CN111832663A - Capsule network rolling bearing fault diagnosis system based on gate control circulation unit - Google Patents

Capsule network rolling bearing fault diagnosis system based on gate control circulation unit Download PDF

Info

Publication number
CN111832663A
CN111832663A CN202010753312.5A CN202010753312A CN111832663A CN 111832663 A CN111832663 A CN 111832663A CN 202010753312 A CN202010753312 A CN 202010753312A CN 111832663 A CN111832663 A CN 111832663A
Authority
CN
China
Prior art keywords
layer
fault
capsule
network
gate
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.)
Pending
Application number
CN202010753312.5A
Other languages
Chinese (zh)
Inventor
王超群
李彬彬
焦斌
梁昱
程鹏宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Dianji University
Original Assignee
Shanghai Dianji University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Dianji University filed Critical Shanghai Dianji University
Priority to CN202010753312.5A priority Critical patent/CN111832663A/en
Publication of CN111832663A publication Critical patent/CN111832663A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G01M13/04—Bearings
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention relates to a fault diagnosis system for a capsule network rolling bearing based on a gated circulation unit, which comprises a gated layer, a convolution layer and a capsule network which are sequentially connected, wherein the capsule network comprises a primary capsule layer and a digital capsule layer, and the gated layer extracts the fault characteristics of a bearing; the convolution layer compresses the extracted fault characteristics; the primary capsule layer converts the extracted fault features from scalar features into vector features; the digital capsule layer identifies the fault characteristics converted into the vector characteristics and outputs corresponding fault types; the gate control layer is provided with a gate control circulation unit, the gate control circulation unit controls a target output state value of the gate control layer through an update gate and a reset gate, and the input of each layer of network in the capsule network comprises the output and the coupling coefficient of the previous layer of network. Compared with the prior art, the method has the advantages of improving the noise resistance, accuracy, universality, generalization capability and the like of bearing fault diagnosis.

