A shield machine main bearing fault diagnosis method based on long tail data distribution

Abstract

This application relates to a fault diagnosis method for the main bearing of a tunnel boring machine (TBM) based on long-tail data distribution, belonging to the field of fault diagnosis technology. This method collects multi-sensor time-series data of the TBM main bearing and constructs a fault diagnosis model including a multi-node topology construction module, a dynamic spatiotemporal feature evolution capture module, and a dual-branch spatiotemporal graph neural network feature representation module. A combined strategy of weight balancing and cyclically varying gradient calibration learning rate is used to train the model. The data to be diagnosed is input into the trained model, and the fault diagnosis result is output. By modeling the spatiotemporal dependence of sensors through multi-node topology, capturing the evolutionary features of the working condition through the dynamic spatiotemporal module, and fusing local and global fault features through the dual-branch network, the combined training strategy addresses the gradient imbalance problem of long-tail data, effectively improving the accuracy, model stability, and robustness of fault identification at the tail of the TBM main bearing. This method is suitable for online condition monitoring and early fault warning of the TBM main bearing.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method for the main bearing of a tunnel boring machine based on long-tail data distribution. Background Technology

[0002] Tunnel boring machines (TBMs) are indispensable in major infrastructure projects such as urban rail transit and river-crossing tunnels. As the core load-bearing component of the TBM, the operating status of the main bearing directly determines the construction progress and project safety. Main bearings mostly adopt a three-row roller structure. Once they fail and the machine is shut down, replacement and repair will take several months and cause huge economic losses.

[0003] The core challenge in diagnosing main bearing faults in tunnel boring machines lies in the fact that fault data exhibits a typical long-tail distribution: the proportion of normal state samples is extremely high, while samples of fault categories such as rolling element scratches and abrasions are extremely scarce. Early fault characteristics are easily masked by wear signals, and by the time they develop into obvious faults, they are already irreversible.

[0004] Existing fault diagnosis methods have the following shortcomings: First, traditional vibration monitoring equipment is difficult to adapt to complex tunneling conditions, and existing data generation and resampling methods cannot overcome the limitations of the sample space; Second, deep learning models such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Transformers rely on balanced samples and are prone to overfitting due to scarce fault samples. Transfer learning and ensemble learning methods do not fully consider the evolution and temporal dependence of sensor features, resulting in insufficient learning of tail fault features; Third, under long-tail distribution, the gradient competition between the head and tail categories is unbalanced, the model parameter update is dominated by the head category, the tail feature representation ability is weak, and it is easy to get trapped in local extrema, making it impossible to achieve accurate diagnosis of main bearing tail faults.

[0005] Therefore, there is an urgent need in related technologies for a way to improve the accuracy and stability of tunnel boring machine main bearing fault diagnosis under long-tail data distribution. Summary of the Invention

[0006] Therefore, it is necessary to provide a method for diagnosing the main bearing faults of tunnel boring machines (TBMs) based on long-tailed data distribution, which can improve the accuracy and stability of fault diagnosis under long-tailed data distribution.

[0007] Firstly, this application provides a method for diagnosing main bearing faults in tunnel boring machines based on long-tail data distribution. The method includes: Collect time-series data from multiple sensors on the main bearing of the tunnel boring machine; A fault diagnosis model is constructed, comprising a multi-node topology construction module, a dynamic spatiotemporal feature evolution capture module, and a dual-branch spatiotemporal graph neural network feature representation module; The fault diagnosis model is trained using a combination strategy of weight balancing and cyclically varying gradient calibration learning rate. Input the data to be diagnosed into the trained fault diagnosis model, and output the fault diagnosis results of the tunnel boring machine's main bearing.

[0008] Optionally, in one embodiment of this application, the multi-node topology construction module includes: The collected multi-sensor time series data is divided into sub-time windows. High-dimensional features are extracted by joint encoding of convolutional neural network and long short-term memory network. Temporal information is preserved by combining position encoding. The correlation between sensors is calculated and the topology is constructed.

