CN114997214A

CN114997214A - Fault diagnosis method and device for residual error intensive network

Info

Publication number: CN114997214A
Application number: CN202210484462.XA
Authority: CN
Inventors: 袁彩艳; 范新安; 陈远方
Original assignee: Beijing Machinery Equipment Research Institute
Current assignee: Beijing Machinery Equipment Research Institute
Priority date: 2022-05-06
Filing date: 2022-05-06
Publication date: 2022-09-02

Abstract

The disclosure relates to a residual error intensive network fault diagnosis method, a residual error intensive network fault diagnosis device, electronic equipment and a storage medium. According to the method, an original signal, an envelope spectrum conversion result and a discrete cosine S conversion result are combined into a multi-channel signal containing multi-domain information of a time domain, a frequency domain and a time-frequency domain, and then the multi-channel signal is input into a residual error intensive model for deep learning, so that hidden fault information is better mined; considering that the number of features increases along with the increase of the network depth and the difference exists between different features in information transmission, the invention provides the steps of performing attention mechanism distribution on learned features, realizing the difference division of the features and finally performing fault identification by utilizing softmax. The method can extract more comprehensive fault characteristics, and has better fault identification effect compared with the traditional method.

Description

Fault diagnosis method and device for residual error intensive network

Technical Field

The present disclosure relates to the field of fault diagnosis, and in particular, to a method and an apparatus for fault diagnosis of a residual error intensive network, an electronic device, and a computer-readable storage medium.

Background

The bearing is a key part of mechanical transmission equipment, loss and influence caused by faults of the bearing are huge, and the bearing fault diagnosis and early warning are significant.

With the great effect of the deep Convolutional network (CNN) in the field of image processing and the like, researchers have introduced it into the field of fault diagnosis and have conducted intensive research. Yang par et al propose to learn the raw data and implement fault classification by using a convolution capsule network for the rolling bearing fault. Lu and the like map original monitoring signals into a two-dimensional matrix in a sectional combination mode, and the fault diagnosis of the bearing is realized through the automatic learning high-dimensional feature representation of a deep CNN network. Ince et al propose to use one-dimensional CNN to extract and classify features together to complete real-time bearing fault diagnosis. Li et al propose a bearing fault diagnosis model integrating a deep neural network and a CNN to complete fault diagnosis of a bearing.

Through analysis of documents, the current diagnostic method based on the deep neural network mostly realizes the improvement of diagnostic capability by constructing a complex network model, continuously deepening the network depth and increasing various treatments among layers, although the diagnostic accuracy is improved, the deep structure also shows a plurality of problems along with the increase of the number of network layers: on one hand, the problem of gradient dispersion or explosion is easily caused, so that the network training is very difficult; in addition, the depth structure causes network degradation, loss in training does not decrease or increase reversely, training errors of samples are increased, and recognition accuracy is reduced; and the training of the large-scale deep network consumes a large amount of system resources, and the problems greatly limit the deep development of the CNN in the field of fault diagnosis.

Accordingly, there is a need for one or more methods to address the above-mentioned problems.

It is to be noted that the information disclosed in the above background section is only for enhancement of understanding of the background of the present disclosure, and thus may include information that does not constitute prior art known to those of ordinary skill in the art.

Disclosure of Invention

An object of the present disclosure is to provide a residual error intensive network fault diagnosis method, apparatus, electronic device, and computer readable storage medium, thereby overcoming, at least to some extent, one or more of the problems due to the limitations and disadvantages of the related art.

