A real-time health status monitoring method and system for the main drive of a large shield tunneling machine

Abstract

The present invention provides a method and system for real-time health status monitoring of the main drive of a large shield tunneling machine, which relates to the technical field of shield main drive status monitoring, and includes obtaining input signals including the stress wave signal S in the horizontal direction at the main drive bearing H0 , the stress wave signal S in the vertical direction V0 and the current signal I0 of the shield machine PLC system; and inputting the monitored input signals into a diagnostic model, and judging the health status of the main bearing in real time according to the prediction result of the diagnostic model. In the present invention, by innovatively combining stress wave sensors to obtain the stress wave signal S in the horizontal direction H , the stress wave signal S in the vertical direction V , cooperating with the current signal I of the main drive, comprehensive coverage monitoring of the shield machine main drive system is realized, the problem of the deficiency of traditional single-sensor monitoring means is made up for, and PCA is used for multi-modal fusion to extract comprehensive features. Through an integrated learning diagnostic model, the accuracy of the fault diagnosis result is improved, and the accuracy and stability of the diagnosis result are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of shield main drive status monitoring, and specifically to a real-time health status monitoring method and system for large shield main drives. Background Art

[0002] As a key device in modern tunnel engineering, the main drive system of a large shield machine undertakes the important tasks of driving the cutter head to rotate and overcoming the formation resistance. The health status of the main drive system is directly related to the construction efficiency and engineering safety of the shield machine. However, due to the complex shield construction environment, the main drive system is in a high-load, low-speed, and high-noise working state for a long time, and problems such as bearing wear, gear failure, and motor failure are likely to occur. If these problems are not detected in time, it may lead to equipment shutdown and even cause serious engineering accidents.

[0003] Currently, the monitoring technologies for the health status of large shield main drives mainly include single-sensor monitoring and fault diagnosis technologies based on traditional signal processing methods. For example, vibration signals are collected using acceleration sensors, and fault features are identified through spectral analysis or wavelet transform. However, such methods have the following deficiencies: 1. Limitations of single signals: The signals collected by a single sensor are difficult to comprehensively reflect the health status of the main drive system. Especially under the superposition of multiple fault modes or complex environmental noise interference, the diagnostic accuracy significantly decreases; 2. Imperfect feature extraction methods: Some feature extraction methods may be too simplistic, resulting in the loss of some important information in the original data, thus affecting the accuracy of subsequent tasks; 3. Lack of real-time performance: Traditional methods usually require offline processing and cannot achieve real-time monitoring and early warning of the health status of the main drive.

[0004] In recent years, the development of multi-sensor fusion technology and artificial intelligence algorithms has provided new ideas for the health status monitoring of the main drive; by collecting multi-modal data such as vibration, temperature, and current through multiple sensors, and using machine learning algorithms to achieve feature fusion and intelligent diagnosis, the comprehensiveness and accuracy of monitoring can be effectively improved. However, the following problems still exist in the application of existing technologies: Difficult data fusion: The fusion of multi-modal data needs to solve the problems of synchronization, correlation, and redundancy of different sensor data, and the robustness and efficiency of the fusion algorithm still need to be improved; Limited model performance: Existing diagnostic models are insufficient in dealing with non-stationary signals and small sample data, and it is difficult to meet the real-time monitoring requirements under complex working conditions; Lack of optimization strategies: There is a lack of effective weight optimization strategies for the decision fusion of multi-model diagnostic results, resulting in limited accuracy and stability of the final diagnostic results. Therefore, there is an urgent need for a real-time health monitoring method based on multi-sensor and multi-modal fusion, combined with ensemble learning algorithms, to solve the problems of difficult data fusion, insufficient diagnostic model performance, and lack of decision optimization in existing technologies, so as to comprehensively improve the accuracy and real-time performance of the health monitoring of large shield main drive systems. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method and system for real-time health status monitoring of the main drive of a large shield tunneling machine, which can solve the problems of difficult data fusion, insufficient performance of the diagnostic model, and lack of decision-making optimization in the prior art, so as to comprehensively improve the accuracy and real-time performance of the health monitoring of the main drive system of the large shield tunneling machine.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for real-time health status monitoring of the main drive of a large shield tunneling machine includes the following steps:

