Pile foundation concrete precision pouring control method based on multi-source sensing fusion

Abstract

This invention belongs to the field of concrete pouring technology, and particularly relates to a precise control method for pile foundation concrete pouring based on multi-source sensor fusion. Addressing the problems of large measurement errors, frequent over-pouring, and significant concrete waste in traditional pile foundation pouring, this invention employs a high-frequency ultrasonic probe array coaxially fixed to the outer wall of the pouring guide pipe to uniformly elevate and scan, acquiring three-dimensional point cloud data of the entire pile foundation. After two levels of outlier removal and adaptive multi-scale spectral filtering preprocessing, combined with the expansion of the over-pouring section point cloud, a high-precision three-dimensional model of the pile foundation is constructed. Cross-sectional features at each depth are extracted to generate a pumping flow rate feedforward compensation curve, and a segmented adaptive improved PID controller is constructed with real-time concrete interface elevation and rise speed as feedback quantities to dynamically regulate the pumping flow rate. This invention achieves automated, intelligent, and precise control of the entire pile foundation pouring process, effectively improving pouring control accuracy, significantly reducing concrete waste and construction costs, and solving the measurement and pouring control problems of traditional construction.

Description

Technical Field

[0001] This invention belongs to the field of concrete pouring technology, and particularly relates to a method for precise control of pile foundation concrete pouring based on multi-source sensor fusion. Background Technology

[0002] During the construction of bored piles, the concrete interface height is not visible during concrete pouring. Sediment at the pile bottom and impurities deposited in the mud during pouring accumulate to a certain thickness on the concrete surface, easily forming laitance at the pile head, leading to widespread over-pouring. Common methods for monitoring over-pouring include the volumetric method and the heavy hammer method. If the pile borehole diameter changes during construction, the method of controlling the pile top elevation based on the volume of poured concrete becomes inaccurate. On construction sites, the pile top elevation of bored piles is mostly measured manually using measuring ropes. This traditional method relies entirely on the worker's experience, and the results vary from person to person, making it not only inaccurate but also time-consuming and labor-intensive. Especially in the construction of underwater bored piles with low-elevation pile tops, under-pouring can directly lead to the scrapping of the pile, causing serious construction accidents. To ensure the quality of the concrete at the top of the pile, it is generally necessary to overfill by 0.8 to 1.0m. However, due to differences in measurement methods and levels, and in order to avoid scrapping the pile foundation, the height of overfilling often far exceeds 1.0m, and in some cases even reaches more than 5.0m. This not only causes serious waste of concrete materials, but also makes earthwork excavation difficult, increases the cost of pile head breaking and transportation, and greatly increases the construction cost of the project. Summary of the Invention

[0003] In view of the technical problems existing in the background art, the present invention proposes a method for precise control of pile foundation concrete pouring based on multi-source sensor fusion.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0005] S1. A high-frequency ultrasonic probe array, coaxially fixed to the outer wall of the casting guide pipe, is used to continuously lift and scan from the bottom of the pile foundation at a constant speed to obtain the original three-dimensional point cloud data of the entire pile foundation.

[0006] S2. Perform preprocessing operations on the original three-dimensional point cloud data of the pile foundation to construct a high-precision three-dimensional point cloud model of the pile foundation; the preprocessing operations on the three-dimensional point cloud data include outlier removal and filtering operations.

[0007] S3. Based on the high-precision three-dimensional point cloud model of the pile foundation, extract the cross-sectional features of each depth segment of the pile foundation, and combine the preset concrete interface rising speed to obtain the benchmark pumping flow rate of the corresponding depth segment, and generate the feedforward compensation curve of the pumping flow rate of the entire pile foundation.

[0008] S4. Construct an adaptive improved PID controller with real-time concrete interface elevation and real-time interface rise rate as feedback quantities and pump flow rate as the controlled quantity. Input the feedforward compensation amount of pump flow rate for the corresponding depth segment into the controller in real time. The adaptive improved PID controller adaptively adjusts the control parameters and controls the pump flow rate in real time, so that the concrete interface rise rate is stabilized within the preset range until the concrete interface reaches the design elevation of the pile top, thus completing the precise pouring control.

[0009] Preferably, the specific implementation of anomaly removal in step S2 includes:

[0010] First, coordinate axis transformation is performed. A cylindrical coordinate system is established with the central axis of the coaxially fixed ultrasonic probe array casting guide as the Z-axis, the designed position at the pile bottom as the origin, and the vertically upward direction as the positive Z-axis. The Cartesian coordinates of the original three-dimensional point cloud of the pile foundation are then transformed. Convert to cylindrical coordinates ,in, The radial distance from that point. This is the circumferential angle at that point;

[0011] Along the Z-axis depth direction, using preset equidistant depth units. The point cloud of the entire pile foundation is segmented to obtain k consecutive depth elements, where k is the total pile length L and the depth element. The ratio, where each depth unit corresponds to a local point cloud: The point cloud depth value within each local point cloud set is at Within the range;

