A method for predicting the state of a shield tail seal in slurry shield construction

Abstract

This invention discloses a method for predicting the tail seal status in slurry shield tunneling, belonging to the field of shield tunneling technology. Addressing the problem that existing technologies rely on manual experience and cannot provide early warnings of seal failure, this method achieves its goal through the following steps: collecting multi-source time-series data of shield tunneling parameters and geological parameters; performing timestamp alignment, cleaning, standardization, and feature engineering on the data to construct a comprehensive feature vector including original parameters and derived features; inputting the feature sequence into a CNN-LSTM hybrid model for training to achieve prediction of tail grease pressure and classification output of seal failure probability; classifying the state into three levels—safe, caution, and dangerous—based on the failure probability and providing real-time warnings. This invention can identify sealing risks several steps in advance, integrates multi-source information to improve prediction accuracy, and possesses good engineering applicability and adaptive optimization capabilities.

Description

Technical Field

[0001] This invention relates to the field of tunnel boring machine (TBM) construction technology, and in particular to a method for predicting the tail seal status in slurry TBM construction. Background Technology

[0002] The shield tunneling method has advantages such as fast construction speed, small impact on the surrounding environment and high degree of automation. However, projects constructed using the shield tunneling method are usually more complex and the material consumption cost during construction is relatively high. Therefore, it is very important to scientifically and rationally evaluate the consumption of engineering materials.

[0003] During slurry shield tunneling, grease is one of the main consumable materials for the shield, serving functions such as lubrication, sealing, waterproofing, and slurry isolation, thus playing a significant role in improving the quality of shield construction and tunneling efficiency.

[0004] Currently, the control of the shield tail seal mainly relies on the experience of the tunnel boring machine operator, who manually adjusts the seal by observing the tail pressure index. This method cannot predict the risk of seal failure in advance, and by the time a leak is detected, serious consequences have often already occurred. In addition, the shield tail seal condition is affected by a complex coupling of multiple factors such as geology, tunneling, and grouting, and existing technologies lack an effective means to accurately predict this by integrating these multi-source time-series information. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention aims to provide a method for predicting the tail seal status in slurry shield tunneling. By collecting multi-source time-series data during slurry shield tunneling, preprocessing and feature engineering the data, constructing a feature vector sequence, and establishing a CNN-LSTM model to predict the tail grease pressure value / seal failure probability, the method achieves the goal of predicting the tail seal status.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] A method for predicting the tail seal status in slurry shield tunneling includes the following steps:

[0008] S1: Collect multi-source time-series data of shield tunneling construction;

[0009] S2: Preprocess and feature engineering the collected data to construct a sequence of feature vectors;

[0010] S3: Input the constructed feature vector sequence into the CNN-LSTM model for training and learning;

[0011] S4: Based on the model's output, obtain the shield tail grease pressure value or seal failure probability for several future rings.

[0012] Preferably, the multi-source time-series data includes shield tunneling parameters and geological parameters;

[0013] The tunneling parameters include: advance speed, penetration depth, cutterhead torque, total thrust, slurry chamber pressure, air chamber pressure, slurry inlet flow rate, slurry outlet flow rate, slurry inlet specific gravity, and slurry outlet specific gravity;

[0014] The geological parameters include: the depth of the tunnel centerline and the soil type.

[0015] Preferably, the preprocessing includes:

[0016] The tunneling parameters and geological parameters collected at different frequencies are timestamped and unified into a sequence in "rings".

[0017] Z-score normalization is applied to continuous numerical parameters;

[0018] The "soil layer type" is numerically represented using unique thermal coding.

[0019] Preferably, the feature engineering includes constructing derived features, wherein the derived features include at least one of the following:

[0020] The mud-water flow difference is the difference between the inlet flow rate and the outlet flow rate.

[0021] The mud-water specific gravity difference is the difference between the inlet slurry specific gravity and the outlet slurry specific gravity;

[0022] The thrust-velocity ratio is the ratio of total thrust to propulsion speed.

[0023] Parameter volatility is the standard deviation of the key parameter over the past several cycles.

