Fault diagnosis method and system based on multi-scale dual feature enhancement

Abstract

The invention provides a fault diagnosis method and system based on multi-scale dual feature enhancement, and relates to the technical field of fault diagnosis. According to the method, fault defects of the rolling bearing are classified through a fault recognition model, wherein the fault recognition model comprises a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; the discriminative feature enhancement mechanism is seamlessly embedded into the feature selection module and the multi-scale feature extraction module to enhance fault features, and the discriminative feature enhancement mechanism comprises two modules, namely a multi-spectrum channel attention mechanism and a dual-path incentive attention mechanism. According to the method, rich fault feature information of different scales is extracted, fine-grained attention on interdependence between channels is realized through a discriminative feature enhancement mechanism, and important information of a pulse excitation section in a fault signal on a time axis can be emphasized, so that the influence of irrelevant information such as noise and working condition change on a final task is effectively reduced, and the reliability of the final task is improved. And the model fault diagnosis precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method and system based on multi-scale dual feature enhancement. Background Art

[0002] As the core component of rotating machinery, rolling bearings are widely used in a variety of industrial equipment such as transportation, energy and power, and aviation power, and are known as the "joints of industry". However, the long-term operation of rolling bearings in harsh working environments undoubtedly greatly increases the possibility of failure. Among the failure cases of mechanical equipment, 30% to 40% of the failures are caused by bearing failures. Therefore, accurate diagnosis of rolling bearing failures has always been an important engineering topic.

[0003] Traditional fault diagnosis methods rely too much on expert experience and have low efficiency in feature extraction, making it difficult to meet the requirements of high efficiency and high precision in fault diagnosis. Deep learning methods have well solved the shortcomings of traditional fault diagnosis. Therefore, rolling bearing fault diagnosis methods based on deep learning, such as long short-term memory networks (LSTM), deep belief networks (DBN), convolutional neural networks (CNN), etc., are widely used and studied in bearing diagnosis.

[0004] Among them, CNN and its variants dominate the bearing fault diagnosis methods based on deep learning. However, the existing fault diagnosis models based on convolutional neural networks still have shortcomings: (1) Lack of discriminative feature learning mechanism. Different features in CNN have different importance for fault diagnosis tasks. Traditional CNN cannot adaptively focus on learning more discriminative features and ignore useless features. (2) Lack of rich feature extraction for reliable fault diagnosis. Since vibration signals have nonlinear and non-stationary characteristics, and traditional CNN extracts features at a single scale, when the convolution kernel size is too large, some local information will be lost, and when it is too small, it is easy to ignore global features, resulting in low reliability of fault diagnosis. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the shortcomings of the prior art, the present invention provides a fault diagnosis method and system based on multi-scale dual feature enhancement, which solves the technical problem of low reliability of fault diagnosis based on convolutional neural network to construct a fault diagnosis model.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, the present invention provides a fault diagnosis based on multi-scale dual feature enhancement, comprising:

[0010] Acquire vibration signals of rolling bearings including several different fault defects in a certain period of time;

[0011] Preprocessing several vibration signals to obtain a data set;

[0012] The multi-scale convolutional neural network with a pre-constructed discriminative feature enhancement mechanism is trained and optimized through the data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of the rolling bearing;

[0013] Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

[0014] Preferably, the feature selection module includes a wide convolution module, and the wide convolution module is used to perform feature screening on the vibration signal, and the expression is:

[0015] x′ i =P(f(BN(Conv(x i ))))

[0016] Among them, P(·) is the maximum pooling operation, BN(·) is the batch normalization operation; f(·) is the ReLU activation function, Conv(·) is the convolution operation, and x i is the ith vibration signal, x′ i Represents the features filtered out after the above operations.

[0017] Preferably, the multi-scale feature extraction module includes different-scale convolution modules with a three-branch parallel structure.

[0018] Preferably, the feature fusion and output module includes a fusion channel, a global average pooling layer, and a fully connected layer; in the feature fusion and output module, a multi-spectral channel attention mechanism is used to adaptively adjust the features fused along the channel; the adjusted fusion features are processed by reducing feature parameters through the global average pooling layer, the output of the global average pooling layer is input to the fully connected layer, and the output of the fully connected layer is processed by Softmax. x The function outputs the fault classification result.

