Intelligent diagnosis method and system for cutter fault of shield tunneling machine
Through the AE sensor and EMD/ICA noise reduction processing technology combined with the multi-traditional excitation function superposition method of four-layer quantum neuron network, the problems of nonstability and low data transmission efficiency in the tool fault diagnosis of shield machine are solved, achieving higher fault detection accuracy and signal clarity.
Patent Information
- Application Number
- CN202411469044.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-21
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the fault diagnosis of the tool of the shield machine is unstable, resulting in low accuracy of fault judgment, especially due to the limited data volume and limited storage space, low transmission efficiency, and insufficient accuracy of the analysis results.
AE sensor is used to collect data, and noise reduction is performed through the excitation method of EMD and ICA, and the data is decomposed into an inherent modal function with meaningful instantaneous frequency, construct a virtual noise channel and combine it with the original data, and use ICA to separate useful signals. The superposition and additional momentum methods of multi-traditional excitation functions are used in a four-layer quantum neuron network to improve the convergence efficiency and diagnostic accuracy of the neuron network.
It improves the accuracy of the evaluation of tool fault diagnosis of shield machine and signal noise reduction, and improves the accuracy of fault detection and transmission efficiency under limited data conditions.
Smart Images

Figure CN119004041A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of tool fault diagnosis, and in particular to an intelligent diagnosis method and system for tool faults of shield machines. Background Art
[0002] With the continuous development and utilization of urban underground space, shield machines, as an important underground engineering construction equipment, are widely used in subways, tunnels and other projects. The shield machine cutter is an important component of the shield machine, and its working state directly affects the construction quality and efficiency. Therefore, it is necessary to perform quality inspection on the shield machine cutter. When fault detection is performed on the shield machine cutter, the collected data needs to be transmitted to the cloud platform to perform abnormality analysis on the collected data. Due to limited storage space, the amount of data transmitted is limited. When analyzing the limited amount of data, the accuracy of the analysis results is not high. Therefore, data is usually compressed to reduce data storage space and improve transmission efficiency, so that more data can be stored in a limited space, making the fault detection results of the shield machine cutter more accurate.
[0003] The existing means of tool fault identification mainly include artificial neural network methods, multi-sensor information fusion methods, fuzzy discrimination methods, etc. Quantum neural networks integrate quantum mechanics and neural network related knowledge. Compared with classical neural networks, they have the following advantages: exponential memory capacity and recall speed; high efficiency in acquiring new skills and analyzing data.
[0004] However, the artificial neural network method uses a single feature to calculate the neuron's activation function, which makes the network inherently fuzzy and unstable. Therefore, an intelligent fault diagnosis method is needed to reduce the instability of fault judgment and increase the accuracy of judgment. Summary of the invention
[0005] The purpose of the present invention is to provide a shield machine cutter fault intelligent diagnosis method and system to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solutions: an intelligent diagnosis system for tool failure of a shield machine, comprising a data acquisition module, a data processing module and a diagnosis module, wherein the data acquisition module comprises an AE sensor distributed on the shield machine, the data processing module comprises a preamplifier, a main amplifier, a data acquisition card and a computer, and the diagnosis module comprises an A / D converter and a host computer; The host computer diagnoses tool faults through a neural network analysis model, and the computer uses EMD and ICA induced polarization methods to perform noise reduction processing on data.
[0007] Preferably, a diagnostic method of a shield machine tool fault intelligent diagnostic system is characterized by comprising the following steps: S1: Use AE sensors to obtain shield machine cutter operation data, and use sensors deployed on the shield machine to obtain shield machine cutter operation data; S2: The AE sensor converts the AE signal into an electrical signal and uses an A / D converter to send the result to a computer, where the EMD and ICA methods are used to reduce noise on the data; The tool operation data is decomposed into several intrinsic mode functions with meaningful instantaneous frequencies. According to the correlation between each IMF and the IP excitation signal, a suitable intrinsic mode function is selected to be regarded as noise, and a virtual noise channel is constructed. The virtual noise channel and the original tool operation data are regarded as the input data of ICA. Finally, ICA is used to extract useful tool operation data from the artificially constructed mixed multidimensional data. S3: Construct a neural network, including input layer, output layer, and first and second hidden layers. Let the input layer vector be ; The output layer vector is ; The transfer function uses a Sigmod function, and the network output is: ; ; Where: is the connection weight from the input layer to the first hidden layer; is the connection weight from the first hidden layer to the second hidden layer; is the connection weight from the second hidden layer to the output layer; is the number of interval quanta; is the quantum interval; is the steepness factor; Assume the error energy function is: ; Where: P is the number of fault diagnosis input samples; is the expected output of the pth sample at the sth node of the network output layer; is the actual output of the pth sample at the sth node of the network output layer; The error energy function is used to obtain the three connection weights , ,and The partial derivative of : ; ; ; S4: Perform fault diagnosis on multiple independent tool operation data feature vectors based on the pre-trained neural network and output the fault diagnosis results.
