A shield tunnel segment disease integrated diagnosis and treatment evaluation method and system

Abstract

The application belongs to the field of tunnel engineering operation and maintenance, and provides a shield tunnel segment disease integrated diagnosis and treatment evaluation method and system, which comprises obtaining multi-modal original data of shield tunnel segment structure for fusion to obtain disease perception data; based on the disease perception data, using a pre-trained machine learning model and the weight of the health evaluation index to perform fault diagnosis to obtain a fault diagnosis result; based on the fault diagnosis result, using a pre-trained hybrid prediction model to perform disease trend prediction, and generating a pre-treatment scheme according to the prediction result; evaluating the implementation effect of the pre-treatment scheme to generate a pre-treatment efficiency evaluation report. Through deep coupling of data driving and intelligent algorithm, the application realizes early warning, accurate intervention and closed-loop optimization of shield tunnel segment diseases, and fundamentally improves the initiative, refinement and full life cycle economic benefits of tunnel operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of tunnel engineering operation and maintenance technology, specifically relating to an integrated diagnosis and assessment method and system for shield tunnel segment defects. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the accelerating pace of global urbanization, the development and utilization of urban underground space has become a key strategic measure to alleviate land scarcity, improve traffic congestion, and address urban problems such as environmental degradation. Shield tunneling technology, due to its significant advantages such as minimal impact on the surrounding environment, high construction safety, and wide adaptability to geological formations, is widely used in the construction of major infrastructure projects such as subways, highways, railways, and integrated utility tunnels. As the main body of the tunnel structure, shield tunnel segments, during long-term service, are subjected to the combined effects of complex geological loads, groundwater pressure, disturbances from nearby construction, and the degradation of material properties, making them highly susceptible to defects such as cracks, water leakage, misalignment, and damage. Therefore, how to accurately and efficiently ensure the safety, durability, and reliability of shield tunnel segments throughout their long operating cycle has become a core challenge that urgently needs to be overcome in the field of tunnel operation and maintenance management.

[0004] Currently, the assessment and management methods for shield tunnel segment defects still have significant limitations. First, the fragmentation of the assessment process is prominent. Traditional defect assessment processes often separate defect detection, causal diagnosis, safety assessment, and repair, which are completed independently by different teams or at different stages of the project. This fragmented approach leads to a break in the data chain, and a lack of effective correlation between the superficial symptoms and the underlying mechanisms of the defects. For example, the detection stage only provides surface information such as crack width and leakage location, while subsequent causal diagnosis and safety assessment, due to the lack of synchronous interaction with the detection data, struggle to accurately determine whether the defects are caused by abnormal internal forces, underlying cavities, or material deterioration. This results in highly subjective diagnostic conclusions, easily leading to misjudgments or omissions, and causing subsequent treatment measures to be either blind or insufficient. Second, the assessment dimensions are superficial. Existing technologies mostly focus on the phenomenological description and qualitative analysis of segment defects, failing to effectively integrate deeper information that reflects the true stress state of the structure. Key mechanical indicators such as the presence of voids behind tunnel segments, excessive stress in reinforcing steel, and decreased structural stiffness are often excluded from conventional assessment systems. This superficial assessment method cannot objectively and quantitatively reveal the actual impact of defects on the overall load-bearing capacity and long-term durability of the structure, making it difficult to provide accurate early warnings of the structure's true health condition. Finally, the limitations of existing intelligent research urgently need to be overcome. In recent years, with the rapid development of artificial intelligence technology, image recognition and data mining methods based on machine learning have shown great potential in tunnel defect detection and preliminary diagnosis, opening up new paths for improving the intelligence level of assessment work. However, existing research and applications mostly focus on a specific defect type (such as crack identification only) or an isolated link in the assessment chain (such as optimizing only the detection algorithm), and have not yet formed a complete closed loop from data perception, mechanism diagnosis, intelligent decision-making to treatment assessment. This research paradigm is difficult to cope with the real-world challenges of frequent, concurrent, and complex coupled evolutionary defects in actual engineering projects, and cannot provide systematic technical support for the full life-cycle health management of shield tunnels. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an integrated diagnosis and assessment method and system for shield tunnel segment defects. This invention constructs a closed-loop management system that breaks down traditional assessment barriers, enabling seamless data flow and business collaboration throughout the entire process of "perception—diagnosis—treatment—assessment." The aim is to integrate multi-source heterogeneous data, fuse structural analysis models with intelligent algorithms, and form a systematic and holistic solution. This solution accurately diagnoses the causes of defects, scientifically assesses structural conditions, optimizes treatment strategies, and verifies the effectiveness through a closed-loop process, thereby comprehensively improving the safe operation and maintenance level of shield tunnels in complex service environments.

