Tunnel boring machine tool remaining service life prediction method based on deep learning

Abstract

The invention relates to the technical field of tunnel boring machine life prediction, and discloses a tunnel boring machine tool remaining service life prediction method based on deep learning. The method comprises the following steps: acquiring internal and external related data of the tunnel boring machine, processing missing values of the acquired data, removing noise and abnormal values, performing normalization processing, performing deep learning model training by applying the processed data, constructing a deep learning model, dividing the data into a training set, a verification set and a test set, and performing model training optimization. Inputting training data, optimizing model parameters by using a back propagation algorithm, judging the convergence condition of the model and the performance on a verification set, determining whether to continue training or not, predicting the residual service life of the tunnel boring machine cutter, predicting the test set by using the trained model, and calculating the residual service life prediction value of each cutter. And carrying out prediction value judgment, returning to the model training optimization step when abnormity occurs, and applying the trained model to an actual production environment.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel boring machine life prediction, and specifically to a method for predicting the remaining service life of a tunnel boring machine tool based on deep learning. Background Art

[0002] Tunnel boring machine is an important mechanical equipment used in underground engineering construction, mainly used for tunnel excavation. Tunnel boring machine creates a uniform space in the soil, builds tunnels, and installs lining structures to ensure the stability of the tunnel. Tunnel boring machine is generally widely used in urban subways, railways, highways, water conservancy projects and other projects. The use of tunnel boring machine can be carried out in a closed environment, reducing interference with the surface, while improving the efficiency and safety of construction. Among them, the cutter is the core component of the tunnel boring machine for soil cutting. Its performance directly determines the cutting efficiency of the tunnel boring machine. The cutter must be able to effectively cope with different soil types, such as soft soil, hard soil, rock, etc. The design and processing accuracy of the cutter affect the final shape and quality of the tunnel. Its stability is crucial to the precise control during the construction process. By monitoring the wear of the cutter, potential problems can be discovered in advance and preventive measures can be taken to avoid construction interruptions or equipment damage.

[0003] At present, many TBM tool remaining service life prediction models have failed to establish an effective feedback mechanism and are unable to adjust the prediction model according to the latest operating data and the actual status of the tool. As a result, the long-term reliability of the TBM tool remaining service life prediction has decreased and the tool cannot be replaced in time, which has reduced the working efficiency of the TBM. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for predicting the remaining service life of a tunnel boring machine tool based on deep learning. The method has the following features: by collecting real-time monitoring data on the vibration, temperature, pressure and rotation speed of the tunnel boring machine tool, soil type, soil parameters, water level and temperature data, processing the missing values ​​of the above collected data, removing noise and outliers, and performing normalization processing to make them within the same range, putting the processed data into a deep learning model for training, generating statistical mean and variance features through time series analysis, building a deep learning model, selecting the TensorFlow deep learning framework and the LSTM model architecture, dividing the data into a training set, a validation set and a test set, performing model training optimization, inputting the training data and using the back propagation algorithm to optimize the model parameters, and judging The convergence of the model and the performance on the validation set are used to decide whether to continue training and predict the remaining service life of the tunnel boring machine tool. The trained model is used to predict the test set, and the predicted value of the remaining service life of each tool is calculated and judged. When an abnormality occurs, the model training optimization step is returned to, and the trained model is applied to the actual production environment. The tool status is monitored in real time and the remaining service life of the tunnel boring machine tool is updated. Through the prediction of the remaining service life of the tool, the effectiveness and accuracy of the prediction value are judged and real-time feedback is given. The prediction model is adjusted according to the latest operating data and the actual status of the tool, so that the predicted value of the remaining service life of the tunnel boring machine tool is reliable, the tool can be replaced in time, and the working efficiency of the tunnel boring machine is improved, etc., which solves the above problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for predicting the remaining service life of a tunnel boring machine tool based on deep learning, comprising the following steps:

[0006] S1. Collect real-time monitoring data of tunnel boring machine tool vibration, temperature, pressure, speed, soil type, soil parameters, water level and temperature data;

[0007] S2. Calculate the missing values ​​of all the collected data, remove noise and outliers, and perform normalization so that the above data are all within the range of [0, 1];

[0008] S3, build a deep learning model for tool life prediction, select the TensorFlow deep learning framework and LSTM model architecture, and divide all the processed data in S2 into training set, validation set and test set;

[0009] S4: Perform deep learning model training optimization, input training data and use back propagation algorithm to optimize model parameters, judge the convergence of the model and the performance on the validation set, and enter S5 after the deep learning model training is completed, otherwise continue to loop S4;

[0010] S5. Use the trained model to predict the test set, calculate the predicted value of the remaining service life of each tool, and make a judgment on the predicted value. If an abnormality occurs, return to S4 for retraining.

[0011] S6. Apply the trained model to the actual use environment to monitor the tool status in real time and update the remaining service life of the tunnel boring machine tool.

[0012] Preferably, the missing value calculation expression of the collected data in S2 is as follows:

[0013]

[0014] In the formula, Sx i represents the data after missing filling, n represents the number of valid data, Sx j Indicates missing data, including tunnel boring machine tool vibration data, temperature data, pressure data, speed data, soil type data, soil parameter data, water level data and temperature data, and j indicates the number of missing values.

[0015] Preferably, the formula for removing all collected data noise in S2 is as follows:

[0016]

[0017] In the formula, Qz i represents the denoised tunnel boring machine tool vibration data, temperature data, pressure data, speed data, soil type data, soil parameter data, water level data and temperature data, N represents the size of the moving window, Qz t Represents the original value of the time series data at time t.

[0018] Preferably, the formula for removing all abnormal values ​​of collected data in S2 is as follows:

[0019]

[0020] In the formula, z represents Z-Score, Dq represents the current value, Jz represents the data mean, and σ represents the data standard deviation.

[0021] Preferably, the formula for normalizing all collected data in S2 is as follows:

[0022]

[0023] In the formula, Gy represents the normalized value, Ys represents the original data, min(Ys) and max(Ys) represent the minimum and maximum values ​​in the original data set.

[0024] Preferably, the S3 constructs a deep learning model and divides the data into a training set ratio of 70%, a validation set ratio of 15% and a test set ratio of 15%, and the data of the training set, validation set and test set are processed tunnel boring machine tool vibration data, temperature data, pressure data, rotation speed data, soil type data, soil parameter data, water level data and temperature data.

[0025] Preferably, S4 determines the convergence of the model and the performance on the validation set by calculating the mean square error loss function, and the formula is as follows:

[0026]

[0027] In the formula, L train represents the mean square error, m represents the number of samples, which is derived from the proportion of the training set, k represents the counting subscript, and Zs k represents the true value, which is extracted from the training set, Yc k Represents the model prediction value.

[0028] Preferably, the convergence judgment criteria of the S4 model are as follows:

[0029] ΔL train =|L train (t)-L train (t-1)|

[0030] ΔL val =|L val (t)-L val (t-1)|

[0031] Among them, ΔL train Represents the change in training loss, ΔL val Represents the change in validation loss, t represents the current epoch index. When the change is less than the convergence threshold ∈ in three consecutive epochs, the model is judged to be converged, and the model no longer needs to be trained.

[0032] Preferably, S5 uses the trained model to predict the test set and calculate the predicted value of the remaining service life of each tool, and the formula is as follows:

[0033] RUL j =f(X j ,θ)

[0034] In the formula, RUL j Represents the predicted value of the remaining service life of the tool, X j represents the input feature vector of the jth tool, f represents the trained prediction model, and θ represents the parameters of the model, which are optimized through training data during the training process.

