Method and system for acquiring motion trail of rock-soil particles in shield tunneling process

Abstract

The invention relates to the field of rock-soil particle motion trail acquisition, and discloses a rock-soil particle motion trail acquisition method and system in a shield tunneling process, and the method comprises the steps: multi-source data collection and fusion preparation, rock-soil particle preliminary identification and classification, unmarked motion tracking algorithm initialization, and real-time motion trail calculation and correction. The method comprises the steps of uncertain parameter real-time estimation, simulation and actual measurement data comparison verification, motion trail visualization and analysis report generation, and adaptive optimization and feedback control. According to the method, multi-source data acquisition and fusion preparation is made firstly, a data foundation is laid, then through links of rock-soil particle identification and classification, motion tracking algorithm initialization and the like in sequence, functions of particle motion tracking, trajectory calculation and correction, parameter estimation and the like are realized by utilizing multiple means, and through simulation comparison verification, visualization and self-adaptive optimization feedback control, the real-time performance of the system is improved. And finally, the construction efficiency and quality are improved, construction decision is assisted, construction is guaranteed, and data management is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of obtaining the movement trajectories of geotechnical particles. Specifically, it relates to a method and system for obtaining the movement trajectories of geotechnical particles during shield tunneling. Background Art

[0002] With the development of shield tunneling projects towards large-scale and complex directions, many limitations of traditional technologies have emerged in obtaining the movement trajectories of geotechnical particles. Due to the complex construction environment of shield tunneling and the variable properties of geotechnical bodies, it is impossible to accurately capture the details of particle movement relying on single sensors or simple measurement methods in the past, and it is difficult to meet the tunneling requirements under different stratum conditions. As a result, there is a lack of accurate data support for surface settlement control, cutter wear prediction, and tunneling parameter optimization during the construction process, thus giving rise to the technical need for more advanced and comprehensive methods and systems for obtaining the movement trajectories of geotechnical particles.

[0003] However, there are problems of marker interference in the existing technologies for obtaining the movement trajectories of geotechnical particles, which will cause the markers to fall off and change the physical properties of geotechnical particles, resulting in the inability to accurately track the particle movement trajectories and the inability to truly reflect their original movement states; the visibility limitations make it difficult to comprehensively and continuously observe the particle movement trajectories. Due to the obstruction of the internal structure of the shield machine and the influence of dust and mud, it is impossible to completely obtain the trajectories of the entire process from the excavation face to the soil discharge port; the marker efficiency problem makes the marking of a large number of geotechnical particles cumbersome and time-consuming, and the marking difficulty is different for different particle sizes. When changing the stratum, re-marking is required, which greatly reduces the research efficiency. The differences between the model assumptions of numerical simulation methods and the actual situation will lead to deviations between the simulation results and the actual particle movement trajectories due to the simplified treatment of the shapes of geotechnical particles and the complex physical processes of shield tunneling; the uncertainty of parameter acquisition is due to the fact that the working parameters of geotechnical materials and shield machines are difficult to accurately measure and often change, resulting in a large deviation between the simulated trajectories and the actual situation; the calculation resource and time limitations mean that high-precision simulation requires a large amount of calculation resources and a long calculation period, and it cannot meet the requirements of real-time monitoring and timely adjustment of construction parameters. These deficiencies will seriously affect the accuracy, integrity, timeliness, and acquisition efficiency of obtaining the movement trajectories of geotechnical particles during shield tunneling, and are not conducive to the effective development of construction and the reasonable adjustment of parameters.

[0004] Therefore, it is necessary to design a method and system for obtaining the movement trajectories of geotechnical particles during shield tunneling to solve the problems that the marker interference problem of the existing technologies for obtaining the movement trajectories of geotechnical particles will affect the tracking and reflection of the original movement state, the visibility limitations make it difficult to comprehensively and continuously observe the trajectories, the low marker efficiency reduces the research efficiency, and the numerical simulation methods have differences between model assumptions and the actual situation, uncertainty in parameter acquisition, and are restricted by calculation resources and time. These all seriously affect the accuracy, integrity, timeliness, and efficiency of trajectory acquisition, and are not conducive to construction development and parameter adjustment. Summary of the Invention

[0005] In view of this, the present invention proposes a method and system for obtaining the movement trajectory of rock and soil particles during shield tunneling, aiming to solve the problems that the interference of the existing rock and soil particle movement trajectory acquisition technology markers will affect the tracking and reflection of the original movement state, the visibility limitation makes it difficult to comprehensively and continuously observe the trajectory, the low marking efficiency reduces the research efficiency, and the numerical simulation method has differences between model assumptions and reality, uncertain parameter acquisition, and is limited by computing resources and time. These all seriously affect the accuracy, integrity, timeliness and efficiency of trajectory acquisition, and are not conducive to the construction development and parameter adjustment.

[0006] On the one hand, the present invention proposes a method for obtaining the movement trajectory of rock and soil particles during shield tunneling, including:

[0007] Planning the installation layout of sensors and accurately calibrating them, and building an industrial Ethernet data transmission network and a large-capacity storage device;

[0008] Preprocessing the laser scanning point cloud data by using a variety of filtering algorithms, extracting and combining geometric and image features to construct a classification model, and accurately identifying and classifying rock and soil particles;

[0009] Preprocessing and annotating the initial image sequence of shield tunneling, dividing the data set to train an algorithm combining optical flow method and convolutional neural network, optimizing the parameters and evaluating the performance to initialize the markerless motion tracking algorithm;

[0010] During shield tunneling, data is collected and processed at time intervals, the movement trajectory of rock and soil particles is calculated and corrected according to the data of pressure and vibration sensors to adapt to the change of the working state of the shield machine;

[0011] Sorting out the geological exploration and shield machine historical data and constructing a prior probability distribution function, collecting sensor data in real time and inputting it into the Bayesian estimation model, and updating the parameters and posterior probability distribution to estimate the uncertainty parameters;

[0012] Constructing a numerical simulation model according to the shield machine and formation conditions and setting parameters, comparing the simulated and measured movement trajectories, and when the deviation exceeds the threshold, checking and optimizing the relevant links and then re-simulating and verifying;

[0013] Developing a visualization platform to display the movement trajectory of rock and soil particles, generating an analysis report containing multi-faceted information, and archiving it after automatic and manual review to ensure accuracy;

[0014] Constructing a multi-objective adaptive optimization control model based on the model predictive control framework, calculating and feedbacking the shield machine working parameter adjustment scheme according to the real-time data, and coordinating each link of the system to achieve closed-loop control to improve the construction efficiency and quality.

[0015] Furthermore, the planning of the installation layout of sensors and accurate calibration, and the building of an industrial Ethernet data transmission network and a large-capacity storage device include:

[0016] Sensor installation and calibration. During the design stage of the shield machine, determine the optimal installation layout of sensors. For laser scanning sensors, according to the internal space geometry of the shield machine and the possible distribution range of the rock and soil mass, use computer-aided design software to analyze the simulated installation positions to ensure that the scanning angle and range of the sensors can cover key areas without obvious blind spots. The laser scanning sensors scan the corresponding areas to collect point cloud data. When installing pressure sensors, based on the results of the mechanical structure analysis of the cutter head and the shield wall, determine the force concentration points of the cutter head tools and the key monitoring points of the circumferential and longitudinal stresses of the shield wall for embedded installation. After installation, use a pressure calibrator to perform static and dynamic calibration on each pressure sensor. The vibration sensors are installed on the main drive system of the shield machine, key parts of the shield body shell, and the drive device and shell of the screw conveyor, and are connected by welding or high-strength bolts. After installation, calibrate them on a vibration test platform;

[0017] Construction of data transmission and storage equipment. Build a high-speed data transmission network based on industrial Ethernet. The network topology structure adopts a redundant ring structure. Lay fiber optic cables along the tunnel axis direction inside the shield machine and connect them through fiber optic splicing boxes and switches. Divide the data of different types of sensors into different virtual local area networks. Each node in the network is assigned a unique IP address, and strict network access permission settings are made. Only authorized devices can access and transmit data. Configure a large-capacity data storage device using redundant array of independent disks (RAID) technology. The storage capacity of the disk array is estimated according to the expected duration of the shield machine tunneling, the data acquisition frequency, and the data volume size, and sufficient storage space is reserved to meet the continuous data storage requirements for at least several months. Establish a structured data storage table to classify and store data in multiple dimensions such as time series, sensor type, and shield machine parts. The data storage format adopts the standardized HDF5 format. Establish a data backup mechanism, and use a combination of off-site backup and regular full and incremental backups to back up the data to external storage media.

[0018] Furthermore, the use of multiple filtering algorithms to preprocess the laser scanning point cloud data, extract and combine geometric and image features to construct a classification model, and accurately identify and classify rock and soil particles includes:

[0019] Point cloud preprocessing: The statistical filtering algorithm is used to remove outliers from the laser-scanned point cloud data. Calculate the average distance between each point in the point cloud data and its neighboring points. According to the overall distribution characteristics of the point cloud data, set a distance threshold. Points whose distance exceeds this threshold are regarded as outliers and removed from the dataset. The filtering algorithm is iteratively executed multiple times until the number of outliers in the remaining point cloud data meets the requirements. The passthrough filtering algorithm is used to remove the reflected point cloud generated by the internal equipment of the shield machine according to the coordinate range of the point cloud. Obtain the spatial coordinate range of the internal equipment based on the CAD model of the shield machine, and then set the range parameters of the passthrough filtering to filter out the point cloud data located within the equipment coordinate range. For some equipment reflection points that are difficult to accurately filter by the coordinate range, further screening is carried out in combination with the intensity information of the point cloud. The density-based clustering algorithm is used to segment the point cloud into different clusters. Set the parameters of the DBSCAN algorithm according to the expected density range of the rock and soil particle point cloud. After the point cloud is segmented into multiple clusters by the clustering algorithm, further screen out the point cloud clusters belonging to individual rock and soil particles according to the size, shape, and point density of the clusters, and remove the clusters that are too large or too small;

[0020] Feature extraction and classification model construction, calculating the detailed geometric features of the extracted rock and soil particle point cloud data, including the minimum bounding box size, sphericity, surface curvature, ratio of the major axis to the minor axis length of the particle, and surface concavity and convexity. Using the principal component analysis algorithm to determine the principal direction of the particle, constructing the covariance matrix of the point cloud data, calculating the eigenvalues and eigenvectors of the covariance matrix, where the eigenvectors correspond to the principal direction of the point cloud data. By projecting the point cloud data onto the principal direction, the size information of the particle in different directions is obtained, and then the ratio of the major axis to the minor axis length is calculated. The surface concavity and convexity feature is determined by calculating the curvature change rate of the point cloud surface. Combining image analysis technology, extracting the color features of the particles from the images synchronously collected with the point cloud data, converting the color images into different color spaces for analysis. In the RGB color space, calculating the color histogram of the particle region, statistically analyzing the pixel value distribution of each color channel to obtain the color mean and variance. In the HSV color space, focusing on analyzing the features of the hue and saturation channels. The texture feature extraction uses the gray-level co-occurrence matrix algorithm. First, converting the color image into a gray image, and then calculating the gray-level co-occurrence matrix in different directions and at different distances, and extracting texture parameters such as contrast, correlation, energy, and entropy from the gray-level co-occurrence matrix. When combining the geometric features and image features into a feature vector, reasonable combination and normalization processing are performed according to the importance of the features and the relevance of the data. Using the support vector machine algorithm to construct a classification model, using the training data for model training, and selecting the appropriate kernel function and penalty parameter C through the cross-validation method. The cross-validation adopts k-fold cross-validation, dividing the training data into k subsets, using k - 1 subsets as the training set each time, and the remaining one subset as the validation set, repeating k times, calculating the classification accuracy on each validation set, and selecting the combination of the kernel function and penalty parameter with the highest accuracy. Using the random forest algorithm to construct a model, determining parameters such as the number of decision trees and the maximum depth of the tree, and adjusting the parameters by evaluating the out-of-bag error. By continuously adjusting the number and depth of the decision trees and observing the change of the out-of-bag error, selecting the parameter combination that minimizes the out-of-bag error. After training is completed, using the test data to evaluate the model, calculating indicators such as the accuracy, recall rate, and F1 value of the model, and the calculation formula is F1 = 2 * (accuracy * recall rate) / (accuracy + recall rate).

