Bearing parameter decoupling method and system for multi-structure SAW

Through the bearing parameter decoupling method of multi-structure SAW, the improved network model and EHML framework are used to solve the problems of complex and inapplicable harsh environments in the traditional method, and the high-precision decoupling and classification of bearing parameters are achieved, improving the improvement of the improvement of the improvement of the bearing parameters. The accuracy and noise immunity of signal decoupling and classification are obtained.

CN118981670APending Publication Date: 2024-11-19EAST CHINA JIAOTONG UNIVERSITY
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411471283.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-22
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

When traditional methods obtain bearing temperature, strain, humidity and other parameters, they require multiple sensors and complex lead interfaces, and are not suitable for harsh environments, making it difficult to achieve real-time monitoring and decoupling.

Method used

The bearing parameter decoupling method of multi-structure SAW is adopted to preprocess and encode by obtaining mixed time signals, and the signal reconstruction and decoupling is performed using Markov transition field and improved dual-channel Inception-attention residual network model. Combining the decoupled EHML framework and multi-scale feature adaptive integrated domain network, high-precision decoupling and classification of parameters are achieved.

Benefits of technology

Significantly improves the accuracy of signal decoupling and classification, enhances noise immunity and robustness, is suitable for complex and multidimensional data, and is suitable for a wider range of complex sensor signal decoupling tasks, especially in harsh environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981670A_ABST
    Figure CN118981670A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-structure SAW bearing parameter decoupling method and system, and the method comprises the steps: obtaining at least two bearing parameter mixed time signals, and carrying out the preprocessing of the time signals, and obtaining a target time signal; encoding the target time signal to obtain a two-dimensional signal diagram, and reconstructing the two-dimensional signal diagram by using a Markov transition field to obtain a reconstructed mixed signal; and according to a preset improved dual-channel Inception-attention residual network model, the reconstructed mixed signal is decoupled, and a classification result is obtained. The problems that in the prior art, the testing size is large, and real-time and accurate testing of multiple parameters cannot be achieved can be solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of measurement and sensing, and in particular relates to a bearing parameter decoupling method and system of a multi-structure SAW. Background Art

[0002] In rotating parts of industrial machinery and equipment, the temperature of the bearings rises sharply due to friction during rotation, thus affecting the normal operation of the device. On the other hand, long-term loads cause the structure of the bearings to deform, and in severe cases, failures or even fractures may occur. Therefore, real-time monitoring of bearing temperature, strain and other parameters is of great significance for the structural design of the bearings and the assessment of their operating health.

[0003] By obtaining multiple parameters such as temperature, strain, and humidity during the rotation of the bearing, the working condition of the bearing can be understood in time, so that timely measures can be taken to prevent it from being damaged. For example, in "A device and method for real-time in-situ measurement of temperature and mechanical parameters on the same side", a temperature / vibration coplanar integrated wireless surface acoustic wave sensing module is proposed, which can realize the detection of vibration signals in variable temperature environments. The traditional method can only use temperature sensors, strain sensors, humidity sensors, and vibration sensors to obtain each parameter separately, and there are many interfaces and complex leads, which are not suitable for harsh environments. Summary of the invention

[0004] The present invention provides a bearing parameter decoupling method and system of a multi-structure SAW, which is used to solve the technical problem that temperature sensors, strain sensors, humidity sensors and vibration sensors are used to obtain various parameters respectively, and there are many interfaces, complex leads, and they are not suitable for use in harsh environments.

[0005] In a first aspect, the present invention provides a bearing parameter decoupling method of a multi-structure SAW, comprising: Acquire a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; Encoding the target time signal to obtain a two-dimensional signal graph, and reconstructing the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The reconstructed mixed signal is decoupled according to the preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a decoupled EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

[0006] In a second aspect, the present invention provides a bearing parameter decoupling system of a multi-structure SAW, comprising: A preprocessing module, configured to obtain a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; A reconstruction module is configured to encode the target time signal to obtain a two-dimensional signal graph, and reconstruct the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The deconstruction module is configured to decouple the reconstructed mixed signal according to a preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the preprocessing generalization efficient signal decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a preprocessing generalized efficient signal decoupling EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

[0007] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the bearing parameter decoupling method of the multi-structure SAW of any embodiment of the present invention.

[0008] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor executes the steps of the bearing parameter decoupling method of the multi-structure SAW of any embodiment of the present invention.

[0009] The bearing parameter decoupling method and system of the multi-structure SAW of the present application proposes a dual-channel Inception-attention residual network model, and improves it using the decoupling EHML framework and the multi-scale feature adaptive integrated domain network. The improved model shows significant advantages over the existing technology when decoupling multi-dimensional mixed signals such as temperature, humidity, pressure, and vibration measured by sensors. First, the EHML framework can decouple multiple physical parameters with high precision, reduce mutual interference between different signals, and ensure the independence and clarity of each parameter. In the complex background of multiple mixed signals, traditional technologies often find it difficult to capture global and local features at the same time. The model can effectively extract global and subtle features from different levels through the improved Inception module and the multi-scale feature adaptive integrated domain network, greatly improving the accuracy of signal decoupling and classification.

[0010] By introducing the attention mechanism, the model further optimizes the focus on key features, can suppress noise interference in complex, multi-dimensional data, highlight important feature information, and performs well in the decoupling of vibration signals. In contrast, traditional methods are more sensitive to noise and abnormal data, while the integrated learning mechanism of this model significantly enhances its noise resistance and robustness, and can maintain efficient and stable decoupling and classification effects in the presence of data noise or missing data.

[0011] In addition, the domain adaptation strategy of the model greatly improves its generalization ability in different data distributions or cross-domain tasks, and can adaptively handle signal changes in different working conditions and environments, ensuring consistent performance in a variety of application scenarios. Traditional technologies often degrade in performance when facing cross-domain data due to distribution differences, and this model significantly improves this problem through adaptive learning.

[0012] Finally, the adaptability and scalability of the model not only enable it to process multi-dimensional signals from existing sensors, but also can be easily extended to other complex multi-task decoupling scenarios, such as processing dynamic signal changes in real-time monitoring systems. Overall, the model surpasses existing technologies in terms of decoupling accuracy, feature extraction capability, noise resistance, cross-domain generalization capability, and adaptability, and is suitable for a wider range of complex sensor signal decoupling tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A flow chart of a bearing parameter decoupling method of a multi-structure SAW provided by an embodiment of the present invention; Figure 2 A structural diagram of a symmetrically distributed multi-structure IDT surface acoustic wave sensor according to an embodiment of the present invention; Figure 3 A schematic diagram of a dual-channel Inception-attention residual network model according to an embodiment of the present invention; Figure 4 A structural block diagram of a bearing parameter decoupling system of a multi-structure SAW provided by an embodiment of the present invention; Figure 5 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0016] See also Figure 1 , which shows a flow chart of a bearing parameter decoupling method of a multi-structure SAW of the present application.

[0017] like Figure 1 As shown, the bearing parameter decoupling method of the multi-structure SAW specifically includes the following steps: Step S101, obtaining a time signal of a mixture of at least two bearing parameters, and preprocessing the time signal to obtain a target time signal.

[0018] In this step, a time signal of a mixture of bearing temperature parameters, bearing strain parameters, bearing humidity parameters and bearing vibration parameters is obtained. The time signal is filtered, denoised and normalized to obtain a target time signal, wherein the expression of filtering and denoising is: , In the formula, is the Butterworth low-pass filter transfer function, is the sampling frequency of the filter, is the order of the filter, is the cut-off frequency; The expression of the standardization process is: , In the formula, For mixed signals, is the mean value of the signal, is the standard deviation of the signal, is the normalized signal.

[0019] In a specific embodiment, Figure 2 As shown, a symmetrical distributed multi-structure IDT surface acoustic wave sensor is designed, wherein the interdigital transducers of the designed sensor are divided into two structures, fan-shaped and rectangular, and are distributed in a symmetrical form on the same substrate 105, specifically including: A symmetrically distributed multi-structure IDT surface acoustic wave sensor is designed, wherein the multi-structure IDT surface acoustic wave sensor includes a temperature sensing unit 101, a strain sensing unit 102, a humidity sensing unit 103, and a vibration sensing unit 104, and the temperature sensing unit 101, the strain sensing unit 102, the humidity sensing unit 103, and the vibration sensing unit 104 are symmetrically distributed on the same substrate; Among them, the IDT of the temperature sensing unit 101 and the IDT of the humidity sensing unit 103 are both fan-shaped structures, the IDT of the strain sensing unit 102 and the IDT of the vibration sensing unit 104 are both rectangular structures, and the electrode width, spacing, and logarithm are all different, so that the frequency of each unit is different.

[0020] Among them, the substrates of the four sensors are made of lanthanum gallium silicate crystal material, and the propagation speed of sound is , , , , set the frequencies of the four sensors to , , , ,Depend on So, their wavelengths are , , , ; The logarithms of the interdigital transducers of the four sensors are , , , The interdigital electrodes are Pt electrodes with a pore size of , the width of the interdigital electrode is , the interdigital spacing is ,in The logarithm of the reflector is , , , .

[0021] Step S102: Encode the target time signal to obtain a two-dimensional signal graph, and reconstruct the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal.

[0022] In this step, the filtered, denoised and standardized data ( is the time step) to encode and obtain a two-dimensional signal graph; Will The value range is divided into quantile bins and the i-th value Map to the corresponding box , and then calculate the migration probability between quantile boxes in a Markov chain manner along the time step, and arrange each migration probability in time order to obtain the Markov matrix, which is expressed as: , In the formula, is the value at the first moment, is the value at the second moment, is the transition probability matrix, is the transition probability from state 1 to state 1, is the probability of transitioning from state 1 to state L, is the probability of state L transferring to state 1, is the probability of state L transferring to state L, For the current region Data points in In the area The probability of an element in following, is the time step The data on is the time step The data on.

[0023] Step S103, decoupling the reconstructed mixed signal according to a preset improved dual-channel Inception-attention residual network model to obtain a classification result.

[0024] In this step, refer to Figure 3 , the reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, and the positive values ​​are kept unchanged. The secondary path is the identity mapping of the input signal. The main path and the secondary path are added to obtain the output of the residual channel, and the decoupling EHML framework is used for decoupling, specifically: , In the formula, is the system residual unit output, is what the residual unit has learned. Output for the network; The decoupled EHML framework is based on the integration of hybrid graph neural network (GNN) and tissue-scale data, and the physical constraint data enhancement (DA) technology is combined with a set of preprocessing and custom algorithms. The expression of the decoupled EHML framework is: , , , , In the formula, For the Layers have relationships Node The encoding latent vector of is the ReLU activation function, is the dropout layer, is the fusion function that fuses each neighbor’s information, and Neighbor nodes and nodes The vector representation of For the The relationship vector of the layer nodes, is the mapping function, and are the feature vector and relationship vector of the node, For Node The embedding For Node About Relationships Neighbors, The parameter matrix representing a particular relationship, A parameter matrix representing a particular relationship.

[0025] The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a decoupled EHML framework for decoupling. Convolution expression: , In the formula, For output, is the Relu function, x is the input, is the bias term; Max Pooling: , Where: For the Layer The first feature matrix, is the width of the pooling region, For the Pooling kernels, Output result for pooling; Obtain the first output signal of the residual channel and the second output signal of the Inception channel, and use the convolution attention module to integrate the first output signal and the second output signal. The formula is: , , , In the formula, is the integrated signal, is the output signal of the Inception channel, and They are channel and spatial attention module operations respectively. is the weighted multiplication operator symbol for the feature map, express After the output of the channel attention module, express Output from the spatial attention module; The global average pooling method is used to reduce the dimension of the integrated signal data. The formula is: , In the formula, For the Layer The first feature matrix, is the width of the pooling region, It is Pooling kernels, is the output result of pooling; According to the output signal of the pooling layer and the fully connected layer, the output signal of the fully connected layer is obtained; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result. The expression for classification using the pre-built multi-scale feature adaptive integrated domain network is: , , In the formula, For the quantization of signal data, is the balance coefficient, and is the signal distribution of input and output, and They are and The coefficient matrix of and is the unreachable posterior probability of input and output, represents the reproducing kernel Hilbert space, is the function that maps the original features to RKHS, and are the input and output feature sets, and is the balancing weight for each class in the input and output, and are the learning rates for input and output, and is the feature set of the corresponding category, and is the learning rate of the corresponding category, There are four types of signals, c=1 is temperature signal, c=2 is strain signal, c=3 is humidity signal, c=4 is vibration signal, is the source domain label space, is the input signal in the source domain, is the label space of the target domain, is the output signal of the target domain.

[0026] In summary, the method of the present application uses a surface acoustic wave sensor to measure the data during the movement of a rotating bearing. Among them, the surface acoustic wave sensor has many advantages such as small size, low cost, simple process, good robustness, etc. It can be well fitted on the surface of the bearing to realize wireless passive measurement, so a symmetrically distributed multi-structure IDT surface acoustic wave sensor is designed, in which the finger transducer of the designed sensor is divided into two structures, fan-shaped and rectangular, and is distributed in a symmetrical form on the same substrate, so that the four sensing units can have a frequency interval, which is convenient for subsequent separation; the sensor and the microstrip antenna are installed on the surface of the inner ring of the bearing, and the sensor is used to measure the changes in the four parameters of temperature, strain, humidity, and vibration during the operation of the bearing, and a time signal of a mixture of the four parameters is obtained; in the process of data collection through the sensor, it will definitely be affected by noise to varying degrees. In order to minimize the impact of noise, the measured data is first filtered and denoised using a Butterworth low-pass filter to make the data distribution smoother, and then the data is standardized using the Z-Score standardization method to reduce the impact of feature data values ​​on the recognition effect and further improve the recognition accuracy; then the signal data is reconstructed using the Markov transition field; the reconstructed signal is decoupled according to the preset improved dual-channel Inception-attention residual network model. The network does not require manual guidance for feature extraction and has good adaptability. It can identify and classify signal features of different signal types, and the dual-channel superposition method can more accurately extract signal features and obtain clear results.

[0027] See also Figure 4, which shows a structural block diagram of a multi-structure SAW bearing parameter decoupling system of the present application.

[0028] like Figure 4 As shown, the bearing parameter decoupling system 200 includes a preprocessing module 210 , a reconstruction module 220 and a deconstruction module 230 .

[0029] The preprocessing module 210 is configured to obtain a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; A reconstruction module 220 is configured to encode the target time signal to obtain a two-dimensional signal graph, and reconstruct the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The deconstruction module 230 is configured to decouple the reconstructed mixed signal according to a preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the preprocessing generalization efficient signal decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a preprocessing generalized efficient signal decoupling EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

[0030] It should be understood that Figure 4 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 4 The modules in it will not be described in detail here.

[0031] In some other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor executes the bearing parameter decoupling method of the multi-structure SAW in any of the above method embodiments; As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows: Acquire a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; Encoding the target time signal to obtain a two-dimensional signal graph, and reconstructing the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The reconstructed mixed signal is decoupled according to the preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a decoupled EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

[0032] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the multi-structure SAW bearing parameter decoupling system, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the multi-structure SAW bearing parameter decoupling system via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0033] Figure 5 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 5 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 5 In the example, the connection through the bus is taken as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the bearing parameter decoupling method of the multi-structure SAW of the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to the user setting and function control of the bearing parameter decoupling system of the multi-structure SAW. The output device 340 may include a display device such as a display screen.

[0034] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0035] As an implementation mode, the electronic device is applied to a bearing parameter decoupling system of a multi-structure SAW, and is used for a client, and includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: Acquire a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; Encoding the target time signal to obtain a two-dimensional signal graph, and reconstructing the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The reconstructed mixed signal is decoupled according to the preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a decoupled EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

[0036] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A bearing parameter decoupling method for multi-structure SAW, characterized in that: include: Acquire a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; Encoding the target time signal to obtain a two-dimensional signal graph, and reconstructing the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The reconstructed mixed signal is decoupled according to the preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a decoupled EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

2. A bearing parameter decoupling method for multi-structure SAW according to claim 1, characterized in that: The step of obtaining a time signal of a mixture of at least two bearing parameters comprises: Obtain a mixed time signal of bearing temperature parameters, bearing strain parameters, bearing humidity parameters and bearing vibration parameters.

3. A bearing parameter decoupling method for multi-structure SAW according to claim 2, characterized in that: Before obtaining the time signal of the mixture of at least two bearing parameters, the method further includes: Design a symmetrically distributed multi-structure IDT surface acoustic wave sensor, wherein the multi-structure IDT surface acoustic wave sensor includes a temperature sensing unit, a strain sensing unit, a humidity sensing unit, and a vibration sensing unit, and the temperature sensing unit, the strain sensing unit, the humidity sensing unit, and the vibration sensing unit are symmetrically distributed on the same substrate; Among them, the interdigital transducers of the temperature sensing unit and the interdigital transducers of the humidity sensing unit are both fan-shaped structures, the interdigital transducers of the strain sensing unit and the interdigital transducers of the vibration sensing unit are both rectangular structures, and the electrode width, spacing, and logarithm are all different, so that the frequency of each unit is different.

4. A method for decoupling bearing parameters of a multi-structure SAW according to claim 1, characterized in that: The preprocessing of the time signal to obtain a target time signal comprises: The time signal is subjected to filtering, denoising and standardization processing to obtain a target time signal, wherein the expression of the filtering and denoising processing is: , In the formula, is the Butterworth low-pass filter transfer function, is the sampling frequency of the filter, is the order of the filter, is the cut-off frequency; The expression of the standardization process is: , In the formula, For mixed signals, is the mean of the signal, is the standard deviation of the signal, is the normalized signal.

5. The method for decoupling bearing parameters of a multi-structure SAW according to claim 1, characterized in that: The reconstructing the two-dimensional signal graph by using the Markov transition field to obtain the reconstructed mixed signal comprises: Will The value range is divided into quantile bins and the i-th value Map to the corresponding box , and then calculate the migration probability between quantile boxes in a Markov chain manner along the time step, and arrange each migration probability in time order to obtain the Markov matrix, which is expressed as: , In the formula, is the value at the first moment, is the value at the second moment, is the transition probability matrix, is the transition probability from state 1 to state 1, is the probability of transitioning from state 1 to state L, is the probability of state L transferring to state 1, is the probability of state L transferring to state L, For the current region Data points in In the area The probability of an element in following, is the time step The data on is the time step The data on.

6. A method for decoupling bearing parameters of a multi-structure SAW according to claim 1, characterized in that: The expression of the decoupled EHML framework is: , , , , In the formula, For the Layers have relationships Node The encoding latent vector of is the ReLU activation function, is the dropout layer, is the fusion function that fuses each neighbor’s information, and Neighbor nodes and nodes The vector representation of For the The relationship vector of the layer nodes, is the mapping function, and are the feature vector and relationship vector of the node, For Node Embedded, For Node About Relationships Neighbors, The parameter matrix representing a particular relationship, A parameter matrix representing a particular relationship.

7. A method for decoupling bearing parameters of a multi-structure SAW according to claim 1, characterized in that: in, The expression for classification using the pre-built multi-scale feature adaptive integrated domain network is: , , , In the formula, For the quantization of signal data, is the balance coefficient, and is the signal distribution of input and output, and They are and The coefficient matrix of and is the unreachable posterior probability of input and output, represents the reproducing kernel Hilbert space, is the function that maps the original features to RKHS, and are the input and output feature sets, and is the balancing weight for each class in the input and output, and are the learning rates for input and output, and is the feature set of the corresponding category, and is the learning rate of the corresponding category, There are four types of signals, c=1 is temperature signal, c=2 is strain signal, c=3 is humidity signal, c=4 is vibration signal, is the source domain label space, is the input signal in the source domain, is the label space of the target domain, is the output signal of the target domain.

8. A multi-structure SAW bearing parameter decoupling system, characterized in that: include: A preprocessing module, configured to obtain a time signal of a mixture of at least two bearing parameters, and preprocess the time signal to obtain a target time signal; A reconstruction module is configured to encode the target time signal to obtain a two-dimensional signal graph, and reconstruct the two-dimensional signal graph using a Markov transition field to obtain a reconstructed mixed signal; The deconstruction module is configured to decouple the reconstructed mixed signal according to a preset improved dual-channel Inception-attention residual network model to obtain a classification result, wherein the decoupling process is specifically as follows: The reconstructed mixed signal is input from the residual channel and the Inception channel respectively. The residual channel is divided into two paths. The main path uses the convolution layer to perform preliminary feature extraction on the reconstructed mixed signal, and then the gradient of each layer is normalized to a fixed value according to the batch normalization module, and the negative values ​​in the convolution result are removed according to the activation function ReLU, while the positive values ​​remain unchanged. The secondary path is the identity mapping of the input signal. The output of the residual channel is obtained by adding the main path and the secondary path, and the preprocessing generalization efficient signal decoupling EHML framework is used for decoupling; The Inception channel uses a convolutional layer to initially extract features from the reconstructed mixed signal, performs a maximum pooling operation, extracts signal features based on the Inception module, and finally uses a preprocessing generalized efficient signal decoupling EHML framework for decoupling. Obtain a first output signal of the residual channel and a second output signal of the Inception channel, and integrate the first output signal and the second output signal using a convolutional attention module; The global average pooling method is used to reduce the dimension of the integrated signal data, and the output signal of the fully connected layer is obtained based on the output signal of the pooling layer and the fully connected layer; According to the output signal of the fully connected layer, the pre-built multi-scale feature adaptive integrated domain network is used for classification to obtain the classification result.

9. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Finger vein recognition method based on residual attention mechanism

    CN112949570A

  • Equipment fault diagnosis method based on improved end-to-end ResNet-BiLSTM dual-channel model

    CN115392333A

  • Mechanical equipment fault diagnosis method based on parallel network and transfer learning

    CN115758212A

  • Gearbox variable speed composite fault diagnosis method

    CN117309377A

  • Rolling bearing domain generalization fault diagnosis method based on decoupling characterization

    CN118171096A