Multi-source signal processing method fusing eigenmode functions and point-symmetry modes

By integrating intrinsic mode functions and point-symmetric modes into a multi-source signal processing method, a multi-source signal SDP image is generated and the MobileViT-SCConv model is used. This solves the limitation of a single data source in rolling bearing fault diagnosis and achieves more efficient fault identification and diagnosis.

CN119106320BActive Publication Date: 2026-11-10HUNAN UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411101029.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2026-11-10
Estimated Expiration
2044-08-12

AI Technical Summary

Technical Problem

In the diagnosis of rolling bearing faults under complex working conditions, existing technologies cannot fully reflect complex fault modes with a single data source, leading to missed diagnoses and misdiagnoses. Furthermore, multi-source signal processing cannot take into account both global and local characteristics.

Method used

A multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes is adopted to generate a fused multi-source signal SDP image. The MobileViT-SCConv classification model is used for fault diagnosis. Through the normalization of multi-source data, empirical mode decomposition and feature extraction, an attention-based classification model is constructed.

Benefits of technology

It improves the accuracy and robustness of fault diagnosis, reduces the impact of a single data source, dynamically adjusts attention weights, improves the extraction of global and local features, and enhances classification accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119106320B_ABST
    Figure CN119106320B_ABST
Patent Text Reader

Abstract

The application discloses a multi-source signal processing method fusing an intrinsic modal function and a point symmetry mode, and comprises the following steps: acquiring multi-source signal data of a rolling bearing, and generating a fused multi-source signal SDP image by using a fusion intrinsic modal function multi-source signal processing method based on the point symmetry mode according to the multi-source signal data; classifying and predicting the multi-source signal SDP image by using a MobileViT-SCConv classification model; randomly dividing the multi-source data fusion image into a training set, a verification set and a test set, updating the model parameters of the training set by using the MobileViT-SCConv classification model, iteratively training according to the result on the verification set, and then evaluating the classification prediction effect by using the test set, so as to realize fault diagnosis of the rolling bearing. The fusion intrinsic modal function is added, and the phenomenon that the multi-source signal processing is difficult to simultaneously consider global features and local features is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for rolling bearings, a key component in rotating machinery, and in particular discloses a multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes. Background Technology

[0002] Rolling bearings are critical components widely used in mechanical equipment, and their operating condition directly affects the performance and reliability of the entire mechanical system. However, due to complex working environments and long-term operation, rolling bearings often experience various failures, such as spalling, wear, and fatigue cracks. These failures not only reduce equipment efficiency but can also lead to serious safety hazards.

[0003] Traditional methods for diagnosing rotating machinery faults primarily reveal the changes in amplitude and frequency over time in monitoring signals by analyzing their temporal or frequency characteristics. Traditional analytical methods include wavelet transform, variational mode decomposition, and empirical mode decomposition. While these methods can extract fault features to some extent, they still face challenges in practical applications. For example, under varying speeds and loads, monitoring signals in industrial settings often exhibit non-stationarity and are susceptible to noise interference, making feature extraction difficult. In contrast, visual information in equipment status signals can reflect the global characteristics of the fault state, thus mitigating the impact of changing operating conditions. Therefore, utilizing visual information to extract fault features is an appropriate method for more effectively identifying and diagnosing mechanical faults. However, most researchers use time-frequency images based on Cartesian coordinates, which lack an intuitive mapping relationship with actual fault features.

[0004] The Symmetrized Dot Pattern (SDP) method can directly transform the original signal into a mirror-symmetric snowflake image, accurately reflecting signal changes through visual differences in the image. However, SDP-based rolling bearing fault diagnosis methods primarily rely on a single data source, such as vibration or temperature signals. A single data source often fails to comprehensively reflect the complex fault modes of rolling bearings, limiting the accuracy and robustness of fault diagnosis. Furthermore, accurate fault diagnosis requires feature extraction and classification of the processed signal image. Deep learning-based classifiers have been applied in bearing fault diagnosis. While deep learning-based methods have demonstrated excellent feature extraction and classification capabilities, they typically face challenges when processing multi-source data. These deep learning methods often rely on a single type of data source, such as vibration signals, and cannot fully utilize comprehensive information from different sensors or different signal types (such as torque and current).