[0024] Preferably, the feature vector sequence includes:

[0025] The standardized tunneling parameters, the burial depth in the standardized geological parameters, the soil layer type after thermal coding, and the derived features.

[0026] Preferably, the CNN-LSTM model inputs a sequence of feature vectors through a sliding window, which covers data from several previous rings.

[0027] Preferably, the CNN-LSTM model is a multi-task learning model that simultaneously outputs the predicted value of the shield tail grease pressure and the probability of seal failure.

[0028] Preferably, the loss function of the model is a weighted sum of regression loss and classification loss, wherein the regression loss uses mean squared error and the classification loss uses binary cross-entropy.

[0029] Preferably: In step S4, the shield tail sealing status level is classified according to the sealing failure probability:

[0030] The state is safe when the failure probability P < 0.3;

[0031] When 0.3 ≤ P < 0.7, the state is "Attention";

[0032] When P ≥ 0.7, the status is dangerous and an alarm is triggered.

[0033] Preferably, the method further includes providing operation suggestions or early warning information to the tunnel boring machine operator based on the prediction results.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] This invention provides a method for predicting the tail seal status in slurry shield tunneling. By analyzing time-series data trends, it can provide early warning several cycles before leakage occurs, giving operators sufficient response time. By integrating multi-dimensional parameters of geology, tunneling, and grouting to comprehensively consider the complex factors of tail seal, it avoids the limitations of judging by a single parameter. Through a deep learning model, it can be continuously updated and optimized as tunneling data accumulates, adapting to different geological conditions and engineering characteristics.

[0036] The shield tunneling parameter prediction method provided by this invention uses a BP neural network method to deeply mine the tunneling parameters of the shield machine, perform correlation analysis on the tunneling parameters, and perform regression analysis based on the correlation analysis to find the relationship between different tunneling parameters. The method uses machine learning to predict the tunneling parameters, so as to provide the shield machine operator with recommended values ​​for the tunneling parameters and improve the efficiency of shield tunneling. Attached Figure Description

[0037] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a method for predicting the tail seal state in slurry shield tunneling, as described in this invention.

[0039] Figure 2 This is a diagram of the CNN structure described in this invention;

[0040] Figure 3 This is a diagram of the LSTM structure described in this invention;

[0041] Figure 4 This is a structural diagram of the CNN-LSTM model described in this invention. Detailed Implementation

[0042] The present invention will be further described below with reference to the accompanying drawings:

[0043] A method for predicting the tail seal status in slurry shield tunneling, such as... Figure 1 As shown, it includes the following steps:

[0044] S1: Collect multi-source time-series data of the tunnel boring machine (TBM), mainly selecting two parts: TBM tunneling parameters and geological parameters. Tunneling parameters mainly include advance speed, penetration depth, cutterhead torque, total thrust, slurry chamber pressure, air chamber pressure, slurry inlet flow rate, slurry outlet flow rate, slurry inlet specific gravity, and slurry outlet specific gravity; geological parameters mainly include tunnel centerline burial depth and soil layer type (clay layer, silt layer, gravel layer).

[0045] S2: The tunneling and geological parameters collected at different frequencies are timestamped and unified into a sequence based on "rings". The collected data is cleaned, and all continuous numerical parameters other than "soil type" are Z-score standardized. The categorical variable "soil type" is converted into numerical features using one-hot encoding. Features are constructed by calculating the slurry flow rate difference (inlet flow rate - outlet flow rate), slurry specific gravity difference (inlet specific gravity - outlet specific gravity), thrust-velocity ratio (total thrust / advance speed), and parameter fluctuation (standard deviation of key parameters over the past N rings).

[0046] S3: After S2, each ring of data is no longer an isolated parameter, but a comprehensive feature vector (20-dimensional) that integrates the original data, statistics and derived information.

