Shield tunneling machine cutter wear prediction method and system

Abstract

The invention provides a shield tunneling machine cutter wear prediction method and system, and belongs to the technical field of shield tunneling. According to the method, the shield tunneling parameters and the vibration signals are fused, intelligent prediction from shield tunneling process data to the cutter abrasion degree is achieved, a new solution is provided for accurate evaluation of the cutter state and efficient cutter replacement in the shield tunneling process, meanwhile, important technical support is provided for shield construction under the complex stratum condition, and the method is suitable for popularization and application. And the cutter changing benefit in the shield tunneling process can be maximized.

Description

Technical Field

[0001] The invention belongs to the technical field of shield machines, and in particular relates to a shield machine tool wear prediction method and system. Background Art

[0002] Shield tunneling is the process of interaction between the shield machine cutter and the stratum. The degree of cutter wear directly affects the construction progress and efficiency of the project. In hard rock strata, shield tunneling is usually accompanied by frequent opening and changing of cutters. The opening and changing of cutters is technically difficult and risky, which seriously affects the safe and efficient construction of the entire shield tunneling process. Therefore, accurately determining the timing of cutter change is crucial to reducing project losses and improving construction efficiency.

[0003] At present, the timing of tool change mainly depends on the experience of on-site construction personnel. Some studies have attempted to establish a mapping relationship between shield tunneling parameters and tool wear by monitoring the entire tool wear process, and use regression fitting models to predict tool wear progress. However, these methods still face two major bottlenecks: first, the limited size of data samples leads to insufficient generalization of the model; second, the empirical formula has weak fitting ability, and the prediction accuracy is low when applied across projects.

[0004] Therefore, it is necessary to provide a shield machine tool wear prediction method and system to solve the above problems. Summary of the invention

[0005] The present invention provides a shield machine tool wear prediction method and system, which predicts the degree of tool wear by fusing excavation parameters and vibration signals, thereby accurately determining the tool change timing, and can effectively solve at least one technical problem involved in the background technology.

[0006] In order to solve the above-mentioned technical problems, the present invention is achieved as follows: A shield machine tool wear prediction method comprises the following steps: Step S1, collecting tunneling parameters and vibration signals during tunneling of the shield machine; Step S2, collaboratively filtering the tunneling parameters and vibration signals based on the singular value decomposition method to capture the principal component signals related to tool wear in the tunneling parameters and vibration signals; Step S3, based on the work log of the shield machine in the actual engineering scenario, the traditional empirical formula for tool wear is corrected, and the corrected tool wear formula is used to generate tool wear degree data; Step S4, extracting features from the vibration signal and the main component signal of the tunneling parameter respectively, fusing the extracted vibration signal features and tunneling parameter features through matrix splicing, constructing a unified feature representation, and then using a fully connected neural network to establish a mapping model from the fusion features to the tool wear degree, using the tool wear formula to generate a large amount of tool wear degree data under different tunneling parameters and vibration signals, and training the mapping model; Step S5, for any shield machine tool wear prediction process, the fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model, and the prediction result of the tool wear degree is output.

[0007] As a preferred improvement, the excavation parameters include the average thrust speed, total thrust force, cutter head torque, cutter head inclination, cutter head speed, cutter head penetration, thrust cylinder stroke and articulated cylinder stroke; the vibration signal includes the cutting vibration signal between the tool and the formation and the tool rotation vibration signal.

[0008] As a preferred improvement, step S2 specifically includes the following steps: Step S21, obtaining power spectrum data of the excavation parameters and the vibration signal, and fitting the power spectrum of the excavation parameters and the vibration signal respectively using a Gaussian function superposition model; Step S22, calculating the half bandwidth of each Gaussian component in the excavation parameter and the vibration signal based on the fitted Gaussian function model, and determining the main component frequency band range of the excavation parameter and the vibration signal based on the half bandwidth; Step S23, stacking the one-dimensional excavation parameters and vibration signals into a three-dimensional tensor using a bitwise loop method; Step S24, using a three-dimensional tensor singular value decomposition method to decompose the obtained three-dimensional tensor into three tensor matrices USV; Step S25, searching for columns located within the frequency band of the principal component in the spectrum of the tensor matrix U, extracting them separately and regenerating the signals using the reverse loop method to obtain principal component signals of the excavation parameters and the vibration signals.

