CN115114965B - Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium - Google Patents
Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN115114965B CN115114965B CN202211037112.5A CN202211037112A CN115114965B CN 115114965 B CN115114965 B CN 115114965B CN 202211037112 A CN202211037112 A CN 202211037112A CN 115114965 B CN115114965 B CN 115114965B
- Authority
- CN
- China
- Prior art keywords
- fault
- network model
- gearbox
- convolution
- volume block
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/021—Gearings
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/028—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/72—Wind turbines with rotation axis in wind direction
Abstract
The invention discloses a wind turbine generator gearbox fault diagnosis method, a device, equipment and a storage medium, wherein the model comprises a first characteristic extraction network model for extracting first fault characteristic information of a signal to be detected; the second characteristic extraction network model is used for extracting second fault characteristic information of the signal to be detected, wherein the depth of the second fault characteristic information is greater than that of the first fault characteristic information; the weighted fusion model is used for carrying out weighted fusion on the first fault characteristic information and the second fault characteristic information to obtain third fault characteristic information; and the classification model is used for carrying out gearbox fault classification according to the third fault characteristic information to obtain a fault diagnosis result. According to the wind turbine generator gearbox fault diagnosis model, the first fault characteristic information and the second fault characteristic information with different characteristic depths are respectively extracted, then the first fault characteristic information and the second fault characteristic information are subjected to weighted fusion and then classified identification is carried out, and compared with a single model, the accuracy is improved, and the identification accuracy is higher.
Description
Technical Field
The invention relates to the technical field of fault detection, in particular to a wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium.
Background
The gear transmission system is an important component of a large double-fed wind turbine generator and is the key for realizing energy transmission and conversion. The gearbox is an important transmission device of the wind turbine generator, because the gearbox works under the working conditions of low speed, heavy load, severe environment and the like for a long time, the probability of failure of the gearbox is high due to the complex working conditions and the severe working environment, and once the gearbox serving as a hub is damaged, the machine set is shut down, personnel safety accidents and other serious consequences are caused, so that the gearbox has very important significance for effectively identifying the failure of the gearbox.
The structural parameters of each component of the gear box are different, so that the frequency and the amplitude of vibration signals caused by each part are different. Therefore, when a certain component is in fault, a specific vibration signal corresponding to the component can be generated, and the fault range and the property of the gearbox can be obtained through analyzing the vibration signal. Most of gear box fault diagnosis models in the prior art adopt a single network or an optimized single network for feature extraction, but because the operating condition of a gear box is complex, and the generated faults are often mixed faults, the single network is difficult to extract fault classification features in all aspects, and the identification accuracy of the models is low.
Disclosure of Invention
In view of this, embodiments of the present invention provide a method, an apparatus, a device and a storage medium for diagnosing a fault of a gearbox of a wind turbine generator, so as to solve the problem of low accuracy of diagnosing and identifying a fault of a gearbox.
The technical scheme provided by the invention is as follows:
the first aspect of the embodiments of the present invention provides a wind turbine generator gearbox fault diagnosis model, including: the first characteristic extraction network model is used for extracting first fault characteristic information of the signal to be detected; the second characteristic extraction network model is used for extracting second fault characteristic information of the signal to be detected, wherein the depth of the second fault characteristic information is larger than that of the first fault characteristic information; the weighted fusion model is used for carrying out weighted fusion on the first fault characteristic information and the second fault characteristic information to obtain third fault characteristic information; and the classification model is used for carrying out gearbox fault classification according to the third fault characteristic information to obtain a fault diagnosis result.
Optionally, the first feature extraction network model is a modified Resnet50 network model, the first feature extraction network model includes a first volume block, a second volume block, a third volume block, a fourth volume block, and a fifth volume block, the first volume block includes a first convolution kernel, a batch normalization function, an activation function, and a maximum pooling layer; the second convolution block comprises a plurality of first trunk networks and a plurality of identity mapping units which are respectively fused with the first trunk networks correspondingly; the third convolution block includes a second backbone network; the fourth convolution block comprises a plurality of third trunk networks and a plurality of identity mapping units which are respectively fused with the third trunk networks correspondingly; the fifth convolution block comprises a plurality of fourth trunk networks and a plurality of identity mapping units which are respectively fused with the fourth trunk networks correspondingly.
Optionally, the second feature extraction network model is a modified Resnet50 network model, the second feature extraction network model includes a sixth volume block, a seventh volume block, an eighth volume block, a ninth volume block, and a tenth volume block, the sixth volume block includes a plurality of second convolution kernels, a batch normalization function, an activation function, and a maximum pooling layer, the second convolution kernels are smaller than the first convolution kernels; the seventh convolution block comprises a plurality of fifth trunk networks and a plurality of identity mapping units which are respectively fused with the fifth trunk networks correspondingly; the eighth convolution block comprises a plurality of sixth trunk networks and a plurality of identity mapping units which are respectively fused with the sixth trunk networks correspondingly; the ninth convolution block comprises a plurality of seventh main networks and a plurality of identity mapping units which are respectively fused with the seventh main networks correspondingly; the tenth convolution block includes a plurality of eighth trunk networks and a plurality of identity mapping units respectively fused with the eighth trunk networks.
Optionally, the weighted fusion model is specifically configured to assign weights to all the first fault feature information and the second fault feature information, and multiply and accumulate the first fault feature information and the second fault feature information by the corresponding weights, respectively, to obtain third fault feature information.
The second aspect of the embodiments of the present invention provides a wind turbine generator gearbox fault diagnosis method, including: receiving a signal to be detected of the gear box; and inputting the detection signal into a wind turbine generator gearbox fault diagnosis model according to the first aspect of the embodiment of the invention to obtain a fault diagnosis result.
Optionally, the receiving a signal to be detected of the gearbox includes: receiving a one-dimensional vibration signal of the gearbox; and converting the one-dimensional vibration signal into a two-dimensional image, wherein the signal to be detected is the two-dimensional image.
Optionally, converting the one-dimensional vibration signal into a two-dimensional image comprises: and normalizing the one-dimensional vibration signal by adopting a symmetrical dot matrix image analysis method to obtain a two-dimensional image in a polar coordinate form.