[0008] Preferably, the shield machine tool operation data in S1 includes the tool rotation speed, cutting force and cutting temperature.
[0009] Preferably, S2 includes the following specific steps: S2.1: Independent component analysis, decomposing the mixed signal into several independent components through optimization algorithm; S2.2: Pseudo-noise channel construction: construct a virtual noise channel that is as similar as possible to the noise; S2.3: Establish a denoising method combining EMD and JCA, taking the virtual noise channel and the original observation signal as the ICA observation matrix, separating the useful signal and achieving the purpose of signal denoising.
[0010] Preferably, in order to improve the training efficiency and prevent the network from falling into a local optimal solution, the additional momentum and the adaptive learning speed method are integrated in S3 to improve the network convergence efficiency. The specific steps are as follows: S3.1: ; Where: is the increment of connection weight; S3.2: Improve the momentum of back propagation by the following method, ; ; Where: and is the connection weight before and after training; mc is the momentum factor; lr is the learning rate; The momentum factor and learning rate are updated as follows: ; ; ; ; in: and is the momentum factor; and is the reduction and increase of learning speed; em is the maximum error rate.
[0011] Preferably, in the neural network, the activation function of each layer is selected as a logarithmic function, the initial weight is obtained by a random function, the maximum error rate is 1.05, and the momentum factor constant is =0, =0.9, the initial value of learning speed is 0.5, the learning speed reduction rate is 0.8, the learning speed growth rate is 1.2, and the steepness factor sf=0.8.
[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. The shield machine tool fault intelligent diagnosis method and system adopts the superposition of multiple traditional excitation functions as the excitation function of the hidden quantum neurons in the four-layer quantum neural network, integrates the additional momentum and learning method, improves the convergence efficiency of the neural network, and improves the accuracy of the judgment.
[0013] 2. The shield machine tool fault intelligent diagnosis method and system, in the process of tool fault diagnosis, uses the EMD method to decompose the observation signal into multiple IMFs, and then according to the prior information of the excitation signal of the induced polarization transmitter, takes the mutual correlation coefficient between each IMF and the excitation signal as an indicator, separates the IMF less than a certain threshold to construct a virtual noise channel, and uses the virtual noise channel and the original observation signal as the ICA observation matrix, so that the one-dimensional observation signal is expanded to multi-dimensional observation, and finally applies the ICA technology to separate the useful signal to achieve the purpose of signal noise reduction. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0015] Figure 1 It is a flow chart of the intelligent diagnosis method of tool fault of the present invention; Figure 2 This is a data noise reduction process flow chart of the present invention; Figure 3 The hardware structure diagram of the fault diagnosis system of the present invention; Figure 4 This is a diagram of the neural network structure of the present invention. DETAILED DESCRIPTION
[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0017] See also Figure 1-Figure 4 , the present invention provides a technical solution: A shield machine tool fault intelligent diagnosis system includes a data acquisition module, a data processing module and a diagnosis module, the data acquisition module includes an AE sensor distributed on the shield machine, the data processing module includes a preamplifier, a main amplifier, a data acquisition card and a computer, and the diagnosis module includes an A / D converter and a host computer; The host computer diagnoses tool faults through a neural network analysis model, and the computer uses EMD and ICA methods to reduce noise on the data.
[0018] A diagnostic method for a shield machine tool fault intelligent diagnostic system comprises the following steps: S1: Use AE sensors to obtain shield machine tool operation data. Use sensors deployed on the shield machine to obtain shield machine tool operation data. The shield machine tool operation data includes tool speed, cutting force, and cutting temperature.
[0019] S2: The AE sensor converts the AE signal into an electrical signal and uses an A / D converter to send the result to a computer, where the EMD and ICA methods are used to reduce noise on the data; The specific steps include: S2.1: Independent component analysis, decomposing the mixed signal into several independent components through optimization algorithm; Independent component analysis is a multidimensional signal processing method that establishes an objective function according to the principle of statistical independence and decomposes a mixed signal into several independent components through an optimization algorithm.