[0009] As a preferred improvement, in step S21, the Gaussian function superposition model is expressed as: ; In the formula, represents the fitting function of the power spectrum; Indicates the ordinal index of the significant peak in the power spectrum; represents the total number of significant peaks; Indicates the power value of the significant peak; represents the power mean of significant peaks; Indicates the standard deviation of the significant peak power value; The fitting function of the power spectrum is solved by the Levenberg–Marquardt algorithm.

[0010] As a preferred improvement, in step S22, the calculation process of the half-broadband is expressed as: ; In the formula, represents half bandwidth; The frequency band range of the principal component is expressed as: , where Indicates the larger of the frequencies corresponding to the highest points in the frequency spectrum of the excavation parameters and the vibration signal.

[0011] As a preferred improvement, in step S23, the bitwise rotation method specifically includes the following steps: defining the window length, step size and number of cycles, cyclically shifting the excavation parameters and vibration signals according to the defined window length, step size and cycle coefficient, generating multiple groups of window segments, and stacking the window matrices of the excavation parameters and vibration signals along the third dimension to form a three-dimensional tensor.

[0012] As a preferred improvement, in step S24, the three-dimensional tensor singular value decomposition method decomposes the three-dimensional tensor into the product of three-order tensors based on the tube product of the tensor, and the decomposition form is expressed as: ; In the formula, Represents a three-dimensional tensor; express orthogonal matrix of order; express orthogonal matrix of order; Represents a matrix consisting of non-negative diagonal elements in descending order A rectangular diagonal matrix with singular values ​​on the diagonal.

[0013] As a preferred improvement, step S3 specifically includes the following steps: Step S31, collect the propulsion speed time series of each ring of the shield machine, perform wavelet transform on the propulsion speed time series of each ring, take the low-frequency approximation of the last layer of wavelet transform to form a low-frequency approximation sequence, and calculate the standard deviation of the low-frequency approximation sequence; the calculation process of the standard deviation is expressed as: ; In the formula, Indicates the shield machine The standard deviation of the low-frequency approximation sequence of the ring; Indicates the shield machine The first elements; Indicates the shield machine The average value of the low-frequency approximate sequence of the ring; N represents the shield machine The total number of elements in the low-frequency approximation sequence of the ring; Step S32, the standard deviations of all loops are combined to form a standard deviation sequence, and Fourier denoising is performed on the standard deviation sequence. The process of Fourier denoising is expressed as: ; ; ; In the formula, represents Fourier transform; represents the sequence after Fourier transformation; It means that the sequence after Fourier transformation is truncated, and all elements from the 1 / 3 length position to the end of the sequence are assigned 0; represents inverse Fourier transform; represents the sequence after inverse Fourier transformation; Step S33, the standard deviation after noise reduction is calculated by Get the final tool wear index .

[0014] As a preferred improvement, a one-dimensional convolutional neural network is used to perform feature extraction on vibration signals; and a fully connected neural network is used to perform feature extraction on excavation parameters.

[0015] A system for executing the above-mentioned shield machine tool wear prediction method comprises: Signal acquisition module, used to collect tunneling parameters and vibration signals during the tunneling process of the shield machine; The principal component extraction module is used to perform collaborative filtering on the tunneling parameters and vibration signals based on the singular value decomposition method, and to capture the principal component signals related to tool wear in the tunneling parameters and vibration signals; The tool wear degree data generation module is used to correct the traditional tool wear empirical formula based on the work log of the shield machine in the actual engineering scenario, and use the corrected tool wear formula to generate tool wear degree data; The mapping model building module is used to extract features from the main component signals of the vibration signal and the excavation parameter respectively, fuse the extracted vibration signal features and the excavation parameter features through matrix splicing, build a unified feature representation, and then use a fully connected neural network to establish a mapping model from the fusion features to the tool wear degree. The tool wear formula is used to generate a large amount of tool wear degree data under different excavation parameters and vibration signals, and train the mapping model; The prediction module is used to predict the tool wear of any shield machine tool. The fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model to output the prediction result of the tool wear degree.

