CN109947086A - Mechanical breakdown migration diagnostic method and system based on confrontation study - Google Patents
Mechanical breakdown migration diagnostic method and system based on confrontation study Download PDFInfo
- Publication number
- CN109947086A CN109947086A CN201910289486.8A CN201910289486A CN109947086A CN 109947086 A CN109947086 A CN 109947086A CN 201910289486 A CN201910289486 A CN 201910289486A CN 109947086 A CN109947086 A CN 109947086A
- Authority
- CN
- China
- Prior art keywords
- tape label
- training
- model
- data
- source domain
- 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.)
- Granted
Links
Landscapes
- Test And Diagnosis Of Digital Computers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of mechanical breakdown migration diagnostic methods and system based on confrontation study, wherein, this method comprises: the original signal for obtaining mechanical breakdown under different operating conditions carries out source domain training dataset and aiming field test data set that analysis generates the source domain training dataset of tape label under different operating conditions, not tape label；Fault diagnosis model is generated according to the source domain training dataset and back-propagation algorithm of tape label training depth convolutional neural networks model；Fault diagnosis model is trained according to the source domain training dataset of not tape label and aiming field test data set；The fault diagnosis model after training is finely adjusted according to the source domain training dataset of tape label and back-propagation algorithm；By the fault diagnosis model after the aiming field test data set input fine tuning of not tape label, the fault category of sample to be tested is exported.This method obtains domain invariant features by confrontation learning method, realizes the migration before not same area, realizes the intelligent diagnostics to variable working condition mechanical breakdown.
Description
Technical field
The present invention relates to mechanical fault diagnosis technical field, in particular to a kind of mechanical breakdown based on confrontation study
Migrate diagnostic method and system.
Background technique
With the gradually development of industrial technology, the demand to industrial equipment constantly rises, and present industrial system is integrated
Change scale is increasing, and individual equipment structure also becomes increasingly complex, and the degree of coupling in same system between distinct device is also got over
Come higher, on the one hand these factors are that mechanical equipment realizes that complex behavior establishes solid foundation, but also lead to whole system simultaneously
The probability to break down greatly increases.
The usually long-term continuous and steady operation of existing industrial system, failure are frequent low, but once break down, failure is bad
It is fast to change speed, failure causes to threaten greatly, if the major accident of fatal crass will be will lead to not in time by controlling.Therefore guarantee industry system
System long-period stable operation, the generation to avoid a nasty accident have important social effect, and can bring huge economic effect
Benefit and social benefit.
In order to ensure the safety of industrial system while push the development of intelligence manufacture, more and more industry complication systems
It goes to establish equipment running status monitoring system using industrial platform of internet of things, this makes the work of industrial system acquisition with storage magnanimity
Industry equipment operating data provides sufficient data source for the method for diagnosing faults based on data-driven.But opportunity is often adjoint
Challenge, since industrial system has the features such as environment high complexity, information incompleteness so that based on data-driven therefore
Barrier diagnostic techniques encounters huge challenge in real application.For the fault diagnosis technology based on data-driven, mainly
Problems faced is from data itself, and the characteristics of due to industrial equipment, resulting in the fault sample of equipment, to collect difficulty big, existing
The physical fault data class deposited is insufficient and sample is few, this will lead to " incomplete " property of data, and transfer learning method is to solve
The key technology of this problem.
Deep learning model based on data-driven is strong for the dependence of sample, needs abundant and a large amount of sample data
Acquisition could be trained to meet model actually required.The good result of depth model needs while meeting two conditions: 1) having and fill
The training sample of foot；2) training data should be similar as far as possible to the distribution of target application data.But in actual failure
In diagnostic task, the two conditions are often difficult to meet simultaneously, for example, industrial system usually requires to carry out function according to actual needs
The adjustment of rate load, and the variation of load changes the data space distribution that acquisition system is obtained, when using low negative
When collecting the obtained model of data training under the conditions of lotus and being applied to the data under diagnosis high-load condition, the performance of model will be by
Greatly influence.However, huge expense will be consumed by collecting data again under target application test condition, many times it is even more
It can not be achieved.Trained diagnostic model is carried out according to source domain training data and target application test data as a result,
Migration is to solve the important means of this problem.
Currently, migration troubleshooting issue still belongs to the elementary step, there are also very more technological difficulties needs further investigate into
Row is broken through.A kind of mechanical breakdown migration diagnostic method under variable working condition based on confrontation study proposed in this paper is exactly to solve this
A kind of effective technology method of one technical problem.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, an object of the present invention is to provide a kind of mechanical breakdowns based on confrontation study to migrate diagnostic method,
This method obtains domain invariant features by confrontation learning method, to realize the migration before not same area, realizes to variable working condition
The intelligent diagnostics of mechanical breakdown.
It is another object of the present invention to propose a kind of mechanical breakdown migration diagnostic system based on confrontation study.
In order to achieve the above objectives, one aspect of the present invention embodiment proposes a kind of mechanical breakdown migration based on confrontation study
Diagnostic method, comprising: S1 obtains the original signal of sample to be tested mechanical breakdown under different operating conditions, to the different operating conditions
The original signal of lower mechanical breakdown carries out the source that analysis generates the source domain training dataset of tape label under different operating conditions, not tape label
The aiming field test data set of domain training dataset, not tape label；
S2, according to the source domain training dataset and back-propagation algorithm of the tape label training depth convolutional neural networks mould
Type generates fault diagnosis model；
S3, according to the aiming field test data set, right of the source domain training dataset of the not tape label, the not tape label
The confrontation network of moving method and Wasserstein the distance guiding of anti-study carries out migration instruction to the fault diagnosis model
Practice；
S4, according to the source domain training dataset of the tape label and back-propagation algorithm to the failure after transfer training
Diagnostic model is finely adjusted；
S5, by the fault diagnosis model after the aiming field test data set input fine tuning of the not tape label, output
The fault category of the sample to be tested.
The mechanical breakdown based on confrontation study of the embodiment of the present invention migrates diagnostic method, by obtaining under different operating conditions
The original signal of mechanical breakdown generates the aiming field test data of the source domain training data and not tape label of tape label；Utilize band
The source domain training data of label uses back-propagation algorithm optimization as the convolutional neural networks of basic diagnostic model；Using not
The source domain training data and aiming field test data of tape label are obtained by the confrontation learning method of Wasserstein distance guidance
Domain invariant features are obtained, realize cross-domain migration；Using the source domain training data of tape label, using back-propagation algorithm to migration mould
Type is finely adjusted, and avoids migration problem；By the model after no labeled targets numeric field data input migration to be tested, failure is obtained
Classification results.