[0007] S1. Obtain the stress wave signal S in the horizontal direction at the main drive bearing, H0 the stress wave signal S in the vertical direction, V0 and the current signal I0 of the shield machine PLC system;

[0008] S2. Input the signals obtained in step S1 into the diagnostic model, and based on the prediction results of the diagnostic model, judge the health status of the main bearing in real time;

[0009] The construction of the diagnostic model includes the following steps:

[0010] Step 1. Obtain the stress wave signal S in the horizontal direction at the main drive bearing, H the stress wave signal S in the vertical direction, V and the current signal I of the shield machine PLC system, and perform min-max normalization processing on the stress wave signal S in the horizontal direction at the main drive bearing, H the stress wave signal S in the vertical direction, V and the current signal I of the shield machine PLC system;

[0011] Step 2. Perform dimensionality reduction on the stress wave signal S in the horizontal direction at the main drive bearing through PCA (abbreviation for Principal Component Analysis, representing principal component analysis), H the stress wave signal S in the vertical direction, V fuse the stress wave signal S in the horizontal direction at the main drive bearing, H and the stress wave signal S in the vertical direction into a single stress wave signal S, V ; F ;

[0012] Step 3. Divide the fused stress wave signal S into manageable sub-signal segments based on the sampling frequency and the defined time interval, generate a data set composed of multiple sub-S F segments, and use the continuous wavelet packet method to perform on the sub-S in the data set F segments, and use the continuous wavelet packet method to perform on the sub-S in the data set segments, and use the continuous wavelet packet method to perform on the sub-S in the data set segments, and use the continuous wavelet packet method to perform on the sub-SF Perform a transformation to generate a two-dimensional time-frequency diagram;

[0013] Step 4: Train the generated two-dimensional time-frequency diagram dataset on the Resnet-50 model to generate a ResNet model;

[0014] Step 5: Use the time-frequency feature index formula to calculate 27 feature indices I from each stress wave signal S F data segment, train an SVM model using the feature index I, and adopt a one-vs.-many strategy to generate an SVM model;

[0015] Step 6: Use the time-frequency feature index formula to extract 27 feature indices II from each current signal I data segment, train a decision tree model using the feature index II, and the decision tree model generates a decision tree model by constructing multiple branch nodes and gradually dividing the feature space;

[0016] Step 7: Ensemble learning diagnosis model: Use ensemble learning to combine the above three models through decision-level fusion to improve the accuracy of the fault diagnosis results; the prediction formula is as follows:

[0017] ①

[0018] In formula ①, P(x) is the finally predicted category; is the category probability predicted by the ResNet model; is the category probability predicted by the SVM model; is the category probability predicted by the decision tree model; a is the weight coefficient assigned to the ResNet model; b is the weight coefficient assigned to the SVM model; c is the weight coefficient assigned to the decision tree model.

[0019] A further technical solution of the present invention: In step 7 of S2, the values of a, b, and c are optimized by a genetic algorithm to make the diagnostic performance of the ensemble model reach the optimal, and the specific optimization model is as follows:

[0020] ②

[0021] ③

[0022] Among them, is the optimization objective function;

[0023] is defined as:

[0024] ④

[0025] is based on the predicted category Calculated classification accuracy;

[0026] The definition of accuracy is:

[0027] ⑤

[0028] : F1 score calculated based on the predicted class The definition of the F1 score is:

[0029] The definition of the F1 score is:

[0030] ⑥

[0031] Among them, ;

[0032] TP is the number of true positives; TN is the number of true negatives; FP is the number of false positives; FN is the number of false negatives; α and β are weight coefficients used to balance the importance of accuracy and F1 score;

[0033] The fusion process is optimized by a genetic algorithm to determine the optimal weight coefficients a, b, and c of the base models.

[0034] Preferably, in the first step, the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system are filtered.

[0035] Preferably, the cross-entropy loss function used in the Resnet-50 model in the fourth step is defined as follows:

[0036] ⑦

[0037] In Equation ⑦, N is the total number of samples in the dataset; M is the total number of classes in the classification task; Indicator function, takes a value of 0 or 1. When the true class of sample i is c, is 1, otherwise is 0; pic is the predicted probability that sample i belongs to class c, provided by the softmax output.

[0038] Preferably, the 27 feature indicators in the fifth step include 14 time-domain features and 13 frequency-domain features, and the 27 feature indicators in the sixth step include 14 time-domain features and 13 frequency-domain features.