[0006] According to some embodiments, the first aspect of the present invention provides an integrated diagnosis and assessment method for defects in shield tunnel segments, employing the following technical solution: An integrated diagnosis and assessment method for shield tunnel segment defects includes: Multimodal raw data of shield tunnel segment structure are acquired and fused to obtain defect perception data; Based on the disease perception data, fault diagnosis is performed using a pre-trained machine learning model and the weights of health evaluation indicators to obtain fault diagnosis results. Based on the fault diagnosis results, a pre-trained hybrid prediction model is used to predict the disease trend, and a pre-treatment plan is generated based on the prediction results. An evaluation report on the effectiveness of the pre-treatment plan is generated based on its implementation.

[0007] Furthermore, the multimodal raw data of the obtained shield tunnel segment structure are fused to obtain defect perception data; Acquire multimodal raw data of shield tunnel segment structure; The original multimodal data is preprocessed, and the preprocessed multimodal data is fused to obtain disease perception data.

[0008] Furthermore, the multimodal raw data includes the geometry of the shield tunnel segment structure, images of apparent defects, internal defects, and environmental and load conditions.

[0009] Furthermore, the fault diagnosis based on the disease perception data, using a pre-trained machine learning model and the weights of health evaluation indicators, yields the following fault diagnosis results: Based on tunnel health levels, key factors of tunnel health, and tunnel health assessment indicators, a health evaluation system for shield tunnels is constructed. Using an improved analytic hierarchy process, the weights of different health evaluation indicators in the health evaluation system of shield tunnels are calculated. Based on the disease perception data, fault diagnosis is performed using a pre-trained machine learning model and the weights of health evaluation indicators to obtain fault diagnosis results.

[0010] Furthermore, the tunnel health level is divided into four levels; The key factors for tunnel health include material deterioration, water leakage, structural deformation, and settlement. Material degradation includes the width of segment cracks, the area of ​​spalling and missing pieces, and the concrete strength ratio; Leakage includes pH value, leakage volume per unit time, and leakage area; Structural deformation includes circumferential joint width, longitudinal joint width, transverse misalignment, radial misalignment, bolt stress-strength ratio, horizontal convergence deformation rate, and vertical convergence deformation rate. Settlement includes cumulative settlement and differential settlement.

[0011] Furthermore, the step of using a pre-trained hybrid prediction model to predict disease trends based on fault diagnosis results, and generating a pre-treatment plan based on the prediction results, includes: The fault diagnosis results are preprocessed to obtain the preprocessed fault diagnosis results. Based on the pre-processed fault diagnosis results, the disease trend is predicted using a pre-trained hybrid prediction model to obtain the prediction results; The prediction results are matched with the treatment plans in the case library to obtain the optimal treatment plan. A digital twin simulation is then performed based on the treatment plan to generate a pre-treatment plan.

[0012] According to some embodiments, the second aspect of the present invention provides an integrated diagnosis and assessment system for tunnel segment defects, employing the following technical solution: An integrated diagnosis and assessment system for shield tunnel segment defects includes: The data sensing module is configured to acquire and fuse multimodal raw data of the shield tunnel segment structure to obtain defect sensing data. The fault diagnosis module is configured to perform fault diagnosis based on the disease perception data, using a pre-trained machine learning model and the weights of health evaluation indicators, and obtain the fault diagnosis results. The fault treatment module is configured to predict disease trends based on fault diagnosis results using a pre-trained hybrid prediction model, and generate a pre-treatment plan based on the prediction results. The treatment evaluation module is configured to evaluate the effectiveness of the pre-treatment plan and generate a pre-treatment efficacy evaluation report.

[0013] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the integrated diagnosis and assessment method for shield tunnel segment defects as described in the first embodiment above.

[0015] According to some embodiments, a fourth aspect of the present invention provides a computer device.

[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the integrated diagnosis and treatment assessment method for shield tunnel segment defects as described in the first embodiment above.

[0017] According to some embodiments, a fifth aspect of the present invention provides a computer program product or computer program.

[0018] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the integrated diagnosis and treatment assessment method for shield tunnel segment defects as described in the first embodiment above.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a closed-loop system integrating perception, diagnosis, treatment, and evaluation, breaking the fragmented nature of traditional shield tunnel segment defect management. Through end-to-end data integration and business collaboration, it achieves systematic management from defect data collection, cause analysis, solution formulation to effect verification, effectively avoiding insufficient diagnostic basis and misjudgments / omissions caused by data fragmentation, and significantly improving the integrity and consistency of tunnel operation and maintenance management.

[0020] This invention integrates high-precision sensing technology with intelligent image recognition in the sensing stage, enabling efficient extraction and digital representation of surface and internal defects information of pipe segments. Compared to traditional methods that only describe surface defects, this solution constructs a structured defect database, providing a complete and reliable data foundation for subsequent analysis. This solves the problem of existing detection methods lacking correlation with deeper information such as structural internal forces and underlying voids, thus enabling defect assessment to move from superficial to in-depth analysis.

[0021] This invention introduces a multi-field information fusion assessment and diagnostic model based on machine learning and deep mining in the diagnostic stage, and establishes a multi-level, multi-dimensional grading standard by combining the analytic hierarchy process (AHP). Through advanced algorithms such as graph neural networks and long short-term memory networks, it deeply mines the mapping relationship between the evolution patterns of defects and disaster-causing factors, enabling accurate identification of the service status of tunnel segments and effective early warning of safety risks. This effectively addresses the complex characteristics of the frequent, concurrent, and coupled evolution of defects in shield tunnel segments, overcoming the limitations of existing research that focuses only on single defects or single aspects.