[0005] In summary, how to consider the complementarity between different sensors and extract effective fault vibration features from multi-source data under complex working conditions and unstable operating environments is a major challenge in rolling bearing fault diagnosis. Summary of the Invention

[0006] This invention provides a multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes, aiming to solve at least one of the aforementioned defects in the prior art.

[0007] This invention relates to a multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes, comprising the following steps: Multi-source signal data of rolling bearings are acquired, and a fused multi-source signal SDP image is generated based on the multi-source signal processing method of fusion intrinsic mode function based on point symmetric mode. The source signal data includes vibration, torque, current, force signals and data acquisition description documents collected by multiple sensors. The MobileViT-SCConv classification model is used to classify and predict multi-source signal SDP images; Multi-source data fusion images are randomly divided into training, validation, and test sets. The MobileViT-SCConv classification model is used to update the model parameters of the training set. The training is iteratively performed based on the results on the validation set. Then, the test set is used to evaluate the classification prediction effect to achieve fault diagnosis of rolling bearings.

[0008] Furthermore, the steps of acquiring multi-source signal data of rolling bearings and generating a fused multi-source signal SDP image based on the multi-source signal processing method using a point-symmetric mode fusion intrinsic mode function fusion method include: The raw signals collected by each sensor are normalized, and the normalized multi-source signal data are used to generate a fused multi-source signal SDP image through the SDP method. Empirical mode decomposition is performed on the SDP image of multi-source signals, which is decomposed into a finite number of intrinsic mode functions; The first intrinsic mode function is selected, and it is normalized. Then, the first intrinsic mode function containing torque and force signals is printed in the inner circle of the SDP image of multi-source data fusion. The first intrinsic mode function containing vibration, first current and second current signals is printed in the outer circle of the polar coordinate image.

[0009] Furthermore, in the step of normalizing the raw signals acquired by each sensor and generating a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method, the raw signals are normalized using the following formula:

[0010] in, and These are the normalized dataset and the original dataset, respectively. One sample, and These are the maximum and minimum values ​​of the training data, respectively.

[0011] Furthermore, in the step of normalizing the raw signals acquired by each sensor and generating a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method, the normalized multi-source signal data... , , , , A fused multi-source signal SDP image is generated using the SDP method, wherein... , , , , These refer to data from the torque, force, vibration, first current, and second current sensors, respectively.

[0012]

[0013] in, yes In polar coordinates, the polar radius, It is the first in the original dataset One sample, The initial rotation angle, It is the time lag factor used for plotting. It is the first in the original dataset One sample, It is the magnification factor. This is the sum of the total number of sensors; Let be the positive angle of the point in polar coordinates. The negative angle of the point in polar coordinates. For the first One sensor.

[0014] Furthermore, in the step of performing empirical mode decomposition on the multi-source signal SDP image, decomposing it into a finite number of intrinsic mode functions, the intrinsic mode functions are:

[0015] in, These are intrinsic mode functions. yes The quantity of the ingredients It is the first One eigenmode function It is the residual component.

[0016] Furthermore, the first intrinsic mode function (IMF) is selected, normalized, and then printed on the inner ring of the SDP image containing torque and force signals, respectively. In the step of printing the first IMF containing vibration, first current, and second current signals on the outer ring of the polar coordinate image, the first IMF includes... , , , and ,in , , , and Each refers to , , , , The first eigenmode function after empirical mode decomposition.