Description

Capsule network rolling bearing fault diagnosis system based on gate control circulation unit
Technical Field
The invention relates to the field of fault diagnosis of mechanical equipment, in particular to a capsule network rolling bearing fault diagnosis system based on a gated circulation unit.
Background
The traditional fault diagnosis method mainly adopts a mode of manually extracting faults, such as a series of time domain and frequency domain analysis on vibration signals, current signals and temperature signals of a motor to obtain characteristics containing fault information, and then diagnoses the faults, wherein the methods comprise empirical mode decomposition, Teager energy operators, wavelet packet transformation and the like. With the development of artificial intelligence technology, more and more new algorithms are applied to the field of fault diagnosis. The algorithms input the fault into the model to extract the characteristics, and then optimize the characteristic to finally complete fault diagnosis. Classical algorithms include PCA, na iotave bayes, decision trees, support vector machines, deep neural networks, convolutional neural networks, and the like.
The traditional method for manually extracting the faults has strong specialty and is complex. Although some new algorithms based on artificial intelligence technology prove to be good in diagnosis effect in a fault diagnosis model, most research is carried out under an ideal condition, and the noise problem which may exist in an industrial practical environment is not considered. In addition, the prior art often only considers the situation of one load, does not consider the situation that the load is not the same during training and actual testing, and lacks the generalization capability of the model.
Disclosure of Invention
The invention aims to overcome the defects that noise and a plurality of loads in the industrial environment influence the accuracy of fault judgment in the prior art, and provides a fault diagnosis system for a capsule network rolling bearing based on a gated cycle unit.
The purpose of the invention can be realized by the following technical scheme:
the utility model provides a capsule network antifriction bearing fault diagnosis system based on gate circulation unit, includes gate layer, convolution layer and the capsule network that connects gradually, the capsule network includes primary capsule layer and digital capsule layer, wherein:
the gate control layer is used for extracting fault characteristics of the bearing;
a convolution layer for compressing the extracted fault features;
the primary capsule layer converts the extracted fault features from scalar features into vector features;
the digital capsule layer identifies the fault characteristics converted into the vector characteristics and outputs corresponding fault types;
the gate control layer is provided with a gate control circulation unit, the gate control circulation unit controls a target output state value of the gate control layer through an update gate and a reset gate, and the input of each layer of the capsule network comprises the output and the coupling coefficient of the previous layer of the network.
The original tensor size of the fault feature is 28 x 28, the number of nodes of the hidden layer of the gating layer is 128, and the fault feature with the tensor size of 10 x 10 is output.
The update gate determines an output historical state value which needs to be reserved for the target output state value, and a calculation formula is specifically as follows:
Zt＝(Wzxt+Uzht-1+bz)
wherein Z istTo update the state value of the gate, as a sigmoid function, xtIs the input vector at time t, WzAnd UzTo update the weight matrix of the gate, bzIs an offset term, ht-1Is the output historical state value of the gating layer.
Further, the reset gate determines an output candidate state value of the gate layer according to the output historical state value, and a calculation formula is specifically as follows:
rt＝(Wrxt+Urht-1+br)
wherein r istTo reset the state value of the gate, WrAnd UrTo reset the weight matrix of the gate, brIs the bias term.
Further, the calculation formula of the candidate state value is specifically as follows:
Figure BDA0002610710440000021
wherein the content of the first and second substances,
Figure BDA0002610710440000022
a candidate state value of the gate layer, tanh is an activation function, an exclusive OR operation, WcAnd UcAs a weight matrix of candidate states, bcIs the bias term.
Further, a calculation formula of the target output state value of the gate layer is specifically as follows:
