Main driving system design method based on operation and maintenance data of tunnel boring machine

Abstract

The invention relates to a main driving system design method based on operation and maintenance data of a tunnel boring machine, and the method comprises the steps: collecting the data of a motor, a speed reducer, a bearing and a sealing key part of the main driving system of the tunnel boring machine, collecting the operation condition data, and installing various types of sensors to collect the physical parameter data; and the collected data is preprocessed and then transmitted to the data processing center. The method provided by the invention can realize higher prediction precision under the condition of fewer data samples, has better stability and adaptability, and can effectively cope with the complicated operating environment and working condition change of the main driving system of the tunnel boring machine.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel boring machines, and in particular to a main drive system design method based on tunnel boring machine operation and maintenance data, aiming to achieve efficient management, fault prediction and performance optimization of the main drive system through in-depth mining and analysis of operation and maintenance data. Background Art

[0002] As a key equipment in modern tunnel construction, the main drive system of a tunnel boring machine is responsible for providing powerful power to drive the cutterhead to rotate and cut rock and soil. The main drive system has a complex structure, including multiple key components such as motors, reducers, bearings, and seals. It faces many challenges in long-term operation under harsh tunnel construction environments.

[0003] Traditional management of the main drive system of a tunnel boring machine mainly relies on regular maintenance plans and post-fault repair modes. Regular maintenance plans are often based on experience to set fixed time intervals or operating hours for inspection, maintenance, and replacement of wearing parts. This method is difficult to accurately adapt to the actual wear and aging conditions of the components of the main drive system, and is prone to over-maintenance or under-maintenance. For example, some components may still be in good condition within the prescribed maintenance cycle, but they are forced to be replaced, resulting in a waste of resources; while other components may be damaged in advance due to special working conditions, but they are not discovered and handled in time because the maintenance time has not arrived, resulting in downtime, affecting the progress of the project and increasing maintenance costs.

[0004] The post-fault repair mode is to perform repairs only after the main drive system fails, which not only leads to the interruption of tunneling operations, but also often further expands the fault due to the lack of early warning and effective prevention of faults, increasing the difficulty and cost of repairs, and may even damage the overall structure of the tunnel boring machine. In addition, traditional management methods make it difficult to monitor and optimize the performance of the main drive system in real time, and cannot give full play to the maximum efficiency of the main drive system.

[0005] With the rapid development of digital technology and big data analysis, more and more equipment management has begun to use data-driven methods to achieve intelligent management. However, in the field of tunnel boring machine main drive system management, the existing data application methods still have many shortcomings. For example, some data processing methods simply collect and store data, and fail to deeply explore the fault characteristics and performance change laws hidden behind the data; some fault diagnosis algorithms have low accuracy and reliability when facing the complex working conditions and multi-source heterogeneous data of the tunnel boring machine main drive system; moreover, the existing management methods lack systematic consideration of the entire life cycle of the main drive system, and cannot achieve the optimal management strategy as a whole. Therefore, there is an urgent need for an innovative main drive system and management method based on tunnel boring machine operation and maintenance data to improve the management level, reliability and operation efficiency of the main drive system. Summary of the invention

[0006] The present invention provides a main drive system design method based on tunnel boring machine operation and maintenance data, including a data acquisition unit, a data processing center, and a control and decision-making unit, wherein the data acquisition unit is deployed in the motor, reducer, bearing, and sealing key parts of the tunnel boring machine main drive system and collects operating condition data, and various types of sensors are installed to collect physical parameter data, and the collected data is pre-processed and transmitted to the data processing center.

[0007] Furthermore, the data processing center is equipped with high-performance computing servers and large-capacity storage devices to clean and organize the data, use a combination of principal component analysis and independent component analysis to perform feature extraction and dimensionality reduction, and establish performance degradation models for each component of the main drive system.

[0008] Furthermore, the control and decision-making unit formulates control strategies and maintenance decisions for the main drive system based on the analysis results of the data processing center, adjusts operating parameters in the event of an abnormality or fault warning, and arranges maintenance plans based on performance degradation models and fault prediction results.

[0009] On the other hand, the present invention also provides a main drive system design method based on tunnel boring machine operation and maintenance data, including data acquisition and preprocessing steps. The data acquisition unit collects sensor data and operating condition data according to a set sampling frequency, and verifies, filters and standardizes the data. The sampling frequency is set according to the operating characteristics of the main drive system.

[0010] Furthermore, it also includes feature extraction and fault diagnosis steps. The data processing center uses a feature extraction method based on multi-scale morphological filtering and Hilbert-Huang transform to process the preprocessed data, construct feature vectors, and use a fault diagnosis model that integrates convolutional neural networks and long short-term memory networks to perform fault diagnosis.

[0011] Furthermore, it also includes performance prediction and optimization steps, establishing a main drive system performance prediction model based on a performance prediction method combining support vector machine regression and particle swarm optimization algorithm, and optimizing the performance of the main drive system according to the prediction results and fault diagnosis information.

[0012] Furthermore, the structural element of the multi-scale morphological filtering is a circle with a radius of 2, 4, 6, 8 or adjusted to 3, 5, 7, 9, 11, 13 according to geological conditions. The data is decomposed into 4 or 6 scales, and 5 or 6 intrinsic mode functions IMFs are obtained after Hilbert-Huang transform, and their characteristic parameters are calculated to construct feature vectors.

[0013] Furthermore, in the fault diagnosis model that integrates convolutional neural network and long short-term memory network, the CNN part is equipped with 3 convolutional layers, the convolution kernel sizes are 3×3, 5×5, and 3×3, respectively, the number of convolution kernels are 16, 32, and 64, respectively, ReLU is used as the activation function, and the LSTM part is equipped with 2 layers of LSTM units, the number of units in each layer is 128, and finally the fully connected layer is connected to output the fault type and severity.

[0014] Furthermore, in the performance prediction method based on the combination of support vector machine regression and particle swarm optimization algorithm, radial basis function is used as the kernel function of SVR, the number of particles is set to 30, the maximum number of iterations is 100, the learning factor c1=c2=1.5, and a high-precision power prediction model is constructed by optimizing the penalty factor C of SVR and the kernel function parameter γ.