In addition, the mechanical breakdown migration diagnostic method according to the above embodiment of the present invention based on confrontation study can also have
There is following additional technical characteristic:
Further, in one embodiment of the invention, the S2 further comprises:
S21, by the source domain training dataset X of the tape labelsIt is randomly assigned, generates 70% training dataWith
30% test data
S22, from training dataIt is middle to extract the data that size is mPass through the back-propagation algorithm pair
Convolution Feature Mapping network unit in the convolutional neural networks model is optimized with the optimization of full link sort network unit；
S23, by the convolutional neural networks model after optimization to test dataIt is predicted and is calculated accurately
Rate thens follow the steps S22 if accuracy rate is less than default accuracy rate, conversely, then deconditioning saves all parameters obtained, it is raw
At the fault diagnosis model.
Further, in one embodiment of the invention, the S3 further comprises:
Using the source domain data of the not tape label and the target numeric field data of the not tape label, learnt based on the confrontation
Moving method the fault diagnosis model that step S2 is generated is migrated, using Wasserstein distance guiding
Confrontation network explore complex characteristic space, the distributional difference between different numeric field data features reduced by dual training strategy
Domain invariant features are obtained, finally realize cross-domain migration.
Further, in one embodiment of the invention, the step S3 and step S4 are specifically included:
Step 1, from the source domain training dataset X of the tape labelsWith the target numeric field data to be tested of the not tape label
Collect XtIt is middle to extract the transfer training data that size is mWith
Step 2, network D is evaluated using back-propagation algorithm optimization domainjWith full articulamentum Fcj:
Step 3, from the source domain training dataset X of the tape labelsFrom size is extracted for the microcall training data of m
Step 4, optimize the convolution Feature Mapping net in the convolutional neural networks model using the back-propagation algorithm
Network unit M and full link sort network unit C；
Step 5, when network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then stop instructing
Practice, keeps parameter obtained, complete transfer training and the fine tuning of the fault diagnosis model；Otherwise return step 1 is held again
Row step 1~5.
Further, in one embodiment of the invention, the step S3 further include:
The fault diagnosis model after transfer training realizes the mapping of feature by the depth convolutional neural networks,
Network is evaluated to instruct domain invariant features learning process, when the distribution nothing between source domain training data and aiming field training data in domain
When deviation, then the constant Feature Mapping in domain is obtained again.
In order to achieve the above objectives, another aspect of the present invention embodiment propose it is a kind of based on confrontation study mechanical breakdown move
Move diagnostic system, comprising: data generation module, the original letter of mechanical breakdown under the different operating conditions for obtaining sample to be tested
Number, the source domain training number that analysis generates tape label under different operating conditions is carried out to the original signal of mechanical breakdown under the different operating conditions
According to the aiming field test data set of collection, not the source domain training dataset of tape label, not tape label；
Model generation module, for according to the source domain training dataset and back-propagation algorithm of the tape label training depth
Convolutional neural networks model generates fault diagnosis model；
Model training module, for the target according to not the source domain training dataset of tape label, the not tape label
The confrontation network of domain test data set, the moving method of confrontation study and Wasserstein distance guiding is to the fault diagnosis
Model carries out transfer training；
Model finely tunes module, for being instructed according to the source domain training dataset and back-propagation algorithm of the tape label to migration
The fault diagnosis model after white silk is finely adjusted；
Diagnostic module, for the fault diagnosis after the aiming field test data set input fine tuning by the not tape label
Model exports the fault category of the sample to be tested.
The mechanical breakdown based on confrontation study of the embodiment of the present invention migrates diagnostic system, by obtaining under different operating conditions
The original signal of mechanical breakdown generates the aiming field test data of the source domain training data and not tape label of tape label；Utilize band
The source domain training data of label uses back-propagation algorithm optimization as the convolutional neural networks of basic diagnostic model；Using not
The source domain training data and aiming field test data of tape label are obtained by the confrontation learning method of Wasserstein distance guidance
Domain invariant features are obtained, realize cross-domain migration；Using the source domain training data of tape label, using back-propagation algorithm to migration mould
Type is finely adjusted, and avoids migration problem；By the model after no labeled targets numeric field data input migration to be tested, failure is obtained
Classification results.
In addition, the mechanical breakdown migration diagnostic system according to the above embodiment of the present invention based on confrontation study can also have
There is following additional technical characteristic:
Further, in one embodiment of the invention, the model generation module, comprising: data allocation unit, excellent
Change unit and determination unit；
The data allocation unit, for by the source domain training dataset X of the tape labelsIt is randomly assigned, generates 70%
Training dataWith 30% test data
The optimization unit is used for from training dataIt is middle to extract the data that size is mBy described anti-
To propagation algorithm to the convolution Feature Mapping network unit and full link sort network unit in the convolutional neural networks model
Optimization optimizes；
The determination unit, for passing through the convolutional neural networks model after optimizing to test dataIt carries out pre-
Accuracy rate is surveyed and calculates, when accuracy rate is greater than default accuracy rate, then deconditioning saves all parameters obtained, described in generation
Fault diagnosis model.
Further, in one embodiment of the invention, the model training module is also used to,
Using the source domain data of the not tape label and the target numeric field data of the not tape label, learnt based on the confrontation
Moving method the fault diagnosis model of generation is migrated, using the confrontation of Wasserstein distance guiding
Network explores complex characteristic space, is obtained by distributional difference that dual training strategy reduces between different numeric field data features
Domain invariant features finally realize cross-domain migration.
Further, in one embodiment of the invention, the model training module and the model finely tune module also
For,
From the source domain training dataset X of the tape labelsWith the aiming field data set X to be tested of the not tape labeltIn
Extract the transfer training data that size is mWithAnd network D is evaluated using back-propagation algorithm optimization domainj
With full articulamentum Fcj:
From the source domain training dataset X of the tape labelsFrom size is extracted for the microcall training data of m
Using the back-propagation algorithm optimize convolution Feature Mapping network unit M in the convolutional neural networks model with connect entirely
Sorter network unit C；
When network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then deconditioning, keeps
Parameter obtained completes transfer training and the fine tuning of the fault diagnosis model.
