to realize the control of the prefabrication quality of the small box girder. Specifically, it includes the following steps:

[0066] Step S1: Construct a designed BIM model of the small box girder according to the bridge design drawings; convert the BIM design model of the small box girder into a discrete point cloud. Among them, a parametric small box girder model is created using BIM software, and the model includes details such as geometric features (bottom surface, side wall, top surface), prestressed duct, and steel bar mesh. The method of converting the BIM design model of the small box girder into a discrete point cloud is: interpolate points one by one for all triangular patches of the BIM model according to the centroid coordinate rule, and the point spacing is the same as the point spacing of the formwork point cloud in the prefabrication yard, so as to obtain the point cloud corresponding to the BIM geometric model.

[0067] The specific implementation process of this step is:

[0068] Step S1.1: Create a structure to store the model information of the designed BIM model of the small box girder, and the model information includes but is not limited to vertices, normals, and texture coordinates;

[0069] Step S1.2: Read the triangular facet data of the small box girder BIM model and determine the number of facets. Specifically, read and parse the obj file of the model to obtain the face data of the small box girder and determine the number of triangular facets.

[0070] Step S1.3: Obtain the point cloud corresponding to the BIM geometric model. Specifically, create a function to traverse the indices of all facets, perform equidistant interpolation on each triangular facet; set a parameter to control the interpolation density, determine the optimal parameter value through experiments, and finally output a TXT file containing the virtual point cloud, which is merged with the measured point cloud for subsequent registration. Among them, the best effect is achieved when the interpolation density parameter is 14. Specifically, interpolate points one by one for all triangular facets of the BIM model according to the barycentric coordinate rule, with the point spacing being the same as that of the precast yard template point cloud, to obtain the point cloud corresponding to the BIM geometric model.

[0071] Step S2: Obtain the precast yard template point cloud and preprocess the precast yard template point cloud.

[0072] In this step, 3D laser scanning can replicate the real scene. Through processes such as on-site point cloud scanning, stitching, denoising, data resampling, data analysis, and result inspection, the scanned point cloud is processed to obtain the 3D point cloud of the precast yard.

[0073] The specific implementation process of this step is as follows:

[0074] Step S2.1: Collect on-site data. Specifically, it includes:

[0075] Step S2.1.1: Conduct exploration and point selection and control point layout around the precast yard;

[0076] Step S2.1.2: Use the scanner to scan on-site to collect the data required for the 3D point cloud of the precast yard;

[0077] Step S2.1.3: Perform supplementary scanning for the specular reflection area (such as the bottom plate coating) to reduce data loss;

[0078] Step S2.2: Preprocess the 3D point cloud of the precast yard to obtain a 3D point cloud model. Among them, the preprocessing includes one or more of denoising, downsampling, and stitching. More specifically, by performing denoising, downsampling, and stitching processing in sequence, it realizes the functions of filtering environmental noise, reducing the point cloud density, improving the calculation efficiency, and aligning multi-station scanning data based on control points to form a complete template point cloud.

[0079] Step S2.3: Analyze the flatness of the template. Specifically, first perform plane fitting and then calculate the surface flatness. Among them, the least squares method is used for the plane fitting algorithm. Flatness calculation: Statistically calculate the distances from all points to the plane and calculate the average deviation.

[0080] In this step, it is required that the surface of the template pedestal in the prefabrication yard should be smooth and flat, the allowable deviation of the flatness within a length of 2 m should not exceed 2 mm, and the deflection of the base or bottom formwork should be ensured not to be greater than 2 mm.