[0012] For each local point cloud By combining pile foundation design parameters and radial statistical characteristics, the first-level outlier point elimination is completed. First, the local point cloud is calculated. Mean radial distance of all points within with standard deviation Combined with the outer diameter of the pouring pipe Maximum allowable over-excavation radius for pile foundations Set a hard threshold constraint; if the point cloud satisfies... or These are directly identified as outliers and removed, resulting in a set of valid points after initial screening. ;

[0013] For the effective point cloud after initial screening Each point to be determined within Delineate the circumferential neighborhood and Belonging to the same depth unit, circumferential angle deviation is Define the depth region from the set of points within the range. To and Circumferential angle deviation at Within the range, the depth value deviation is The set of points within the range, where, Preset the circumferential neighborhood angle threshold; calculate the points to be determined respectively. Mean radial deviation in the circumferential neighborhood Mean radial deviation within the depth neighborhood The outlier degree of the two neighborhoods of the point to be determined is calculated by weighted summation based on the mean radial deviation in the circumferential neighborhood and the mean radial deviation in the depth neighborhood. For the effective point cloud after initial screening, calculate the mean outlier degree of the two neighborhoods of all undecided points within it. with standard deviation Set an adaptive two-neighborhood outlier threshold. ,in, The preset confidence coefficient is used if the point to be determined satisfies... If a point is identified as a local outlier, it is deleted, resulting in a set of valid points after further screening. .

[0014] Preferably, a filtering operation is performed after outlier removal. This filtering operation employs an adaptive multi-scale geometric signal separation mechanism based on local graph Laplacian spectral decomposition, specifically including:

[0015] For each target point in the effective point cluster after secondary screening Construct the initial k-nearest neighbor region Calculate points in the neighborhood and Spatial distance, dot product of normal vectors, and local curvature difference;

[0016] Construct an undirected weighted graph from point cloud data. Where the vertex set V corresponds to all points in the point cloud, and the edge set E is the set of point pairs that satisfy the topological association relationship, only if the neighboring points Belongs to target point When the k-nearest neighbor is used, the target point With neighboring points Edges are formed and included in edge set E; no edges are formed between points that are not topologically related; the weighted adjacency matrix W is adaptively constructed based on local geometric features.

[0017] Constructing elements in the weighted neighborhood matrix W Only when neighboring points Belongs to target point k-nearest neighbors It is not equal to 0, otherwise it is 0. The non-zero weight is defined as a product of three parts, including spatial distance weight, normal vector consistency weight and curvature similarity weight.

[0018] Based on the constructed weighted graph, the symmetric normalized graph Laplacian matrix is ​​calculated and eigenvalue decomposition is performed, transforming the spatial domain geometric signal of the point cloud to the spectral domain, obtaining eigenvalue and eigenvector sets corresponding to different geometric frequencies. The symmetric normalized graph Laplacian matrix is: Where I is the identity matrix, for Perform eigenvalue decomposition to obtain an ascending sequence of eigenvalues ​​and a corresponding sequence of orthogonal eigenvectors.

[0019] For each target point Calculate its local geometric complexity index Consistency between curvature and normal: ,in, These are the weighting coefficients;

[0020] For each feature vector Design an adaptive threshold function Targeting points The threshold is: ,in, Based on the threshold, The optimal cutoff eigenvalue is determined by estimating the global noise level of the point cloud. The standard deviation of the spectral kernel;

[0021] For each feature vector The coefficients are then subjected to soft thresholding to obtain the thresholded coefficients: ,in, These are the original spectral coefficients. ,in, Representative point The original coordinate vector;

[0022] The point cloud is reconstructed using inverse spectral transformation, and the denoised coordinates are reconstructed for each coordinate component using an iterative mechanism through inverse spectral transformation. Then, the reconstructed point cloud is used as new input for iterative optimization. In each iteration, the weighted adjacency matrix, the graph Laplacian matrix, and the threshold are updated. The process continues until the global average coordinate offset error is satisfied or the number of iterations reaches the preset upper limit, at which point cloud data is obtained.

[0023] Preferably, after constructing the high-precision three-dimensional point cloud model of the pile foundation in step S2, a point cloud expansion operation is also required. This is achieved by setting a preset over-irrigation elevation of the pile foundation, extracting the cross-sectional features of the final segment of the pile top in the high-precision three-dimensional point cloud model of the pile foundation, calculating the mean values ​​of the center coordinates and fitting radii of each cross-section within this interval, and using them as the reference cross-sectional parameters of the over-irrigation segment model. With the positive Z-axis as the extension direction, a continuous over-irrigation segment point cloud is generated from the scanning endpoint elevation to the preset over-irrigation elevation to construct a complete three-dimensional point cloud model.

[0024] Preferably, the specific implementation of step S3, which involves extracting the cross-sectional features of the pile foundation at each depth based on the high-precision three-dimensional point cloud model of the pile foundation, includes:

[0025] Along the Z-axis of the pile foundation column coordinate system, the high-precision three-dimensional point cloud model is divided into continuous depth section cross-sections, and the depth value corresponding to each cross-section is determined to form an ordered depth section sequence for the entire pile foundation.