[0047] The feature vector includes the following parts:

[0048] ①Standardized original tunneling parameters (10 dimensions): advance speed, penetration depth, cutterhead torque, total thrust, slurry chamber pressure, air chamber pressure, slurry inlet flow rate, slurry outlet flow rate, slurry inlet specific gravity, and slurry outlet specific gravity;

[0049] ②Standardized geological parameters (1D): burial depth;

[0050] ③ Encoded geological parameters (3D): Unique thermal coding of soil layer type;

[0051] ④ Derivative characteristics (6 dimensions): Slurry flow rate difference, slurry specific gravity difference, thrust-velocity ratio, torque_past 10 rings_standard deviation, slurry chamber pressure_past 10 rings_standard deviation, total thrust_past 10 rings_standard deviation;

[0052] like Figure 2As shown, Convolutional Neural Networks (CNNs) have significant advantages in processing high-dimensional data, extracting local features, and automating feature learning. Based on the features mentioned above, a 20-dimensional vector is finally constructed. Traditional methods are prone to the "curse of dimensionality" problem when processing high-dimensional data, resulting in high computational complexity. CNNs, through convolutional layers and pooling layers, can reduce the number of model parameters, reduce computational complexity, and at the same time retain key information of the data.

[0053] like Figure 3 As shown, Long Short-Term Memory (LSTM) networks, as a special type of recurrent neural network (RNN), have significant advantages in processing time-series data and capturing long-term relational dependencies;

[0054] The core of LSTM lies in the "gating" mechanism, where the forget gate's main function is to selectively discard historical information, the update gate's function is to selectively store new input information at the current position, and the output gate's function is to output the hidden layer at the current position.

[0055] )

[0056] In the formula, Represents the state of the Forgotten Gate. It is the parameter matrix of the forget gate, the forget gate state, and the current input. The hidden layer at the previous position related;

[0057] The formula for calculating the updated door is as follows:

[0058]

[0059] This represents updating the state of the door. It updates the parameter matrix of the gate. Represents cell state information;

[0060] The processed feature sequence is input into the CNN-LSTM model. Figure 4 The CNN-LSTM model is trained as follows: Figure 4 As shown, the CNN-LSTM is connected to a fully connected layer, and the model's output includes predicted values ​​of grease pressure at the tail of the shield. (Regression Output) and Shield Tail Seal Failure Probability (Categorized output);

[0061] ;

[0062] );

[0063] Since it is multi-task learning, the total loss is the weighted sum of regression loss and classification loss;

[0064] Regression loss (L reg Mean squared error (MSE) is used:

[0065] ;

[0066] The classification loss (Lcls) uses binary cross-entropy:

[0067] ;

[0068] Total loss (Ltotal):

[0069] ;

[0070] S4: According to the model's output, when the predicted failure probability P < 0.3, the shield tail seal is in a safe state; when 0.3 ≤ P < 0.7, the shield tail seal is in a caution state, prompting the driver to pay attention to changes in relevant parameters; when P ≥ 0.7, the state is in danger, issuing an alarm to the main driver and prompting timely implementation of appropriate solutions.

[0071] S5: Provide operational suggestions or early warning information to the tunnel boring machine operator based on the prediction results.

[0072] The specific steps of step S2 are as follows:

[0073] (1) Data alignment

[0074] The tunneling parameters (high-frequency acquisition) and geological parameters (low-frequency acquisition) collected at different frequencies are timestamped and unified into a sequence based on "rings".

[0075] (2) Data cleaning

[0076] 1) Outlier handling: For numerical tunneling parameters, the box plot method is used for identification. Data points exceeding ±1.5 times the interquartile range (IQR) of the upper and lower quartiles are considered mild outliers, and data points exceeding ±3 times the IQR are considered extreme outliers. Extreme outliers are directly treated as missing values, and mild outliers are filled with data from the preceding and following loops according to the actual situation.

[0077] 2) Missing value handling: Use forward filling and linear interpolation to fill in some missing values. If the entire loop is missing and the data before and after it is stable, delete the data of that loop directly.

[0078] (3) Data standardization / normalization

[0079] To prevent features with large numerical ranges from overshadowing features with small numerical ranges and to accelerate model convergence, continuous numerical parameters are standardized.