[0019] Preferably, the expression of the discriminative feature enhancement mechanism is:

[0020] X DFEM=F DPEAM (F MSCAM (X)

[0021] X DFEM represents the enhanced features after the discriminative feature mechanism, F MSCAM (·) indicates that the data is processed by the multi-spectral channel attention mechanism, F DPEAM (·) indicates that the data is processed by the dual-path incentivized attention mechanism.

[0022] Preferably, the data processing process in the multi-spectral channel attention mechanism includes:

[0023] The time series signal input is converted into the frequency domain through Fourier transform, which includes:

[0024] The input feature X is represented as X = [X 0 , X 1 , ..., X c-1 ],in i∈{0, 1, …, c-1}, N is the signal length, convert the time series input into the frequency domain:

[0025]

[0026] Where FFT(·) is the FFT function, is the jth frequency component on the i-th channel;

[0027] The frequency domain components are divided into blocks and weighted, including:

[0028] According to the size of the frequency component, the frequency components of each channel are divided into three parts: low frequency band, medium frequency band and high frequency band, and the weight of each frequency band is obtained by weighting

[0029]

[0030]

[0031] Where σ is the sigmoid function, f(·) is the ReLU activation function, and Conv(·) is the convolution operation; represents the weight of the kth frequency band on the ith channel, k = 1, 2, 3, representing the low frequency band, the middle frequency band, and the high frequency band respectively;

[0032] The weights of each frequency band are combined to achieve channel recalibration, including:

[0033]

[0034]

[0035]

[0036] in, is the final weight of each channel, It is element-by-element multiplication, and the weight W ci Multiplying with the corresponding frequency component to obtain the recalibrated frequency characteristic X(F i ), and the time-series recalibration feature vector is obtained through inverse Fourier transform IFFT(·)

[0037] Using the idea of ​​residual learning, we introduce residual connections to obtain the multi-spectral channel attention feature X MSCAM :

[0038]

[0039] Preferably, the data processing process in the dual-path incentive attention mechanism includes:

[0040] The feature X adaptively weighted by the channel MSCAM First, perform convolution operation along the channel domain to obtain the convolution feature X′:

[0041] X′=f(Conv 1×1 (X MSCAM ))

[0042] Among them, X MSCAM =[X 0 , X 1 , …, X n-1 ],in Indicates the jth position on the timing signal;

[0043] Then perform a dual-branch convolution operation on the feature X′ to obtain the dual-branch convolution feature X":

[0044] X″=Conv 1×1 (concat[f(Conv 3×1 (X′)), f(Conv 7×1 (X′))])

[0045] Perform time recalibration on the dual convolution feature X″ to obtain the time recalibration feature vector X N :

[0046] W s =σ(X″)

[0047]

[0048] Among them, s∈{0, 1,…, N-1},

[0049] The time-order recalibration feature vector is residually connected to the input feature to obtain the dual-path incentive attention X DPEAM :

[0050] X DPEAM =X MSCAM +X N .

[0051] In a second aspect, the present invention provides a fault diagnosis system based on multi-scale dual feature enhancement, comprising:

[0052] A data acquisition module, used to acquire vibration signals of rolling bearings including several different fault defects in a certain period of time;

[0053] A preprocessing module, used to preprocess a number of vibration signals to obtain a data set;

[0054] A model training module is used to train and optimize a multi-scale convolutional neural network with a pre-built discriminative feature enhancement mechanism through a data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of the rolling bearing;

[0055] Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

[0056] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for fault diagnosis based on multi-scale dual feature enhancement, wherein the computer program enables a computer to execute the fault diagnosis method based on multi-scale dual feature enhancement as described above.

[0057] In a fourth aspect, the present invention provides an electronic device, comprising:

[0058] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include methods for executing the fault diagnosis method based on multi-scale dual feature enhancement as described above.