According to an aspect of the present disclosure, there is provided a residual error intensive network fault diagnosis method, including:

a data preprocessing step, namely acquiring a detection signal of a bearing, performing discrete cosine S transform and envelope spectrum extraction on the detection signal to generate a multi-domain detection signal, and superposing the multi-domain detection signal and the detection signal respectively to generate a multi-channel signal which is used as the input of a bearing fault diagnosis model;

extracting fault characteristics, namely extracting fault characteristics in the multi-channel signal from a bearing fault diagnosis model of a residual error dense network based on a multi-stage residual error dense block;

weighting the fault characteristics, namely weighting the weights of the fault characteristics according to the channel attention and the space attention of a convolution attention module in a bearing fault diagnosis model;

and a fault diagnosis step, namely classifying the attention-sequenced fault characteristics based on a softmax function so as to complete the identification and classification of the fault mode.

In an exemplary embodiment of the present disclosure, the data preprocessing step of the method further includes:

collecting detection signals of a bearing, performing discrete cosine S transformation and envelope spectrum extraction on the detection signals to generate multi-domain detection signals, and respectively superposing the multi-domain detection signals and the detection signals to generate a multi-channel signal x _i And extracting features by using convolution kernels with the step length of 1, 1 × 3 to obtain shallow vector features:

wherein σ _r For Relu activation function, θ ^inp ＝{k ^inp ，b ^inp And the parameters are parameters to be trained of the bearing fault diagnosis model.

In an exemplary embodiment of the present disclosure, the fault feature extraction step of the method further includes:

in a bearing fault diagnosis model of a residual dense network based on a multi-stage residual dense block, a multichannel signal X is input to the residual dense network in the bearing fault diagnosis model, low-level features fea (X) are extracted through convolution 1, fea (X) are extracted through convolution 2 to high-level features fea1(X), the low-level features fea (X) and the high-level features fea1(X) are connected in parallel to realize dense connection to obtain fea2(X), convolution 3 performs linear operation on the input X to obtain fea3(X), fea2(X) and fea3(X) form a residual block structure through series connection, and fault features F in the output multichannel signal are as follows:

y＝{σ _r [σ _r (w ₁ *x+b ₁ )*w ₂ +b ₂ ∪σ _r (w ₁ *x+b ₁ )}+{σ _r (w ₃ *x+b ₃ )}

wherein, w ₁ ，b ₁ ，w ₂ ，b ₂ ,w ₃ ，b ₃ Parameters to be trained, σ, for convolution 1, convolution 2 and convolution 3, respectively _r Is the activation function Relu.

In an exemplary embodiment of the disclosure, the weighting the weight of the fault feature according to the channel attention and the spatial attention of the convolution attention module in the bearing fault diagnosis model in the attention ranking step of the method includes:

aggregating spatial information of the feature graph based on the global average pool and the global maximum pool operations, generating a spatial context descriptor for the input feature F:

and

inputting the characteristics of the two pooling layers into a shared network for performing ascending and descending dimensional operation, wherein the shared network consists of a plurality of layers of sensors and a hidden layer;

the two output feature graphs are summed element by element to combine output feature vectors, and the feature vectors are activated based on a sigmoid functionObtaining the channel attention weight coefficient M _c (F)：

Wherein: w is a group of ₀ ∈R ^c/r×c ，W ₁ ∈R ^c×c/r ，W ₀ Relu is used as an activation function, and sigma is a Sigmoid function;

the channel attention weight coefficient M _c (F) And multiplying the input feature F element by element to obtain a refined feature F'.

taking the refined feature F' output after the channel attention processing as the input of a space attention processing part, and generating the channel with the number of 1 based on the channel information of the Avgpool and Maxpool operation aggregation feature diagram

And

will be described in

And

merging the feature maps into a feature map with 2 channels, and compressing the feature map in channel dimension by a convolution kernel with the step length of 2 and the size of 1 × 7, wherein the number of the compressed features is 1;

feature vector is activated based on sigmoid function to obtain spatial attention coefficient M _s (F′)：

Where σ denotes a sigmoid function, f ^1*7 Represents a convolution of size 1 x 7;

weighting the spatial attention by a factor M _s (F ') element-by-element multiplication is performed on the input features F', and finally the spatial attention graph is generated.