[0026] For each depth section layer, select all point cloud data in the 3D point cloud model whose depth values ​​fall within the depth range of that section layer to form a two-dimensional planar point set for that depth section.

[0027] The least squares method is used to fit a circle to the two-dimensional plane point set of each depth section, and the coordinates of the fitted circle center and the fitted radius of the section are obtained by solving the problem.

[0028] After removing outliers in the center coordinates and fitting radius using a clustering algorithm, the average values ​​of the remaining fitted center coordinates and fitting radius are used to obtain the final fitted center coordinates and fitting radius of the depth section layer.

[0029] The cross-sectional area of ​​the section at each depth is calculated based on the final fitted radius, thus obtaining the cross-sectional characteristic data for each depth segment.

[0030] As a preferred method, the high-precision 3D point cloud model is divided into continuous depth cross-section layers. The division of the depth value corresponding to each cross-section layer is determined by obtaining the fitting radius from the two-dimensional plane point set. The difference between the fitting radii of the upper and lower two-dimensional plane point sets is judged. If it is less than the fitting radius difference threshold, it is determined to be the same depth segment; otherwise, it is a new depth segment.

[0031] Preferably, step S3, in combination with the preset concrete interface rising speed, obtains the benchmark pumping flow rate for the corresponding depth segment and generates the feedforward compensation curve for the pumping flow rate of the entire pile foundation. Specifically, the cross-sectional area is multiplied by the preset concrete interface rising speed to obtain the benchmark pumping flow rate for that depth segment. The benchmark pumping flow rate is mapped one-to-one with the corresponding depth according to the depth segment order, and the feedforward compensation curve for the pumping flow rate of the entire pile foundation is generated by interpolation.

[0032] Preferably, the specific implementation of step S4, adaptive adjustment of control parameters by the PID controller, is as follows:

[0033] Based on the extracted cross-sectional characteristics of each depth section of the pile foundation, and combined with the feedforward compensation curve of the pumping flow rate of the entire pile foundation, the entire pile foundation is divided into three continuous control segments, including the initial grouting segment at the pile bottom, the stable grouting segment of the pile body, and the over-grouting segment at the pile top.

[0034] The initial grouting section at the bottom of the pile adopts a PD control structure, the stable pouring section of the pile body adopts a standard PID control structure, and the over-grouting control section at the top of the pile adopts a PI control structure, with preset initial parameters and upper and lower limits of parameters respectively.

[0035] The core control deviation is set by combining the ascent speed deviation and elevation deviation using a weighted summation method.

[0036] For each segment, a multi-objective optimization function is constructed, the objective function being to minimize the ascent speed error and elevation deviation. The Pareto optimal PID parameter set for each segment is obtained by solving the particle swarm optimization algorithm.

[0037] Based on the PID parameter set obtained from the current segmentation and the core control deviation, the feedback control term is calculated. ;

[0038] Finally, the current segmented pumping flow rate feedforward compensation amount is... As a feedforward term, it is superimposed on the feedback control term, and then subjected to the output amplitude locking limit of the current segment to generate the final pumping flow control value. ,in, The upper and lower limit ranges of the output corresponding to the current segment are used to obtain the final control value.

[0039] Compared with existing technologies, the advantages and positive effects of this invention are as follows: It uses precise 3D point cloud modeling as its core, relies on a high-frequency ultrasonic array to collect raw data of the entire pile, and constructs a high-precision model through two-level outlier removal and adaptive spectral filtering. This is combined with pile vertex cloud expansion to complete the geometric information of the over-pouring section, accurately restoring the actual shape of the pile foundation. Based on cross-sectional features, a pumping flow feedforward compensation curve is generated to adapt to changes in pile diameter. Combined with segmented adaptive improved PID dual-feedback control, the pumping flow is dynamically adjusted using interface elevation and rise speed as feedback quantities. This fundamentally solves the problems of large errors and frequent over-pouring in traditional manual measurements, significantly improving pouring control accuracy, reducing concrete waste and construction costs, and achieving automated, intelligent, and precise control of the entire pile foundation pouring process. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating a method for precise concrete pouring control in pile foundations based on multi-source sensor fusion. Detailed Implementation

[0042] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0043] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0044] In this embodiment, addressing the problem of inaccurate control over concrete over-pouring height, this application provides a method for precise control of pile foundation concrete pouring based on multi-source sensor fusion, such as... Figure 1 As shown.

[0045] First, a high-frequency ultrasonic probe array, coaxially fixed to the outer wall of the pouring guide pipe, is used to continuously lift and scan from the bottom of the pile foundation at a constant speed to obtain the original three-dimensional point cloud data of the entire pile foundation; the endpoint of the continuous lifting scan is the set height of the pile foundation over-pouring.