[0059] (III) Beneficial effects

[0060] The present invention provides a fault diagnosis method and system based on multi-scale dual feature enhancement. Compared with the prior art, it has the following beneficial effects:

[0061] The fault diagnosis method and system based on multi-scale dual feature enhancement of the present invention classifies the fault defects of rolling bearings through a pre-trained fault identification model, wherein the fault identification model includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; the discriminative feature enhancement mechanism is seamlessly embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault features, wherein the discriminative feature enhancement mechanism includes two modules, namely a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism. The present invention extracts rich fault feature information of different scales, and through the discriminative feature enhancement mechanism, it not only realizes the fine-grained attention to the mutual dependence between channels, but also emphasizes the important information of the pulse excitation segment in the fault signal on the time axis, thereby effectively reducing the influence of irrelevant information such as noise and working condition changes on the final task, and improving the model fault diagnosis accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A block diagram of a fault diagnosis method based on multi-scale dual feature enhancement according to an embodiment of the present invention;

[0064] Figure 2 Detailed structural diagram of a multi-scale convolutional neural network of a discriminative feature enhancement mechanism in an embodiment of the present invention;

[0065] Figure 3 This is the specific structural diagram of the multi-spectral channel attention mechanism;

[0066] Figure 4 The specific structural diagram of the dual-path incentive attention mechanism. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. 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.

[0068] The embodiments of the present application provide a fault diagnosis method and system based on multi-scale dual feature enhancement, thereby solving the technical problem of low reliability of fault diagnosis in the current fault diagnosis model constructed based on convolutional neural networks, and improving the reliability of model fault diagnosis by extracting rich fault feature information at different scales.

[0069] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0070] The current rolling bearing fault diagnosis methods based on deep learning have the following main defects:

[0071] (1) The existing fault diagnosis model based on convolutional neural network is not rich enough in feature extraction. The ability to extract features from a single scale is limited. Due to the nonlinear and non-stationary characteristics of vibration signals, a single-scale convolutional neural network is not sufficient to extract rich features for reliable fault diagnosis.

[0072] (2) During the feature extraction process, the information value contained in features at different channels and scales is not equal. However, the fault diagnosis model based on convolutional neural network treats all features equally due to the lack of discriminative feature enhancement mechanism, which makes the model difficult to train due to feature redundancy, thus affecting the improvement of fault diagnosis performance. In addition, the compression operation of the traditional channel attention mechanism may cause a large amount of information loss, which has a negative impact on the prediction of channel weights.

[0073] (3) In actual industrial applications, strong noise pollution and variable working conditions are inevitable. Although there are a large number of algorithms for noise resistance, most of them are only applicable to noise environments with a signal-to-noise ratio greater than -4dB and are not applicable to more severe noise environments. Secondly, due to the obvious differences in signal distribution under different working conditions, the domain adaptability of the diagnosis method is poor, making it difficult to apply in actual engineering environments. Finally, these two scenarios often do not appear independently in actual industry, but there are very limited models that can simultaneously achieve robust noise resistance and cross-condition diagnosis.

[0074] (4) Deep learning-based fault diagnosis models are considered “black box” models due to their lack of interpretability. This “black box” nature leads to a lack of explanation of the model’s internal decision-making process, thus limiting its application in industries that require a high degree of transparency and interpretability.

[0075] In order to solve the above problems and enhance the model's feature extraction capability, strong noise resistance, domain adaptability and interpretability, the embodiment of the present invention proposes a fault diagnosis and system based on multi-scale dual feature enhancement based on a discriminative feature enhancement mechanism. The improvements of the embodiment of the present invention are mainly reflected in the following aspects:

[0076] (1) A multi-scale feature learning strategy was established. First, a wide convolution module was introduced to screen the features of the original signal and obtain long-term features from the vibration signal, thereby capturing low-frequency information as much as possible while reducing the impact of noise. Then, a three-branch multi-scale structure was set up to extract multi-scale features of different fine-grainedness from the non-stationary signal. Convolution kernels of different sizes learn fault features at different scales in parallel. The features extracted from multiple different receptive fields have the ability to capture global and local information, thereby obtaining rich fault feature representation information.