[0017] Furthermore, the steps for classifying and predicting multi-source signal SDP images using the MobileViT-SCConv classification model include: The multi-source signal SDP image is input into a 3x3 convolutional layer for preliminary feature extraction; The inverted residual blocks of MobileNetV2 are used to perform feature transformation and downsampling on the initial features: first, the initial features are expanded by 1x1 convolution, then spatial features are extracted by 3x3 depthwise separable convolution, and finally the number of channels of the feature map is restored to the original size by 1x1 convolution. By analyzing input features Apply a 3x3 convolution kernel, where This represents higher-level features extracted step-by-step through multiple convolutional layers and non-linear activation functions, achieving local perception and feature extraction. , , and These are the height and width of the input image, respectively. and These represent the number of input and output channels, respectively. The feature map is flattened and divided into several non-overlapping image blocks. Then, the pixels at the same position in each image block are... The input is fed into a Transformer encoder for encoding to obtain the output. ;in, This represents the global structural information features obtained through the Transformer encoder. , It is the size of the flattened image patch; , It is the number of image patches; It is an image block width, It is an image block of high; Will Obtained by superposition ,in, represent The structural information features obtained through superposition will The features are then reassembled into the initial features through a 1x1 convolution operation. The shape, and through the cat operation, is compared with the input features. The features are combined and then fused using a 3x3 convolutional layer to obtain the output features. ,in, This represents the local and global structural information features obtained through multi-layer convolution and Transformer encoders; Output features Spatial refinement features are obtained through the SRU operation in the SCConv module, and then channel refinement features are obtained through the CRU operation. The feature map is then converted into a fixed-length feature vector through a global average pooling layer. Finally, the feature vector is fed into a fully connected layer for classification.

[0018] Furthermore, the feature map is flattened and divided into several non-overlapping image blocks. Then, the pixels at the same position in each image block are... The input is fed into a Transformer encoder for encoding to obtain the output. In the steps, output It is calculated using the following formula:

[0019] in, The first one after processing by the Transformer encoder Output information features at each location It refers to the first in the flattened image block The input feature vector at each position, Indicates to Encoding operations, For pixels at the same position in each image patch, This refers to the size of the flattened image block.

[0020] The beneficial effects achieved by this invention are as follows: This invention provides a multi-source signal processing method that integrates intrinsic mode functions (IMFs) and point-symmetric modes. It acquires multi-source signal data from rolling bearings and generates a fused multi-source signal SDP image using an IMF-integrated multi-source signal processing method based on point-symmetric modes. The MobileViT-SCConv classification model is then used to classify and predict the multi-source signal SDP image. The fused multi-source image is randomly divided into training, validation, and test sets. The MobileViT-SCConv classification model is used to update the model parameters of the training set. Training is iteratively performed based on the results on the validation set, and the classification prediction effect is evaluated using the test set, thus achieving fault diagnosis of rolling bearings. This invention's multi-source signal processing method, which integrates IMFs and point-symmetric modes, improves upon the technical problem that a single data source cannot fully reflect complex fault modes, leading to missed diagnoses and misdiagnoses. It effectively reduces the impact of a single data source on fault diagnosis. The addition of IMF integration addresses the difficulty of simultaneously considering both global and local features in multi-source signal processing. By constructing an attention-based classification model to extract fault features from fused images, reducing information loss through channel and spatial reconstruction units, and dynamically adjusting attention weights, the accuracy of fault identification is improved, further enhancing the classification accuracy of fault diagnosis. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the multi-source signal processing method of the present invention that integrates intrinsic mode functions and point symmetric modes; Figure 2 This is a comparison chart of feature visualization results in the multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes in this invention. Figure 3 This is a comparison diagram of the confusion matrix results in the multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes in this invention; Figure 4 This is a comparison chart of prediction results in the multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes of the present invention. Detailed Implementation

[0022] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0023] like Figures 1 to 4 As shown, the first embodiment of the present invention proposes a multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes, including the following steps: Step S100: Acquire multi-source signal data of the rolling bearing, and generate a fused multi-source signal SDP image based on the multi-source signal processing method of fusion intrinsic mode function based on point symmetric mode according to the multi-source signal data; the source signal data includes vibration, torque, current, force signals and data acquisition description documents collected by multiple sensors.

[0024] The multi-source signal data of the rolling bearing is acquired, including vibration, torque, current, force signals collected by multiple sensors and data acquisition documentation. The fused multi-source signal SDP (Symmetrized Dot Pattern) image is generated using the Fusion of Intrinsic Mode Functions Multisource Signal Symmetrized Dot Pattern (IMFMSSDP) processing method based on point symmetry proposed in this embodiment.