[0009] Optionally, in one embodiment of this application, the dynamic spatiotemporal feature evolution capture module includes: Calculate the sensor spatial relationship weight matrix and the cross-sub-time window temporal relationship weight matrix based on the topology; By fusing the spatial and temporal weight matrices, a dynamic spatiotemporal relationship weighted matrix is ​​obtained. The dynamic spatiotemporal relationship weighted matrix is ​​then fused with the correlation matrix to generate a weighted adjacency matrix.

[0010] Optionally, in one embodiment of this application, the dual-branch spatiotemporal graph neural network feature representation module includes: Based on a weighted adjacency matrix, local spatiotemporal interaction features are extracted through a fixed sliding window, and a multi-level message passing mechanism is used to complete node feature aggregation and updating.

[0011] Optionally, in one embodiment of this application, the dual-branch spatiotemporal graph neural network feature representation module further includes: Two parallel feature extraction branches are set up to extract the local and global time features of the main bearing fault, respectively. The two types of features are then concatenated and input into the classifier to obtain the fault classification result.

[0012] Optionally, in one embodiment of this application, the combined strategy of weight balancing and cyclic gradient calibration learning rate includes a gradient balancing mechanism and a cyclic gradient calibration learning rate strategy.

[0013] Optionally, in one embodiment of this application, the gradient balancing mechanism dynamically adjusts the training weights of head and tail class samples by designing a cross-entropy loss function, thereby balancing the contribution of different class samples to the model parameter update.

[0014] Optionally, in one embodiment of this application, the cyclic gradient calibration learning rate strategy adjusts the learning rate by periodically restarting and decaying to optimize the model parameter update process.

[0015] Secondly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the steps of the methods described in the various embodiments above.

[0016] Thirdly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the methods described in the various embodiments above.

[0017] The above-mentioned method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution has the following advantages compared with existing technologies: First, high-precision tail fault identification: Under extreme long-tail ratio (100:1), the fault identification accuracy reaches 99.248%, which is 10%~13% higher than the traditional CNN-LSTM model, effectively solving the problems of missed detection and false detection of tail faults; Second, high stability under multiple working conditions: the standard deviation of the model accuracy is ≤2.97% under different long-tail ratios, and the shield machine main bearing has strong adaptability to dynamic working conditions. Third, strong robustness of long-tail data: diagnostic performance fluctuation is ≤0.7% under imbalanced datasets, overcoming overfitting and gradient imbalance caused by sample scarcity, and meeting engineering monitoring needs. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution in one embodiment. Figure 2 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] In one embodiment, such as Figure 1 As shown, a method for diagnosing main bearing faults in tunnel boring machines based on long-tail data distribution is provided, including the following steps: S101: Collect time series data from multiple sensors on the main bearing of the tunnel boring machine.

[0021] In this embodiment of the application, a shield tunneling damage simulation test bench is used to test the tunneling machine at three speeds (1, 2, 3 r·min). -1 Nine operating conditions were constructed using an orthogonal combination of three axial thrust levels (10, 20, 30 kN). Wire cutting technology was used to process local defects on the surface of the rolling elements, namely scratches of 3mm×3mm, 3mm×5mm, 3mm×7mm, 5mm×3mm, and 7mm×3mm, and abrasions of 2mm, 4mm, and 6mm width. Each defective roller was run to a steady state under all nine operating conditions, and triaxial vibration signals were collected in real time by an accelerometer installed in the bearing housing.

[0022] Among them, the normal state is the head category, and the other eight fault states are all tail categories. The fault types and numbers are shown in Table 1:

[0023] Table 1. Failure Type Numbers of Shield Tunnel Main Bearing Meanwhile, to simulate the long-tail distribution of tunnel boring machine main bearing failure data, four scenarios with long-tail ratios {10:1, 20:1, 50:1, 100:1} were constructed, and the dataset was divided into a training set, a balanced test set, and a long-tail test set.