[0015] Furthermore, in tunneling projects under different geological conditions, such as subway tunnels or mountain tunnels, the sensor type, sampling frequency, data processing parameters, and model structure are adjusted according to the actual situation to achieve effective intelligent management of the main drive system of the tunnel boring machine.

[0016] Beneficial technical effects:

[0017] A comprehensive and efficient data acquisition system has been constructed, which can collect various types of data of the main drive system of the tunnel boring machine in real time. The data acquisition unit of the present invention can more comprehensively reflect the operating status of the main drive system, and provide a richer information basis for subsequent data analysis and management decisions. Advanced data cleaning, feature extraction and analysis algorithms are adopted, such as the feature extraction method combining PCA and ICA, which can effectively process multi-source heterogeneous data and extract more representative and sensitive fault feature information. For the first time, multi-scale morphological filtering and Hilbert-Huang transform are combined and applied to the operation and maintenance data feature extraction of the main drive system of the tunnel boring machine. Multi-scale morphological filtering can analyze data at different scales, effectively remove noise and interference, and highlight the local characteristics of the data; Hilbert-Huang transform can decompose non-stationary signals into intrinsic mode functions, and further extract the time-frequency characteristics of the signal. The combination of the two can more comprehensively and accurately extract the fault feature information in the operation data of the main drive system. The proposed fault diagnosis model integrating convolutional neural network and long short-term memory network is an innovative solution for fault diagnosis of the main drive system of the tunnel boring machine. The performance prediction method based on the combination of support vector machine regression and particle swarm optimization algorithm can utilize the good generalization ability of SVR and the parameter optimization ability of PSO to build a high-precision main drive system performance prediction model. By modeling and optimizing the nonlinear relationship between the performance indicators of the main drive system and the operation and maintenance data, it is possible to more accurately predict the performance change trend of the main drive system in the future, discover potential performance problems in advance, and provide strong support for timely optimization measures. Compared with traditional performance prediction methods, the method of the present invention can achieve higher prediction accuracy with fewer data samples, and has better stability and adaptability, and can effectively cope with the complex operating environment and working condition changes of the main drive system of the tunnel boring machine. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the process. DETAILED DESCRIPTION

[0019] Example 1

[0020] 1. Main drive system architecture

[0021] Data acquisition unit: deployed in key parts of the main drive system of the tunnel boring machine, including motors, reducers, bearings, seals, etc., and installed with various types of sensors, such as temperature sensors, vibration sensors, pressure sensors, torque sensors, etc., to collect various physical parameter data during the operation of the main drive system. These sensors have high sensitivity and high precision, and can accurately monitor the operating status of the main drive system in real time. For example, temperature sensors can accurately measure temperature changes in motor windings, bearings and other parts, and promptly detect abnormal temperature increases caused by overload, friction and other reasons; vibration sensors can capture vibration signals generated by reducer gear meshing, bearing operation, etc., and judge fault conditions such as wear, looseness or imbalance of components by analyzing information such as vibration frequency, amplitude and phase.

[0022] At the same time, the operating condition data of the main drive system is collected, such as excavation speed, cutter head speed, propulsion pressure, geological condition information, etc. These operating condition data are closely related to the performance of the main drive system and can provide an important reference for fault diagnosis and performance optimization. For example, different geological conditions will have a significant impact on the load of the main drive system. Hard rock formations will cause the main drive system to bear greater torque and impact, while soft soil formations may cause uneven cutting resistance of the cutter head, affecting the smooth operation of the main drive system. The data acquisition unit pre-processes all the collected data, such as filtering, amplification, analog-to-digital conversion, etc., and then transmits it to the data processing center through a high-speed data transmission channel.

[0023] Data processing center: equipped with high-performance computing servers and large-capacity storage devices, used to receive, store and process massive operation and maintenance data from data acquisition units. Further clean and organize the data to remove noise data, outliers and duplicate data to ensure data accuracy and integrity. For example, use outlier detection methods based on statistical principles, such as the 3σ criterion, to identify and remove abnormal data points that are significantly deviated from the normal data range; for noise data generated by sensor failure or signal interference, use a smoothing filter algorithm to remove it to improve data quality.

[0024] Advanced data mining and analysis algorithms are used to deeply process the data and extract the fault characteristics and performance indicators of the main drive system. For example, a method combining principal component analysis (PCA) and independent component analysis (ICA) is used to extract features and reduce the dimension of multivariate sensor data, converting the high-dimensional data space into a low-dimensional feature space, while retaining the main information of the data, reducing the complexity of data processing and highlighting the fault characteristic information. Through the analysis of historical operation and maintenance data, a performance degradation model of each component of the main drive system is established to predict the remaining service life and failure probability of the components.

[0025] Control and decision-making unit: According to the results of the data processing center's analysis, formulate the control strategy and maintenance decision of the main drive system. When an abnormality or fault warning is detected in the main drive system, timely adjust the operating parameters of the main drive system, such as reducing the motor speed, adjusting the propulsion pressure, etc., to reduce the load of the faulty component and avoid further deterioration of the fault. For example, when the bearing temperature is too high, the control unit can automatically reduce the cutter head speed to reduce the friction and heat of the bearing, and increase the flow of the cooling system to cool the bearing. Based on the performance degradation model and fault prediction results, reasonably arrange the maintenance plan of the main drive system, and determine the time, content and required resources for maintenance. For example, when it is predicted that the remaining service life of a key component is short, prepare the corresponding spare parts and maintenance tools in advance, and arrange replacement or repair at a suitable construction interval to achieve preventive maintenance and reduce the losses caused by sudden failures.

[0026] 2. Management Method Process

[0027] 1. Data collection and preprocessing steps:

[0028] The data acquisition unit collects various sensor data and working condition data of the main drive system of the tunnel boring machine in real time according to the set sampling frequency. The sampling frequency is reasonably set according to the operating characteristics of the main drive system and the sensitivity of data changes. For example, for parameters that change relatively slowly, such as temperature and pressure, the sampling frequency can be set to 1-10 times per second; for parameters that change quickly, such as vibration, the sampling frequency can be increased to 100-1000 times per second to ensure that subtle changes in the data can be captured.

[0029] The collected data is pre-processed in real time, including data verification, filtering and standardization. Data verification is used to check the integrity and correctness of the data to ensure that the data is not lost or erroneous during transmission; filtering uses digital filtering algorithms, such as low-pass filtering, high-pass filtering or band-pass filtering, to remove noise interference in the data; standardization converts data of different ranges and units into a standard numerical range to facilitate subsequent data processing and analysis.