[0021] Further, the preprocessing and annotation of the initial image sequence of the shield tunneling, dividing the data set to train the algorithm combining the optical flow method and the convolutional neural network, optimizing the parameters and evaluating the performance to initialize the unmarked motion tracking algorithm, including:

[0022] Data preprocessing and annotation: Preprocess the image sequence at the initial stage of shield tunneling collected. Image enhancement includes adjusting brightness, contrast, and color balance. The brightness of the image is adjusted using the adaptive histogram equalization algorithm. Image cropping operations are precisely cropped according to the position and size of the key areas inside the shield machine, removing the irrelevant background areas in the image, reducing the data volume, and highlighting the areas where the rock and soil particles are located. Image deblurring is performed using the blind deconvolution algorithm, which automatically estimates the blur function based on the blur characteristics of the image and performs deconvolution operations on the image to restore the clarity of the image. The image is converted into a grayscale image, and the RGB three channels of the color image are converted into grayscale values using the weighted average method. Label the rock and soil particles in the image. Manually label some key image frames, marking the exact position and bounding box size of the particles. Semi-automatic annotation uses the labeled image frames to train the object detection model. Use the pre-trained convolutional neural network as the basic network and perform fine-tuning to adapt to the detection task of rock and soil particles. Apply the model to the unlabeled image frames to automatically generate the initial annotation information of the particles, and then manually review and correct it. The annotation information includes the position, outside the bounding box, and the class label of the particles. Different types of rock and soil particles are labeled as different classes respectively;

[0023] Algorithm training and optimization: Divide the labeled data into a training set, a validation set, and a test set according to the ratio of 70%, 20%, and 10%. Use the stratified sampling method to ensure that the ratio of various types of rock and soil particles in the training set, validation set, and test set is approximately the same as that in the original data. Use the training set to train the algorithm combining the optical flow method and the convolutional neural network. In the convolutional neural network part, try different network architectures and select the optimal architecture according to the accuracy on the validation set. The adaptive learning rate adjustment strategy is used to optimize the algorithm parameters. Set the initial learning rate and decay factor, and determine the number of iterations according to the convergence of the model on the validation set. By observing the change of the loss function value on the validation set, when the loss function value no longer decreases significantly within several consecutive iteration steps, it is considered that the model has converged and stop training. Use the L1 / L2 regularization technique to prevent overfitting, adjust the regularization coefficient, observe the influence of regularization on the accuracy and generalization ability of the model on the validation set, and select the appropriate regularization coefficient. Use the test set to evaluate the performance of the trained algorithm, calculate the mean average precision and intersection over union. When the indicators do not meet the expected requirements, further analyze the reasons, make targeted adjustments and then retrain until the performance of the algorithm on the test set meets the reliability requirements of practical applications.

[0024] Furthermore, data is collected and processed at time intervals during shield tunneling, and the movement trajectories of rock and soil particles are calculated and corrected based on the data of pressure and vibration sensors to adapt to the changes in the working state of the shield machine, including:

[0025] Real-time data processing. When collecting laser scanning data, check the data integrity, verify the spatial consistency of the point cloud data, compare the coordinate change ranges of the same feature points in adjacent scan frames. If it exceeds the preset threshold, mark the data as abnormal. Check the clarity of the image data. Based on the evaluation of image contrast and edge sharpness, use the image spectrum analysis method to calculate the Fourier spectrum of the image. If the high-frequency components are missing or abnormally attenuated, it indicates that the image is blurred and mark it when it is lower than the set clarity threshold. When the data is marked as unqualified, supplementary collection is carried out with a delay of no more than 0.5 seconds based on the original collection time point. The maximum number of supplementary collections is 3 times. If it is still unqualified, an alarm is issued to prompt manual intervention to check the equipment. Data compression uses a hybrid compression algorithm. First, compress the laser scanning data based on geometric features. For image data, first perform region segmentation and use different compression ratios for the background region and the particle region respectively. The background region uses a high compression ratio, and the particle region uses a low compression ratio. Data encryption uses the AES-256 encryption algorithm to encrypt the compressed data. The encryption key is generated dynamically and updated every 10 minutes. The key seed is generated based on the running time and current position of the shield machine;

[0026] Motion trajectory calculation and correction logic. Calculate the displacement vector of the geotechnical particles between adjacent frames. Use the Lucas-Kanade algorithm based on the pyramid. First, construct an image pyramid, calculate the optical flow from the top layer, and gradually refine it to the lower layer. Combine the laser scanning data to determine the three-dimensional position change of the particles. Use the high-precision spatial coordinate information of the laser scanning to calibrate the displacement vector calculated from the image. When the pressure sensor detects abnormal pressure, the position of the pressure change is located through the installation position information of the sensor. The abnormal degree is determined by comparing the pressure amplitude with the normal tunneling pressure range. The correction coefficient is determined using a linear or non-linear mapping relationship according to the abnormal degree of the pressure. When processing the vibration sensor data, use the time-frequency analysis method to convert the vibration signal to the time-frequency domain, analyze the frequency components and amplitude changes of the vibration. If the high-frequency vibration energy exceeds 50% of the normal threshold, it is judged that the cutter head cutting is uneven, and a random perturbation in the particle movement direction is increased. The perturbation amplitude is determined according to the high-frequency vibration energy size and is between 5° and 15°. If the low-frequency vibration amplitude exceeds 30% of the normal threshold, it indicates that the overall advancement of the shield machine is blocked, and the estimated value of the particle movement speed inside the entire shield machine is reduced by 20% - 30% to reflect the slowdown of particle movement when the shield machine has difficulty advancing.

[0027] Furthermore, the arrangement of geological exploration and shield machine historical data and the construction of the prior probability distribution function, real-time collection of sensor data and input into the Bayesian estimation model, update the parameters and posterior probability distribution to estimate the uncertainty parameters, including:

[0028] Prior information collation and model construction: When collating the data of geological exploration reports, statistical distribution information is counted, the correlation between geotechnical parameters of different strata is analyzed, and the prior probability distribution function of the Bayesian estimation model is constructed. For complex geotechnical parameters, a mixed probability distribution model is adopted. Different distribution parameters are determined according to different stratum regions or depth ranges, and the prior distribution is constructed in a weighted combination manner. The weights of the prior distribution are determined according to the reliability of geological exploration data and the uniformity of strata.

[0029] Real-time estimation and update: Sensor data is input into the Bayesian estimation model as observation values, and the data is preprocessed. The pressure data is filtered to remove measurement noise. The Kalman filter algorithm is used, and the filter parameters are set according to the dynamic characteristics of pressure changes. The vibration data is subjected to feature extraction, and the frequency, amplitude, peak factor, and kurtosis of vibration are extracted. An association model is established between these features and geotechnical parameters and shield machine operating parameters. The model update uses the Gibbs sampling algorithm in the Markov chain Monte Carlo method, and a sufficient number of sampling times is set to ensure the convergence of the posterior probability distribution. After each sampling, the confidence interval of the parameter estimation value is calculated, and the confidence level is set to 95%. If the new sampling value exceeds the confidence interval, the model parameters or the prior distribution are adjusted, the parameter range or weight of the prior distribution is adjusted, and sampling is restarted until the parameter estimation value is stable within the confidence interval and the posterior probability distribution no longer changes significantly.

[0030] Furthermore, a numerical simulation model is constructed and parameters are set according to the shield machine and stratum conditions. When the deviation between the simulated and measured movement trajectories exceeds the threshold, relevant links are checked and optimized, and then re-simulated and verified, including:

[0031] Numerical simulation model construction and parameter setting. Finite element modeling is carried out. The cutter head, shield and other structures of the shield machine are finely modeled using solid elements. Considering the anisotropic characteristics of the material, for the cutters on the cutter head, contact elements are used to simulate their interaction with the rock and soil mass. The friction coefficient of the contact elements is determined according to the actual friction test results of the cutter material and the rock and soil type, accurate to 0.01. Discrete element modeling is carried out. The mechanical parameters of the rock and soil particles are set according to the classification results and the real-time estimated parameters. The contact model between particles adopts the Hertz-Mindlin contact model, and the rolling friction of the particles is considered. Boundary conditions are set. The thrust force of the shield machine is dynamically adjusted according to the set value of the actual construction and the real-time monitored propulsion speed. The constraint conditions of the surrounding rock and soil mass are simulated using spring-damper elements. The spring and damping coefficients are determined according to the elastic modulus and damping ratio of the formation. The elastic modulus and damping ratio are adjusted according to the geological exploration data and the real-time estimated parameters to accurately reflect the constraint effect of the formation on the shield machine. In the initial condition setting, the initial stress state of the rock and soil mass is determined according to the in-situ stress measurement data, and the particle distribution is initialized according to the initial shape of the rock and soil mass obtained by laser scanning to ensure that the initial state of the simulation is consistent with the actual situation;

[0032] Comparative analysis and deviation processing. Calculate the distance deviation and speed deviation between trajectory points. Using the spatial vector calculation method, considering the three-dimensional characteristics of particle motion, calculate the Euclidean distance and the deviation components of the trajectory points in different coordinate axis directions. The speed deviation is calculated by comparing the instantaneous speed and the average speed. Calculate the instantaneous speed and the average speed of the simulated trajectory and the measured trajectory in the same time period respectively, and compare the differences between the two. When the deviation exceeds the threshold range, check the abnormal process of the sensor data, check for data loss and fault alarms, and conduct data rationality checks. For algorithm problem analysis, the sensitivity analysis method is used. Change the key parameters in the algorithm one by one, observe the influence on the trajectory calculation results, determine the key parameters causing the deviation and carry out optimization and adjustment.

[0033] Furthermore, the developed visualization platform displays the motion trajectories of the rock and soil particles, generates an analysis report containing multi-faceted information, and archives it after automatic and manual review to ensure accuracy, including:

[0034] Visualization platform construction, development of a visualization platform. The 3D graphics rendering engine adopts physically based rendering technology. In the internal view of the shield machine, it displays the movement trajectories of rock and soil particles, as well as the temperature distribution and stress distribution of the internal structure of the shield machine, and updates the display in real-time by connecting with sensor data. For the classification display function, when displaying by particle type, different colors and shapes are used to represent different types of particles. When displaying by the range of movement speed, speed intervals are set, and particles in different speed intervals are represented by different transparencies or blinking frequencies. In terms of the interaction function, after the user clicks on a specific particle, the movement information and detailed physical properties of the particle are displayed;

[0035] Analysis report template and content generation, recording general information, construction team information of the shield machine, construction environment temperature and humidity, summarizing the average movement speed, movement direction distribution and residence time distribution of rock and soil particle movement, analyzing the acceleration change of particle movement. The acceleration is obtained by calculating the second derivative of the trajectory data, and the magnitude and direction changes of the acceleration are analyzed to understand the force change situation of the particles inside the shield machine. To evaluate the influence of parameter uncertainty, a sensitivity analysis method is used to quantify the influence degree of different parameter uncertainties on the particle movement trajectory, and the parameters are sorted according to the magnitude of the sensitivity coefficient, and their changes in different construction stages and the influence on the working state of the shield machine are analyzed. Analyze the cutting efficiency, soil discharge smoothness and energy consumption of the shield machine during operation. According to the power monitoring data of the shield machine, calculate the energy consumption per unit tunneling distance, and combine the particle movement trajectory analysis to analyze the relationship between energy consumption and particle movement state. The automated part of the analysis report generation is written in a scripting language to write an automated report generation program, which automatically fills the content of the report template according to the data processing and analysis results to generate a preliminary report. Manually review the accuracy of the data, the rationality of the analysis method and the feasibility of the suggestions, and sign and confirm after passing the review. The final report is archived in two forms: electronic document and paper document. The electronic document is in PDF format, and the paper document is archived and saved according to the construction project and time sequence.