[0025] Step S200: Use the MobileViT-SCConv classification model to classify and predict the multi-source signal SDP image.

[0026] The components of the empirical mode decomposition (EMD) of the multi-source signals are added to the SDP image to synthesize an SDP-EMD (SymmetrizedDot Pattern-Empirical Mode Decomposition) image, thereby fusing global and local fault feature information of the multi-source signals. The MobileViT-SCConv (Mobile Vision Transformer Self-Calibrated Convolution) classification model proposed in this embodiment is used to classify the multi-source signal SDP image obtained in step S100.

[0027] Step S300: Randomly divide the multi-source data fusion image into a training set, a validation set, and a test set. Update the model parameters of the training set using the MobileViT-SCConv classification model. Iterate the training based on the results on the validation set. Then, use the test set to evaluate the classification prediction effect to achieve fault diagnosis of rolling bearings.

[0028] The multi-source data fusion image generated in step S200 is randomly divided into a training set, a validation set, and a test set. The model parameters are updated using MobileViT-SCConv on the training set. The model is trained iteratively based on the results on the validation set. The classification prediction effect is then evaluated using the test set, thereby achieving fault diagnosis.

[0029] By using multi-source data fusion images as training samples, the improved MobileViT-SCConv classification model is used to classify bearing fault categories, thereby achieving efficient and accurate fault diagnosis of rolling bearings.

[0030] Further, please see Figures 1 to 4 The multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes proposed in this embodiment includes step S100: Step S110: Normalize the raw signals collected by each sensor, and generate a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method.

[0031] In the step of normalizing the raw signals acquired by each sensor and generating a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method, the raw signals are normalized using the following formula: (1) In formula (1), and These are the normalized dataset and the original dataset, respectively. One sample, and These are the maximum and minimum values ​​of the training data, respectively.

[0032] Normalized signal data , , , , A fused multi-source signal SDP image is generated using the SDP method, namely: (2) (3) In formulas (2) and (3), , , , , These correspond to the data collected by the torque sensor, force sensor, vibration sensor, first current sensor, and second current sensor, respectively. and These are the maximum and minimum values ​​of the sample time-domain signal, respectively. yes In polar coordinates, the polar radius, The initial rotation angle and the time lag factor are given. =4, magnification factor =35°, the sum of the total number of sensors =5, It refers to the first One sensor, when When the values ​​are 1, 2, 3, 4, and 5, they correspond to vibration, torque, the first current, the second current, and the force sensor, respectively. It refers to the first Sample data from the nth sensor, the first one in the original dataset One sample.

[0033] Step S120: Perform empirical mode decomposition on the multi-source signal SDP image, decomposing it into a finite number of intrinsic mode functions.

[0034] The acquired raw signal is decomposed using Empirical Mode Decomposition (EMD) to decompose the complex signal into a finite number of Intrinsic Mode Functions (IMFs). The IMFs are as follows: (4) In formula (4), These are intrinsic mode functions. yes The quantity of the ingredients It is the first One eigenmode function It is the residual component.

[0035] Step S130: Select the first intrinsic mode function of the intrinsic mode function, normalize the first intrinsic mode function, and then print the first intrinsic mode function containing torque and force signals in the inner circle of the SDP image of multi-source data fusion; print the first intrinsic mode function containing vibration, first current and second current signals in the outer circle of the polar coordinate image.

[0036] The first intrinsic mode function after EMD decomposition of each multi-source data is selected, i.e. , , , , To better display the multi-source data SDP image with fused intrinsic mode functions, the first intrinsic mode function after decomposition is first normalized. Then, the first intrinsic mode function containing torque and force signals is printed in the inner circle of the multi-source data fused SDP image, and the first intrinsic mode function containing vibration, first current and second current signals is printed in the outer circle of the polar coordinate image.

[0037] Preferably, see Figures 1 to 4 The multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes proposed in this embodiment includes step S200 as follows: Step S210: Input the multi-source signal SDP image into a 3x3 convolutional layer for preliminary feature extraction.