A third aspect of an embodiment of the present invention provides a wind turbine generator gearbox fault diagnosis device, including: the signal receiving module is used for receiving a signal to be detected of the gearbox; and the diagnosis module is used for inputting the detection signal to the wind turbine generator gearbox fault diagnosis model according to the first aspect of the embodiment of the invention to obtain a fault diagnosis result.
A fourth aspect of an embodiment of the present invention provides an electronic device, including: the wind turbine generator gearbox fault diagnosis method comprises a storage and a processor, wherein the storage and the processor are in communication connection with each other, the storage stores computer instructions, and the processor executes the computer instructions so as to execute the wind turbine generator gearbox fault diagnosis method according to the second aspect of the embodiment of the invention.
A fifth aspect of the embodiments of the present invention provides a computer-readable storage medium, which stores computer instructions for causing a computer to execute the wind turbine generator gearbox fault diagnosis method according to the second aspect of the embodiments of the present invention.
According to the technical scheme, the embodiment of the invention has the following advantages:
according to the method, the device, the equipment and the storage medium for diagnosing the fault of the gearbox of the wind turbine generator, provided by the embodiment of the invention, the first fault characteristic information and the second fault characteristic information are respectively extracted through the first characteristic extraction network model and the second characteristic extraction network model, the depth of the second fault characteristic information is greater than that of the first fault characteristic information, then the first fault characteristic information and the second fault characteristic information are subjected to weighted fusion through the weighted fusion model to obtain third fault characteristic information, and finally the classification model carries out gearbox fault classification according to the third fault characteristic information to obtain a fault diagnosis result. According to the wind turbine generator gearbox fault diagnosis model, the first fault characteristic information and the second fault characteristic information with different characteristic depths are respectively extracted, then the first fault characteristic information and the second fault characteristic information are subjected to weighted fusion and then classified identification is carried out, and compared with a single model, the accuracy is improved, and the identification accuracy is higher.
Drawings
In order to express the technical scheme of the embodiment of the invention more clearly, the drawings used for describing the embodiment will be briefly introduced below, and obviously, the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic structural diagram of a wind turbine generator gearbox fault diagnosis model in an embodiment of the invention;
FIG. 2 is a diagram of a convolutional layer structure of a Resnet50 network model in an embodiment of the present invention;
FIG. 3 is a schematic block diagram of a Resnet50 network model in an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a Resnet50 network model according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of BANK1 and BANK2 in the embodiment of the present invention;
FIG. 6 is a schematic structural diagram of a first feature extraction network model according to an embodiment of the present invention;
FIG. 7 is a schematic structural view of BANK3 and BANK4 in the embodiment of the present invention;
FIG. 8 is a schematic structural diagram of a second feature extraction network model according to an embodiment of the present invention;
FIG. 9 is a diagram illustrating the effect of extracting feature information at a lower layer of the Resnet50 network model according to an embodiment of the present invention;
FIG. 10 is a diagram illustrating the effect of extracting feature information at a high level of the Resnet50 network model according to an embodiment of the present invention;
FIG. 11 is a flow chart of a wind turbine generator gearbox fault diagnosis method according to an embodiment of the invention;
FIG. 12 is a diagram illustrating the conversion of one-dimensional vibration signals into two-dimensional images according to an embodiment of the present invention;
FIG. 13 shows a variation of the embodiment of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 14 shows another embodiment of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 15 shows another variation of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 16 shows another embodiment of the present invention
And
effect graph of two-dimensional image under value；
FIG. 17 shows another variation of the present invention
And
obtaining an effect graph of the two-dimensional image;
FIG. 18 shows another embodiment of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 19 shows another variation of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 20 shows another variation of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 21 shows another variation of the present invention
And
taking an effect graph of the two-dimensional image;
FIG. 22 is a two-dimensional image effect graph corresponding to a pitting wear mixed fault in an embodiment of the invention;
FIG. 23 is a diagram illustrating the effect of two-dimensional images corresponding to pitting failures in an embodiment of the present invention;
FIG. 24 is an effect diagram of a two-dimensional image corresponding to a broken tooth wear mixed fault in an embodiment of the invention;
FIG. 25 is a diagram illustrating the effect of a two-dimensional image corresponding to a tooth breakage failure in an embodiment of the present invention;
FIG. 26 is a graph illustrating the effect of a two-dimensional image corresponding to a wear failure in an embodiment of the present invention;
FIG. 27 is a diagram illustrating an effect of a two-dimensional image corresponding to a normal state in an embodiment of the present invention;
FIG. 28 is a chart showing the accuracy curve of classification training in the embodiment of the present invention;
FIG. 29 is a graph showing model accuracy after training is completed according to the embodiment of the present invention;
FIG. 30 is a block diagram of a wind turbine generator gearbox fault diagnosis apparatus according to an embodiment of the present invention;
FIG. 31 is a schematic structural diagram of an electronic device in an embodiment of the invention;
fig. 32 is a schematic structural diagram of a computer-readable storage medium in an embodiment of the present invention.
Detailed Description
For a better understanding of the present invention, reference will now be made to the embodiments of the present invention which are illustrated in the accompanying drawings, wherein like reference numerals refer to like elements throughout, and wherein the same reference numerals refer to like elements throughout. The described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides a wind turbine generator gearbox fault diagnosis model, which comprises the following components:
and the first characteristic extraction network model is used for extracting first fault characteristic information of the signal to be detected.
And the second characteristic extraction network model is used for extracting second fault characteristic information of the signal to be detected, wherein the depth of the second fault characteristic information is greater than that of the first fault characteristic information.