[0080] Z-Score standardization: ;

[0081] in, These are the original eigenvalues. It is the mean of this feature on the training set. It is the standard deviation of the feature on the training set;

[0082] Min-Max normalization: ;

[0083] (4) Categorical variable coding

[0084] One-hot encoding is used for "soil type" to create a feature vector of length 3 to represent the three types of strata:

[0085] Clay: [1, 0, 0]; Silt: [0, 1, 0]; Gravel: [0, 0, 1];

[0086] (5) Feature construction

[0087] Based on existing parameters, four derivative features are constructed: “slurry flow rate difference”, “slurry specific gravity difference”, “thrust-velocity ratio”, and “parameter fluctuation”.

[0088] Slurry flow difference = Inlet slurry flow rate - Outlet slurry flow rate;

[0089] Mud-water specific gravity difference = inlet slurry specific gravity - outlet slurry specific gravity;

[0090] Thrust-speed ratio = Total thrust / Propulsion speed;

[0091] Parameter volatility: Standard deviation of key parameters (cutterhead torque, total thrust, slurry chamber pressure) over the past 10 rings.

[0092] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are only illustrative of the principles of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the present invention, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A method for predicting the tail seal state in slurry shield tunneling, characterized in that, Includes the following steps: S1: Collect multi-source time-series data of shield tunneling construction; S2: Preprocess and feature engineering the collected data to construct a sequence of feature vectors; S3: Input the constructed feature vector sequence into the CNN-LSTM model for training and learning; S4: Based on the model's output, obtain the shield tail grease pressure value or seal failure probability for several future rings.

2. The method for predicting the sealing state of the shield tail according to claim 1, characterized in that, The multi-source time-series data includes shield tunneling parameters and geological parameters; The tunneling parameters include: advance speed, penetration depth, cutterhead torque, total thrust, slurry chamber pressure, air chamber pressure, slurry inlet flow rate, slurry outlet flow rate, slurry inlet specific gravity, and slurry outlet specific gravity; The geological parameters include: the depth of the tunnel centerline and the soil type.

3. The method for predicting the sealing state of the shield tail according to claim 2, characterized in that, The preprocessing includes: The tunneling parameters and geological parameters collected at different frequencies are timestamped and unified into a sequence in "rings"; Z-score normalization is applied to continuous numerical parameters; The "soil layer type" is numerically represented using unique thermal coding.

4. The shield tail sealing state prediction method according to claim 3, characterized in that, The feature engineering includes constructing derived features, which include at least one of the following: The mud-water flow difference is the difference between the inlet flow rate and the outlet flow rate. The mud-water specific gravity difference is the difference between the inlet slurry specific gravity and the outlet slurry specific gravity; The thrust-velocity ratio is the ratio of total thrust to propulsion speed. Parameter volatility is the standard deviation of the key parameter over the past several cycles.

5. The method for predicting the sealing state of the shield tail according to claim 4, characterized in that, The feature vector sequence includes: The standardized tunneling parameters, the burial depth in the standardized geological parameters, the soil layer type after thermal coding, and the derived features.

6. The method for predicting the sealing state of the shield tail according to claim 1, characterized in that, The CNN-LSTM model takes a sequence of feature vectors as input through a sliding window, which covers data from several previous rings.

7. The method for predicting the sealing state of the shield tail according to claim 1, characterized in that, The CNN-LSTM model is a multi-task learning model that simultaneously outputs the predicted value of the shield tail grease pressure and the probability of seal failure.

8. The method for predicting the sealing state of the shield tail according to claim 7, characterized in that, The loss function of the model is a weighted sum of regression loss and classification loss, where the regression loss uses mean squared error and the classification loss uses binary cross-entropy.

9. The method for predicting the sealing state of the shield tail according to claim 1, characterized in that, In step S4, the shield tail seal status level is classified according to the seal failure probability: The state is safe when the failure probability P < 0.3; When 0.3 ≤ P < 0.7, the state is "Attention"; When P ≥ 0.7, the status is dangerous and an alarm is triggered.

10. The method for predicting the sealing state of the shield tail according to claim 1, characterized in that, The method also includes providing operational suggestions or early warning information to the tunnel boring machine operator based on the prediction results.

Images