In an exemplary embodiment of the present disclosure, the method further comprises:

extracting fault features in the multi-channel signal based on a bearing fault diagnosis model of a residual dense network of multi-level residual dense blocks

As inputs, the residual dense unit based on CBAM yields the output:

in the formula, w ₁ ，b ₁ ，w ₂ ，b ₂ ,w ₃ ，b ₃ The parameters to be trained are convolution 1, convolution 2 and convolution 3, respectively.

In an exemplary embodiment of the present disclosure, the method further includes a bearing fault diagnosis model training step of:

gradually updating the weights and the deviations based on the gradient function;

based on a back propagation algorithm, the method is used for monitoring the learning process and adjusting parameters of each layer, and calculating the error of each network iteration;

evaluating the errors of the estimated probability distribution and the target probability distribution by using a cross entropy loss function;

the cross entropy loss function of each training is obtained by forward calculation, and the network parameters are updated by using an error back propagation algorithm until the maximum training steps are reached;

and reducing the value of a cross entropy loss function in the training process based on a small-batch random gradient descent optimization algorithm, so that the estimated distribution and the target distribution are approximate to improve the prediction precision of the model.

In one aspect of the present disclosure, there is provided a residual error intensive network fault diagnosis apparatus including:

the data preprocessing module is used for acquiring detection signals of the bearing, performing discrete cosine S transform and envelope spectrum extraction on the detection signals to generate multi-domain detection signals, and superposing the multi-domain detection signals and the detection signals respectively to generate multi-channel signals which are used as input of a bearing fault diagnosis model;

the fault feature extraction module is used for extracting fault features in the multi-channel signals in a bearing fault diagnosis model of a residual error dense network based on the multi-stage residual error dense blocks;

the fault characteristic weighting module is used for weighting the weight of the fault characteristic according to the channel attention and the space attention of the convolution attention module in the bearing fault diagnosis model;

and the fault diagnosis module is used for classifying the attention-sequenced fault characteristics based on the softmax function so as to complete the identification and classification of the fault mode.

In one aspect of the present disclosure, there is provided an electronic device including:

a processor; and

a memory having computer readable instructions stored thereon which, when executed by the processor, implement a method according to any of the above.

In an aspect of the disclosure, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, realizes the method according to any one of the above.

In the exemplary embodiment of the disclosure, the method for diagnosing the fault of the residual error dense network comprises the steps of forming an original signal, an envelope spectrum conversion result and a discrete cosine S conversion result into a multi-channel signal containing multi-domain information of a time domain, a frequency domain and a time-frequency domain, and then inputting the multi-channel signal into a residual error dense model for deep learning to better mine hidden fault information; considering that the number of features increases along with the increase of the network depth and the difference exists between different features in information transmission, the invention provides the steps of performing attention mechanism distribution on learned features, realizing the difference division of the features and finally performing fault identification by utilizing softmax. The method can extract more comprehensive fault characteristics, and has better fault identification effect compared with the traditional method.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The above and other features and advantages of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

FIG. 1 shows a flow diagram of a method of residual error intensive network fault diagnosis according to an example embodiment of the present disclosure;

2A-2D illustrate a step-by-step algorithmic flow chart of a method of residual error intensive network fault diagnosis according to an exemplary embodiment of the present disclosure;

3A-3B illustrate a classification statistics and recognition accuracy comparison graph of a residual error intensive network fault diagnosis method according to an exemplary embodiment of the present disclosure;

FIG. 4 shows a schematic block diagram of a residual error intensive network fault diagnostic apparatus according to an exemplary embodiment of the present disclosure;

FIG. 5 schematically shows a block diagram of an electronic device according to an exemplary embodiment of the present disclosure; and

fig. 6 schematically illustrates a schematic diagram of a computer-readable storage medium according to an exemplary embodiment of the present disclosure.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The same reference numerals denote the same or similar parts in the drawings, and thus, a repetitive description thereof will be omitted.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the disclosure. One skilled in the relevant art will recognize, however, that the embodiments of the disclosure can be practiced without one or more of the specific details, or with other methods, components, materials, devices, steps, and so forth. In other instances, well-known structures, methods, devices, implementations, materials, or operations are not shown or described in detail to avoid obscuring aspects of the disclosure.