[0035] Preferably, S5 compares the predicted value of the remaining service life of the tool with the actual remaining service life value Sjz of the tool to determine the validity of the prediction model, and the determination criteria are as follows:

[0036] When | RUL j -Sjz j |<1, the model prediction is considered valid. j -Sjz j When |≥1, it means that the model prediction is invalid, and it falls back to S4, RUL j represents the remaining service life of the jth tool predicted by the model, Sjz j It represents the real remaining service life of the jth tool, that is, the life value of the tool in actual work.

[0037] Compared with the prior art, the present invention provides a method for predicting the remaining service life of a tunnel boring machine tool based on deep learning, which has the following beneficial effects:

[0038] The present invention collects real-time monitoring data of tunnel boring machine tool vibration, temperature, pressure and rotation speed, soil type, soil parameters, water level and temperature data, processes the missing values ​​of the above collected data through formulas, removes noise and outliers, and performs normalization processing to make them within the same range, puts the processed data into a deep learning model for training, generates statistical mean and variance characteristics through time series analysis, builds a deep learning model, selects the TensorFlow deep learning framework and LSTM model architecture, divides the data into training set, validation set and test set, performs model training optimization, inputs training data and uses the back propagation algorithm to optimize the model parameters, and judges the convergence of the model and the validation set. The performance of the tool is used to decide whether to continue training and predict the remaining service life of the tunnel boring machine tool. The trained model is used to predict the test set, and the predicted value of the remaining service life of each tool is calculated and judged. When an abnormality occurs, it falls back to the model training optimization step and applies the trained model to the actual production environment. The tool status is monitored in real time and the remaining service life of the tunnel boring machine tool is updated. By predicting the remaining service life of the tool, the effectiveness and accuracy of the predicted value are judged and real-time feedback is given. The prediction model is adjusted according to the latest operating data and the actual status of the tool, so that the predicted value of the remaining service life of the tunnel boring machine tool is reliable, the tool can be replaced in time, and the working efficiency of the tunnel boring machine is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Many current prediction models for the remaining service life of tunnel boring machine tools have failed to establish an effective feedback mechanism and are unable to adjust the prediction model based on the latest operating data and the actual status of the tool, which reduces the long-term reliability of the prediction of the remaining service life of the tunnel boring machine tool and makes it impossible to replace the tool in time, thus reducing the working efficiency of the tunnel boring machine. To this end, a method for predicting the remaining service life of tunnel boring machine tools based on deep learning is proposed. Figure 1 , the method comprises the following steps:

[0042] S1. Collect real-time monitoring data of tunnel boring machine tool vibration, temperature, pressure and speed, and external environmental data such as soil type, soil parameters, water level and temperature;

[0043] By collecting and integrating various status data of the tunnel boring machine cutters, we can gain an in-depth understanding of the performance of the cutters under actual working conditions. These data are monitored by high-precision sensors and transmitted to the central control system in real time to ensure dynamic adjustment and real-time feedback during the construction process. At the same time, by collecting key external environmental data such as soil type, soil parameters, moisture content, density, water level and ambient temperature, the system can evaluate the working efficiency of the cutters under variable geological conditions. Combining this information and using deep learning algorithms, we can establish a cutter performance prediction model to achieve intelligent prediction of the remaining service life of the cutters.

[0044] S2. Calculate the missing values ​​of all the collected data, remove noise and outliers, and perform normalization so that the above data are all within the range of [0, 1];

[0045] The formula for using mean interpolation to process missing values ​​of the above collected data is as follows:

[0046]

[0047] Mean interpolation can effectively fill in missing values ​​in the data, ensure the integrity of the data set, avoid the reduction of sample size due to missing values, and thus ensure the adequacy of model training. In the formula, Sx i represents the data after missing filling, n represents the number of valid data, Sx jIndicates missing data, including tunnel boring machine tool vibration, temperature, pressure, rotation speed, soil type, soil parameters, water level and temperature data. j represents the number of missing values. By filling the missing values ​​with the mean of the same feature, the impact on data distribution can be reduced, the overall trend of the data can be maintained, and the deviation of model training can be reduced.