[0081] Step S3: Register the point cloud model with the BIM model;

[0082] The specific implementation process of this step is as follows:

[0083] Step S3.1: Achieve initial alignment through rough registration; among them, rough registration is performed by the SAC-IA algorithm method or / and the 4PCS algorithm;

[0084] SAC-IA algorithm: Through sampling consistency initial registration, automatically search for corresponding point pairs and calculate the initial transformation matrix. Specifically, first search for the nearest corresponding point pair for each point in the box girder point cloud in the mold point cloud, and then calculate the rigid body transformation matrix using the found point pairs. Apply this transformation matrix to the box girder point cloud and calculate the mean square error of the distances between the transformed box girder point cloud and the mold point cloud. In the algorithm, the process of searching for the nearest point to the final error evaluation is iterated repeatedly until the error between two iterations is less than the threshold, and the iteration process ends.

[0085] 4PCS algorithm: Based on the search for four-point congruent sets, quickly match the geometric features of the point cloud and the model. Use the 4PCS (4-Points Congruent Sets) algorithm for rough registration. Achieve basic matching of the point cloud and the model through rough registration. Specifically, find several groups of corresponding point pairs in the two-site point clouds and calculate a rigid body transformation matrix using these point pairs. Apply this matrix to the point cloud to be registered to make the two-site point clouds fit together as much as possible, thus facilitating the subsequent fine registration work. At the same time, once the initial registration work is completed, the overlapping part of the two-site point clouds can be easily calculated, so it can well provide good fast convergence conditions for the fine registration work.

[0086] The specific implementation process of this step can be as follows:

[0087] Step S3.1.1: Input the two-site point clouds: the point cloud P to be registered and the reference point cloud Q; calculate the normal information of their respective point clouds: N P 、N Q ;

[0088] Step S3.1.2: Calculate the fast point feature histogram FPFH of their respective point clouds: F P 、FQ ;

[0089] Step S3.1.3: Randomly sample m points P1, P2, P3, …, P from the to-be-registered site cloud P m ; and find m points Q1, Q2, Q3, …, Q in the reference station site cloud Q that are closest to the point feature quick histograms of the above m points m ; then calculate the transformation matrix T using these m pairs of points, and apply this matrix T to the to-be-registered site cloud: P' = P * T; Find corresponding points.

[0090] Step S3.1.4: Calculate the registration error ∈ between P and Q', and store the corresponding transformation matrix T

[0091] Step S3.1.5: Determine whether the iteration count i_iter is less than the set count MaxIterCounts. If it is less, return to Step S2.2 and continue to execute backward to Step S2.4; otherwise, jump out of the loop

[0092] Step S3.1.6: Compare all sets of registration errors during the loop, filter out a set of the minimum registration error, output the corresponding transformation matrix T, and apply it to the original to-be-registered site cloud P to complete the initial registration.

[0093] In this step, the specific details can be divided into the following modules:

[0094] Neighborhood search selection.

[0095] For the neighborhood search selection algorithm

[0096] First, clarify the concepts of feature descriptors and neighborhoods. For a given point P q , the set of points within its corresponding neighborhood is represented by .

[0097] These neighborhood point sets can be determined by the following expression:

[0098]

[0099] In the formula, d m is the neighborhood search radius, that is, the maximum distance between the neighborhood point and the query point, and ||·|| x is an example of a Minkowski norm (other different distance norms can also be used).

[0100] In addition to being constrained by the radius, the neighborhood point set can also be described by a neighborhood limit number K, that is, the K points closest to the point P q are used as P qThe set of neighborhood points. Therefore, the point feature descriptor can be represented by a vector function F, which can describe the local geometric information reflected by passing through point P q The set of neighborhood points P around k The reflected local geometric information is as follows:

[0101] F(P q , P k ) = {x1, x2, x3…x n}#

[0102] Where xi i , i ∈ {1…n} respectively represent the values of the i-th dimension of the feature vector.

[0103] Then, to compare whether two points p1 and p2 have the same features, it only needs to compare whether the values of their feature vectors F1 and F2 in each dimension are the same or close.