The signal to be detected is obtained by monitoring a sensor arranged in the gear box, and specifically, the signal to be detected is a two-dimensional image converted from a vibration signal in the gear box. The structural parameters of each component of the gear box are different, so that the frequency and the amplitude of vibration signals caused by each part are different. Therefore, when a certain component is in fault, a specific vibration signal corresponding to the component can be generated, and the fault range and the property of the gearbox can be obtained through analyzing the vibration signal. For example, when the gearbox has faults such as pitting, abrasion, tooth breakage and the like, a specific vibration signal is corresponding to the fault, so that the embodiment of the invention analyzes the vibration signal to further judge the fault type of the gearbox. The first feature extraction network model and the second feature extraction network model are both used for feature extraction of the signal to be detected, the emphasis points of the features extracted by the first feature extraction network model and the feature extracted by the second feature extraction network model are different, the depth of the second fault feature information extracted by the second feature extraction network model is larger than the depth of the first fault feature information extracted by the first feature extraction network model, namely the first feature extraction network model is used for emphasizedly extracting the bottom feature of the image, and the second feature extraction network model is used for emphasizedly extracting the deep feature with more expressive power in the image.
And the weighted fusion model is used for carrying out weighted fusion on the first fault characteristic information and the second fault characteristic information to obtain third fault characteristic information. Specifically, weights are assigned to all the first fault feature information and the second fault feature information, and the first fault feature information and the second fault feature information are multiplied by the corresponding weights respectively and accumulated to obtain third fault feature information. Illustratively, the deepest output features of the first feature extraction network model and the second feature extraction network model are summarized, then different weights are given according to different algorithms, and feature fusion is carried out on first fault feature information and second fault feature information identified by the first feature extraction network model and the second feature extraction network model, wherein the fusion formula is as follows:
wherein Result is the Result after fusion,
in order to be the total number of models,
in order to perform the total number of features to be fused,
is a model
To middle
The weight value occupied by the individual features,
is a model
To middle
And (4) the characteristic value.
And the classification model is used for carrying out gearbox fault classification according to the third fault characteristic information to obtain a fault diagnosis result. Specifically, the classification model adopts a Softmax classifier, after third fault characteristic information is input into the Softmax classifier, the Softmax classifier classifies the detected fault type of the gearbox into a type of fault with the highest similarity by calculating the similarity of the third fault characteristic information and fault characteristics such as pitting, abrasion and tooth breakage according to the received third fault characteristic information, and accordingly classifies the fault of the gearbox.
According to the wind turbine generator gearbox fault diagnosis model provided by the embodiment of the invention, the first fault characteristic information and the second fault characteristic information are respectively extracted through the first characteristic extraction network model and the second characteristic extraction network model, the depth of the second fault characteristic information is greater than that of the first fault characteristic information, then the first fault characteristic information and the second fault characteristic information are subjected to weighted fusion through the weighted fusion model to obtain third fault characteristic information, and finally the classification model is used for carrying out gearbox fault classification according to the third fault characteristic information to obtain a fault diagnosis result. According to the wind turbine generator gearbox fault diagnosis model, the first fault characteristic information and the second fault characteristic information with different characteristic depths are respectively extracted, then weighted fusion is carried out, and classification recognition is carried out, so that compared with a single model, the accuracy is improved, the recognition accuracy is higher, and meanwhile, compared with a double-channel characteristic fusion structure of a first characteristic extraction network model and a second characteristic extraction network model, the double-channel characteristic fusion structure is more excellent in generalization capability and robustness.
In one embodiment, the first feature extraction network model is a modified Resnet50 network model, the first feature extraction network model includes a first volume block, a second volume block, a third volume block, a fourth volume block, and a fifth volume block, the first volume block includes a first convolution kernel, a batch normalization function, an activation function, and a max pooling layer; the second convolution block comprises a plurality of first trunk networks and a plurality of identity mapping units which are respectively fused with the first trunk networks correspondingly; the third convolution block includes a second backbone network; the fourth convolution block comprises a plurality of third trunk networks and a plurality of identity mapping units which are respectively fused with the third trunk networks correspondingly; the fifth convolution block comprises a plurality of fourth backbone networks and a plurality of identity mapping units which are respectively fused with the fourth backbone networks correspondingly.
The existing Resnet50 network model is shown in fig. 2, and mainly includes five volume blocks, conv1, conv2_ x, conv3_ x, conv4_ x and conv5_ x, where average pool represents average poolLayer, FLOPs, represents the number of floating point operations per second. The five volume blocks of the Resnet50 network model are mapped to five phases, shown in fig. 3 and 4, which are phase 0, phase 1, phase 2, phase 3, and phase 4, respectively. Specifically, the convolution blocks in the stage 1 to the stage 4 of the Resnet50 network model are each composed of 2 Bottleneck layers (bottleeck), which are respectively denoted as BANK1 and BANK2. As shown in fig. 5, BANK1 has four variable parameters C, W, C and S, the input shape is (C, W), the output shape is (C1 × 4,W/S, W/S), the left side of BANK1 is a backbone network, which includes three volume blocks and associated RELU activation functions and Batch Normalization processing (BN), and the right side is an identity mapping (identity mapping) unit, which includes one volume block and a RELU activation function. Let the output of the backbone network of BANK1 be
The output of the identity mapping unit is
Then the output of BANK1 is
(ii) a BANK2 differs from BANK1 in that the identity mapping unit on the right only includes the RELU activation function, so the output of BANK2 is
. Wherein the content of the first and second substances,
，
is shown as
The activation function of the layer, BN denotes the process of batch normalization,
when is coming into contact with
In the process, the gradient is constantly 1, so that the saturation condition is avoided, the disappearance of the gradient is overcome, and the convergence speed is accelerated; when the temperature is higher than the set temperature
And in the hard saturation region, the output value is 0, so that the sparsity and generalization performance of the network are improved.
The first feature extraction network model is improved from the Resnet50 network model, which is denoted as D-Resnet in the embodiment of the present invention, and the structure thereof is shown in fig. 6 and 7. The first feature extraction network model D-Resnet also includes a phase 0, a phase 1, a phase 2, a phase 3, and a phase 4, which respectively correspond to the first volume block, the second volume block, the third volume block, the fourth volume block, and the fifth volume block, where the phase 0, the phase 1, the phase 3, and the phase 4 are all the same as the Resnet50 network model. Specifically, the structure of the phase 0 of the first feature extraction network model is, as in the Resnet50 network model, to perform preliminary feature extraction using convolution kernel images with a size of 7 × 7 × 64 and a step size of 2, and to process fault images of the gearbox using a batch normalization function and an activation function RELU, and then perform maximum pooling using a maximum pooling layer MAXPAXOOL with a step size of 2.