[0036] Furthermore, a multi-objective adaptive optimization control model is constructed based on the model predictive control framework, which calculates and feeds back the shield machine working parameter adjustment plan according to real-time data, and coordinates each link of the system to achieve closed-loop control to improve construction efficiency and quality, including:

[0037] Optimization control model construction and parameter setting. The multi-objective adaptive optimization control model is constructed based on the model predictive control framework. The input variables include the influence of conventional variables and the attitude parameters of the shield machine on particle movement and construction effect. The objective function is set in the form of a weighted sum of multiple sub-objective functions. The sub-objective function of the tool wear rate is constructed according to the wear characteristics of the tool material and the relationship between the shield machine working parameters. Controlling the tool wear rate within an acceptable level is used as a constraint condition, and the analytic hierarchy process is used to determine the weight coefficients of the objective function;

[0038] The feedback control collaborates with the system. The optimization control model calculates the optimal working parameter adjustment scheme based on real-time data, and uses an efficient optimization algorithm to solve it. During the solution process, considering the adjustment limitations of the shield machine's working parameters, the adjusted parameters are fed back to the shield machine control system, and an industrial communication protocol is used for data transmission. After receiving the parameter instruction, the control system completes the parameter adjustment and executes a new operation within 0.1 - 0.5 seconds, and feeds back the adjusted parameters and control effect information to the motion trajectory acquisition system. The motion trajectory acquisition system further optimizes the data acquisition frequency and range based on this information, adjusts the algorithm parameters, and optimizes the parameters in the motion tracking algorithm according to the control effect information.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: Through multi-source data acquisition and fusion preparation, various sensors are accurately installed and calibrated, and a reliable data transmission and storage system is constructed, providing a rich and accurate data basis for subsequent analysis; through the preliminary identification and classification of geotechnical particles, various algorithms and technologies are used to extract features and construct a classification model to effectively distinguish particle types; data preprocessing, annotation, algorithm training, and optimization in the initialization of the markerless motion tracking algorithm can accurately track the particle motion; in the real-time motion trajectory calculation and correction link, a combination of various data and algorithms is used to calculate and correct the trajectory to handle abnormal situations; real-time estimation of uncertain parameters uses prior information and Bayesian models to improve the accuracy of parameter estimation; comparison and verification of simulation and measured data can check for deviations and optimize the model and algorithm; motion trajectory visualization and analysis report generation facilitate intuitive understanding and comprehensive evaluation; adaptive optimization and feedback control construct an optimization model and achieve system collaboration, ultimately effectively improving the efficiency and quality of shield tunneling construction, accurately controlling surface settlement and cutter wear rate, ensuring the safety and smooth progress of construction, providing a detailed and scientific basis for construction decision-making, and facilitating data archiving and sharing.

[0040] On the other hand, the present application also provides a system for obtaining the motion trajectory of geotechnical particles during shield tunneling, including:

[0041] Multi-source data acquisition and fusion preparation module: Plan the installation layout of sensors and calibrate them accurately, and build an industrial Ethernet data transmission network and a large-capacity storage device;

[0042] Geotechnical particle preliminary identification and classification module: Use a variety of filtering algorithms to preprocess the laser scanning point cloud data, extract and combine geometric and image features to construct a classification model, and accurately identify and classify geotechnical particles;

[0043] Markerless motion tracking algorithm initialization module, preprocess and annotate the initial image sequence of shield tunneling, divide the data set to train an algorithm combining optical flow method and convolutional neural network, optimize the parameters and evaluate the performance to initialize the markerless motion tracking algorithm;

[0044] Real-time motion trajectory calculation and correction module: Collect and process data at time intervals during shield tunneling, calculate the motion trajectories of rock and soil particles, and correct them based on the data from pressure and vibration sensors to adapt to changes in the working state of the shield machine;

[0045] Uncertainty parameter real-time estimation module: Organize geological exploration and historical data of the shield machine and construct a prior probability distribution function, collect sensor data in real time and input it into the Bayesian estimation model, and update the parameters and posterior probability distribution to estimate uncertainty parameters;

[0046] Simulation and measured data comparison and verification module: Build a numerical simulation model based on the shield machine and formation conditions and set parameters, compare the simulated and measured motion trajectories, and when the deviation exceeds the threshold, check and optimize the relevant links and then re-simulate and verify;

[0047] Motion trajectory visualization and analysis report generation module: Develop a visualization platform to display the motion trajectories of rock and soil particles, generate an analysis report containing various aspects of information, and archive it after ensuring accuracy through automated and manual reviews;

[0048] Adaptive optimization and feedback control module: Build a multi-objective adaptive optimization control model based on the model predictive control framework, calculate and feedback the shield machine working parameter adjustment plan based on real-time data, and coordinate all links of the system to achieve closed-loop control to improve construction efficiency and quality.

[0049] It can be understood that the above-mentioned system and method for obtaining the motion trajectories of rock and soil particles during shield tunneling have the same beneficial effects, which will not be elaborated here. Description of the Drawings

[0050] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0051] Figure 1 It is a flowchart of a method for obtaining the motion trajectories of rock and soil particles during shield tunneling provided by an embodiment of the present invention;

[0052] Figure 2 It is a functional block diagram of a system for obtaining the motion trajectories of rock and soil particles during shield tunneling provided by an embodiment of the present invention. Detailed Embodiments

[0053] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0054] Referring to Figure 1 , in some embodiments of the present application, a method for obtaining the movement trajectory of rock and soil particles during shield tunneling includes:

[0055] Step S100: Plan the installation layout of the sensors and calibrate them precisely, and build an industrial Ethernet data transmission network and a large-capacity storage device.

[0056] Step S200: Preprocess the laser scanning point cloud data using a variety of filtering algorithms, extract and combine geometric and image features to construct a classification model, and accurately identify and classify rock and soil particles.

[0057] Step S300: Preprocess and annotate the initial image sequence of shield tunneling, divide the data set to train an algorithm combining optical flow method and convolutional neural network, optimize the parameters and evaluate the performance to initialize the unmarked motion tracking algorithm.

[0058] Step S400: Collect and process data at time intervals during shield tunneling, calculate the movement trajectory of rock and soil particles and correct it according to the data of pressure and vibration sensors to adapt to the change of the working state of the shield machine.

[0059] Step S500: Sort out the geological exploration and historical data of the shield machine and construct a prior probability distribution function, collect sensor data in real time and input it into the Bayesian estimation model, update the parameters and posterior probability distribution to estimate the uncertainty parameters.

[0060] Step S600: Build a numerical simulation model according to the shield machine and formation conditions and set the parameters, compare the simulated and measured movement trajectories, and when the deviation exceeds the threshold, check and optimize the relevant links and then re-simulate and verify.

[0061] Step S700: Develop a visualization platform to display the movement trajectory of rock and soil particles, generate an analysis report containing multi-faceted information, and archive it after ensuring accuracy through automated and manual reviews.

[0062] Step S800: Build a multi-objective adaptive optimization control model based on the model predictive control framework, calculate and feedback the shield machine working parameter adjustment plan according to the real-time data, and coordinate each link of the system to achieve closed-loop control to improve the construction efficiency and quality.

[0063] Specifically, during the design phase of the shield machine, a multi-disciplinary team (including mechanical engineers, geotechnical engineers, and sensor technology experts) jointly determines the optimal installation layout of the sensors. For the laser scanning sensor, based on the internal space geometry of the shield machine and the possible distribution range of the geotechnical body, computer-aided design (CAD) software is used to analyze the simulated installation positions to ensure that the scanning angle and range of the sensor can cover the key areas without obvious blind spots. The laser scanning sensor scans the corresponding area to collect point cloud data. For example, multiple laser scanning sensors are set at positions such as behind the cutter head, in the middle of the shield machine, and near the soil discharge port. The mounting brackets are made of materials with high strength and low vibration conduction, and the installation angle and position accuracy of the sensor are ensured to be within ±0.5° and ±5 mm through precise machining and assembly processes.

[0064] Specifically, when installing the pressure sensor, based on the results of the mechanical structure analysis of the cutter head and the shield wall, it is determined to be embedded at the stress concentration points of the cutter head tools (such as near the cutting edge of the tool, the connection part between the cutter head spokes and the hub) and the key monitoring points of the circumferential and longitudinal stresses of the shield wall (such as the top, bottom, both sides of the waist of each ring of the shield wall, and the connection part between adjacent rings). The installation process strictly follows the operating procedures of the sensor manufacturer, and a high-precision torque wrench is used to control the installation torque to ensure that the sensor is closely fitted with the structure body, the pressure transmission is uniform and there is no leakage. After installation, a professional pressure calibrator is used to calibrate each pressure sensor statically and dynamically. Static calibration is carried out by gradually applying standard pressure values from 0 to the upper limit of the sensor range, recording the output signal of the sensor, and comparing it with the standard pressure value, and adjusting the zero offset and range coefficient inside the sensor to control the measurement error within ±0.5%. Dynamic calibration simulates the pressure change rate that the shield machine may encounter during actual operation, checks the response time and frequency characteristics of the sensor to ensure that it can accurately capture rapidly changing pressure signals.

[0065] Specifically, the vibration sensors are installed on the main drive system of the shield machine (including the motor, reducer, and main bearing), key parts of the shield body shell (such as the top, bottom, both sides), and the drive device and shell of the screw conveyor. The installation method is welding or connection with high-strength bolts to ensure a rigid connection between the sensor and the structure body, reducing measurement errors caused by loose connections. After installation, calibration is carried out on a special vibration test platform. The platform can generate vibration signals with precisely controlled frequencies (ranging from 0.1 Hz to 1000 Hz) and amplitudes (from 0.01 mm to 10 mm). The vibration sensor is installed on the platform, its output signal is measured, and compared with the standard vibration signal, and the gain, filtering parameters, etc. inside the sensor are adjusted to make the frequency response error of the sensor within ±2% and the amplitude measurement error controlled within ±0.1 mm.

[0066] Specifically, when constructing a high-speed data transmission network based on industrial Ethernet, a fiber optic cable with a high bandwidth (such as 10 Gbps or higher) is selected as the transmission medium, and the network topology adopts a redundant ring structure to improve the reliability and self-healing ability of the network. The fiber optic cable is laid along the tunnel axis inside the shield machine and connected through fiber optic splicing boxes and switches. The switches use industrial-grade products, which have redundant power supplies, hot backup functions, and powerful VLAN (Virtual Local Area Network) partitioning capabilities. The data of different types of sensors are partitioned into different VLANs to reduce network data conflicts and improve the security of data transmission. Each node in the network (including sensors, data acquisition terminals, and central data processing units) is assigned a unique IP address, and strict network access permission settings are made, and only authorized devices can access and transmit data.