[0020] Assuming the observed Dimensional signal ; is composed of n independent signals After linear aliasing of an unknown system, we get: ; Rewritten into matrix form: ; A is the unknown mixing coefficient The goal of ICA is to find a separation matrix W from the data matrix X, so that the output matrix Y=WX is the optimal estimate of the original signal matrix s. In practical applications, the tool operation data signal model is often affected by noise, which is extended to the following noise model: ; Where: u is the noise vector. Limited by the sensor array, most of the data signals obtained are tool operation data signals. Considering the addition of m types of noise, the tool operation data signal is expressed as: ; in: and are the useful signal s and the noise respectively The weight of the tool operation data is used to process such operation data signals. The virtual noise introduced is ; The noise-added model can be rewritten as the standard ICA model without an explicit representation of the additive noise:
[0021]
[0022] Then ICA can estimate the separation matrix by finding the inverse matrix of matrix A, so that is the optimal estimate of the source signal. Since the virtual noise u is artificially introduced, it can be obtained from the output signal The useful signal s is separated from the signal to achieve signal noise reduction.
[0023] S2.2: Pseudo-noise channel construction: construct a virtual noise channel that is as similar as possible to the noise; From the above formula, we can know that EMD can decompose complex signals into linear superposition of multiple intrinsic mode functions IMFs and residual terms with approximate single frequency components: ; in: To observe the signal, is the intrinsic mode component arranged from high to low frequency, res is the remainder, and m is the number of IMFs set. Every time an IMF is separated by EMD, its time spectrum is calculated, where IMF is orthogonal in the frequency domain, and each IMF represents a component of different frequencies in the observed signal.
[0024] The excitation signal s of the transmitting device is preset artificially, so the approximate waveform, frequency and other prior information of the useful signal in the collected mixed signal are known, and the correlation coefficient between each IMF and the excitation signal s can be selected. As a judgment indicator: ; in: Represents the covariance of a and b. Select n intrinsic modal components, perform linear superposition, and construct virtual noise: ; in: The correlation coefficient is less than the set threshold In practical applications, known information can be added to the independent component analysis process to improve the separation effect. A standard signal can be used to pre-observe a hybrid system, and then the system output can be analyzed to have a general understanding of the system noise, which can be used as a certain dimension observation in the virtual noise, which will also improve the effect of independent component separation.
[0025] S2.3: Establish a denoising method combining EMD and JCA, taking the virtual noise channel and the original observation signal as the ICA observation matrix, separating the useful signal and achieving the purpose of signal denoising.
[0026] Firstly, the observed signal is decomposed into multiple IMFs using the EMD method. Then, based on the prior information of the excitation signal of the induced polarization transmitter (the frequency and waveform are known), the IMFs smaller than a certain threshold are separated with the mutual correlation coefficient between each IMF and the excitation signal as the indicator to construct a virtual noise channel. The virtual noise channel and the original observed signal are used as the ICA observation matrix, thus expanding the one-dimensional observation signal to multi-dimensional observation. Finally, the ICA technology is applied to separate the useful signal to achieve the purpose of signal denoising.
[0027] The tool operation data is decomposed into several intrinsic mode functions with meaningful instantaneous frequencies. According to the correlation between each IMF and the IP excitation signal, a suitable intrinsic mode function is selected to be regarded as noise, and a virtual noise channel is constructed. The virtual noise channel and the original tool operation data are regarded as the input data of ICA. Finally, ICA is used to extract useful tool operation data from the artificially constructed mixed multidimensional data. S3: Construct a neural network, including input layer, output layer, and first and second hidden layers. Let the input layer vector be ; The output layer vector is ; The transfer function uses a Sigmod function, and the network output is: ; ; Where: It represents summation, which performs sum operation on a series of values.
[0028] is the connection weight from the input layer to the first hidden layer; is the connection weight from the first hidden layer to the second hidden layer; is the connection weight from the second hidden layer to the output layer; is the number of interval quanta; is the quantum interval; is the steepness factor; Assume the error energy function is: ; Where: P is the number of fault diagnosis input samples; is the expected output of the pth sample at the sth node of the network output layer; is the actual output of the pth sample at the sth node of the network output layer; The error energy function is used to obtain the three connection weights , ,and The partial derivative of : ; ; ; It means to find the partial derivative of the series value e, is the standard deviation, It means to find the partial derivative of the series value w.
[0029] In order to improve the training efficiency and prevent the network from falling into the local optimal solution, the additional momentum and adaptive learning speed methods are integrated to improve the network convergence efficiency. The specific steps are as follows: The specific steps are as follows: S3.1: ; Where: is the increment of connection weight; S3.2: Improve the momentum of back propagation by the following method, ; ; Where: and is the connection weight before and after training; mc is the momentum factor; lr is the learning rate; The momentum factor and learning rate are updated as follows: ; ; ; ; in: and is the momentum factor; and is the reduction and increase of learning speed; em is the maximum error rate.
[0030] In the neural network, the activation function of each layer is selected as a logarithmic function, and the initial weight is obtained by a random function. The maximum error rate is 1.05, and the momentum factor constant =0, =0.9, the initial value of learning speed is 0.5, the learning speed reduction rate is 0.8, the learning speed growth rate is 1.2, and the steepness factor sf=0.8.
