Prestressed intelligent tensioning and grouting quality monitoring method and system

Abstract

The present application relates to the field of image processing technology, and discloses a method and system for monitoring the quality of prestressed intelligent tensioning and grouting. The method includes: performing regional cascade detection and dual time series matching processing on standard image data to obtain positioning parameters and motion features; performing sub-pixel edge reconstruction and multi-view displacement field analysis processing on positioning parameters and motion features to obtain stress state data; performing adaptive grid segmentation and flow field dynamic feature extraction processing on the grouting area image to obtain the flow state parameters of the grouting material; performing spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain quality assessment indicators; performing multi-objective parameter optimization and closed-loop feedback control processing on the quality assessment indicators to obtain a quality control strategy based on construction process parameters. The present application improves the control accuracy and reliability of prestressed construction quality by establishing a collaborative monitoring mechanism for the tensioning process and the grouting process.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method and system for monitoring the quality of prestressed intelligent tensioning and grouting. Background Art

[0002] Prestressed intelligent tensioning and grouting construction is a key process in the construction of prestressed structures. At present, there are many intelligent monitoring methods at home and abroad. In the existing technology, tension force sensors, displacement sensors and other equipment are usually used to monitor the tensioning process, while pressure sensors, flow meters and other equipment are used to monitor the grouting process. These methods mainly rely on a single data source for quality assessment, such as judging the tensioning quality by measuring the force and displacement of the tensioning end, or evaluating the grouting effect by measuring the grouting pressure and flow. In terms of data processing, the existing technology mainly uses relatively simple methods such as threshold judgment and trend analysis for quality assessment, lacking in-depth mining and comprehensive analysis of multi-source data.

[0003] However, the existing technology has obvious shortcomings: First, the monitoring method based on a single data source is difficult to fully reflect the construction quality status. For example, it is difficult to detect local anomalies in the tensioning process by relying solely on force monitoring, and it is also difficult to accurately judge the density of grouting by pressure monitoring alone. Secondly, simple data processing methods cannot effectively handle complex working conditions during the construction process, such as the dynamic change characteristics of tensioning force and the flow laws of grouting materials. In addition, the existing methods lack coordinated analysis of the two processes of tensioning and grouting, and cannot optimize construction parameters from an overall perspective, resulting in insufficient accuracy and reliability of quality control. Summary of the invention

[0004] The present application provides a prestressed intelligent tensioning and grouting quality monitoring method and system, which is used to improve the control accuracy and reliability of prestressed construction quality by establishing a collaborative monitoring mechanism for the tensioning process and the grouting process.

[0005] In the first aspect, the present application provides a prestressed intelligent tensioning and grouting quality monitoring method, which includes: performing nonlinear projection transformation and cross-contrast enhancement processing on multi-perspective images collected at the construction site of the prestressed component to obtain standard image data; performing regional cascade detection and dual time series matching processing on the standard image data to obtain positioning parameters and motion characteristics corresponding to the prestressed component; performing sub-pixel edge reconstruction and multi-perspective displacement field analysis processing on the positioning parameters and motion characteristics to obtain stress state data of the tensioning end; performing adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain flow state parameters of the grouting material; performing spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain quality evaluation indicators; performing multi-objective parameter optimization and closed-loop feedback control processing on the quality evaluation indicators to obtain a quality control strategy based on construction process parameters.

[0006] In a second aspect, the present application provides a prestressed intelligent tensioning and grouting quality monitoring system, the prestressed intelligent tensioning and grouting quality monitoring system comprising:

[0007] A transformation module is used to perform nonlinear projection transformation and cross contrast enhancement processing on multi-view images collected at the construction site of prestressed components to obtain standard image data;

[0008] A matching module, used for performing regional cascade detection and dual time series matching processing on the standard image data to obtain positioning parameters and motion characteristics corresponding to the prestressed component;

[0009] A reconstruction module, used to perform sub-pixel edge reconstruction and multi-view displacement field analysis on the positioning parameters and motion features to obtain stress state data of the tensioning end;

[0010] An extraction module is used to perform adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain the flow state parameters of the grouting material;

[0011] A fusion module, used for performing spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain a quality evaluation index;

[0012] The control module is used to perform multi-objective parameter optimization and closed-loop feedback control processing on the quality evaluation indicators to obtain a quality control strategy based on construction process parameters.

[0013] In the technical solution provided by this application, the standardization and quality improvement of image data are achieved by performing nonlinear projection transformation and cross contrast enhancement processing on the multi-view images collected at the construction site of the prestressed component, providing a high-quality data basis for subsequent target detection and feature extraction; the positioning accuracy of key components of the prestressed component and the accuracy of motion feature extraction are improved by performing regional cascade detection and dual time series matching processing on the standard image data; the accurate measurement and evaluation of the stress state of the tensioning end are achieved by adopting sub-pixel edge reconstruction and multi-view displacement field analysis processing; the adaptive grid segmentation and flow field dynamic feature extraction processing are performed on the grouting area image to accurately capture the flow characteristics and distribution law of the grouting material; the effective integration of stress state data and flow state parameters is achieved through spatiotemporal feature mapping and weight adaptive fusion processing, and a comprehensive quality evaluation system is established; finally, through multi-objective parameter optimization and closed-loop feedback control processing, the precise regulation of construction process parameters is achieved, ensuring the stability and reliability of construction quality. The overall solution significantly improves the monitoring accuracy of the prestressed construction process through the deep fusion and intelligent analysis of multi-source data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of a method for monitoring quality of prestressed intelligent tensioning and grouting in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a prestressed intelligent tensioning and grouting quality monitoring system in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide a method and system for monitoring the quality of prestressed intelligent tensioning and grouting. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the prestressed intelligent tensioning and grouting quality monitoring method includes:

[0019] Step S101, performing nonlinear projection transformation and cross contrast enhancement processing on multi-view images collected at the prestressed component construction site to obtain standard image data;

[0020] Step S102, performing regional cascade detection and dual time series matching processing on the standard image data to obtain positioning parameters and motion characteristics corresponding to the prestressed component;

[0021] Step S103, performing sub-pixel edge reconstruction and multi-view displacement field analysis processing on the positioning parameters and motion features to obtain the stress state data of the tensioning end;

[0022] Step S104, performing adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain flow state parameters of the grouting material;

[0023] Step S105, performing spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain a quality assessment index;

[0024] Step S106: Perform multi-objective parameter optimization and closed-loop feedback control processing on the quality assessment indicators to obtain a quality control strategy based on the construction process parameters.

[0025] It is understandable that the execution subject of the present application may be a prestressed intelligent tensioning and grouting quality monitoring system, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0026] Specifically, these original images are transformed by nonlinear projection, which includes lens distortion correction, image noise removal by Gaussian noise filtering, and brightness normalization so that images at different viewing angles have similar brightness distribution. In the cross-contrast enhancement stage, the image is segmented by regional adaptive threshold, nonlinear grayscale mapping is performed based on the partition data, spatial feature points are extracted and the homography matrix is ​​calculated, and finally spatial reprojection is completed to obtain standardized image data.