[0048] The formula for removing noise from collected data using the moving average method is as follows:

[0049]

[0050] The moving average method calculates the average value of data points to reduce short-term fluctuations and noise, making the data smoother, so that potential trends and cycles can be identified more clearly. In the formula, Qz i represents the denoised tunnel boring machine tool vibration, temperature, pressure, speed, soil type, soil parameters, water level and temperature data, N represents the size of the moving window, Qz t Represents the original value of the time series data at time t. This method can effectively highlight the signal part in the data and help the model better capture the key factors affecting tool performance;

[0051] The formula for removing data outliers using the Z-Score method is as follows:

[0052]

[0053] The Z-Score method standardizes the data to make it conform to the standard normal distribution, which makes it easier to identify outliers and quickly locate data points with significant deviations. In the formula, z represents the Z-Score, Dq represents the current value, Jz represents the data mean, and σ represents the data standard deviation. After removing outliers, the model training process is not easily disturbed by extreme values, thereby improving the robustness and accuracy of the model and reducing prediction risks.

[0054] The formula for normalizing the processed data is as follows:

[0055]

[0056] Normalization can scale data of different dimensions and ranges to the same standard range, so that features can be directly compared, providing a unified input basis for model training. In the formula, Gy represents the normalized value, Ys represents the original data, mmin(Ys) and max(Ys) represent the minimum and maximum values ​​in the original data set. In the deep learning model training process, normalization helps to accelerate the convergence speed of the algorithm, improve training efficiency, and reduce training time.

[0057] S3, build a deep learning model for tool life prediction, select the TensorFlow deep learning framework and LSTM model architecture, and divide all the processed data in S2 into training set, validation set and test set;

[0058] The data of the training set, validation set and test set are derived from the processed data of tunnel boring machine tool vibration, temperature, pressure, rotation speed, soil type, soil parameters, water level and temperature;

[0059] TensorFlow provides a flexible way to build deep learning models, can handle complex modeling requirements, and supports GPU acceleration, which helps speed up the model training process. The LSTM network is specially designed to process and predict time series data. It has the ability to remember long-term and short-term dependencies and can better capture the dynamic changes in the use of tools. The data is divided into a training set with a ratio of 70%, so that the model can learn more data features and patterns and improve its generalization ability. The validation set ratio of 15% can prevent the model from overfitting the training data and ensure that the model can maintain good performance on unseen data. The test set ratio of 15% ensures that the test results of the model are not affected by the training or validation process, thereby providing more reliable performance indicators;

[0060] S4: Perform deep learning model training optimization, input training data and use back propagation algorithm to optimize model parameters, judge the convergence of the model and the performance on the validation set, and enter S5 after the deep learning model training is completed, otherwise continue to loop S4;

[0061] By calculating the mean square error loss function, we can judge the convergence of the model and its performance on the validation set. The formula is as follows:

[0062]

[0063] The mean square error is used as a loss function to quantify the performance of the model by calculating the square of the difference between the predicted value and the true value. This measurement method is very intuitive for regression problems and can effectively reflect the prediction accuracy of the model. In the formula, L train represents the mean square error, m represents the number of samples, which is derived from the proportion of the training set, k represents the counting subscript, and Zs k represents the true value, which is extracted from the training set, Yc k Represents the model prediction value. The mean square error is sensitive to large prediction errors, which means that if there are obvious outliers or prediction errors, the model will automatically be punished more, prompting the model to pay more attention to reducing the impact of these large errors during the learning process. This helps the model accurately capture the true performance of the tool in the prediction of the remaining service life of the tunnel boring machine tool;

[0064] The convergence criteria of the model are as follows:

[0065] ΔL train =|L train (t)-L train (t-1)|

[0066] ΔL val =|L val (t)-L val (t-1)|

[0067] Where, ΔL train Represents the change in training loss, ΔL val Represents the change in validation loss, t represents the current epoch index. When the change is less than the convergence threshold ∈ in three consecutive epochs, the model is judged to be converged. The convergence situation can provide a basis for hyperparameter adjustment. If the model fails to converge, it may be necessary to adjust hyperparameters such as learning rate and batch size and continue to train the model. A good convergence mode usually points to a suitable hyperparameter configuration, thereby improving model performance.