The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in the form of software, or in one or more software-hardened modules, or in different networks and/or processor devices and/or microcontroller devices.

In the present exemplary embodiment, a method for diagnosing a fault of a residual error intensive network is first provided; referring to fig. 1, the method for diagnosing a fault of a residual error intensive network may include the steps of:

a data preprocessing step S110, collecting detection signals of the bearing, carrying out discrete cosine transform and envelope spectrum extraction on the detection signals to generate multi-domain detection signals, and superposing the multi-domain detection signals and the detection signals respectively to generate multi-channel signals which are used as the input of a bearing fault diagnosis model;

a fault feature extraction step S120, wherein fault features in the multichannel signals are extracted from a bearing fault diagnosis model of a residual error dense network based on a plurality of stages of residual error dense blocks;

a fault feature weighting step S130, wherein the weights of the fault features are weighted according to the channel attention and the space attention of a convolution attention module in the bearing fault diagnosis model;

and a fault diagnosis step S140, classifying the attention-ranked fault features based on a softmax function so as to complete the identification and classification of the fault modes.

Next, a residual error intensive network fault diagnosis method in the present exemplary embodiment will be further described.

The first embodiment is as follows:

in the data preprocessing step S110, a detection signal of the bearing may be acquired, discrete cosine transform and envelope spectrum extraction are performed on the detection signal to generate a multi-domain detection signal, and the multi-domain detection signal and the detection signal are respectively superimposed to generate a multi-channel signal as an input of the bearing fault diagnosis model.

In an embodiment of the present example, the data preprocessing step of the method further comprises:

In the fault feature extraction step S120, a fault feature in the multi-channel signal may be extracted in a bearing fault diagnosis model based on a residual dense network of multi-level residual dense blocks.

In the embodiment of the present example, the fault feature extraction step of the method further includes:

in a bearing fault diagnosis model of a residual error dense network based on a multi-stage residual error dense block, a multi-channel signal X is input into the residual error dense network in the bearing fault diagnosis model, low-level features fea (X) are extracted through convolution 1, fea (X) is extracted through convolution 2 to high-level features fea1(X), the low-level features fea (X) and the high-level features fea1(X) are connected in parallel to realize dense connection to obtain fea2(X), convolution 3 performs linear operation on the input X to obtain fea3(X), fea2(X) and fea3(X) form a residual block structure through series connection, and fault features F in the multi-channel signal are output:

wherein, w ₁ ，b ₁ ，w ₁ ，b ₂ ,w ₃ ，b ₃ Parameters to be trained, σ, for convolution 1, convolution 2 and convolution 3, respectively _r Is the activation function Relu. A large number of features can be extracted through the stacked multi-stage residual error dense blocks, but the features do not contain equally important fault information, so that the method introduces an attention mechanism to process the features, and realizes the differential division of the features.

In the embodiment of the present example, in order to effectively learn the fault features from the multichannel signal, the core of the model is to construct a deep one-dimensional network, and learn and abstract the signal features layer by layer, but simply stack the residual blocks to form a chain structure, and it is likely that the learned effective fault features will be lost during the transmission process of the deep network, the present invention proposes that the residual learning modules are connected in a dense manner to form a residual dense structure as shown in fig. 2A. The residual error dense block in the graph is composed of a plurality of convolution layers, batch standardization, an activation function and quick connection. The method improves the information flow through the network and relieves the problem of gradient disappearance; and because the previous convolution mapping is reused, the re-learning of redundant characteristics is avoided, and abundant vibration signal details are provided for the later layer.

In the step S130 of weighting the fault characteristics, the weights of the fault characteristics may be weighted according to the channel attention and the spatial attention of the convolution attention module in the bearing fault diagnosis model.