Further, in one embodiment of the invention, the model training module, is also used to,
The fault diagnosis model after transfer training realizes the mapping of feature by the depth convolutional neural networks,
Network is evaluated to instruct domain invariant features learning process, when the distribution nothing between source domain training data and aiming field training data in domain
When deviation, then the constant Feature Mapping in domain is obtained again.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is to migrate diagnostic method flow chart according to the mechanical breakdown based on confrontation study of one embodiment of the invention；
Fig. 2 is to migrate diagnostic method process according to the mechanical breakdown based on confrontation study of one specific embodiment of the present invention
Figure；
Fig. 3 is the totality side that diagnostic model is migrated according to the mechanical breakdown based on confrontation study of one embodiment of the invention
Case schematic diagram；
Fig. 4 be according in D → B migration task of one embodiment of the invention to the basic model feature of source domain training data
Visualize schematic diagram；
Fig. 5 is according to special to the basic model of aiming field test data in D → B migration task of one embodiment of the invention
Sign visualization schematic diagram；
Fig. 6 be according in D → B migration task of one embodiment of the invention to the migration models feature of source domain training data
Visualize schematic diagram；
Fig. 7 is according to special to the migration models of aiming field test data in D → B migration task of one embodiment of the invention
Sign visualization schematic diagram；
Fig. 8 is to migrate diagnostic system structural representation according to the mechanical breakdown based on confrontation study of one embodiment of the invention
Figure.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The mechanical breakdown migration diagnosis based on confrontation study proposed according to embodiments of the present invention is described with reference to the accompanying drawings
Method and system.
The mechanical breakdown migration based on confrontation study for describing to propose according to embodiments of the present invention with reference to the accompanying drawings first is examined
Disconnected method.
Fig. 1 is to migrate diagnostic method flow chart according to the mechanical breakdown based on confrontation study of one embodiment of the invention.
As shown in Figure 1, should based on confrontation study mechanical breakdown migration diagnostic method the following steps are included:
In step sl, the original signal for obtaining sample to be tested mechanical breakdown under different operating conditions, under different operating conditions
The original signal of mechanical breakdown carries out the source domain that analysis generates the source domain training dataset of tape label under different operating conditions, not tape label
The aiming field test data set of training dataset, not tape label.
Specifically, the original signal for obtaining the mechanical breakdown under different operating conditions, arranges and generates tape label under the 1st kind of operating condition
Source domain training datasetWith the aiming field data set to be tested of the 2nd kind of operating condition not tape labelWhereinWithIt is data set X respectivelysAnd XtI-th and jth sample, have n data in each sample
Point, msAnd mtIt is the total sample number of two datasets respectively,It is data set XsI-th of sample fault category label.
In step s 2, according to the source domain training dataset and back-propagation algorithm of tape label training depth convolutional Neural net
Network model generates fault diagnosis model.
Specifically, the source domain training data of tape label is utilizedUsing back-propagation algorithm training by
Convolutional layer can be capable of the event of Accurate classification with the depth convolutional neural networks model for connecting layer building entirely, acquisition to training data
Hinder diagnostic model.
It is understood that convolutional neural networks are a kind of nerves calculated using convolutional calculation method substitution universal matrix
Network.Typical convolutional calculation for image classification for, two-dimensional convolution is defined as:
For the migration troubleshooting issue of mechanical breakdown, processing is 1 dimension vibration signal, and therefore, the present invention is implemented
Example uses one-dimensional convolutional neural networks.By formula (1) it can easily be seen that two-dimensional convolution is changed into one-dimensional convolution when m is equal to 1.
One-dimensional convolutional calculation formula is as follows:
Wherein,It is that first of convolutional layer scale belongs to for n × 1 × jJ-th of convolution kernel；It is i-th defeated
The signal segment entered；bijIt is deviation；φ is activation primitive.
For our inventive embodiments failure modes problem, need that full articulamentum is added after continuous convolutional layer
The formula of output category result, full articulamentum is defined as follows:
yl=φ (Wlyl-1+bl) (3)
Wherein, WlIt is two layers of complete direct weight parameter of articulamentum, yl-1It is the Feature Mapping on upper layer, blIt is the inclined of current layer
Difference.
It is trained using convolutional neural networks of the back-propagation method to failure modes, sorter network includes for extracting
(its parameter is θ to the convolution Feature Mapping network unit M of the multilayer fault signature of data setM) and for establishing fault signature and sample
(its parameter is θ to the full link sort network unit C mapped between this labelc), it is pre- that the loss of model is defined as SoftMax function
Survey the cross entropy between probability distribution and the One-hot coding of source domain data label:
Wherein,It is indicator function,It is the prediction distribution value of kth dimension, K is categorical measure.
Further, in one embodiment of the invention, step S2 is specifically included:
S21, by the source domain training dataset X of tape labelsIt is randomly assigned, generates 70% training dataWith 30%
Test data
S22, from training dataIt is middle to extract the data that size is mOptimized using back-propagation algorithm and is rolled up
Convolution Feature Mapping network unit M and full link sort network unit C in product neural network update convolution according to following formula
Neural network:
S23, by the convolutional neural networks model after optimization to test dataAccuracy rate is predicted and is calculated, if
Accuracy rate is less than default accuracy rate Threc, S22 is thened follow the steps, conversely, then deconditioning saves all parameters obtained, it is raw
At fault diagnosis model.
In step s3, according to the source domain training dataset of not tape label, not the aiming field test data set of tape label, right
The confrontation network of moving method and Wasserstein the distance guiding of anti-study carries out transfer training to fault diagnosis model.
Further, using the source domain data of not tape label and target numeric field data, the moving method pair based on confrontation study
The basic diagnostic model that upper step obtains is migrated, and explores complicated spy using the confrontation network of Wasserstein distance guiding
Space is levied, domain invariant features are obtained by distributional difference that dual training strategy reduces between different numeric field data features, finally
Realize cross-domain migration.
In step s 4, according to the source domain training dataset of tape label and back-propagation algorithm to the failure after transfer training
Diagnostic model is finely adjusted.
Specifically, the source domain training data of tape label is utilizedUsing back-propagation algorithm to migration
The diagnostic model obtained afterwards is finely adjusted, and guarantees its applicability to source domain data set, and avoided migration problem.