[0022] This invention innovatively deploys a GNN-ST-LSTM dual-engine prediction model in the treatment phase, achieving accurate spatiotemporal prediction of disease development trends and intelligently matching the optimal repair solution based on the prediction results. Simultaneously, by encapsulating the entire process data into reusable digital cases, it enables the rapid generation of pre-optimized solutions in new scenarios, promoting the co-evolution of the prediction model and the treatment knowledge base, and significantly improving the targeting of treatment solutions and the level of intelligent decision-making.

[0023] This invention constructs a multi-dimensional and quantitative performance verification mechanism in the evaluation phase, and uses a digital twin platform to achieve a three-dimensional visual comparison between actual results and predicted targets. By forming a closed-loop feedback of evaluation results, successful cases strengthen the knowledge base weight, and unsatisfactory cases trigger the automatic optimization of diagnostic models, predictive models, and decision rules, it achieves a leap from one-time handling to continuous intelligent maintenance, fundamentally improving the initiative, refinement, and full life-cycle economic benefits of tunnel operation and maintenance. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 This is a flowchart of an integrated diagnosis and assessment method for tunnel segment defects in shield tunnels, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the shield tunnel health evaluation system in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0030] Example 1 like Figure 1As shown, this embodiment provides an integrated diagnosis and assessment method for tunnel segment defects. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to a terminal, or to a system including a terminal, server, and system, and is implemented through interaction between the terminal and server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps: Multimodal raw data of shield tunnel segment structure are acquired and fused to obtain defect perception data; Based on the disease perception data, fault diagnosis is performed using a pre-trained machine learning model and the weights of health evaluation indicators to obtain fault diagnosis results. Based on the fault diagnosis results, a pre-trained hybrid prediction model is used to predict the disease trend, and a pre-treatment plan is generated based on the prediction results. An evaluation report on the effectiveness of the pre-treatment plan is generated based on its implementation.

[0031] The integrated diagnosis and assessment method for shield tunnel segment defects described in this embodiment is implemented as follows: Step S1: Obtain and fuse multimodal raw data of shield tunnel segment structure to obtain defect perception data; The intelligent sensing link is the data foundation of the entire technology system. Its core task is to achieve full-domain, multi-dimensional, and non-destructive information collection of shield tunnel segments from apparent damage to internal defects through a multi-source sensor collaborative detection network. Through data fusion and feature extraction, it outputs a structured defect feature data stream, providing a complete and reliable data foundation for subsequent diagnosis, treatment, and evaluation.

[0032] Step S1.1: Obtain multimodal raw data of the shield tunnel segment structure; The collected multimodal raw data can be categorized as follows: Geometric morphology: including point cloud and coordinate elevation; Apparent disease images: These are images of apparent damage. Internal defects and materials: radar profile data and elastic wave signals reflecting the structure behind the segments and the distribution of reinforcing bars; Environment and loads: This includes environmental and load data such as stress, strain, surrounding soil pressure, and chemical composition of seepage water for the segment structure.

[0033] Each type of data contains a wealth of engineering information, such as the length, width, and direction of cracks, the location and size of cavities, and the convergence deformation rate of tunnel segments.

[0034] This stage begins by deploying a collaborative detection network composed of various high-performance sensors through a multi-source sensing layer. This network comprehensively utilizes 3D laser scanning technology to accurately capture the overall and local geometric changes of the tunnel, acquiring point cloud data for extracting geometric parameters such as convergence deformation and misalignment. Ground-penetrating radar penetrates the concrete surface to detect hidden defects such as internal voids and looseness, acquiring radar profile data reflecting the structure behind the tunnel segments and the distribution of reinforcing steel. Elastic wave technology is used to sense the deterioration of material mechanical properties, assessing concrete strength and internal defects by measuring wave velocity and attenuation. High-definition imaging technology clearly records apparent defects such as cracks, spalling, and leakage, acquiring high-definition image data of the tunnel's inner wall. Furthermore, embedded strain gauges, earth pressure cells, and water quality analyzers are used to acquire environmental and load data such as stress and strain of the tunnel segment structure, surrounding earth pressure, and the chemical composition of seepage water. These technologies together form a complementary sensing matrix, achieving non-destructive acquisition of comprehensive, multi-dimensional information about the shield tunnel segment structure, from macroscopic deformation to microscopic materials, and from internal defects to apparent damage.

[0035] Step S1.2: Preprocess the multimodal raw data, and fuse the preprocessed multimodal data to obtain disease perception data; In the data fusion and feature extraction layer, the main task is to transform the raw data collected by the multi-source sensing layer into quantifiable feature indicators with clear engineering significance.