[0039] Another object of the present invention is to provide a real-time health status monitoring system for the main drive of a large shield tunneling machine, comprising: a stress wave sensor fixedly installed on the outer ring end face of the main bearing;

[0040] a dynamic data acquisition module for acquiring the detection signal of the stress wave sensor and the current signal of the shield machine PLC system;

[0041] an edge computing device communicatively connected to the dynamic data acquisition module, and the edge computing device is used to execute the above-mentioned diagnostic model.

[0042] Preferably, the stress wave sensor is connected to the dynamic data acquisition module through a shielded cable.

[0043] The present invention provides a real-time health status monitoring method and system for the main drive of a large shield tunneling machine. It has the following beneficial effects:

[0044] 1. In the present invention, by innovatively combining a stress wave sensor to obtain the stress wave signal S in the horizontal direction H and the stress wave signal S in the vertical direction V , the stress wave sensor is used to monitor the stress wave signal of the main bearing, and in cooperation with the current signal I of the main drive, it realizes the comprehensive coverage monitoring of the main drive system of the shield tunneling machine, making up for the deficiency of the traditional single-sensor monitoring method; and for the stress wave signal S in the horizontal direction H and the stress wave signal S in the vertical direction V , PCA is used for multi-modal fusion to extract comprehensive features; and through real-time online monitoring, it can give an early warning before a fault occurs, avoid downtime maintenance caused by sudden faults, and greatly reduce the unplanned downtime, thereby improving the efficiency and continuity of shield tunneling construction; the present invention uses an integrated learning diagnostic model, and through integrated learning, combines three models through decision-level fusion to improve the accuracy of the fault diagnosis result. The fusion process is optimized by a genetic algorithm to determine the optimal weight coefficients a, b, and c of the basic models, thereby ensuring the accuracy and stability of the diagnosis result. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a distribution diagram of a three-axis vibration sensor and a stress wave sensor in a real-time health status monitoring system for the main drive of a large shield tunneling machine proposed by the present invention;

[0046] Figure 2 It is a schematic distribution diagram of a three-axis vibration sensor and a stress wave sensor on the main bearing in a real-time health status monitoring system for the main drive of a large shield tunneling machine proposed by the present invention;

[0047] Figure 3Schematic diagram of PCA fusion in a real-time health status monitoring method for the main drive of a large shield tunneling machine proposed by the present invention;

[0048] Figure 4 In a real-time health status monitoring method for the main drive of a large shield tunneling machine proposed by the present invention, the continuous wavelet packet method is used to transform the sub-S on the data set to generate a schematic diagram of a two-dimensional time-frequency map; F Schematic diagram of the transformation to generate a two-dimensional time-frequency map;

[0049] Figure 5 Schematic diagram of the Resnet-50 model architecture in a real-time health status monitoring method for the main drive of a large shield tunneling machine proposed by the present invention;

[0050] Figure 6 Schematic diagram of the SVM model in a real-time health status monitoring method for the main drive of a large shield tunneling machine proposed by the present invention;

[0051] Figure 7 Schematic diagram of the decision tree model in a real-time health status monitoring method for the main drive of a large shield tunneling machine proposed by the present invention. Detailed implementation manners

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0053] Embodiment 1:

[0054] As Figures 1 - 7 shown, the embodiment of the present invention provides a real-time health status monitoring method for the main drive of a large shield tunneling machine, including the following steps:

[0055] S1. Obtain the stress wave signal S in the horizontal direction at the main drive bearing H0 , the stress wave signal S in the vertical direction V0 and the current signal I0 of the shield machine PLC system. The stress wave signal S in the horizontal direction at the main drive bearing H0 , the stress wave signal S in the vertical direction V0 and the current signal I0 of the shield machine PLC system constitute the monitoring input signal;

[0056] S2. Input the input signal into the diagnostic model, and based on the prediction result of the diagnostic model, judge the health status of the main bearing in real time.

[0057] The diagnostic model is executed by an edge computing device. After inputting an input signal into the diagnostic model, the diagnostic model outputs a prediction result, and the prediction result covers multiple health states of the main drive bearing (such as normal, slightly damaged, moderately damaged, severely damaged, failed).

[0058] The construction of the diagnostic model includes the following steps:

[0059] Step 1: Obtain the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system, and perform min-max normalization on the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system.