It is understood that Wasserstein distance is a kind of distribution measure.Wasserstein measurement is a kind of
The metric form of distance between measurement data is distributed on Polish metric space (M, ρ).P and Q is distributed for two different probabilities
Measurement, Wasserstein range formula is defined as follows:
Wherein, ρ (x, y) be distance function, x and y from sample set M, ∏ (P, Q) be all edges that M × M concentrates P and Q
Probability.In an embodiment of the present invention, confrontation network training process is instructed using 1 rank Wasserstein distance function.Full
Under sufficient Kantorovich-Rubinstein duality, 1 rank Wasserstein distance function formula is as follows:
Wherein, the upper bound covers all 1-Lipschitz function f:x → R.
Wasserstein distance instructs dual training plan to get over, the objective function for the confrontation network that the embodiment of the present invention uses
It is defined as follows:
Wherein, D is 1-Lipschitz function, PgIt is the distribution for generating modelZ~p (z).In order to meet
A penalty term is added in Lipschitz constraint condition in objective function, asks to solve constraint condition bring optimization difficulty
Topic.Overall objective function as a result, is defined as:
λ is penalty coefficient.
In the migration fault diagnosis model of proposition, the mapping of feature, domain evaluation are realized using depth convolutional neural networks
Network instructs the learning processes of domain invariant features.When the distribution zero deflection between source domain and aiming field, will just domain be obtained not
The Feature Mapping of change.Meanwhile domain evaluator is the Wasserstein distance for estimating source domain and target domain characterization distribution.Source
Characteristic of field distributionIt is distributed with target domain characterizationWasserstein distance by maximize domain evaluate network D loss letter
Number LwdEstimate, wherein fs=F (xs) and ft=F (xt).Loss function is defined as follows:
Wherein, xsAnd xtCome from source domain XsWith aiming field XtSample data.In order to meet Wasserstein distance
Lipschitz constraint condition, propose on original objective function increase constraint condition penalty termFormula defines such as
Under:
Wherein,It is from distributionStochastical sampling.
Domain invariant features are obtained by dual training policy learning.Mainly consist of two parts during transfer training,
Firstly, training domain evaluator network goes to maximize the Wasserstein distance in two domains, then, fixed field evaluator network
Parameter reduces Wasserstein distance by adjusting the Feature Mapping in depth model.The learning strategy of domain invariant features is public
Formula is expressed as follows:
Wherein, λ is penalty coefficient.
For basic diagnostic model convolutional neural networks by migrating full link sort network before source domain and aiming field
Unit C obtains domain invariant features, Fcj(its parameter) it is j-th of full articulamentum in fully-connected network, it is commented using multiple domains
Valence network Dj(its parameter) go to estimate the Wasserstein distance between source domain and aiming field.Dual training strategy as a result,
Loss function is defined as:
Wherein,It is gradient penalty term, λ is penalty coefficient.
The fine tuning to migration models proposed is fighting net during invariant features fight repetitive exercise in domain each time
After network is optimized and revised, using source domain training dataset, convolution mind is adjusted by the back-propagation algorithm of formula (5) and formula (6)
Through the convolution Feature Mapping network unit M and full link sort network unit C in network, to be finely adjusted to migration models.
Further, in one embodiment of the invention, step S3 and step S4 are specifically included:
Step 1, from the source domain training dataset X of tape labelsWith the aiming field data set X to be tested of not tape labeltMiddle pumping
Taking size is the transfer training data of mWith
Step 2, network D is evaluated using back-propagation algorithm optimization domainjWith full articulamentumAnd more according to following formula
New network:
Step 3, from the source domain training dataset X of tape labelsFrom size is extracted for the microcall training data of m
Step 4, using back-propagation algorithm optimization convolutional neural networks model in convolution Feature Mapping network unit M with
Full link sort network unit updates convolutional neural networks according to following formula:
Step 5, when network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then stop instructing
Practice, keeps parameter obtained, complete transfer training and the fine tuning of fault diagnosis model；Otherwise return step 1 re-executes step
Rapid 1~5.
In step s 5, by the fault diagnosis model after the aiming field test data set input fine tuning of not tape label, output
The fault category of sample to be tested.
Specifically, by target numeric field data to be testedThe diagnostic model after moving is inputted, test can be obtained
The fault category of sample.
Diagnostic method is migrated to the mechanical breakdown of the invention based on confrontation study in the following with reference to the drawings and specific embodiments
It is described in detail.
It is a kind of to migrate diagnostic model overall plan as shown in Fig. 2 for the mechanical breakdown learnt under variable working condition based on confrontation.
Use depth convolutional neural networks as basic diagnosis model, wherein including multiple convolutional layers, full articulamentum and last
Softmax output layer utilizes source domain training dataset XsThe cross entropy of computation model prediction output and data label, by reversed
Propagation algorithm optimizes basic diagnostic model, and the mesh for capableing of Accurate classification to source domain training data is reached by repetitive exercise
Mark.On the basis of basic model, learn the constant spy in domain in each layer feature in depth model respectively using dual training strategy
Sign evaluates network using domain to evaluate the Wasserstein distance of source domain and target domain characterization distribution, gradually by repetitive exercise
Reduce cross-domain distribution distance.Purpose is to learn plan by confrontation on the basis of can be to source domain training data Accurate classification
Slightly, the feature distribution for adjusting simultaneously learning objective domain test data, makes it close to the feature distribution of source domain data, examines to solve failure
Due to the incomplete caused migration problem of data in disconnected practical application.As a result, in the dual training reduction of every generation not same area
After Wasserstein distance, increase is finely adjusted migration models using tape label source domain data, guarantees it to source domain data
Applicability avoided migration problem.
By taking the rolling bearing fault diagnosis task under certain 4 kinds different operating conditions as an example, to the mechanical breakdown based on confrontation network
Migration diagnostic method is verified.Under the conditions of 4 kinds of different driving motor loads, the vibration of bearing is acquired by vibrating sensor
Signal, sample frequency are set as 12kHz.It include 3 kinds of inner ring defect, outer ring defect and rolling volume defect events in Mishap Database
Barrier, each failure make 3 kinds of failure sizes, are having 10 kinds of classifications altogether plus normal condition, 500 sample datas of every kind of classification,
Specific as shown in table 1, table 1 is the description of different operating condition lower bearing fault data collection.