[0036] First, for the physical characteristics of the raw data of each modality, the corresponding data preprocessing methods are adopted as follows: For geometric morphology data, to address the issue of unifying the spatiotemporal reference for multi-source data, a spatiotemporal synchronization calibration algorithm based on feature point matching is adopted:

[0037] In the formula, The optimal spatiotemporal transformation matrix. and Data collected by different sensors under the same modal data. The index sequence number corresponding to the matching point, ranging from... To match the number of point pairs, Spatial transformation operation functions, This represents the value of the transformation matrix that minimizes the error function. The algorithm ensures that all sensor data resides in a unified coordinate system and along the time axis.

[0038] Extract geometric parameters such as convergence deformation and misalignment from point cloud data; Based on ground-penetrating radar signals from internal defects and material data, the location and size of cavities and loose areas are identified through techniques such as migration imaging and inversion; material performance indicators such as wave velocity and attenuation are calculated from elastic wave data. For surface disease image data, a segmentation model is used to accurately identify surface cracks and quantify their length and width. The segmentation model can be, but is not limited to, the U-Net++ segmentation model, whose loss function incorporates continuity constraints.

[0039] In the formula, For continuous weights, The gradient is the image gradient. This model can accurately identify and quantify crack geometric parameters.

[0040] Finally, an adaptive weighted fusion algorithm for multimodal data is used to transform the preprocessed multimodal data into standardized disease features:

[0041]

[0042] In the formula, For the first Standardized characteristics of sensor-like devices For the first Adaptive weights for sensor-like devices For sensor signal-to-noise ratio, The adjustment coefficients are converted into standardized feature vectors or structured data units, and these results are presented in the form of structured disease data units. After preprocessing, the raw data is directly input into the hybrid evaluation model in the second part.

[0043] Four types of data collected from different sensors at the same time point were spliced ​​and packaged into a structured comprehensive data record, retaining their respective confidence weights, and then stored in the database.

[0044] Step S2: Based on the disease perception data, use the pre-trained machine learning model and the weights of health evaluation indicators to perform fault diagnosis and obtain the fault diagnosis results; The intelligent diagnosis stage, based on perceived data, utilizes a hybrid evaluation model that integrates the improved Analytic Hierarchy Process (AHP) and interpretable machine learning to accurately determine the health status and analyze the causes of pipeline defects. This stage systematically improves upon traditional AHP, aiming to enhance the objectivity, robustness, and accuracy of weight calculation, thereby providing more reliable input features for subsequent machine learning models.

[0045] Step S2.1: Based on tunnel health level, key factors of tunnel health, and tunnel health assessment indicators, construct a shield tunnel health evaluation system; Build as Figure 2 The shield tunnel health evaluation hierarchical structure model shown divides the evaluation system into target layer (tunnel health level), criterion layer (material deterioration, water leakage, structural deformation, settlement) and indicator layer (a total of 15 specific indicators).

[0046] The target layer adopts a four-level classification method, dividing the tunnel structure health status into Level 1 (no defects or minor defects), Level 2 (general defects), Level 3 (relatively serious defects), and Level 4 (serious defects).

[0047] The criteria layer consists of four key factors that affect the health status of shield tunnels; The indicator layer further refines this into specific, quantifiable parameters, for example, Material degradation includes the width of segment cracks, the area of ​​spalling and missing pieces, and the concrete strength ratio; Leakage includes pH value, leakage volume per unit time, and leakage area; Structural deformation includes circumferential joint width, longitudinal joint width, transverse misalignment, radial misalignment, bolt stress-strength ratio, horizontal convergence deformation rate, and vertical convergence deformation rate. Settlement includes cumulative settlement and differential settlement. These index layer data all come from the structured disease database output in Part 1, and have clear spatiotemporal attributes and confidence levels.

[0048] Step S2.2: Calculate the weights of different health assessment indicators in the shield tunnel health assessment system using the improved analytic hierarchy process. The improved weight calculation method of the analytic hierarchy process includes two major improvements: first, it uses multi-expert decision-making and fuzzy consistency matrix to construct a comprehensive judgment matrix; second, it introduces particle swarm optimization algorithm to solve for the optimal weight vector.

[0049] In the multi-expert decision-making process, M domain experts are invited, each focusing on the same criterion level. Each indicator is used to independently construct a judgment matrix based on the 1-9 scaling method. as follows:

[0050] in, Experts Indicators relative to indicators The degree of importance. To eliminate logical contradictions that may exist in the judgment of a single expert (such as... Each judgment matrix is ​​transformed into a fuzzy consistency matrix. as follows:

[0051]

[0052] Furthermore, a consistency adjustment algorithm (such as the weighted average method) is used to ensure that the matrix satisfies... and .

[0053] Then, experts are assigned corresponding weights based on their authority, years of experience, and other factors. By weighted and fused multiple fuzzy consistency matrices, the comprehensive fuzzy judgment matrix is ​​obtained as follows:

[0054] in, ,satisfy This comprehensive matrix not only incorporates collective wisdom but also exhibits good consistency.