[0060] ⑧

[0061] In Equation 8, represents the output signal after min-max normalization processing, is less than 1 and greater than 0, X represents the input signal, X max , X min respectively represent the maximum and minimum in the input signal X.

[0062] Step 2: Perform dimensionality reduction on the stress wave signal S in the horizontal direction at the main drive bearing through PCA H , the stress wave signal S in the vertical direction V , fuse the stress wave signal S in the horizontal direction at the main drive bearing H and the stress wave signal S in the vertical direction V into a single stress wave signal S F , realizing multi-modal fusion of data and extracting comprehensive features.

[0063] Step 3: Divide the fused stress wave signal S F into manageable sub-signal segments based on the sampling frequency and the defined time interval, generating a data set composed of multiple sub-S F segments , and use the continuous wavelet packet method to transform the sub-S on the data set F to generate a two-dimensional time-frequency diagram.

[0064] Step 4: Train the generated two-dimensional time-frequency diagram data set on the Resnet-50 model to generate a ResNet model.

[0065] The cross-entropy loss function used in the Resnet-50 model is defined as follows:

[0066] ⑦

[0067] In Equation ⑦, N is the total number of samples in the dataset; M is the total number of classes in the classification task; indicator function, takes values of 0 or 1. When the true class of sample i is c, it is 1, otherwise it is 0; pic is the predicted probability that sample i belongs to class c, provided by the softmax output.

[0068] Based on historical operation data and expert experience, a database containing 27 eigenvalue (time domain, frequency domain, time-frequency domain features) can be constructed, covering different health states of the main drive bearing (such as normal, slightly damaged, moderately damaged, severely damaged, failed);

[0069] Label definition: The health state of the main bearing is divided into 5 categories as the output labels of the model.

[0070] The 27 eigenvalues include 14 time domain features and 13 frequency domain features; among them, Table 1 is the time domain index and Table 2 is the frequency domain index;

[0071] Table 1 Time domain index

[0072]

[0073] Table 2 Frequency domain index

[0074]

[0075] Step Five: Using the time-frequency feature index formula, calculate 27 feature indices one from each stress wave signal S F data segment, and use the feature indices one to train the SVM model. To achieve the multi-classification task, the one-versus-all strategy is adopted, thus generating the SVM model;

[0076] Step Six: Using the time-frequency feature index formula, extract 27 feature indices two from each current signal I data segment, and use the feature indices two to train the decision tree model. To achieve the multi-classification task, the decision tree model generates the third model by constructing multiple branch nodes and gradually dividing the feature space.

[0077] Step Seven: Ensemble learning diagnosis model: Use ensemble learning to combine the above three models through decision-level fusion to improve the accuracy of the fault diagnosis results;

[0078] ①

[0079] In Equation ①, P(x) is the finally predicted category; P ResNet (x) is the category probability predicted by the ResNet model; P SVM (x) is the category probability predicted by the SVM model; P DT (x) is the category probability predicted by the decision tree model; a is the weight coefficient assigned to the ResNet model; b is the weight coefficient assigned to the SVM model; c is the weight coefficient assigned to the decision tree model;

[0080] The values of a, b, and c are optimized by the genetic algorithm to optimize the diagnostic performance of the integrated model. The specific optimization model is as follows:

[0081] ②

[0082] ③

[0083] Among them, is the optimization objective function;

[0084] is defined as:

[0085] ④

[0086] is the classification accuracy calculated based on the predicted category ;

[0087] The definition of accuracy is:

[0088] ⑤

[0089] : The F1 score calculated based on the predicted category ;

[0090] The definition of

[0091] ⑥

[0092] Among them, ;

[0093] TP is the number of true positive examples; TN is the number of true negative examples; FP is the number of false positive examples; FN is the number of false negative examples; α and β are weight coefficients used to balance the importance of accuracy and F1 score;

[0094] The fusion process is optimized by the genetic algorithm to determine the optimal weight coefficients a, b, and c of the base models.

[0095] In step one, it also includes: filtering the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system.

[0096] Real-time monitoring and early warning: Deploy the diagnostic model in the edge computing device to achieve real-time monitoring of the health status of the main drive system of the large shield and early warning of faults.

[0097] The model training process mainly includes: data processing and offline training:

[0098] Data processing

[0099] Use historical operation data to construct a main drive main bearing database, including stress wave signals, current signals, and fault annotation information; through steps such as data cleaning and feature extraction (such as time domain, frequency domain, and time-frequency domain features), generate a training dataset.

[0100] Model training:

[0101] Adopt an integrated learning method, combine multi-modal data for offline training, and optimize model parameters to improve the accuracy and robustness of fault diagnosis.

[0102] Embodiment 2:

[0103] As Figures 1 - 2 shown, the embodiment of the present invention provides a real-time health status monitoring system for the main drive of a large shield, including: a stress wave sensor, a dynamic data acquisition module, and an edge computing device.

[0104] The stress wave sensor is fixedly installed on the outer ring end face of the main bearing. The stress wave sensor captures the early fault signals of the main bearing (such as high-frequency signals generated by friction and impact). For example, a stress wave sensor with a model number of 370A, a high-frequency response (>10 kHz), and a sensitivity of 11.5 mV / g. The dynamic data acquisition module is used to collect the detection signals of the stress wave sensor and the current signal of the shield machine PLC system. The VSE153 vibration analysis module can be used, which can support signal acquisition from 0 to 12000 Hz and includes stress wave and vibration data channels. The shield machine PLC system is configured with a DC power supply and a control cabinet, which can detect the current signal of the shield machine PLC system and obtain the current signal I of the shield machine PLC system. The edge computing device is communicatively connected to the dynamic data acquisition module. The edge computing device is used to execute the diagnostic model in Embodiment 1, that is, according to the stress wave signal S in the horizontal direction at the main drive bearing obtained in real time H0 , the stress wave signal S in the vertical direction V0 and the current signal I of the shield machine PLC system 0，Input it into the diagnostic model, and judge the health status of the main drive of the large shield tunneling machine based on the output of the diagnostic model.

[0105] Stress wave sensors are installed on both the main bearing and the main drive motor of the shield tunneling machine. The stress wave sensors are connected to the dynamic data acquisition module through shielded cables. The signals are amplified and filtered and then converted into digital signals. The dynamic acquisition module transmits the data to the PLC controller through the Modbus TCP protocol, and then it is transmitted to the remote monitoring server through the fiber optic network. The data server and the Web server are responsible for real-time data storage and display of the analysis results respectively.

[0106] In this solution, stress wave sensors with high durability are selected, such as those with a temperature tolerance range of -40°C to +150°C, a moisture resistance performance reaching the IP67 protection level, and capable of withstanding vibration shocks above 10 g, which are suitable for the shield tunneling construction environment with high noise, high humidity, high temperature and strong vibration. At the same time, the signal processing module of the stress wave sensor has a background noise suppression ratio of 40 dB, effectively filtering out environmental interference and ensuring the accuracy and reliability of the monitoring data.

[0107] In this solution, the stress wave sensor has high sensitivity (10 mV / g) and a wide frequency response range (1 kHz to 1 MHz), which is suitable for capturing low-frequency micro fault signals of the main bearing; the vibration sensor has a sensitivity of 100 mV / g and a frequency range of 10 Hz to 10 kHz, which can accurately monitor the vibration state of the motor. The two work together to increase the fault detection rate to over 95%, significantly enhancing the diagnostic accuracy and reliability.

[0108] System operation steps

[0109] Arrange stress wave sensors on the outer ring end face of the main bearing and connect them to the dynamic data acquisition module. After completing the hardware debugging, start the system. The stress wave sensors continuously collect physical quantity signals and display the operating status of the main bearing and the motor in real time. The system conducts multi-factor trend analysis on the collected signals, judges the fault location and type, generates a diagnostic report, provides equipment maintenance suggestions based on the monitoring results, and regularly updates the sensor calibration parameters and system software.