[0067] Specifically, when configuring a large-capacity data storage device, an enterprise-level disk array is selected, and RAID (Redundant Array of Independent Disks) technology, such as RAID5 or RAID6, is adopted to improve the reliability and read / write speed of data storage. The storage capacity of the disk array is estimated according to the expected duration of the shield machine tunneling, the data acquisition frequency, and the data volume size. Generally, enough storage space is reserved to meet the continuous data storage requirements for at least several months. The data storage management system uses specialized database management software, such as MySQL or Oracle, to establish structured data storage tables and classify and store data according to multiple dimensions such as time series (accurate to the millisecond level), sensor type, and shield machine location. The data storage format adopts the standardized HDF5 (Hierarchical Data Format 5) format, which supports the efficient storage and management of large-scale complex data and can facilitate data compression, indexing, and query operations. At the same time, a data backup mechanism is established, and a combination of off-site backup and regular full and incremental backups is adopted to back up the data to external storage media (such as tape libraries or cloud storage) to ensure the security and integrity of the data.

[0068] It is understandable that planning the installation layout of sensors and calibrating them precisely can ensure that sensors accurately collect various key data of the rock and soil mass during shield tunneling, such as pressure, vibration, laser scanning point cloud, etc., providing a rich and reliable information source for accurately analyzing the movement trajectory of rock and soil particles; building an industrial Ethernet data transmission network can achieve high-speed and stable data transmission, ensuring the timeliness and integrity of data, and avoiding interference with trajectory analysis caused by data loss or delay; a large-capacity storage device can properly store a large amount of data, which is not only convenient for subsequent data backtracking analysis and processing, but also provides a solid guarantee for the accumulation of shield tunneling construction data over a long time and on a large scale, thus comprehensively improving the accuracy, continuity and reliability of obtaining the movement trajectory of rock and soil particles, and effectively promoting the efficient and safe development of shield tunneling projects.

[0069] Specifically, when using the statistical filtering algorithm to remove outliers from the laser scanning point cloud data, first calculate the average distance between each point in the point cloud data and its neighboring points (for example, take the points within a radius of 0.1 meters as neighboring points). Then, according to the overall distribution characteristics of the point cloud data, set a distance threshold (such as 3 times the standard deviation of the average distance), and regard the points whose distance exceeds this threshold as outliers and remove them from the data set. To improve the filtering effect, the filtering algorithm can be iteratively executed multiple times until the number of outliers in the remaining point cloud data meets the requirements (such as the proportion of the total number of points is less than 0.1%).

[0070] Specifically, when using the pass-through filtering algorithm to remove the reflected point cloud generated by the internal equipment of the shield machine according to the coordinate range of the point cloud, obtain the spatial coordinate range of the internal equipment (such as the minimum bounding box coordinates of equipment such as motors and hydraulic pipelines) based on the CAD model of the shield machine, and then set the range parameters of the pass-through filtering to filter out the point cloud data located within the coordinate range of these equipment. For some equipment reflected points that are difficult to accurately filter by the coordinate range, further screening is carried out in combination with the intensity information of the point cloud. Since the intensity of the equipment reflected points is usually significantly different from that of the rock and soil mass point cloud, by setting an intensity threshold (such as determining the average value and standard deviation of the intensity of the equipment reflected points according to experimental data, and taking the points whose intensity values are outside the range of the average value ± 2 standard deviations as suspected equipment reflected points for removal), most of the equipment reflected point cloud can be effectively removed, and the point cloud data within the spatial range where the rock and soil mass is located can be retained.

[0071] When using a density-based clustering algorithm (such as DBSCAN) to segment the point cloud into different clusters, the parameters of the DBSCAN algorithm (such as the neighborhood radius and the minimum number of points) are set according to the expected density range of the geotechnical particle point cloud. After the point cloud is segmented into multiple clusters by the clustering algorithm, according to the characteristics of the clusters (such as volume, number of points), shape (such as sphericity, ratio of major axis length to minor axis length), and point density, etc., the point cloud clusters belonging to a single geotechnical particle are further screened. Remove some clusters that are too large or too small. The too-large clusters may be structures that are not completely filtered inside the shield machine or large blocks formed by the aggregation of multiple particles, and the too-small clusters may be noise or fragmented small particle debris.

[0072] Specifically, for the extracted geotechnical particle point cloud data, when calculating its detailed geometric characteristics, in addition to the minimum bounding box size, sphericity, and surface curvature, it also includes characteristics such as the ratio of major axis length to minor axis length of the particle and surface concavity and convexity. When using the principal component analysis (PCA) algorithm to determine the principal direction of the particle, first construct the covariance matrix of the point cloud data, and then calculate the eigenvalues and eigenvectors of the covariance matrix. The eigenvectors correspond to the principal direction of the point cloud data. By projecting the point cloud data onto the principal direction, the size information of the particle in different directions can be obtained, and then characteristics such as the ratio of major axis length to minor axis length can be calculated. The surface concavity and convexity characteristics can be determined by calculating the curvature change rate of the point cloud surface.

[0073] Specifically, when combining image analysis technology to extract the color characteristics of particles from the images synchronously collected with the point cloud data, the color images are converted into different color spaces (such as RGB, HSV, etc.) for analysis. In the RGB color space, calculate the color histogram of the particle region, and statistically analyze the pixel value distribution of each color channel (red, green, blue) to obtain characteristics such as color mean and variance. In the HSV color space, focus on analyzing the characteristics of the hue (H) and saturation (S) channels because these two channels are more sensitive to the color differences of geotechnical particles. The texture feature extraction uses the gray-level co-occurrence matrix (GLCM) algorithm. First, convert the color image into a gray image, and then calculate the gray-level co-occurrence matrices in different directions (such as 0°, 45°, 90°, 135°) and different distances (such as 1 pixel, 2 pixels, etc.). Extract texture parameters such as contrast, correlation, energy, and entropy from the gray-level co-occurrence matrix. These parameters can reflect the roughness, directionality, and regularity of the particle surface texture and other characteristics. When combining these geometric characteristics and image characteristics into a feature vector, reasonable combination and normalization processing are performed according to the importance of the features and the relevance of the data.

[0074] Specifically, when constructing a classification model using the Support Vector Machine (SVM) algorithm, the model is first trained using training data. During the training process, appropriate kernel functions (such as linear kernel, polynomial kernel, radial basis function kernel, etc.) and penalty parameter C are selected through the method of cross-validation. The cross-validation adopts k-fold cross-validation (such as k = 10), divides the training data into k subsets, each time uses k - 1 subsets as the training set, and the remaining one subset as the validation set, repeats k times, calculates the classification accuracy rate on each validation set, and selects the combination of kernel function and penalty parameter with the highest accuracy rate. When constructing a model using the random forest algorithm, parameters such as the number of decision trees (such as 100 to 500 trees) and the maximum depth of the tree (such as 5 to 15 layers) are determined. The parameters are adjusted by evaluating the Out-of-Bag Error. The Out-of-Bag Error refers to the difference between the prediction result of a sample on a certain decision tree and the true label when the sample is not used to construct that decision tree during the construction of the random forest. By continuously adjusting the number and depth of the decision trees and observing the change of the Out-of-Bag Error, the parameter combination that minimizes the Out-of-Bag Error is selected. After the training is completed, the model is evaluated using the test data, and indicators such as the accuracy rate, recall rate, and F1 value of the model are calculated. The accuracy rate refers to the proportion of the number of correctly classified samples to the total number of samples, the recall rate refers to the proportion of the number of correctly classified positive samples to the actual number of positive samples, and the F1 value is an evaluation indicator that comprehensively considers the accuracy rate and the recall rate. The calculation formula is F1 = 2 * (accuracy rate * recall rate) / (accuracy rate + recall rate). Ensure that these indicators of the model on the test set reach a relatively high level (such as accuracy rate > 90%, recall rate > 85%, F1 value > 0.88) to meet the reliability requirements of practical applications.

[0075] It can be understood that by preprocessing the laser scanning point cloud data using a variety of filtering algorithms, noise interferences such as outliers and device reflection points can be effectively removed, and the data can be purified to highlight the information of rock and soil particles; extracting and combining geometric and image features to construct a classification model can comprehensively capture multi-dimensional characteristics of the particles such as shape, size, color, and texture, realizing the accurate identification and classification of rock and soil particles. This not only lays a solid foundation for accurately tracking the movement trajectories of individual particles subsequently, but also can deeply analyze the differences in their movement laws based on the particle types, and further provides valuable basis and precise guidance for parameter optimization, risk prediction, and quality control in shield tunneling construction, effectively ensuring the efficient and safe progress of the project.

[0076] Specifically, when preprocessing the image sequence in the initial stage of shield tunneling, in addition to adjusting the brightness, contrast, etc., color balance adjustment is also performed during image enhancement to reduce color deviation caused by different light sources inside the shield machine (such as incandescent lamps, LED lamps, etc.). The adaptive histogram equalization algorithm is used to adjust the brightness of the image, making the brightness distribution of the image more uniform and improving the overall contrast of the image. The image cropping operation is precisely cropped according to the position and size of the key areas inside the shield machine (such as the cutter head area, the soil outlet area, etc.), removing the irrelevant background areas in the image (such as the non-critical structural parts of the shield machine, some areas of the tunnel wall, etc.), reducing the data volume and highlighting the area where the rock and soil particles are located. When using the blind deconvolution algorithm for image deblurring, the blur function is automatically estimated according to the blur characteristics of the image (such as the size and direction of the blur kernel), and the image is deconvolved to restore the clarity of the image. When converting the image to a grayscale image, the weighted average method is used to convert the RGB three channels of the color image into grayscale values.

[0077] Specifically, when annotating the rock and soil particles in the image, for manual annotation, first, experienced engineers manually annotate some key image frames (such as selecting 1 frame every 10 frames). During the annotation process, professional image annotation tools such as Label Img or VGGImageAnnotator (VIA) are used to mark the precise position of the particles (represented by the upper left and lower right coordinates of the bounding box) and the size of the bounding box. For semi-automatic annotation, a target detection model (such as the deep learning-based FasterR-CNN model) is trained using the annotated image frames. During the training process, a pre-trained convolutional neural network (such as the ResNet network pre-trained on the ImageNet dataset) is used as the basic network and fine-tuned to adapt to the detection task of rock and soil particles. The model is applied to the unannotated image frames to automatically generate the initial annotation information of the particles, and then manually reviewed and corrected to ensure the accuracy of the annotation data. In addition to the position and the bounding box, the annotation information also includes the class label of the particles (according to the classification results in step two). Different types of rock and soil particles (such as sand grains, gravels, clay particles, etc.) are respectively marked as different classes so that the subsequent algorithm training can distinguish the movement characteristics of different types of particles.

[0078] Specifically, when dividing the labeled data into a training set, a validation set, and a test set at a ratio of 70%, 20%, and 10%, a stratified sampling method is adopted to ensure that the proportions of various types of geotechnical particles in the training set, validation set, and test set are roughly the same as those in the original data, so as to avoid the impact of unbalanced data distribution on model training and evaluation. When using the training set to train the algorithm that combines the optical flow method and the convolutional neural network, in the convolutional neural network part, different network architectures are tried, such as ResNet, VGG, etc., and the optimal architecture is selected according to the accuracy on the validation set. For the ResNet network, different numbers of layers can be tried (such as ResNet-18, ResNet-34, ResNet-50, etc.), and their performance on the validation set can be compared. When adjusting the hyperparameters of the network, the stride of the convolutional layer can be set to 1 or 2, the size of the pooling layer can be selected as 2x2 or 3x3, and the number of neurons in the fully connected layer is adjusted according to the number of particle categories and data characteristics (such as set to 128, 256, or 512, etc.). In terms of optimizing the algorithm parameters, an adaptive learning rate adjustment strategy is adopted, such as learning rate decay (gradually decreasing the learning rate as the number of training steps increases), and appropriate initial learning rate (such as 0.001) and decay factor (such as 0.9) are set. The number of iterations is determined according to the convergence of the model on the validation set, generally set to 100 - 500 times. By observing the change of the loss function value (such as cross-entropy loss) on the validation set, when the loss function value no longer decreases significantly within a continuous number of iteration steps (such as 10 - 20 steps), it is considered that the model has converged and the training can be stopped. When using the L1 / L2 regularization technique to prevent overfitting, the regularization coefficient is adjusted (such as the L2 regularization coefficient is set to 0.01), and the impact of regularization on the accuracy and generalization ability of the model on the validation set is observed, and an appropriate regularization coefficient is selected.