[0055] After obtaining the comprehensive fuzzy judgment matrix, the weight vector is solved using the particle swarm optimization algorithm. Particle swarm optimization is a global optimization algorithm based on swarm intelligence, suitable for solving nonlinear and nonconvex optimization problems. In this paper, each particle represents a possible weight vector. ,satisfy and The fitness function is defined as the error between the comprehensive fuzzy judgment matrix and the consistency matrix constructed from the weights:

[0056] This function measures the degree of consistency between the current weight vector and the judgment of the expert group; the smaller the value, the better the weight reflects the expert consensus.

[0057] Particle swarm optimization (PSO) searches for the optimal solution that minimizes the fitness function by iteratively updating the velocity and position of particles. Let the nth particle be an example of this. The position of each particle is The speed is The individual's historical best position is The group's historical best position is Then the velocity and position update formulas are:

[0058]

[0059] in, For inertial weights, , As a learning factor, , A random number between [0,1] This represents the current iteration count of the algorithm. During iteration, the position of each particle is normalized to satisfy the weights and constraints. When the fitness value converges or the maximum number of iterations is reached, the optimal position of the population is output. As the final weight vector This algorithm effectively overcomes the dependence of traditional eigenvalue methods on the strict consistency of the judgment matrix, and can find the globally optimal weight allocation under fuzzy judgment matrices.

[0060] After completing the weight calculation, the raw data of the current shield tunnel segment structure index layer is extracted from the structured defect database output in Part 1. This data has been preprocessed and standardized; for example, crack width is expressed in millimeters and leakage rate in liters per day. For each criterion layer, the data of its subordinate indexes are normalized (eliminating the influence of dimensions), and then a weighted sum is performed using the weight vector obtained from the improved AHP to obtain the comprehensive score for that criterion layer. Taking the material degradation criterion as an example, its comprehensive score is:

[0061] in, This is the value of the segment crack width after normalization. This represents the normalized value of the area of ​​the peeled-off block. This represents the normalized concrete strength ratio. yes The weights are then calculated. This yields the scores for the four criterion layers, forming a four-dimensional feature vector, namely the criterion score feature vector. This serves as input for subsequent machine learning models.

[0062] Step S2.3: Based on the disease perception data, use the pre-trained machine learning model and the weights of the health evaluation indicators to perform fault diagnosis and obtain the fault diagnosis results; Based on historical shield tunnel criterion layer data and their corresponding health levels, a hybrid evaluation model is constructed and trained. This invention employs a stacking ensemble learning strategy to fuse multiple heterogeneous base models to improve prediction performance. The first layer base models are Random Forest, XGBOOST, and LightGBM. These three models are based on Bagging and Boosting frameworks, respectively, and can mine data features from different perspectives. Random Forest, by constructing multiple decision trees and combining their prediction results, has strong anti-overfitting ability; XGBOOST uses a gradient boosting framework, gradually optimizing through serial learning of residuals, resulting in high accuracy; LightGBM introduces one-sided gradient sampling and mutually exclusive feature binding techniques on the basis of XGBOOST, making it fast to train and suitable for large-scale data. The second layer meta-model is Logistic Regression, which can learn the optimal combination weights output by the base models to achieve the final classification.

[0063] Before model training, a stratified sampling strategy was used to divide the historical dataset into training and test sets in a 7:3 ratio to ensure consistent distribution across categories. On the training set, 10-fold cross-validation combined with grid search was used for hyperparameter optimization. For example, for Random Forest, optimized hyperparameters included the number of trees (100-500), maximum depth (3-10), and minimum number of leaf samples (2-10); for XGBoost, hyperparameters included the learning rate (0.01-0.3), maximum depth (3-10), and subsampling ratio (0.6-1.0); LightGBM's hyperparameters were similar. Grid search iterated through all parameter combinations, calculating the average accuracy of each combination under 10-fold cross-validation, and selecting the best-performing parameter combination as the final model parameters. After the base model was trained, its predicted probabilities for the training set were used as new features to train the second-layer logistic regression meta-model. The logistic regression loss function used cross-entropy, and the weights were optimized using gradient descent. and bias The formula is:

[0064] in, For the first The probability vectors output by each base model belonging to each health level. for Function (multi-class classification) or Function (binary classification depends on the situation) This is the bias term. The final model output is the predicted probability distribution of the health level of the current tunnel segment, and the level corresponding to the highest probability is taken as the evaluation result.

[0065] After determining the optimal combination of hyperparameters, the performance of the machine learning model is validated on the test set. Four evaluation metrics—accuracy, precision, recall, and F1 score—are used for final evaluation of the machine learning model. The calculation methods for these evaluation metrics are as follows: 1. Accuracy Accuracy refers to the proportion of samples that are correctly predicted out of all samples.

[0066]

[0067] 2. Precision Precision refers to the proportion of samples that are predicted to be positive but are actually positive.

[0068]

[0069] 3. Recall Recall rate refers to the proportion of samples that are actually positive that are correctly predicted to be positive.

[0070]

[0071] 4. F1 score (F1-Score) The F1 score is the harmonic mean of precision and recall.