[0024] The number of dataset partitions is shown in Tables 2 and 3:

[0025] Table 2. Division of Training Set Sample Quantity for Tunnel Boring Machine Main Bearings

[0026] Table 3. Sample size breakdown of the long-tail test set for the main bearing of the tunnel boring machine. S102: Construct a fault diagnosis model that includes a multi-node topology construction module, a dynamic spatiotemporal feature evolution capture module, and a dual-branch spatiotemporal graph neural network feature representation module.

[0027] In one embodiment of this application, the multi-node topology construction module includes: The collected multi-sensor time series data is divided into sub-time windows. High-dimensional features are extracted by joint encoding of convolutional neural network and long short-term memory network. Temporal information is preserved by combining position encoding. The correlation between sensors is calculated and the topology is constructed.

[0028] In one embodiment of this application, the preprocessed multivariate time series master bearing data is... Cut into Each time window is divided into equal-length windows, and each time window is further divided into sub-time windows to obtain segmented signal characteristics:

[0029] in, Indexing sub-time windows , Sensor Index , The length of the sub-time window.

[0030] pass and The joint encoding strategy extracts high-dimensional features, and positional encoding is introduced to preserve temporal information:

[0031] in, (D is the feature dimension) For position encoding, It consists of 3 layers of one-dimensional convolution. It is a 2-layer LSTM.

[0032] The correlation between sensors is calculated using dot product similarity, and after normalization, the following is obtained:

[0033] in, Representation Feature Transformation Network For sub-time windows Sensors With sub-time windows Sensors The correlation coefficient, The function implements normalization.

[0034] Calculate the correlation between different sub-time windows and different sensors, and construct the topology. : Feature set :

[0035] Correlation matrix :

[0036] In one embodiment of this application, the dynamic spatiotemporal feature evolution capture module includes: Calculate the sensor spatial relationship weight matrix and the cross-sub-time window temporal relationship weight matrix based on the topology; By fusing the spatial and temporal weight matrices, a dynamic spatiotemporal relationship weighted matrix is ​​obtained. The dynamic spatiotemporal relationship weighted matrix is ​​then fused with the correlation matrix to generate a weighted adjacency matrix.

[0037] In one embodiment of this application, the amplitude of the signal change of the shield tunnel main bearing fault diagnosis sensor may exhibit asynchronous time-delay effects due to changes in dynamic operating conditions. A dynamic spatiotemporal weighting matrix is ​​designed. By integrating spatial interaction and temporal evolution characteristics, fault-related spatiotemporal patterns are enhanced. A spatial weight matrix is ​​calculated for the spatial dependencies of sensors within the same sub-time window. :

[0038] Meanwhile, in order to ensure that the sensor has a weight of 1 for itself and to suppress normalization error, the following steps are taken: After adjustment, the spatial relationship weight matrix is ​​obtained. :

[0039] in, For Hadamard product, Sub-time window Internal sensor and Spatial relationship weights between them.

[0040] A time weight matrix is ​​constructed to address the temporal dependencies across sub-time windows. :

[0041] in, It is a learnable, dynamic, spatiotemporal relationship adjustment matrix. For the Sigmoid function, for and The interactive splicing features.

[0042] By integrating the spatial and temporal weight matrices, a dynamic spatiotemporal relationship weighted matrix is ​​obtained. and correlation matrix Merge to generate a weighted adjacency matrix:

[0043] right and By fusion This enables refined modeling of spatiotemporal dependencies for the final weighted adjacency matrix.

[0044]

[0045] In one embodiment of this application, the dual-branch spatiotemporal graph neural network feature representation module includes: Based on a weighted adjacency matrix, local spatiotemporal interaction features are extracted through a fixed sliding window, and a multi-level message passing mechanism is used to complete node feature aggregation and updating.

[0046] In one embodiment of this application, the dual-branch spatiotemporal graph neural network feature representation module further includes: Two parallel feature extraction branches are set up to extract the local and global time features of the main bearing fault, respectively. The two types of features are then concatenated and input into the classifier to obtain the fault classification result.