[0068] S5. Use the trained model to predict the test set, calculate the predicted value of the remaining service life of each tool, compare the predicted value of the remaining service life of the tool with the actual remaining service life value Sjz of the tool, and predict the effectiveness of the model. If an abnormality occurs, return to S4 for retraining.

[0069] The remaining service life prediction value of each tool is calculated as follows:

[0070] HUL j =f(X j ,θ)

[0071] Predicting the service life of each tool can better adjust the construction plan, ensure that the tool is replaced at the most appropriate time, avoid construction delays, and improve overall project efficiency. In the formula, RUL j represents the remaining service life prediction value of the tool, X j represents the input feature vector of the jth tool, which includes tool structure parameters, tool vibration parameters, tool speed, tool temperature and tool pressure, f represents the trained prediction model, θ represents the parameters of the model, including the corresponding weight values ​​of the parameters, unit states and neuron bias items, which are obtained through training data optimization during the training process. With this prediction, enterprises can shift from traditional post-maintenance to predictive maintenance, reduce maintenance costs and improve the continuity and efficiency of operations, forming a more scientific maintenance strategy;

[0072] The criteria for judging the predictive validity of the model are as follows:

[0073] When | RUL j -Sjzj |<1, the model prediction is considered valid. j -Sjz j When |≥1, it means that the model prediction is invalid, and it falls back to S4, RUL j represents the remaining service life of the jth tool predicted by the model, Sjz j It represents the real remaining service life of the jth tool, that is, the life value of the tool in actual work. Setting the effectiveness standard can form a feedback mechanism for continuous evaluation. By monitoring and evaluating the prediction results, the model can be adjusted and optimized in time to improve the overall prediction performance.

[0074] S6. Apply the trained model to the actual use environment to monitor the tool status in real time and update the remaining service life of the tunnel boring machine tool;

[0075] The training steps of the prediction model f are as follows:

[0076] S11. Define the LSTM model. The input layer receives the preprocessed feature data. The input shape is (time step, feature dimension). The LSTM layer selects 60% of the LSTM units, sets the number of memory units, and configures dropout to reduce overfitting. The fully connected layer connects the LSTM layer and the output layer, adjusts the information passed downward, and the output layer outputs the remaining service life prediction value of the tool.

[0077] S21. After building the model, compile it, configure the loss function and optimizer, use mean square error as the loss function for regression problems, select Adam optimizer, and set a learning rate of 70% to ensure effectiveness during the training process;

[0078] S31. Input training data, train the model and perform forward propagation and back propagation to adjust model parameters. After each cycle, use 15% of the validation set to evaluate the performance of the model.

[0079] S41. Introduce an early stopping mechanism. When the validation loss does not improve within 10 cycles, terminate the training early and obtain the relevant model parameters.

[0080] By predicting the remaining service life of the tool, judging the effectiveness and accuracy of the prediction value and providing real-time feedback, and adjusting the prediction model according to the latest operating data and the actual status of the tool, the remaining service life prediction value of the tunnel boring machine tool is reliable, the tool can be replaced in time, and the working efficiency of the tunnel boring machine is improved.