[0072]

[0073] In the formula, This represents a positive sample that the model predicts to be positive. This represents a negative sample that the model predicts to be positive. This represents a negative sample that the model predicts to be negative. This represents a positive sample that the model predicts to be negative.

[0074] The model output includes: the current health level of the tunnel segment (levels 1-4), the probability distribution of each health level, and the contribution of each criterion layer score to the final health level. The contribution can be interpreted by calculating the importance of each criterion layer feature in the base model (e.g., feature importance in random forests) or by using SHAP values, thus helping engineers understand which factors dominate the current health status. For example, if the model outputs a level 3 (relatively severe damage), and the contribution analysis shows that water leakage and structural deformation are the main driving factors, then subsequent treatment can focus on leakage and deformation issues to achieve precise intervention.

[0075] Step S3: Based on the fault diagnosis results, use the pre-trained hybrid prediction model to predict the disease trend, and generate a pre-treatment plan based on the prediction results; Part Three: The Precision Treatment stage is based on the disease classification and health status assessment results output by the intelligent diagnosis stage. It involves developing and implementing repair plans for diagnosed shield tunnel segment diseases. The core of this stage is a GNN-ST-LSTM hybrid prediction model. This model performs in-depth processing on the assessment results from Part Two to achieve accurate spatiotemporal prediction of disease development trends.

[0076] Step S3.1: Preprocess the fault diagnosis results to obtain preprocessed fault diagnosis results; The data layer receives and processes structured output from the intelligent diagnosis stage, primarily including the spatial location of the pipe segment unit, its current health level, the quantified values ​​of each criterion layer, and historical time-series data of these indicators. Different preprocessing strategies are employed for the disease data classified into different health levels (levels 1-4) in the diagnosis stage. For level 1-2 diseases, the data fluctuations are small, so Gaussian kernel density interpolation is used to complete the missing values.

[0077] Its estimation formula is:

[0078] In the formula, For the time points where interpolation is needed, The number of known data points. For the first Timestamps of known data points For the first Observations of known data points (such as crack width, settlement value, etc.). It is a Gaussian kernel function. It preserves the subtle variations in the data to the greatest extent possible. For level 2-3 diseases, data changes intensify and coupling may occur, so robust weighted smoothing is used to remove outliers; For diseases at levels 3-4, the data correlation is strong. A multi-factor collaborative correction method is adopted to ensure that the preprocessed data can accurately reflect the coupling state of the diseases.

[0079] Step S3.2: Based on the preprocessed fault diagnosis results, use the pre-trained hybrid prediction model to predict the disease trend and obtain the prediction results; The model layer, based on the high-quality dataset processed by the data layer, constructs a dual-engine collaborative prediction model for predicting disease development. This includes two parts: basic model selection and training model optimization and validation. Then, the model layer constructs a GNN-ST-LSTM hybrid prediction model based on the processed high-quality dataset. This model consists of two parts: 1. GNN Part (Spatial Dependency Modeling): The tunnel segment rings are treated as nodes in a graph, and the adjacency relationships between segments are considered as edges. A graph convolutional network is used to learn the mutual influence between defects in adjacent segments. The propagation rule is:

[0080] in, , For degree matrix, For the first Layer weight matrix.

[0081] 2. Partial (Spatiotemporal Dynamic Modeling): Spatial features output by GNN As input, it is fed into a Long Short-Term Memory (ST-LSTM) network with spatial attention mechanism. The core computational unit of ST-LSTM is similar to that of a traditional LSTM, containing an input gate, a forget gate, an output gate, and a memory unit. Its update formula is:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] in, It includes spatial features and environmental factors from the GNN output. for The hidden state of a moment contains spatiotemporal information from all past moments.

[0088] 3. Topological prior regularization term:

[0089] In the formula, For tunnel segments and The structural distance between them This is the regularization coefficient. This constraint ensures that the disease propagation patterns learned by the model do not seriously violate the basic mechanical principles of tunnel structures, thus maintaining physical rationality even in data-sparse regions. Model optimization and validation employ grid search combined with 10-fold cross-validation, with the mean square error between predicted and actual monitored values ​​≤0.05 and the accuracy of risk escalation probability prediction ≥85% as evaluation indicators.

[0090] The specific process by which the hybrid prediction model processes the evaluation results obtained in the second part is as follows: For the target segment node Enter its past Disease characteristic sequence at each time step , and the corresponding data of its adjacent nodes.

[0091] At each time step, the GNN aggregates the disease features of neighboring nodes, updates the node features, and obtains new features that incorporate spatial influences. This feature is then fed as input to the ST-LSTM network at each time step. The recurrent structure of the ST-LSTM learns the evolutionary trend of the disease over time. Ultimately, the future will be output. The predicted values ​​of disease status at each time step include the predicted values ​​of indicators at each criterion level (such as crack width and water seepage in the next 3 months), the evolution probability of health level (such as the probability that the current level 3 disease will be upgraded to level 4 in 6 months is 85%), and risk area identification.