The structure of the volume block in the stage 1, the stage 3 and the stage 4 is also the same as that of the Resnet50 network model, the stage 1 sequentially comprises 1 BANK1 and two BANKs 2, 3 groups of first trunk networks with the size of 1 × 1 × 64, 3 × 3 × 64 and 1 × 1 × 256 are respectively used, and the first trunk networks with the step length of 1 extract deep features and are fused with feature information transmitted by corresponding identity mapping units; stage 3 sequentially comprises 1 BANK1 and 5 BANK2, and uses 6 groups of third trunk networks with the size of 1 × 1 × 256, 3 × 3 × 256 and 1 × 1 × 1024 and the step length of 2 to extract features and fuse the features with feature information transmitted by corresponding identity mapping units; the stage 4 comprises 1 BANK1 and 2 BANKs 2 in turn, uses 3 groups of 1 × 1 × 512, 3 × 3 × 512, 1 × 1 × 2048 in size, has a step size of 2, and fuses with the characteristic information transmitted from the corresponding identity mapping unit. The output of stage 4 is finally processed through the averaging pooling layer.
In particular, the first feature extraction network model differs from the Resnet50 network model in that the volume block of stage 2, including only 4 sets of second backbone networks of 1 × 1 × 128, 3 × 3 × 128, 1 × 1 × 512 size and 2 step size, has no corresponding identity mapping unit. The convolution block of stage 2 of the first feature extraction network model is composed of BANK3 and BANK4, and sequentially comprises 1 BANK3 and 3 BANKs 4. The structures of BANK3 and BANK4 are shown in FIG. 7, BANK3 and BANK4 are respectively BANK1 and BANK2 and are obtained by removing right identical mapping units, the difference between BANK3 and BANK4 and BANK1 and BANK2 is that no right identical mapping unit exists, and the outputs of BANK3 and BANK4 after the right identical mapping units are removed are both
Wherein, in the step (A),
is the output of the backbone network of BANK3 and BANK4. The first feature extraction network model in the embodiment of the present invention may be represented by the following formula:
in the above formula, the first and second carbon atoms are,
in the case of a gearbox fault type,
being the fully connected layer of the Resnet50,
for the residual mapping function, i.e. the backbone network corresponding to BANK1 and BANK2 as described above,
is the corresponding identity map. In a convolutional network, the extracted information is not all useful information, and a method for reducing negative information by adding branches is adoptedOverhead of network performance is increased. For the Resnet50 network model, the convolution kernel at the lower layer extracts more spatial features and less classified features, so that the first feature extraction network model of the embodiment of the invention reduces the occupation ratio of spatial feature information and negative information in the network by discarding the identity mapping unit in the volume block of the stage 2, thereby reducing the performance overhead of the network, and meanwhile, because a part of the identity mapping unit is discarded, the depth of the first feature extraction network model for extracting image features is lower than that of the existing Resnet50, and the first feature extraction network model focuses on extracting the bottom layer features of the image.
In an embodiment, the second feature extraction network model is an improved Resnet50 network model, the second feature extraction network model includes a sixth volume block, a seventh volume block, an eighth volume block, a ninth volume block, and a tenth volume block, the sixth volume block includes a plurality of second convolution kernels, a batch normalization function, an activation function, and a maximum pooling layer, the second convolution kernels are smaller than the first convolution kernels; the seventh convolution block comprises a plurality of fifth trunk networks and a plurality of identity mapping units which are respectively fused with the fifth trunk networks correspondingly; the eighth convolution block comprises a plurality of sixth trunk networks and a plurality of identity mapping units which are respectively fused with the sixth trunk networks correspondingly; the ninth convolution block comprises a plurality of seventh trunk networks and a plurality of identity mapping units which are respectively fused with the seventh trunk networks correspondingly; the tenth convolution block includes a plurality of eighth trunk networks and a plurality of identity mapping units respectively fused with the eighth trunk networks.
The second feature extraction network model is improved from the Resnet50 network model, and is denoted as E-Resnet in the embodiment of the present invention, as shown in fig. 8, the second feature extraction network model in the embodiment of the present invention also includes a phase 0, a phase 1, a phase 2, a phase 3, and a phase 4, which correspond to a sixth volume block, a seventh volume block, an eighth volume block, a ninth volume block, and a tenth volume block, respectively, and is different from the Resnet50 network model in that, in the phase 0, that is, the sixth volume block, a 7 × 7 large convolution kernel is changed into 3 small convolution kernels, and the structures of the sixth volume block, the seventh volume block, the eighth volume block, the ninth volume block, and the tenth volume block corresponding to the remaining phase 1, the phase 2, the phase 3, and the phase 4 are the same as the volume blocks in the phase 1, the phase 2, the phase 3, and the phase 4 of the Resnet50 network model.
For the convolutional neural network, different layers of convolutional kernels acquire different strengths of image features, the lower layer of convolutional kernels pay more attention to extracting the spatial information of the image, and the classification feature extraction effect is poor; and the high-level convolution kernel emphasizes the extraction of image classification characteristics, so that the spatial information is relatively weak to obtain. The Resnet50 low-level, high-level extracted feature information is visualized as shown in FIGS. 9 and 10. The feature information extracted by Resnet50 at the lower layer is mainly the outline of the image, and the feature information extracted by Resnet50 at the higher layer is more gathered to the thickness and the geometric center of the petal of the image. The first layer convolution kernel of the Resnet50 is 7 × 7, the convolution kernel is large, although the feature information is obtained much, most of the feature information is spatial information, and the corresponding calculation amount is also large, so that the performance of the model is poor.