Table 1
The framework proposed according to embodiments of the present invention, for bearing failure diagnosis problem, as shown in figure 3, using convolution mind
Through network struction basis diagnostic model, the framework and details of convolutional neural networks are as shown in table 2, and table 2 is convolutional Neural net
Network model details.It goes to estimate the Wasserstein distance between the cross-domain distribution of each layer, domain evaluation using domain evaluation network
Network architecture Domain Details are as shown in table 3, and table 3 is that network details are evaluated in domain.
Table 2
Table 3
The convolutional neural networks training step of basis diagnosis:
Step 1: by source domain training dataset XsIt is randomly assigned, obtains 70% training dataWith 30% test number
According to
Step 2: from training dataIt is middle to extract the data that size is m
Step 3: using the convolution Feature Mapping network unit M in back-propagation algorithm optimization convolutional neural networks and entirely
Link sort network unit C updates convolutional neural networks according to following formula:
Step 4: using convolutional neural networks to test dataAccuracy rate is predicted and is calculated, when accuracy rate is less than
Given threshold ThrecWhen, return step 2 re-execute the steps 2~4；Otherwise deconditioning saves all parameters obtained, complete
At basic diagnostic model pre-training.
The dual training and trim step of migration models:
Step 1: from source domain training dataset XsWith aiming field test data set XtIt is middle to extract the transfer training number that size is m
According toWith
Step 2: evaluating network D using back-propagation algorithm optimization domainjWith full articulamentum Fcj, net is updated according to following formula
Network:
Step 3: from source domain training dataset XsFrom size is extracted for the microcall training data of m
Step 4: the convolution Feature Mapping network unit M optimized using back-propagation algorithm and full link sort network unit
C updates convolutional neural networks according to following formula:
Step 5: when the Wesserstein distance of domain evaluation network output is less than given threshold ThrewWhen, deconditioning is protected
Parameter obtained is held, the training and fine tuning of migration models are completed；Otherwise return step 1 re-execute the steps 1~5.
Hyper parameter setting in the training process are as follows: learning rate α=0.0001, mini-batch size is m=64, punishment
Coefficient lambda=10, basic diagnostic model test accuracy rate threshold value Threc=99%, Wasserstein distance threshold Threw=
0.0001。
Design the method that mutually migration is carried out between 4 type difference floor data collection to verify the present embodiment.Such as A →
B, using data set A as source domain training dataset, data set B is as aiming field test data set.Comparison between calculation results such as 4 institute of table
Show, table 4 is accuracy rate of diagnosis of the basic model from migration models in different migration tasks.The embodiment of the present invention as the result is shown
Moving method can greatly promote basic model to the accuracy rate of diagnosis of aiming field test data.
| A→B | A→C | A→D | B→A | B→C | B→D | |
| Basic model | 87.93% | 89.00% | 80.73% | 97.47% | 99.40% | 89.00% |
| Migration models | 99.73% | 99.67% | 100.00% | 99.13% | 100.00% | 99.93% |
| C→A | C→B | C→D | D→A | D→B | D→C | |
| Basic model | 97.00% | 97.20% | 89.53% | 90.20% | 75.53% | 79.26% |
| Migration models | 98.53% | 99.80% | 100.00% | 98.07% | 98.27% | 99.53% |
Fig. 4~7 are visually to be tied using s-TNE technology to diagnostic model domain invariant features in D → B migration task
Fruit figure.Fig. 4 and Fig. 5 be respectively basic model to the feature visualization of source domain training data and aiming field test data as a result, Fig. 6
It is respectively visualization result of the migration models to source domain training data and target detection data with Fig. 7.It can be obvious by result
Find out, target detection data and the basic diagnostic model feature distribution gap of source domain training data are significant, especially with classification 7,8,9
The most prominent, feature distribution space length is very big, it is clear that basic model is difficult to Accurate classification identification to aiming field test data.
And after method through the embodiment of the present invention is migrated, distribution distance of all categories in feature space is almost the same, this
Sample enables the model after migration being capable of accurately classification diagnosis to source domain training data and aiming field test data.
The mechanical breakdown based on confrontation study proposed according to embodiments of the present invention migrates diagnostic method, by obtaining not
With the original signal of mechanical breakdown under operating condition, the target domain test number of the source domain training data and not tape label of tape label is generated
According to；Using the source domain training data of tape label, use back-propagation algorithm optimization as the convolutional Neural net of basic diagnostic model
Network；Using the source domain training data and aiming field test data of not tape label, pass through the confrontation of Wasserstein distance guidance
Learning method obtains domain invariant features, realizes cross-domain migration；Using the source domain training data of tape label, calculated using backpropagation
Method is finely adjusted migration models, avoids migration problem；By the mould after no labeled targets numeric field data input migration to be tested
Type obtains failure modes result.
The mechanical breakdown migration diagnosis based on confrontation study proposed according to embodiments of the present invention referring next to attached drawing description
System.
Fig. 8 is to migrate diagnostic system structural representation according to the mechanical breakdown based on confrontation study of one embodiment of the invention
Figure.
As shown in figure 8, should based on confrontation study mechanical breakdown migration diagnostic system include: data generation module 100,
Model generation module 200, model training module 300, model fine tuning module 400 and diagnostic module 500.
Wherein, data generation module 100 is used to obtain the original signal of mechanical breakdown under the different operating conditions of sample to be tested,
Analysis is carried out to the original signal of mechanical breakdown under different operating conditions and generates the source domain training dataset of tape label under different operating conditions, no
The aiming field test data set of the source domain training dataset of tape label, not tape label.
Model generation module 200 is used for according to the source domain training dataset and back-propagation algorithm of tape label training depth volume
Product neural network model, generates fault diagnosis model.
Model training module 300 is used for the target domain test according to the not source domain training dataset of tape label, not tape label
The confrontation network of data set, the moving method of confrontation study and Wasserstein distance guiding moves fault diagnosis model
Move training.
Model finely tunes module 400 and is used for source domain training dataset and back-propagation algorithm according to tape label to transfer training
Fault diagnosis model afterwards is finely adjusted.
Diagnostic module 500 is used for the fault diagnosis model after the aiming field test data set input fine tuning by not tape label,
Export the fault category of sample to be tested.
The mechanical breakdown migration diagnostic system 10 based on confrontation study obtains domain invariant features by confrontation learning method,
To realize the migration before not same area, the intelligent diagnostics to variable working condition mechanical breakdown are realized.