[0110] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

Claims

1. A real-time health status monitoring method for the main drive of a large shield tunneling machine, characterized in that It includes the following steps: S1. Obtain the stress wave signal S in the horizontal direction at the main drive bearing H0 , the stress wave signal S in the vertical direction V0 and the current signal I0 of the shield machine PLC system; S2. Input the signals obtained in step S1 into the diagnostic model, and based on the prediction results of the diagnostic model, judge the health status of the main bearing in real time; The construction of the diagnostic model includes the following steps: Step 1: Obtain the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system, and perform min-max normalization processing on the stress wave signal S in the horizontal direction at the main drive bearing H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system; Step 2: Perform dimensionality reduction on the stress wave signals S in the horizontal direction at the main drive bearing through PCA H and the stress wave signals S in the vertical direction V to fuse the stress wave signals S in the horizontal direction and the stress wave signals S in the vertical direction at the main drive bearing into a single stress wave signal S H ; V F ;​ Step 3: Divide the fused stress wave signal S F into manageable sub-signal segments based on the sampling frequency and the defined time interval, generating a dataset composed of multiple sub-S F segments , and use the continuous wavelet packet method to transform the sub-S on the dataset F to generate a two-dimensional time-frequency diagram; Step 4. Train the generated two-dimensional time-frequency diagram data set on the Resnet-50 model to generate a ResNet model; Step 5: Using the time-frequency feature index formula, calculate 27 feature indices I from each stress wave signal S F data segment, use the feature index I to train the SVM model, and adopt the one-versus-all strategy to generate the SVM model; Step 6. Use the time-frequency feature index formula to extract 27 feature indices II from each current signal I data segment, and use the feature indices II to train a decision tree model. The decision tree model generates a decision tree model by constructing multiple branch nodes and gradually dividing the feature space; Step 7. Integrated learning diagnostic model: Use integrated learning to combine the above three models through decision-level fusion to improve the accuracy of the fault diagnosis results. The prediction formula is as follows: ① In Equation ①, P(x) is the finally predicted class; is the class probability predicted by the ResNet model; is the class probability predicted by the SVM model; is the class probability predicted by the decision tree model; a is the weight coefficient assigned to the ResNet model; b is the weight coefficient assigned to the SVM model; c is the weight coefficient assigned to the decision tree model.

2. The real-time health status monitoring method for the main drive of a large shield tunneling machine according to claim 1, characterized in that: In step 7 of S2, the values of a, b, and c are optimized by a genetic algorithm to make the diagnostic performance of the integrated model reach the optimal. The specific optimization model is as follows: The values of a, b, and c are optimized by a genetic algorithm to make the diagnostic performance of the integrated model reach the optimal. The specific optimization model is as follows: ② ③ Among them, is the optimization objective function; Defined as: ④ is based on the predicted class Calculated classification accuracy; The definition of accuracy is: ⑤ : F1 score calculated based on the predicted class ; is defined as: ⑥ Among them, ; TP is the number of true positives; TN is the number of true negatives; FP is the number of false positives; FN is the number of false negatives; α and β are weight coefficients used to balance the importance of accuracy and F1 score; The fusion process is optimized by a genetic algorithm to determine the optimal weight coefficients a, b, and c of the basic models.

3. A real-time health status monitoring method for the main drive of a large shield tunneling machine according to claim 1 or 2, characterized in that: The stress wave signal S in the horizontal direction at the main drive bearing is also included in the first step H , the stress wave signal S in the vertical direction V and the current signal I of the shield machine PLC system are filtered 4. A real-time health status monitoring method for the main drive of a large shield tunneling machine according to claim 1 or 2, characterized in that The cross-entropy loss function used in the Resnet-50 model in step 4 is defined as follows: ⑦ In Equation ⑦, N is the total number of samples in the dataset; M is the total number of classes in the classification task; Indicator function, which takes a value of 0 or 1. When the true class of sample i is c, it is 1; otherwise it is 0; pic is the predicted probability that sample i belongs to class c, provided by the softmax output.

5. A real-time health status monitoring method for the main drive of a large shield tunneling machine according to claim 1 or 2, characterized in that: The 27 feature indices I in step 5 include 14 time-domain features and 13 frequency-domain features, and the 27 feature indices II in step 6 include 14 time-domain features and 13 frequency-domain features.

6. A real-time health status monitoring system for the main drive of a large shield tunneling machine, characterized in that, It includes: A stress wave sensor fixedly installed on the outer ring end face of the main bearing; A dynamic data acquisition module for collecting the detection signals of the stress wave sensor and the current signals of the shield machine PLC system; An edge computing device communicatively connected to the dynamic data acquisition module, and the edge computing device is used to execute the diagnostic model described in any one of claims 1-5.

7. The real-time health status monitoring system for the main drive of a large shield tunneling machine according to claim 6, characterized in that: The stress wave sensor is connected to the dynamic data acquisition module through a shielded cable.

Images