In the embodiment of the present example, for a large number of features obtained after residual intensive network learning, in order to improve the classification efficiency of subsequent networks, a Convolutional Attention Module (CBAM) is introduced to implement feature optimization selection. The CBAM is mainly composed of series channel attention and spatial attention, and achieves an adaptive refinement effect by redistribution and positioning of feature weights, and the structure of the CBAM is shown in fig. 2B.

In this exemplary embodiment, in the attention ranking step of the method, weighting the weight of the fault feature according to the channel attention and the spatial attention of the convolution attention module in the bearing fault diagnosis model includes:

and

inputting the characteristics of the two pooling layers into a shared network for performing ascending and descending dimensional operation, wherein the shared network consists of a plurality of layers of sensors and a hidden layer; to reduce parameter overhead, the activation size of the concealment is set to R ^c/r×1×1 Where r is the compressibility.

The two output feature graphs are summed element by element to combine output feature vectors, and the feature vectors are activated based on a sigmoid function to obtain a channel attention weight coefficient M _c (F)：

Wherein: w ₀ ∈R ^c×r/c ，W ₁ ∈R ^c×c/r ，W ₀ Relu is used as an activation function, and sigma is a Sigmoid function;

In this exemplary embodiment, the step of attention ranking of the method, the weighting the weight of the fault feature according to the channel attention and the spatial attention of the convolution attention module in the bearing fault diagnosis model includes:

spatial attention discrimination focuses attention on spatial positions of features, and different weights are given to position information of the features, so that a network learns feature information useful for classification tasks according to weight distribution. Taking the refined feature F' output after the channel attention processing as the input of a space attention processing part, and generating a channel number of 1 based on channel information of an Avgpool and Maxpool operation aggregation feature map

And

will be described in

And

merging into 2-channel characteristic diagram, and performing channel dimension by a convolution kernel with step size of 2 and size of 1 × 7Compressing, wherein the number of the compressed characteristic channels is 1;

activating the feature vector based on sigmoid function to obtain spatial attention coefficient M _s (F′)：

weighting the spatial attention by a factor M _s (F ') element-by-element multiplication is carried out on the input features F', and finally the space attention is generated. By distinguishing the importance of the channel and the position of the features, the difference learning of the features in subsequent processing is realized, and the learning efficiency is improved.

In an embodiment of the present example, the method further comprises:

As inputs, the residual dense unit based on CBAM yields the output:

In the present exemplary embodiment, CBAM can be aggregated in principle into any CNN architecture, achieving good attention discrimination. On the basis of fully considering the functions and the functions of the residual error dense unit, the invention provides that the CBAM is added into the last-stage residual error dense unit to realize the differential division of a large number of characteristics.

The characteristics extracted after the multi-stage residual error intensive unit are set as

Then it is input into the residual dense unit with CBAM to get the output:

In order to reduce the parameters to be trained, the characteristics obtained by each level of residual error intensive unit

And derived features with CBAM residual dense cells

Dividing the data into a plurality of non-overlapping sections, and returning the maximum value of each section of elements, namely the maximum pooling process, namely:

in the formula (I), the compound is shown in the specification,

as feature vectors

J-th row and n columns of elements; s is the length of the non-overlapping segment, here taken to be 2.

In the fault diagnosis step S140, the attention-ranked fault features may be subjected to a classification process based on the softmax function to complete the identification and classification of the fault pattern.

In the exemplary embodiment, as shown in fig. 4, the process of bearing fault diagnosis, the network gradually updates the weights and the deviations according to the gradient function during training, and the back propagation algorithm is used to monitor the learning process and adjust the parameters of each layer, and calculate the error of each network iteration. And evaluating the errors of the estimated probability distribution and the target probability distribution by using a cross entropy loss function. The cross entropy loss function of each training is obtained by forward calculation, the network parameters are updated by using an error back propagation algorithm until the maximum training step number, and meanwhile, the value of the cross entropy loss function in the training process is reduced by using a small-batch random gradient descent optimization algorithm, so that the estimated distribution and the target distribution are closer and closer, and the prediction precision of the model is gradually improved.