Further, in one embodiment of the invention, model generation module, comprising: data allocation unit, optimization are single
Member and determination unit；
Data allocation unit, for by the source domain training dataset X of tape labelsIt is randomly assigned, generates 70% training number
According toWith 30% test data
Optimize unit, is used for from training dataIt is middle to extract the data that size is mIt is calculated by backpropagation
Method optimizes the convolution Feature Mapping network unit in convolutional neural networks model with the optimization of full link sort network unit；
Determination unit, for passing through the convolutional neural networks model after optimizing to test dataIt is predicted and is calculated
Accuracy rate, when accuracy rate is greater than default accuracy rate, then deconditioning saves all parameters obtained, generates fault diagnosis mould
Type.
Further, in one embodiment of the invention, model training module is also used to,
Using the source domain data and the target numeric field data of not tape label of not tape label, the moving method pair based on confrontation study
The fault diagnosis model of generation is migrated, and explores complex characteristic sky using the confrontation network of Wasserstein distance guiding
Between, domain invariant features are obtained by distributional difference that dual training strategy reduces between different numeric field data features, it is final to realize
Cross-domain migration.
Further, in one embodiment of the invention, model training module and model fine tuning module are also used to,
From the source domain training dataset X of tape labelsWith the aiming field data set X to be tested of not tape labeltMiddle extraction size is
The transfer training data of mWithAnd network D is evaluated using back-propagation algorithm optimization domainjWith full articulamentum Fcj:
From the source domain training dataset X of tape labelsFrom size is extracted for the microcall training data of mIt uses
Back-propagation algorithm optimizes convolution Feature Mapping network unit M and full link sort network unit in convolutional neural networks model
C；
When network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then deconditioning, keeps
Parameter obtained completes transfer training and the fine tuning of fault diagnosis model.
Further, in one embodiment of the invention, model training module is also used to,
The fault diagnosis model after transfer training realizes the mapping of feature by the depth convolutional neural networks,
Network is evaluated to instruct domain invariant features learning process, when the distribution nothing between source domain training data and aiming field training data in domain
When deviation, then the constant Feature Mapping in domain is obtained again.
It should be noted that the aforementioned explanation to the mechanical breakdown migration diagnostic method embodiment based on confrontation study
The system for being also applied for the embodiment, details are not described herein again.
The mechanical breakdown based on confrontation study proposed according to embodiments of the present invention migrates diagnostic system, by obtaining not
With the original signal of mechanical breakdown under operating condition, the target domain test number of the source domain training data and not tape label of tape label is generated
According to；Using the source domain training data of tape label, use back-propagation algorithm optimization as the convolutional Neural net of basic diagnostic model
Network；Using the source domain training data and aiming field test data of not tape label, pass through the confrontation of Wasserstein distance guidance
Learning method obtains domain invariant features, realizes cross-domain migration；Using the source domain training data of tape label, back-propagation algorithm is used
Migration models are finely adjusted, migration problem was avoided；Model after no labeled targets numeric field data input to be tested is migrated,
Obtain failure modes result.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. a kind of mechanical breakdown based on confrontation study migrates diagnostic method, which comprises the following steps:
S1 obtains the original signal of sample to be tested mechanical breakdown under different operating conditions, to mechanical breakdown under the different operating conditions
Original signal carry out the source domain training data that analysis generates the source domain training dataset of tape label under different operating conditions, not tape label
The aiming field test data set of collection, not tape label；
S2 trains depth convolutional neural networks model according to the source domain training dataset and back-propagation algorithm of the tape label,
Generate fault diagnosis model；
S3 is learned according to the source domain training dataset of the not tape label, the aiming field test data set of the not tape label, confrontation
The confrontation network of moving method and Wasserstein the distance guiding of habit carries out transfer training to the fault diagnosis model；
S4, according to the source domain training dataset of the tape label and back-propagation algorithm to the fault diagnosis after transfer training
Model is finely adjusted；
S5, by the fault diagnosis model after the input fine tuning of the aiming field test data set of the not tape label, described in output
The fault category of sample to be tested.
2. the method according to claim 1, wherein the S2 further comprises:
S21, by the source domain training dataset X of the tape labelsIt is randomly assigned, generates 70% training dataWith 30%
Test data
S22, from training dataIt is middle to extract the data that size is mBy the back-propagation algorithm to the volume
Convolution Feature Mapping network unit in product neural network model is optimized with the optimization of full link sort network unit；
S23, by the convolutional neural networks model after optimization to test dataAccuracy rate is predicted and is calculated, if
Accuracy rate is less than default accuracy rate, thens follow the steps S22, conversely, then deconditioning saves all parameters obtained, described in generation
Fault diagnosis model.
3. the method according to claim 1, wherein the S3 further comprises:
Using the source domain data of the not tape label and the target numeric field data of the not tape label, based on moving for the confrontation study
Shifting method migrates the fault diagnosis model that step S2 is generated, using pair of the Wasserstein distance guiding
Anti- network explores complex characteristic space, is obtained by distributional difference that dual training strategy reduces between different numeric field data features
Domain invariant features are taken, finally realize cross-domain migration.
4. the method according to claim 1, wherein the step S3 and step S4 are specifically included:
Step 1, from the source domain training dataset X of the tape labelsWith the aiming field data set X to be tested of the not tape labeltIn
Extract the transfer training data that size is mWith
Step 2, network D is evaluated using back-propagation algorithm optimization domainjWith full articulamentum Fcj:
Step 3, from the source domain training dataset X of the tape labelsFrom size is extracted for the microcall training data of m
Step 4, optimize the convolution Feature Mapping network list in the convolutional neural networks model using the back-propagation algorithm
First M and full link sort network unit C；
Step 5, when network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then deconditioning, is protected
Parameter obtained is held, transfer training and the fine tuning of the fault diagnosis model are completed；Otherwise return step 1 re-executes step
Rapid 1~5.
5. the method according to claim 1, wherein the step S3 further include:
The fault diagnosis model after transfer training realizes the mapping of feature by the depth convolutional neural networks, and domain is commented
Valence network instructs domain invariant features learning process, when the distribution zero deflection between source domain training data and aiming field training data
When, then the constant Feature Mapping in domain is obtained again.
6. a kind of mechanical breakdown based on confrontation study migrates diagnostic system characterized by comprising
Data generation module, the original signal of mechanical breakdown under the different operating conditions for obtaining sample to be tested, to the difference
The original signal of mechanical breakdown carries out analysis and generates the source domain training dataset of tape label, not tape label under different operating conditions under operating condition
Source domain training dataset, not tape label aiming field test data set；
Model generation module, for according to the source domain training dataset and back-propagation algorithm of the tape label training depth convolution
Neural network model generates fault diagnosis model；
Model training module, for being surveyed according to the aiming field of the not source domain training dataset of tape label, the not tape label
The confrontation network of data set, the moving method of confrontation study and Wasserstein distance guiding is tried to the fault diagnosis model
Carry out transfer training；
Model finely tune module, for the source domain training dataset and back-propagation algorithm according to the tape label to transfer training after
The fault diagnosis model be finely adjusted；
Diagnostic module, for the fault diagnosis mould after the aiming field test data set input fine tuning by the not tape label
Type exports the fault category of the sample to be tested.
7. system according to claim 6, which is characterized in that the model generation module, comprising: data allocation unit,
Optimize unit and determination unit；
The data allocation unit, for by the source domain training dataset X of the tape labelsIt is randomly assigned, generates 70% training
DataWith 30% test data
The optimization unit is used for from training dataIt is middle to extract the data that size is mPass through the reversed biography
Algorithm is broadcast to the convolution Feature Mapping network unit and the optimization of full link sort network unit in the convolutional neural networks model
It optimizes；
The determination unit, for passing through the convolutional neural networks model after optimizing to test dataIt is predicted simultaneously
Accuracy rate is calculated, when accuracy rate is greater than default accuracy rate, then deconditioning saves all parameters obtained, generates the failure
Diagnostic model.
8. system according to claim 6, which is characterized in that the model training module is also used to,
Using the source domain data of the not tape label and the target numeric field data of the not tape label, based on moving for the confrontation study
Shifting method migrates the fault diagnosis model of generation, using the confrontation network of the Wasserstein distance guiding
Complex characteristic space is explored, domain is obtained not by distributional difference that dual training strategy reduces between different numeric field data features
Become feature, finally realizes cross-domain migration.
9. system according to claim 6, which is characterized in that the model training module and model fine tuning module are also
For,
From the source domain training dataset X of the tape labelsWith the aiming field data set X to be tested of the not tape labeltIt is middle to extract greatly
The small transfer training data for mWithAnd network D is evaluated using back-propagation algorithm optimization domainjAnd Quan Lian
Meet a layer Fcj:
From the source domain training dataset X of the tape labelsFrom size is extracted for the microcall training data of mIt uses
The back-propagation algorithm optimizes convolution Feature Mapping network unit M and full link sort in the convolutional neural networks model
Network unit C；
When network D is evaluated in domainjThe Wesserstein distance of output is less than preset threshold ThrewWhen, then deconditioning, holding are obtained
The parameter obtained, completes transfer training and the fine tuning of the fault diagnosis model.
10. system according to claim 6, which is characterized in that the model training module is also used to,
The fault diagnosis model after transfer training realizes the mapping of feature by the depth convolutional neural networks, and domain is commented
Valence network instructs domain invariant features learning process, when the distribution zero deflection between source domain training data and aiming field training data
When, then the constant Feature Mapping in domain is obtained again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289486.8A CN109947086B (en) | 2019-04-11 | 2019-04-11 | Mechanical fault migration diagnosis method and system based on counterstudy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289486.8A CN109947086B (en) | 2019-04-11 | 2019-04-11 | Mechanical fault migration diagnosis method and system based on counterstudy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109947086A true CN109947086A (en) | 2019-06-28 |
| CN109947086B CN109947086B (en) | 2020-07-28 |
Family
ID=67014823
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910289486.8A Active CN109947086B (en) | 2019-04-11 | 2019-04-11 | Mechanical fault migration diagnosis method and system based on counterstudy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109947086B (en) |
Cited By (53)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414383A (en) * | 2019-07-11 | 2019-11-05 | 华中科技大学 | Convolutional neural networks based on Wasserstein distance fight transfer learning method and its application |
| CN110555273A (en) * | 2019-09-05 | 2019-12-10 | 苏州大学 | bearing life prediction method based on hidden Markov model and transfer learning |
| CN110567720A (en) * | 2019-08-07 | 2019-12-13 | 东北电力大学 | method for diagnosing depth confrontation of fault of fan bearing under unbalanced small sample scene |
| CN110728377A (en) * | 2019-10-21 | 2020-01-24 | 山东大学 | Intelligent fault diagnosis method and system for electromechanical equipment |
| CN110738107A (en) * | 2019-09-06 | 2020-01-31 | 上海衡道医学病理诊断中心有限公司 | microscopic image recognition and segmentation method with model migration function |
| CN110751207A (en) * | 2019-10-18 | 2020-02-04 | 四川大学 | Fault diagnosis method for anti-migration learning based on deep convolution domain |
| CN110796199A (en) * | 2019-10-30 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Image processing method and device and electronic medical equipment |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
| CN111027678A (en) * | 2019-12-04 | 2020-04-17 | 湃方科技(北京)有限责任公司 | Data migration method and device |
| CN111046581A (en) * | 2019-12-27 | 2020-04-21 | 国网江苏省电力有限公司电力科学研究院 | Power transmission line fault type identification method and system |