[0081] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the remaining service life of a tunnel boring machine tool based on deep learning, characterized in that: The following steps are involved: S1. Collect real-time monitoring data of tunnel boring machine tool vibration, temperature, pressure, speed, soil type, soil parameters, water level and temperature data; S2. Calculate the missing values ​​of all the collected data, remove noise and outliers, and perform normalization so that the above data are all within the range of [0, 1]; S3, build a deep learning model for tool life prediction, select the TensorFlow deep learning framework and LSTM model architecture, and divide all the processed data in S2 into training set, validation set and test set; S4: Perform deep learning model training optimization, input training data and use back propagation algorithm to optimize model parameters, judge the convergence of the model and the performance on the validation set, and enter S5 after the deep learning model training is completed, otherwise continue to loop S4; S5. Use the trained model to predict the test set, calculate the predicted value of the remaining service life of each tool, compare the predicted value of the remaining service life of the tool with the actual remaining service life value Sjz of the tool, judge the effectiveness of the prediction model, and return to S4 for retraining if an abnormality occurs; S6. Apply the trained model to the actual use environment to monitor the tool status in real time and update the remaining service life of the tunnel boring machine tool.

2. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 1, characterized in that: The missing value calculation expression of the collected data in S2 is as follows: In the formula, Sx i represents the data after missing filling, n represents the number of valid data, Sx j Indicates missing data, including tunnel boring machine tool vibration data, temperature data, pressure data, speed data, soil type data, soil parameter data, water level data and temperature data, and j indicates the number of missing values.

3. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 2, characterized in that: The formula for removing all collected data noise by S2 is as follows: In the formula, Qz i represents the denoised tunnel boring machine tool vibration data, temperature data, pressure data, speed data, soil type data, soil parameter data, water level data and temperature data, N represents the size of the moving window, Qz t Represents the original value of the time series data at time t.

4. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 3, characterized in that: The formula for removing all abnormal values ​​of collected data in S2 is as follows: In the formula, z represents Z-Score, Dq represents the current value, Jz represents the data mean, and σ represents the data standard deviation.

5. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 4, characterized in that: The formula for normalizing all collected data by S2 is as follows: In the formula, Gy represents the normalized value, Ys represents the original data, min(Ys) and max(Ys) represent the minimum and maximum values ​​in the original data set.

6. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 5, characterized in that: The S3 constructs a deep learning model and divides the data into a training set ratio of 70%, a validation set ratio of 15% and a test set ratio of 15%. The data of the training set, validation set and test set are processed tunnel boring machine tool vibration data, temperature data, pressure data, rotation speed data, soil type data, soil parameter data, water level data and temperature data.

7. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 6, characterized in that: S4 determines the convergence of the model and the performance on the validation set by calculating the mean square error loss function. The formula is as follows: In the formula, L train represents the mean square error, m represents the number of samples, which is derived from the proportion of the training set, k represents the counting subscript, and Zs k represents the true value, which is extracted from the training set, Yc k Represents the model prediction value.

8. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 7, characterized in that: The convergence judgment criteria of the S4 model are as follows: ΔL train =|L train (t)-L train (t-1)| ΔL val =|L val (t)-L val (t-1)| Where, ΔL train Represents the change in training loss, ΔL val Represents the change in validation loss, t represents the current epoch index. When the change is less than the convergence threshold ∈ in three consecutive epochs, the model is judged to be converged, and the model no longer needs to be trained.

9. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 8, characterized in that: S5 uses the trained model to predict the test set and calculates the predicted value of the remaining service life of each tool. The formula is as follows: NUMBER j =f(X j ,θ) In the formula, RUL j represents the remaining service life prediction value of the tool, X j represents the input feature vector of the jth tool, f represents the trained prediction model, and θ represents the parameters of the model, which are optimized through training data during the training process.

10. The method for predicting the remaining service life of a tunnel boring machine tool based on deep learning according to claim 9, characterized in that: The S5 compares the predicted value of the remaining service life of the tool with the actual remaining service life value Sjz of the tool to determine the validity of the prediction model. The determination criteria are as follows: When | RUL j -Sjz j When |<1, the model prediction is considered valid. j -Sjz j When |≥1, it means that the model prediction is invalid, and it falls back to S4, RUL j represents the remaining service life of the jth tool predicted by the model, Sjz j It represents the real remaining service life of the jth tool, that is, the life value of the tool in actual work.

Images