[0030] 2. Feature extraction and fault diagnosis steps:

[0031] The data processing center uses innovative feature extraction algorithms to extract features from preprocessed data. A feature extraction method based on the combination of multiscale morphological filtering and Hilbert-Huang transform (HHT) is proposed. First, multiscale morphological filtering is used to decompose the data at multiple scales to extract data features at different scales, which can effectively remove background noise and local interference in the data and highlight the edge and detail information of the signal. Then, the Hilbert-Huang transform is performed on the results after multiscale morphological filtering to decompose the non-stationary signal into a finite number of intrinsic mode functions (IMFs). By calculating the characteristic parameters of the IMFs, such as the instantaneous frequency and instantaneous amplitude, a feature vector that can reflect the operating status of the main drive system is constructed.

[0032] A fault diagnosis model based on deep learning is used to perform fault diagnosis on the extracted feature vectors. A fault diagnosis model that integrates convolutional neural network (CNN) and long short-term memory network (LSTM) is constructed. CNN is good at extracting local and spatial features of data and can automatically learn fault mode features in sensor data; LSTM can effectively process time series data and capture long-term dependencies in data, which is suitable for dynamic fault diagnosis during the operation of the main drive system. The feature vector is input into the fusion model, and the model is trained with a large amount of historical operation and maintenance data to learn the mapping relationship between different fault types and feature vectors, so as to achieve accurate diagnosis and classification of main drive system faults. For example, it can accurately identify common fault types such as motor winding short circuit, reducer gear wear, bearing damage, seal failure, and determine the severity and location of the fault.

[0033] 3. Performance prediction and optimization steps:

[0034] Based on the historical operation and maintenance data stored in the data processing center, a performance prediction model of the main drive system is established. A performance prediction method based on the combination of support vector machine regression (SVR) and particle swarm optimization algorithm (PSO) is adopted. First, a nonlinear regression model between the performance indicators of the main drive system (such as power, efficiency, torque, etc.) and the operation and maintenance data is established using SVR. SVR has good generalization ability and small sample learning ability, and can build a more accurate prediction model under limited data samples. Then, the particle swarm optimization algorithm is used to optimize the model parameters of SVR (such as kernel function parameters, penalty factors, etc.) to improve the prediction accuracy and stability of the model. Through this performance prediction model, the performance change trend of the main drive system in the future period is predicted according to the current operation and maintenance data, and the trend of performance degradation is discovered in advance, providing a basis for taking corresponding optimization measures.

[0035] The performance of the main drive system is optimized based on the performance prediction results and fault diagnosis information. When it is predicted that the performance of the main drive system will decline, the operating parameters of the main drive system are adjusted, such as optimizing the control strategy of the motor, adjusting the transmission ratio of the reducer, and improving the oil supply of the lubrication system, so as to improve the operating efficiency and reliability of the main drive system. For example, when it is found that the power consumption of the main drive system is too large under a certain working condition, the power supply frequency and voltage of the motor are optimized to make it operate near the optimal efficiency point to reduce energy consumption; at the same time, according to the wear of the bearing, the oil supply and oil temperature of the lubrication system are adjusted to improve the lubrication conditions of the bearing and extend the service life of the bearing.

[0036] (I) Detailed modeling and solution steps

[0037] 1. Feature extraction based on multi-scale morphological filtering and Hilbert-Huang transform (HHT):

[0038] Select appropriate structural elements. The shape and size of the structural elements are determined according to the characteristics of the data and the type of noise to be removed. For example, to remove local interference in vibration data, a small-sized structural element of circular or square shape can be selected.

[0039] Set multi-scale parameters and perform multi-scale decomposition on the data by changing the scale of the structural element. Generally, start from a smaller scale and gradually increase the scale. For example, set the scale range to [2,10] and perform multi-scale decomposition with a step size of 1. At each scale, use morphological opening and closing operations to process the data to obtain data features at different scales. Through this multi-scale decomposition, background noise and local interference in the data can be effectively removed, and the edge and detail information of the signal can be highlighted.

[0040] Hilbert-Huang Transform (HHT):

[0041] The Hilbert-Huang transform is performed on the results of multi-scale morphological filtering. First, the non-stationary signal is decomposed into a finite number of intrinsic mode functions (IMFs) through empirical mode decomposition (EMD). The EMD algorithm is based on the local extreme points of the signal itself, and constructs the upper and lower envelopes by continuously screening out the local maximum and minimum points, and then obtains the IMF.

[0042] Calculate the characteristic parameters such as instantaneous frequency and instantaneous amplitude of each IMF. The instantaneous frequency can be obtained by taking the phase derivative of the Hilbert transform of the IMF, and the instantaneous amplitude can be directly obtained from the result after the Hilbert transform. Through these characteristic parameters, construct a characteristic vector that can reflect the operating state of the main drive system. For example, arrange the instantaneous frequencies and instantaneous amplitudes of different IMFs in a certain order to form a multi-dimensional characteristic vector, such as [IMF1_instantaneous frequency, IMF1_instantaneous amplitude, IMF2_instantaneous frequency, IMF2_instantaneous amplitude, ...].

[0043] Fault diagnosis model integrating convolutional neural network (CNN) and long short-term memory network (LSTM): CNN part: Construct the network structure of CNN, including input layer, convolution layer, pooling layer, fully connected layer and output layer. The input layer receives the feature vector, whose dimension is determined according to the dimension of the feature vector obtained in the feature extraction step. For example, if the feature vector is 100-dimensional, the number of nodes in the input layer is set to 100.

[0044] In the convolution layer, set the appropriate convolution kernel size (such as 3×3 or 5×5) and the number of convolution kernels (such as 16, 32, etc.) to extract the local and spatial features of the data through convolution operations. The convolution kernel slides on the input data to perform convolution operations and obtain the feature map after convolution.

[0045] The pooling layer uses the maximum pooling or average pooling method to downsample the convolutional feature map to reduce the data dimension while retaining the main features. For example, a 2×2 maximum pooling window is used to reduce the size of the feature map by half. After being processed by multiple convolutional and pooling layers, the pooled features are mapped to the output layer through the fully connected layer. The number of nodes in the output layer is determined by the number of fault types that need to be diagnosed. For example, to diagnose 5 common fault types, the output layer is set to 5 nodes.