The embodiment of the invention extracts the characteristics of the fault image of the gearbox by changing the 7 multiplied by 7 convolution kernel in the Resnet50 network stage 0 into 3 multiplied by 3 small convolution kernels. The transformed second feature extraction network model has the following parameters:
where M is the size of the input feature map, N is the number of convolution kernels, and the number of convolution kernel parameters in Resnet50 is
Compared with Resnet50, the parameter quantity of the second feature extraction network model is reduced, the model performance is improved, the classification speed is accelerated, and meanwhile, the same backbone network and identity mapping units as Resnet50 are maintained from stage 1 to stage 4, so that 3 x 3 small convolution kernels are adopted by the second feature extraction network model to replace 7 x 7 large convolution kernels in stage 0, the calculated quantity is reduced, the extraction of image features by model low-layer convolution kernels is weakened, and the occupation ratio of space information in extracted features is reduced.
According to the wind turbine generator gearbox fault diagnosis model, the first characteristic extraction network model and the second characteristic extraction network model are obtained through improvement based on the existing Resnet50, the first fault characteristic information and the second fault characteristic information are respectively extracted, then weighted fusion is carried out, and classification recognition is carried out, while the gearbox fault diagnosis model in the prior art mostly adopts a single network or an optimized single network to carry out characteristic extraction, but the fault diagnosis model of the wind turbine generator gearbox is complex in operation condition, the generated fault is often a mixed fault, the single network is difficult to comprehensively extract fault classification characteristics, the recognition accuracy of the model is low, and the model is poor in robustness and generalization capability.
The embodiment of the invention also provides a wind turbine generator gearbox fault diagnosis method, as shown in fig. 11, the method comprises the following steps: receiving a signal to be detected of the gearbox; the detection signal is input into the wind turbine generator gearbox fault diagnosis model to obtain a fault diagnosis result. Wherein, the signal to be detected is obtained by receiving vibration signals through different sensors arranged in the gear box. Illustratively, the vibration signals for each state of the gearbox are monitored by 6 sensors, namely, input shaft X and Y directional displacement, wind turbine side bearing X directional acceleration, wind turbine side bearing Y directional acceleration, generator side bearing X directional acceleration, and generator side bearing Y directional acceleration. The specific processing steps after the signal to be detected is input to the wind turbine generator gearbox fault diagnosis model refer to the corresponding parts of the wind turbine generator gearbox fault diagnosis model embodiment, which are not described herein again.
According to the fault diagnosis method for the gearbox of the wind turbine generator, firstly, a signal to be detected of the gearbox is received, first fault characteristic information and second fault characteristic information are extracted through a first characteristic extraction network model and a second characteristic extraction network model respectively based on the signal to be detected, then the first fault characteristic information and the second fault characteristic information are subjected to weighted fusion through a weighted fusion model to obtain third fault characteristic information, and finally, the classification model is used for carrying out fault classification on the gearbox according to the third fault characteristic information to obtain a fault diagnosis result. According to the wind turbine generator gearbox fault diagnosis method, the first fault characteristic information and the second fault characteristic information are extracted respectively, then weighted fusion is carried out, classification recognition is carried out, and compared with a single model, the accuracy is improved to some extent, and the recognition accuracy is higher.
In one embodiment, receiving a signal to be detected of a gearbox includes:
and receiving a one-dimensional vibration signal of the gearbox. Specifically, 6 sensors are adopted to monitor one-dimensional vibration signals of the gearbox in each state, namely displacement in the X direction and the Y direction of the input shaft, acceleration in the X direction of the fan-side bearing, acceleration in the Y direction of the fan-side bearing, acceleration in the X direction of the generator-side bearing and acceleration in the Y direction of the generator-side bearing, so that one-dimensional vibration signals of the gearbox are obtained. The gear transmission system is an important component of a large double-fed wind turbine generator and is the key for realizing energy transmission and conversion. The gearbox is an important transmission device of the wind turbine generator, and because the gearbox works under working conditions of low speed, heavy load, severe environment and the like for a long time, the fault probability of the gearbox is high, the fault of the gearbox can cause the whole wind turbine generator to stop running, and the economic benefit of a wind power plant can be seriously influenced by the faults and unplanned shutdown. Therefore, the method has important practical significance for effectively identifying the gearbox fault. The structural parameters of each component of the gear box are different, so that the frequency and the amplitude of vibration signals caused by each part are different. Therefore, when a certain component is in fault, a specific vibration signal corresponding to the component can be generated, and the fault range and the property of the gearbox can be obtained through analyzing the vibration signal.
And converting the one-dimensional vibration signal into a two-dimensional image, wherein the signal to be detected is the two-dimensional image. Specifically, a Symmetric Dot Pattern (SDP) method is adopted to normalize the one-dimensional vibration signal to obtain a two-dimensional image in a polar coordinate form. The one-dimensional vibration signals are converted into the two-dimensional images, so that the relation between the amplitude and the frequency of different fault signals is displayed through the two-dimensional images, the two-dimensional images converted by the method can effectively express the fault characteristics of the gearbox, and the fault characteristics of the gearbox can be more intuitively expressed on the premise of not losing original signals. Meanwhile, a one-dimensional vibration signal is converted into a two-dimensional image (namely an SDP image), so that a plurality of groups of data can be represented in the same image, the influence of noise can be greatly reduced, and the identification speed and the classification accuracy are improved.
The mapping conversion method of the SDP algorithm normalization method is mainly based on formulas (1) - (4).
In the above formula:
the radius is under a polar coordinate system;
is the first of a time domain sequence
The serial number of each sampling point; sampling point
Corresponding to an acquisition value of
(ii) a While
And
respectively a maximum acquisition value and a minimum acquisition value in the signal sequence;
is a time interval parameter;
to enlarge factor
；
Represents the first
A mirror symmetry plane;
is the total number of mirror symmetry planes;
is as follows
A mirror symmetry plane angle;
is composed of
The counterclockwise deflection angle of;
is composed of
Clockwise deflection angle.
By transformation of equations (1) - (4), one-dimensional vibration signal can be converted into
The one-dimensional time-domain vibration signal is converted into a two-dimensional image shown in fig. 12. For SDP algorithm, parameter values
、
And
the selection of (A) is very important for the characteristic representation of the two-dimensional image.