[0079] Specifically, finally, the test set is used to evaluate the performance of the trained algorithm, and indicators such as the mean average precision (mAP) and intersection over union (IoU) are calculated. The mean average precision is a comprehensive indicator to measure the performance of the object detection algorithm, which comprehensively considers the detection accuracy and recall rate of different types of particles. The intersection over union is an indicator to measure the overlapping degree between the predicted bounding box and the true bounding box. Generally, it is required that IoU > 0.5 to consider the detection result to be correct. If the indicators do not meet the expected requirements, the reasons are further analyzed, such as inaccurate data annotation, unreasonable algorithm parameter settings, etc., and targeted adjustments are made and retraining is carried out until the performance of the algorithm on the test set meets the reliability requirements of practical applications.

[0080] It is understandable that preprocessing and annotating the initial image sequence of shield tunneling can effectively improve the image quality, highlight the rock and soil particle regions, and accurately define their positions and categories, providing a high-quality data basis for algorithm training. Dividing the dataset to train the algorithm combining optical flow method and convolutional neural network can fully integrate the advantages of both, accurately capturing the particle motion characteristics. Optimizing the parameters can further improve the accuracy and adaptability of the algorithm, and evaluating the performance ensures the reliability of the algorithm. Thus, initializing the unmarked motion tracking algorithm creates a good start for subsequent real-time and accurate tracking of the rock and soil particle motion trajectories, greatly enhancing the effectiveness and stability of the entire trajectory acquisition method, and effectively guaranteeing the smooth progress and precise control of shield tunneling construction.

[0081] Specifically, when the data acquisition system collects laser scan data, in addition to checking the data integrity, it also verifies the spatial consistency of the point cloud data. By comparing the coordinate change ranges of the same feature points in adjacent scan frames, if it exceeds the preset threshold (such as 0.05 meters), it indicates that the data may be abnormal and is marked. For the clarity check of image data, in addition to the evaluation based on image contrast and edge sharpness, an image spectrum analysis method is also used. Calculate the Fourier spectrum of the image. If the high-frequency components are missing or abnormally attenuated, it indicates that the image may be blurred and is marked when it is below the set clarity threshold. When the data is marked as unqualified in quality, the supplementary acquisition process will be carried out no more than 0.5 seconds after the original acquisition time point to ensure the timeliness of the data as much as possible. The maximum number of supplementary acquisitions is 3 times. If it is still unqualified, an alarm will be issued to prompt manual intervention to check the equipment.

[0082] Specifically, the data compression adopts a hybrid compression algorithm. First, the laser scan data is compressed based on geometric features. For example, the adjacent and similar points are merged and encoded using the spatial distribution law of the point cloud. For image data, first, region segmentation is carried out, and different compression ratios are used for the background region and the particle region respectively. The background region uses a high compression ratio (such as 20:1), and the particle region uses a low compression ratio (such as 5:1) to retain the key information of the particles while reducing the data volume. The data encryption adopts the AES-256 encryption algorithm to encrypt the compressed data. The encryption key is dynamically generated and updated every 10 minutes. The key seed is generated based on information such as the running time and current position of the shield machine to ensure the security of data transmission.

[0083] Specifically, when calculating the displacement vector of geotechnical particles between adjacent frames, the optical flow method uses the pyramid-based Lucas-Kanade algorithm. First, an image pyramid is constructed, and the optical flow is calculated starting from the top layer and gradually refined to the lower layers to improve the calculation accuracy and handle large displacement situations. When determining the three-dimensional position change of particles by combining laser scanning data, the displacement vector obtained from the image calculation is calibrated using the high-precision spatial coordinate information of the laser scanning. For example, by matching the particle feature points identified in the image with the corresponding points in the laser scanning point cloud, a coordinate transformation relationship is established to convert the displacement in the image plane to the three-dimensional space displacement, and then the movement speed and direction of the particles are calculated. The speed calculation is accurate to 0.01 m / s, and the direction calculation accuracy is within ±2°.

[0084] Specifically, when the pressure sensor detects abnormal pressure, the position of the pressure change is located through the installation position information of the sensor, and the abnormal degree is determined by comparing the pressure amplitude with the normal tunneling pressure range. The correction coefficient is determined using a linear or non-linear mapping relationship according to the abnormal degree of the pressure. For example, if the abnormal pressure amplitude is between 10% and 20%, the correction coefficient is set to 0.8 - 0.9, and the estimated value of the particle movement speed in the corresponding area is adjusted by a corresponding proportion. The direction adjustment is based on the pressure gradient direction, and the pressure gradient direction is determined by calculating the pressure difference between adjacent sensors in the pressure sensor array, causing the particle movement direction to shift towards the lower pressure area.

[0085] Specifically, when processing the vibration sensor data, a time-frequency analysis method (such as the short-time Fourier transform) is used to convert the vibration signal to the time-frequency domain to analyze the frequency components and amplitude changes of the vibration. If the energy of high-frequency vibration (such as 100 Hz - 1000 Hz) exceeds 50% of the normal threshold, it is judged that the cutter head cutting is uneven, and a random perturbation of the particle movement direction is increased. The perturbation amplitude is determined according to the high-frequency vibration energy and is between 5° and 15°. If the amplitude of low-frequency vibration (such as 0 - 10 Hz) exceeds 30% of the normal threshold, it indicates that the overall advancement of the shield machine is blocked, and the estimated value of the particle movement speed inside the entire shield machine is reduced by 20% - 30% to reflect the slowdown of particle movement when the shield machine has difficulty advancing.

[0086] It can be understood that when data is collected and processed at time intervals during shield tunneling, the state information of the geotechnical body at different tunneling stages can be obtained in a timely manner, providing continuous and dynamic data support for accurately calculating the movement trajectory of geotechnical particles; correcting the movement trajectory based on the pressure and vibration sensor data can fully consider the complex mechanical environment changes inside the shield machine, effectively respond to the changes in the particle movement state caused by factors such as abnormal pressure and vibration fluctuations, making the calculated movement trajectory more in line with the actual situation, greatly improving the accuracy and reliability of obtaining the movement trajectory of geotechnical particles, and thus providing a strong basis and guarantee for parameter adjustment, risk warning, and quality control during the shield tunneling construction process.

[0087] Specifically, when organizing the data of the geological exploration report, in addition to statistical distribution information, the correlation between the geotechnical parameters of different strata is also analyzed. For example, there may be a specific functional relationship between the internal friction angle and the particle gradation and water content of the soil and rock. This relationship is determined through regression analysis and incorporated into the prior information. For the records of the previous working parameters of the shield machine, not only the basic parameters are recorded, but also the abnormal events during the construction process (such as cutter jamming, gushing, etc.) and the parameter settings at that time are recorded, so as to consider the influence of these special situations on parameter estimation in the model construction.

[0088] Specifically, when constructing the prior probability distribution function of the Bayesian estimation model, for complex geotechnical parameters (such as geomechanical parameters with multiple influencing factors), a mixed probability distribution model is adopted. For example, for the internal friction angle, in some strata, the characteristics of normal distribution and uniform distribution may exist simultaneously. The distribution parameters are determined separately according to different strata regions or depth ranges, and the prior distribution is constructed in a weighted combination manner. The weight of the prior distribution is determined according to the reliability of the geological exploration data and the uniformity of the strata. For example, for the regions with detailed geological exploration data and good strata uniformity, the weight of the corresponding distribution is relatively high (such as 0.7 - 0.8).

[0089] Specifically, when the sensor data is input into the Bayesian estimation model as the observed value, the data is preprocessed. The pressure data is filtered to remove measurement noise. The Kalman filter algorithm is used, and the filter parameters are set according to the dynamic characteristics of the pressure change, such as the state transition matrix and the measurement noise covariance matrix, to improve the accuracy of the pressure data. The vibration data is subjected to feature extraction. In addition to frequency and amplitude, features such as the peak factor and kurtosis of the vibration are also extracted. An association model is established between these features and the geotechnical parameters and the working parameters of the shield machine. For example, there may be a certain relationship between the peak factor of the vibration and the hardness of the soil and rock and the wear degree of the shield machine cutter. This relationship is determined through experimental data and historical experience and incorporated into the Bayesian estimation model as supplementary information.

[0090] Specifically, the model update adopts the Gibbs sampling algorithm in the Markov chain Monte Carlo (MCMC) method. A sufficient number of sampling times (such as 10,000 - 50,000 times) is set to ensure the convergence of the posterior probability distribution. After each sampling, the confidence interval of the parameter estimation value is calculated, and the confidence level is set to 95%. If the new sampling value exceeds the confidence interval, the model parameters or the prior distribution are adjusted. For example, the parameter range or weight of the prior distribution is adjusted, and sampling is carried out again until the parameter estimation value is stable within the confidence interval and the posterior probability distribution no longer changes significantly.

[0091] It is understandable that organizing the historical data of geological exploration and shield machines and constructing a prior probability distribution function can make full use of past experience and formation information to provide a reasonable initial speculation framework for the estimation of uncertain parameters. Real-time collection of sensor data and input into the Bayesian estimation model can incorporate the dynamic information during actual tunneling. By updating the parameters and posterior probability distribution, uncertain parameters such as geotechnical mechanical parameters can be accurately estimated. This can not only effectively make up for the information gaps caused by geological complexity and construction uncertainty, but also provide a key basis for more accurately calculating the movement trajectories of geotechnical particles, greatly enhancing the scientific nature and reliability of the entire trajectory acquisition method, and effectively ensuring the safe and efficient progress of shield tunneling construction.

[0092] Specifically, during finite element modeling, solid elements are used to finely model the structures such as the cutter head and shield of the shield machine, considering the anisotropic characteristics of the materials. For example, the elastic modulus and Poisson's ratio of the steel of the cutter head may vary in different directions, and these parameters are accurately set according to the material test data. For the cutters on the cutter head, contact elements are used to simulate their interaction with the geotechnical body, and the friction coefficient of the contact elements is determined according to the actual friction test results of the cutter material and geotechnical type, accurate to 0.01.

[0093] Specifically, during discrete element modeling, the mechanical parameters of the geotechnical particles are set according to the classification results of step two and the real-time estimated parameters. The Hertz-Mindlin contact model is used for the contact model between particles, and the rolling friction of the particles is considered. For example, different stiffness coefficients, friction coefficients, and rolling friction coefficients are set for sand grains and gravel particles respectively, and these coefficients are determined according to the results of indoor geotechnical mechanical tests and dynamically adjusted according to the real-time estimated parameters during shield tunneling.