[0038] The multi-source signal SDP image generated in step S100 is input into a 3x3 convolutional layer for processing to extract preliminary features.

[0039] Step S220: Use the inverted residual block of MobileNetV2 to perform feature transformation and downsampling on the preliminary features: First, the preliminary features are expanded by 1x1 convolution, then spatial features are extracted by 3x3 depthwise separable convolution, and finally the number of channels of the feature map is restored to the original size by 1x1 convolution.

[0040] The inverted residual blocks of MobileNetV2 are used to perform feature transformation and downsampling on the preliminary features extracted in step S100. The preliminary features are first expanded by 1x1 convolution, then spatial features are extracted by 3x3 depthwise separable convolution, and finally the number of channels of the feature map is restored to the original size by 1x1 convolution.

[0041] Step S230: By analyzing the input features Apply a 3x3 convolution kernel, where This represents higher-level features extracted step-by-step through multiple convolutional layers and non-linear activation functions, achieving local perception and feature extraction. , , and These are the height and width of the input image, respectively. and These represent the number of input and output channels, respectively.

[0042] By analyzing input features By applying 3x3 convolutional kernels, local perception and feature extraction are achieved. In this embodiment, the height of the input image =256, Width of the input image =256, Number of input channels =3, Number of output channels =16.

[0043] Step S240: Flatten the feature map and divide it into several non-overlapping image blocks. Then, the pixels at the same position in each image block are... The input is fed into a Transformer encoder for encoding to obtain the output. ;in This represents the global structural information features obtained through the Transformer encoder. , It is the size of the flattened image patch; , It is the number of image patches; It is an image block width, It is an image block of high.

[0044] It is a dimension A real matrix or tensor; in this embodiment, the size of the flattened image patch. =4, = =2, , It represents the number of image patches. It is a dimension A real matrix or tensor.

[0045] Output It is calculated using the following formula: (5) In formula (5), The first one after processing by the Transformer encoder Output information features at each location Indicates to Encoding operations, For pixels at the same position in each image patch, This refers to the size of the flattened image block.

[0046] Step S250, Obtained by superposition ,in, represent The structural information features obtained by superposition will The features are then reassembled into the initial features through a 1x1 convolution operation. The shape, and through the cat operation with The features are combined and then fused using a 3x3 convolutional layer to obtain the output features. ,in, This represents the local and global structural information features obtained through multi-layer convolution and Transformer encoders.

[0047] The `cat` command, short for `catenate` or `concatenate`, is a command in Linux systems used to concatenate, display, or manipulate text files. It can combine the contents of multiple text files and display them on the terminal, or write the result to a new file.

[0048] Step S260: Output features Spatial refinement features are obtained through the SRU operation in the SCConv module, and then channel refinement features are obtained through the CRU operation. The feature map is then converted into a fixed-length feature vector through a global average pooling layer. Finally, the feature vector is fed into a fully connected layer for classification.

[0049] The SCConv (Spatial and Channel Reconstruction Convolution) module can obtain more discriminative features through simple latent-space representation enhancement.

[0050] Cross-Refinement Unit (CRU) is an operation that optimizes and extracts features through a series of operations. CRU primarily involves feature optimization and extraction, using specific algorithms and computational processes to finely process input features to obtain more accurate and useful feature representations. This operation is widely used in machine learning and deep learning, especially when processing complex data such as images and text. CRU helps models better understand and extract key information from the data, thereby improving model performance and accuracy.

[0051] To demonstrate the effectiveness of the method of this invention, vibration, torque, force, and two current sensor data from the Paderborn dataset in Germany are used and compared with different classification methods for further description. The Paderborn rolling bearing dataset has three system states: inner ring fault, outer ring fault, and rolling element fault. The vibration and two current sensor datasets are sliced ​​into 2048 data points, and the torque and force sensor datasets are sliced ​​into 128 data points, thus dividing the dataset into training and test sets. The method of this invention is used to classify the rolling bearing dataset for faults. For the dataset, the original signals are first processed using the IMFMSDP (Fusion of Intrinsic Mode Functions Multisource Signal Symmetrized Dot Pattern) method to generate image data of fused multisource signals. Then, the MobileViT-SCConv algorithm is used to extract features for classification and diagnosis. A comparative analysis was conducted on the classification accuracy of the method of this invention compared with AlexNet, DenseNet, GoogleNet, ResNet18, CNN, MobileViT, and MobileNetV2. The method of this invention achieves a classification accuracy of 99.43%, which is 1.71%, 0.88%, 1.62%, 0.96%, 1.49%, 1.80%, and 1.84% higher than the aforementioned methods, respectively. This indicates that the method of this invention has better prediction accuracy and can efficiently diagnose faults. The feature visualization results are comparable to... Figure 2 As shown, the confusion matrix results are for example... Figure 3 As shown, the prediction results are for example... Figure 4 As shown.

[0052] The multi-source signal processing method fusing intrinsic mode functions (IMFs) and point-symmetric modes provided in this embodiment, compared with existing technologies, acquires multi-source signal data of rolling bearings and generates a fused multi-source signal SDP image based on the IMF fusion method using point-symmetric modes. The MobileViT-SCConv classification model is used to classify and predict the multi-source signal SDP image. The fused multi-source data image is randomly divided into training, validation, and test sets. The MobileViT-SCConv classification model is used to update the model parameters of the training set, and training is iteratively performed based on the results on the validation set. Finally, the test set is used to evaluate the classification and prediction effect, thus achieving fault diagnosis of rolling bearings. The multi-source signal processing method fusing IMFs and point-symmetric modes provided in this embodiment improves the problem that a single data source cannot fully reflect complex fault modes, thus causing missed diagnoses and misdiagnoses, effectively reducing the impact of a single data source on fault diagnosis. The addition of fused IMFs improves the phenomenon that multi-source signal processing struggles to simultaneously consider both global and local features. By constructing an attention-based classification model to extract fault features from fused images, reducing information loss through channel and spatial reconstruction units, and dynamically adjusting attention weights, the accuracy of fault identification is improved, further enhancing the classification accuracy of fault diagnosis.

[0053] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A multi-source signal processing method that integrates intrinsic mode functions and point-symmetric modes, characterized in that, Includes the following steps: Multi-source signal data of rolling bearings are acquired, and a fused multi-source signal SDP image is generated based on the multi-source signal data using a point-symmetric mode-based fusion intrinsic mode function multi-source signal processing method; the source signal data includes vibration, torque, current, force signals collected by multiple sensors and data acquisition description documents; The MobileViT-SCConv classification model is used to classify and predict the multi-source signal SDP image. The multi-source data fusion image is randomly divided into a training set, a validation set, and a test set. The model parameters of the training set are updated using the MobileViT-SCConv classification model. The model is trained iteratively based on the results on the validation set. Then, the classification prediction effect is evaluated using the test set to achieve fault diagnosis of rolling bearings. The steps of acquiring multi-source signal data of rolling bearings and generating a fused multi-source signal SDP image based on the multi-source signal data using a point-symmetric mode-based fusion intrinsic mode function multi-source signal processing method include: The raw signals collected by each sensor are normalized, and the normalized multi-source signal data are used to generate a fused multi-source signal SDP image through the SDP method. The multi-source signal SDP image is subjected to empirical mode decomposition, which decomposes it into a finite number of intrinsic mode functions; The first intrinsic mode function of the intrinsic mode function is selected, and the first intrinsic mode function is normalized. Then, the first intrinsic mode function containing torque and force signals is printed in the inner circle of the SDP image of multi-source data fusion. The first intrinsic mode function containing vibration, first current and second current signals is printed in the outer circle of the polar coordinate image. In the step of normalizing the raw signals collected by each sensor and generating a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method, the normalized multi-source signal data... , , , , A fused multi-source signal SDP image is generated using the SDP method, wherein... , , , , These refer to data from the torque, force, vibration, first current, and second current sensors, respectively. in, yes In polar coordinates, the polar radius, It is the first in the original dataset One sample, The initial rotation angle, It is the time lag factor used for plotting. It is the first in the original dataset One sample, It is the magnification factor. This represents the sum of the total number of sensors; Let be the positive angle of the point in polar coordinates. The negative angle of the point in polar coordinates. For the first One sensor; In the step of selecting the first intrinsic mode function of the intrinsic mode functions, firstly, the first intrinsic mode function is normalized, and then the first intrinsic mode function containing torque and force signals is printed on the inner circle of the SDP image of multi-source data fusion; in the step of printing the first intrinsic mode function containing vibration, first current and second current signals on the outer circle of the polar coordinate image, the first intrinsic mode function includes , , , and ,in , , , and Each refers to , , , , The first eigenmode function after empirical mode decomposition.

2. The multi-source signal processing method that integrates intrinsic mode functions and point-symmetric modes as described in claim 1, characterized in that, In the step of normalizing the raw signals acquired by each sensor and generating a fused multi-source signal SDP image from the normalized multi-source signal data using the SDP method, the raw signals are normalized using the following formula: in, and These are the normalized dataset and the original dataset, respectively. One sample, and These are the maximum and minimum values ​​of the training data, respectively.

3. The multi-source signal processing method that integrates intrinsic mode functions and point-symmetric modes as described in claim 1, characterized in that, In the step of performing empirical mode decomposition on the multi-source signal SDP image to decompose it into a finite number of intrinsic mode functions, the intrinsic mode functions are: in, These are intrinsic mode functions. yes The quantity of the ingredients It is the first One eigenmode function It is the residual component.

4. The multi-source signal processing method that integrates intrinsic mode functions and point symmetric modes as described in claim 1, characterized in that, The step of classifying and predicting the multi-source signal SDP image using the MobileViT-SCConv classification model includes: The multi-source signal SDP image is input into a 3x3 convolutional layer for preliminary feature extraction. The inverted residual blocks of MobileNetV2 are used to perform feature transformation and downsampling on the preliminary features: first, the preliminary features are expanded by a 1x1 convolution, then spatial features are extracted by a 3x3 depthwise separable convolution, and finally the number of channels of the feature map is restored to the original size by a 1x1 convolution. By analyzing input features By applying 3x3 convolutional kernels, local perception and feature extraction are achieved. ,in, This represents higher-level features extracted step-by-step through multiple convolutional layers and non-linear activation functions. This represents the local structural information features obtained through multiple convolutions. and These are the height and width of the input image, respectively. and These represent the number of input and output channels, respectively. The feature map is flattened and divided into several non-overlapping image blocks. Then, the pixels at the same position in each image block are... The input is fed into a Transformer encoder for encoding to obtain the output. ;in, Representing the global structural information features obtained through the Transformer encoder, where R is the set of real numbers. , It is the size of the flattened image patch; , It is the number of image patches; It is an image block width, It is an image block of high; Will Obtained by superposition ,in, represent The structural information features obtained by superposition will The features are then reassembled into the initial features through a 1x1 convolution operation. The shape, and through the cat operation, is compared with the input features. The features are combined and then fused using a 3x3 convolutional layer to obtain the output features. ,in, This represents the local and global structural information features obtained through multi-layer convolution and Transformer encoders; Output features Spatial refinement features are obtained through the SRU operation in the SCConv module, and then channel refinement features are obtained through the CRU operation. The feature map is then converted into a fixed-length feature vector through a global average pooling layer. Finally, the feature vector is fed into a fully connected layer for classification.

5. The multi-source signal processing method that integrates intrinsic mode functions and point-symmetric modes as described in claim 4, characterized in that, The feature map is flattened and divided into several non-overlapping image blocks. Then, the pixels at the same position in each image block are... The input is fed into a Transformer encoder for encoding to obtain the output. In the steps, the output It is calculated using the following formula: in, The first one after processing by the Transformer encoder Output information features at each location It refers to the first in the flattened image block The input feature vector at each position, Indicates to Encoding operations, For pixels at the same position in each image patch, This refers to the size of the flattened image patch.

Citation Information

Patent Citations

  • Multi-scale gearbox fault intelligent diagnosis method

    CN115808305A

  • Bearing diagnosis method and system based on multi-modal and multi-scale fusion network

    CN117763494A