[0077] (2) A discriminative feature enhancement mechanism (DFEM) is proposed. Since the collected vibration signals often contain irrelevant information such as environmental noise and rotation speed, in order to make the model focus on learning features that are strongly related to diagnostic identification information and improve the fault classification effect, a discriminative feature enhancement mechanism is introduced and embedded in the backbone network to improve the accuracy of bearing fault diagnosis. The mechanism is composed of a cascade of a multi-spectral channel attention mechanism (MSCAM) and a dual-path excitation attention mechanism (DPEAM).

[0078] ①Multi-spectral channel attention mechanism (MSCAM)

[0079] Compared with the traditional channel attention module, the proposed channel attention module is designed from the perspective of spectrum to avoid the loss of a large amount of information due to the compression operation of the traditional channel feature representation, which leads to the lack of diversity in the channel feature representation. The time-frequency analysis method is used to convert the time series signal into the frequency domain, and multiple frequency components are used to enhance the channel feature representation, while solving the problem that the time series signal features are easily masked by noise. Secondly, the different frequency components of different channels are captured through the attention mechanism to obtain a more refined channel attention inference. Compared with the traditional channel attention mechanism, this method expands the traditional channel feature representation from the global average pooling (GAP) compression channel to a single scalar to multiple frequency components, thereby significantly enhancing the channel fault characterization capability and helping to improve the diagnostic capability of the model.

[0080] ②Dual Path Encouragement Attention Mechanism (DPEAM)

[0081] Since the pulse excitation segments of the time series signal in the time dimension reflect the fault characteristics, these pulse excitation segments are the key to fault diagnosis. Therefore, it is necessary to enhance the signal segments where the fault characteristics are located to improve the accuracy and efficiency of diagnosis. The dual-path excitation attention mechanism (DPEAM) proposed in the embodiment of the present invention aims to help the model focus on learning the pulse excitation segments closely related to the fault through a dual-branch structure, accurately locate the fault impact segment, and also improve the interpretability of the network feature learning process.

[0082] In summary, the discriminative feature enhancement mechanism (DFEM) takes the time series signal as input and makes full use of the time domain and frequency domain information of the fault signal. It can not only pay attention to the interdependence between channels, but also extract the important information of the signal pulse excitation part on the time axis, thereby effectively reducing the impact of irrelevant information such as noise and working condition changes on the target task. At the same time, the introduction of the attention mechanism improves the interpretability of the neural network feature learning mechanism.

[0083] (3) Considering the plug-and-play nature of the attention mechanism, the discriminative feature enhancement mechanism of the present invention can be seamlessly embedded into other existing neural network architectures to enhance the network's extraction and learning of discriminative features, and by extracting feature weights, visualize the key activation areas that play a role in the model when performing the final task, thereby enhancing the interpretability of the model.

[0084] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0085] The embodiment of the present invention proposes a fault diagnosis method based on multi-scale dual feature enhancement, such as Figure 1 As shown, the method includes:

[0086] S1. Acquire vibration signals of rolling bearings with several different fault defects in a certain period of time;

[0087] S2. Preprocessing a number of vibration signals to obtain a data set;

[0088] S3. Training and optimizing a pre-built multi-scale convolutional neural network based on a discriminative feature enhancement mechanism through a data set to obtain a fault recognition model, wherein the fault recognition model is used to classify fault defects of rolling bearings;

[0089] Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

[0090] The following is a detailed description of each step:

[0091] In one embodiment, S1, obtaining vibration signals of rolling bearings with several different fault defects in a certain period of time. Specifically including:

[0092] Fault defects are introduced artificially at different positions of the rolling bearing (common and easily damaged positions such as the outer ring, inner ring, rolling element, etc.), and an acceleration sensor is installed at the signal collection point of the equipment to collect vibration signals over a certain period of time, and the signal is uploaded to a computer.

[0093] In one embodiment, S2, pre-processing a plurality of vibration signals to obtain a data set.

[0094] Specifically include:

[0095] S201, preprocessing a number of vibration signals for data enhancement, including:

[0096] Considering that the collected signal is of a short period of time, the amount of data is relatively small, and the deep learning network often requires a massive amount of data for model training, the embodiment of the present invention adopts random starting point continuous sampling technology for data expansion.