[0094] Specifically, when setting the boundary conditions, the propulsion force of the shield machine is dynamically adjusted according to the set value of the actual construction and the real-time monitored propulsion speed. For example, if the propulsion speed is lower than expected, the propulsion force is appropriately increased, but the increase amplitude of the propulsion force does not exceed 10% of the rated value. The constraint conditions of the surrounding geotechnical body are simulated using spring-damper elements, and the spring and damping coefficients are determined according to the elastic modulus and damping ratio of the formation. The elastic modulus and damping ratio are adjusted according to the geological exploration data and real-time estimated parameters to accurately reflect the constraint effect of the formation on the shield machine. In the initial condition setting, the initial stress state of the geotechnical body is determined according to the in-situ stress measurement data, and the particle distribution is initialized according to the initial geotechnical body morphology obtained by laser scanning to ensure that the initial state of the simulation is consistent with the actual situation.

[0095] Specifically, when calculating the distance deviation and velocity deviation between trajectory points, a spatial vector calculation method is adopted, taking into account the three-dimensional characteristics of particle motion. In addition to the Euclidean distance, the deviation components of the trajectory points in different coordinate axis directions are also calculated to analyze the source of deviation in more detail. The velocity deviation is calculated by comparing the instantaneous velocity and the average velocity. The instantaneous velocity and the average velocity of the simulated trajectory and the measured trajectory are calculated respectively within the same time period, and the differences between the two are compared.

[0096] Specifically, when the deviation exceeds the threshold range, during the process of checking for abnormal sensor data, in addition to checking for data loss and fault alarms, data rationality checks are also carried out. For example, the change rate of sensor data in adjacent time periods is compared. If the change rate exceeds the normal range (such as the pressure change rate exceeds 1 MPa / s, and the vibration acceleration change rate exceeds 10 m / s 2 / s), it is considered that the data may be abnormal. For the analysis of algorithm problems, a sensitivity analysis method is adopted. The key parameters in the algorithm (such as the window size in the optical flow method, the threshold in the convolutional neural network, etc.) are changed one by one, and the influence on the trajectory calculation result is observed to determine the key parameters causing the deviation and optimize and adjust them. For the inspection of the assumed conditions of the numerical simulation model, in addition to the constitutive model and the contact model, the scale effect of the model is also checked. For example, whether the ratio of the particle size in the discrete element model to the actual geotechnical particle size is appropriate. If the scale effect is obvious, the particle size is adjusted or a multi-scale modeling method is used for correction, and then the simulation and comparison verification are carried out again until the deviation between the simulated trajectory and the measured trajectory is within the acceptable range. The acceptable standard for the deviation is that the average value of the distance deviation is less than 0.05 meters, and the average value of the velocity deviation is less than 0.03 meters per second.

[0097] It can be understood that by constructing a numerical simulation model and setting parameters based on the shield machine and the formation conditions, the motion trajectory of geotechnical particles can be simulated at the theoretical level, and the construction conditions can be predicted in advance. By comparing the simulated and measured motion trajectories, the differences between the two can be intuitively found. When the deviation exceeds the threshold, after checking and optimizing the relevant links such as sensor data, algorithms, and the assumed conditions of the numerical simulation model, and then re-simulating and verifying, it helps to accurately locate the root cause of the problem and correct it in time, continuously improving the accuracy of obtaining the motion trajectory of geotechnical particles, providing reliable data reference and technical support for shield tunneling construction, effectively reducing construction risks, and improving project quality and efficiency.

[0098] Specifically, the 3D graphics rendering engine adopts physically based rendering (PBR) technology, which can realistically simulate the reflection, refraction, and scattering effects of light on the internal surface of the shield machine and the surface of rock and soil particles, improving the realism of visualization. In the internal view of the shield machine, in addition to showing the movement trajectories of rock and soil particles, information such as the temperature distribution and stress distribution of the internal structure of the shield machine is also displayed, and this information is updated and displayed by connecting to sensor data in real time. For example, the color mapping method is used to map the temperature data to the surface color of the shield machine structure, where red represents the high-temperature area and blue represents the low-temperature area, facilitating users to intuitively understand the working state inside the shield machine.

[0099] Specifically, in the classification display function, when displaying by particle type, different colors and shapes are used to represent different types of particles. When displaying by the range of movement speeds, speed intervals are set, and particles in different speed intervals are represented by different transparencies or blinking frequencies, so that users can quickly distinguish particle groups with different speeds. In terms of the interaction function, after the user clicks on a specific particle, in addition to displaying the movement information, the detailed physical properties of the particle are also displayed, and these property information are obtained and displayed based on the classification and recognition results in Step 2.

[0100] Specifically, in the basic information section, in addition to the conventional information, information such as the construction team information of the shield machine, the construction environment temperature and humidity, etc. are also recorded to comprehensively understand the construction background. In the summary of the movement characteristics of rock and soil particles, in addition to the average movement speed, movement direction distribution, and residence time distribution, the acceleration change of particle movement is also analyzed. The acceleration is obtained by calculating the second derivative of the trajectory data, and the magnitude and direction changes of the acceleration are analyzed to understand the force change of the particles inside the shield machine.

[0101] Specifically, in the part of evaluating the influence of parameter uncertainty, the sensitivity analysis method is used to quantify the influence degree of different parameter uncertainties on the particle movement trajectory. For example, the sensitivity coefficient of each parameter is calculated. The larger the sensitivity coefficient, the greater the influence of the uncertainty of this parameter on the trajectory. The parameters are sorted according to the size of the sensitivity coefficient, and key attention is paid to the parameters with higher sensitivity coefficients, and their changes in different construction stages and the influence on the working state of the shield machine are analyzed.

[0102] Specifically, in the part of analyzing the working state of the shield machine, in addition to the cutting efficiency and the smoothness of soil discharge, the energy consumption of the shield machine is also analyzed. According to the power monitoring data of the shield machine, the energy consumption per unit tunneling distance is calculated, and the relationship between the energy consumption and the particle motion state is analyzed in combination with the particle motion trajectory analysis. For example, if the particle motion is blocked, it may lead to an increase in the energy consumption of the shield machine. Through this relationship analysis, suggestions for energy-saving construction are provided. The automation part of the report generation uses a scripting language (such as Python) to write an automated report generation program, which automatically fills the content of the report template according to the data processing and analysis results to generate a preliminary report. The manual review part is reviewed by at least two professional engineers. The review content includes the accuracy of the data, the rationality of the analysis method, the feasibility of the suggestions, etc. After passing the review, they sign and confirm. The final report is archived in two forms: electronic document and paper document. The electronic document is in PDF format, which is convenient for sharing and transmission. The paper document is archived and stored according to the construction project and time sequence.

[0103] Specifically, the multi-objective adaptive optimization control model is constructed based on the framework of model predictive control (MPC). When determining the input variables, in addition to the conventional variables, the influence of the attitude parameters of the shield machine (such as pitch angle, yaw angle) on the particle motion and construction effect is also considered. For example, an excessive pitch angle of the shield machine may cause particles to accumulate forward or backward inside the shield machine, affecting the smoothness of soil discharge. The pitch angle is incorporated into the input variables for adjustment in the optimization control.

[0104] Specifically, the objective function is set in the form of a weighted sum of multiple sub-objective functions. For example, the sub-objective function for controlling the ground settlement is constructed based on the relationship between the ground settlement and the working parameters of the shield machine obtained from empirical formulas or numerical simulations. Limiting the ground settlement within the allowable range (such as ±10 mm) is used as a constraint condition. The sub-objective function of the tool wear rate is constructed based on the relationship between the wear characteristics of the tool material and the working parameters of the shield machine (such as cutter head rotation speed, thrust force, particle hardness, etc.). Controlling the tool wear rate within an acceptable level (such as the wear amount does not exceed 1 mm per 100 meters of tunneling) is used as a constraint condition. Through a large number of numerical simulations and actual engineering experience data, the analytic hierarchy process (AHP) is used to determine the weight coefficients of the objective function. For example, in areas with strict requirements for ground settlement, the weight coefficient of the objective function for controlling the ground settlement is set to 0.6 - 0.7, and the weight coefficient of the objective function of the tool wear rate is set to 0.3 - 0.4; in strata where the tool wear problem is more prominent, the weight coefficients are adjusted accordingly to balance the relationship between different objectives.

[0105] Specifically, when the optimization control model calculates the optimal working parameter adjustment scheme based on real-time data, an efficient optimization algorithm (such as the interior point method or sequential quadratic programming method) is used for solution. During the solution process, the adjustment limitations of the shield machine's working parameters are considered. For example, the adjustment range of the cutter head rotation speed cannot exceed ±20% of the rated speed, and the adjustment range of the propulsion speed cannot exceed ±30% of the designed speed, etc. When feeding the adjusted parameters back to the shield machine control system, an industrial communication protocol (such as Profibus-DP or Modbus TCP) is used for data transmission to ensure the real-time and accuracy of data transmission. After receiving the parameter instructions, the control system completes the parameter adjustment and executes new operations within 0.1 - 0.5 seconds.

[0106] Specifically, at the same time, the adjusted parameters and control effect information are fed back to the motion trajectory acquisition system, and the motion trajectory acquisition system further optimizes the data acquisition frequency and range based on this information. For example, if it is found that the particle motion state changes violently in a certain area, increase the laser scanning data acquisition frequency in this area (such as increasing from once every 0.1 seconds to once every 0.05 seconds) and the image acquisition range (such as expanding the camera shooting angle by 10° - 20°). In terms of algorithm parameter adjustment, the parameters in the motion tracking algorithm are optimized according to the control effect information. For example, if it is found that the calculation deviation of the particle motion trajectory is relatively large in some cases, adjust the parameters in the optical flow method (such as window size, number of iterations, etc.) or the weight parameters in the convolutional neural network to improve the accuracy of the motion trajectory calculation, realize the collaborative work and closed-loop control of the entire system, and continuously improve the efficiency and quality of shield tunneling construction.

[0107] It can be understood that constructing a multi-objective adaptive optimization control model based on the model predictive control framework can comprehensively consider various factors, accurately calculate and feedback the shield machine working parameter adjustment scheme based on real-time data, and make the shield machine operation adapt to the movement of rock and soil particles; realizing closed-loop control by coordinating all links of the system can not only effectively reduce ground settlement and reduce tool wear, but also improve the shield tunneling speed, thereby comprehensively improving the construction efficiency and quality, providing a stable and optimized construction environment for obtaining the motion trajectory of rock and soil particles during shield tunneling, ensuring the accuracy and continuity of trajectory acquisition, and promoting the efficient, safe and orderly development of the entire project.

[0108] In the above embodiment, a method for obtaining the motion trajectory of rock and soil particles during shield tunneling first prepares for multi-source data collection and fusion to lay a data foundation, and then sequentially goes through links such as rock and soil particle identification and classification, initialization of the motion tracking algorithm, etc., and uses various means to realize functions such as particle motion tracking, trajectory calculation correction, parameter estimation, etc. Through simulation comparison verification, visualization and adaptive optimization feedback control, the construction efficiency and quality are ultimately improved, which helps construction decision-making, ensures construction and is conducive to data management.