[0046] Specifically, the first step was to complete the hardware deployment and on-site pre-calibration of the probe array. Eight high-frequency ultrasonic probes were fixed equidistantly and coaxially along the outer wall of the casting guide pipe at 360 degrees, forming a ring scanning array. The probe beams were perpendicularly pointed to the pile hole wall, and the operating frequency was selected at 300kHz to balance ranging accuracy and penetration in the underwater mud environment. A 30cm safety distance was reserved between the array and the bottom of the guide pipe to avoid damage from impacts during lowering and to eliminate blind spots in the pile bottom scanning. After the guide pipe was lowered to the designed elevation of the pile bottom, on-site calibration was completed: the ultrasonic propagation velocity was calibrated based on the on-site mud density and temperature, and the ranging zero point and installation deviation were calibrated. The lifting mechanism, depth encoder, angle encoder, and probe acquisition system were synchronously linked to achieve microsecond-level timing synchronization and eliminate data misalignment errors. Subsequently, uniform speed scanning and acquisition were initiated. The guide pipe was lifted vertically upward along the pile hole axis at a constant speed by the lifting mechanism. During the lifting process, the probe array was synchronously triggered at a sampling frequency of 10kHz to form original three-dimensional point cloud data of the pile foundation covering the entire pile hole.

[0047] Secondly, preprocessing operations are performed on the original three-dimensional point cloud data of the pile foundation to construct a high-precision three-dimensional point cloud model of the pile foundation; the preprocessing operations on the three-dimensional point cloud data include outlier removal and filtering operations.

[0048] Furthermore, the specific implementation of anomaly point removal includes: firstly, performing coordinate axis transformation, establishing a cylindrical coordinate system with the central axis of the casting guide pipe with the coaxially fixed ultrasonic probe array as the Z-axis, the designed position at the bottom of the pile as the origin, and the vertically upward direction as the positive Z-axis, and converting the Cartesian coordinates of the original three-dimensional point cloud of the pile foundation. Convert to cylindrical coordinates The conversion formula is: ,in, The radial distance from that point. Let be the circumferential angle of that point.

[0049] Along the Z-axis depth direction, using preset equidistant depth units. The point cloud of the entire pile foundation is segmented to obtain k consecutive depth elements, where k is the total pile length L and the depth element. The ratio, where each depth unit corresponds to a local point cloud: The point cloud depth value within each local point cloud set is at Within the range.

[0050] For each local point cloud By combining pile foundation design parameters and radial statistical characteristics, the first-level outlier point elimination is completed. First, the local point cloud is calculated. Mean radial distance of all points within with standard deviation The calculation method is as follows: ,in, For local point cloud aggregation The total number of point clouds within; combined with the outer diameter of the casting guide pipe. Maximum allowable over-excavation radius for pile foundations Set a hard threshold constraint; if the point cloud satisfies... or These are directly identified as outliers and removed, resulting in a set of valid points after initial screening. .

[0051] For the effective point cloud after initial screening Each point to be determined within The process involves constructing circumferential and deep neighborhoods, calculating the outlier degree of the two neighborhoods, and performing secondary outlier removal. Specifically, for the points to be determined... Delineate the circumferential neighborhood and Belonging to the same depth unit, circumferential angle deviation is Define the depth region from the set of points within the range. To and Circumferential angle deviation at Within the range, the depth value deviation is The set of points within the range, where, Preset the circumferential neighborhood angle threshold; calculate the points to be determined respectively. Mean radial deviation in the circumferential neighborhood Mean radial deviation within the depth neighborhood The calculation formula is: ,in, For the circumferential neighborhood The number of point clouds within, For deep neighborhood The number of point clouds within the area; the outlier degree of the point to be judged is calculated by weighted summation based on the mean radial deviation in the circumferential neighborhood and the mean radial deviation in the depth neighborhood. For the effective point cloud after initial screening, calculate the mean outlier degree of the two neighborhoods of all undecided points within it. with standard deviation Set an adaptive two-neighborhood outlier threshold. ,in, This is a preset confidence coefficient. If the point to be judged satisfies... If a point is identified as a local outlier, it is deleted, resulting in a set of valid points after further screening. .

[0052] Furthermore, after outlier removal, a filtering operation is required. This filtering operation employs an adaptive multi-scale geometric signal separation mechanism based on local graph Laplacian spectral decomposition, specifically including: for each target point in the effective point cloud after re-screening... Construct the initial k-nearest neighbor region Calculate points in the neighborhood and spatial distance dot product of normal vectors and local curvature difference ,in, for The Gaussian curvature is calculated using the eigenvalues ​​of the neighborhood covariance matrix.

[0053] Construct an undirected weighted graph from point cloud data. The vertex set V corresponds to all points in the point cloud, and the edge set E is the set of point pairs that satisfy topological relationships, where each edge belongs to the target only if its neighboring points belong to the target. point When the k-nearest neighbor is used, the target point With neighboring points Edges are formed and included in the edge set E; no edges are formed between points that are not topologically related; the weighted adjacency matrix W is adaptively constructed based on local geometric features.