[0047] In one embodiment of this application, the spatiotemporal graph neural network uses a graph convolutional network (GCN) as the basic graph structure feature extraction unit, and integrates the time series modeling capabilities of a long short-term memory network (LSTM). The core consists of three parts: a local feature extraction layer, a multi-level message passing layer, and a two-branch feature fusion and classification layer. against Using size The step size is The moving slider extracts and expands local spatiotemporal features within the topological structure:

[0048] in, , For feature expansion function, , Indicates each sliding window covers One sensor and A series of consecutive sub-time windows.

[0049] Aggregating neighbor features using a central node, and introducing multi-order powers. Control the message propagation range and update node characteristics:

[0050] in, For the first eigenmap function of order, , This is the updated feature set.

[0051] To reduce the interference of local noise within sub-time windows on feature learning, in the time dimension... To achieve multi-scale extraction of local and global temporal features, a mean-averaging operation is performed. This embodiment sets up two parallel feature extraction branches. Branch 1 serves as the local feature extraction branch, with its core neural network structure denoted as TGNN1 (Local Temporal Graph Neural Network Unit); Branch 2 serves as the global feature extraction branch, with its core neural network structure denoted as TGNN2 (Global Temporal Graph Neural Network Unit). Both TGNN1 and TGNN2 use the same graph-temporal convolutional component (denoted as Conv_GraphST) and temporal graph neural network component (denoted as TGNN_mk_v2). The only difference lies in the stride of Conv_GraphST: TGNN1 uses a smaller stride to preserve temporal detail, while TGNN2 uses a larger stride to achieve temporal downsampling, thus focusing on local and global temporal features respectively. The concatenated data is then input into a classifier to obtain the classification result. :

[0052]

[0053] in, For feature splicing operations, For classifiers, Number of fault categories For the first The predicted probability of the class of each sample.

[0054] In practical applications, the Conv_GraphST time window size is set to 2, the sliding step size of TGNN1 is 1, the sliding step size of TGNN2 is 2, the hidden layer dimension of the TGNN_mk_v2 ​​node aggregation layer is set to 32, and the output feature dimension is set to 16; the classifier: the hidden layer dimension of the fully connected layer is 32, the activation function is ReLU, the output layer dimension is consistent with the number of fault categories, and finally the predicted probability of each category is output through the Softmax activation function.

[0055] S103: The fault diagnosis model is trained using a combination strategy of weight balancing and cyclically varying gradient calibration learning rate.

[0056] In one embodiment of this application, the combined strategy of weight balancing and cyclic gradient calibration learning rate includes a gradient balancing mechanism and a cyclic gradient calibration learning rate strategy.

[0057] In one embodiment of this application, the gradient balancing mechanism dynamically adjusts the training weights of head and tail class samples by designing a cross-entropy loss function, thereby balancing the contribution of different class samples to the model parameter update.

[0058] In one embodiment of this application, to address the gradient imbalance problem in long-tail data, a balancing mechanism and a Cyclic Decaying Gradient Calibration Learning Rate (CGCLR) strategy are used to enhance learning of tail samples: In a long-tailed data distribution, the head and tail categories are severely imbalanced. A cross-entropy loss function is designed as follows:

[0059] in, The sample represents the model's prediction. for Probability distribution of categories It is the loss function with respect to the parameters gradient, Indicates the learning rate. Indicates model parameters, Indicates the first The sample corresponds to the first The true label indicative value of the category. For model parameters. gradient:

[0060] in, The total gradient for model training, For a single sample The gradient.

[0061] Calculate the total gradient update and model parameter update, and quantify the contributions of the head and tail gradients:

[0062]

[0063]

[0064] Among them, when hour, The head class dominates, and model parameter updates are heavily biased towards the head class space.

[0065] In one embodiment of this application, the cyclic gradient calibration learning rate strategy adjusts the learning rate by periodically restarting and decaying, thereby optimizing the model parameter update process.