[0104] Let Γ be the similarity expression of points p1 and p2, and d be a scale metric function, then:

[0105] Γ = d(F1, F2)

[0106] Therefore, when Γ approaches a certain minimum value, that is, Γ → 0, then the feature descriptors of these two points are considered similar. On the contrary, if Γ is a very large value, then these two points are considered to be very different from each other, that is, they should be on different geometric surfaces. By finding the neighborhood point set of a single point, the geometric features of the local surface of this point can be well expressed by the point feature descriptor. Therefore, point cloud registration can rely on this feature variable when looking for corresponding relationships, find the homologous point pairs with the same features in the two-site point clouds, and thus solve the registration problem. Then, the normal and fast point feature histogram calculations of the point cloud are carried out. When using the FLANN neighborhood search library for neighborhood search work, when looking for the point set with similar features to the m points sampled by the point cloud P to be registered in the reference point cloud Q of the benchmark site, neighborhood search also needs to be carried out. The neighborhood search here mainly searches for the points closest to the fast point feature histogram of the sample point as the neighborhood of the sample point. When calculating the registration error ∈ between P' and Q, the FLANN library is used to search for the nearest neighborhood points of each point in the transformed point cloud P to be registered in the reference point cloud Q of the benchmark site.

[0107] Search for homologous point pair sets.

[0108] For the search for homologous point pair sets, after the normal and histogram calculations of the point cloud are completed, the sample points of the point cloud P to be registered need to be selected.

[0109] According to the principle of measurement control, the control points should be evenly distributed, and the corresponding control points in the registration work should also follow this principle. Therefore, geometric constraints are required when randomly sampling the sample points of the reference station point cloud, and a distance threshold ε between the sample points needs to be set. dist , and this threshold is determined according to the point cloud density, and it is only necessary to ensure that the sampled points are not distributed in a narrow area.

[0110] If no points can be sampled within the set threshold, the algorithm should be designed to automatically reduce the set threshold and continue sampling until appropriate sample points are collected.

[0111] Since the purpose of sampling points and subsequent finding point pairs is to calculate a transformation matrix, and the rigid body transformation matrix between point sets includes three rotation angles and three translation amounts, the necessary number of point pairs for solving the transformation matrix should be 3 groups of three-dimensional point pairs. Therefore, only three points need to be sampled in the algorithm, and the number of sampled points can also be appropriately increased.

[0112] Search for the corresponding point sets with similar features of the three sample points P1, P2, and P3 in the point cloud P of the station to be registered in the reference station point cloud Q. k neighborhood values are searched for each sample point, that is, the following point pair set is formed:

[0113] C = {c i ∥c i = <P I , Q i1 , Q i2 ,…, Q ik >, 0 < i < 3} #

[0114] In the above formula, each sampled point corresponds to k most similar feature points. Here, a point is randomly sampled from its corresponding k neighborhood points as the point pair of the sample point, so as to construct the corresponding relationship between the two point clouds of the stations.

[0115] Calculation of the transformation matrix.

[0116] For the calculation of the transformation matrix, when the corresponding relationship between the two point clouds of the stations is available, the corresponding transformation matrix T is calculated.

[0117] The three-dimensional transformation matrix mainly consists of three modules: rotation, translation, and scaling. If the original coordinates of the three-dimensional point are [x y z] and the new coordinates after transformation are [x' y' z']. Then there are:

[0118] ① The formula for rotation transformation is:

[0119]

[0120] ② The formula for translation transformation is:

[0121]

[0122] ([T x T y T z is the translation amount of three - dimensional coordinate values)

[0123] ③ The scaling transformation formula is:

[0124]

[0125] ([S x S y S z is the scaling coefficient on the three - dimensional axis)

[0126] Step S3.2: Achieve refined alignment through fine registration; among them, fine registration is performed by the ICP algorithm or / and the registration algorithm based on key points.