Figure BDA0002610710440000023
wherein h istThe state value is output for the target of the gating layer.
The calculation formula of the output vector of each layer of the capsule network is specifically as follows:
Figure BDA0002610710440000024
Figure BDA0002610710440000025
Figure BDA0002610710440000026
wherein v isjFor the output vector of each layer of the network, sjAs intermediate vectors in each layer of the network, cijIn order to be able to determine the coupling coefficient,
Figure BDA0002610710440000031
as a prediction vector, WijIs a weight matrix of each layer network, uiIs the output vector of the previous layer capsule network.
The modulus values of the output vector are between [0, 1).
Further, a dynamic routing module for determining a coupling coefficient is arranged in the capsule network, the coupling coefficient is obtained by normalizing an initial vector, the initial vector is updated through a prediction vector and an output vector, and the specific process is as follows:
Figure BDA0002610710440000032
Figure BDA0002610710440000033
wherein, bijIs the initial vector of the j-th cycle, bij-1Is the initial vector of the last cycle.
The convolution layer has a convolution kernel size of 5 x 5 and a depth of 256.
And the convolution layer is provided with one batch normalization layer, and finally, the fault characteristics with the size of 6 multiplied by 6 tensor are output.
The number of channels output from the primary capsule layer is 32, each channel comprises 8 characteristic numbers, and the vectors with the characteristic numbers of 8 are encapsulated in one primary capsule.
The number of the capsules of the digital capsule layer is 10, and the vector dimension is 16.
The total loss function of the digital capsule layer comprises an edge loss function and a reconstruction loss function, and the calculation formula of the total loss function is as follows:
L＝Lc+αLr
wherein L is the total loss function of the digital capsule layer, LcAs a function of edge loss, LrTo reconstruct the loss function, α is the weight of the reconstruction loss function.
Further, the calculation formula of the edge loss function is specifically as follows:
Lc＝Tcmax(0，m+-||vc||)2+λ(1-Tc)max(0，||vc||-m-)2
wherein，TcThe real label is a real label, when the input sample class is consistent with the output fault class, the real label takes 1, otherwise, the real label is 0, | vcI is the modular length of the fault feature, i.e. the probability corresponding to the fault type, m+And m-Respectively an upper boundary and a lower boundary of the edge loss function, wherein lambda is a proportionality coefficient, and the proportion of the upper boundary and the lower boundary is adjusted;
the calculation formula of the reconstruction loss function is specifically as follows:
Lr＝(yout-xin)2
wherein, youtTo reconstruct an image, xinIs an input image.
The reconstruction process is composed of three full connection layers, and the corresponding activation functions are a relu function, a relu function and a sigmoid function.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, a gated circulation unit and a capsule network are combined, fault characteristics of a fault bearing are fully extracted through an update gate and a reset gate of the gated circulation unit, the extracted fault characteristics are converted into vector characteristics capable of representing the direction and the state of an object from scalar characteristics through a primary capsule layer after the convolution layer is compressed, the fault characteristics converted into the vector characteristics are identified through a digital capsule layer, a corresponding fault type is output, and the noise immunity, the accuracy, the universality and the generalization capability of bearing fault diagnosis are improved.
Drawings
FIG. 1 is a schematic structural view of the present invention;
FIG. 2 is a schematic diagram of a gated cycle cell according to the present invention;
FIG. 3 is a schematic diagram of the structure of the capsule network of the present invention;
fig. 4 is a schematic diagram of an overlapped sampling according to the present invention.
Reference numerals:
1-a gate-controlling layer; 2-a convolutional layer; 3-primary capsule layer; 4-digital capsule layer.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
As shown in fig. 1, a capsule network rolling bearing fault diagnosis system based on a gated circulation unit can improve noise immunity, universality and generalization capability of bearing fault diagnosis, and comprises a gated layer 1, a convolutional layer 2 and a capsule network which are sequentially connected, wherein the capsule network comprises a primary capsule layer 3 and a digital capsule layer 4, and wherein:
the gate control layer 1 is used for extracting fault characteristics of the bearing;
a convolution layer 2 for compressing the extracted fault features;
the primary capsule layer 3 is used for converting the extracted fault features into vector features from scalar features;
the digital capsule layer 4 identifies the fault characteristics converted into the vector characteristics and outputs corresponding fault types;
as shown in fig. 2, the gated layer 1 is provided with a gated loop unit, and the gated loop unit controls the target output state value of the gated layer 1 through an update gate and a reset gate, and the input of each layer of the capsule network comprises the output and the coupling coefficient of the previous layer of the capsule network.
The original tensor size of the fault feature is 28 × 28, the number of hidden layer nodes of the gated layer 1 is 128, and the fault feature with the tensor size of 10 × 10 is output.
And updating the output historical state value which needs to be reserved for determining the target output state value by the gate, wherein the calculation formula is specifically as follows:
Zt＝(Wzxt+Uzht-1+bz)
wherein Z istTo update the state value of the gate, as a sigmoid function, xtIs the input vector at time t, WzAnd UzTo update the weight matrix of the gate, bzIs an offset term, ht-1The output historical state value of gating layer 1.
The reset gate determines an output candidate state value of the gating layer 1 according to the output historical state value, and a calculation formula is specifically as follows:
rt＝(Wrxt+Urht-1+br)
wherein r istTo reset the state value of the gate, WrAnd UrTo reset the weight matrix of the gate, brIs the bias term.
The calculation formula of the candidate state value is specifically as follows:
Figure BDA0002610710440000051
wherein the content of the first and second substances,
Figure BDA0002610710440000052
a candidate state value of gate layer 1, tanh is an activation function, an exclusive OR operation, WcAnd UcAs a weight matrix of candidate states, bcIs the bias term.
The calculation formula of the target output state value of the gating layer 1 is specifically as follows:
Figure BDA0002610710440000053
wherein h istThe state value is output for the target of the gating layer.
As shown in fig. 3, the calculation formula of the output vector of each layer in the capsule network is specifically as follows:
Figure BDA0002610710440000054
Figure BDA0002610710440000055
Figure BDA0002610710440000056
wherein v isjFor the output vector of each layer of the network, sjAs intermediate vectors in each layer of the network, cijIn order to be able to determine the coupling coefficient,
Figure BDA0002610710440000057
as a prediction vector, WijIs a weight matrix of each layer network, uiIs the output vector of the previous layer capsule network.
The modulus values of the output vector are between [0, 1).
The capsule network is provided with a dynamic routing module for determining a coupling coefficient, the coupling coefficient is obtained by normalizing an initial vector, the initial vector is updated through a prediction vector and an output vector, and the specific process is as follows:
Figure BDA0002610710440000061
Figure BDA0002610710440000062
wherein, bijIs the initial vector of the j-th cycle, bij-1Is the initial vector of the last cycle.
The convolution layer 2 has a convolution kernel size of 5 × 5 and a depth of 256.
The convolutional layer 2 is provided with one batch normalization layer, and finally outputs failure characteristics of 6 × 6 tensor size.
The number of channels output from the primary capsule layer 3 is 32, each channel comprises 8 features, the total number of the features is 256, and the vectors with the feature number of 8 are packaged in one primary capsule, so that 1152 primary capsules are total.
The number of capsules of the digital capsule layer 4 is 10, and the vector dimension is 16.
The total loss function of the digital capsule layer 4 includes an edge loss function and a reconstruction loss function, and the calculation formula of the total loss function is as follows:
L＝Lc+αLr
wherein L is the total loss function of the digital capsule layer 4, LcAs a function of edge loss, LrFor the reconstruction loss function, α ═ 0.0005 is the weight of the reconstruction loss function.
The calculation formula of the edge loss function is specifically as follows:
Lc＝Tcmax(0，m+-||vc||)2+λ(1-Tc)max(0，||vc||-m-)2
wherein, TcThe real label is a real label, when the input sample class is consistent with the output fault class, the real label takes 1, otherwise, the real label is 0, | vcI is the modular length of the fault feature, i.e. the probability corresponding to the fault type, m+And m-Respectively an upper boundary and a lower boundary of the edge loss function, respectively taking 0.9 and 0.1 when | | vc||>0.9 or v | |c||<At 0.1, the loss function is 0, and λ ═ 0.5 is a proportionality coefficient;
the calculation formula of the reconstruction loss function is specifically as follows:
Lr＝(yout-xin)2
wherein, youtTo reconstruct an image, xinIs an input image.
The reconstruction process is composed of three fully-connected layers, the unit number of the three fully-connected layers is 256, 512 and 784, and the corresponding activation functions are a relu function, a relu function and a sigmoid function.
Example one
Setting the batch size to be 64, training all fault samples for 20 times, optimizing a total loss function through an Adam optimizer for 2 times of iteration of a dynamic routing algorithm module in the capsule network, wherein the learning rate is set to be 0.001, and a dynamic attenuation mode is adopted, and the attenuation rate is 10-8Meanwhile, a GRU network, a convolutional neural network and a deep neural network are adopted for carrying out comparison experiments.
The number of output nodes of the GRU network is 10, 10 fault states are represented, an Adam optimizer is used, the learning rate is set to be 0.001, and a cross entropy loss function is used as a loss function;
the convolutional neural network uses two layers of convolutional networks and two layers of full-link layers, the loss function uses a cross entropy loss function, and the specific parameter settings are shown in table 1:
TABLE 1 convolutional neural network parameter Table
Serial number Layer type Size of Depth of field Step size Output size All 0 filling
1 Convolutional layer 5 256 1 (24，24) Whether or not
2 Pooling layer 2 － 2 (12，12) Is that
3 Convolutional layer 3 32 1 (10，10) Whether or not
4 Pooling layer 2 － 2 (5，5) Is that
5 Full connection layer 512 － － 512 －
6 Full connection layer 10 － － 10 －
The deep neural network uses three full-connection layers, the number of nodes on the first layer is set to be 1024, the number of nodes on the second layer is set to be 512, the third layer is an output layer, the number of output nodes is 10, namely 10 states, the first two layers of activation functions are relu functions, the third layer of activation functions are softmax functions, an Adam optimizer is used, the learning rate is set to be 0.001, and the loss functions use cross entropy loss functions.
Acquiring a rolling bearing data set of the university of Keyssy storage, adopting drive end data with a signal sampling frequency of 12kHz, wherein the data types are divided into four states of a normal state, an inner ring fault, an outer ring fault and a rolling body fault, and are divided into three states of a fault diameter of 0.007, 0.014 and 0.021 inch in each fault state, so that ten states are provided, adopting drive end acceleration data of three loads of 0hp, 1hp and 2hp, wherein labels are 0-9 and respectively represent the normal state and 9 fault states, the data adopts a continuous sampling interception method, the sampling step length is 784, each state takes 1000 signal samples, the labels are added to finally obtain a matrix of 10000 x (784+1), and 10000 data are calculated as 7: 2: the proportion of 1 is divided into a training set, a verification set and a test set, namely 7000 training samples, 2000 verification samples and 1000 test samples are included, and the specific specification of the data set is shown in table 2:
TABLE 2 bearing Fault data sheet
Figure BDA0002610710440000071
Figure BDA0002610710440000081
In the original state, the data under 0hp, 1hp and 2hp are respectively adopted to carry out the experiment, and the experimental results are specifically shown in table 3:
TABLE 3 accuracy in original state table (%)
Figure BDA0002610710440000082
The experimental result shows that in the original state, the accuracy of the model of the invention is close to that of the GRU and CNN models, and the accuracy of the DNN model is lower;
in a noise state, Gaussian white noise is adopted as noise interference, no noise is added to a training set, only the Gaussian white noise with a plurality of signal-to-noise ratios (SNRs) is added to a verification set and a test set, and the calculation formula of the SNRs is as follows:
Figure BDA0002610710440000083
wherein, PsTo output signal power, PnIs the noise power;