In an embodiment of the present example, the method further comprises a bearing fault diagnosis model training step of:

The second embodiment:

in the embodiment of the present example, the method proposed by the present invention was verified, and experimental data of different public data sets were used for processing and analysis.

The inventive data set came from the bearing data center at the university of Keiss Cauchy, USA. The test equipment at the university of Keysuchen storage includes two motors, a coupler and other equipment. Accelerometers are used to collect signals of different fault types. The drive end bearing (DE) accelerometer is mounted on the DE housing in a vertical orientation. A fan end bearing (FE) accelerometer is mounted on the FE housing in a vertical orientation. The data sampling frequency of a bearing at the driving end is 12KHz, and four state types including a normal state, a rolling element fault, an inner ring fault and an outer ring fault are simulated, wherein the rolling element, the inner ring and the outer ring are sampled under different damage diameters (0.007 inches, 0.014 inches and 0.021 inches).

The invention defines the fault type as an inner ring fault and adjusts the working load and the damage diameter. At three lesion diameters (0.007 inch, 0.014 inch, 0.021 inch) and 3 loads (0HP,1HP,2HP), 9 classes were constructed, plus the normal operating conditions at 3 loads (0HP,1HP,2HP), for a total of 12 classes (denoted 0-11), with 100 samples for each class of fault, 80% of which were used as the training data set. 10% of the samples were used as the validation set and 10% as the test set, each sample containing 1024 points of the original time signal. The detailed description is shown in table 1.

TABLE 1 identification status and Classification tag for dataset I

In the embodiment of the present example, the network configuration parameters and the diagnosis result. The data processing is carried out on the constructed data set I for a plurality of times according to the method, the total classification accuracy is measured by adopting the ratio of the number of the classified correct samples to the total number of the tested samples, and the average identification accuracy of the results of a plurality of times of experiments is counted to be 99.24%. The specific parameters of the diagnostic model in the experiment are shown in table 2.

Table 2 network parameter table of the present invention

The convolution kernel size in the above table is 1 x 3 because the 1 x 3 convolution kernel filter is the smallest kernel for center, left/right, etc. different directional pattern capture, and the use of a small convolution kernel filter increases the non-linearity inside the network, making the network more discriminative. The number of feature maps increases at a rate of 2m (m 16) per addition of one residual dense block. And adding a pooling layer with the step size of 2 after each residual error dense block to reduce the training parameters. To avoid overfitting of the data, dropout was used after the fully-connected layer F2, with a scale of 0.5 for hierarchical training in the network.

The confusion matrix of the classification result of a certain time is shown in figure 3A, the abscissa in the figure is a predicted data label, the ordinate in the figure is a real data label, the diagonal in the figure represents the number of correctly classified samples, and the rest of each block displays the number of samples classified into other categories by errors. The method has high classification accuracy, and only a small part of samples are classified wrongly.

In the present exemplary embodiment, the method of the present invention is compared with a conventional method. The invention improves the traditional model, provides a multi-channel signal input depth network model formed by the original time domain signal and the transform domain information, introduces the CBAM into the residual dense network model, increases the diversity of data and performs self-adaptive refinement on the extracted characteristics, thereby improving the bearing fault identification accuracy. In order to verify the effectiveness of the method provided by the invention, the diagnostic effect of the method is compared with that of other more common deep network models. At present, three models which are commonly used in bearing fault diagnosis comprise a convolutional neural network, convolutional self-coding and a residual error network, wherein the convolutional self-coding has good generalization and robustness. The structure of each model used for comparison is shown in table 6 for high consistency with the structure of the inventive process.

Table 6 common deep learning model structure table

In the present exemplary embodiment, 10 replicates of each method are shown in fig. 3B.