[0054] Constructing elements in the weighted neighborhood matrix W Only when hour It is not equal to 0, otherwise it is 0. The non-zero weight is defined as the product of three parts, including the spatial distance weight. ,in To adapt to spatial scale, the normal vector consistency weights ,in The standard deviation of the normal kernel and the curvature similarity weights. ,in, It is an adaptive curvature scale.

[0055] Based on the constructed weighted graph, the symmetric normalized graph Laplacian matrix is ​​calculated and eigenvalue decomposition is performed, transforming the spatial domain geometric signal of the point cloud to the spectral domain, obtaining eigenvalue and eigenvector sets corresponding to different geometric frequencies. The symmetric normalized graph Laplacian matrix is: Where I is the identity matrix, for Perform eigenvalue decomposition to obtain an ascending sequence of eigenvalues. With the corresponding orthogonal eigenvector sequence .

[0056] For each target point Calculate its local geometric complexity index Consistency between curvature and normal: ,in, This is the weighting coefficient, with a value ranging from 0.3 to 0.7. Represented as All points belonging to the k-neighborhood.

[0057] For each feature vector Design an adaptive threshold function Targeting points The threshold is: ,in, The base threshold is set to 0.5 to 1 times the global average coordinate offset of the point cloud. The optimal cutoff eigenvalue is determined by estimating the global noise level of the point cloud. denoted as the standard deviation of the spectral kernel.

[0058] For each feature vector The coefficients are then subjected to soft thresholding to obtain the thresholded coefficients: ,in, These are the original spectral coefficients. ,in, Representative point The original coordinate vector; the point cloud is reconstructed through inverse spectral transformation, and the denoised coordinates are reconstructed for each coordinate component through inverse spectral transformation using an iterative mechanism: Then, the reconstructed point cloud is used as new input for iterative optimization. In each iteration, the weighted adjacency matrix, the graph Laplacian matrix, and the threshold are updated. The process continues until the global average coordinate offset error is satisfied or the number of iterations reaches the preset upper limit, at which point cloud data is obtained.

[0059] After constructing a high-precision 3D point cloud model of the pile foundation, point cloud expansion is required. This involves setting a preset over-irrigation elevation for the pile foundation, extracting the cross-sectional features of the final segment of the pile top in the high-precision 3D point cloud model, calculating the average coordinates of the fitted center and the fitted radius of each cross-section within this interval, and using these as the reference cross-sectional parameters for the over-irrigation segment model. Extending along the positive Z-axis, a continuous over-irrigation segment point cloud is generated from the scanning endpoint elevation to the preset over-irrigation elevation, thus constructing a complete 3D point cloud model. Specifically, pouring control needs to cover the final over-irrigation elevation. If the model only extends to the scanning endpoint, it will lead to a lack of feedforward compensation for the pumping flow rate at the end of the over-irrigation segment, causing a sudden change in the interface rise rate and inaccurate elevation control. Through point cloud expansion, a continuous model covering the entire interval from the pile bottom to the over-irrigation elevation can be constructed, ensuring the continuity of feedforward control and the accuracy of over-irrigation segment control, thereby reducing the problems of excessive or insufficient over-irrigation at the root. First, a pre-set over-grouting elevation for the pile foundation is established. This elevation is higher than the design elevation of the pile top and not lower than the endpoint elevation of the ultrasonic scan, ensuring that the expanded section covers the entire over-grouting control range. Then, the cross-sectional features of the final segment of the pile top are extracted from the high-precision 3D point cloud model of the pile foundation. The average values ​​of the center coordinates and the fitting radius of each cross-section within this interval are calculated and used as the reference cross-sectional parameters for the over-grouting section model, ensuring consistency between the expanded model and the actual shape of the pile body. Third, using the cylindrical coordinate system of the original pile body model, with the positive Z-axis as the extension direction, continuous over-grouting section cross-sectional point clouds are generated from the scanning endpoint elevation to the pre-set over-grouting elevation, using equidistant depth units consistent with the pile body. This constructs a seamlessly integrated 3D point cloud model of the over-grouting section with the pile body model. Fourth, the over-grouting section point cloud is merged with the original high-precision 3D point cloud model of the pile foundation to obtain a complete 3D point cloud model covering the entire pouring area, providing complete data support for the subsequent generation of the feedforward compensation curve for the entire pumping flow rate.

[0060] Based on a high-precision 3D point cloud model of the pile foundation, the cross-sectional features of each depth segment of the pile foundation are extracted. Combined with the preset concrete interface rise velocity, the benchmark pumping flow rate of the corresponding depth segment is obtained, and a feedforward compensation curve for the pumping flow rate of the entire pile foundation is generated. Specifically, along the Z-axis of the pile foundation column coordinate system, the high-precision 3D point cloud model is divided into continuous depth segment cross-sectional layers, and the depth value corresponding to each cross-sectional layer is determined to form an ordered depth cross-sectional sequence of the entire pile foundation. For each depth cross-sectional layer, all point cloud data in the 3D point cloud model whose depth value falls within the depth range of that cross-sectional layer are selected to form a two-dimensional plane point set for that depth cross-section. The least squares method is used to perform circular fitting on the two-dimensional plane point set of each depth cross-section to obtain the fitting circle center coordinates and fitting radius of the cross-section. After removing outliers in the circle center coordinates and fitting radius through a clustering algorithm, the average values ​​of the remaining fitting circle center coordinates and fitting radii are calculated to obtain the final fitting circle center coordinates and fitting radius of the depth cross-section layer. The cross-sectional area of ​​the depth cross-section is calculated based on the final fitting radius to obtain the cross-sectional feature data of each depth segment. The instantaneous value of the reference pumping flow rate is calculated by taking the cross-sectional characteristic data of the i-th depth segment, i.e., the cross-sectional area. The cross-sectional area and the rising speed of the pre-set concrete interface Multiply to obtain the baseline pumping flow rate for that depth range. The unit of cross-sectional area is square meters, the unit of rising speed is meters per second, and the unit of reference pumping flow rate is cubic meters per second. The reference pumping flow rate is mapped to the corresponding depth in the order of depth segments, and the feedforward compensation curve of pumping flow rate for the entire pile foundation is generated by interpolation.

[0061] Furthermore, the high-precision 3D point cloud model is divided into continuous depth cross-section layers. The division of the depth value corresponding to each cross-section layer is determined by obtaining the fitting radius from the two-dimensional plane point set. The difference between the fitting radii of the upper and lower two-dimensional plane point sets is judged. If it is less than the fitting radius difference threshold, it is determined to be the same depth segment; otherwise, it is a new depth segment.

[0062] Furthermore, the specific implementation of the feedforward compensation curve for the pumping flow rate of the entire pile foundation is as follows: multiply the cross-sectional area by the preset concrete interface rise rate to obtain the reference pumping flow rate for that depth segment; map the reference pumping flow rate to the corresponding depth one by one according to the depth segment order; and interpolate to generate the feedforward compensation curve for the pumping flow rate of the entire pile foundation.

[0063] Finally, an adaptive improved PID controller is constructed with the real-time elevation of the concrete interface and the real-time rising speed of the interface as feedback quantities, and the pumping flow rate as the controlled quantity. The feedforward compensation of the pumping flow rate for the corresponding depth segment is input into the controller in real time. The adaptive improved PID controller adaptively adjusts the control parameters and controls the pumping flow rate in real time, so that the rising speed of the concrete interface is stabilized within the preset range until the concrete interface reaches the design elevation of the pile top, thus completing the precise pouring control.

[0064] Specifically, based on the extracted cross-sectional characteristics of each depth section of the pile foundation, and combined with the feedforward compensation curve of the pumping flow rate of the entire pile foundation, the entire pile foundation T is divided into three continuous control segments. The division rules are as follows: less than or equal to 0.2T is the initial grouting segment at the pile bottom, greater than 0.2T and less than 0.9T is the stable grouting segment of the pile body, and greater than or equal to 0.9T is the over-grouting control segment at the pile top. PID control structures, initial parameters and constraint boundaries are preset for each segment.

[0065] The initial grouting section at the pile bottom uses a PD control structure to preset initial parameters. , upper and lower limits of parameters , ; Pile body stabilization pouring section: adopts standard PID control structure, with preset initial parameters , , upper and lower limits of parameters , , The over-irrigation control section at the pile top adopts a PI control structure with preset initial parameters. , upper and lower limits of parameters , .

[0066] By combining weighted summation with the upward speed deviation and elevation deviation To set the core control deviation , The weights are set accordingly. A multi-objective optimization function is constructed for each segment, with the objective function being to minimize the rise rate error and elevation deviation. The Pareto optimal PID parameter set for each segment is obtained by solving the particle swarm optimization algorithm. Based on the PID parameter set obtained for the current segment and the core control deviation... Calculate feedback control terms For the initial grouting section at the pile bottom: For the stable pouring section of the pile body: For the over-grouting control section at the pile top: ;in The rate of change of error, , , , , , , To obtain the Pareto optimal PID parameters for each segment using the particle swarm optimization algorithm.

[0067] Finally, the current segmented pumping flow rate feedforward compensation amount is... As a feedforward term, it is superimposed on the feedback control term, and then subjected to the output amplitude locking limit of the current segment to generate the final pumping flow control value. ,in, The upper and lower limit ranges of the output corresponding to the current segment are respectively set, and the final control value is sent to the pumping system in real time for execution. The controller continues to cycle the above steps until the concrete interface reaches the design elevation of the pile top, thus completing the precise pouring control of the pile foundation concrete.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for precise control of pile foundation concrete pouring based on multi-source sensor fusion, characterized in that, Includes the following steps: S1. A high-frequency ultrasonic probe array, coaxially fixed to the outer wall of the casting guide pipe, is used to continuously lift and scan from the bottom of the pile foundation at a constant speed to obtain the original three-dimensional point cloud data of the entire pile foundation. S2. Perform preprocessing operations on the original three-dimensional point cloud data of the pile foundation to construct a high-precision three-dimensional point cloud model of the pile foundation; the preprocessing operations on the three-dimensional point cloud data include outlier removal and filtering operations. S3. Based on the high-precision three-dimensional point cloud model of the pile foundation, extract the cross-sectional features of each depth segment of the pile foundation, and combine the preset concrete interface rising speed to obtain the benchmark pumping flow rate of the corresponding depth segment, and generate the feedforward compensation curve of the pumping flow rate of the entire pile foundation. S4. Construct an adaptive improved PID controller with real-time concrete interface elevation and real-time interface rise rate as feedback quantities and pump flow rate as the controlled quantity, and input the feedforward compensation amount of pump flow rate for the corresponding depth segment into the controller in real time. The adaptive improved PID controller adaptively adjusts the control parameters and controls the pump flow rate in real time, so that the concrete interface rise rate is stabilized within the preset range until the concrete interface reaches the design elevation of the pile top, thus completing precise pouring control.

2. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 1, characterized in that, The specific implementation of anomaly removal in step S2 includes: First, coordinate axis transformation is performed. A cylindrical coordinate system is established with the central axis of the coaxially fixed ultrasonic probe array casting guide as the Z-axis, the designed position at the pile bottom as the origin, and the vertically upward direction as the positive Z-axis. The Cartesian coordinates of the original three-dimensional point cloud of the pile foundation are then transformed. Convert to cylindrical coordinates ,in, The radial distance from that point. This is the circumferential angle at that point; Along the Z-axis depth direction, using preset equidistant depth units. The point cloud of the entire pile foundation is segmented to obtain k consecutive depth elements, where k is the total pile length L and the depth element. The ratio, where each depth unit corresponds to a local point cloud: The point cloud depth value within each local point cloud set is at Within the range; For each local point cloud By combining pile foundation design parameters and radial statistical characteristics, the first-level outlier point elimination is completed. First, the local point cloud is calculated. Mean radial distance of all points within with standard deviation Combined with the outer diameter of the pouring pipe Maximum allowable over-excavation radius for pile foundations Set a hard threshold constraint; if the point cloud satisfies... or These are directly identified as outliers and removed, resulting in a set of valid points after initial screening. ; For the effective point cloud after initial screening Each point to be determined within Delineate the circumferential neighborhood and Belonging to the same depth unit, circumferential angle deviation is Define the depth region from the set of points within the range. To and Circumferential angle deviation Within the range, the depth value deviation is The set of points within the range, where, Preset the circumferential neighborhood angle threshold; calculate the points to be determined respectively. Mean radial deviation in the circumferential neighborhood Mean radial deviation within the depth neighborhood The outlier degree of the two neighborhoods of the point to be determined is calculated by weighted summation based on the mean radial deviation in the circumferential neighborhood and the mean radial deviation in the depth neighborhood. For the effective point cloud after initial screening, calculate the mean outlier degree of the two neighborhoods of all undecided points within it. with standard deviation Set an adaptive two-neighborhood outlier threshold. ,in, The preset confidence coefficient is used if the point to be determined satisfies... If a point is identified as a local outlier, it is deleted, resulting in a set of valid points after further screening. .

3. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 2, characterized in that, After outlier removal, a filtering operation is required. This filtering operation employs an adaptive multi-scale geometric signal separation mechanism based on local graph Laplacian spectral decomposition, specifically including: For each target point in the effective point cluster after secondary screening Construct the initial k-nearest neighbor region Calculate points in the neighborhood and Spatial distance, dot product of normal vectors, and local curvature difference; Construct an undirected weighted graph from point cloud data. Where the vertex set V corresponds to all points in the point cloud, and the edge set E is the set of point pairs that satisfy the topological association relationship, only if the neighboring points Belongs to target point When the target point is within the k-nearest neighbor region. With neighboring points Edges are formed and included in edge set E; no edges are formed between points that are not topologically related; the weighted adjacency matrix W is adaptively constructed based on local geometric features. Constructing elements in the weighted neighborhood matrix W Only when neighboring points Belongs to target point k-nearest neighbors It is not equal to 0, otherwise it is 0. The non-zero weight is defined as a product of three parts, including spatial distance weight, normal vector consistency weight and curvature similarity weight. Based on the constructed weighted graph, the symmetric normalized graph Laplacian matrix is ​​calculated and eigenvalue decomposition is performed, transforming the spatial domain geometric signal of the point cloud to the spectral domain, obtaining eigenvalue and eigenvector sets corresponding to different geometric frequencies. The symmetric normalized graph Laplacian matrix is: Where I is the identity matrix, for Perform eigenvalue decomposition to obtain an ascending sequence of eigenvalues ​​and a corresponding sequence of orthogonal eigenvectors. For each target point Calculate its local geometric complexity index Consistency between curvature and normal: ,in, These are the weighting coefficients; For each feature vector Design an adaptive threshold function Targeting points The threshold is: ,in, Based on the threshold, The optimal cutoff eigenvalue is determined by estimating the global noise level of the point cloud. The standard deviation of the spectral kernel; For each feature vector The coefficients are then subjected to soft thresholding to obtain the thresholded coefficients: ,in, These are the original spectral coefficients. ,in, Representative point The original coordinate vector; The point cloud is reconstructed using inverse spectral transformation, and the denoised coordinates are reconstructed for each coordinate component using an iterative mechanism through inverse spectral transformation. Then, the reconstructed point cloud is used as new input for iterative optimization. In each iteration, the weighted adjacency matrix, the graph Laplacian matrix, and the threshold are updated. The process continues until the global average coordinate offset error is satisfied or the number of iterations reaches the preset upper limit, at which point cloud data is obtained.

4. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 1, characterized in that, After constructing the high-precision three-dimensional point cloud model of the pile foundation in step S2, a point cloud expansion operation is required. This is achieved by setting the over-irrigation elevation of the pile foundation, extracting the cross-sectional features of the final segment of the pile top in the high-precision three-dimensional point cloud model of the pile foundation, calculating the mean values ​​of the center coordinates and fitting radii of each cross-section within this interval, and using them as the reference cross-sectional parameters of the over-irrigation segment model. With the positive Z-axis as the extension direction, a continuous over-irrigation segment point cloud is generated from the scanning endpoint elevation to the preset over-irrigation elevation to construct a complete three-dimensional point cloud model.

5. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 1, characterized in that, The specific implementation of extracting the cross-sectional features of each depth segment of the pile foundation based on the high-precision three-dimensional point cloud model of the pile foundation in step S3 includes: Along the Z-axis of the pile foundation column coordinate system, the high-precision three-dimensional point cloud model is divided into continuous depth section cross-sections, and the depth value corresponding to each cross-section is determined to form an ordered depth section sequence for the entire pile foundation. For each depth section layer, select all point cloud data in the 3D point cloud model whose depth values ​​fall within the depth range of that section layer to form a two-dimensional planar point set for that depth section. The least squares method is used to fit a circle to the two-dimensional plane point set of each depth section, and the coordinates of the fitted circle center and the fitted radius of the section are obtained by solving the problem. After removing outliers in the center coordinates and fitting radius using a clustering algorithm, the average values ​​of the remaining fitted center coordinates and fitting radius are used to obtain the final fitted center coordinates and fitting radius of the depth section layer. The cross-sectional area of ​​the section at each depth is calculated based on the final fitted radius, thus obtaining the cross-sectional characteristic data for each depth segment.

6. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 5, characterized in that, The high-precision 3D point cloud model is divided into continuous depth cross-section layers. The division of the depth value corresponding to each cross-section layer is determined by obtaining the fitting radius from the two-dimensional plane point set. The difference between the fitting radii of the upper and lower two-dimensional plane point sets is judged. If it is less than the fitting radius difference threshold, it is determined to be the same depth segment; otherwise, it is a new depth segment.

7. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 1, characterized in that, The specific implementation of step S3, which combines the preset concrete interface rising speed to obtain the benchmark pumping flow rate for the corresponding depth segment and generate the feedforward compensation curve for the pumping flow rate of the entire pile foundation, is as follows: multiply the cross-sectional area by the preset concrete interface rising speed to obtain the benchmark pumping flow rate for that depth segment; map the benchmark pumping flow rate to the corresponding depth one by one according to the depth segment order; and interpolate to generate the feedforward compensation curve for the pumping flow rate of the entire pile foundation.

8. The method for precise pouring control of pile foundation concrete based on multi-source sensor fusion according to claim 1, characterized in that, The specific implementation of step S4, adaptive improvement of the PID controller adaptively adjusting the control parameters, is as follows: Based on the extracted cross-sectional characteristics of each depth section of the pile foundation, and combined with the feedforward compensation curve of the pumping flow rate of the entire pile foundation, the entire pile foundation is divided into three continuous control segments, including the initial grouting segment at the pile bottom, the stable grouting segment of the pile body, and the over-grouting segment at the pile top. The initial grouting section at the bottom of the pile adopts a PD control structure, the stable pouring section of the pile body adopts a standard PID control structure, and the over-grouting control section at the top of the pile adopts a PI control structure, with preset initial parameters and upper and lower limits of parameters respectively. The core control deviation is set by combining the ascent speed deviation and elevation deviation using a weighted summation method. For each segment, a multi-objective optimization function is constructed, the objective function being to minimize the ascent speed error and elevation deviation. The Pareto optimal PID parameter set for each segment is obtained by solving the particle swarm optimization algorithm. Based on the PID parameter set obtained from the current segmentation and the core control deviation, the feedback control term is calculated. ; Finally, the current segmented pumping flow rate feedforward compensation amount is... As a feedforward term, it is superimposed on the feedback control term, and then subjected to the output amplitude locking limit of the current segment to generate the final pumping flow control value. ,in, The upper and lower limit ranges of the output corresponding to the current segment are used to obtain the final control value.

Images