[0031] S4: Extract and separate the features of the fault data after noise reduction based on the independent component analysis method to obtain multiple independent feature vectors; S5: Perform fault diagnosis on multiple independent tool operation data feature vectors based on the pre-trained neural network and output the fault diagnosis results.
[0032] The specific steps of S4 and S5 have been disclosed in the patent document with publication number CN114595779A, so they will not be described in detail here.
[0033] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0034] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An intelligent diagnosis system for shield machine tool faults, comprising a data acquisition module, a data processing module and a diagnosis module, characterized in that: The data acquisition module includes AE sensors distributed on the shield machine, the data processing module includes a preamplifier, a main amplifier, a data acquisition card and a computer, and the diagnosis module includes an A / D converter and a host computer; The host computer diagnoses tool faults through a neural network analysis model, and the computer uses EMD and ICA induced polarization methods to perform noise reduction processing on data.
2. The diagnostic method of a shield machine tool fault intelligent diagnostic system according to claim 1 is characterized in that: The following steps are involved: S1: Use AE sensors to obtain shield machine cutter operation data, and use sensors deployed on the shield machine to obtain shield machine cutter operation data; S2: The AE sensor converts the AE signal into an electrical signal and uses an A / D converter to send the result to a computer, where the EMD and ICA methods are used to reduce noise on the data; The tool operation data is decomposed into several intrinsic mode functions with meaningful instantaneous frequencies. Then, according to the correlation between each IMF and the IP excitation signal, the intrinsic mode function is selected and regarded as noise to construct a virtual noise channel. The virtual noise channel and the original tool operation data are regarded as the input data of ICA. Finally, ICA is used to extract useful tool operation data from the artificially constructed mixed multidimensional data. S3: Construct a neural network, including input layer, output layer, and first and second hidden layers. Let the input layer vector be ; The output layer vector is ; The transfer function uses a Sigmod function, and the network output is: ; ; Where: is the connection weight from the input layer to the first hidden layer; is the connection weight from the first hidden layer to the second hidden layer; is the connection weight from the second hidden layer to the output layer; is the number of interval quanta; is the quantum interval; is the steepness factor; Assume the error energy function is: ; Where: P is the number of fault diagnosis input samples; is the expected output of the pth sample at the sth node of the network output layer; is the actual output of the pth sample at the sth node of the network output layer; The error energy function is used to obtain the three connection weights , ,and The partial derivative of : ; ; ; S4: Perform fault diagnosis on multiple independent tool operation data feature vectors based on the pre-trained neural network and output the fault diagnosis results.
3. The intelligent diagnosis method for shield machine tool failure according to claim 2 is characterized by: The shield machine cutter operation data in S1 includes the cutter rotation speed, cutting force and cutting temperature.
4. The intelligent diagnosis method for shield machine tool failure according to claim 2 is characterized by: The S2 includes the following specific steps: S2.1: Independent component analysis, decomposing the mixed signal into several independent components through optimization algorithm; S2.2: Pseudo-noise channel construction, construct a virtual noise channel that is identical to the noise; S2.3: Establish a denoising method combining EMD and JCA, taking the virtual noise channel and the original observation signal as the ICA observation matrix, separating the useful signal and achieving the purpose of signal denoising.
5. The intelligent diagnosis method for shield machine tool failure according to claim 2 is characterized by: In order to improve the training efficiency and prevent the network from falling into the local optimal solution, the additional momentum and adaptive learning speed method are integrated in S3 to improve the network convergence efficiency. The specific steps are as follows: S3.1: ; Where: is the increment of connection weight; S3.2: Improve the momentum of back propagation by the following method, ; ; Where: and is the connection weight before and after training; mc is the momentum factor; lr is the learning rate; The momentum factor and learning rate are updated as follows: ; ; ; ; in: and is the momentum factor; and is the reduction and increase of learning speed; em is the maximum error rate.
6. The intelligent diagnosis method for shield machine tool failure according to claim 2 is characterized by: In the neural network, the activation function of each layer is selected as a logarithmic function, and the initial weight is obtained by a random function. The maximum error rate is 1.05, and the momentum factor constant =0, =0.9, the initial value of learning speed is 0.5, the learning speed reduction rate is 0.8, the learning speed growth rate is 1.2, and the steepness factor sf=0.8.
Citation Information
Patent Citations
Shield tunneling machine cutter fault analysis and diagnosis method
CN114595779A
Vibration signal detection and fault diagnosis system in shield tunneling process
CN104457973A
Vehicle-mounted-ground auxiliary inverter remote fault diagnosis system
CN105572493A
Cited By
Shield tunneling machine operation and maintenance management system based on data analysis
CN119204592A