[0109] Another preferred method based on the above embodiments is as follows. Refer to Figure 2 As shown, this embodiment provides a system for obtaining the movement trajectory of rock and soil particles during shield tunneling, including:

[0110] Multi-source data acquisition and fusion preparation module: Plan the installation layout of sensors and calibrate them precisely, and build an industrial Ethernet data transmission network and a large-capacity storage device;

[0111] Rock and soil particle preliminary identification and classification module: Use a variety of filtering algorithms to preprocess the laser scanning point cloud data, extract and combine geometric and image features to build a classification model, and accurately identify and classify rock and soil particles;

[0112] Unmarked motion tracking algorithm initialization module: Preprocess and annotate the initial image sequence of shield tunneling, divide the data set to train the algorithm combining optical flow method and convolutional neural network, optimize the parameters and evaluate the performance to initialize the unmarked motion tracking algorithm;

[0113] Real-time motion trajectory calculation and correction module: Collect and process data at time intervals during shield tunneling, calculate the movement trajectory of rock and soil particles and correct it according to the data of pressure and vibration sensors to adapt to the change of the working state of the shield machine;

[0114] Uncertainty parameter real-time estimation module: Sort out the geological exploration and historical data of the shield machine and build a prior probability distribution function, collect sensor data in real time and input it into the Bayesian estimation model, update the parameters and posterior probability distribution to estimate the uncertainty parameters;

[0115] Simulation and measured data comparison and verification module: Build a numerical simulation model according to the shield machine and stratum conditions and set parameters, compare the simulated and measured motion trajectories, and when the deviation exceeds the threshold, check and optimize the relevant links and then re-simulate and verify;

[0116] Motion trajectory visualization and analysis report generation module: Develop a visualization platform to display the movement trajectory of rock and soil particles, generate an analysis report containing multi-faceted information, and archive it after automatic and manual review to ensure accuracy;

[0117] Adaptive optimization and feedback control module: Build a multi-objective adaptive optimization control model based on the model predictive control framework, calculate and feedback the shield machine working parameter adjustment plan according to the real-time data, and cooperate with each link of the system to achieve closed-loop control to improve the construction efficiency and quality.

[0118] It can be understood that the above system and method for obtaining the movement trajectory of rock and soil particles during shield tunneling have the same beneficial effects, which will not be elaborated here.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for obtaining the motion trajectory of rock and soil particles during shield tunneling, characterized in that: include: Plan the installation layout of sensors and accurately calibrate them, build industrial Ethernet data transmission networks and large-capacity storage devices; Use a variety of filtering algorithms to preprocess laser scanning point cloud data, extract and combine geometric and image features to build a classification model, and accurately identify and classify rock and soil particles; Preprocess and annotate the initial image sequence of shield tunneling, divide the data set to train the algorithm combining optical flow method and convolutional neural network, optimize parameters and evaluate performance to initialize the markerless motion tracking algorithm; During shield tunneling, data is collected and processed at time intervals, the movement trajectory of rock and soil particles is calculated and corrected based on the pressure and vibration sensor data to adapt to changes in the working state of the shield machine; Organize geological survey and shield machine historical data and construct a priori probability distribution function, collect sensor data in real time and input it into the Bayesian estimation model, update parameters and posterior probability distribution to estimate uncertainty parameters; Construct a numerical simulation model and set parameters based on the shield machine and stratum conditions, compare the simulated and measured motion trajectories, and check and optimize related links and re-simulate and verify when the deviation exceeds the threshold; Develop a visualization platform to display the movement trajectory of rock and soil particles, generate analysis reports containing various information, and archive them after automatic and manual review to ensure accuracy; A multi-objective adaptive optimization control model is constructed based on the model predictive control framework. The shield machine working parameter adjustment plan is calculated and fed back according to real-time data. The coordinated system links realize closed-loop control to improve construction efficiency and quality.

2. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The planning of the sensor installation layout and accurate calibration, the construction of an industrial Ethernet data transmission network and a large-capacity storage device, include: Sensor installation and calibration: Determine the optimal installation layout of the sensor during the shield machine design phase. For laser scanning sensors, use computer-aided design software to simulate the installation position analysis based on the internal space geometry of the shield machine and the possible distribution range of the rock and soil, to ensure that the scanning angle and range of the sensor can cover the key area without obvious blind spots. The laser scanning sensor scans the corresponding area and collects point cloud data. When installing the pressure sensor, determine the force concentration point of the cutterhead tool and the key monitoring points of the circumferential and longitudinal stress of the shield wall based on the mechanical structure analysis results of the cutterhead and the shield wall for embedded installation. After installation, use a pressure calibrator to perform static and dynamic calibration on each pressure sensor. The vibration sensor is installed on the main drive system of the shield machine, the key parts of the shield shell, and the drive device and shell of the screw conveyor, and is connected by welding or high-strength bolts. After installation, it is calibrated on the vibration test platform; Data transmission and storage equipment are set up to build a high-speed data transmission network based on industrial Ethernet. The network topology adopts a redundant ring structure. Fiber optic cables are laid inside the shield machine along the axis of the tunnel and connected through fiber optic junction boxes and switches. The data of different types of sensors are divided into different virtual LANs. Each node in the network is assigned a unique IP address and strict network access rights are set. Only authorized devices can access and transmit data. Large-capacity data storage devices are configured using independent disk redundant array technology. The storage capacity of the disk array is estimated based on the expected duration of the shield machine excavation, data collection frequency and data volume. Sufficient storage space is reserved to meet the continuous data storage needs of at least several months. A structured data storage table is established, and data is classified and stored according to multiple dimensions such as time series, sensor type, and shield machine location. The data storage format adopts the standardized HDF5 format. A data backup mechanism is established, and data is backed up to external storage media by a combination of off-site backup and regular full and incremental backup.

3. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The aforementioned use of multiple filtering algorithms to pre-process laser scanning point cloud data, extract and combine geometric and image features to construct a classification model, and accurately identify and classify rock and soil particles includes: Point cloud preprocessing: use statistical filtering algorithm to remove outliers in laser scanning point cloud data, calculate the average distance between each point in the point cloud data and its neighboring points, set a distance threshold according to the overall distribution characteristics of the point cloud data, and regard points whose distance exceeds the threshold as outliers and remove them from the data set. Perform the filtering algorithm iteratively for multiple times until the number of outliers in the remaining point cloud data meets the requirements. Use the straight-through filtering algorithm to remove the reflection point cloud generated by the internal equipment of the shield machine according to the coordinate range of the point cloud. Obtain the spatial coordinate range of the internal equipment according to the CAD model of the shield machine, and then set the range parameters of the straight-through filtering to filter out the point cloud data within the coordinate range of the equipment. For some equipment reflection points that are difficult to filter accurately through the coordinate range, further screening is carried out in combination with the intensity information of the point cloud. Use the density-based clustering algorithm to divide the point cloud into different clusters. Set the parameters of the DBSCAN algorithm according to the expected density range of the rock and soil particle point cloud. After the point cloud is divided into multiple clusters by the clustering algorithm, further screen the point cloud clusters belonging to a single rock and soil particle according to the size, shape and point density of the cluster, and remove clusters that are too large or too small. Feature extraction and classification model construction, calculate the detailed geometric features of the extracted rock and soil particle point cloud data, including the minimum bounding box size, sphericity, surface curvature, the ratio of the major axis to the minor axis of the particle, and the surface concavity. Use the principal component analysis algorithm to determine the main direction of the particle, construct the covariance matrix of the point cloud data, and calculate the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvector corresponds to the main direction of the point cloud data. By projecting the point cloud data onto the main direction, the size information of the particle in different directions is obtained, and then the ratio of the major axis to the minor axis is calculated. The surface concavity feature is determined by calculating the curvature change rate of the point cloud surface. Combined with image analysis technology, the color features of the particles are extracted from the images collected synchronously with the point cloud data, and the color images are converted into different color spaces for analysis. In the RGB color space, the color histogram of the particle area is calculated, and the distribution of pixel values ​​in each color channel is statistically analyzed to obtain the color mean and variance. In the HSV color space, the features of the hue and saturation channels are analyzed in detail. The texture feature extraction uses the grayscale co-occurrence matrix algorithm. First, the color image is converted into a grayscale image, and then the grayscale in different directions and distances is calculated. Co-occurrence matrix, extract texture parameters such as contrast, correlation, energy and entropy from the grayscale co-occurrence matrix, when geometric features and image features are combined into feature vectors, they are reasonably combined and normalized according to the importance of features and the correlation of data, and the support vector machine algorithm is used to build the classification model. The training data is used for model training, and the appropriate kernel function and penalty parameter C are selected by cross-validation. The cross-validation adopts k-fold cross-validation, and the training data is divided into k subsets. Each time, k-1 subsets are used as training sets, and the remaining subset is used as validation sets. Repeat k times, calculate the classification accuracy on each validation set, select the kernel function and penalty parameter combination with the highest accuracy, build the model using the random forest algorithm, determine the number of decision trees, the maximum depth of the tree and other parameters, adjust the parameters by evaluating the out-of-bag error, and continuously adjust the number and depth of decision trees, observe the changes in the out-of-bag error, and select the parameter combination that minimizes the out-of-bag error. After the training is completed, the model is evaluated using the test data, and the accuracy, recall, F1 value and other indicators of the model are calculated. The calculation formula is F1=2*(accuracy*recall) / (accuracy+recall).

4. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The aforementioned preprocessing and labeling of the initial image sequence of shield tunneling, dividing the data set to train the algorithm combining the optical flow method and the convolutional neural network, optimizing parameters and evaluating performance to initialize the unmarked motion tracking algorithm, includes: Data preprocessing and annotation: preprocess the collected image sequence of the initial stage of shield tunneling. Image enhancement includes adjusting brightness, contrast and color balance. The adaptive histogram equalization algorithm is used to adjust the brightness of the image. The image cropping operation is performed accurately according to the position and size of the key area inside the shield machine, removing irrelevant background areas in the image, reducing the amount of data and highlighting the area where the rock and soil particles are located. The image deblurring process uses a blind deconvolution algorithm to automatically estimate the blur function based on the blur characteristics of the image, and perform a deconvolution operation on the image to restore the clarity of the image and convert the image into a grayscale image. The weighted The averaging method converts the RGB channels of the color image into grayscale values, annotates the rock and soil particles in the image, and manually annotates some key image frames to mark the precise position and bounding box size of the particles. Semi-automatic annotation uses the annotated image frames to train the target detection model, and uses the pre-trained convolutional neural network as the basic network to perform fine-tuning to adapt to the detection task of rock and soil particles. The model is applied to unlabeled image frames to automatically generate initial annotation information for the particles, which is then manually reviewed and corrected. The annotation information includes the position, outside the bounding box, and the category label of the particles, and different types of rock and soil particles are labeled as different categories. Algorithm training and optimization: The labeled data are divided into training set, validation set and test set according to the ratio of 70%, 20% and 10%, and the stratified sampling method is adopted to ensure that the ratio of various types of rock and soil particles in the training set, validation set and test set is roughly the same as that in the original data. The training set is used to train the algorithm combining the optical flow method and the convolutional neural network. In the convolutional neural network part, different network architectures are tried, and the optimal architecture is selected according to the accuracy on the validation set. The adaptive learning rate adjustment strategy is adopted to optimize the algorithm parameters, and the initial learning rate and attenuation factor are set. The number of iterations is determined according to the convergence of the model on the validation set. By observing the changes in the loss function value on the validation set, when the loss function value no longer decreases significantly within several consecutive iterations, the model is considered to have converged, and training is stopped. The L1 / L2 regularization technology is used to prevent overfitting, and the regularization coefficient is adjusted. The impact of regularization on the accuracy and generalization ability of the model on the validation set is observed, and an appropriate regularization coefficient is selected. The performance of the trained algorithm is evaluated using the test set, and the average precision mean and intersection-over-union ratio are calculated. When the indicators do not meet the expected requirements, the reasons are further analyzed, and targeted adjustments are made and retraining is performed until the performance of the algorithm on the test set meets the reliability requirements of actual applications.

5. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The shield machine collects and processes data at time intervals during tunneling, calculates the movement trajectory of rock and soil particles, and corrects it based on the pressure and vibration sensor data to adapt to the changes in the working state of the shield machine, including: Real-time data processing: when collecting laser scanning data, check data integrity, verify spatial consistency of point cloud data, compare coordinate variation range of the same feature point in adjacent scanning frames, if it exceeds the preset threshold, mark data abnormality, check image data clarity, based on image contrast and edge sharpness assessment, use image spectrum analysis method to calculate Fourier spectrum of image, if high-frequency component missing or abnormal attenuation, indicating image blur, mark below the set clarity threshold, when data is marked as unqualified, supplementary collection process based on the original collection time delay of no more than 0.5 seconds, the maximum number of supplementary collection is 3 times, if still unqualified, alarm is issued, prompting manual intervention inspection equipment, data compression uses hybrid compression algorithm, firstly compress laser scanning data based on geometric features, for image data, first perform regional segmentation, use different compression ratios for background area and particle area, use high compression ratio for background area, use low compression ratio for particle area, data encryption uses AES-256 encryption algorithm, encrypts compressed data, encryption key uses dynamic generation method, updated every 10 minutes, generates key seed based on shield machine operation time and current position; The motion trajectory calculation and correction logic calculates the displacement vector of rock and soil particles between adjacent frames. The pyramid-based Lucas-Kanade algorithm is used to first build an image pyramid, calculate the optical flow from the top layer, and gradually refine it to the lower layers. The three-dimensional position change of particles is determined in combination with the laser scanning data. The displacement vector calculated by the image is calibrated using the high-precision spatial coordinate information of the laser scanning. When the pressure sensor detects abnormal pressure, the position of the pressure change is located through the installation position information of the sensor. The pressure amplitude is compared with the normal excavation pressure range to determine the degree of abnormality. The correction coefficient is used according to the degree of pressure abnormality. The linear or nonlinear mapping relationship is determined. When processing the vibration sensor data, the time-frequency analysis method is used to convert the vibration signal into the time-frequency domain, and the frequency component and amplitude change of the vibration are analyzed. If the high-frequency vibration energy exceeds 50% of the normal threshold, it is judged that the cutterhead cutting is uneven, and the random disturbance of the particle movement direction is increased. The disturbance amplitude is determined according to the high-frequency vibration energy, between 5°-15°. If the low-frequency vibration amplitude exceeds 30% of the normal threshold, it means that the overall advancement of the shield machine is hindered, and the estimated value of the particle movement speed inside the entire shield machine is reduced by 20% to 30% to reflect the slowdown of particle movement when the shield machine is difficult to advance.

6. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The method of collating geological survey and shield machine historical data and constructing a priori probability distribution function, collecting sensor data in real time and inputting it into the Bayesian estimation model, updating parameters and posterior probability distribution to estimate uncertainty parameters includes: Prior information collation and model construction: When collating geological survey report data, statistical distribution information is collected, the correlation between geotechnical parameters of different strata is analyzed, and the prior probability distribution function of the Bayesian estimation model is constructed. For complex geotechnical parameters, a mixed probability distribution model is used to determine different distribution parameters according to different stratum areas or depth ranges, and the prior distribution is constructed in a weighted combination manner. The weight of the prior distribution is determined according to the reliability of the geological survey data and the uniformity of the strata; Real-time estimation and update: sensor data is input into the Bayesian estimation model as observation values, and the data is preprocessed. The pressure data is filtered to remove measurement noise. The Kalman filter algorithm is used to set the filter parameters according to the dynamic characteristics of pressure changes. The vibration data is used for feature extraction, and the frequency, amplitude, peak factor and kurtosis of vibration are extracted. These features are associated with geotechnical parameters and shield machine working parameters to establish an association model. The model is updated using the Gibbs sampling algorithm in the Markov chain Monte Carlo method, and a sufficient number of sampling times is set to ensure the convergence of the posterior probability distribution. After each sampling, the confidence interval of the parameter estimate is calculated, and the confidence level is set to 95%. If the new sampling value exceeds the confidence interval, the model parameters or prior distribution are adjusted, the parameter range or weight of the prior distribution is adjusted, and re-sampling is performed until the parameter estimate is stable within the confidence interval and the posterior probability distribution no longer changes significantly.

7. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The aforementioned method of constructing a numerical simulation model and setting parameters based on the shield machine and the formation conditions, comparing the simulated and measured motion trajectories, and re-simulating and verifying after checking and optimizing the relevant links when the deviation exceeds the threshold value includes: The numerical simulation model is constructed and the parameters are set. Finite element modeling is performed. The cutterhead, shield and other structures of the shield machine are finely modeled using solid units. Considering the anisotropic properties of the material, the tool on the cutterhead uses contact units to simulate its interaction with the rock and soil. The friction coefficient of the contact unit is determined according to the actual friction test results of the tool material and the rock and soil type, accurate to 0.01, and discrete element modeling is performed. The mechanical parameters of the rock and soil particles are set according to the classification results and the parameters estimated in real time. The contact model between particles uses the Hertz-Mindlin contact model, and the rolling friction of the particles is considered. The boundary conditions are set, and the thrust of the shield machine is dynamically adjusted according to the set value of the actual construction and the thrust speed monitored in real time. The constraint conditions of the surrounding rock and soil are simulated by spring-damper units. The spring and damping coefficients are determined according to the elastic modulus and damping ratio of the stratum. The elastic modulus and damping ratio are adjusted according to the geological survey data and real-time estimated parameters to accurately reflect the constraint effect of the stratum on the shield machine. In the initial condition setting, the initial stress state of the rock and soil is determined according to the ground stress measurement data, and the particle distribution is initialized according to the initial rock and soil morphology obtained by laser scanning to ensure that the initial state of the simulation is consistent with the actual situation. Comparative analysis and deviation processing, calculate the distance deviation and speed deviation between trajectory points, use the space vector calculation method, consider the three-dimensional characteristics of particle motion, calculate the Euclidean distance and the deviation components of trajectory points in different coordinate axis directions, and use the instantaneous speed and average speed comparison method to calculate the instantaneous speed and average speed of the simulated trajectory and the measured trajectory in the same time period, and compare the difference between the two. When the deviation exceeds the threshold range, check the abnormal process of sensor data, check data loss and fault alarms, and perform data rationality checks. Algorithm problem analysis uses sensitivity analysis methods to change the key parameters in the algorithm one by one, observe the impact on the trajectory calculation results, determine the key parameters that cause the deviation, and optimize and adjust them.

8. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The development visualization platform displays the movement trajectory of rock and soil particles, generates analysis reports containing various information, and archives them after automatic and manual review to ensure accuracy, including: Visualization platform construction, development of visualization platform, 3D graphics rendering engine adopts physical rendering technology, in the internal view of shield machine, it displays the movement trajectory of rock and soil particles and the temperature distribution and stress distribution of the internal structure of shield machine, and updates the display through real-time connection with sensor data, classification display function, when displaying by particle type, different colors and shapes are used to represent different types of particles, when displaying by movement speed range, speed range is set, particles in different speed ranges are represented by different transparency or flashing frequency, in terms of interactive function, after the user clicks on a specific particle, the movement information and detailed physical characteristics of the particle are displayed; Analysis report template and content generation, record general information, shield machine construction team information, construction environment temperature and humidity, summarize the average movement speed, movement direction distribution and residence time distribution of rock and soil particles, analyze the acceleration change of particle movement, calculate the acceleration by the second-order derivative of the trajectory data, analyze the magnitude and direction of the acceleration to understand the force change of particles in the shield machine, evaluate the impact of parameter uncertainty, use sensitivity analysis method to quantify the impact of different parameter uncertainties on particle movement trajectory, sort parameters according to the sensitivity coefficient, analyze their changes in different construction stages and their impact on the working state of the shield machine, analyze The cutting efficiency, soil discharge smoothness and energy consumption of the shield machine are taken into consideration. According to the power monitoring data of the shield machine, the energy consumption per unit excavation distance is calculated. The relationship between energy consumption and particle motion state is analyzed in combination with the particle motion trajectory. The automated part of the analysis report generation uses a scripting language to write an automated report generation program. The report template content is automatically filled in according to the data processing and analysis results to generate a preliminary report. The data accuracy, rationality of the analysis method and feasibility of the suggestions are manually reviewed. After the review is passed, the report is signed and confirmed. The final report is archived in both electronic and paper forms. The electronic document is in PDF format, and the paper document is archived according to the construction project and time sequence.

9. The method for obtaining the motion trajectory of rock and soil particles during shield tunneling according to claim 1, characterized in that: The multi-objective adaptive optimization control model is constructed based on the model predictive control framework, and the shield machine working parameter adjustment plan is calculated and fed back according to real-time data, so as to realize closed-loop control of each link of the coordinated system and improve the construction efficiency and quality, including: Optimization control model construction and parameter setting. The multi-objective adaptive optimization control model is constructed using a framework based on model predictive control. The input variables include the influence of conventional variables and the attitude parameters of the shield machine on particle movement and construction effect. The objective function is set as the weighted sum of multiple sub-objective functions. The sub-objective function of the tool wear rate is constructed based on the relationship between the wear characteristics of the tool material and the working parameters of the shield machine. The tool wear rate is controlled at an acceptable level as a constraint condition, and the hierarchical analysis method is used to determine the weight coefficient of the objective function. Feedback control works in coordination with the system, and the optimization control model calculates the optimal working parameter adjustment plan based on real-time data, and uses an efficient optimization algorithm to solve it. During the solution process, the adjustment limits of the shield machine's working parameters are considered, and the adjusted parameters are fed back to the shield machine control system. The industrial communication protocol is used for data transmission. After receiving the parameter instructions, the control system completes the parameter adjustment and executes the new operation within 0.1-0.5 seconds, and feeds back the adjusted parameters and control effect information to the motion trajectory acquisition system. The motion trajectory acquisition system further optimizes the data acquisition frequency and range based on this information, adjusts the algorithm parameters, and optimizes the parameters in the motion tracking algorithm based on the control effect information.

10. A system for acquiring the motion trajectory of rock and soil particles during shield tunneling, characterized in that: include: Multi-source data acquisition and fusion preparation module, planning the installation layout of sensors and accurate calibration, building industrial Ethernet data transmission network and large-capacity storage device; The module for preliminary identification and classification of rock and soil particles uses a variety of filtering algorithms to preprocess laser scanning point cloud data, extract and combine geometric and image features to build a classification model, and accurately identify and classify rock and soil particles; The markerless motion tracking algorithm initialization module preprocesses and annotates the initial image sequence of shield tunneling, divides the data set to train the algorithm combining optical flow method and convolutional neural network, optimizes parameters and evaluates performance to initialize the markerless motion tracking algorithm; Real-time motion trajectory calculation and correction module, which collects and processes data at time intervals during shield tunneling, calculates the motion trajectory of rock and soil particles and corrects it based on pressure and vibration sensor data to adapt to changes in the working state of the shield machine; The uncertainty parameter real-time estimation module organizes geological survey and shield machine historical data and constructs a priori probability distribution function, collects sensor data in real time and inputs it into the Bayesian estimation model, updates parameters and posterior probability distribution to estimate uncertainty parameters; The simulation and measured data comparison and verification module builds a numerical simulation model and sets parameters based on the shield machine and stratum conditions, compares the simulation and measured motion trajectories, and checks and optimizes related links and re-simulates and verifies when the deviation exceeds the threshold; The motion trajectory visualization and analysis report generation module develops a visualization platform to display the motion trajectory of rock and soil particles, generates analysis reports containing various information, and archives them after automatic and manual review to ensure accuracy; The adaptive optimization and feedback control module builds a multi-objective adaptive optimization control model based on the model predictive control framework, calculates and feeds back the shield machine working parameter adjustment plan based on real-time data, and coordinates all links of the system to achieve closed-loop control to improve construction efficiency and quality.

Images