The degree of scatter of the two-dimensional image points is affected,
the value of (a) is generally determined according to the dimension of the feature vector. For example, if 6 sensors are used to obtain the vibration signal, then the selection is made
Therefore, it is
Wherein
. Taking the normal state of the gearbox as an example, for
And
the influence of the difference in the values of (a) on the two-dimensional image is shown in fig. 13 to 21. Through verification, when the parameters are
，
The gearbox condition is characterized most significantly and therefore determined
，
，
The one-dimensional vibration signal is converted into a two-dimensional image resembling a "petal".
Specifically, a specific two-dimensional image will be described by taking a gear failure among failures of rotating members of a gear box as an example.
According to the difference of the gear states, 6 data sets are selected in total, wherein the data sets comprise 1 normal state and 5 fault states, 2 of the data sets are mixed fault states which are respectively pitting corrosion, abrasion, tooth breakage, pitting abrasion mixed fault, tooth breakage abrasion mixed fault and normal state. The two-dimensional image can realize the fusion of multiple feature vectors and show the features of different sensor vibration signals in each state. In the two-dimensional image in each state, different fault types can be distinguished by comparing the curvature, the thickness, the geometric center, the shape of the two-dimensional image and the like of petals of the two-dimensional image. The vibration signals of the gearbox in each state are monitored by 6 sensors, namely displacement in the X direction and the Y direction of an input shaft, acceleration in the X direction of a fan side bearing, acceleration in the Y direction of the fan side bearing, acceleration in the X direction of a generator side bearing and acceleration in the Y direction of the generator side bearing. And (3) carrying out two-dimensional image conversion on the obtained one-dimensional vibration signal, namely converting the one-dimensional vibration signal into a two-dimensional image, wherein each fault state of the gearbox generates a corresponding two-dimensional image.
Two-dimensional images of the gearbox in different states are shown in fig. 22 to 27. From an analysis of fig. 22 to 27, it can be seen that there are significant differences in the curvature, shape and size of the "petals" between the two-dimensional images of the different states of the gearbox. Observe with
The petals are symmetrical surfaces, obvious differences exist between two-dimensional images corresponding to different states, the petals are thick in a normal state, the petals are short in a pitting state, and obvious differences also exist among lengths, thicknesses and curvatures of the petals in other states. In addition, observe and control
The petals of the symmetrical surface are distributed more discretely corresponding to pitting wear and pitting corrosion, and the petals in other states are distributed more intensively. By converting the one-dimensional vibration signal into the two-dimensional image, under the condition of not reducing the original vibration signal, a large amount of feature extraction time is saved, and simultaneous expression of various feature vectors can be realized, which has important significance for fault identification based on the vibration signal.
And dividing the two-dimensional image generated by conversion into a training set and a test set, wherein the training set accounts for 80% and the test set accounts for 20%. The number of samples of each input training is 20, 1000 rounds (epoch) of training are performed, the learning rate is 0.01, the batch (batch) value is 80, and L2 regularization is adopted to increase the size of a penalty term constraint parameter on a loss function to prevent overfitting of the network. Inputting the test set into the trained wind turbine generator gearbox fault diagnosis model (DE-Resnet) to extract the characteristics of the gearbox fault to obtain a fault diagnosis result. The accuracy of the fault diagnosis model is characterized by the accuracy. In particular, according to the formula
Calculating the accuracy of the model, wherein
In order to identify the exact number of samples,
is the total amount of samples tested. For explaining the fault diagnosis of the wind turbine gearbox provided by the embodiment of the inventionAnd (3) interrupting the advantages of the model, and inputting the same training set and test set into the improved single D-Resnet network, the improved single E-Resnet network and the existing Resnet50 network model for training and testing respectively. The SDP images of 6 different states of the gear box were trained by classification, and the accuracy curve of classification training obtained by calculation is shown in fig. 28, and the accuracy of the model after final training is shown in fig. 29. As can be seen from the accuracy curve of fig. 28, the accuracy based on the two-dimensional image and the four models is substantially higher than 90% after 1000 iterations. As can be analyzed by the graph 29, the accuracy of the model for diagnosing the gearbox fault of the wind turbine generator, namely DE-Resnet, is highest and reaches 97.6%, and compared with the improved single D-Resnet network model, the improved single E-Resnet network model and the existing Resnet50 network model, the accuracy is respectively 2.4%, 2.5% and 5.7%. Therefore, the wind turbine generator gearbox fault diagnosis model provided by the embodiment of the invention can well meet the requirement of gearbox fault classification, the training data is only converted through two-dimensional images and is not subjected to noise processing, and the accuracy of DE-Resnet still reaches 97.6%. Therefore, the model has good adaptability to the environment and good application prospect.
An embodiment of the present invention further provides a wind turbine generator gearbox fault diagnosis device, as shown in fig. 30, the device includes:
and the signal receiving module is used for receiving the signal to be detected of the gearbox. For details, reference is made to the corresponding parts of the above method embodiments, which are not described herein again.
And the diagnosis module is used for inputting the detection signal to the wind turbine generator gearbox fault diagnosis model of the embodiment to obtain a fault diagnosis result. For details, reference is made to the corresponding parts of the above method embodiments, which are not described herein again.
The wind turbine generator gearbox fault diagnosis device provided by the embodiment of the invention comprises the steps of firstly receiving a signal to be detected of a gearbox, respectively extracting first fault characteristic information and second fault characteristic information based on the signal to be detected through a first characteristic extraction network model and a second characteristic extraction network model, then carrying out weighted fusion on the first fault characteristic information and the second fault characteristic information through a weighted fusion model to obtain third fault characteristic information, and finally carrying out gearbox fault classification according to the third fault characteristic information through a classification model to obtain a fault diagnosis result. According to the wind turbine generator gearbox fault diagnosis device, the first fault characteristic information and the second fault characteristic information are respectively extracted, then weighted fusion is carried out, classification identification is carried out, and compared with a single model, the accuracy is improved to some extent, and the identification accuracy is higher.