[0092] The technical advantages of this model are as follows: by coupling the spatial modeling capabilities of GNN and the temporal modeling capabilities of LSTM, it can more accurately predict the coupled evolution trend of shield tunnel segment defects under complex spatiotemporal conditions; the model's prediction results are traceable, for example, the prediction that a certain crack will accelerate its expansion can be explained as "mainly affected by the continuous increase in the misalignment of the adjacent joint A", providing a clear basis for treatment decisions; it realizes the transformation from passive treatment to proactive pre-treatment, and the decision-making level can formulate intervention plans in advance based on the risk escalation probability predicted by the model, significantly reducing operation and maintenance costs and risks.

[0093] Step S3.3: Match the prediction results with the treatment plans in the case library to obtain the optimal treatment plan, and perform digital twin simulation based on the treatment plan to generate a pre-treatment plan.

[0094] The decision-making layer automatically generates or recommends accurate pre-treatment plans based on the disease evolution prediction results output by the model layer. The system takes the current disease status as input, first obtains a robust basic plan by matching historical similar scenarios through case reasoning, and then introduces a reinforcement learning mechanism in the digital twin simulation environment to perform multiple rounds of deduction and optimization of the parameters of the basic plan.

[0095] The execution and feedback layer is responsible for translating the plans formulated by the decision-making layer into specific construction actions, collecting data from the entire process and feeding it back to the system. It automatically encapsulates the entire process data of each treatment task into standardized digital cases, establishing multi-dimensional relationships between cases to form a case network. After matching similar cases, Bayesian optimization is used to adaptively adjust the construction parameters, and incremental learning drives the collaborative and continuous evolution of the prediction model and treatment strategy.

[0096] Step S4: Evaluate the effectiveness of the pre-treatment plan and generate a pre-treatment efficacy evaluation report.

[0097] Part Four: The scientific evaluation phase, with its core objective, aims to systematically and quantitatively verify the effectiveness of dynamic prediction-driven pre-treatment measures, ensuring the achievement of pre-treatment goals. Simultaneously, the verification results are transformed into structured knowledge to drive the iterative optimization of diagnostic models, predictive models, and treatment decision-making logic, forming a continuous improvement mechanism of "evaluation-feedback-optimization." The first step involves constructing a multi-dimensional effectiveness evaluation index system and collecting data. The calculation formulas for the effectiveness evaluation indicators are as follows: 1. Disease relief rate ( )

[0098] In the formula, For the first Weighting of disease types The quantitative value of the disease before treatment, It is the quantitative value of the disease after treatment.

[0099] 2. Structural performance restitution factor (PRC):

[0100] In the formula, Structural performance is scored based on calculations such as elastic wave velocity and strain distribution. For the structural performance evaluation after the pretreatment is implemented, For structural performance evaluation before pretreatment implementation, The theoretical standard score for a structure in a state of perfect health.

[0101] 3. Durability Index (DI):

[0102] In the formula, To correct the degradation rate of material properties, accelerated aging tests can be used to simulate the degradation. The total service life of the design or the expected evaluation period.

[0103] After the pretreatment construction is completed, the testing equipment and technology from Part 1 are used again to periodically retest the target pipe segment. The evaluation is carried out around three core dimensions: assessment of the degree of visual relief of the disease (calculation of crack width shrinkage rate, seepage reduction rate, and settlement stability coefficient, and weighted to obtain the comprehensive disease relief rate, with a target RR≥80%), assessment of the structural performance recovery status (quantitative calculation of PRC, with a target value≥0.85), and assessment of the long-term durability of the pretreatment measures (forming the effective life of the DI prediction measures).

[0104] Subsequently, the evaluation results were visualized and analyzed in depth using digital twin technology. A variational autoencoder was used to detect the difference between the predicted and measured values. If the reconstruction error exceeded a threshold, the treatment effect was deemed unsatisfactory. The retest data was synchronized to the digital twin of the pipe segment disease and compared in three dimensions with the predicted state of the disease before pre-treatment and the expected effect of the digital twin pre-simulation. The differences were highlighted with intuitive colors. Engineers could click on the unsatisfactory areas to trace back to specific parameters and input factors, enabling rapid identification of the root cause of the poor effect.

[0105] Finally, closed-loop feedback and system optimization are implemented: when the desired effect is achieved, successful cases are packaged into case packages and stored in the knowledge base to enhance recommendation weight; when the desired effect is not achieved, the system reverse-optimizes the risk judgment threshold of the diagnostic model, uses deviation data as new samples for incremental training of the prediction model, and optimizes the parameter matching rules of the decision layer, thereby achieving a closed-loop correction across the entire chain from assessment to diagnosis, prediction, and decision-making. The system also supports macro-level statistics and report generation for long-term operational efficiency, and automatically generates pre-treatment efficiency evaluation reports periodically, showcasing the iterative improvement trends of key performance indicators and the system's own capability indicators.