[0016] The beneficial effects of the present invention are: The present invention realizes intelligent prediction of cutter wear degree from shield tunneling process data by integrating shield tunneling parameters and vibration signals, providing a new solution for accurate evaluation of cutter status and efficient cutter changing during shield tunneling. It also provides important technical support for shield construction under complex geological conditions and can maximize the benefits of cutter changing during shield tunneling. DETAILED DESCRIPTION

[0017] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] This embodiment provides a shield machine tool wear prediction method, comprising the following steps: Step S1, collecting tunneling parameters and vibration signals during the tunneling process of the shield machine.

[0019] The tunneling parameters include the average thrust speed, total thrust, cutter head torque, cutter head inclination, cutter head speed, cutter head penetration, thrust cylinder stroke and articulation cylinder stroke. The tunneling parameters are collected from the operation data of the shield machine.

[0020] The vibration signal includes the cutting vibration signal between the tool and the stratum and the tool rotation vibration signal. The cutting vibration signal between the tool and the stratum is collected by a single-axis sensor arranged at the soil compartment partition behind the cutterhead; the tool rotation vibration signal is collected by a three-axis sensor arranged at the motor, gearbox, reducer, bearing and screw conveyor in the excavation system. The acquisition frequency of all sensors is 1000Hz, and 1000 sets of high-resolution vibration data can be collected per second, so as to fully record the dynamic information during the excavation process.

[0021] During the tunneling process of the shield machine, the vibration signal of the tool cutting the stratum and the rotation vibration signal of the tool itself can both reflect the vibration state of the tool. Therefore, by collecting the vibration signals of these two parts at the same time, we can more comprehensively obtain the vibration condition of the tool during the tunneling process.

[0022] Step S2, collaboratively filtering the tunneling parameters and vibration signals based on the singular value decomposition method, capturing the principal component signals related to the tool wear in the tunneling parameters and vibration signals.

[0023] In the traditional signal processing field, usually only a single signal is considered, for example, only the influence of excavation parameters on tool wear is considered, and the subsequent wear prediction process is performed after filtering out the noise. The disadvantage is that the main component of the signal cannot be effectively determined, and it is difficult to determine the selection of the interception frequency. The applicant found in the study that the rotation of the tool is driven by the motor of the motor excavation system and the conduction between the components in the excavation system (including but not limited to the gear box, reducer, bearing), which can be effectively transmitted in all directions to the rear screw conveyor that transports the front slag. The cutting action between the tool and the stratum can be accompanied by strong vibration, which can be axially transmitted to the rear soil compartment partition, etc. Therefore, the wear of the tool is related to the excavation parameters and the vibration signal at the same time. There is a certain coupling correlation between the excavation parameters and the vibration signal. The present invention uses the idea that there are common principal components of the excavation parameters and the vibration signal, and finds and extracts the common principal components of the two signals to identify the working state of the tool.

[0024] Step S2 specifically includes the following steps: Step S21, obtaining the power spectrum data of the excavation parameters and the vibration signal, and using the Gaussian function superposition model to fit the power spectrum of the excavation parameters and the vibration signal respectively.

[0025] The number of Gaussian functions can be selected according to the number of significant peaks in the power spectrum. For example, when the number of significant peaks is three, three Gaussian functions can be selected to fit the power spectrum. The superposition model of Gaussian functions is expressed as: ; In the formula, represents the fitting function of the power spectrum; Indicates the ordinal index of the significant peak; represents the total number of significant peaks; Indicates the power value of the significant peak; represents the power mean of significant peaks; Represents the standard deviation of the significant peak power values.

[0026] The fitting function of the power spectrum can be solved by the Levenberg-Marquardt algorithm, and the solution process belongs to the conventional technology in this field, which will not be described in detail in this embodiment.

[0027] Step S22, calculating the half bandwidth of each Gaussian component in the excavation parameter and the vibration signal based on the fitted Gaussian function model, and determining the main component frequency band range of the excavation parameter and the vibration signal based on the half bandwidth.

[0028] The calculation process of half-broadband is expressed as: ; In the formula, represents half bandwidth; Since the periodic content in the signal occupies most of the energy, the principal components are generally distributed in Between, where Indicates the larger of the excavation parameter and the frequency corresponding to the highest point in the frequency spectrum of the vibration signal; Step S23, stacking the one-dimensional excavation parameters and vibration signals into a three-dimensional tensor using a bitwise loop method; The bitwise rotation method specifically includes the following steps: defining the window length, step size and number of cycles, cyclically shifting the excavation parameters and vibration signals according to the defined window length, step size and cycle coefficient, generating multiple groups of window segments, and stacking the window matrices of the excavation parameters and vibration signals along the third dimension (channel) to form a three-dimensional tensor.

[0029] Step S24, using a three-dimensional tensor singular value decomposition method to decompose the obtained three-dimensional tensor into three tensor matrices USV.

[0030] The three-dimensional tensor singular value decomposition method decomposes the three-dimensional tensor into the product of three-order tensors based on the tube product of the tensor. The decomposition form is expressed as: ; In the formula, Represents a three-dimensional tensor; express orthogonal matrix of order; express orthogonal matrix of order; Represents a matrix consisting of non-negative diagonal elements in descending order Rectangular diagonal matrix, the elements on the diagonal are singular values; Step S23, searching for columns located within the frequency band of the principal component in the spectrum of the tensor matrix U, extracting them separately and regenerating the signals using the reverse loop method to obtain principal component signals of the excavation parameters and the vibration signals.

[0031] Unnecessary noise is filtered out from the main component signals of the excavation parameters and vibration signals, and the working status of the tool can be displayed intuitively.

[0032] Step S3, based on the work log of the shield machine in the actual engineering scenario, the traditional empirical formula for tool wear is corrected, and the tool wear degree data is generated using the corrected tool wear formula.

[0033] Step S3 specifically includes the following steps: Step S31, collecting the propulsion speed time series of each ring of the shield machine, performing wavelet transform on the propulsion speed time series of each ring, taking the low-frequency approximation of the last layer of the wavelet transform to form a low-frequency approximation sequence, and calculating the standard deviation of the low-frequency approximation sequence; The essence of wavelet transform is a multi-level decomposition process. Each layer represents a multi-scale analysis of the signal, peeling off the different frequency components of the signal layer by layer. Specifically: the first layer decomposes the original signal into high-frequency details and low-frequency approximations; the second layer decomposes the low-frequency approximations of the first layer again to obtain higher-frequency high-frequency details and lower-frequency low-frequency approximations; repeat this process until the preset number of decomposition layers is reached.

[0034] In this embodiment, wavelet transform is used for signal denoising, which includes 4 layers in total. Noise is generally reflected in high-frequency details, so this application retains the low-frequency approximation of the 4th layer and discards the high-frequency details. The low-frequency approximation of the 4th layer is highly correlated with the wear state of the tool and can directly reflect the wear condition of the tool.

[0035] The calculation process of standard deviation is expressed as: ; In the formula, Indicates the shield machine The standard deviation of the low-frequency approximation sequence of the ring; Indicates the shield machine The first elements; Indicates the shield machine The average value of the low-frequency approximate sequence of the ring; N represents the shield machine The total number of elements in the low-frequency approximation sequence of the ring; Step S32: combine the standard deviations of all loops to form a standard deviation sequence, and perform Fourier denoising on the standard deviation sequence.

[0036] Fourier denoising converts the original signal from the time domain to the frequency domain through Fourier transform to obtain the spectrum of the signal, then removes the high-frequency noise part by direct truncation, and restores the original signal through inverse Fourier transform.

[0037] The process of Fourier denoising is expressed as ; ; ; In the formula, represents Fourier transform; represents the sequence after Fourier transformation; It means that the sequence after Fourier transformation is truncated, and all elements from the 1 / 3 length position to the end of the sequence are assigned 0; represents inverse Fourier transform; represents the sequence after inverse Fourier transformation.

[0038] Step S33, the standard deviation after noise reduction is calculated by Get the final tool wear index .

[0039] It is understandable that the traditional empirical formula for tool wear is a general formula that fits a large amount of data. In actual engineering scenarios, it often has the problem of insufficient generalization ability, which is manifested in the large oscillation of the tool wear curve and the inability to accurately reflect the actual degree of wear. Therefore, the present invention collects the work log in the actual engineering scenario and corrects it, so that it can better perform the tool wear prediction in the actual engineering scenario.

[0040] It should be noted that the work log generally only records the relevant data when the tool is fully worn (reaching the tool change standard). Therefore, if only the relevant data in the work log is used as the input of the subsequent fully connected neural network, there will be a problem of too few and single data samples. The revised tool wear formula can generate wear data for the entire life cycle of the tool, which can provide a large number of data samples for subsequent training, which is conducive to the rapid fitting of the model.

[0041] Step S4, feature extraction is performed on the main component signals of the vibration signal and the excavation parameter respectively, the extracted vibration signal features and the excavation parameter features are fused through matrix splicing to construct a unified feature representation, and then a mapping model from the fused features to the tool wear degree is established using a fully connected neural network. A large amount of tool wear degree data under different excavation parameters and vibration signals is generated using the tool wear formula, and the mapping model is trained.

[0042] Since the time scales of the excavation parameters and the vibration signal do not match, which is mainly manifested in the inconsistency of frequency, the vibration signal and the excavation parameters need to be pre-processed before feature extraction. The present invention uses the root mean square to calculate the effective value of the signal to align the time scale. The processing process is expressed as follows: ; In the formula, Represents the signal output after alignment; Represents the input signal, represents the index in the signal sequence, Represents the number of samples in the signal sequence.

[0043] Since vibration signals have strong periodicity, the present invention adopts a one-dimensional convolutional neural network (1D CNN) to perform the feature extraction process of vibration signals. The sliding of the convolution kernel during the convolution process can effectively detect short-term trends or fluctuations; its parameter sharing mechanism enables the model to effectively capture position-independent patterns in the input sequence; it does not rely on time step calculation and is more efficient.

[0044] The one-dimensional convolutional neural network consists of two convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. The maximum pooling layer is then used to reduce the spatial dimension of the feature map, gradually extracting more abstract and invariant features from the vibration signal. Finally, the feature map is flattened and passed through two fully connected layers to further optimize the learned feature representation, and finally a 32-dimensional feature vector is output.

[0045] A fully connected neural network is a basic deep learning model, which consists of a series of fully connected layers, and each neuron between layers is connected to each other. It performs well in processing high-dimensional structured data (such as tabular data, etc.), and can effectively model the complex nonlinear relationship between high-dimensional structured tunneling parameters and tool wear. Therefore, the present invention uses a fully connected neural network to extract features from tunneling parameters.

[0046] The fully connected neural network includes a fully connected layer, a batch normalization layer, a ReLU activation function, and a regularization layer connected in sequence. The input tunneling parameters are first mapped to a 64-dimensional feature space and then further reduced to 32 dimensions. Batch normalization is applied after each linear transformation to improve feature extraction stability, and the regularization layer is used to randomly deactivate a small number of neurons during training to reduce overfitting.

[0047] Step S5, for any shield machine tool wear prediction process, the fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model, and the prediction result of the tool wear degree is output.

[0048] In order to verify the effectiveness of the method provided by the present invention, a total of 14 input parameters including shield tunneling parameters (average propulsion speed, total propulsion force, cutter head torque, inclination angle, cutter head speed, penetration, 2# propulsion cylinder stroke, 5# propulsion cylinder stroke, 7# propulsion cylinder stroke, 10# propulsion cylinder stroke, 1# articulated cylinder stroke, 2# articulated cylinder stroke, 3# articulated cylinder stroke, 4# articulated cylinder stroke) in a certain construction project and minute-level effective values ​​of vibration signals of all measuring points (motor, gearbox, reducer, bearing, screw conveyor - XYZ three-axis; upper and lower soil bin partitions - X single axis) were selected to train the model. The R-square (evaluation index) of the model on the test set reached an accuracy of 95%, with excellent performance. In addition, two machine learning models of other methods were set up as comparative experiments, using a random forest algorithm based only on tunneling parameters and an FNN algorithm based only on tunneling parameters, respectively, with an accuracy of less than 90%, further verifying the superiority of the present invention.

[0049] This embodiment also provides a system for executing the above-mentioned shield machine tool wear prediction method, comprising: Signal acquisition module, used to collect tunneling parameters and vibration signals during the tunneling process of the shield machine; The principal component extraction module is used to perform collaborative filtering on the tunneling parameters and vibration signals based on the singular value decomposition method, and to capture the principal component signals related to tool wear in the tunneling parameters and vibration signals; The tool wear degree data generation module is used to correct the traditional tool wear empirical formula based on the work log of the shield machine in the actual engineering scenario, and use the corrected tool wear formula to generate tool wear degree data; The mapping model building module is used to extract features from the main component signals of the vibration signal and the excavation parameter respectively, fuse the extracted vibration signal features and the excavation parameter features through matrix splicing, build a unified feature representation, and then use a fully connected neural network to establish a mapping model from the fusion features to the tool wear degree. The tool wear formula is used to generate a large amount of tool wear degree data under different excavation parameters and vibration signals, and train the mapping model; The prediction module is used to predict the tool wear of any shield machine tool. The fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model to output the prediction result of the tool wear degree.

[0050] The embodiments of the present invention are described above, but the present invention is not limited to the above-mentioned specific implementation modes. The above-mentioned specific implementation modes are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A shield machine tool wear prediction method, characterized in that: The steps include: Step S1, collecting tunneling parameters and vibration signals during tunneling of the shield machine; Step S2, collaboratively filtering the tunneling parameters and vibration signals based on the singular value decomposition method to capture the principal component signals related to tool wear in the tunneling parameters and vibration signals; Step S3, based on the work log of the shield machine in the actual engineering scenario, the traditional empirical formula for tool wear is corrected, and the corrected tool wear formula is used to generate tool wear degree data; Step S4, extracting features from the vibration signal and the main component signal of the tunneling parameter respectively, fusing the extracted vibration signal features and tunneling parameter features through matrix splicing, constructing a unified feature representation, and then using a fully connected neural network to establish a mapping model from the fusion features to the tool wear degree, using the tool wear formula to generate a large amount of tool wear degree data under different tunneling parameters and vibration signals, and training the mapping model; Step S5, for any shield machine tool wear prediction process, the fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model, and the prediction result of the tool wear degree is output.

2. The shield machine tool wear prediction method according to claim 1, characterized in that: The excavation parameters include the average thrust speed, total thrust force, cutter head torque, cutter head inclination, cutter head speed, cutter head penetration, thrust cylinder stroke and articulated cylinder stroke; the vibration signals include the cutting vibration signal between the tool and the formation and the tool rotation vibration signal.

3. The shield machine tool wear prediction method according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21, obtaining power spectrum data of the excavation parameters and the vibration signal, and fitting the power spectrum of the excavation parameters and the vibration signal respectively using a Gaussian function superposition model; Step S22, calculating the half bandwidth of each Gaussian component in the excavation parameter and the vibration signal based on the fitted Gaussian function model, and determining the main component frequency band range of the excavation parameter and the vibration signal based on the half bandwidth; Step S23, stacking the one-dimensional excavation parameters and vibration signals into a three-dimensional tensor using a bitwise loop method; Step S24, using a three-dimensional tensor singular value decomposition method to decompose the obtained three-dimensional tensor into three tensor matrices USV; Step S25, searching for columns located within the frequency band of the principal component in the spectrum of the tensor matrix U, extracting them separately and regenerating the signals using the reverse loop method to obtain principal component signals of the excavation parameters and the vibration signals.

4. The shield machine tool wear prediction method according to claim 3, characterized in that: In step S21, the Gaussian function superposition model is expressed as: ; In the formula, represents the fitting function of the power spectrum; Indicates the ordinal index of the significant peak in the power spectrum; represents the total number of significant peaks; Indicates the power value of the significant peak; represents the power mean of significant peaks; Indicates the standard deviation of significant peak power values; The fitting function of the power spectrum is solved by the Levenberg–Marquardt algorithm.

5. The shield machine tool wear prediction method according to claim 4, characterized in that: In step S22, the calculation process of the half-bandwidth is expressed as: ； In the formula, represents half bandwidth; The frequency band range of the principal component is expressed as: , where Indicates the larger of the frequencies corresponding to the highest points in the frequency spectrum of the excavation parameters and the vibration signal.

6. The shield machine tool wear prediction method according to claim 3, characterized in that: In step S23, the bitwise rotation method specifically includes the following steps: defining the window length, step size and number of cycles, cyclically shifting the excavation parameters and vibration signals according to the defined window length, step size and cycle coefficient, generating multiple groups of window segments, and stacking the window matrices of the excavation parameters and vibration signals along the third dimension to form a three-dimensional tensor.

7. The shield machine tool wear prediction method according to claim 3, characterized in that: In step S24, the three-dimensional tensor singular value decomposition method decomposes the three-dimensional tensor into the product of three-order tensors based on the tube product of the tensor. The decomposition form is expressed as: ； In the formula, Represents a three-dimensional tensor; express orthogonal matrix of order; express orthogonal matrix of order; Represents a matrix consisting of non-negative diagonal elements in descending order A rectangular diagonal matrix with singular values ​​on the diagonal.

8. The shield machine tool wear prediction method according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31, collect the propulsion speed time series of each ring of the shield machine, perform wavelet transform on the propulsion speed time series of each ring, take the low-frequency approximation of the last layer of wavelet transform to form a low-frequency approximation sequence, and calculate the standard deviation of the low-frequency approximation sequence; the calculation process of the standard deviation is expressed as: ； In the formula, Indicates the shield machine The standard deviation of the low-frequency approximation sequence of the ring; Indicates the shield machine The first elements; Indicates the shield machine The average value of the low-frequency approximate sequence of the ring; N represents the shield machine The total number of elements in the low-frequency approximation sequence of the ring; Step S32, the standard deviations of all loops are combined to form a standard deviation sequence, and Fourier denoising is performed on the standard deviation sequence. The process of Fourier denoising is expressed as: ； ； ； In the formula, represents Fourier transform; represents the sequence after Fourier transformation; It means that the sequence after Fourier transformation is truncated, and all elements from the 1 / 3 length position to the end of the sequence are assigned 0; represents inverse Fourier transform; represents the sequence after inverse Fourier transformation; Step S33, the standard deviation after noise reduction is calculated by Get the final tool wear index .

9. The shield machine tool wear prediction method according to claim 1, characterized in that: The vibration signal uses a one-dimensional convolutional neural network to perform feature extraction; the excavation parameter uses a fully connected neural network to perform feature extraction.

10. A system for executing the shield machine tool wear prediction method according to any one of claims 1 to 9, characterized in that: include: Signal acquisition module, used to collect tunneling parameters and vibration signals during the tunneling process of the shield machine; The principal component extraction module is used to perform collaborative filtering on the tunneling parameters and vibration signals based on the singular value decomposition method, and to capture the principal component signals related to tool wear in the tunneling parameters and vibration signals; The tool wear degree data generation module is used to correct the traditional tool wear empirical formula based on the work log of the shield machine in the actual engineering scenario, and use the corrected tool wear formula to generate tool wear degree data; The mapping model building module is used to extract features from the main component signals of the vibration signal and the excavation parameter respectively, fuse the extracted vibration signal features and the excavation parameter features through matrix splicing, build a unified feature representation, and then use a fully connected neural network to establish a mapping model from the fusion features to the tool wear degree. The tool wear formula is used to generate a large amount of tool wear degree data under different excavation parameters and vibration signals, and train the mapping model; The prediction module is used to predict the tool wear of any shield machine tool. The fusion features of the vibration signal and the main component signal of the tunneling parameter are input into the mapping model to output the prediction result of the tool wear degree.

Images