The method has the advantages of highest average accuracy, small fluctuation, good stability and high reliability of single recognition accuracy. The statistics of the multiple experiments are shown in Table 7.

Table 7 compares the statistical results of the experiments with the usual methods

In the embodiment of the example, aiming at the problems that the original bearing data contains single information and the feature extracted by the depth model has difference in information transmission, the invention provides that a time domain, a frequency domain and a time-frequency domain multi-domain fusion signal are used as input, so that the input contains more comprehensive information; the convolution attention processing can realize the importance distinguishing of the feature weight extracted from the residual dense block. The Kaiser university data set and the XJSY data set are adopted to verify that the bearing fault diagnosis and identification accuracy can be effectively improved by the method provided by the invention.

It should be noted that although the various steps of the methods of the present disclosure are depicted in the drawings in a particular order, this does not require or imply that these steps must be performed in this particular order, or that all of the depicted steps must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions, etc.

Furthermore, in the present exemplary embodiment, a residual error intensive network fault diagnosis apparatus is also provided. Referring to fig. 4, the residual error intensive network fault diagnosis apparatus 400 may include: a data preprocessing module 410, a fault feature extraction module 420, a fault feature weighting module 430, and a fault diagnosis module 440. Wherein:

the data preprocessing module 410 is configured to collect a detection signal of a bearing, perform discrete cosine transform and envelope spectrum extraction on the detection signal, generate a multi-domain detection signal, and superimpose the multi-domain detection signal and the detection signal respectively to generate a multi-channel signal, which is used as an input of a bearing fault diagnosis model;

a fault feature extraction module 420, configured to extract a fault feature in the multichannel signal in a bearing fault diagnosis model based on a residual dense network of multi-stage residual dense blocks;

the fault feature weighting module 430 is used for weighting the weight of the fault feature according to the channel attention and the space attention of the convolution attention module in the bearing fault diagnosis model;

and the fault diagnosis module 440 is used for classifying the attention-ranked fault features based on the softmax function so as to complete the identification and classification of the fault modes.

The specific details of each of the residual error intensive network fault diagnosis device modules are already described in detail in a corresponding residual error intensive network fault diagnosis method, and therefore are not described herein again.

It should be noted that although several modules or units of a residual error intensive network fault diagnosis apparatus 400 are mentioned in the above detailed description, such division is not mandatory. Indeed, the features and functionality of two or more modules or units described above may be embodied in one module or unit, according to embodiments of the present disclosure. Conversely, the features and functions of one module or unit described above may be further divided into embodiments by a plurality of modules or units.

In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or program product. Thus, various aspects of the invention may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.

An electronic device 500 according to such an embodiment of the invention is described below with reference to fig. 5. The electronic device 500 shown in fig. 5 is only an example and should not bring any limitation to the functions and the scope of use of the embodiments of the present invention.

As shown in fig. 5, the electronic device 500 is embodied in the form of a general purpose computing device. The components of the electronic device 500 may include, but are not limited to: the at least one processing unit 510, the at least one memory unit 520, a bus 530 connecting various system components (including the memory unit 520 and the processing unit 510), and a display unit 540.

Wherein the storage unit stores program code that is executable by the processing unit 510 to cause the processing unit 510 to perform steps according to various exemplary embodiments of the present invention as described in the above section "exemplary methods" of the present specification. For example, the processing unit 510 may perform steps S110 to S140 as shown in fig. 1.

The memory unit 520 may include readable media in the form of volatile memory units, such as a random access memory unit (RAM)5201 and/or a cache memory unit 5202, and may further include a read-only memory unit (ROM) 5203.

Storage unit 520 may also include a program/utility 5204 having a set (at least one) of program modules 5203, such program modules 5205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.