An embodiment of the present invention further provides an electronic device, including: the memory 12 and the processor 11 are connected with each other in a communication mode, the memory 12 stores computer instructions, and the processor 11 executes the computer instructions so as to execute the wind turbine generator gearbox fault diagnosis method according to the embodiment of the invention. As shown in fig. 31, a memory 12 and a processor 11 are included, wherein the processor 11 and the memory 12 may be connected by a bus or other means. The processor 11 may be a Central Processing Unit (CPU). The processor 11 may also be other general purpose processors, digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof. The memory 12, which is a non-transitory computer storage medium, may be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the corresponding program instructions/modules in embodiments of the present invention. The processor 11 executes various functional applications and data processing of the processor 11 by running the non-transitory software programs, instructions and modules stored in the memory 12, that is, the wind turbine gearbox fault diagnosis method in the above method embodiment is realized. The memory 12 may include a storage program area and a storage data area, wherein the storage program area may store an application program required for operating the device, at least one function; the storage data area may store data created by the processor 11, and the like. Further, the memory 12 may include high speed random access memory 12, and may also include non-transitory memory 12, such as at least one piece of disk memory 12, flash memory device, or other non-transitory solid state memory 12. In some embodiments, the memory 12 may optionally include memory 12 located remotely from the processor 11, and these remote memories 12 may be connected to the processor 11 over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof. One or more modules are stored in memory 12, which when executed by processor 11, perform the wind turbine gearbox fault diagnosis method as in the above-described method embodiments. The specific details of the electronic device may be understood according to the related descriptions and effects corresponding to the method embodiments, and are not described herein again.
An embodiment of the present invention further provides a computer-readable storage medium, as shown in fig. 32, on which a computer program 13 is stored, and when the instructions are executed by a processor, the instructions implement the steps of the wind turbine generator gearbox fault diagnosis method in the foregoing embodiment. The storage medium is also stored with audio and video stream data, characteristic frame data, an interactive request signaling, encrypted data, preset data size and the like. The storage medium may be a magnetic disk, an optical disk, a Read-only memory (ROM), a Random Access Memory (RAM), a flash memory (FlashMemory), a hard disk (hard disk drive, abbreviated as HDD) or a Solid State Drive (SSD), etc.; the storage medium may also comprise a combination of memories of the kind described above. It will be understood by those skilled in the art that all or part of the processes of the methods of the above embodiments may be implemented by a computer program to instruct relevant hardware, and the computer program 13 may be stored in a computer readable storage medium, and when executed, may include the processes of the embodiments of the methods as described above. The storage medium may be a magnetic disk, an optical disk, a Read-only memory (ROM), a Random Access Memory (RAM), a flash memory (FlashMemory), a hard disk (hard disk drive, abbreviated as HDD) or a Solid State Drive (SSD), etc.; the storage medium may also comprise a combination of memories of the kind described above.
The above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (6)
1. A wind turbine generator gearbox fault diagnosis method is characterized by comprising the following steps:
receiving a signal to be detected of the gearbox;
inputting the detection signals into a first feature extraction network model and a second feature extraction network model respectively, extracting first fault feature information of the signals to be detected through the first feature extraction network model, and extracting second fault feature information of the signals to be detected through the second feature extraction network model, wherein the depth of the second fault feature information is greater than that of the first fault feature information;
carrying out weighted fusion on the first fault characteristic information and the second fault characteristic information to obtain third fault characteristic information;
classifying the faults of the gearbox according to the third fault characteristic information to obtain a fault diagnosis result;
the first feature extraction network model is an improved Resnet50 network model and comprises a first volume block, a second volume block, a third volume block, a fourth volume block and a fifth volume block, and the first volume block comprises a first convolution kernel, a batch normalization function, an activation function and a maximum pooling layer; the second convolution block comprises a plurality of first trunk networks and a plurality of identity mapping units which are respectively fused with the first trunk networks correspondingly; the third convolution block includes a second backbone network; the fourth convolution block comprises a plurality of third trunk networks and a plurality of identity mapping units which are respectively fused with the third trunk networks correspondingly; the fifth convolution block comprises a plurality of fourth trunk networks and a plurality of identity mapping units which are respectively fused with the fourth trunk networks correspondingly;
the second feature extraction network model is an improved Resnet50 network model, the second feature extraction network model comprises a sixth volume block, a seventh volume block, an eighth volume block, a ninth volume block and a tenth volume block, the sixth volume block comprises a plurality of second convolution kernels, a batch normalization function, an activation function and a maximum pooling layer, and the second convolution kernels are smaller than the first convolution kernels; the seventh convolution block comprises a plurality of fifth trunk networks and a plurality of identity mapping units which are respectively fused with the fifth trunk networks correspondingly; the eighth convolution block comprises a plurality of sixth trunk networks and a plurality of identity mapping units which are respectively fused with the sixth trunk networks correspondingly; the ninth convolution block comprises a plurality of seventh trunk networks and a plurality of identity mapping units which are respectively fused with the seventh trunk networks; the tenth convolution block includes a plurality of eighth trunk networks and a plurality of identity mapping units respectively fused with the eighth trunk networks.
2. The wind turbine generator gearbox fault diagnosis method according to claim 1, wherein the receiving of the signal to be detected of the gearbox comprises:
receiving a one-dimensional vibration signal of the gearbox;
and converting the one-dimensional vibration signal into a two-dimensional image, wherein the signal to be detected is the two-dimensional image.
3. The wind turbine generator gearbox fault diagnosis method of claim 2, wherein converting the one-dimensional vibration signal into a two-dimensional image comprises:
and normalizing the one-dimensional vibration signal by adopting a symmetrical dot matrix image analysis method to obtain a two-dimensional image in a polar coordinate form.