[0106] Example 2 This embodiment provides an integrated diagnosis and assessment system for shield tunnel segment defects, including: The data sensing module is configured to acquire and fuse multimodal raw data of the shield tunnel segment structure to obtain defect sensing data. The fault diagnosis module is configured to perform fault diagnosis based on the disease perception data, using a pre-trained machine learning model and the weights of health evaluation indicators, and obtain the fault diagnosis results. The fault treatment module is configured to predict disease trends based on fault diagnosis results using a pre-trained hybrid prediction model, and generate a pre-treatment plan based on the prediction results. The treatment evaluation module is configured to evaluate the effectiveness of the pre-treatment plan and generate a pre-treatment efficacy evaluation report.

[0107] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0108] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0109] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0110] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the integrated diagnosis and evaluation method for shield tunnel segment defects as described in Embodiment 1 above.

[0111] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the integrated diagnosis and evaluation method for shield tunnel segment defects as described in Embodiment 1 above.

[0112] Example 5 This embodiment provides a computer program product or computer program, including computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the integrated diagnosis and treatment assessment method for shield tunnel segment defects described in Embodiment 1 above.

[0113] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0114] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0118] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for integrated diagnosis and assessment of defects in shield tunnel segments, characterized in that, include: Multimodal raw data of shield tunnel segment structure are acquired and fused to obtain defect perception data; Based on the disease perception data, fault diagnosis is performed using a pre-trained machine learning model and the weights of health evaluation indicators to obtain fault diagnosis results. Based on the fault diagnosis results, a pre-trained hybrid prediction model is used to predict the disease trend, and a pre-treatment plan is generated based on the prediction results. An evaluation report on the effectiveness of the pre-treatment plan is generated based on its implementation.

2. The integrated diagnosis and assessment method for shield tunnel segment defects as described in claim 1, characterized in that, The multimodal raw data of the shield tunnel segment structure are obtained by fusing them to obtain defect perception data; Acquire multimodal raw data of shield tunnel segment structure; The original multimodal data is preprocessed, and the preprocessed multimodal data is fused to obtain disease perception data.

3. The integrated diagnosis and assessment method for shield tunnel segment defects as described in claim 2, characterized in that, The multimodal raw data includes the geometry of the shield tunnel segment structure, images of apparent defects, internal defects, and environmental and load conditions.

4. The integrated diagnosis and assessment method for shield tunnel segment defects as described in claim 1, characterized in that, The method of performing fault diagnosis based on disease perception data, using a pre-trained machine learning model and the weights of health evaluation indicators, yields fault diagnosis results, including: Based on tunnel health levels, key factors of tunnel health, and tunnel health assessment indicators, a health evaluation system for shield tunnels is constructed. Using an improved analytic hierarchy process, the weights of different health evaluation indicators in the health evaluation system of shield tunnels are calculated. Based on the disease perception data, fault diagnosis is performed using a pre-trained machine learning model and the weights of health evaluation indicators to obtain fault diagnosis results.

5. The integrated diagnosis and assessment method for shield tunnel segment defects as described in claim 4, characterized in that, The tunnel's health status is divided into four levels; The key factors for tunnel health include material deterioration, water leakage, structural deformation, and settlement. Material degradation includes the width of segment cracks, the area of ​​spalling and missing pieces, and the concrete strength ratio; Leakage includes pH value, leakage volume per unit time, and leakage area; Structural deformation includes circumferential joint width, longitudinal joint width, transverse misalignment, radial misalignment, bolt stress-strength ratio, horizontal convergence deformation rate, and vertical convergence deformation rate. Settlement includes cumulative settlement and differential settlement.

6. The integrated diagnosis and assessment method for shield tunnel segment defects as described in claim 1, characterized in that, The process involves using a pre-trained hybrid prediction model to predict disease trends based on fault diagnosis results, and generating a pre-treatment plan based on the prediction results, including: The fault diagnosis results are preprocessed to obtain the preprocessed fault diagnosis results. Based on the pre-processed fault diagnosis results, the disease trend is predicted using a pre-trained hybrid prediction model to obtain the prediction results; The prediction results are matched with the treatment plans in the case library to obtain the optimal treatment plan. A digital twin simulation is then performed based on the treatment plan to generate a pre-treatment plan.

7. An integrated diagnosis and assessment system for shield tunnel segment defects, characterized in that, include: The data sensing module is configured to acquire and fuse multimodal raw data of the shield tunnel segment structure to obtain defect sensing data. The fault diagnosis module is configured to perform fault diagnosis based on the disease perception data, using a pre-trained machine learning model and the weights of health evaluation indicators, and obtain the fault diagnosis results. The fault treatment module is configured to predict disease trends based on fault diagnosis results using a pre-trained hybrid prediction model, and generate a pre-treatment plan based on the prediction results. The treatment evaluation module is configured to evaluate the effectiveness of the pre-treatment plan and generate a pre-treatment efficacy evaluation report.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the integrated diagnosis and treatment assessment method for shield tunnel segment defects as described in any one of claims 1-6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the integrated diagnosis and evaluation method for shield tunnel segment defects as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the integrated diagnosis and evaluation method for shield tunnel segment defects as described in any one of claims 1-6.

Images