CN-122024173-A - Intelligent monitoring system and method for shrinkage cavity of compaction pile

CN122024173ACN 122024173 ACN122024173 ACN 122024173ACN-122024173-A

Abstract

The application discloses an intelligent monitoring system and method for shrinkage cavity of compaction pile, which relate to the technical field of pile foundation construction monitoring, and the method comprises the steps of acquiring multi-time three-dimensional laser scanning point cloud data of the wall of the compaction pile hole, obtaining aperture data corresponding to each depth position at each monitoring time, and constructing an aperture-depth-time three-dimensional data matrix; the method comprises the steps of extracting instantaneous shrinkage rate, spatial gradient and accumulated shrinkage volume based on the three-dimensional data matrix to form a space-time feature matrix, inputting the space-time feature matrix into an LSTM-GRU hybrid neural network to obtain a future time aperture prediction result, performing cluster analysis and anomaly detection based on the space-time feature matrix to obtain a shrinkage pattern recognition result and a comprehensive anomaly score, calculating a risk score by combining the aperture prediction result, the shrinkage pattern recognition result and the comprehensive anomaly score, dividing the risk grade, and outputting a monitoring and early warning result. The method can realize continuous monitoring, trend prediction, abnormal identification and risk early warning of the shrinkage cavity state of the compaction pile.