4. A wind turbine generator system gear box fault diagnosis device is characterized by comprising:
the signal receiving module is used for receiving a signal to be detected of the gearbox;
the diagnosis module is used for inputting the detection signal into a first feature extraction network model and a second feature extraction network model, extracting first fault feature information of the signal to be detected through the first feature extraction network model, extracting second fault feature information of the signal to be detected through the second feature extraction network model, wherein the depth of the second fault feature information is larger than that of the first fault feature information, performing weighted fusion on the first fault feature information and the second fault feature information to obtain third fault feature information, and performing gearbox fault classification according to the third fault feature information to obtain a fault diagnosis result;
the first feature extraction network model is an improved Resnet50 network model and comprises a first volume block, a second volume block, a third volume block, a fourth volume block and a fifth volume block, and the first volume block comprises a first convolution kernel, a batch normalization function, an activation function and a maximum pooling layer; the second convolution block comprises a plurality of first trunk networks and a plurality of identity mapping units which are respectively fused with the first trunk networks correspondingly; the third convolution block includes a second backbone network; the fourth convolution block comprises a plurality of third trunk networks and a plurality of identity mapping units which are respectively fused with the third trunk networks correspondingly; the fifth convolution block comprises a plurality of fourth trunk networks and a plurality of identity mapping units which are respectively fused with the fourth trunk networks correspondingly;
the second feature extraction network model is an improved Resnet50 network model, the second feature extraction network model comprises a sixth volume block, a seventh volume block, an eighth volume block, a ninth volume block and a tenth volume block, the sixth volume block comprises a plurality of second convolution kernels, a batch normalization function, an activation function and a maximum pooling layer, and the second convolution kernels are smaller than the first convolution kernels; the seventh convolution block comprises a plurality of fifth trunk networks and a plurality of identity mapping units which are respectively fused with the fifth trunk networks correspondingly; the eighth convolution block comprises a plurality of sixth trunk networks and a plurality of identity mapping units which are respectively fused with the sixth trunk networks correspondingly; the ninth convolution block comprises a plurality of seventh main networks and a plurality of identity mapping units which are respectively fused with the seventh main networks correspondingly; the tenth convolution block includes a plurality of eighth trunk networks and a plurality of identity mapping units respectively fused with the eighth trunk networks.
5. An electronic device, comprising: a memory and a processor, wherein the memory and the processor are connected with each other in a communication manner, the memory stores computer instructions, and the processor executes the computer instructions to execute the wind turbine gearbox fault diagnosis method according to any one of claims 1 to 3.
6. A computer-readable storage medium storing computer instructions for causing a computer to perform the wind turbine generator gearbox fault diagnosis method of any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211037112.5A CN115114965B (en) | 2022-08-29 | 2022-08-29 | Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211037112.5A CN115114965B (en) | 2022-08-29 | 2022-08-29 | Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115114965A CN115114965A (en) | 2022-09-27 |
| CN115114965B true CN115114965B (en) | 2023-01-31 |
Family
ID=83335495
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211037112.5A Active CN115114965B (en) | 2022-08-29 | 2022-08-29 | Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115114965B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116091445B (en) * | 2023-01-05 | 2024-01-02 | 中国长江电力股份有限公司 | Ship lift gear and gear surface damage identification method based on deep learning |
| CN116223037B (en) * | 2023-05-09 | 2023-09-19 | 山东金帝精密机械科技股份有限公司 | Operation monitoring method and equipment for wind power bearing retainer |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091175A (en) * | 2018-10-23 | 2020-05-01 | 北京嘀嘀无限科技发展有限公司 | Neural network model training method, neural network model classification method, neural network model training device and electronic equipment |
| US10671878B1 (en) * | 2019-01-11 | 2020-06-02 | Capital One Services, Llc | Systems and methods for text localization and recognition in an image of a document |
| CN111753912A (en) * | 2020-06-28 | 2020-10-09 | 中国矿业大学 | Coal slime flotation clean coal ash content prediction method based on deep learning |
| CN114118251A (en) * | 2021-11-22 | 2022-03-01 | 国网辽宁省电力有限公司沈阳供电公司 | Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network |
| CN114004262A (en) * | 2021-11-29 | 2022-02-01 | 华东交通大学 | Gearbox bearing fault detection method and system |
| CN114818774A (en) * | 2022-03-15 | 2022-07-29 | 南京航空航天大学 | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network |
-
2022
- 2022-08-29 CN CN202211037112.5A patent/CN115114965B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115114965A (en) | 2022-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xu et al. | Online fault diagnosis method based on transfer convolutional neural networks | |
| CN115114965B (en) | Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium | |
| Zhang et al. | The multiclass fault diagnosis of wind turbine bearing based on multisource signal fusion and deep learning generative model | |
| Zhao et al. | Intelligent diagnosis using continuous wavelet transform and gauss convolutional deep belief network | |
| Fan et al. | Vibration signal denoising for structural health monitoring by residual convolutional neural networks | |
| Guo et al. | Multitask convolutional neural network with information fusion for bearing fault diagnosis and localization | |
| CN110161343B (en) | Non-invasive real-time dynamic monitoring method for external powered device of intelligent train | |
| Ye et al. | AKSNet: A novel convolutional neural network with adaptive kernel width and sparse regularization for machinery fault diagnosis | |
| CN113505655B (en) | Intelligent bearing fault diagnosis method for digital twin system | |
| CN110657984B (en) | Planetary gearbox fault diagnosis method based on reinforced capsule network | |
| CN111044570A (en) | Defect identification and early warning method and device for power equipment and computer equipment | |
| CN112380767B (en) | Equipment fault diagnosis method and system based on improved generation countermeasure network | |
| Li et al. | Joint attention feature transfer network for gearbox fault diagnosis with imbalanced data | |
| Huang et al. | 1DCNN fault diagnosis based on cubic spline interpolation pooling | |
| CN113158722A (en) | Rotary machine fault diagnosis method based on multi-scale deep neural network | |
| CN114548190A (en) | Wind turbine fault diagnosis method based on self-adaptive residual error neural network | |
| Soother et al. | The importance of feature processing in deep-learning-based condition monitoring of motors | |
| Sun et al. | Curvature enhanced bearing fault diagnosis method using 2D vibration signal | |
| CN116248392A (en) | Network malicious traffic detection system and method based on multi-head attention mechanism | |
| Li et al. | Fault diagnosis for rolling bearings of a freight train under limited fault data: Few-shot learning method | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| Liu et al. | Multi-Scale Fusion Attention Convolutional Neural Network for Fault Diagnosis of Aero-Engine Rolling Bearing | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN116541771A (en) | Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion | |
| Wang et al. | An improved MSCNN and GRU model for rolling bearing fault diagnosis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |