Bearing Fault Diagnosis Method and Storage Medium

Abstract

The present invention provides a bearing fault diagnosis method and a storage medium. The method includes: acquiring a vibration signal of a target bearing, and performing modal decomposition on the vibration signal to obtain a first preset number of intrinsic mode components; extracting features of each intrinsic mode component to form a frequency band feature vector; determining a frequency band attention weight vector according to each intrinsic mode component; modulating the frequency band feature vector by using the frequency band attention weight vector to obtain a frequency band fusion feature; and obtaining a fault recognition result according to the frequency band fusion feature. The present invention adaptively adjusts the frequency band attention weight vector based on the vibration signal, so that the frequency band fusion feature can be adaptively modulated. The frequency band fusion feature is more sensitive to faults and insensitive to working conditions, thereby effectively improving the accuracy of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and particularly relates to a bearing fault diagnosis method and a storage medium. Background Art

[0002] Bearings are key rotating components in industrial mechanical equipment, and their health status plays an important role in the continuity and safety of the operation of the entire mechanical system. Especially in risk-sensitive industrial fields such as aerospace, high-speed rail, wind power generation, and intelligent manufacturing, real-time monitoring and fault diagnosis of the operating status of bearings are of great significance. With the gradual improvement of the informatization and intelligence levels of industrial mechanical equipment, higher requirements are put forward for the diagnostic accuracy, intelligence, working condition generalization, engineering practicability, and credibility of bearing fault diagnosis methods.

[0003] In the prior art, deep learning methods have been widely used in the field of fault diagnosis due to their powerful non-linear deep feature learning ability. However, the feature distributions of bearing vibration signals collected under variable working conditions are inconsistent, and it is difficult for the model training set to cover all operating condition conditions. When the test set working conditions are not included in the training set working conditions, the training set and the test set do not meet the independent and identically distributed conditions, resulting in poor generalization ability of the model and low fault diagnosis accuracy, and it is impossible to ensure the accurate identification of bearing faults. Summary of the Invention

[0004] Embodiments of the present invention provide a bearing fault diagnosis method and a storage medium to solve the problem of low accuracy of fault identification of existing deep learning algorithms for bearings under variable working conditions.

[0005] In a first aspect, embodiments of the present invention provide a bearing fault diagnosis method, including:

[0006] Obtain the vibration signal of the target bearing, and perform modal decomposition on the vibration signal to obtain the first preset number of intrinsic mode components;

[0007] Extract the features of each intrinsic mode component to form a frequency band feature vector;

[0008] Determine a frequency band attention weight vector according to each intrinsic mode component;

[0009] Modulate the frequency band feature vector with the frequency band attention weight vector to obtain a frequency band fusion feature;

[0010] Obtain a fault identification result according to the frequency band fusion feature.

[0011] In a second aspect, an embodiment of the present invention provides a computer-readable storage medium storing a computer program, which when executed by a processor, implements the steps of the bearing fault diagnosis method provided in the first aspect above or any possible implementation manner of the first aspect.

[0012] An embodiment of the present invention provides a bearing fault diagnosis method and a storage medium. The above bearing fault diagnosis method includes: acquiring a vibration signal of a target bearing, and performing modal decomposition on the vibration signal to obtain a first preset number of intrinsic mode components; extracting features of each intrinsic mode component to form a frequency band feature vector; determining a frequency band attention weight vector according to each intrinsic mode component; modulating the frequency band feature vector by using the frequency band attention weight vector to obtain a frequency band fusion feature; and obtaining a fault recognition result according to the frequency band fusion feature. The present invention adaptively adjusts the frequency band attention weight vector based on the vibration signal, that is, adaptively adjusts according to the specific working condition, so that the frequency band fusion feature is insensitive to the working condition but more sensitive to the fault, has strong generalization ability, and effectively improves the accuracy of bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0014] Figure 1 is a flowchart of the implementation of a bearing fault diagnosis method provided by an embodiment of the present invention;

[0015] Figure 2 is a schematic flowchart of an adaptive dilation rate multi-scale dilated convolution module provided by an embodiment of the present invention;

[0016] Figure 3 is a schematic flowchart of the extraction of the frequency band fusion feature and the fault diagnosis provided by an embodiment of the present invention;

[0017] Figure 4 is a schematic structural diagram of a bearing fault diagnosis device provided by an embodiment of the present invention;

[0018] Figure 5 is a schematic diagram of a terminal device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will be described through specific embodiments in conjunction with the accompanying drawings.

[0021] Refer to Figure 1 , which shows the implementation flowchart of a bearing fault diagnosis method provided by an embodiment of the present invention, and is described in detail as follows:

[0022] The above-mentioned bearing fault diagnosis method includes:

[0023] S101: Obtain the vibration signal of the target bearing, and perform modal decomposition on the vibration signal to obtain the first preset number of intrinsic mode components;

[0024] In the embodiments of the present invention, a one-dimensional vibration signal in the vertical direction of the axle box with a fixed length can be collected for analysis.

[0025] When a bearing fails, vibration shock components will be generated in its vibration signal. When the rotational speed and load change, the time interval of the generated vibration shock, the frequency and amplitude of the shock components will also change accordingly. The characteristics in the time domain are that the fault characteristics appear on different time scales at different rotational speeds, and the characteristics in the frequency domain are that the fault characteristics appear in different frequency bands at different rotational speeds. Since the working conditions change randomly, it is impossible to determine which time scale or frequency band is more sensitive to bearing faults under the current working conditions. Therefore, in this application, the empirical mode decomposition algorithm (EMD) is first used to perform modal decomposition on the vibration signal, and the vibration signal is adaptively decomposed into a series of intrinsic mode components .

[0026] Since the first few intrinsic mode components often represent the main characteristics of the signal, the dominant modes (the first few intrinsic mode components) carry most of the energy and key information of the signal, and can reflect the main dynamic characteristics of the bearing. The subsequent intrinsic mode components often correspond to some small, low-frequency vibrations in the system or secondary details in the signal, and may even contain noise components. These low-frequency and secondary modes have little impact on the overall behavior of the bearing. Retaining too many unimportant mode components may introduce unnecessary complexity, mask the main characteristics, and interfere with the understanding of the essence of the signal.

[0027] Therefore, in the embodiments of the present invention, only the first preset number of intrinsic mode components are retained, and the signal components in different frequency bands contained in the vibration signal are adaptively screened out from high to low level by level to form (for example, ) intrinsic mode components of different frequency bands. One intrinsic mode component corresponds to a component signal of one frequency band. In this way, the signal can be decomposed into frequency bands, which can effectively remove noise and secondary information, simplify the analysis process, more clearly grasp the core characteristics of the bearing, reduce the computational complexity on the premise of meeting the feature analysis, and save storage space.

[0028] Since the frequency components contained in each intrinsic mode component change with the change of the vibration signal, it has self-adaptability.

[0029] In a possible implementation manner, the preset number can be 5.

[0030] Based on the application scenario of the present application, the preset number can be 5. Specifically, the number of intrinsic mode components can be set according to actual application requirements, and will not be specifically limited here.

[0031] EMD is an analysis method based on the time scale of the data itself, which can decompose a complex signal into several IMFs (Intrinsic Mode Function, intrinsic mode components) and a residual component. As an adaptive signal decomposition method, empirical mode decomposition can effectively decompose non-stationary signals, extract useful information hidden in the signals, and at the same time play a role in denoising.

[0032] The specific steps of the EMD algorithm are as follows:

[0033] Assume that the input signal is , and use the cubic spline function to find the maximum and minimum values of the input signal for fitting to obtain the upper envelope and the lower envelope of the signal.

[0034] Calculate the average value of the upper and lower envelopes to obtain the envelope mean

[0035] Subtract the envelope mean from the input signal to obtain the remaining signal

[0036] Repeat the above process for the remaining signal until the condition SD of the mode component is less than the threshold value and then stop, to obtain the first IMF component .

[0037] Among them, the SD condition is as follows:

[0038]

[0039] Take the difference between the input signal and to obtain the first-order residual quantity . Use to replace the input signal and perform the operations of the above steps. Repeat times to obtain the -order intrinsic mode function component and -order residual .

[0040]

[0041] Therefore, the expression of the EMD decomposition of the input signal is as follows: .

[0042] The length of the input signal is . After empirical mode decomposition, multiple IMF components are obtained, and the length of each IMF component is also .

[0043] Corresponding to the embodiment of the present invention, another preset number is 5, that is , and the first 5 intrinsic mode function components are obtained.

[0044] S102: Extract the features of each intrinsic mode function component to form a frequency band feature vector;

[0045] This application extracts the features of each intrinsic mode function component, that is, the features of each frequency band, to form a frequency band feature vector.

[0046] In a possible implementation manner, S102 may include:

[0047] S1021: Determine the target dilation rate according to the vibration signal;

[0048] S1022: Set the dilation rate of the multi-scale dilated convolution module to the target dilation rate to obtain an adaptive dilation rate multi-scale dilated convolution module;

[0049] S1023: For any one intrinsic mode function component, input the intrinsic mode function component into the adaptive dilation rate multi-scale dilated convolution module to obtain the features of the intrinsic mode function component;

[0050] The multi-scale dilated convolution module can capture multi-scale features, expand the receptive field, reduce the number of parameters and computational complexity, and at the same time can alleviate the information loss in the pooling layer. Therefore, in the embodiments of the present invention, the multi-scale dilated convolution module is used to extract features of different frequency bands.

[0051] The dilation rate in the multi-scale dilated convolution module is a key parameter, which determines the size of the convolution kernel scale in the dilated convolution. A small dilation rate is suitable for capturing local detailed features, while a large dilation rate helps to obtain more global and large-scale features. Selecting an appropriate dilation rate has an important impact on the receptive field and feature extraction ability of the convolution operation. When the rotational speed increases, the fault characteristic frequency of the signal will increase, and features of a smaller time scale should be extracted. At this time, a smaller dilation rate should be used. When the rotational speed decreases, the fault characteristic frequency of the signal will decrease, and features of a larger time scale should be extracted. At this time, a larger dilation rate should be used. In the embodiments of the present invention, the target dilation rate is determined according to the vibration signal as the dilation rate of the multi-scale dilated convolution module, so that the dilation rate of the multi-scale dilated convolution module can be adaptively adjusted according to the bearing vibration signal characteristics under different working conditions, and the feature extraction ability of the network for bearing vibration signals under different working conditions can be optimized.

[0052] In a possible implementation manner, S1021 may include:

[0053] 1. Determine the center frequency according to the vibration signal;

[0054] 2. Input the center frequency into a fully connected layer, and round the output result to obtain the target dilation rate.

[0055] In a possible implementation manner, the center frequency The calculation formula may include:

[0056]

[0057] Where is the th spectral line of the vibration signal, , is the number of spectral lines, is the frequency value of the th spectral line.

[0058] Specifically, for a certain IMF component , the adaptive dilation rate multi-scale dilated convolution module is used to extract the features of the signal in this frequency band , and the target dilation rate is , and the specific steps are as follows:

[0059] Referring to Figure 2 , the th intrinsic mode component passes through a parameter of (where, is the number of convolutional kernels, is the width of the convolutional kernel) of the Conv1D layer, and the output size is features . Among them, in order to achieve the effect of denoising and reducing the feature length, the width of the convolutional kernel can be selected as a larger value.

[0060] Input the feature into 3 parallel Conv1D layers. The widths of the convolutional kernels of these 3 parallel Conv1D layers are different, which are , and respectively. Convolutional kernels with different widths are used to extract features of different scales, vary with the change of the vibration signal, that is, when the vibration signal changes, , and will also change their values, and features of different time scales can be extracted for different vibration signals.

[0061] For example: in the same bearing fault state, when the rotational speed increases, the fault characteristic frequency of the vibration signal will increase, and the center frequency of the vibration signal usually also increases, the value changes, which in turn causes the widths , and of the convolutional kernels of the 3 parallel Conv1D layers to change, that is, the adaptive dilation rate multi-scale dilated convolution module can adaptively extract multi-scale features within an appropriate scale range for vibration signals with different rotational speeds. The number of channels of the convolutional kernels of the 3 parallel Conv1D layers is all , and the convolutional stride of the Conv1D layer is all 1. Each Conv1D layer in the adaptive dilation rate multi-scale dilated convolution module can extract features of one scale, and the features of each scale are respectively expressed as , and . In order to make the features of each scale have the same length, zero-padding is used during convolution. , and are all of size .

[0062] Use a concatenation layer to connect , and along the channel dimension to form a multi-scale feature , expressed as:

[0063]

[0064] Among them, has a size of , . This splicing layer can aggregate features of all scales to form a multi-scale feature set.

[0065] Input into a Time-Distributed Conv1D (TDConv1D) layer, and output a feature vector with a size of . In the TDConv1D layer, Conv1D is simultaneously applied to the features of channels of to simultaneously extract the global features of each channel. The convolution kernel width of the TDConv1D layer is , and the number of channels is 1. Finally, the feature of the th channel in c is compressed into an element in the feature vector . Input into 2 fully connected layers, and output a multi-scale channel attention weight vector . The multi-scale channel attention weight vector is a vector with a size of . The th element in c represents the sensitivity of the feature of the th channel in c to bearing faults.

[0066] Use the multi-scale channel attention weight vector to optimize and modulate the multi-scale feature . The calculation formula is:

[0067]

[0068] Among them, represents the element-wise product operation.

[0069] In order to further extract multi-scale fusion features from the optimized and modulated multi-scale feature , input into a Conv1D layer for feature extraction and output the features in this frequency band . The purpose of this layer is to learn the dependencies between features of different scales and different channels, and to mine the features of the signal as a whole. The parameters of the Conv1D layer are represented as , and the convolution stride is . And a max - pooling layer with a pooling size of 2 is used to reduce the size of the features. Finally, this layer outputs a feature with a size of , that is, the feature of the th intrinsic mode component; where . .

[0070] S1024: Form frequency - band feature vectors from the features of each intrinsic mode component.

[0071] For example, referring to Figure 3 , corresponding to the first 5 intrinsic mode components , the features of each intrinsic mode component form a frequency - band feature vector .

[0072] S103: Determine the frequency - band attention weight vector according to each intrinsic mode component;

[0073] The frequency - band attention weight vector reflects the relative importance of signals in different frequency bands in the whole signal. For important frequency bands, higher weights are assigned, enabling the model to pay more attention to the features of these frequency bands, so as to better capture and utilize the information valuable for the task; for relatively unimportant frequency bands, lower weights are assigned, reducing the model's attention to these frequency bands, thereby reducing the impact of noise or irrelevant information on the model performance.

[0074] Kurtosis and Energy are two important concepts in signal analysis, which describe the features of signals from different perspectives. Kurtosis is sensitive to impulse signals caused by faults, especially when early faults occur, the kurtosis value changes significantly. The energy of a signal is a measure of the signal strength. When a bearing fails, the energy in the frequency band where the fault characteristic frequency is located will increase significantly;

[0075] The embodiments of the present invention comprehensively consider multiple features such as kurtosis and energy, can more comprehensively understand the characteristics and states of signals, and it is more reasonable to further calculate the frequency - band attention weight vector from kurtosis and energy features.

[0076] In a possible implementation manner, S103 may include:

[0077] S1031: For any one intrinsic mode component, determine the kurtosis index value of this intrinsic mode component and normalize it to obtain the normalized kurtosis of this intrinsic mode component; determine the energy index value of this intrinsic mode component and normalize it to obtain the normalized energy of this intrinsic mode component; add the normalized kurtosis of this intrinsic mode component and the normalized energy to obtain the attention weight of this intrinsic mode component;

[0078] S1032: Obtain the frequency - band attention weight vector according to the attention weights of each intrinsic mode component.

[0079] In a possible implementation, the calculation formula of the normalized kurtosis may include:

[0080]

[0081]

[0082] where is the kurtosis index value of the th intrinsic mode component, is the normalized kurtosis of the th intrinsic mode component; is the root mean square value of the th intrinsic mode component; is the th data point of the th intrinsic mode component; , is the preset quantity; , is the number of analysis points

[0083] Normalized energy The calculation formula of may include:

[0084]

[0085]

[0086] where is the energy index value of the th intrinsic mode component, is the normalized energy of the th intrinsic mode component.

[0087] Exemplarily, referring to Figure 3 , 5 intrinsic mode components are calculated to obtain 5 kurtosis index values , and 5 normalized kurtoses are obtained after normalization ; 5 intrinsic mode components are calculated to obtain 5 energy index values , and 5 normalized energies are obtained after normalization , and the corresponding ones are added to obtain 5 attention weights .

[0088] Among them, normalization is used to ensure that the sum of the kurtosis indexes of 5 intrinsic mode components is 1, and to ensure that the sum of the energy indexes of 5 intrinsic mode components is 1.

[0089] In a possible implementation, S1032 may include:

[0090] 1. Form an initial attention weight vector from the attention weights of each intrinsic mode component;

[0091] 2. Input the initial attention weight vector into two fully connected layers to obtain a frequency band attention weight vector.

[0092] In the embodiment of the present invention, referring to Figure 3 , form an initial attention weight vector from the attention weights of each intrinsic mode component , and input into two fully connected layers to obtain a frequency band attention weight vector .

[0093] The calculation formula of

[0094]

[0095] is as follows: where is the Sigmoid activation function, and the gating mechanism can be implemented through and are the parameters of the two fully connected layers respectively; and are the bias terms respectively; The -th element in represents the sensitivity of the -th frequency band in the frequency band feature vector to bearing faults.

[0096] Referring to Figure 3 , the frequency band attention weight vector is jointly determined by the model parameters (for example, the parameters of the two fully connected layers) and , while is calculated from the energy and kurtosis features of the intrinsic mode components of the vibration signal. After forming a model for the bearing fault identification method provided in the embodiment of the present invention and training the model, the parameters of the model are fixed. Therefore, the frequency band attention weight vector in the trained model only changes with , that is, the frequency band attention weight vector only changes with the change of the model vibration signal. Therefore, the frequency band attention weight vector has better adaptability than the fixed weight.

[0097] S104: Modulate the frequency band feature vector with the frequency band attention weight vector to obtain a frequency band fusion feature;

[0098] In a possible implementation manner, S104 may include:

[0099] S1041: Multiply each element in the frequency band attention weight vector with the corresponding element in the frequency band feature vector respectively, and splice the multiplied elements to obtain the frequency band fusion feature;

[0100] Among them, the number of elements in the frequency band attention weight vector is the same as the number of elements in the frequency band feature vector.

[0101] Reference Figure 3 , and use the frequency band attention weight vector to modulate the frequency band feature vector , and the calculation formula is as follows:

[0102]

[0103] Among them, is the frequency band fusion feature.

[0104] The frequency band feature vector is adaptively modulated to obtain the frequency band fusion feature after modulation optimization . Each element in the frequency band attention weight vector is a value between 0 and 1. Therefore, the frequency band attention weight vector can be regarded as a control gate, which can control the amount of information transmitted by the features of each frequency band to the subsequent layer, so as to achieve the purpose of enhancing the frequency band features sensitive to faults and suppressing the frequency band features and useless features sensitive to working conditions. Therefore, the frequency band fusion feature after modulation optimization is more sensitive to faults and insensitive to changes in working conditions.

[0105] S105: Obtain the fault recognition result according to the frequency band fusion feature.

[0106] Reference Figure 3 , the frequency band fusion feature can be input into two stacked convolutional modules for further feature extraction. The purpose is to learn the dependence relationship between different frequency band features and mine the features of the signal as a whole. Each convolutional module contains 1 convolutional layer, 1 batch normalization layer, 1 activation layer and 1 pooling layer. The extracted features are flattened by the flattening layer and then the classification result is output using 2 fully connected layers and a softmax layer y . The calculation formula of the softmax output layer is:

[0107]

[0108] Among them, represents the number of fault categories; represents the parameter set of the softmax classification layer; Represents the input of the softening maximum layer; Represents the output classification result of the softening maximum layer.

[0109] In the embodiments of the present invention, the value of the adaptive dilation rate establishes an adaptive dilation rate multi-scale dilated convolution module to extract features of each frequency band, and at the same time adaptively adjusts the attention weights of each frequency band, realizes fault feature extraction under invariant working conditions, improves the generalization ability of the working condition domain of fault recognition, improves the accuracy of bearing fault diagnosis under variable working conditions, and at the same time improves the interpretability and credibility of the fault diagnosis.

[0110] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0111] The following is the device embodiment of the present invention. For the details not described in detail, reference can be made to the corresponding method embodiments above.

[0112] Figure 4 The structural schematic diagram of the bearing fault diagnosis device provided by the embodiments of the present invention is shown. For the convenience of description, only the parts related to the embodiments of the present invention are shown and are described in detail as follows:

[0113] As Figure 4 shown, the bearing fault diagnosis device includes:

[0114] A modal decomposition module 21, configured to obtain the vibration signal of the target bearing and perform modal decomposition on the vibration signal to obtain the first preset number of intrinsic mode components;

[0115] An initial feature extraction module 22, configured to extract the features of each intrinsic mode component to form a frequency band feature vector;

[0116] A weight adaptive adjustment module 23, configured to determine a frequency band attention weight vector according to each intrinsic mode component;

[0117] A feature modulation module 24, configured to modulate the frequency band feature vector by using the frequency band attention weight vector to obtain a frequency band fusion feature;

[0118] A fault recognition module 25, configured to obtain a fault recognition result according to the frequency band fusion feature.

[0119] In a possible implementation manner, the initial feature extraction module 22 may include:

[0120] A dilation rate determination unit, configured to determine a target dilation rate according to the vibration signal;

[0121] The dilated convolution model establishment unit is used to set the dilation rate of the multi-scale dilated convolution module to the target dilation rate to obtain the multi-scale dilated convolution module with an adaptive dilation rate;

[0122] The eigen-component extraction unit is used to input any one of the intrinsic mode components into the multi-scale dilated convolution module with an adaptive dilation rate to obtain the features of the intrinsic mode component;

[0123] The eigen-vector formation unit is used to form the frequency band eigen-vector from the features of each intrinsic mode component.

[0124] In a possible implementation manner, the dilation rate determination unit may specifically be used for:

[0125] 1. Determine the center frequency according to the vibration signal;

[0126] 2. Input the center frequency into a fully connected layer and round the output result to obtain the target dilation rate.

[0127] In a possible implementation manner, the calculation formula of the center frequency may include:

[0128]

[0129] where is the th spectral line of the vibration signal, , is the number of spectral lines, is the frequency value of the th spectral line.

[0130] In a possible implementation manner, the weight adaptive adjustment module 23 may include:

[0131] The weight component calculation unit is used to, for any one of the intrinsic mode components, determine the kurtosis index value of the intrinsic mode component and normalize it to obtain the normalized kurtosis of the intrinsic mode component; determine the energy index value of the intrinsic mode component and normalize it to obtain the normalized energy of the intrinsic mode component; add the normalized kurtosis of the intrinsic mode component and the normalized energy to obtain the attention weight of the intrinsic mode component;

[0132] The weight vector formation unit is used to obtain the frequency band attention weight vector according to the attention weights of each intrinsic mode component.

[0133] In a possible implementation manner, the calculation formula of the normalized kurtosis may include:

[0134]

[0135]

[0136] Among them, is the kurtosis index value of the th intrinsic mode component, is the normalized kurtosis of the th intrinsic mode component; is the root mean square value of the th intrinsic mode component; is the th data point of the th intrinsic mode component; , is the preset quantity; , is the number of analysis points

[0137] Normalized energy The calculation formula of can include:

[0138]

[0139]

[0140] Among them, is the energy index value of the th intrinsic mode component, is the normalized energy of the th intrinsic mode component.

[0141] In a possible implementation manner, the weight vector forming unit can be specifically used for:

[0142] 1. Form an initial attention weight vector from the attention weights of each intrinsic mode component;

[0143] 2. Input the initial attention weight vector into a two-layer fully connected layer to obtain a frequency band attention weight vector.

[0144] In a possible implementation manner, the feature modulation module 24 can include:

[0145] A fusion splicing unit, configured to multiply each element in the frequency band attention weight vector by the corresponding element in the frequency band feature vector respectively, and splice the multiplied elements to obtain a frequency band fusion feature;

[0146] Among them, the number of elements in the frequency band attention weight vector is the same as the number of elements in the frequency band feature vector.

[0147] In a possible implementation manner, the preset quantity is 5.

[0148] Figure 5 is a schematic diagram of the terminal device 3 provided by the embodiments of the present invention. AsFigure 5 As shown, the terminal device 3 of this embodiment includes: a processor 30 and a memory 31. The memory 31 is used to store a computer program 32, and the processor 30 is used to call and run the computer program 32 stored in the memory 31, and execute the steps in the above-mentioned embodiments of each bearing fault diagnosis method, such as Figure 1 the steps S101 to S105 shown. Alternatively, the processor 30 is used to call and run the computer program 32 stored in the memory 31 to implement the functions of each module / unit in the above-mentioned device embodiments, such as Figure 4 the functions of the modules 21 to 25 shown.

[0149] Exemplarily, the computer program 32 can be divided into one or more modules / units. One or more modules / units are stored in the memory 31 and executed by the processor 30 to complete the present invention. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 32 in the terminal device 3. For example, the computer program 32 can be divided into Figure 4 the modules / units 21 to 25 shown.

[0150] The terminal device 3 can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art can understand that Figure 5 these are only examples of the terminal device 3 and do not constitute a limitation on the terminal device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the terminal may further include input / output devices, network access devices, a bus, etc.

[0151] The so-called processor 30 may be a central processing unit (CPU), or may also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0152] The memory 31 can be an internal storage unit of the terminal device 3, such as the hard disk or memory of the terminal device 3. The memory 31 can also be an external storage device of the terminal device 3, such as a plug-in hard disk equipped on the terminal device 3, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 31 can also include both the internal storage unit and the external storage device of the terminal device 3. The memory 31 is used to store computer programs and other programs and data required by the terminal. The memory 31 can also be used to temporarily store the data that has been output or will be output.

[0153] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be assigned to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiment and will not be elaborated herein.

[0154] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0155] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0156] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal and method can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical or other forms.

[0157] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0159] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0160] 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 foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A bearing fault diagnosis method, characterized in that: include: Acquire a vibration signal of a target bearing, and perform modal decomposition on the vibration signal to obtain a preset number of eigenmodal components; Extract the characteristics of each eigenmode component to form a frequency band feature vector; According to each eigenmode component, determine the frequency band attention weight vector; The frequency band feature vector is modulated by using the frequency band attention weight vector to obtain a frequency band fusion feature; Obtaining a fault identification result according to the frequency band fusion characteristics; The step of determining the frequency band attention weight vector according to each eigenmode component includes: For any eigenmode component, determine the kurtosis index value of the eigenmode component and normalize it to obtain the normalized kurtosis of the eigenmode component; determine the energy index value of the eigenmode component and normalize it to obtain the normalized energy of the eigenmode component; add the normalized kurtosis and normalized energy of the eigenmode component to obtain the attention weight of the eigenmode component; According to the attention weights of each eigenmodal component, the frequency band attention weight vector is obtained.

2. The bearing fault diagnosis method according to claim 1, characterized in that: The step of extracting the features of each eigenmode component to form a frequency band feature vector includes: determining a target expansion rate according to the vibration signal; The expansion rate of the multi-scale dilated convolution module is set to the target expansion rate to obtain an adaptive expansion rate multi-scale dilated convolution module; For any eigenmode component, input the eigenmode component into the adaptive dilation rate multi-scale dilated convolution module to obtain the characteristics of the eigenmode component; The characteristics of each eigenmode component are used to form the frequency band characteristic vector.

3. The bearing fault diagnosis method according to claim 2, characterized in that: Determining a target expansion rate according to the vibration signal includes: determining a center of gravity frequency based on the vibration signal; The centroid frequency is input into a fully connected layer, and the output result is rounded to obtain the target expansion rate.

4. The bearing fault diagnosis method according to claim 3, characterized in that: The center of gravity frequency The calculation formula include: in, is the vibration signal Spectrum lines, , is the number of spectral lines, For the The frequency value of the spectral line.

5. The bearing fault diagnosis method according to claim 1, characterized in that: The calculation formula of the normalized kurtosis includes: in, For the The kurtosis index value of the eigenmode component, For the The normalized kurtosis of the eigenmode components; For the The RMS value of the eigenmode components; For the The first eigenmode component data points; , is the preset number; , To analyze the points; The normalized energy The calculation formula includes: in, For the The energy index value of the eigenmode component, For the The normalized energy of the eigenmode components.

6. The bearing fault diagnosis method according to claim 1, characterized in that: The step of obtaining the frequency band attention weight vector according to the attention weights of each eigenmode component includes: The attention weights of each intrinsic modal component are formed into an initial attention weight vector; The initial attention weight vector is input into two fully connected layers to obtain the frequency band attention weight vector.

7. The bearing fault diagnosis method according to any one of claims 1 to 4, characterized in that: The step of using the frequency band attention weight vector to modulate the frequency band feature vector to obtain a frequency band fusion feature includes: Multiplying each element in the frequency band attention weight vector with each element in the frequency band feature vector respectively, and concatenating the multiplied elements to obtain the frequency band fusion feature; Among them, the number of elements in the frequency band attention weight vector is the same as the number of elements in the frequency band feature vector.

8. The bearing fault diagnosis method according to any one of claims 1 to 4, characterized in that: The preset number is 5.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the bearing fault diagnosis method as described in any one of claims 1 to 8 are implemented.

Images