[0127] ICP algorithm: Utilize the geometric invariance in rigid body transformation (vector / line segment ratio, Euclidean distance between points). According to the characteristics that the ratio of the intersection points occupied by the line segment remains unchanged and the Euclidean distance between points remains unchanged after rigid transformation, try to find 4 approximately coplanar points (approximate congruent four - point set) in the target point cloud to correspond to them, so as to calculate the transformation matrix using the least - squares method. Based on the RANSAC algorithm framework, multiple groups of bases are iteratively selected, and the optimal transformation is obtained by comparing according to the evaluation criterion of the largest common point set (LCP). Specifically, for two groups of point clouds, an initial transformation (which can be random or estimated) is selected. In each iteration, find the nearest neighbor point (a point in the target point cloud) for each point in the source point cloud, and calculate the error between each pair of points; then, by minimizing these errors (usually the least - squares method), calculate the best transformation (translation and rotation), and finally update the transformation, repeating the iteration until the error converges.

[0128] The specific implementation process of this step can be: Given the existing point cloud P of the site to be registered = {p i |p i ∈R 3 ; i = 1, 2, 3, …, n} and the reference site point cloud Q = {q i |q i ∈R 3 ; i = 1, 2, 3, …, m}, where n ≤ m. The algorithm first traverses each point P i in the point cloud P, and searches for the nearest point Y i in the point cloud Q as the corresponding point. If the symbol is defined as the operation of searching for the nearest neighbor point, then there is: After the point pairs are established, the transformation matrix can be calculated. The calculation of the transformation matrix uses the least - squares principle, and the transformation matrix when the following error measure is minimized is the required matrix.

[0129]

[0130] Solve the error measure The rigid body transformation parameters at the minimum That is:

[0131]

[0132] In the formula, d is the mean square error of the Euclidean distance between the two site point clouds after the corresponding transformation, that is Apply the obtained rigid body transformation to the point cloud P to be registered, denoted as: Then iterate repeatedly until the iteration convergence condition is met.

[0133] That is, the specific implementation process of step S3.2 can be:

[0134] S3.2.1. Initialization: P0 = P,

[0135] S3.2.2. For the point cloud P to be registered k Use the neighborhood search method in the reference site point cloud Q to search for the nearest neighbor points corresponding to each point in the point cloud P' to form adjacent point pairs

[0136] S3.2.3. Calculate the parameters corresponding to the three-dimensional transformation matrix using all the nearest neighbor point pairs

[0137] S3.2.4. Apply R k , T k to the point cloud P to be registered to obtain a new point cloud P to be registered

[0138] S3.2.5. Compare whether the difference between the error value this time and the error value last time is less than the given threshold ε: d k -d k+1 < ε. If it is satisfied, stop the iteration; otherwise, repeatedly iterate from step S3.2.2 to step S3.2.4.

[0139] It should be noted that this algorithm requires the two site point clouds to be in a relatively good initial overlapping state at the beginning, and it is required that the point cloud to be registered is a subset of the reference site point cloud.

[0140] The registration algorithm based on key points completes the registration process:

[0141] According to the geometric characteristics of the small box girder and the mold, the Keypoint-based Registration algorithm is used to perform registration without manual operation. This method realizes automatic registration by automatically detecting key feature points in the point cloud. Commonly used feature extraction methods include SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), and FPFH (Fast Point Feature Histograms), etc. These methods perform registration by extracting local features in the point cloud and matching these features. The mean square error of the Euclidean distance between points and planes is used as the error measure. Based on this least squares principle, the transformation matrix is obtained using the quaternion method. Then, it enters the transformation result judgment module to determine whether the iteration reaches the convergence condition. After meeting the conditions, the final fine registration result is output. Key point matching can extract FPFH feature descriptors, enhancing the robustness of local geometric feature matching.

[0142] The specific implementation process of this step can be as follows: (High-precision alignment is achieved through automatic feature matching, transformation solution, and convergence judgment. Its innovation lies in combining key point features (such as FPFH) and robust filtering strategies, reducing the dependence on manual intervention, and being applicable to the efficient registration of complex structures such as small box girder molds)

[0143] Input the point cloud P of the site to be registered and the reference point cloud Q of the site;

[0144] Initialization: The transformation matrix T0 = I. Data reduction filtering is performed on the point cloud P of the site to be registered and the reference point cloud Q of the site: P' = datafilter(P), Q' = tafilter(Q). Set the iteration convergence threshold: ε Angle 、ε Trans ;

[0145] Start iteratively looping through the following processes a - d until the iteration converges:

[0146] a. Perform three-dimensional transformation on the point cloud P' to be registered: P' k = T k-1 (P' K );

[0147] b. Find the nearest neighbor point pairs: M k = match(P' K , Q');

[0148] c. Filter outlier point pairs: W k = outlier(M k );

[0149] d. Least squares estimation:

[0150] e. Judgment of transformation result:

[0151] Output the final registration matrix: where m is the number of iterations at the time of iterative convergence.

[0152] For the point cloud downsampling part, this algorithm uses random sampling filtering.

[0153] The random sampling set expression is as follows:

[0154] P' = datafilter(P) = {p ∈ P: η(p) < θ} #

[0155] In the formula, η ∈ [0, 1), η is any random value within this range. θ ∈ [0, 1], θ is the ratio setting value for point cloud sampling, and the range is between 0 and 1.

[0156] The normal calculation set expression is as follows:

[0157]

[0158] In the formula, normal(p) is the normal of point p estimated based on the neighborhood around point p.

[0159] By comparing the feature descriptors of key points, the most matching point pairs are found. The point pair matching module mainly uses the point cloud P of the site to be registered to find the corresponding nearest point pairs in the reference site cloud Q, and finally uses them for subsequent transformation matrix calculation.

[0160] There are three modes for searching neighboring point pairs: ① Euclidean distance matching of three-dimensional point coordinates, ② matching of additional feature description information, ③ hybrid matching. Neighborhood point matching is used for fine matching. Its core is to use a KD-TREE linear heap for point cloud space division. When searching for neighbors, a range search mechanism is adopted, and a maximum distance constraint threshold for point pairs is given. The nearest point within this threshold range is searched as the neighboring point. The point pair matching set expression is as follows:

[0161]

[0162] In the formula, d(P n , Q x ) is the Euclidean distance between point P n and point Q x , that is, the index x corresponding to the point closest to P n is selected from the index of the reference site cloud Q.

[0163] Then, the rigid transformation (rotation and translation) is calculated and optimized using the least squares method as shown in the following formula:

[0164]

[0165] Where n i Q is the midpoint of the Q cloud, the benchmark site i The three-dimensional normal vector, R and t are the rotation parameters and translation parameters of the transformation matrix T respectively. Finally, the transformation is adjusted to ensure that the two point clouds are aligned.

[0166] Step S3.3: Evaluate the registration result. Specifically, use the root mean square error to evaluate the registration result.

[0167] In this method, the root mean square error (RMSE) between points is selected as the error evaluation standard. The root mean square error can quantify the spatial distance error between point clouds. Specifically, the root mean square error calculates the distance from each point in the source point cloud to the nearest point in the target point cloud, and averages the squares of these distances and then takes the square root.

[0168] The specific implementation process of this step is:

[0169] For each point pi in the source point cloud, find the point qi in the target point cloud that is closest to it

[0170] Calculate the distance between each pair of points di = ||pi-qi||;

[0171] Calculate the average of the squared distances of all pairs of points and then take the square root:

[0172] Among them, NN is the number of points in the point cloud, and di is the distance between point pi and its nearest neighbor point qi.

[0173] Step S4: Select control points on the registered point cloud model as the basis for external model control lofting, and perform calculation and evaluation on the on-site lofting data.

[0174] The specific implementation process of this step is:

[0175] Step S4.1, select template assembly control points and extract coordinates.

[0176] In the overlapped area of the registered point cloud and the BIM model, select five pairs of control points (such as the four corners of the base plate and the center point of the side wall). Output the three-dimensional coordinates (X, Y, Z) and deviation values of the control points as the lofting reference.

[0177] Step S4.2: Output external template layout adjustment suggestions.

[0178] In this step, deviation analysis is first performed, and then adjustment suggestions are generated and dynamically fed back. Specifically, the coordinates of the measured control points are compared with the theoretical coordinates of the BIM model to calculate the translation amount and rotation angle. Adjustment instructions (such as the movement amount of the jig and the tightness of the bolts) are automatically generated through the "Quality Control Procedure for Prefabricated Small Box Girder". After adjustment, re-scanning verification is carried out until the error meets the specifications.

[0179] A quality control method for small box girders provided by the present invention scans the outer formwork of the prefabricated small box girder to obtain the point cloud inside the outer formwork of the small box girder, compares the laser point cloud with the designed BIM model, and puts forward adjustment suggestions for the closure of the jig to achieve the control of the quality of prefabricated small box girders. The completed prefabricated small box girder and the pier cap at the bridge position are scanned, the laser point cloud is compared with the designed box girder model, and the size deviation of the completed prefabricated small box girder is determined. According to the control benchmark for the installation of small box girders, the optimized boundary conditions are determined, a mathematical algorithm is developed and optimized to obtain the installation data, providing data support for the physical installation.

[0180] The present invention also provides a quality control system for small box girders, including: a BIM model module 1, a three-dimensional point cloud module 2, a registration module 3, and an evaluation module 4. The BIM model module 1 is used to construct a designed BIM model of the small box girder according to the bridge design drawings; then, model interpolation and completion are performed on the designed BIM model of the small box girder. The three-dimensional point cloud module 2 is used to obtain the point cloud of the formwork in the prefabrication yard and preprocess the point cloud of the formwork in the prefabrication yard. The registration module 3 is used to register the point cloud model with the BIM model. The evaluation module 4 is used to select control points on the registered point cloud model as the benchmark for the layout control of the outer formwork and calculate and evaluate the on-site layout data.

[0181] A quality control method and system for small box girders provided by the present invention, compared with the prior art, uses BIM model interpolation to generate point clouds, combines the point cloud-point cloud registration algorithm to complete high-precision registration, and evaluates the registration effect based on the root mean square error. The closed-loop quality control from design to construction reduces manual intervention and can achieve full-process digitization. Through algorithm fusion, the registration error ≤ 4 cm, meeting the engineering requirements and effectively improving the accuracy. Avoiding traditional measurement rework, shortening the template adjustment time by more than 50%, and improving the efficiency. It solves the pain points of low accuracy and poor efficiency of traditional methods.

[0182] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A quality control method for small box girders, characterized in that, Including the following steps: According to the bridge design drawings, construct the BIM model for the design of small box girders; then convert the BIM design model of the small box girders into discrete point clouds; Obtain the point cloud of the precast yard formwork and preprocess the point cloud of the precast yard formwork; Register the point cloud model with the BIM model; Select control points on the registered point cloud model as the benchmark for the external formwork control lofting, and calculate and evaluate the on-site lofting data.

2. The quality control method for small box girders according to claim 1, characterized in that, The step of constructing the BIM model for the design of small box girders according to the bridge design drawings; then converting the BIM design model of the small box girders into discrete point clouds includes: Create a structure to store the model information of the BIM model for the design of small box girders; Read the triangular facet data of the small box girder BIM model and determine the number of facets; Obtain the point cloud corresponding to the BIM geometric model.

3. The quality control method for small box girders according to claim 2, characterized in that, The step of obtaining the point cloud corresponding to the BIM geometric model includes: Create a function to traverse the indices of all facets and perform equidistant interpolation on each triangular facet; Output as a TXT file containing the virtual point cloud, which is merged with the measured point cloud for subsequent registration.