Inventors

HAN XIDONG
ZHANG HANG
WANG XIN
REN XIUFANG

Assignees

陕西天地地质有限责任公司
陕西省煤田地质集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. An intelligent monitoring method for shrinkage cavity of compaction pile is characterized by comprising the following steps: Acquiring multi-time three-dimensional laser scanning point cloud data of the pile hole wall of the compaction pile, denoising, registering and fitting cross sections of the point cloud data to obtain aperture data corresponding to each depth position at each monitoring time, and constructing an aperture-depth-time three-dimensional data matrix; extracting instantaneous shrinkage rate, spatial gradient and accumulated shrinkage based on the three-dimensional data matrix to form a space-time feature matrix; inputting the space-time feature matrix into an LSTM-GRU hybrid neural network to obtain a future moment aperture prediction result; Executing DBSCAN algorithm and Gaussian mixture model clustering based on the space-time feature matrix to obtain a shrinkage cavity mode recognition result; Performing statistical threshold detection, isolated forest detection and self-encoder reconstruction error detection based on the space-time feature matrix, and performing weighted voting fusion on detection results to obtain comprehensive anomaly scores; and calculating a risk score based on the aperture prediction result, the shrinkage cavity mode recognition result and the comprehensive anomaly score, dividing the risk grade, and outputting a corresponding monitoring and early warning result.
2. The method of claim 1, wherein denoising, registering, and cross-section fitting the point cloud data comprises: The method comprises the steps of carrying out outlier rejection and local smoothing on original point clouds, carrying out unified coordinate registration on each frame of point clouds and point clouds at different monitoring moments in the same scanning based on depth position information and posture information of a scanning device, projecting hole wall points corresponding to each depth position to a cross-section plane, carrying out circular fitting or elliptical fitting, and determining equivalent aperture data corresponding to the depth position according to fitting results.
3. The method of claim 2, wherein said constructing an aperture-depth-time three-dimensional data matrix comprises: The method comprises the steps of arranging aperture data according to depth coordinates and monitoring time to form an aperture matrix taking depth positions and monitoring time as indexes, taking a designed aperture or an initial monitoring aperture as a reference aperture, carrying out normalization processing on the aperture matrix, and determining shrinkage ratio of each depth position at each monitoring time based on a difference value between the reference aperture and a current aperture.
4. The method of claim 3, wherein the extracting the instantaneous shrinkage rate, the spatial gradient, and the cumulative shrinkage forms a spatio-temporal feature matrix comprising: The aperture-depth-time three-dimensional data matrix is subjected to change rate calculation along a time dimension to obtain an instantaneous shrinkage cavity rate, change rate calculation along a depth dimension to obtain a spatial gradient, cumulative shrinkage cavity amount is obtained based on the difference value between the aperture at the initial monitoring moment and the aperture at the current monitoring moment, and the normalized aperture, the instantaneous shrinkage cavity rate, the spatial gradient and the cumulative shrinkage cavity amount are combined to form a space-time characteristic matrix.
5. The method of claim 4, wherein inputting the spatio-temporal feature matrix into the LSTM-GRU hybrid neural network to obtain the future time aperture prediction result comprises: And taking the historical space-time characteristic sequence of the target depth position as an input sequence, extracting time-dependent characteristics through an LSTM network, inputting into a GRU network for time sequence compression and mapping, and outputting aperture predicted values at one or more future moments through a full connection layer.
6. The method of claim 5, wherein performing DBSCAN algorithm and gaussian mixture model clustering based on the spatio-temporal feature matrix to obtain a shrinkage cavity pattern recognition result comprises: The method comprises the steps of performing a DBSCAN algorithm based on a normalized space-time feature matrix to obtain a high-density sample cluster and noise samples, performing Gaussian mixture model clustering on the basis of the high-density sample cluster to obtain the attribution probability of each sample to different clustering categories, and determining a stable uniform shrinkage cavity mode, a rapid local shrinkage cavity mode, an abnormal shrinkage cavity mode or a transition mode according to the attribution probability and the noise sample marks.
7. The method of claim 6, wherein the performing statistical threshold detection, isolated forest detection, and self-encoder reconstruction error detection based on the spatio-temporal feature matrix and weighted voting fusion of the detection results comprises: The method comprises the steps of carrying out statistics threshold detection on a space-time feature matrix based on statistics distribution of historical normal samples to obtain a first anomaly score, carrying out isolation forest detection on the space-time feature matrix to obtain a second anomaly score, inputting the space-time feature matrix into a self-encoder trained by normal samples to obtain a third anomaly score according to differences between input features and reconstruction features, and carrying out normalization weighting fusion on the first anomaly score, the second anomaly score and the third anomaly score to obtain a comprehensive anomaly score.
8. The method of claim 7, wherein the weighting voting fusion is performed on the detection results to obtain a comprehensive anomaly score, comprising: Setting corresponding weight coefficients for statistic threshold detection, isolated forest detection and self-encoder reconstruction error detection respectively, determining initial values of the weight coefficients according to detection coincidence rate in a history labeling sample, carrying out rolling correction on the weight coefficients according to an abnormality judgment result and a rechecking result in a preset updating period, and calculating the comprehensive abnormality score according to a weighted sum of each weight coefficient and the corresponding abnormality score.
9. The method of claim 8, wherein calculating and ranking risk scores based on aperture prediction results, shrinkage cavity pattern recognition results, and composite anomaly scores comprises: The method comprises the steps of taking prediction deviation corresponding to aperture prediction results, abnormal density corresponding to comprehensive abnormal scores and pattern categories corresponding to shrinkage cavity pattern recognition results as risk factors, calculating risk scores according to weighted combination of the risk factors, setting a grading threshold value for the risk scores based on the grading number of historical monitoring data, classifying the risk grades into low risk, medium risk, high risk and serious risk according to the grading threshold value, and outputting monitoring and early warning results according to the risk grades.
10. An intelligent monitoring system for shrinkage cavity of compaction pile, which is characterized by comprising: The data acquisition and processing module is used for acquiring multi-time three-dimensional laser scanning point cloud data of the pile hole wall of the compaction pile, denoising, registering and fitting a cross section on the point cloud data to obtain aperture data corresponding to each depth position at each monitoring time, and constructing an aperture-depth-time three-dimensional data matrix; the characteristic construction module is used for extracting instantaneous shrinkage rate, spatial gradient and accumulated shrinkage based on the three-dimensional data matrix to form a space-time characteristic matrix; the prediction analysis module is used for inputting the space-time feature matrix into an LSTM-GRU hybrid neural network to obtain a future moment aperture prediction result; The pattern recognition module is used for executing a DBSCAN algorithm and Gaussian mixture model clustering based on the space-time feature matrix to obtain a shrinkage cavity pattern recognition result; the anomaly detection module is used for executing statistical threshold detection, isolated forest detection and self-encoder reconstruction error detection based on the space-time feature matrix, and carrying out weighted voting fusion on detection results to obtain comprehensive anomaly scores; And the risk assessment and early warning module is used for calculating the risk score and dividing the risk grade based on the aperture prediction result, the shrinkage cavity mode recognition result and the comprehensive anomaly score, and outputting a corresponding monitoring early warning result.