[0046] LSTM part: Based on CNN, the features output by the pooling layer are used as the input of LSTM. The number of hidden layer nodes of LSTM is set and adjusted according to the time series characteristics of the data and the training effect. For example, it is set to 128 hidden layer nodes. LSTM processes time series data through its internal gating mechanism (input gate, forget gate, output gate) to capture long-term dependencies in the data. At each time step, LSTM updates the hidden state and output state of the current time step based on the input data and the hidden state of the previous time step.

[0047] Model training: Collect a large amount of historical operation and maintenance data, including normal operation data and data under different fault types. Divide this data into training set and validation set according to a certain ratio (such as 70% for training and 30% for validation). Input the feature vectors in the training set into the constructed fusion CNN and LSTM model, and use a suitable loss function (such as cross entropy loss function) and optimization algorithm (such as Adam optimization algorithm) to train the model. During the training process, adjust the parameters of the model (such as the number of convolution kernels, the number of hidden layer nodes, etc.) according to indicators such as the accuracy of the validation set until the model achieves good performance on the validation set (such as an accuracy rate of more than 90%).

[0048] Select multiple sets of actual operation and maintenance data of the main drive system of the tunnel boring machine, each set of data covers different operating conditions and fault conditions. Divide each set of data into training set, test set and validation set according to a certain ratio (such as 80% for training and testing, 20% for final verification).

[0049] Method 1: Fault diagnosis method based on traditional artificial feature extraction (such as simple statistical features) and support vector machine (SVM). Artificial feature extraction mainly calculates statistical features such as mean, variance, peak value, etc. of data, and then inputs these features into SVM for fault diagnosis.

[0050] Method 2: Only convolutional neural network (CNN) is used for fault diagnosis without combining LSTM, that is, the preprocessed data (without feature extraction based on multi-scale morphological filtering and Hilbert-Huang transform) is directly input into CNN for training and diagnosis.

[0051] The present invention proposes a feature extraction method based on the combination of multi-scale morphological filtering and Hilbert-Huang transform (HHT) and a fault diagnosis model integrating convolutional neural network (CNN) and long short-term memory network (LSTM) (hereinafter referred to as the present invention method).

[0052] Evaluation indicators: Accuracy: Calculate the ratio of the number of correctly diagnosed fault samples to the total number of fault samples. The higher the accuracy, the better the accuracy of fault diagnosis. Recall: Calculate the ratio of the number of samples of the fault type that are correctly diagnosed to the number of samples of the actual fault type. The higher the recall, the more comprehensive the detection of various fault types. F1 value: An indicator that combines accuracy and recall, calculated by the formula F1 = 2 × (accuracy × recall) / (accuracy + recall). The higher the F1 value, the better the overall performance of the fault diagnosis model.

[0053] Test results: After comparing multiple sets of data, the accuracy of the method of the present invention is more than 92% on average, while the average accuracy of method one is about 80% and the average accuracy of method two is about 85%. This is because the method of the present invention can more accurately extract the feature vector reflecting the operating status of the main drive system through an innovative feature extraction method, and the model integrating CNN and LSTM can better learn the mapping relationship between these features and fault types. In terms of recall rate, the method of the present invention is more than 90% on average, the average recall rate of method one is about 75%, and the average recall rate of method two is about 80%. The high recall rate of the method of the present invention is due to the fact that its feature extraction method can highlight the key features of the signal, and the effective processing of LSTM on time series data, so that the model can detect various fault types more comprehensively. In terms of F1 value, the method of the present invention is more than 91% on average, the average F1 value of method one is about 78%, and the average F1 value of method two is about 83%. In summary, the method of the present invention is significantly superior to the traditional manual feature extraction and the method using only CNN in the comprehensive performance of fault diagnosis, and can more accurately and comprehensively diagnose the fault type of the main drive system of the tunnel boring machine.

[0054] 3. Performance prediction and optimization steps

[0055] (I) Detailed modeling and solution steps

[0056] Performance prediction model based on support vector machine regression (SVR) combined with particle swarm optimization (PSO): SVR modeling: Determine the input vector and output vector. The input vector consists of the operation and maintenance data of the main drive system, such as sensor data such as temperature, pressure, vibration, and data such as running time and working conditions. The output vector is the performance indicator to be predicted, such as power, efficiency, torque, etc. For example, for power prediction, the input vector may be [temperature, pressure, vibration, running time, working condition], and the output vector is [power]. Select a suitable kernel function, such as linear kernel function, polynomial kernel function, radial basis kernel function (RBF), etc. According to the distribution characteristics and nonlinearity of the data, in general, for data with a certain nonlinear relationship, it is more appropriate to use the RBF kernel function. Determine other parameters of SVR, such as the penalty factor C and the kernel function parameter γ. The penalty factor C is used to control the degree of penalty for outliers in the training sample, and the kernel function parameter γ affects the shape and fitting effect of the kernel function. Through cross-validation and other methods, the value range of C and γ is preliminarily determined, such as C in the range of [1,10] and γ in the range of [0.1,1].

[0057] PSO optimizes SVR parameters: Initialize the particle swarm. Determine the size of the particle swarm, that is, the number of particles, which is generally set to 20-50 particles. Each particle represents a set of SVR parameter combinations (C, γ). For example, a particle may be [C = 3, γ = 0.5]. Set the particle speed and position update rules. According to the principle of the particle swarm optimization algorithm, the particle speed and position update are affected by its own historical optimal position, the group's historical optimal position, and the current position. The speed and position update of the particle in each iteration is calculated by the formula to make it evolve towards a better parameter combination. Define the fitness function. The mean square error (MSE) of the SVR model on the validation set is used as the fitness function, that is, the fitness function = MSE. The smaller the MSE, the higher the prediction accuracy of the SVR model, and the better the parameter combination represented by the particle. Perform particle swarm optimization iterations. Through multiple iterations (generally set to 100-500 iterations), the particle speed and position are continuously updated so that the particle swarm can search for the optimal SVR parameter combination in the parameter space. In each iteration, the quality of each particle is evaluated according to the fitness function, and the historical optimal position of the group and its own historical optimal position are updated.

