Shield tunneling machine hydraulic system fault diagnosis method based on singular value decomposition and XGBoost
By combining singular value decomposition and XGBoost, the problems of high-dimensional data redundancy and nonlinear coupling in the hydraulic system of tunnel boring machines were solved, enabling high-precision fault diagnosis and real-time monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-20
- Publication Date
- 2026-04-03
AI Technical Summary
Existing technologies are insufficient to effectively reduce the redundancy of high-dimensional data in the hydraulic system of tunnel boring machines, and their diagnostic accuracy and stability are inadequate, making them unable to adapt to complex nonlinear coupling conditions.
Singular value decomposition is used to reduce the dimensionality of hydraulic system data, and the Extreme Gradient Boosting Tree (XGBoost) model is used for fault diagnosis. The main features are extracted by singular value decomposition and classified using the nonlinear learning capability of XGBoost.
It achieves high-precision fault diagnosis of hydraulic systems, reduces computational complexity, improves the generalization performance and noise resistance of the model, and is suitable for online real-time monitoring of tunnel boring machines.
Smart Images
Figure CN121786642A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tunnel boring equipment monitoring and fault diagnosis technology, and more specifically to a fault diagnosis method for the hydraulic system of a tunnel boring machine based on singular value decomposition and limit gradient lifting tree. Background Technology
[0002] As the core equipment in tunnel construction, the tunnel boring machine (TBM) relies on its propulsion hydraulic system for critical tasks such as cutterhead propulsion, attitude adjustment, segment assembly, and muck removal. It is the core power subsystem throughout the entire TBM operation. The operational status of the hydraulic system directly affects the safety and tunneling efficiency of the TBM. Any malfunction (such as internal or external leakage in the hydraulic cylinders, main pump abnormalities, valve blockage, or excessively high oil temperature) can lead to tunneling interruptions or even safety accidents. Therefore, real-time monitoring and fault diagnosis of the TBM's hydraulic system are of paramount engineering importance.
[0003] With the increasing intelligence of tunnel boring machines, a large number of sensors are installed on the hydraulic system to collect multi-dimensional operational data such as pressure, flow rate, temperature, displacement, and vibration. The resulting multi-source time-series data is characterized by high dimensionality, high redundancy, strong noise, and significant nonlinear coupling, posing a great challenge to the modeling and real-time calculation of fault diagnosis algorithms.
[0004] Currently, the academic and engineering communities mainly adopt the following technical approaches for fault diagnosis of the hydraulic system of tunnel boring machines: Threshold determination and rule-based expert system approach: Traditional methods mainly set sensor signal thresholds through empirical rules or expert knowledge, and determine a fault when the detected value exceeds the normal range. For example, an alarm is triggered based on hydraulic oil pressure fluctuations or flow rate changes exceeding a set threshold. This type of method is simple to implement, but it relies on human experience, the threshold is difficult to adjust dynamically, and it cannot handle nonlinear relationships under complex coupled operating conditions, resulting in high false alarm and false negative rates.
[0005] Linear dimensionality reduction methods based on principal component analysis (PCA) or singular value decomposition (SVD): Some researchers use PCA or SVD to reduce the dimensionality of high-dimensional hydraulic data to extract key features, and then use support vector machines (SVM) or neural networks for classification. Although this type of method is effective in reducing data dimensionality, it has two prominent problems: (1) the dimensionality reduction process only considers linear features and ignores the nonlinear coupling characteristics of hydraulic system operation data; (2) the subsequent classification algorithm is prone to overfitting on high-dimensional complex data, resulting in unstable diagnostic accuracy and sensitivity to noise.
[0006] Classification and diagnostic methods based on a single machine learning model: In recent years, some studies have used algorithms such as random forest (RF), gradient boosting tree (GBDT), backpropagation neural network or convolutional neural network (CNN) to learn and classify hydraulic system data. These methods can model nonlinear features to a certain extent, but they also have problems: (1) the high dimension of the input features leads to long model training time and complex parameters; (2) the lack of an effective dimensionality reduction mechanism, and the large amount of noise and redundant information in the original features reduces the model's generalization ability.
[0007] Therefore, how to provide a new method that can reduce input dimensionality, filter redundant information, improve diagnostic accuracy and model stability, and be suitable for online monitoring of shield tunneling hydraulic systems in complex environments is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0008] In view of the above problems, the present invention is proposed to provide a fault diagnosis method for the hydraulic system of a tunnel boring machine based on singular value decomposition and XGBoost to overcome or at least partially solve the above problems, so as to improve the diagnostic accuracy, reduce the computational complexity and enhance the noise resistance.