[0027] Then, regional cascade detection is carried out. By constructing a multi-scale image pyramid, a sliding window is used to scan each scale layer to obtain candidate regions. The feature descriptors of these regions are extracted and filtered through classifier cascade to obtain preliminary target regions. Accurate target positioning parameters are obtained through bounding box regression and non-maximum suppression processing. In the dual temporal matching stage, the appearance features of the targets in adjacent frames are extracted, and the feature similarity is calculated to obtain the inter-frame matching pairs. At the same time, the motion state prediction is performed, and the predicted value is associated and optimized with the actual detection result to finally obtain the motion characteristics of the target. In the sub-pixel edge reconstruction link, the local grayscale interpolation of the target area is performed, the gradient intensity distribution is calculated, the edge point data set is obtained by sub-pixel coordinate positioning, and the initial contour is obtained by high-order curve fitting. The geometric feature extraction and curvature analysis of the contour are performed, and the refined target contour is obtained through multi-scale decomposition and edge refinement. In the multi-view displacement field analysis, the contour feature points under different perspectives are registered, the parallax is calculated based on the epipolar constraint to obtain the depth map, and the three-dimensional point cloud is reconstructed to obtain the spatial deformation variable. After temporal registration and vector field construction, the stress state data of the tension end is finally obtained.

[0028] For monitoring of grouting area, firstly, regional stratification and morphological processing are performed to obtain the target area mask, and threshold segmentation is performed based on grayscale changes to obtain the initial flow field area. Local grid units are formed by pixel density calculation and adaptive re-segmentation, and local flow direction characteristics are obtained by calculating grayscale gradients to construct flow field direction maps. The flow fields at adjacent moments are differentiated to obtain the change amount, and the velocity field data is obtained by time-space correlation. The vorticity distribution of the flow field is calculated and clustered, and the flow state parameters of the grouting material are finally obtained through boundary tracking and time series analysis. In the data fusion stage, the stress state data is decomposed into time series and periodic analysis, and the flow state parameters are spatially mapped and time-series reconstructed. The two types of data are mapped to a unified feature space and normalized. The core features are obtained by correlation analysis and principal component extraction, and the feature importance is obtained based on historical data comparison. Dynamic weight update and constraint optimization are performed, and finally quality evaluation indicators are formed through weighted combination and redundancy elimination. In the control strategy generation stage, the quality evaluation indicators are decomposed into multiple parameters and parameter constraints are determined, mapped to construction process parameters, and parameter conflicts are detected. Through discrete sampling and sensitivity analysis, the parameter control target is obtained. In closed-loop feedback control, the deviation between the control amount and the actual state is calculated, the deviation change trend is analyzed, the compensation amount is calculated and the stability is evaluated, and finally a quality control strategy is generated through feedback compensation and superposition fusion.

[0029] For example, when monitoring the tensioning end at a prestressed bridge construction site, the original images collected by three 4K industrial cameras were processed, and the standard deviation of image brightness was reduced from the original 87 to 12, ensuring the consistency of image quality. In the target detection stage, through multi-scale feature extraction, the positioning accuracy of the tensioning jack reached the sub-pixel level. The displacement field analysis showed that the end displacement was 2.47mm during the change of tension force from 0 to 200kN, which was close to the theoretical calculated value of 2.51mm. In the grouting process monitoring, through grid segmentation and flow field analysis, it was detected that there were tiny bubbles when the slurry flow rate was 0.15m / s, and the pressure was adjusted in time. The comprehensive evaluation index after data fusion shows that the construction quality is within the 95% confidence interval, and the closed-loop control controls the fluctuation range of the tensioning pressure within ±2%, ensuring the stability of the construction quality.

[0030] In the embodiment of the present application, the image data is standardized and improved in quality by performing nonlinear projection transformation and cross contrast enhancement processing on the multi-view images collected at the construction site of the prestressed component, providing a high-quality data basis for subsequent target detection and feature extraction; the positioning accuracy of the key components of the prestressed component and the accuracy of motion feature extraction are improved by performing regional cascade detection and dual time series matching processing on the standard image data; the sub-pixel edge reconstruction and multi-view displacement field analysis processing are adopted to achieve accurate measurement and evaluation of the stress state of the tensioning end; the grouting area image is adaptively meshed and the flow field dynamic feature extraction is processed to accurately capture the flow characteristics and distribution law of the grouting material; through spatiotemporal feature mapping and weight adaptive fusion processing, the stress state data and flow state parameters are effectively integrated, and a comprehensive quality assessment system is established; finally, through multi-objective parameter optimization and closed-loop feedback control processing, the construction process parameters are precisely regulated, ensuring the stability and reliability of the construction quality. The overall solution significantly improves the monitoring accuracy of the prestressed construction process through the deep fusion and intelligent analysis of multi-source data.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) Perform lens distortion correction on the multi-view images to obtain an initial corrected image, and perform Gaussian noise filtering on the initial corrected image to obtain a denoised image;

[0033] (2) Perform brightness normalization on the denoised image to obtain a grayscale balanced image, and perform local contrast calculation on the grayscale balanced image to obtain a contrast feature map;

[0034] (3) Performing regional adaptive threshold segmentation processing on the contrast feature map to obtain image partition data, and performing nonlinear grayscale mapping processing on the image partition data to obtain an enhanced image;

[0035] (4) Extracting feature points from the enhanced image to obtain a set of spatial feature points, and calculating the homography matrix of the set of spatial feature points to obtain image transformation parameters;

[0036] (5) The enhanced image is spatially reprojected based on the image transformation parameters to obtain a reprojected image, and the reprojected image is fused to obtain standard image data.

[0037] Specifically, at the construction site of prestressed components, image data of key parts such as prestressed tendons, tensioning ends, and grouting holes are collected from different angles through high-definition industrial cameras. First, lens distortion correction is performed. Lens distortion mainly includes two forms: radial distortion and tangential distortion. Radial distortion causes barrel or pillow-shaped deformation in the edge area of ​​the image, while tangential distortion causes asymmetric deformation of the image. Distortion correction obtains distortion parameters through camera calibration, remaps the coordinates of each pixel in the original image, and eliminates the distortion effect. Random noise still exists in the corrected image, and Gaussian filtering is required for noise reduction. Gaussian filtering is a linear smoothing filter that achieves noise suppression by calculating the weighted average of all pixels in the Gaussian kernel centered on the target pixel. The filter window size and standard deviation are key parameters. Usually a window size of 5×5 or 7×7 is used, and the standard deviation value is dynamically adjusted according to the noise intensity.

[0038] The denoised image is subjected to brightness normalization to adjust the brightness distribution of the image collected under different viewing angles and different lighting conditions to a uniform range. Brightness normalization first counts the grayscale histogram of the image, calculates the maximum, minimum and mean values, and then linearly maps the pixel values ​​to a standard range, usually the interval of 0-255. The normalized image is subjected to local contrast calculation, which is an important feature that describes the degree of brightness and darkness changes in the local area of ​​the image. The calculation process is to set a sliding window around each pixel point, count the maximum and minimum values ​​of the pixels in the window, obtain the contrast value of the point according to the contrast definition formula, and finally generate the contrast feature map of the entire image. The contrast feature map reflects the prominence of the target edge and texture in the image, and it is necessary to extract the effective target area through regional adaptive threshold segmentation. Adaptive threshold segmentation is different from global threshold segmentation. It dynamically calculates the local threshold based on the grayscale distribution characteristics in the neighborhood of the pixel point. In specific implementation, a sliding window is set, the mean and standard deviation of the pixels in the window are calculated, and the threshold of the point is obtained by combining the dynamic range of the image and the preset weight coefficient, and then binarization is performed. The segmented image area is further enhanced by nonlinear grayscale mapping, and the pixel values ​​are nonlinearly transformed using gamma correction to enhance the contrast and detail features of the target area.

[0039] Feature points are extracted from the enhanced image. Feature points are pixels with significant local features in the image, such as corner points, edge points, etc. The feature point extraction uses the FAST corner detection algorithm, which determines whether it is a corner point by comparing the grayscale values ​​in the circular neighborhood around the pixel point. The extracted feature points constitute a set of spatial feature points, and the correspondence between feature points between images from different perspectives needs to be calculated. The homography matrix describes the projection transformation relationship between the two image planes. The mapping relationship between feature point pairs is solved by the least squares method or the RANSAC algorithm, and a 3×3 homography matrix is ​​obtained as the image transformation parameter. Based on the calculated transformation parameters, the enhanced image is spatially reprojected. The reprojection process uniformly transforms images from different perspectives to the same reference plane to eliminate the geometric deformation caused by the perspective difference. The reprojected image also needs to be fused. The fusion adopts a weighted average strategy. The weight coefficient is dynamically allocated according to the image quality of the overlapping area, and finally generates standardized image data.

[0040] In an example of monitoring the tension end at a prestressed box girder construction site, 4K resolution images from three different viewing angles were first acquired. The radial distortion coefficient was obtained through camera calibration. =-0.2164, =0.1286, =-0.0382, tangential distortion coefficient =0.0021, =-0.0018, and the original image is distorted. A Gaussian filter with a window size of 7×7 and a standard deviation of σ=1.5 is used for noise reduction. Brightness normalization adjusts the average brightness of the three view images to 127, and the standard deviation is controlled within 45. A 15×15 sliding window is used for local contrast calculation. In the regional adaptive threshold segmentation, the local window size is set to 25×25, and the weight coefficient is 0.2. Through feature point matching and image fusion, standardized image data with a resolution of 3840×2160 is obtained.

[0041] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0042] (1) Performing regional hierarchical processing on the standard image data to obtain a multi-scale image pyramid, and performing sliding window scanning processing on each scale layer image in the multi-scale image pyramid to obtain a set of candidate regions;

[0043] (2) Perform feature map extraction on the candidate region set to obtain the region feature descriptor, and perform classifier cascade screening on the region feature descriptor to obtain the target region set;

[0044] (3) Perform bounding box regression on the target area set to obtain target detection box parameters, and perform non-maximum suppression on the target detection box parameters to obtain positioning parameters;

[0045] (4) Perform appearance feature extraction on the positioning parameters of adjacent frames to obtain the target appearance descriptor, and perform similarity calculation on the target appearance descriptor to obtain inter-frame matching pairs;

[0046] (5) Perform motion state prediction processing on the inter-frame matching pairs to obtain the target trajectory prediction value, and then perform correlation optimization processing on the target trajectory prediction value and the actual detection result to obtain the motion feature.

[0047] Specifically, the standardized image data is first subjected to regional hierarchical processing to construct a multi-scale image pyramid. An image pyramid is a sequence of images with decreasing resolutions, which is obtained by continuously downsampling the original image. During the downsampling process, the resolution is halved at each level, and Gaussian smoothing is performed to avoid aliasing effects. For the original image with 4K resolution, a 5-layer image pyramid is usually constructed, and the image size of each layer is 3840×2160, 1920×1080, 960×540, 480×270 and 240×135 pixels, respectively. A sliding window scan is performed on each layer of the pyramid image. The size of the sliding window is set according to the target feature scale. Typical window sizes include 64×64, 128×128 and 256×256 pixels, and the sliding step size is set to 1 / 4 of the window size. Through multi-scale scanning, a large number of candidate regions are obtained, which cover all possible locations in the image that may contain targets. Feature maps are extracted from the obtained candidate regions. Feature maps are numerical representations that describe the visual features of image regions. The extraction process includes multiple convolutional layers, each of which extracts features at different levels. The first layer extracts low-level features such as edges and colors, with a data dimension of 64 channels; the second layer extracts mid-level features such as textures and shapes, with the data dimension expanded to 128 channels; the third layer extracts semantic-level features, with a data dimension of 256 channels. The extracted features are compressed in size through a pooling operation to generate a fixed-dimensional regional feature descriptor. The feature descriptor is input into a cascade classifier for screening. The cascade classifier consists of multiple simple classifiers connected in series. Simple features are used in the early stage to quickly exclude obvious negative samples, and complex features are used in the later stage for fine classification, and finally the target region set is screened out.

[0048] The bounding box regression is performed on the screened target region set. The bounding box regression aims to refine the position and size of the target detection frame. The regression network inputs the feature descriptor and outputs the correction of the frame coordinates, including the offset of the center point coordinates and the scaling factor of the width and height. The corrected detection frame is subjected to non-maximum suppression, and the intersection-over-union ratio between the detection frames is calculated. When the intersection-over-union ratio exceeds the threshold (usually set to 0.5), the detection frame with the highest score is retained, and other overlapping detection frames are suppressed, and finally the accurate positioning parameters of each target are obtained. Based on the positioning parameters of adjacent frames in the video sequence, the appearance features of the target region are extracted as the appearance descriptor. The appearance descriptor includes color histogram features (256 dimensions for each of the three RGB channels), HOG features (histogram of oriented gradients, 2916 dimensions) and depth features (2048 dimensions). The cosine similarity between the target region descriptors of different frames is calculated, and the similarity threshold is set (usually 0.8). The targets with similarity exceeding the threshold are paired to form inter-frame matching pairs.

[0049] Kalman filtering is used to predict the motion state of the inter-frame matching pairs, and the state vector contains the target position, velocity and acceleration information. The target position of the next frame is predicted based on the historical trajectory data, and the predicted position is associated and optimized with the actual detection results. The Hungarian algorithm is used to solve the data association problem, and finally the motion characteristics of the target are obtained, including motion trajectory, velocity and acceleration information. In the actual prestressed component construction monitoring, the tracking example of the tension end shows that in the image sequence with a resolution of 3840×2160, a total of 12,800 candidate regions are generated through 5 layers of image pyramids and 3 sizes of sliding windows. After feature extraction and cascade classification, 56 target regions are screened out, and finally three key monitoring points are located through bounding box regression and non-maximum suppression. These monitoring points are tracked in a 30fps video sequence, the dimension of the apparent descriptor is 5220, and the average similarity of inter-frame matching is 0.92. The average position error predicted by the Kalman filter is 1.8 pixels, which realizes the fine tracking and monitoring of the tensioning process.

[0050] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0051] (1) Perform local grayscale interpolation processing on the target area indicated by the positioning parameters to obtain a high-resolution grayscale distribution map, and perform gradient operator calculation processing on the high-resolution grayscale distribution map to obtain edge intensity distribution data;

[0052] (2) Perform sub-pixel coordinate positioning processing on the edge intensity distribution data to obtain an edge point data set, and perform high-order curve fitting processing on the edge point data set to obtain an initial contour curve;

[0053] (3) Perform geometric feature extraction on the initial contour curve to obtain a set of key points, and perform curvature analysis on the key point set to obtain feature description data;

[0054] (4) Perform multi-scale decomposition processing on the feature description data to obtain a hierarchical feature map, and perform edge refinement processing on the hierarchical feature map to obtain target contour data;

[0055] (5) Perform feature point registration processing on the target contour data to obtain corresponding point sets under different viewing angles, and perform parallax calculation processing on the corresponding point sets based on epipolar constraints to obtain depth mapping data;

[0056] (6) Performing 3D point cloud reconstruction on the depth mapping data to obtain the target space deformation variable, and performing time series registration on the target space deformation variable to obtain a displacement change sequence;

[0057] (7) Performing vector field construction processing on the displacement change sequence to obtain a displacement vector field, and performing spatial filtering processing on the displacement vector field to obtain a smooth displacement field;

[0058] (8) The smooth displacement field is converted into a stress-strain relationship to obtain the target force distribution, and the target force distribution is filtered in the time domain to obtain the force state data of the tensioning end.

[0059] Specifically, in the process of prestressed component construction monitoring, the area determined by the target positioning parameters is first subjected to local grayscale interpolation processing, and the bicubic interpolation algorithm is used to increase the resolution of the original image by 4 times. The grayscale value influence of the surrounding 16 pixels is considered during the interpolation process to generate a more delicate grayscale distribution map. The Sobel operator is used to calculate the gradient of the high-resolution grayscale distribution map, and the gradient value is calculated in the horizontal and vertical directions respectively. The edge strength and direction information are obtained comprehensively to form the edge strength distribution data. For the edge strength distribution data, the sub-pixel edge positioning algorithm is used for fine positioning. The algorithm performs grayscale value fitting near the edge point, and solves the precise coordinates of the grayscale gradient maximum point through the interpolation polynomial, and the positioning accuracy can reach 0.1 pixel. The acquired edge point data set is then fitted using a cubic spline curve. The spline curve maintains the continuity of the second-order derivative at the node, ensuring the smooth transition of the contour and forming the initial contour curve.

[0060] Extract geometric features from the initial contour curve, including key feature points such as turning points, inflection points and extreme points of the curve. By calculating the first-order derivative and second-order derivative of the curve at each point, identify the locations where the curvature changes significantly, and use these locations as key points. Perform curvature analysis on the key point set, calculate the curvature value and curvature change rate of each point, and generate feature description data that describes the local geometric characteristics of the contour. The feature description data is decomposed at multiple scales through wavelet transform, and a 4-layer wavelet decomposition tree is constructed, each layer containing approximation coefficients and detail coefficients. The decomposition process extracts contour features at different scales to form a hierarchical feature map. The feature map is subjected to edge refinement, and a refinement algorithm is used to reduce the edge width to a single pixel while maintaining the connectivity and shape characteristics of the edge to obtain accurate target contour data.

[0061] The feature points in the target contour data are matched across perspectives using the SIFT algorithm. The algorithm extracts the local gradient direction histogram of the feature points as descriptors for feature matching. The epipolar geometric constraints are considered during matching, and the epipolar lines of the corresponding points are calculated using the basic matrix to ensure that the matching points meet the epipolar constraints. The parallax, that is, the relative displacement of the corresponding points in the image plane, is calculated for the matching point pairs that meet the constraints to generate depth mapping data. The depth mapping data is used to reconstruct a three-dimensional point cloud using the triangulation principle, and the three-dimensional coordinates of the spatial points are calculated in combination with the known camera parameters. The point cloud data is processed through registration and denoising to remove outliers and obtain the spatial deformation of the target. The deformation data of the continuous time series is processed using the ICP registration algorithm, and the rigid transformation matrix between the point clouds at adjacent moments is calculated to obtain a displacement sequence that changes over time.

[0062] The displacement change sequence constructs a continuous displacement vector field through interpolation algorithm to describe the motion state of each point on the target surface. The vector field is smoothed by Gaussian filtering to filter out noise and mutations, and a smooth displacement field distribution is obtained. Based on the material mechanics model, the displacement field is converted into stress-strain distribution, and the stress state of the target surface is calculated. The stress data is subjected to time-domain Kalman filtering to eliminate instantaneous fluctuations and obtain stable stress state data. In the monitoring of the tensioning process of a prestressed beam end, the original image resolution is 3840×2160 pixels, which reaches 15360×8640 pixels after 4 times interpolation. Edge detection obtains about 8500 sub-pixel edge points, and curve fitting uses a cubic spline curve with 50 control points. Feature point matching finds a total of 326 pairs of corresponding points between three perspectives, and the reconstructed point cloud contains about 12,000 spatial points. During the 200-second tensioning process, a 600-frame image sequence was collected, and the calculated maximum displacement was 2.47 mm. The corresponding tension change was 0-200 kN. The stress distribution showed that the maximum stress at the tensioning end was 785 MPa, and the measurement accuracy was better than 0.1 mm.

[0063] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0064] (1) Performing regional stratification processing on the pre-collected grouting area image to obtain image partition data, and performing morphological processing on the image partition data to obtain the target area mask;

[0065] (2) Perform threshold segmentation processing on the target area mask according to the regional grayscale change to obtain the regional grid structure, and perform boundary constraint processing on the regional grid structure to obtain the initial flow field area;

[0066] (3) Calculate the pixel density of the initial flow field area to obtain a regional density distribution map, and adaptively re-segment the regional density distribution map to obtain local grid units;

[0067] (4) Perform grayscale gradient calculation on the local grid cells to obtain local flow direction features, and perform direction field construction on the local flow direction features to obtain a flow field direction map;

[0068] (5) Perform differential processing on the flow field direction diagrams at adjacent moments to obtain the flow field variation, and perform spatiotemporal correlation processing on the flow field variation to obtain velocity field data;

[0069] (6) Perform curl calculation on the velocity field data to obtain the flow field vorticity distribution, and perform cluster analysis on the flow field vorticity distribution to obtain the flow characteristic area;

[0070] (7) Perform boundary tracking processing on the flow feature area to obtain the feature area evolution sequence, and perform time series analysis processing on the feature area evolution sequence to obtain flow pattern data;

[0071] (8) Extract characteristic quantities from the flow pattern data to obtain flow state quantities, and perform time domain statistical processing on the flow state quantities to obtain flow state parameters of the grouting material.

[0072] Specifically, the image data of the grouting area is processed. Through the multi-scale decomposition method, the original image is divided into different resolution levels, and the image size of each layer is half of the previous layer. This layered processing can not only capture the detailed features at different scales, but also significantly reduce the computational complexity of subsequent processing. Each layer of the image is first Gaussian smoothed to eliminate the influence of noise, and then downsampled to avoid aliasing effects. This layered process can be expressed as:

[0073] ;

[0074] in, represents the n-th level pyramid image, is the Gaussian smoothing kernel, is the smoothing parameter, x represents the horizontal coordinate on the image, and y represents the vertical coordinate on the image.

[0075] In order to accurately extract the grouting area, the layered image is morphologically processed. First, the edge noise and burrs are removed by corrosion operation, and then the small holes in the area are filled by dilation operation, and finally the complete target area mask is obtained. The mathematical expression of morphological operation is:

[0076] ;

[0077] in, is the mask result, and are the erosion and dilation operators respectively, is the input image, Represents a compound operation.

[0078] After obtaining the target area mask, fine area division is required. Adaptive threshold segmentation method is used to dynamically calculate the threshold according to the grayscale distribution characteristics of the local area. This method is more suitable for processing images with uneven lighting than global threshold. For each local area, the mean and standard deviation are calculated, and the threshold is determined by combining the adjustment parameters:

[0079] ;

[0080] in, is the local threshold, is the local mean, is the local standard deviation, and To adjust the parameters.

[0081] The segmented area needs to be subjected to density analysis to calculate the distribution characteristics of pixels in each local area. The density distribution map is constructed by using kernel function weighting and considering the spatial correlation of pixels:

[0082] ;

[0083] in, is the density value, is the weighted kernel function, is the feature image, and W is the calculation window. For each grid cell, the grayscale gradient is calculated to obtain the local flow direction feature. The direction of local fluid movement is determined by analyzing the horizontal and vertical gradients:

[0084] ;

[0085] in, is the flow angle, and They are the horizontal and vertical gradients respectively.

[0086] By comparing the flow field directional diagrams at adjacent moments, the change characteristics of the flow field can be obtained. The directional fields at two moments are differentially calculated to obtain the change in the flow field, and then combined with the time interval to establish the velocity field distribution:

[0087] ;

[0088] in, is the velocity vector, is the direction change, is a spatial scale factor. This calculation method takes into account the spatial and temporal characteristics of the flow field. The velocity field data can be used to analyze the rotation characteristics of the flow field. By calculating the curl of the velocity field, the vortex area can be identified. The vorticity represents the local rotation intensity of the fluid and is of great significance for the evaluation of grouting quality:

[0089] ;

[0090] in, is the vorticity value, and are the horizontal and vertical components of the velocity field, respectively.

[0091] Dynamically track the flow feature area and record the changes in its shape and position over time. This change reflects the flow characteristics of the grouting material and can be used to evaluate the uniformity and stability of the grouting process:

[0092] ;

[0093] in, For evolutionary characteristics, is the weight coefficient of different features, is the time series characteristic function, and N is the total number of characteristics. Finally, the flow characteristics are statistically analyzed to extract key state parameters. These parameters include velocity distribution, vorticity intensity, flow uniformity, etc. The overall flow state of the grouting material is obtained through time domain statistics:

[0094] ;

[0095] in, is the state parameter, R(t) is the time domain weight function, is the total flow duration;

[0096] Take a prestressed box girder grouting project as an example: a 4K resolution camera was used to collect images of the grouting process, and the images were divided into different scale levels through a 4-layer pyramid decomposition. The adaptive threshold segmentation used a local calculation window of 15×15 pixels, and the adjustment parameter k was set to 0.15 to effectively separate the grouting area. The density analysis used a Gaussian weight kernel of 11×11, which accurately reflected the distribution characteristics of the grouting material. Real-time monitoring data showed that the flow velocity at the beginning of grouting was 0.25m / s, and as the slurry gradually filled, the flow velocity dropped to 0.15m / s. A vortex peak of 0.8rad / s was detected within 5cm around the grouting hole. The vortex phenomenon was successfully eliminated by adjusting the grouting pressure from 0.6MPa to 0.4MPa. The entire grouting process lasted 280 seconds, and the flow uniformity index remained above 95%, meeting the engineering quality requirements.

[0097] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0098] (1) Decompose the stress state data into time series to obtain the mechanical characteristic sequence, and perform periodic analysis on the mechanical characteristic sequence to obtain the stress change pattern;

[0099] (2) Perform spatial distribution mapping on the flow state parameters to obtain the flow characteristic matrix, and then perform time series reconstruction on the flow characteristic matrix to obtain the flow change pattern;

[0100] (3) Perform feature space mapping on the force change pattern and the flow change pattern to obtain a unified feature expression, and perform dimension normalization on the unified feature expression to obtain a standardized feature set;

[0101] (4) Perform correlation analysis on the standardized feature set to obtain a feature correlation matrix, and perform principal component extraction on the feature correlation matrix to obtain a core feature vector;

[0102] (5) Perform historical data comparison processing on the core feature vector to obtain the feature importance distribution, and then divide the feature importance distribution into intervals to obtain the initial weight value;

[0103] (6) Dynamically update the initial weight value based on the state change to obtain the real-time weight coefficient, and perform constraint optimization on the real-time weight coefficient to obtain the balanced weight value;

[0104] (7) Performing weighted combination processing on the standardized feature set based on the balanced weight value to obtain fused feature data, and performing redundancy elimination processing on the fused feature data to obtain a reduced feature set;

[0105] (8) Perform multi-level quantitative processing on the reduced feature set to obtain a hierarchical index system, and then normalize the hierarchical index system to obtain a quality assessment index.

[0106] Specifically, the stress state data on the prestressed tendons are decomposed in time series. This process decomposes the complex stress time series data into three basic components: the overall trend term (reflecting the long-term change trend of the force value), the periodic change term (reflecting the cyclical force value fluctuation) and the random fluctuation term (reflecting the random changes caused by various interference factors). By performing periodic analysis on these decomposed data, the main cycles and laws of the tension force changes can be identified, thereby obtaining the stress change pattern. At the same time, the flow state parameters in the grouting process are spatially distributed. These parameters include physical quantities such as the flow velocity, pressure distribution, and flow direction of the grouting material. By establishing a spatial mapping relationship, these discrete monitoring data are converted into continuous characteristic field distributions to form a flow characteristic matrix. This matrix not only contains information in the spatial dimension, but also needs to be reconstructed in the time dimension to reflect the dynamic change characteristics in the grouting process, and finally form a complete flow change pattern.

[0107] In order to achieve comprehensive analysis of force data and flow data, it is necessary to unify the data of these two different physical quantities into the same feature space. This unification is achieved through feature space mapping, which converts data of different dimensions and scales into comparable feature spaces. Subsequently, the mapped data is dimensionally normalized to eliminate the influence of dimensions, so that different features are comparable and form a standardized feature set. The standardized feature set is subjected to correlation analysis to examine the degree of correlation between different features. This analysis constructs a feature correlation matrix by calculating the correlation coefficients between features. Then, this matrix is ​​subjected to principal component analysis to extract the most representative core feature vectors, reducing the data dimension while retaining key information.

[0108] Based on historical construction data, the extracted core feature vectors are analyzed for importance. By comparing the impact of each feature on construction quality in historical cases, the feature importance distribution is obtained. This distribution is divided into intervals to form the initial feature weight value, which provides a basis for subsequent feature fusion. During the construction process, the weight value needs to be dynamically adjusted according to the real-time status. This adjustment is based on the deviation between the current state and the target state. The optimal weight update amount is calculated through the optimization algorithm, and the constraints of the weight change are considered to finally obtain a balanced weight value.

[0109] The obtained balance weight values ​​are used to perform weighted combination of the standardized feature set to achieve effective feature fusion. In the fusion process, it is also necessary to remove redundant information, extract the most representative feature combination, and form a streamlined feature set. Finally, a multi-level quantitative analysis is performed on the streamlined feature set to establish a hierarchical quality evaluation system. This system rationally organizes feature indicators at different levels and finally forms a unified quality evaluation indicator through normalization.

[0110] In a prestressed box girder project, the monitoring data of a tension control point was analyzed: the time series decomposition of the force data showed that the overall trend showed a characteristic of first rapid rise and then stability, with a main period of 120 seconds and an amplitude change of ±5kN. The flow state parameters were centrally symmetrically distributed in space, and the grouting speed gradually decreased from the grouting point to the outside, from 0.25m / s to 0.15m / s. The data after feature space mapping was standardized to the [-1,1] interval, and the correlation analysis showed that the correlation coefficient between tension and grouting speed was -0.72. By comparing historical data, the tension feature weight was determined to be 0.6 and the grouting feature weight was 0.4. In practical applications, when a 10% fluctuation in tension was detected, the weight was dynamically adjusted by 0.05, so that the comprehensive quality assessment index was stabilized above 0.92.

[0111] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0112] (1) Perform multi-parameter decomposition processing on the quality assessment indicators to obtain indicator sub-item data, and perform target interval definition processing on the indicator sub-item data to obtain parameter constraint conditions;

[0113] (2) Mapping the parameter constraints to the construction process to obtain a set of process parameters, and performing conflict detection on the set of process parameters to obtain a feasible parameter domain;

[0114] (3) Discrete sampling is performed on the feasible parameter domain to obtain a candidate parameter group, and parameter sensitivity analysis is performed on the candidate parameter group to obtain an impact factor matrix;

[0115] (4) Calculate the target value of the influencing factor matrix to obtain the parameter control target, and perform multi-target conversion on the parameter control target to obtain the control quantity sequence;

[0116] (5) Calculate the deviation between the control quantity sequence and the actual execution state to obtain the state deviation value, and perform trend analysis on the state deviation value to obtain the deviation change characteristics;

[0117] (6) Calculate the compensation amount for the deviation change characteristics to obtain a correction parameter value, and perform stability evaluation on the correction parameter value to obtain an adjustment gain coefficient;

[0118] (7) Feedback compensation is performed on the adjustment gain coefficient to obtain a dynamic correction value, and the dynamic correction value is superimposed and fused to obtain a control compensation value;

[0119] (8) The control compensation value is parameter reconstructed to obtain an updated control quantity, and the updated control quantity is converted into a construction process to obtain a quality control strategy.

[0120] Specifically, in prestressed structures, these indicators include uniformity of tension force (reflecting the consistency of force distribution), density of grouting (reflecting the integrity of slurry filling), stability of construction (reflecting the degree of fluctuation of process parameters), etc. For each sub-item, it is necessary to determine its target range based on specification requirements and engineering experience, and clarify the upper and lower limits. For example, the deviation of tension uniformity shall not exceed ±5% of the design value, and the grouting density requirement shall not be less than 95%. This decomposition and constraint can be expressed as:

[0121] ;

[0122] in, As a comprehensive quality indicator, is the item weight, For sub-item indicators, and are constraint boundaries, and N is the number of sub-item indicators. These quality requirements are converted into specific construction process parameters. This process involves the relationship between multiple construction parameters. For example, the tensioning speed affects the uniformity of force values, and the grouting pressure determines the filling degree of the slurry. At the same time, possible conflicts between parameters should be considered. For example, a faster tensioning speed is conducive to improving construction efficiency, but may reduce the accuracy of force value control. This process mapping relationship is expressed as:

[0123] ;

[0124] in, is the process parameter function, is the mapping coefficient, is the process parameter, is the correction term, and m is the number of process parameters. After determining the feasible domain of the process parameters, discrete sampling and sensitivity analysis are required. Discrete sampling is to select typical parameter combinations within the feasible domain of the parameters, and these combinations need to cover all areas of the parameter space. For example, for tensioning speeds in the range of 3-8mm / min, a sample point is taken every 0.5mm / min; for grouting pressures in the range of 0.3-0.7MPa, a point is taken every 0.05MPa. Perform sensitivity analysis on these parameter combinations to evaluate the degree of influence of each parameter change on the final quality. This sensitivity analysis can be expressed as:

[0125] ;

[0126] in, is the sensitivity coefficient, For the output response, is the input parameter.

[0127] According to the influencing factor matrix, the control target value of each parameter is calculated. This process requires balancing multiple objectives, such as ensuring the uniformity of tension and the density of grouting. Each objective has its weight, reflecting its relative importance. Through multi-objective optimization, these objectives are converted into specific control quantity sequences to guide the parameter adjustment of the construction process. During the construction process, the deviation between the actual execution status and the control target is continuously monitored. This deviation includes amplitude deviation (reflecting the gap between the current value and the target value) and change trend (reflecting the direction and speed of parameter change). By analyzing the time series characteristics of the deviation, the change trend of the parameter can be predicted and potential quality risks can be discovered in time.

[0128] Calculate the required compensation for the deviation found. The compensation calculation needs to consider the size of the deviation, the speed of change and the cumulative effect. For example, when the tension shows a trend of continuous increase, the compensation amount must not only consider the current deviation, but also estimate possible changes in the future. The calculated compensation amount also needs to be evaluated for stability to ensure that the compensation action does not cause drastic fluctuations in the system. Feedback adjustment is performed on the compensation results, and this process adopts a dynamic gain control strategy. The gain coefficient determines the strength of the feedback compensation. A larger gain can quickly eliminate the deviation, but may cause overshoot; a smaller gain has a slower response, but can maintain the stability of the system. The calculation of the compensation value needs to consider the cumulative effect of multiple moments.

[0129] Finally, all compensation and adjustment results are integrated to form a complete control strategy. This strategy needs to be converted into specific construction process parameters, including the adjustment value of tensioning speed, the correction amount of grouting pressure, etc. These parameter adjustments must meet the requirements of process specifications to ensure the operability of construction.

[0130] In the construction of a prestressed box girder, the above control strategy was used to achieve precise control of construction quality. When a 6% fluctuation in tension was detected, sensitivity analysis showed that this was mainly caused by changes in tension speed (sensitivity coefficient 0.85). The control system immediately calculated the compensation amount, adjusted the tension speed from 6mm / min to 4.5mm / min, and used a gain coefficient of 0.8 for smoothing. After about 30 seconds of adjustment, the tension fluctuation was reduced to 2.8%. At the same time, the density was increased by 8 percentage points by adjusting the grouting rate from 0.2m / s to 0.15m / s during the grouting process. During the entire construction process, the comprehensive quality index remained above 0.92, proving the effectiveness of the control strategy.

[0131] The above describes the prestressed intelligent tensioning and grouting quality monitoring method in the embodiment of the present application. The following describes the prestressed intelligent tensioning and grouting quality monitoring system in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the prestressed intelligent tensioning and grouting quality monitoring system includes:

[0132] The transformation module 201 is used to perform nonlinear projection transformation and cross contrast enhancement processing on the multi-view images collected at the prestressed component construction site to obtain standard image data;

[0133] A matching module 202 is used to perform regional cascade detection and dual time series matching processing on the standard image data to obtain positioning parameters and motion characteristics corresponding to the prestressed component;

[0134] A reconstruction module 203 is used to perform sub-pixel edge reconstruction and multi-view displacement field analysis on the positioning parameters and motion features to obtain stress state data of the tensioning end;

[0135] The extraction module 204 is used to perform adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain the flow state parameters of the grouting material;

[0136] A fusion module 205 is used to perform spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain a quality evaluation index;

[0137] The control module 206 is used to perform multi-objective parameter optimization and closed-loop feedback control processing on the quality assessment index to obtain a quality control strategy based on construction process parameters.

[0138] Through the cooperation of the above components, the image data is standardized and improved in quality by performing nonlinear projection transformation and cross contrast enhancement on the multi-view images collected at the construction site of the prestressed component, providing a high-quality data basis for subsequent target detection and feature extraction; the positioning accuracy of the key components of the prestressed component and the accuracy of motion feature extraction are improved by performing regional cascade detection and dual time series matching on the standard image data; the sub-pixel edge reconstruction and multi-view displacement field analysis are used to achieve accurate measurement and evaluation of the stress state of the tensioning end; the grouting area image is adaptively meshed and the flow field dynamic feature extraction is performed to accurately capture the flow characteristics and distribution law of the grouting material; the effective integration of stress state data and flow state parameters is achieved through spatiotemporal feature mapping and weight adaptive fusion processing, and a comprehensive quality assessment system is established; finally, the precise regulation of construction process parameters is achieved through multi-objective parameter optimization and closed-loop feedback control processing, ensuring the stability and reliability of construction quality. The overall solution significantly improves the monitoring accuracy of the prestressed construction process through the deep fusion and intelligent analysis of multi-source data.

[0139] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0140] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A prestressed intelligent tensioning and grouting quality monitoring method, characterized in that: The prestressed intelligent tensioning and grouting quality monitoring method comprises: The multi-view images collected at the construction site of prestressed components are processed by nonlinear projection transformation and cross-contrast enhancement to obtain standard image data; The standard image data is subjected to regional cascade detection and dual time series matching processing to obtain positioning parameters and motion features corresponding to the prestressed component, including: performing regional layering processing on the standard image data to obtain a multi-scale image pyramid, and performing sliding window scanning processing on each scale layer image in the multi-scale image pyramid to obtain a candidate region set; performing feature map extraction processing on the candidate region set to obtain a regional feature descriptor, and performing classifier cascade screening processing on the regional feature descriptor to obtain a target region set; performing bounding box regression processing on the target region set to obtain a target detection frame parameter, and performing non-maximum suppression processing on the target detection frame parameter to obtain the positioning parameter; performing appearance feature extraction processing on the positioning parameters of adjacent frames to obtain a target appearance descriptor, and performing similarity calculation processing on the target appearance descriptor to obtain an inter-frame matching pair; performing motion state prediction processing on the inter-frame matching pair to obtain a target trajectory prediction value, and performing association optimization processing on the target trajectory prediction value and the actual detection result to obtain the motion feature; Sub-pixel edge reconstruction and multi-view displacement field analysis are performed on the positioning parameters and motion features to obtain the force state data of the tensioning end, including: local grayscale interpolation processing is performed on the target area indicated by the positioning parameters to obtain a high-resolution grayscale distribution map, and gradient operator calculation processing is performed on the high-resolution grayscale distribution map to obtain edge intensity distribution data; sub-pixel coordinate positioning processing is performed on the edge intensity distribution data to obtain an edge point data set, and high-order curve fitting processing is performed on the edge point data set to obtain an initial contour curve; geometric feature extraction processing is performed on the initial contour curve to obtain a key point set, and curvature analysis processing is performed on the key point set to obtain feature description data; multi-scale decomposition processing is performed on the feature description data to obtain a hierarchical feature map, The hierarchical feature map is subjected to edge refinement processing to obtain target contour data; the target contour data is subjected to feature point registration processing to obtain corresponding point sets under different viewing angles, and the corresponding point sets are subjected to parallax calculation processing based on epipolar constraints to obtain depth mapping data; the depth mapping data is subjected to three-dimensional point cloud reconstruction processing to obtain target spatial deformation variables, and the target spatial deformation variables are subjected to time series registration processing to obtain a displacement change sequence; the displacement change sequence is subjected to vector field construction processing to obtain a displacement vector field, and the displacement vector field is subjected to spatial filtering processing to obtain a smooth displacement field; the smooth displacement field is subjected to stress-strain relationship conversion processing to obtain target force distribution, and the target force distribution is subjected to time domain filtering processing to obtain force state data of the tensioning end; The method performs adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain the flow state parameters of the grouting material, including: performing regional stratification processing on the pre-collected grouting area image to obtain image partition data, and performing morphological processing on the image partition data to obtain a target area mask; performing threshold segmentation processing on the target area mask according to regional grayscale changes to obtain a regional grid structure, and performing boundary constraint processing on the regional grid structure to obtain an initial flow field area; performing pixel density calculation processing on the initial flow field area to obtain a regional density distribution map, and performing adaptive re-segmentation processing on the regional density distribution map to obtain a local grid unit; performing grayscale gradient calculation processing on the local grid unit to obtain a local flow direction feature. Characteristic, and perform directional field construction processing on the local flow direction characteristics to obtain a flow field direction diagram; perform differential processing on the flow field direction diagrams at adjacent moments to obtain a flow field change, and perform time-space correlation processing on the flow field change to obtain velocity field data; perform curl calculation processing on the velocity field data to obtain a flow field vortex distribution, and perform cluster analysis processing on the flow field vortex distribution to obtain a flow feature area; perform boundary tracking processing on the flow feature area to obtain a feature area evolution sequence, and perform time series analysis processing on the feature area evolution sequence to obtain flow pattern data; perform feature extraction processing on the flow pattern data to obtain a flow state quantity, and perform time domain statistical processing on the flow state quantity to obtain the flow state parameters of the grouting material; The stress state data and flow state parameters are subjected to spatiotemporal feature mapping and weight adaptive fusion processing to obtain quality assessment indicators, including: performing time series decomposition processing on the stress state data to obtain a mechanical feature sequence, and performing periodic analysis processing on the mechanical feature sequence to obtain a stress change pattern; performing spatial distribution mapping processing on the flow state parameters to obtain a flow feature matrix, and performing time series reconstruction processing on the flow feature matrix to obtain a flow change pattern; performing feature space mapping processing on the force change pattern and the flow change pattern to obtain a unified feature expression, and performing dimensional normalization processing on the unified feature expression to obtain a standardized feature set; performing correlation analysis processing on the standardized feature set to obtain a feature association matrix, and The principal component extraction process is performed on the feature association matrix to obtain a core feature vector; the core feature vector is compared with historical data to obtain a feature importance distribution, and the feature importance distribution is divided into intervals to obtain an initial weight value; the initial weight value is dynamically updated based on a state change to obtain a real-time weight coefficient, and the real-time weight coefficient is constrained to optimize to obtain a balanced weight value; the standardized feature set is weighted combined based on the balanced weight value to obtain fused feature data, and the fused feature data is redundancy eliminated to obtain a reduced feature set; the reduced feature set is quantified at multiple levels to obtain a hierarchical index system, and the hierarchical index system is normalized to obtain the quality assessment index; The quality assessment index is subjected to multi-objective parameter optimization and closed-loop feedback control processing to obtain a quality control strategy based on construction process parameters, including: performing multi-parameter decomposition processing on the quality assessment index to obtain index sub-item data, and performing target interval definition processing on the index sub-item data to obtain parameter constraints; performing construction process mapping processing on the parameter constraints to obtain a process parameter set, and performing conflict detection processing on the process parameter set to obtain a feasible parameter domain; performing discrete sampling processing on the feasible parameter domain to obtain a candidate parameter group, and performing parameter sensitivity analysis processing on the candidate parameter group to obtain an influencing factor matrix; performing target value calculation processing on the influencing factor matrix to obtain a parameter control target, and The parameter control target is subjected to multi-target conversion processing to obtain a control quantity sequence; the control quantity sequence and the actual execution state are subjected to deviation calculation processing to obtain a state deviation value, and the state deviation value is subjected to trend analysis processing to obtain a deviation change characteristic; the deviation change characteristic is subjected to compensation calculation processing to obtain a correction parameter value, and the correction parameter value is subjected to stability evaluation processing to obtain an adjustment gain coefficient; the adjustment gain coefficient is subjected to feedback compensation processing to obtain a dynamic correction amount, and the dynamic correction amount is subjected to superposition fusion processing to obtain a control compensation value; the control compensation value is subjected to parameter reconstruction processing to obtain an updated control amount, and the updated control amount is subjected to construction process conversion processing to obtain the quality control strategy.

2. The prestressed intelligent tensioning and grouting quality monitoring method according to claim 1 is characterized in that: The multi-view images collected at the prestressed component construction site are subjected to nonlinear projection transformation and cross-contrast enhancement processing to obtain standard image data, including: Performing lens distortion correction processing on the multi-view image to obtain an initial corrected image, and performing Gaussian noise filtering processing on the initial corrected image to obtain a denoised image; Performing brightness normalization processing on the denoised image to obtain a grayscale balanced image, and performing local contrast calculation processing on the grayscale balanced image to obtain a contrast feature map; Performing a region adaptive threshold segmentation process on the contrast feature map to obtain image partition data, and performing a nonlinear grayscale mapping process on the image partition data to obtain an enhanced image; Performing feature point extraction processing on the enhanced image to obtain a spatial feature point set, and performing homography matrix calculation processing on the spatial feature point set to obtain image transformation parameters; The enhanced image is spatially reprojected based on the image transformation parameters to obtain a reprojected image, and the reprojected image is image fusion-processed to obtain the standard image data.

3. A prestressed intelligent tensioning and grouting quality monitoring system, used to implement the prestressed intelligent tensioning and grouting quality monitoring method as described in any one of claims 1-2, characterized in that: The prestressed intelligent tensioning and grouting quality monitoring system includes: A transformation module is used to perform nonlinear projection transformation and cross contrast enhancement processing on multi-view images collected at the construction site of prestressed components to obtain standard image data; A matching module is used to perform regional cascade detection and dual time series matching processing on the standard image data to obtain positioning parameters and motion features corresponding to the prestressed component, including: performing regional layering processing on the standard image data to obtain a multi-scale image pyramid, and performing sliding window scanning processing on each scale layer image in the multi-scale image pyramid to obtain a candidate region set; performing feature map extraction processing on the candidate region set to obtain a regional feature descriptor, and performing classifier cascade screening processing on the regional feature descriptor to obtain a target region set; performing bounding box regression processing on the target region set to obtain target detection frame parameters, and performing non-maximum suppression processing on the target detection frame parameters to obtain the positioning parameters; performing appearance feature extraction processing on the positioning parameters of adjacent frames to obtain a target appearance descriptor, and performing similarity calculation processing on the target appearance descriptor to obtain an inter-frame matching pair; performing motion state prediction processing on the inter-frame matching pair to obtain a target trajectory prediction value, and performing association optimization processing on the target trajectory prediction value and the actual detection result to obtain the motion feature; The reconstruction module is used to perform sub-pixel edge reconstruction and multi-view displacement field analysis on the positioning parameters and motion features to obtain the force state data of the tensioning end, including: performing local grayscale interpolation processing on the target area indicated by the positioning parameters to obtain a high-resolution grayscale distribution map, and performing gradient operator calculation processing on the high-resolution grayscale distribution map to obtain edge intensity distribution data; performing sub-pixel coordinate positioning processing on the edge intensity distribution data to obtain an edge point data set, and performing high-order curve fitting processing on the edge point data set to obtain an initial contour curve; performing geometric feature extraction processing on the initial contour curve to obtain a key point set, and performing curvature analysis processing on the key point set to obtain feature description data; performing multi-scale decomposition processing on the feature description data to obtain hierarchical features A map is obtained by performing edge refinement processing on the hierarchical feature map to obtain target contour data; feature point registration processing is performed on the target contour data to obtain corresponding point sets under different viewing angles, and parallax calculation processing is performed on the corresponding point sets based on polar constraints to obtain depth mapping data; three-dimensional point cloud reconstruction processing is performed on the depth mapping data to obtain target space deformation variables, and time series registration processing is performed on the target space deformation variables to obtain a displacement change sequence; vector field construction processing is performed on the displacement change sequence to obtain a displacement vector field, and spatial filtering processing is performed on the displacement vector field to obtain a smooth displacement field; stress-strain relationship conversion processing is performed on the smooth displacement field to obtain target force distribution, and time domain filtering processing is performed on the target force distribution to obtain force state data of the tensioning end; The extraction module is used to perform adaptive grid segmentation and flow field dynamic feature extraction processing on the pre-collected grouting area image to obtain the flow state parameters of the grouting material, including: performing regional stratification processing on the pre-collected grouting area image to obtain image partition data, and performing morphological processing on the image partition data to obtain the target area mask; performing threshold segmentation processing on the target area mask according to the regional grayscale change to obtain the regional grid structure, and performing boundary constraint processing on the regional grid structure to obtain the initial flow field area; performing pixel density calculation processing on the initial flow field area to obtain a regional density distribution map, and performing adaptive re-segmentation processing on the regional density distribution map to obtain a local grid unit; performing grayscale gradient calculation processing on the local grid unit to obtain a local The flow direction characteristics of the local flow direction characteristics are processed by direction field construction to obtain a flow field direction diagram; the flow field direction diagrams at adjacent moments are processed by difference to obtain a flow field change, and the flow field change is processed by time-space correlation to obtain velocity field data; the velocity field data is processed by curl calculation to obtain a flow field vortex distribution, and the flow field vortex distribution is processed by cluster analysis to obtain a flow feature area; the flow feature area is processed by boundary tracking to obtain a feature area evolution sequence, and the feature area evolution sequence is processed by time series analysis to obtain flow pattern data; the flow pattern data is processed by feature extraction to obtain a flow state quantity, and the flow state quantity is processed by time domain statistics to obtain the flow state parameters of the grouting material; The fusion module is used to perform spatiotemporal feature mapping and weight adaptive fusion processing on the stress state data and flow state parameters to obtain quality evaluation indicators, including: performing time series decomposition processing on the stress state data to obtain a mechanical feature sequence, and performing periodic analysis processing on the mechanical feature sequence to obtain a stress change pattern; performing spatial distribution mapping processing on the flow state parameters to obtain a flow feature matrix, and performing time series reconstruction processing on the flow feature matrix to obtain a flow change pattern; performing feature space mapping processing on the stress change pattern and the flow change pattern to obtain a unified feature expression, and performing dimensional normalization processing on the unified feature expression to obtain a standardized feature set; performing correlation analysis processing on the standardized feature set to obtain a feature association matrix, and The feature association matrix is ​​subjected to principal component extraction processing to obtain a core feature vector; the core feature vector is subjected to historical data comparison processing to obtain a feature importance distribution, and the feature importance distribution is subjected to interval division processing to obtain an initial weight value; the initial weight value is subjected to dynamic update processing based on state changes to obtain a real-time weight coefficient, and the real-time weight coefficient is subjected to constraint optimization processing to obtain a balanced weight value; the standardized feature set is subjected to weighted combination processing based on the balanced weight value to obtain fused feature data, and the fused feature data is subjected to redundancy elimination processing to obtain a reduced feature set; the reduced feature set is subjected to multi-level quantization processing to obtain a hierarchical index system, and the hierarchical index system is subjected to normalization processing to obtain the quality assessment index; The control module is used to perform multi-objective parameter optimization and closed-loop feedback control processing on the quality assessment index to obtain a quality control strategy based on construction process parameters, including: performing multi-parameter decomposition processing on the quality assessment index to obtain index sub-item data, and performing target interval definition processing on the index sub-item data to obtain parameter constraints; performing construction process mapping processing on the parameter constraints to obtain a process parameter set, and performing conflict detection processing on the process parameter set to obtain a feasible parameter domain; performing discrete sampling processing on the feasible parameter domain to obtain a candidate parameter group, and performing parameter sensitivity analysis processing on the candidate parameter group to obtain an influence factor matrix; performing target value calculation processing on the influence factor matrix to obtain parameter control objectives The target is obtained by multi-target conversion, and the parameter control target is converted into a control quantity sequence; the control quantity sequence and the actual execution state are calculated to obtain a state deviation value, and the state deviation value is analyzed to obtain a deviation change characteristic; the deviation change characteristic is calculated to obtain a correction parameter value, and the correction parameter value is evaluated for stability to obtain an adjustment gain coefficient; the adjustment gain coefficient is feedback compensated to obtain a dynamic correction value, and the dynamic correction value is superimposed and fused to obtain a control compensation value; the control compensation value is reconstructed to obtain an updated control value, and the updated control value is converted into a construction process to obtain the quality control strategy.

Images