Description

Intelligent monitoring system and method for shrinkage cavity of compaction pile Technical Field The application relates to the technical field of pile foundation construction monitoring, in particular to an intelligent monitoring system and method for shrinkage cavity of a compaction pile. Background In the compaction pile construction process, the pile hole wall is influenced by soil stress redistribution, groundwater change, construction disturbance and time effect, and local shrinkage holes, uneven pore diameters or hole wall deformation are easy to occur within a certain time after hole forming. Once shrinkage cavities are not found in time, the quality of subsequent filling, tamping and piling can be affected, and local defects of pile bodies can be caused when the shrinkage cavities are serious, so that continuous monitoring of pore diameter changes after pore forming is needed. The existing field practice mostly adopts a manual rechecking mode, a single through hole inspection mode or a discrete detection mode based on a small number of measuring points, can only reflect the hole state at a certain moment or a certain local position, is difficult to obtain continuous change information of the aperture along the depth direction and the time direction at the same time, and has limited identification capability for shrinkage cavities, local burst shrinkage cavities and evolution differences of different hole segments which occur in a delayed mode. On the other hand, the existing monitoring means are still imperfect in the aspects of unified comparison of multi-time data, shrinkage cavity development trend analysis, abnormal mode distinction and risk early warning, and quantitative, continuous and traceable evaluation of pile hole shrinkage cavity states is difficult to form. Therefore, a technical scheme capable of continuously acquiring, analyzing and outputting early warning results for the hole diameter change is needed to be used for monitoring scenes after the compaction pile is formed. Disclosure of Invention The embodiment of the application provides an intelligent monitoring system and method for shrinkage cavity of a compaction pile, which are used for at least solving part of technical problems in the related art. According to a first aspect of the embodiment of the application, an intelligent monitoring method for shrinkage cavity of a compaction pile is provided, which comprises the following steps: Acquiring multi-time three-dimensional laser scanning point cloud data of the pile hole wall of the compaction pile, denoising, registering and fitting cross sections of the point cloud data to obtain aperture data corresponding to each depth position at each monitoring time, and constructing an aperture-depth-time three-dimensional data matrix; extracting instantaneous shrinkage rate, spatial gradient and accumulated shrinkage based on the three-dimensional data matrix to form a space-time feature matrix; inputting the space-time feature matrix into an LSTM-GRU hybrid neural network to obtain a future moment aperture prediction result; Executing DBSCAN algorithm and Gaussian mixture model clustering based on the space-time feature matrix to obtain a shrinkage cavity mode recognition result; Performing statistical threshold detection, isolated forest detection and self-encoder reconstruction error detection based on the space-time feature matrix, and performing weighted voting fusion on detection results to obtain comprehensive anomaly scores; and calculating a risk score based on the aperture prediction result, the shrinkage cavity mode recognition result and the comprehensive anomaly score, dividing the risk grade, and outputting a corresponding monitoring and early warning result. The method comprises the steps of carrying out denoising, registering and section fitting on point cloud data, carrying out outlier rejection and local smoothing on original point cloud, carrying out unified coordinate registration on each frame of point cloud and point cloud at different monitoring moments in the same scanning based on depth position information and posture information of a scanning device, projecting hole wall points corresponding to each depth position to a section plane, carrying out circular fitting or elliptic fitting, and determining equivalent aperture data of corresponding depth positions according to fitting results. As an alternative, the construction of the aperture-depth-time three-dimensional data matrix includes: The method comprises the steps of arranging aperture data according to depth coordinates and monitoring time to form an aperture matrix taking depth positions and monitoring time as indexes, taking a designed aperture or an initial monitoring aperture as a reference aperture, carrying out normalization processing on the aperture matrix, and determining shrinkage ratio of each depth position at each monitoring time based on a difference value between the reference aperture and