4. A quality control method for small box girders according to claim 2, characterized in that, The step of obtaining the point cloud of the precast yard formwork and preprocessing the point cloud of the precast yard formwork includes: Collect on-site data; Preprocess the three-dimensional point cloud of the precast yard to obtain a three-dimensional point cloud model; among them, the preprocessing includes one or more of denoising, downsampling, and splicing; First perform plane fitting, and then calculate the surface flatness.

5. A quality control method for small box girders as described in claim 4, characterized in that, The step of collecting on-site data includes: Conduct exploration and point selection and control point layout around the precast yard; Use a scanner to scan on-site and collect the data required for the three-dimensional point cloud of the precast yard; Perform supplementary scanning for the specular reflection area to reduce data loss.

6. The quality control method for small box girders according to claim 4, characterized in that The step of registering the point cloud model with the BIM model includes: Achieve initial alignment through rough registration; among them, rough registration is performed by the SAC-IA algorithm method or / and the 4PCS algorithm; Achieve refined alignment through fine registration; among them, fine registration is performed by the ICP algorithm or / and the registration algorithm based on key points; Use the root mean square error to evaluate the registration result.

7. A quality control method for small box girders as described in claim 6, characterized in that, The step of achieving initial alignment through rough registration includes: Input two site clouds: the site cloud to be registered P and the reference site cloud Q; calculate the normal information of each point cloud: N P 、N Q ; Calculate the Fast Point Feature Histogram (FPFH) of the respective point clouds: F P 、F Q ; Randomly sample m points P1, P2, P3, …, P from the point cloud P of the station to be registered respectively m ; and find m points Q1, Q2, Q3, …, Q in the point cloud Q of the reference station that are closest to the point feature fast histograms of the above m points m ; then calculate the transformation matrix T using these m pairs of points, and apply the matrix T to the point cloud of the station to be registered: P' = P * T; find corresponding points Calculate the registration error ∈ between P and Q', and store the corresponding transformation matrix T; Judge whether the iteration number i_iter is less than the set number MaxIterCounts, and jump out of the loop if it is not less than; Compare all sets of registration errors in the loop process, select a set of the smallest registration errors, output the corresponding transformation matrix T, and apply it to the original point cloud P to be registered to complete the initial registration.

8. A quality control method for small box girders as claimed in claim 7, characterized in that, The step of achieving refined alignment through fine registration includes: Initialization: P0 = P, k = 0; Point cloud P of the station to be registered k Using the neighborhood search method in the reference station cloud Q, the nearest neighbor points corresponding to each point in the point cloud P' are searched to form adjacent point pairs Calculate the parameters corresponding to the three-dimensional transformation matrix using all the nearest neighbor pairs Apply R k and T k to the point cloud P to be registered, obtaining a new point cloud of the site to be registered Compare whether the difference between the current error value and the previous error value is less than the given threshold ε: d k -d k+1 < ε. If satisfied, stop the iteration; It should be noted that this algorithm requires the two-site point clouds to be in a relatively good initial superposition state at the beginning, and it is required that the point cloud to be registered is a subset of the reference site point cloud.

9. A quality control method for small box girders according to claim 1, characterized in that, The step of selecting control points on the registered point cloud model as the benchmark for the external formwork control lofting and calculating and evaluating the on-site lofting data includes: Select the control points for formwork assembly and extract the coordinates; Output the adjustment suggestions for the external formwork lofting.

10. A quality control system for small box girders, characterized in that, Including: The BIM model module is used to construct the BIM model for the design of small box girders according to the bridge design drawings; then convert the BIM design model of the small box girders into discrete point clouds; 3D point cloud module, which is used to obtain the prefabrication yard template point cloud and preprocess the prefabrication yard template point cloud; Registration module, which is used to register the point cloud model with the BIM model; Evaluation module, which is used to select control points on the registered point cloud model as the benchmark for external mold control lofting and calculate and evaluate the on-site lofting data.

Images