[0009] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a fault diagnosis method for the hydraulic system of a tunnel boring machine based on singular value decomposition and XGBoost, comprising the following steps: S1: Collect multi-source sensor operating data of the shield machine's hydraulic system to form a raw data matrix; S2: Perform singular value decomposition on the original data matrix, and select the top-ranked singular values based on their cumulative contribution rate. k Using the principal singular values and their corresponding eigenvectors, construct a low-dimensional feature matrix after dimensionality reduction. k >1; S3: Input the low-dimensional feature matrix into the pre-trained limit gradient boosting tree classification model, and output the fault diagnosis results of the shield machine hydraulic system by the limit gradient boosting tree classification model.
[0010] Preferably, in step S1, the multi-source sensor operating data includes sensor data from the propulsion pump, propulsion cylinder, control valve, and hydraulic lines; the data includes at least two of the following: pressure, pressure change rate, flow rate, temperature, displacement, and vibration parameters.
[0011] Preferably, in step S1, the rows of the original data matrix correspond to time samples, and the columns correspond to different sensor features.
[0012] Preferably, between step S1 and step S2, a preprocessing step is further included for the original data matrix, wherein the preprocessing includes at least one of denoising, standardization, and outlier removal.
[0013] Preferably, in step S2, the original data matrix is decomposed into singular values as follows:
[0014] In the formula, the matrix A The original data matrix, matrix U Let Σ be the basis vector matrix of the original data matrix in the new feature space, and let Σ be the singular value matrix, whose main diagonal elements are principal singular values. V is the orthogonal basis matrix of the original feature space; The low-dimensional feature matrix after dimensionality reduction From the matrix U The former k Columns and Singular Value Matrices S The former k The product of the principal singular values is obtained.
[0015] Preferably, in step S2, the step of selecting the previous value based on the cumulative contribution rate of the singular values... k The steps for obtaining a principal singular value include: Sort the singular values according to their cumulative contribution rate; Set a cumulative contribution rate threshold, and select the factors that cause the cumulative contribution rate to reach the threshold. k One primary singular value.
[0016] Preferably, in step S3, the extreme gradient boosting tree classification model uses classification and regression trees as base learners.
[0017] Preferably, the extreme gradient boosting tree classification model performs a training step on a historical dataset containing samples with the low-dimensional feature matrix and their corresponding fault category labels. The objective function of the training includes a loss term and a regularization term. During the training process, the first-order gradient information and second-order gradient information of the loss function are used for optimization.
[0018] Preferably, in step S3, the fault diagnosis result includes the probability value that the hydraulic system is in normal operation or in a specific fault state; the specific fault state includes at least one of the following: internal leakage of hydraulic cylinder, external leakage of hydraulic cylinder, main pump abnormality, valve group blockage, and excessively high oil temperature.
[0019] Preferably, the method is used for online real-time fault monitoring and diagnosis of the tunnel boring machine's propulsion hydraulic system.
[0020] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: This invention achieves matrix decomposition of high-dimensional hydraulic system data through singular value decomposition (SVD), automatically filters the feature components corresponding to the main singular values, effectively removes noise and redundant features while retaining core information, purifies high-dimensional feature information, and significantly reduces the input dimension of the model.
[0021] This invention utilizes XGBoost's nonlinear learning and ensemble optimization mechanism at the model level to improve the model's generalization performance, enabling high-precision classification and intelligent diagnosis of hydraulic systems under multiple operating conditions. This effectively overcomes the shortcomings of existing technologies, such as feature redundancy, severe overfitting, high computational complexity, and insufficient real-time performance.
[0022] The singular value decomposition of this invention can filter out the interference caused by abnormal fluctuations, making the input features more stable; XGBoost is trained on the dimensionality-reduced low-dimensional data, with low computational cost, and can achieve fault diagnosis in a short time. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0024] Figure 1 This is a flowchart of the fault diagnosis method for the hydraulic system of a tunnel boring machine based on singular value decomposition and limit gradient lifting tree according to the present invention. Figure 2 A block diagram for fault diagnosis based on SVD-XGBoost provided for embodiments of the present invention. Detailed Implementation
[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0026] This invention, based on a multi-scale consideration of the data center, server rack, and server, discloses a fault diagnosis method for the hydraulic system of a tunnel boring machine (TBM) based on singular value decomposition and XGBoost. By performing multi-dimensional feature acquisition, matrix dimensionality reduction, feature compression, and intelligent classification on the operating data of the TBM hydraulic system, high-precision diagnosis of the hydraulic system's operating status is achieved. For example... Figure 1 As shown, it includes the following steps: S1: Collect multi-source sensor operating data of the shield machine's hydraulic system to form a raw data matrix; S2: Perform singular value decomposition on the original data matrix, select the top k principal singular values and their corresponding eigenvectors based on the cumulative contribution rate of the singular values, and construct a low-dimensional feature matrix after dimensionality reduction, k>1; S3: Input the low-dimensional feature matrix into the pre-trained limit gradient boosting tree classification model, and the limit gradient boosting tree classification model outputs the fault diagnosis results of the shield machine hydraulic system.
[0027] In one embodiment, in step S1, the multi-source sensor operating data includes sensor data from the propulsion pump, propulsion cylinder, control valve, and hydraulic lines; the data includes at least two of the following: pressure, pressure change rate, flow rate, temperature, displacement, and vibration parameters.
[0028] In one embodiment, in step S1, the rows of the original data matrix correspond to time samples, and the columns correspond to different sensor features.
[0029] In step S1, operational data is collected from sensors deployed on multiple key components of the tunnel boring machine's propulsion hydraulic system. The collected signals include various parameters such as propulsion pump pressure, pressure change rate, oil temperature, oil flow rate, and rodless chamber pressure in the propulsion cylinder. By continuously collecting these multi-dimensional characteristic signals, a raw data matrix is constructed. A Each row represents a time sample, the number of samples is denoted as m, and each column represents a different sensor feature, the number of features is denoted as n.
[0030] In one embodiment, between step S1 and step S2, a preprocessing step is further included for the original data matrix. Preprocessing includes at least one of denoising, standardization, and outlier removal. During the data preprocessing stage, the original matrix is denoised, standardized, and outlier removed to ensure it has stable dimensions and numerical ranges, thereby guaranteeing the effectiveness and comparability of subsequent algorithmic processing.
[0031] In one embodiment, after obtaining the preprocessed data matrix A Next, in step S2, the high-dimensional matrix is reduced in dimensionality using singular value decomposition (SVD). SVD reduces the original matrix... A Decomposed into a product of three matrices:
[0032] In the formula, the matrix A The original data matrix, matrix U Let the original data matrix be the basis vector matrix in the new feature space. S The matrix is a singular value matrix, with its main diagonal elements being principal singular values. Vis the orthogonal basis matrix of the original feature space; The low-dimensional feature matrix after dimensionality reduction From the matrix U The former k The first column and the singular value matrix Σ k The product of the principal singular values is obtained.
[0033] In this embodiment, in step S2, the previous one is selected based on the cumulative contribution rate of the singular values. k The steps for obtaining a principal singular value include: Sort the outliers according to their cumulative contribution rate; Set a cumulative contribution rate threshold, and select the top performers that cause the cumulative contribution rate to reach the threshold. k One primary singular value.
[0034] In step S2, the elements on the main diagonal of the singular value matrix are, in order, singular values. s 1, s 2,..., s r ,and s 1≥ s 2≥...≥ s r >0.
[0035] Each singular value s i This corresponds to a set of independent energy components in the original data, and its magnitude reflects the degree to which the component contributes to the overall data variation.
[0036] By calculating the cumulative contribution rate of singular values, i.e., the proportion of the sum of the top few singular values to the sum of all singular values, this embodiment selects the top few singular values whose cumulative contribution rate reaches a set threshold. k A low-dimensional feature subspace is constructed using the principal singular values and their corresponding eigenvectors. A threshold value is typically set between 90% and 99%. Figure 2 As shown. At this point, the dimension-reduced matrix... From the matrix U The former k Columns and Singular Value Matrices S The former k The product of the principal singular values is obtained.
[0037] This process achieves effective compression of high-dimensional sensor data, which retains the main information of the hydraulic system's operating status while significantly reducing noise and redundant features, thus obtaining a low-dimensional, information-intensive, and computationally friendly feature representation.
[0038] In one embodiment, in step S3, the extreme gradient boosting tree classification model uses classification and regression trees as base learners.
[0039] In one embodiment, the extreme gradient boosting tree classification model performs a training step on a historical dataset containing samples with low-dimensional feature matrices and their corresponding fault category labels. The objective function of the training includes a loss term and a regularization term. During the training process, the first-order gradient information and second-order gradient information of the loss function are used for optimization.
[0040] In one embodiment, in step S3, the fault diagnosis result includes the probability value that the hydraulic system is in normal operation or a specific fault state; the specific fault state includes at least one of the following: internal leakage of hydraulic cylinder, external leakage of hydraulic cylinder, main pump abnormality, valve group blockage, and excessively high oil temperature.
[0041] In the specific execution of step S3, the above dimensionality reduction results are... Each row is used as a new input sample vector in the Extreme Gradient Boosting Tree (XGBoost) classification model for fault identification in hydraulic systems. The XGBoost model is an ensemble learning algorithm based on the gradient boosting principle. It approximates the true distribution of samples by constructing and stacking multiple decision trees. The output of each tree serves as the input for the next tree, minimizing the overall prediction error. In each iteration, the model learns from the prediction residuals of the previous stage, making the final prediction results more accurate.
[0042] In the implementation of this invention, such as Figure 2 As shown, the XGBoost model uses a Classification and Regression Tree (CART) as its basic learner and incorporates both a loss term and a regularization term into the objective function. The loss term measures the difference between the model's predicted values and the true labels, while the regularization term controls the complexity of the tree and prevents overfitting. During training, the algorithm utilizes both the first-order gradient (the first derivative of the loss function with respect to the predicted values) and the second-order gradient (the second derivative) to accelerate model convergence and improve classification accuracy. Finally, the model outputs a classification result vector, where each component represents the probability of the hydraulic system being in a specific operating condition, such as "normal operation," "internal leakage in the hydraulic cylinder," or "external leakage in the hydraulic cylinder."
[0043] During the model training phase, manually labeled historical data was divided into training and validation sets. Through multiple rounds of iterative training and cross-validation, the hyperparameters of XGBoost were optimized, including the learning rate, maximum tree depth, number of leaf nodes, sample sampling ratio, and L1 / L2 regularization coefficients. The trained model demonstrated high accuracy and robustness on the validation set, achieving a diagnostic accuracy of over 97%.
[0044] In one embodiment, the method is used for online real-time fault monitoring and diagnosis of the tunnel boring machine propulsion hydraulic system.
[0045] Through the above process, this invention realizes a fault diagnosis mechanism from high-dimensional raw data to low-dimensional feature extraction, and then to intelligent classification and recognition. Singular value decomposition ensures the stability and noise resistance of feature extraction, while limiting gradient boosting trees ensure the nonlinear expressive power and generalization performance of classification. The combination of the two forms a high-precision, high-efficiency, and engineering-feasible fault diagnosis framework, providing a new technical path for intelligent monitoring and early fault identification of tunnel boring machine hydraulic systems.
[0046] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0047] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A fault diagnosis method for a tunnel boring machine hydraulic system based on singular value decomposition and XGBoost, characterized in that, Includes the following steps: S1: Collect multi-source sensor operating data of the tunnel boring machine's hydraulic system to form a raw data matrix; S2: Perform singular value decomposition on the original data matrix, and select the top-performing singular values based on their cumulative contribution rate. k Using the principal singular values and their corresponding eigenvectors, construct a low-dimensional feature matrix after dimensionality reduction. k >1; S3: Input the low-dimensional feature matrix into the pre-trained limit gradient boosting tree classification model, and output the fault diagnosis results of the shield machine hydraulic system by the limit gradient boosting tree classification model.
2. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, In step S1, the multi-source sensor operating data includes sensor data from the propulsion pump, propulsion cylinder, control valve, and hydraulic lines; the data includes at least two of the following: pressure, pressure change rate, flow rate, temperature, displacement, and vibration parameters.
3. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, In step S1, the rows of the original data matrix correspond to time samples, and the columns correspond to different sensor features.
4. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, Between step S1 and step S2, there is also a step of preprocessing the original data matrix, which includes at least one of denoising, standardization and outlier removal.
5. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, In step S2, the original data matrix is decomposed into singular values as follows: In the formula, the matrix A The original data matrix, matrix U Let Σ be the basis vector matrix of the original data matrix in the new feature space, and let Σ be the singular value matrix, whose main diagonal elements are principal singular values. V is the orthogonal basis matrix of the original feature space; The low-dimensional feature matrix after dimensionality reduction From the matrix U The former k Columns and Singular Value Matrices Σ The former k The product of the principal singular values is obtained.
6. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 5, characterized in that, In step S2, the selection of the previous values based on the cumulative contribution rate of the singular values... k The steps for obtaining a principal singular value include: Sort the singular values according to their cumulative contribution rate; Set a cumulative contribution rate threshold, and select the factors that cause the cumulative contribution rate to reach the threshold. k One primary singular value.
7. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, In step S3, the extreme gradient boosting tree classification model uses classification and regression trees as base learners.
8. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, The extreme gradient boosting tree classification model performs a training step on a historical dataset containing samples with the low-dimensional feature matrix and their corresponding fault category labels. The objective function of the training includes a loss term and a regularization term. During the training process, the first-order gradient information and second-order gradient information of the loss function are used for optimization.
9. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to claim 1, characterized in that, In step S3, the fault diagnosis result includes the probability value that the hydraulic system is in normal operation or in a specific fault state; the specific fault state includes at least one of the following: internal leakage of hydraulic cylinder, external leakage of hydraulic cylinder, main pump abnormality, valve group blockage, and excessively high oil temperature.
10. The method for fault diagnosis of tunnel boring machine hydraulic system based on singular value decomposition and XGBoost according to any one of claims 1 to 9, characterized in that, The method is used for online real-time fault monitoring and diagnosis of the propulsion hydraulic system of tunnel boring machines.