Bus 550 may be any of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 500 may also communicate with one or more external devices 570 (e.g., keyboard, pointing device, Bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 500, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 500 to communicate with one or more other computing devices. Such communication may occur via input/output (I/O) interfaces 550. Also, the electronic device 500 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the internet) via the network adapter 560. As shown, the network adapter 560 communicates with the other modules of the electronic device 500 over a bus 550. It should be appreciated that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

In an exemplary embodiment of the present disclosure, there is also provided a computer-readable storage medium having stored thereon a program product capable of implementing the above-described method of the present specification. In some possible embodiments, aspects of the invention may also be implemented in the form of a program product comprising program code means for causing a terminal device to carry out the steps according to various exemplary embodiments of the invention described in the above-mentioned "exemplary methods" section of the present description, when said program product is run on the terminal device.

Referring to fig. 6, a program product 600 for implementing the above method according to an embodiment of the present invention is described, which may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited in this respect, and in an apparatus of the present invention, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

A computer readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

Furthermore, the above-described drawings are only schematic illustrations of processes involved in methods according to exemplary embodiments of the invention, and are not intended to be limiting. It will be readily appreciated that the processes illustrated in the above figures are not intended to indicate or limit the temporal order of the processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, e.g., in multiple modules.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is to be limited only by the terms of the appended claims.

Claims

1. A residual error intensive network fault diagnosis method, the method comprising:

2. The method of claim 1, wherein the data preprocessing step of the method further comprises:

3. The method of claim 1, wherein the fault feature extraction step of the method further comprises:

wherein, w ₁ ，b ₁ ，w ₂ ，b ₂ ，w ₃ ，b ₃ The parameters to be trained are convolution 1, convolution 2 and convolution 3, respectively.

4. The method of claim 1, wherein the step of attention ranking of the method wherein weighting the weight of the fault feature according to the channel attention and the spatial attention of the convolution attention module in the bearing fault diagnosis model comprises:

and

Wherein: w ₀ ∈R ^c/r×c ，W ₁ ∈R ^c×c/r ，W ₀ Relu is used as an activation function, and sigma is a Sigmoid function;

injecting the channelGravity weight coefficient M _c (F) And multiplying the input feature F element by element to obtain a refined feature F'.

5. The method of claim 4, wherein the step of attention ranking of the method wherein weighting the weights of the fault signatures according to channel and spatial attention of a convolution attention module in a bearing fault diagnosis model comprises:

And

will be described in

And

merging the feature maps into a feature map with 2 channels, and compressing the feature map in channel dimension through a convolution kernel with the step length of 2 and the size of 1 × 7, wherein the number of the compressed features is 1;

feature vector is activated based on sigmoid function to obtain spatial attention coefficient

weighting the spatial attention weight coefficient M _s (F') performing element-by-element with the input feature FThe elements are multiplied to finally generate a spatial attention map.

6. The method of claim 5, wherein the method further comprises:

extracting fault features in the multi-channel signal based on a bearing fault diagnosis model of a residual error dense network of multi-stage residual error dense blocks

As inputs, the residual dense unit based on CBAM yields the output:

in the formula, w ₁ ，b ₁ ，w ₂ ，b ₂ ，w ₃ ，b ₃ The parameters to be trained are convolution 1, convolution 2 and convolution 3, respectively.

7. The method of claim 1, further comprising a bearing fault diagnosis model training step of:

8. A residual error intensive network fault diagnosis apparatus, the apparatus comprising:

the data preprocessing module is used for acquiring detection signals of the bearing, performing discrete cosine transform and envelope spectrum extraction on the detection signals to generate multi-domain detection signals, and superposing the multi-domain detection signals and the detection signals to generate multi-channel signals which are used as input of a bearing fault diagnosis model;

the fault feature weighting module is used for weighting the weight of the fault feature according to the channel attention and the space attention of the convolution attention module in the bearing fault diagnosis model;

9. An electronic device, comprising

A processor; and

a memory having computer readable instructions stored thereon which, when executed by the processor, implement the method of any of claims 1 to 7.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 7.