[0058] Performance optimization: When the performance prediction model predicts that the performance of the main drive system will decline, the operating parameters that need to be adjusted are determined based on the prediction results and fault diagnosis information. For example, if the power consumption is predicted to be too large, combined with the fault diagnosis information, it is determined that it is a motor problem, then consider adjusting the motor control strategy. For motor control strategy adjustment, such as optimizing the power supply frequency and voltage of the motor, by establishing a mathematical model of the motor (such as an equivalent circuit model of the motor), analyze the changes in performance indicators such as power and efficiency of the motor under different power supply frequencies and voltages. According to the analysis results, determine the power supply frequency and voltage values ​​that can make the motor run near the optimal efficiency point, and then adjust the power supply frequency and voltage of the motor through the control system. For adjusting the transmission ratio of the reducer, according to the relationship between the transmission ratio of the reducer and performance indicators such as torque and power, combined with the predicted performance degradation (such as insufficient torque), calculate the appropriate transmission ratio adjustment value, and then adjust the transmission ratio of the reducer through the mechanical structure. For improving the oil supply and oil temperature of the lubrication system, according to the wear of the bearing (provided by the fault diagnosis information) and the working principle of the lubrication system, establish a mathematical model of the lubrication system, and analyze the lubrication effect and service life of the bearing under different oil supply and oil temperature. Based on the analysis results, the oil supply and oil temperature values ​​that can improve the bearing lubrication conditions are determined, and then the oil supply and oil temperature of the lubrication system are adjusted through the control system.

[0059] Efficiency comparison test: Select multiple groups of actual operation and maintenance data of the main drive system of the tunnel boring machine, and each group of data covers different operating conditions and performance changes. Divide each group of data into training set, test set and validation set according to a certain proportion (such as 80% for training and testing, and 20% for final verification). The following two performance prediction methods are used for comparative tests: Method 1: Only support vector machine regression (SVR) is used for performance prediction, and the SVR parameters are not optimized through the particle swarm optimization algorithm (PSO), that is, the default SVR parameter settings are used. The performance prediction method based on the combination of support vector machine regression (SVR) and particle swarm optimization algorithm (PSO) proposed in the present invention (hereinafter referred to as the method of the present invention).

[0060] Evaluation indicators: Mean square error (MSE): Calculate the average of the squared difference between the predicted value and the actual value. The smaller the MSE, the higher the prediction accuracy. Mean absolute error (MAE): Calculate the average of the absolute difference between the predicted value and the actual value. The smaller the MAE, the higher the prediction accuracy. Prediction accuracy: Calculate the proportion of the number of samples that predict the correct performance change trend (increase, decrease or remain unchanged) to the total number of samples. The higher the prediction accuracy, the more accurately the performance change trend can be predicted.

[0061] In terms of mean square error (MSE), the method of the present invention averages below 0.05, while the average MSE of method one is about 0.1. This is because the method of the present invention optimizes the parameters of SVR through the particle swarm optimization algorithm, so that the SVR model can better fit the data and improve the prediction accuracy. In terms of mean absolute error (MAE), the method of the present invention averages below 0.03, while the average MAE of method one is about 0.08. Similarly, the low MAE of the method of the present invention benefits from its optimization of SVR parameters, which makes the predicted value closer to the actual value. In terms of prediction accuracy, the method of the present invention averages above 90%, while the average prediction accuracy of method one is about 70%. The high prediction accuracy of the method of the present invention is because its optimized SVR model can more accurately capture the nonlinear relationship between the performance indicators of the main drive system and the operation and maintenance data, thereby more accurately predicting the performance change trend. Based on the above efficiency comparison test results, the present invention has obvious advantages in fault diagnosis, performance prediction and optimization of the main drive system of the tunnel boring machine, and can effectively improve the operation and maintenance level and operation efficiency of the main drive system.

[0062] III. Beneficial Technical Effects of the Invention

[0063] 1. Innovative data collection and processing architecture:

[0064] A comprehensive and efficient data acquisition system has been built, which can collect various types of data of the main drive system of the tunnel boring machine in real time, including sensor data and working condition data, and ensure the timeliness and accuracy of the data through high-speed data transmission and preprocessing. Compared with the traditional data acquisition method, the data acquisition unit of the present invention can more comprehensively reflect the operating status of the main drive system, providing a richer information basis for subsequent data analysis and management decisions.

[0065] In the data processing center, advanced data cleaning, feature extraction and analysis algorithms are used, such as the feature extraction method combining PCA and ICA, which can effectively process multi-source heterogeneous data and extract more representative and sensitive fault feature information, overcoming the limitations of traditional data processing methods in processing complex data and improving the accuracy of fault diagnosis and performance prediction.

[0066] 2. Innovation of feature extraction algorithm based on multi-scale morphological filtering and HHT:

[0067] For the first time, multiscale morphological filtering and Hilbert-Huang transform are combined and applied to the feature extraction of operation and maintenance data of the main drive system of a tunnel boring machine. Multiscale morphological filtering can analyze data at different scales, effectively remove noise and interference, and highlight the local characteristics of the data; Hilbert-Huang transform can decompose non-stationary signals into intrinsic mode functions and further extract the time-frequency characteristics of the signal. The combination of the two can more comprehensively and accurately extract the fault feature information in the operation data of the main drive system, providing a more reliable basis for fault diagnosis. Compared with traditional feature extraction methods, such as Fourier transform, wavelet transform, etc., the feature extraction algorithm of the present invention can better handle non-stationary and nonlinear data, and improve the recognition and extraction accuracy of fault features.

[0068] 3. Innovation of fault diagnosis model integrating CNN and LSTM:

[0069] The proposed fault diagnosis model that integrates convolutional neural network and long short-term memory network is an innovative solution for fault diagnosis of the main drive system of tunnel boring machines. CNN can automatically learn the local features and spatial distribution features of data, and LSTM can effectively process the long-term dependencies in time series data. The combination of the two can give full play to their respective advantages and accurately diagnose and classify complex faults in the operation of the main drive system. This fusion model can adapt to the fault diagnosis needs of the main drive system under different working conditions, has strong generalization ability and robustness, and can effectively solve the problems of low accuracy and poor adaptability of traditional fault diagnosis models when facing complex fault modes and variable working conditions.

[0070] 4. Innovation of performance prediction method based on SVR-PSO:

[0071] The performance prediction method based on the combination of support vector machine regression and particle swarm optimization algorithm can utilize the good generalization ability of SVR and the parameter optimization ability of PSO to build a high-precision main drive system performance prediction model. By modeling and optimizing the nonlinear relationship between the performance indicators of the main drive system and the operation and maintenance data, it is possible to more accurately predict the performance change trend of the main drive system in the future, discover potential performance problems in advance, and provide strong support for timely optimization measures. Compared with traditional performance prediction methods, the method of the present invention can achieve higher prediction accuracy with fewer data samples, and has better stability and adaptability, and can effectively cope with the complex operating environment and working condition changes of the main drive system of the tunnel boring machine.

[0072] Embodiment 2:

[0073] Data collection and preprocessing: Temperature sensors, vibration sensors, pressure sensors, and torque sensors are installed on the main drive system of the subway tunnel boring machine. The temperature, vibration, oil pressure, and torque data of the motor, reducer, bearing, and other parts are collected at a sampling frequency of 5 times per second. At the same time, the working condition data such as the excavation speed, cutter head speed, and geological conditions are collected. For example, a platinum resistance temperature sensor is installed on the motor winding, and the measurement accuracy can reach ±0.1℃; an acceleration vibration sensor is installed on the bearing seat, and the measurement range is ±50m / s 2 , resolution 0.01m / s 2 .

[0074] The collected data were preprocessed in real time, and the 3σ criterion was used for outlier detection to remove obviously abnormal data points. The Butterworth low-pass filter algorithm was used with the cutoff frequency set to 100 Hz to remove high-frequency noise interference in the data. All data were standardized so that their values ​​ranged from 0 to 1.

[0075] Feature extraction and fault diagnosis: The preprocessed data is processed using a feature extraction method that combines multiscale morphological filtering with Hilbert-Huang transform. The structural element of the multiscale morphological filter is set to a circle with radii of 2, 4, 6, and 8, and the data is decomposed into 4 scales; then the decomposition result is subjected to Hilbert-Huang transform to obtain 5 intrinsic mode functions (IMFs), and the characteristic parameters such as instantaneous frequency, instantaneous amplitude, and energy of each IMF are calculated to construct a 20-dimensional feature vector.

[0076] The feature vector is input into the fault diagnosis model that integrates CNN and LSTM. The CNN part sets three convolution layers, with convolution kernel sizes of 3×3, 5×5, and 3×3, respectively, and the number of convolution kernels is 16, 32, and 64, respectively. ReLU is used as the activation function; the LSTM part sets two layers of LSTM units, with 128 units in each layer, and finally connects a fully connected layer to output the fault type and severity. By training on the operation and maintenance data of the past 100 excavation sections of the main drive system of the subway tunnel boring machine, the model can accurately diagnose common fault types such as motor overheating, reducer gear wear, and bearing failure, with an accuracy rate of more than 95%.

[0077] 1. Performance prediction and optimization:

[0078] The power prediction model of the main drive system is established by using the performance prediction method based on SVR-PSO. The operation and maintenance data such as oil temperature, oil pressure, cutter head speed, and tunneling speed are selected as input variables, the power of the main drive system is selected as the output variable, and the radial basis function (RBF) is used as the kernel function of SVR. The penalty factor C and kernel function parameter γ of SVR are optimized by particle swarm optimization algorithm. The number of particles is set to 30, the maximum number of iterations is 100, and the learning factor c1=c2=1.5. After optimization, the prediction root mean square error (RMSE) of the model is reduced from 0.15 to below 0.05.

[0079] According to the performance prediction results, when it is predicted that the power of the main drive system will decrease, the motor's operating efficiency is optimized by adjusting the motor's power supply frequency and voltage; at the same time, according to the wear condition of the bearing, the oil supply of the lubrication system is increased, the oil temperature is reduced, and the lubrication performance of the bearing is improved, thereby achieving performance optimization of the main drive system and improving the construction efficiency and equipment reliability of the subway tunnel excavation project.

[0080] Embodiment 3:

[0081] Data acquisition and preprocessing: Due to the complex geological conditions of mountain tunnels, more types of sensors are installed on the main drive system of the tunnel boring machine, such as displacement sensors for monitoring the axial and radial displacement of bearings, and strain gauges for measuring stress changes in key components. The sampling frequency is increased to 10 times per second to capture data changes more timely. For example, the measurement accuracy of the displacement sensor can reach ±0.01mm, and the measurement accuracy of the strain gauge is ±1με.

[0082] The collected data are subjected to outlier detection, filtering and standardization processing using the same data preprocessing method as in Example 1. At the same time, in view of the frequent changes in geological conditions during mountain tunneling, the working condition data are classified and labeled in more detail, such as rock hardness grade, joint and fissure development degree, etc., so as to better analyze the impact of geological conditions on the main drive system.

[0083] Feature extraction and fault diagnosis: In the feature extraction process, the parameters of multi-scale morphological filtering and Hilbert-Huang transform are adjusted according to the data characteristics of the main drive system of the mountain tunnel boring machine. The number of scales of the multi-scale morphological filter is increased to 6, and the radius of the structural element is 3, 5, 7, 9, 11, and 13 respectively, in order to better extract the data features under complex geological conditions; after the Hilbert-Huang transform, 6 intrinsic mode functions IMF are obtained, and their characteristic parameters are calculated to construct a 25-dimensional feature vector.

[0084] The structure of the fault diagnosis model is the same as that of Example 1, but a processing module for geological condition data is added. The geological condition data is encoded and input into the model together with the feature vector, so that the model can consider the impact of geological conditions on the occurrence of faults. After training on the operation and maintenance data of the main drive system of the mountain tunnel boring machine in the past 80 excavation sections, the model's accuracy in fault diagnosis reached more than 93%, and it can effectively diagnose main drive system faults caused by changes in geological conditions, such as bearing damage caused by rock impact, and uneven force on the reducer caused by uneven geology.

[0085] Example 4

[0086] 1. Multi-source data perception and fusion module

[0087] 1. Data Collection

[0088] Various types of sensors are deployed at various stages of the manufacturing site, including equipment status sensors (monitoring the operating parameters and fault information of processing equipment and assembly equipment), material sensors (inventory quantity and location information of raw materials and parts), personnel positioning sensors (worker's working position and status), and environmental sensors (temperature, humidity and other environmental parameters that affect manufacturing). At the same time, data such as order information, production plans, and process documents are obtained from information systems such as the enterprise resource planning (ERP) system and the manufacturing execution system (MES).

[0089] 2. Data fusion and preprocessing

[0090] Data fusion technology is used to integrate data from different sources to eliminate data redundancy and inconsistency. The collected data is pre-processed, including data cleaning (removing outliers and noise data), data standardization (unifying the scale of data of different magnitudes) and data encoding (converting non-numeric data into a form suitable for calculation), providing a high-quality data foundation for subsequent scheduling decisions.

[0091] 2. Manufacturing Task Modeling and Decomposition Module

[0092] 1. Task Modeling

[0093] According to the design drawings and manufacturing process requirements of the tunnel boring machine, a manufacturing task model is established. The model includes the manufacturing tasks, assembly tasks, and quality inspection tasks of each component, and clarifies the input and output, process parameters, quality standards, and time limits of each task.

[0094] 2. Task decomposition and correlation analysis

[0095] Decompose the overall manufacturing task into multiple subtasks, and analyze the sequence, parallel relationship and resource dependency between subtasks. Build a task relationship network to clearly express the logical structure of the entire manufacturing process and provide task-level constraint information for the scheduling algorithm.

[0096] (III) Resource dynamic modeling and management module

[0097] 1. Resource classification and modeling

[0098] Classify and model all types of resources involved in the manufacturing process, including production equipment (machine tools of different types and specifications, welding equipment, etc.), human resources (workers of different skill levels), material resources (raw materials, parts and their supplier information), and auxiliary resources (such as jigs and fixtures, transportation tools, etc.). Define the attributes of each resource, such as the processing capacity of the equipment, the skill range of the personnel, the supply cycle of the material, etc.

[0099] 2. Dynamic monitoring and updating of resource status

[0100] Track resource status changes in real time, and update resource availability, remaining workload, location, and other information in a timely manner through sensor data and information system feedback. Establish a resource status prediction model to predict resource status changes in the future based on historical data and current trends, and provide forward-looking information for scheduling decisions.

[0101] (IV) Intelligent Scheduling Module

[0102] 1. Scheduling model construction

[0103] The intelligent scheduling model is constructed by comprehensively considering manufacturing tasks and resource information. The model takes minimizing the manufacturing cycle, maximizing resource utilization, and ensuring product quality as optimization goals, while considering complex constraints such as equipment maintenance plans, material supply restrictions, and personnel working hours.

[0104] 2. Scheduling decisions driven by new algorithm models

[0105] The following three unreported algorithm models are used to implement intelligent scheduling decisions and generate detailed scheduling plans, including the start time, execution resources, completion time, and material distribution plan of each subtask.

[0106] (V) Scheduling execution and monitoring adjustment module

[0107] 1. Issuance and execution of dispatch instructions

[0108] The generated scheduling plan is converted into specific scheduling instructions, and the instructions are issued to the corresponding execution units through the interface with the equipment control system, personnel management system and material distribution system of the manufacturing site, ensuring that each link carries out work in an orderly manner according to the scheduling plan.

[0109] 2. Execution process monitoring and exception handling

[0110] Monitor the execution of the manufacturing process in real time, and promptly detect abnormal situations such as production delays, equipment failures, and material shortages by comparing the actual progress with the scheduling plan. When an abnormality occurs, activate the emergency response mechanism, use the dynamic scheduling algorithm to adjust the original scheduling plan, reallocate resources, and adjust the task sequence to ensure the continuity of the manufacturing process.

[0111] 4. Three scheduling solutions based on the new algorithm model

[0112] 1. Scheduling model based on hyper-heuristic evolutionary fuzzy system

[0113] 1. Model Principle

[0114] The hyper-heuristic evolutionary fuzzy system combines the advantages of the hyper-heuristic algorithm and the evolutionary fuzzy system. The hyper-heuristic algorithm is used to select and combine low-level heuristic algorithms at a high level to adapt to different scheduling scenarios. The evolutionary fuzzy system optimizes fuzzy rules and membership functions through evolutionary algorithms, so that it can handle the ambiguity and uncertainty in manufacturing scheduling, such as the fuzzy evaluation of worker skill levels and the fuzzy estimation of equipment failure risks.

[0115] 2. Algorithm Implementation

[0116] First, a set of multiple low-level heuristic scheduling algorithms is constructed. The hyper-heuristic algorithm selects appropriate heuristic algorithms from the set for combination according to the characteristics of the current manufacturing tasks and resources. At the same time, the evolutionary algorithm is used to evolve the rule base and membership function of the fuzzy system. During the scheduling process, the information of tasks and resources is input into the evolutionary fuzzy system, and scheduling decisions such as task allocation, resource selection and time scheduling are obtained based on fuzzy reasoning.

[0117] 3. Advantages

[0118] The model can adaptively select appropriate scheduling strategies in complex scheduling environments and effectively deal with various fuzzy and uncertain factors in tunnel boring machine manufacturing. Compared with traditional scheduling algorithms, it has stronger robustness and versatility, and can improve the quality and efficiency of scheduling decisions.

[0119] 2. Dynamic Scheduling Model Integrating Echo State Network and Ant Colony Optimization

[0120] 1. Model Principle

[0121] The echo state network (ESN) is a special recursive neural network that has the ability to quickly model dynamic systems. The manufacturing process is regarded as a dynamic system and is modeled using ESN to learn the changing patterns of manufacturing tasks and resources in time series. The ant colony optimization algorithm searches for the optimal solution by simulating the pheromone propagation mechanism of ants in the process of searching for food. In this model, ants search for scheduling solutions in the solution space composed of tasks and resources, and the update of pheromones is affected by the output of ESN, so that the ant colony optimization algorithm can better adapt to the dynamic changes of the manufacturing process.

[0122] 2. Algorithm Implementation

[0123] First, the echo state network is trained, taking historical manufacturing data (including task progress, resource status, etc.) as input and outputting predictions of future manufacturing status. Then, in the ant colony optimization algorithm, ants select the next task and resource for scheduling decisions based on the pheromone concentration and the prediction results of ESN. After each iteration, the pheromone concentration is updated according to the quality of the scheduling plan, and the new manufacturing status is updated and predicted using ESN, guiding the ant colony to search for a better scheduling plan.

[0124] 3. Advantages

[0125] This fusion model can make full use of the modeling ability of the echo state network for dynamic systems and the search ability of the ant colony optimization algorithm to achieve dynamic and accurate scheduling of the tunnel boring machine manufacturing process. It can effectively respond to real-time changes in the manufacturing process, such as urgent order insertion, sudden equipment failure, etc., and improve the flexibility and response speed of scheduling.

[0126] 3. Hybrid Scheduling Model Based on Quantum Annealing and Particle Swarm Optimization

[0127] 1. Model Principle

[0128] The quantum annealing algorithm uses quantum fluctuation characteristics to search for the optimal solution in the solution space, and has the potential to quickly find the global optimal solution in complex energy terrain. The particle swarm optimization algorithm simulates the foraging behavior of bird flocks, allowing particles to fly in the solution space and adjust the flight direction based on their own and the group's experience to find the optimal solution. Combining the two, the quantum annealing algorithm is used to explore the high-quality solution area globally, and the particle swarm optimization algorithm performs a detailed search in the area found by the quantum annealing algorithm to improve the accuracy of the solution.

[0129] 2. Algorithm Implementation

[0130] In the quantum annealing part, the scheduling problem is mapped to the spin state space of the quantum bit. By setting the appropriate Hamiltonian and annealing plan, the system gradually converges to a low-energy state under the action of quantum fluctuations, that is, a better scheduling solution. Then, the solution obtained by quantum annealing is used as the initial population of the particle swarm optimization algorithm, and the speed and position update rules of the particles are set to make the particles evolve towards a better scheduling solution in the solution space. In the whole process, the quality of the solution is evaluated according to the scheduling objective function (such as manufacturing cycle, resource utilization, etc.) to guide the search direction of the algorithm.

[0131] 3. Advantages

[0132] The hybrid model combines the global search capability of the quantum annealing algorithm with the local optimization capability of the particle swarm optimization algorithm, and can more effectively search for the optimal scheduling solution in the complex scheduling problem space. For scheduling problems such as tunnel boring machine collaborative manufacturing with a large number of constraints and complex objectives, this hybrid algorithm can improve the quality of the scheduling solution and reduce the calculation time.

[0133] The present invention has been described in detail above in conjunction with the embodiments. However, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be changed to form a plurality of specific embodiments, which are all within the common variation range of the present invention and will not be described in detail herein.

Claims

1. A main drive system design method based on tunnel boring machine operation and maintenance data, characterized in that: It involves the following coherent and coordinated steps: S1 Data acquisition and preprocessing: The data acquisition unit sets the sampling frequency to collect sensor data and working condition data in real time according to the main drive system of the tunnel boring machine; the collected data uses low-pass, high-pass, and band-pass digital filtering algorithms to remove noise interference, and is converted into a standard value range through standardization processing; The S2 data processing center uses multi-scale morphological filtering to perform multi-scale decomposition on pre-processed data, remove background noise and local interference, extract features of different scales, and then decompose the non-stationary signal into intrinsic mode functions through Hilbert-Huang transform, calculate its instantaneous frequency and amplitude parameters to construct feature vectors; using the fault diagnosis model that integrates convolutional neural network and long short-term memory network, relying on historical operation and maintenance data training, learn the mapping relationship between fault type and feature vector, and accurately diagnose common faults such as motor winding short circuit and reducer gear wear; S3 performance prediction and optimization: First use SVR to build a nonlinear regression model of performance indicators and operation and maintenance data, and then use PSO to optimize its parameters to predict the future performance trend of the system.

2. The main drive system design method based on tunnel boring machine operation and maintenance data according to claim 1 is characterized in that: It also includes using a hyper-heuristic algorithm to select appropriate heuristic algorithms from a set for combination according to the characteristics of current manufacturing tasks and resources, and using an evolutionary algorithm to perform evolutionary operations on the rule base and membership function of the fuzzy system; In the scheduling process, the information of tasks and resources is input into the evolutionary fuzzy system, and the scheduling decision is obtained based on fuzzy reasoning.

3. The main drive system design method based on tunnel boring machine operation and maintenance data according to claim 1, characterized in that: It also includes performance prediction and optimization steps. A performance prediction model of the main drive system is established based on a performance prediction method that combines support vector machine regression with a particle swarm optimization algorithm. The performance of the main drive system is optimized according to the prediction results and fault diagnosis information.

4. The main drive system design method based on tunnel boring machine operation and maintenance data according to claim 1, characterized in that: The structural element of the multi-scale morphological filtering is a circle with a radius of 2, 4, 6, 8 or adjusted to 3, 5, 7, 9, 11, 13 according to geological conditions. The data is decomposed into 4 or 6 scales, and 5 or 6 intrinsic mode functions IMF are obtained after Hilbert-Huang transform, and their characteristic parameters are calculated to construct feature vectors.

5. The main drive system design method based on tunnel boring machine operation and maintenance data according to claim 1, characterized in that: In the fault diagnosis model integrating convolutional neural network and long short-term memory network, three convolutional layers are set in the CNN part, and the convolution kernel sizes are 3×3, 5×5, and 3×3 respectively, and the number of convolution kernels are 16, 32, and 64 respectively. ReLU is used as the activation function, and two layers of LSTM units are set in the LSTM part, and the number of units in each layer is 128. Finally, the fully connected layer is connected to output the fault type and severity.

6. The main drive system design method based on tunnel boring machine operation and maintenance data according to claim 1, characterized in that: In the performance prediction method based on the combination of support vector machine regression and particle swarm optimization algorithm, radial basis function is used as the kernel function of SVR, the number of particles is set to 30, the maximum number of iterations is 100, the learning factor c1=c2=1.5, and a high-precision power prediction model is constructed by optimizing the penalty factor C of SVR and the kernel function parameter γ.

7. The method according to any one of claims 1 to 6, characterized in that: In tunneling projects under different geological conditions, such as subway tunnels or mountain tunnels, the sensor type, sampling frequency, data processing parameters and model structure are adjusted according to the actual situation to achieve effective intelligent management of the main drive system of the tunnel boring machine.

Images