[0066] In one embodiment of this application, a cyclically varying gradient calibration learning rate strategy is employed to overcome the limitations of the definition of local extrema. Within the first restart cycle Adjustment coefficient for each training cycle:

[0067] in, Indicates the number of restarts during the decay period; This indicates the initial decay period, set to 30. This represents the decay period multiplication factor, which can be set to 1.2.

[0068] The final learning rate is calculated using the adjustment factor:

[0069] in, Indicates the first Maximum learning rate per restart cycle This represents the learning rate decay factor. for , Set to 0.001, Set it to 0.5.

[0070] Initially, a large learning rate is provided to adapt features; in the middle stage, the learning rate is decayed to weaken the dominance of the head gradient; and periodic restarts are used to avoid local optima.

[0071] S104: Input the data to be diagnosed into the trained fault diagnosis model and output the fault diagnosis results of the tunnel boring machine main bearing.

[0072] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0073] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 2As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0074] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0075] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0076] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0077] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0078] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0079] 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 computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for diagnosing main bearing faults in tunnel boring machines based on long-tail data distribution, characterized in that, The method includes: Collect time-series data from multiple sensors on the main bearing of the tunnel boring machine; A fault diagnosis model is constructed, comprising a multi-node topology construction module, a dynamic spatiotemporal feature evolution capture module, and a dual-branch spatiotemporal graph neural network feature representation module; The fault diagnosis model is trained using a combination strategy of weight balancing and cyclically varying gradient calibration learning rate. Input the data to be diagnosed into the trained fault diagnosis model, and output the fault diagnosis results of the tunnel boring machine's main bearing.

2. The method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution according to claim 1, characterized in that, The multi-node topology construction module includes: The collected multi-sensor time series data is divided into sub-time windows. High-dimensional features are extracted by joint encoding of convolutional neural network and long short-term memory network. Temporal information is preserved by combining position encoding. The correlation between sensors is calculated and the topology is constructed.

3. The method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution according to claim 1, characterized in that, The dynamic spatiotemporal feature evolution capture module includes: Calculate the sensor spatial relationship weight matrix and the cross-sub-time window temporal relationship weight matrix based on the topology; By fusing the spatial and temporal weight matrices, a dynamic spatiotemporal relationship weighted matrix is ​​obtained. The dynamic spatiotemporal relationship weighted matrix is ​​then fused with the correlation matrix to generate a weighted adjacency matrix.

4. The method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution according to claim 3, characterized in that, The dual-branch spatiotemporal graph neural network feature representation module includes: Based on a weighted adjacency matrix, local spatiotemporal interaction features are extracted through a fixed sliding window, and a multi-level message passing mechanism is used to complete node feature aggregation and updating.

5. The method for diagnosing main bearing faults of a tunnel boring machine based on long-tail data distribution according to claim 4, characterized in that, The dual-branch spatiotemporal graph neural network feature representation module further includes: Two parallel feature extraction branches are set up to extract the local and global time features of the main bearing fault, respectively. The two types of features are then concatenated and input into the classifier to obtain the fault classification result.

6. The method for diagnosing main bearing faults of a tunnel boring machine based on long-tail data distribution according to claim 1, characterized in that, The combined strategy of weight balancing and cyclic gradient calibration learning rate includes a gradient balancing mechanism and a cyclic gradient calibration learning rate strategy.

7. The method for diagnosing the main bearing fault of a tunnel boring machine based on long-tail data distribution according to claim 6, characterized in that, The gradient balancing mechanism dynamically adjusts the training weights of head and tail class samples by designing a cross-entropy loss function, thereby balancing the contribution of different class samples to the model parameter updates.

8. The method for diagnosing main bearing faults of a tunnel boring machine based on long-tail data distribution according to claim 6, characterized in that, The cyclic gradient calibration learning rate strategy adjusts the learning rate through periodic restarts and decays, optimizing the model parameter update process.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images