gaussian white noise signals of 0dB to 8dB are added in the verification set and the test set, and the experimental results are shown in table 4 specifically:
table 4 accuracy table in noise state (%)
Figure BDA0002610710440000084
The experimental result shows that the accuracy of each model is reduced along with the reduction of the signal-to-noise ratio in the noise environment of 0dB to 8dB, the accuracy of each model can reach more than 80% in the noise environment of 4dB or more, but the accuracy of the model provided by the invention is the highest in the four models in the noise environment of 0dB to 8dB, particularly, the accuracy of GRU, CNN and DNN is greatly reduced when the signal-to-noise ratio is 0dB, and the model provided by the invention can also keep 94.3750% of accuracy, which indicates that the invention can also keep higher accuracy in the environment with larger noise pollution;
under the variable load state, when the load of the motor is changed, the rotating speed is also changed, so that signals measured by the sensor are also changed, and the experiment is respectively carried out when 0hp and 1hp are used as training sets and 2hp is used as a verification set and a test set; when 0hp and 2hp are used as training sets, 1hp is used as a verification and test set; when 1hp and 2hp are used as training sets and 0hp is used as a verification set and a test set, the experiment results are specifically shown in table 5:
TABLE 5 accuracy Rate table in varying load State (%)
Figure BDA0002610710440000091
The experimental result shows that when the load is changed, the accuracy of DNN is reduced fastest, other three models have higher accuracy under the variable load state, and particularly the model provided by the invention can achieve 90% of accuracy under three conditions, which shows that the invention has better generalization capability.
Example two
Acquiring bearing operation data of a laboratory motor data acquisition platform, performing data enhancement by an overlapping sampling method as shown in fig. 4 due to less data volume, wherein the sliding step length is 99, finally obtaining a total sample number of 8000, namely the sample numbers of a normal state and 7 fault states are respectively 1000, and setting labels to be 0-7. 8000 data were recorded as 7: 2: the proportion of 1 is divided into training set, verification set and test set, that is, containing 5600 training samples, 1600 verification samples and 800 test samples, and the specific specification of the data set is shown in table 6:
table 6 motor bearing fault data table
Label (R) Type of failure Training set Verification set Test set
0 Is normal 700 200 100
1 Stator turn-to-turn short circuit 2 turns 700 200 100
2 Stator 4 turns of turn-to-turn short circuit 700 200 100
3 Stator turn-to-turn short circuit 8 turns 700 200 100
4 Air gap eccentricity 700 200 100
5 Rotor broken bar 700 200 100
6 Bearing cage fracture 700 200 100
7 Bearing wear failure 700 200 100
The model, the GRU, the CNN and the DNN provided by the invention are respectively tested under the conditions of no noise and noise, the noise is Gaussian white noise, the signal-to-noise ratios are 0dB, 2dB, 4dB, 6dB and 8dB, the training times of the sample are 40 times, and the rest is the same as the first embodiment. The results of the experiment are shown in tables 7 and 8:
TABLE 7 diagnostic accuracy table (%) -of motor bearing in original state
Model (model) GRUCAPS GRU CNN DNN
Rate of accuracy 99.8698 90.2344 99.8698 94.2708
TABLE 8 diagnostic accuracy table (%) in noise state of motor bearing
Figure BDA0002610710440000101
The experimental result shows that under the condition of no noise, the accuracy of the model provided by the invention is close to that of CNN, and the accuracy of GRU and DNN is relatively low; under the condition of noise, the accuracy of GRU, CNN and DNN is reduced quickly, and the accuracy of the model provided by the invention can be kept above 80% under the noise environment, which shows that the invention has better universality.
In addition, it should be noted that the specific embodiments described in the present specification may have different names, and the above descriptions in the present specification are only illustrations of the structures of the present invention. All equivalent or simple changes in the structure, characteristics and principles of the invention are included in the protection scope of the invention. Various modifications or additions may be made to the described embodiments or methods may be similarly employed by those skilled in the art without departing from the scope of the invention as defined in the appending claims.

Claims (10)

1. The utility model provides a capsule network antifriction bearing fault diagnosis system based on gate circulation unit, its characterized in that, is including the gate layer (1), convolution layer (2) and the capsule network that connect gradually, the capsule network includes primary capsule layer (3) and digital capsule layer (4), wherein:
the gate control layer (1) is used for extracting fault characteristics of the bearing;
a convolution layer (2) for compressing the extracted fault features;
the primary capsule layer (3) converts the extracted fault features from scalar features into vector features;
the digital capsule layer (4) identifies the fault characteristics converted into the vector characteristics and outputs corresponding fault types;
the gating layer (1) is provided with a gating circulation unit, the gating circulation unit controls a target output state value of the gating layer (1) through an updating gate and a resetting gate, and the input of each layer of network in the capsule network comprises the output and the coupling coefficient of the previous layer of network.
2. The system for diagnosing the fault of the rolling bearing of the capsule network based on the gated cyclic unit according to claim 1, wherein the update gate determines an output historical state value which is required to be reserved for a target output state value, and a calculation formula is specifically as follows:
Zt＝(Wzxt+Uzht-1+bz)
wherein Z istTo update the state value of the gate, as a sigmoid function, xtIs the input vector at time t, WzAnd UzTo update the weight matrix of the gate, bzIs an offset term, ht-1Is the output historical state value of the gating layer (1).
3. The system for diagnosing the fault of the capsule network rolling bearing based on the gated cyclic unit as claimed in claim 2, wherein the reset gate determines the output candidate state value of the gated layer (1) according to the output historical state value, and the calculation formula is specifically as follows:
rt＝(Wrxt+Urht-1+br)
wherein r istTo reset the state value of the gate, WrAnd UrTo reset the weight matrix of the gate, brIs the bias term.
4. The system for diagnosing the fault of the capsule network rolling bearing based on the gated cyclic unit according to claim 3, wherein the calculation formula of the candidate state values is specifically as follows:
Figure FDA0002610710430000011
wherein the content of the first and second substances,
Figure FDA0002610710430000012
is a candidate state value of the gate layer (1), tanh is an activation function, an exclusive OR operation, WcAnd UcAs a weight matrix of candidate states, bcIs the bias term.
5. The system for diagnosing the fault of the capsule network rolling bearing based on the gated cyclic unit according to claim 4, wherein the calculation formula of the target output state value of the gated layer (1) is specifically as follows:
Figure FDA0002610710430000021
wherein h istThe state value is output for the target of the gating layer.
6. The system for diagnosing the fault of the rolling bearing of the capsule network based on the gated cyclic unit according to claim 1, wherein a calculation formula of an output vector of each layer of the capsule network is specifically as follows:
Figure FDA0002610710430000022
Figure FDA0002610710430000023
Figure FDA0002610710430000024
wherein v isjFor the output vector of each layer of the network, sjAs intermediate vectors in each layer of the network, cijIn order to be able to determine the coupling coefficient,
Figure FDA0002610710430000025
as a prediction vector, WijIs a weight matrix of each layer network, uiIs the output vector of the previous layer capsule network.
7. The system for diagnosing the rolling bearing fault of the capsule network based on the gated cyclic unit according to claim 6, wherein a dynamic routing module for determining a coupling coefficient is arranged in the capsule network, the coupling coefficient is obtained by normalizing an initial vector, and the initial vector is updated through a prediction vector and an output vector, and the specific process is as follows:
Figure FDA0002610710430000026
Figure FDA0002610710430000027
wherein, bijIs the initial vector of the j-th cycle, bij-1Is the initial vector of the last cycle.
8. The capsule network rolling bearing fault diagnosis system based on the gated cyclic unit of claim 1, wherein a batch normalization layer is provided in the convolution layer (2) in number of one layer.
9. The system for diagnosing the fault of the capsule network rolling bearing based on the gated cyclic unit according to claim 1, wherein the total loss function of the digital capsule layer (4) comprises an edge loss function and a reconstruction loss function, and the calculation formula of the total loss function is as follows:
L＝Lc+αLr
wherein L is the total loss function of the digital capsule layer (4), LcAs a function of edge loss, LrTo reconstruct the loss function, α is the weight of the reconstruction loss function.
10. The system for diagnosing the fault of the capsule network rolling bearing based on the gated cyclic unit according to claim 9, wherein the calculation formula of the edge loss function is specifically as follows:
Lc＝Tcmax(0，m+-||vc||)2+λ(1-Tc)max(0，||vc||-m-)2
wherein, TcFor a genuine label, | | vcI is the modular length of the fault signature, m+And m-Respectively an upper boundary and a lower boundary of the edge loss function, wherein lambda is a proportionality coefficient;
the calculation formula of the reconstruction loss function is specifically as follows:
Lr＝(yout-xin)2
wherein, youtTo reconstruct an image, xinIs an input image.
CN202010753312.5A 2020-07-30 2020-07-30 Capsule network rolling bearing fault diagnosis system based on gate control circulation unit Pending CN111832663A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010753312.5A CN111832663A (en) 2020-07-30 2020-07-30 Capsule network rolling bearing fault diagnosis system based on gate control circulation unit

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010753312.5A CN111832663A (en) 2020-07-30 2020-07-30 Capsule network rolling bearing fault diagnosis system based on gate control circulation unit

Publications (1)

Publication Number Publication Date
CN111832663A true CN111832663A (en) 2020-10-27

Family

ID=72920665

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010753312.5A Pending CN111832663A (en) 2020-07-30 2020-07-30 Capsule network rolling bearing fault diagnosis system based on gate control circulation unit

Country Status (1)

Country Link
CN (1) CN111832663A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112565128A (en) * 2020-11-28 2021-03-26 西北工业大学 Radio signal modulation recognition network based on hybrid neural network and implementation method
CN113255882A (en) * 2021-04-30 2021-08-13 南通大学 Bearing fault diagnosis method based on improved convolution capsule network
CN113255458A (en) * 2021-04-29 2021-08-13 燕山大学 Bearing fault diagnosis method based on multi-view associated feature learning
CN113592071A (en) * 2021-06-22 2021-11-02 广州大学 Equipment fault recognition model training and recognition method, system, device and medium
CN114227382A (en) * 2022-01-18 2022-03-25 湖北汽车工业学院 Cutter damage monitoring system and method based on novel capsule network

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108960077A (en) * 2018-06-12 2018-12-07 南京航空航天大学 A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network
CN111026095A (en) * 2019-12-30 2020-04-17 太原科技大学 Fault diagnosis method with noise label based on recurrent neural network
CN111046916A (en) * 2019-11-20 2020-04-21 上海电机学院 Motor fault diagnosis method and system based on void convolution capsule network
CN111259157A (en) * 2020-02-20 2020-06-09 广东工业大学 Chinese text classification method based on hybrid bidirectional circulation capsule network model
CN111459144A (en) * 2020-05-01 2020-07-28 西北工业大学 Airplane flight control system fault prediction method based on deep cycle neural network
CN111458148A (en) * 2020-04-26 2020-07-28 上海电机学院 CBAM-based convolutional neural network rolling bearing fault diagnosis method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108960077A (en) * 2018-06-12 2018-12-07 南京航空航天大学 A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network
CN111046916A (en) * 2019-11-20 2020-04-21 上海电机学院 Motor fault diagnosis method and system based on void convolution capsule network
CN111026095A (en) * 2019-12-30 2020-04-17 太原科技大学 Fault diagnosis method with noise label based on recurrent neural network
CN111259157A (en) * 2020-02-20 2020-06-09 广东工业大学 Chinese text classification method based on hybrid bidirectional circulation capsule network model
CN111458148A (en) * 2020-04-26 2020-07-28 上海电机学院 CBAM-based convolutional neural network rolling bearing fault diagnosis method
CN111459144A (en) * 2020-05-01 2020-07-28 西北工业大学 Airplane flight control system fault prediction method based on deep cycle neural network

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112565128A (en) * 2020-11-28 2021-03-26 西北工业大学 Radio signal modulation recognition network based on hybrid neural network and implementation method
CN113255458A (en) * 2021-04-29 2021-08-13 燕山大学 Bearing fault diagnosis method based on multi-view associated feature learning
CN113255882A (en) * 2021-04-30 2021-08-13 南通大学 Bearing fault diagnosis method based on improved convolution capsule network
CN113592071A (en) * 2021-06-22 2021-11-02 广州大学 Equipment fault recognition model training and recognition method, system, device and medium
CN114227382A (en) * 2022-01-18 2022-03-25 湖北汽车工业学院 Cutter damage monitoring system and method based on novel capsule network

Similar Documents

Publication Publication Date Title
CN111832663A (en) Capsule network rolling bearing fault diagnosis system based on gate control circulation unit
CN112418013B (en) Complex working condition bearing fault diagnosis method based on meta-learning under small sample
CN111046916A (en) Motor fault diagnosis method and system based on void convolution capsule network
CN111458148A (en) CBAM-based convolutional neural network rolling bearing fault diagnosis method
CN113203566B (en) Motor bearing fault diagnosis method based on one-dimensional data enhancement and CNN
CN111753891B (en) Rolling bearing fault diagnosis method based on unsupervised feature learning
CN110657984A (en) Planetary gearbox fault diagnosis method based on reinforced capsule network
CN112577748A (en) Reinforced lightweight multi-scale CNN-based rolling bearing fault diagnosis method
CN112308147A (en) Rotating machinery fault diagnosis method based on integrated migration of multi-source domain anchor adapter
CN113723489A (en) Rolling bearing fault identification method for improved relation network
CN112132102B (en) Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN112926728B (en) Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor
CN114386526A (en) Combined convolution neural network diagnosis method for rotary machine fault
CN116304861A (en) Time-frequency characteristic fusion fault diagnosis method based on self-attention
CN113032917A (en) Electromechanical bearing fault detection method based on generation countermeasure and convolution cyclic neural network and application system
CN114091504A (en) Rotary machine small sample fault diagnosis method based on generation countermeasure network
CN112651426A (en) Fault diagnosis method for rolling bearing of wind turbine generator
Yang et al. Few-shot learning for rolling bearing fault diagnosis via siamese two-dimensional convolutional neural network
CN115545070A (en) Intelligent diagnosis method for unbalance-like bearing based on comprehensive balance network
CN113758709A (en) Rolling bearing fault diagnosis method and system combining edge calculation and deep learning
CN114459760A (en) Rolling bearing fault diagnosis method and system under strong noise environment
CN111783941B (en) Mechanical equipment diagnosis classification method based on probability confidence convolutional neural network
CN113221996A (en) Bearing fault diagnosis method and system based on multi-head attention mechanism
CN114781427B (en) Wind generating set antifriction bearing fault diagnosis system based on information fusion technology
CN115204234A (en) Fault diagnosis method for wind driven generator

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20201027

RJ01 Rejection of invention patent application after publication