[0097] S202 Data normalization: In order to speed up the convergence of the network, the expanded data is normalized using z-score.

[0098] S202, dividing the data set. The normalized total sample set is randomly divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set and the validation set are used to prepare for training the neural network, and the test set is used to test the performance of the model.

[0099] In one embodiment, S3, a multi-scale convolutional neural network with a pre-constructed discriminative feature enhancement mechanism is trained and optimized through a data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of rolling bearings. Specifically, it includes:

[0100] In the embodiment of the present invention, the specific structure of the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism is as follows: Figure 2 As shown, it includes feature selection module, multi-scale feature extraction module, feature fusion and output module. In addition, the discriminative feature enhancement mechanism is seamlessly embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, wherein the discriminative feature enhancement mechanism is composed of a cascade of a multi-spectral channel attention mechanism (MSCAM) and a dual-path excitation attention mechanism (DPEAM). It can be expressed as:

[0101] X DFEM =F DPEAM (F MSCAM (X)

[0102] In the embodiment of the present invention, the multi-spectral channel attention mechanism (MSCAM) is used to improve the problem that the channel feature representation of the traditional attention mechanism lacks diversity and ignores the local receptive field from the global receptive field, thereby limiting the model's ability to perceive subtle local differences in fault characteristics. The specific structure of the multi-spectral channel attention mechanism is shown in 3. First, the time series signal input is converted into the frequency domain through Fourier transform, and then the frequency domain components are divided and weighted. Specifically including:

[0103] Assume that the input feature X is represented as X = [X 0 , X 1 , ..., X c-1 ],in i∈{0, 1, …, c-1}, N is the signal length. First, convert the time series input into the frequency domain:

[0104]

[0105] Where FFT(·) is the FFT function, is the jth frequency component on the ith channel.

[0106] Then, the frequency components of each channel are divided into three parts: low frequency band, medium frequency band and high frequency band according to the size of the frequency components, and weighted:

[0107]

[0108]

[0109] Among them, σ is the sigmoid function, f(·) is the ReLU activation function, and Conv(·) is the convolution operation. By weighting the different frequency bands of each channel, fine-grained adjustment and capture of frequency domain features on different channels are achieved, and the contribution of each frequency band to the final task is dynamically learned from the local receptive field, resulting in a richer and more accurate channel feature representation.

[0110] Then the weights of each frequency band are combined to achieve channel recalibration.

[0111]

[0112]

[0113]

[0114] in, is the final weight of each channel, It is element-by-element multiplication, and the weight W ci Multiplying with the corresponding frequency component to obtain the recalibrated frequency characteristic X(F i), and the time-series recalibration feature vector is obtained through inverse Fourier transform IFFT(·) This step combines the weights of each frequency band to adjust the importance of each channel, so that the model can more effectively process and emphasize important frequency components, thereby improving the signal processing effect. The weighted frequency domain representation is converted to the time domain through the inverse Fourier transform layer, thereby adaptively enhancing or suppressing specific feature channels and improving the network's ability to respond to key features.

[0115] Finally, using the idea of ​​residual learning, residual connections are introduced to enhance the training stability of the model while retaining the original information.

[0116]

[0117] In MSCAM, fine-grained adjustment of cross-channel frequency characteristics is achieved through time-frequency transformation, frequency division weighting and weight synthesis to adaptively recalibrate the characteristic response of the channel.

[0118] The dual-path excitation attention mechanism of the embodiment of the present invention can help the network accurately locate the fault segment, focus on learning the pulse excitation segment closely related to the fault, effectively improve the efficiency and reliability of feature learning, and also improve the interpretability of the network feature learning process.

[0119] The specific structure of the dual-path incentive attention mechanism is as follows Figure 4 As shown in the figure, first, 1*1 convolution is used to obtain the spatial features of channel compression. Secondly, a dual-branch convolution is set up to use convolution kernels of different sizes to capture fault features at different scales, so that the model can focus on local details and global structures at the same time, enhance the feature representation of pulse excitation fragments, and improve the expression ability of the model and the positioning accuracy of the excitation fragments. Then, the weight is multiplied by the corresponding position of the original input to obtain the weighted excitation pulse signal. Finally, the residual connection is used to add the original input and the weighted excitation pulse signal, retaining the relevant information between the time series data. The specific steps are as follows:

[0120] Assume the input feature X MSCAM =[X 0 , X 1 , …, X n-1 ],in Represents the jth position on the timing signal. The basic steps are as follows:

[0121] Step 1: Perform convolution operation along the channel domain

[0122] X′=f(Conv 1×1 (X MSCAM ))

[0123] Step 2: Two-branch convolution operation

[0124] X″=Conv 1×1 (concat[f(Conv 3×1 (X′)), f(Conv 7×1 (X′))])

[0125] Step 3: Timing Recalibration

[0126] W s =σ(X″)

[0127]

[0128] where s∈{0, 1, …, N-1},

[0129] Step 4: Residual Connection

[0130] X DPEAM =X MSCAM +X N

[0131] In the fault identification model, the process of processing the vibration signal includes: firstly using the wide convolution module to screen the features from the original vibration signal in the feature selection module, which can be expressed as:

[0132] x′ i =P(f(BN(Conv(x i ))))

[0133] Then, in the multi-scale feature extraction module, a multi-scale convolution module with a three-branch parallel structure is used for feature extraction. The discriminative feature enhancement mechanism is seamlessly embedded behind the wide convolution module in the feature selection module and each multi-scale convolution module in the multi-scale feature extraction module to adaptively enhance the fault features. Secondly, the purpose of feature fusion is to learn and fuse complementary fault features of different scales. However, simple splicing cannot completely fuse these complementary features, and redundant features are not conducive to model learning. Therefore, in the feature fusion and output module, MSCAM is used to adaptively adjust the features fused along the fusion channel, and then the GAP (Global Average Pooling) layer is used to reduce the feature parameters after adaptive fusion to prevent model overfitting to a certain extent. Finally, the output of the fully connected layer is connected to the Softmax function to realize fault classification.

[0134] In the embodiment of the present invention, the training of the fault identification model is carried out under the Windows 11 operating system, the CPU is i7-137000k, the frequency is 3.40GHz, and the GPU is NVIDIA RTX 4060. The code is written in Python 3.8 and PyTorch. In order to enhance the reliability of the experimental results and reduce the impact of randomness, all experiments were carried out 10 times. (The cross entropy loss function is used in the training process, the Aadm optimizer is selected to update the model parameters, the number of training rounds is set to 100, the batch size is 32, the learning rate is 0.001, and the learning rate training strategy is set to be whenever the loss on the validation set does not decrease in 10 consecutive iterations, the learning rate will be reduced to 0.1 of the current value. The training set and validation set data in step S2 are input into the model for training. During the training process, the model parameters are updated by the principle of gradient back propagation, and the accuracy of the model on the validation set and the training set is evaluated, and the model with the highest accuracy in the validation set greater than the training set is saved.

[0135] The embodiment of the present invention further provides a fault diagnosis system based on multi-scale dual feature enhancement, comprising:

[0136] A data acquisition module, used to acquire vibration signals of rolling bearings including several different fault defects in a certain period of time;

[0137] A preprocessing module, used to preprocess a number of vibration signals to obtain a data set;

[0138] A model training module is used to train and optimize a multi-scale convolutional neural network with a pre-built discriminative feature enhancement mechanism through a data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of the rolling bearing;

[0139] Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

[0140] It can be understood that the fault diagnosis system based on multi-scale dual feature enhancement provided in the embodiment of the present invention corresponds to the above-mentioned fault diagnosis method based on multi-scale dual feature enhancement. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the fault diagnosis based on multi-scale dual feature enhancement, and will not be repeated here.

[0141] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program for fault diagnosis based on multi-scale dual feature enhancement, wherein the computer program enables a computer to execute the fault diagnosis method based on multi-scale dual feature enhancement as described above.

[0142] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the program includes a method for performing fault diagnosis based on multi-scale dual feature enhancement as described above.

[0143] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0144] 1. The embodiment of the present invention proposes a multi-scale learning strategy to improve the model fault diagnosis accuracy by extracting rich fault feature information at different scales. The specific strategy is as follows: First, a wide convolution module is set to perform feature screening on the original vibration signal. Compared with the traditional small convolution kernel, this setting captures features on a longer time scale and better suppresses noise interference. Secondly, three-branch filters with different kernel sizes are used to extract feature information of different levels of granularity in parallel at different spatial scales to improve the network model's perception of features of different sizes.

[0145] 2. The embodiment of the present invention aims at the problem that the traditional convolutional neural network treats the features under different channels and scales equally, but fails to emphasize the role of key features in the final classification task, resulting in a large number of redundant features affecting the model training, thereby affecting the fault diagnosis performance. A discriminative feature enhancement mechanism is proposed to adaptively enhance the fault features and weaken the meaningless features, thereby giving the deep learning model the ability to intelligently perceive the fault features. First, from the perspective of the frequency domain, the channel attention is expanded from the original GAP method to compress the channel into a scalar to multiple frequency components. While enriching the feature representation, the model's ability to perceive subtle local differences in fault features is improved with the local receptive field, and the importance of different channels is adaptively adjusted. Secondly, the feature learning ability of the fault pulse is enhanced by accurately locating the fault pulse segment of the time series signal. Finally, although the module takes the original time series signal as input, it realizes the fusion and enhancement of different fault features more comprehensively from the time domain and frequency domain, so that the model can adaptively focus on learning more discriminative features and ignore useless features.

[0146] 3. The embodiment of the present invention addresses the problem that the deep learning model lacks interpretability, which not only reduces the credibility of intelligent fault results, but also limits its application in fault diagnosis with high reliability requirements. By visualizing the attention weight distribution of the proposed discriminative feature mechanism in different channels and different time sequences, the feature extraction process of the proposed model is successfully decoded, and an intuitive and physically interpretable result is obtained, thereby improving the learning ability and interpretability of the neural network.

[0147] 4. The discriminative feature enhancement mechanism proposed in the embodiment of the present invention can be seamlessly embedded into the existing deep learning model as a plug-and-play module to improve performance, and provides an effort to explore the inherent mechanism of the model in feature learning and classification. Experiments have also proved that it is worthwhile to embed the module into the existing model, because the module only requires a small amount of parameters and computational overhead, and the model embedded in the module can capture the key features required to achieve the target task, while ignoring irrelevant features, and stably and effectively improve the diagnostic ability of the model.

[0148] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0149] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. 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 fault diagnosis based on multi-scale dual feature enhancement, characterized in that: include: Acquire vibration signals of rolling bearings including several different fault defects in a certain period of time; Preprocessing several vibration signals to obtain a data set; The multi-scale convolutional neural network with a pre-constructed discriminative feature enhancement mechanism is trained and optimized through the data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of the rolling bearing; Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

2. The fault diagnosis based on multi-scale dual feature enhancement as claimed in claim 1, characterized in that: The feature selection module includes a wide convolution module, which is used to perform feature screening on the vibration signal, and the expression is: x′ i =P(f(BN(Conv(x i )))) Among them, P(·) is the maximum pooling operation, BN(·) is the batch normalization operation; f(·) is the ReLU activation function, Conv(·) is the convolution operation, and x i is the ith vibration signal, x′ i Represents the features filtered out after the above operations.

3. The fault diagnosis based on multi-scale dual feature enhancement as claimed in claim 1, characterized in that: The multi-scale feature extraction module includes convolution modules of different scales with a three-branch parallel structure.

4. The fault diagnosis based on multi-scale dual feature enhancement as claimed in claim 1, characterized in that: The feature fusion and output module includes a fusion channel, a global average pooling layer, and a fully connected layer; in the feature fusion and output module, a multi-spectral channel attention mechanism is used to adaptively adjust the features fused along the channel; The adjusted fusion features are processed by reducing feature parameters through the global average pooling layer. The output of the global average pooling layer is input to the fully connected layer. The output of the fully connected layer is processed by Softmax. x The function outputs the fault classification result.

5. The fault diagnosis based on multi-scale dual feature enhancement as claimed in claim 1, characterized in that: The expression of the discriminative feature enhancement mechanism is: X DFEM =F DPEAM (F MSCAM (X)) X DFEM represents the enhanced features after the discriminative feature mechanism, F MSCAM (·) indicates that the data is processed by the multi-spectral channel attention mechanism, F DPEAM (·) indicates that the data is processed by the dual-path incentivized attention mechanism.

6. The fault diagnosis based on multi-scale dual feature enhancement as claimed in any one of claims 1 to 5, characterized in that: The data processing process in the multi-spectral channel attention mechanism includes: The time series signal input is converted into the frequency domain through Fourier transform, which includes: The input feature X is represented as X = [X 0 , X 1 , …, X c-1 ],in i∈{0, 1, …, c-1}, N is the signal length, convert the time series input into the frequency domain: Where FFT(·) is the FFT function, is the jth frequency component on the i-th channel; The frequency domain components are divided into blocks and weighted, including: According to the size of the frequency component, the frequency components of each channel are divided into three parts: low frequency band, medium frequency band and high frequency band, and the weight of each frequency band is obtained by weighting Where σ is the sigmoid function, f(·) is the ReLU activation function, and Conv(·) is the convolution operation; represents the weight of the kth frequency band on the ith channel, k = 1, 2, 3, representing the low frequency band, the middle frequency band, and the high frequency band respectively; The weights of each frequency band are combined to achieve channel recalibration, including: in, is the final weight of each channel, It is element-by-element multiplication, and the weight W ci Multiplying with the corresponding frequency component to obtain the recalibrated frequency characteristic X(F i ), and the time-series recalibration feature vector is obtained through inverse Fourier transform IFFT(·) Using the idea of ​​residual learning, we introduce residual connections to obtain the multi-spectral channel attention feature X MSCAM :

7. The fault diagnosis based on multi-scale dual feature enhancement as claimed in any one of claims 1 to 5, characterized in that: The data processing process in the dual-path incentive attention mechanism includes: The feature X adaptively weighted by the channel MSCAM First, perform convolution operation along the channel domain to obtain the convolution feature X′: X′=f(Conv 1×1 (X MScAM )) Among them, X MSCAM =[X 0 , X 1 , …, X n-1 ],in Indicates the jth position on the timing signal; Then perform a dual-branch convolution operation on the feature X′ to obtain the dual-branch convolution feature X″: X″=Conv 1×1 (concat[f(Conv 3×1 (X′)),f(Conv 7×1 (X′))]) Perform time recalibration on the dual convolution feature X" to obtain the time recalibration feature vector X N : W s =σ(X″) Among them, s∈{0, 1,…, N-1}, The time-order recalibration feature vector is residually connected to the input feature to obtain the dual-path incentive attention X DPEAM : X DPEAM =X MSCAM +X N 。 8. A fault diagnosis system based on multi-scale dual feature enhancement, characterized in that: include: A data acquisition module, used to acquire vibration signals of rolling bearings including several different fault defects in a certain period of time; A preprocessing module, used to preprocess a number of vibration signals to obtain a data set; A model training module is used to train and optimize a multi-scale convolutional neural network with a pre-built discriminative feature enhancement mechanism through a data set to obtain a fault recognition model, wherein the fault recognition model is used to classify the fault defects of the rolling bearing; Among them, the multi-scale convolutional neural network of the pre-constructed discriminative feature enhancement mechanism includes a feature selection module, a multi-scale feature extraction module and a feature fusion and output module; wherein the discriminative feature enhancement mechanism is embedded in the feature selection module and the multi-scale feature extraction module to enhance the fault characteristics, and the discriminative feature enhancement mechanism includes a multi-spectral channel attention mechanism and a dual-path excitation attention mechanism.

9. A computer-readable storage medium, characterized in that: It stores a computer program for fault diagnosis based on multi-scale dual feature enhancement, wherein the computer program enables a computer to execute the fault diagnosis method based on multi-scale dual feature enhancement as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the fault diagnosis method based on multi-scale dual feature enhancement as described in any one of claims 1 to 7.

Images