| CN111060318A (en) * | 2020-01-09 | 2020-04-24 | 山东科技大学 | Bearing fault diagnosis method based on deep countermeasure migration network |
| CN111158964A (en) * | 2019-11-26 | 2020-05-15 | 北京邮电大学 | Disk failure prediction method, system, device and storage medium |
| CN111314113A (en) * | 2020-01-19 | 2020-06-19 | 赣江新区智慧物联研究院有限公司 | Internet of things node fault detection method and device, storage medium and computer equipment |
| CN111428803A (en) * | 2020-03-31 | 2020-07-17 | 山东大学 | Wasserstein distance-based depth domain adaptive image classification method |
| CN111442926A (en) * | 2020-01-11 | 2020-07-24 | 哈尔滨理工大学 | Fault diagnosis method for rolling bearings of different models under variable load based on deep characteristic migration |
| CN111444780A (en) * | 2020-03-06 | 2020-07-24 | 同济大学 | Bearing fault diagnosis method based on deep sparse noise reduction self-coding network |
| CN111444952A (en) * | 2020-03-24 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Method and device for generating sample identification model, computer equipment and storage medium |
| CN111598161A (en) * | 2020-05-14 | 2020-08-28 | 哈尔滨工业大学（威海） | Engine gas circuit state diagnosis system based on CNN transfer learning |
| CN111614215A (en) * | 2020-05-11 | 2020-09-01 | 东南大学 | Method for designing driving motor for electric vehicle based on generation of countermeasure network |
| CN111612035A (en) * | 2020-04-18 | 2020-09-01 | 华为技术有限公司 | Method for training migration model, method and device for detecting fault |
| CN111626345A (en) * | 2020-05-15 | 2020-09-04 | 北京航空航天大学 | Multi-stage deep convolution transfer learning fault diagnosis method between different bearing devices |
| CN111651937A (en) * | 2020-06-03 | 2020-09-11 | 苏州大学 | Method for diagnosing similar self-adaptive bearing fault under variable working conditions |
| CN111814396A (en) * | 2020-07-02 | 2020-10-23 | 重庆大学 | Centrifugal fan fault early warning method based on transfer learning |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN111949796A (en) * | 2020-08-24 | 2020-11-17 | 云知声智能科技股份有限公司 | Resource-limited language speech synthesis front-end text analysis method and system |
| CN111949459A (en) * | 2020-08-10 | 2020-11-17 | 南京航空航天大学 | Hard disk failure prediction method and system based on transfer learning and active learning |
| CN111998936A (en) * | 2020-08-25 | 2020-11-27 | 四川长虹电器股份有限公司 | Equipment abnormal sound detection method and system based on transfer learning |
| CN112149350A (en) * | 2020-09-21 | 2020-12-29 | 浙江工业大学 | Model migration method for solving data dependence problem in fault diagnosis of heat pump system |
| CN112149726A (en) * | 2020-09-21 | 2020-12-29 | 浙江工业大学 | Totally-enclosed compressor fault diagnosis method based on knowledge sharing and model migration |
| CN112161784A (en) * | 2020-09-07 | 2021-01-01 | 华南理工大学 | Mechanical fault diagnosis method based on multi-sensor information fusion migration network |
| CN112257851A (en) * | 2020-10-29 | 2021-01-22 | 重庆紫光华山智安科技有限公司 | Model confrontation training method, medium and terminal |
| CN112329329A (en) * | 2020-09-22 | 2021-02-05 | 东北大学 | Simulation data driven rotary machine depth semi-supervised migration diagnosis method |
| CN112330063A (en) * | 2020-11-25 | 2021-02-05 | 新智数字科技有限公司 | Equipment failure prediction method, device and computer readable storage medium |
| CN112668633A (en) * | 2020-12-25 | 2021-04-16 | 浙江大学 | Adaptive graph migration learning method based on fine granularity field |
| CN112733943A (en) * | 2021-01-13 | 2021-04-30 | 浙江工业大学 | Heat pump fault diagnosis model migration method based on data mixed shearing technology |
| CN112836896A (en) * | 2021-03-03 | 2021-05-25 | 西门子工厂自动化工程有限公司 | Method for maintaining equipment and system for maintaining equipment |
| CN113076920A (en) * | 2021-04-20 | 2021-07-06 | 同济大学 | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model |
| CN113204280A (en) * | 2021-05-08 | 2021-08-03 | 山东英信计算机技术有限公司 | Method, system, equipment and medium for diagnosing power failure |
| CN113392881A (en) * | 2021-05-27 | 2021-09-14 | 重庆大学 | Rotary machine fault diagnosis method based on transfer learning |
| CN113538353A (en) * | 2021-07-05 | 2021-10-22 | 华北电力大学（保定） | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
| CN113569887A (en) * | 2021-01-18 | 2021-10-29 | 腾讯科技（深圳）有限公司 | Picture recognition model training and picture recognition method, device and storage medium |
| CN114021285A (en) * | 2021-11-17 | 2022-02-08 | 上海大学 | Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning |
| CN114021610A (en) * | 2021-09-10 | 2022-02-08 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | Fan fault recognition model training method and system based on transfer learning |
| CN114136622A (en) * | 2021-08-10 | 2022-03-04 | 南京航空航天大学 | DBN-DTL-based aeroengine gas circuit fault diagnosis method |
| CN114167837A (en) * | 2021-12-02 | 2022-03-11 | 中国路桥工程有限责任公司 | Intelligent fault diagnosis method and system for railway signal system |
| CN114239859A (en) * | 2022-02-25 | 2022-03-25 | 杭州海康威视数字技术股份有限公司 | Time sequence data prediction method and device based on transfer learning and storage medium |
| CN114305446A (en) * | 2021-12-25 | 2022-04-12 | 肇庆星网医疗科技有限公司 | Atrial fibrillation detection method and system based on artificial intelligence |
| WO2022141669A1 (en) * | 2020-12-31 | 2022-07-07 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
| CN115349111A (en) * | 2020-03-30 | 2022-11-15 | 三菱电机株式会社 | Diagnostic system, diagnostic method, and program |
| CN115562029A (en) * | 2022-10-17 | 2023-01-03 | 杭州天然气有限公司 | Intelligent control method and system for natural gas turbine expansion generator set |
| CN116009480A (en) * | 2023-03-24 | 2023-04-25 | 中科航迈数控软件(深圳)有限公司 | Fault monitoring method, device and equipment of numerical control machine tool and storage medium |
| CN116992953A (en) * | 2023-09-27 | 2023-11-03 | 苏州捷杰传感技术有限公司 | Model training method, fault diagnosis method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009205615A (en) * | 2008-02-29 | 2009-09-10 | Internatl Business Mach Corp <Ibm> | Change analysis system, method, and program |
| CN105628383A (en) * | 2016-02-01 | 2016-06-01 | 东南大学 | Bearing fault diagnosis method and system based on improved LSSVM transfer learning |
| CN108009633A (en) * | 2017-12-15 | 2018-05-08 | 清华大学 | A kind of Multi net voting towards cross-cutting intellectual analysis resists learning method and system |
| CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
| CN108875918A (en) * | 2018-08-14 | 2018-11-23 | 西安交通大学 | It is a kind of that diagnostic method is migrated based on the mechanical breakdown for being adapted to shared depth residual error network |
| CN109165695A (en) * | 2018-09-17 | 2019-01-08 | 重庆交通大学 | Piler method for diagnosing faults based on fault tree and transfer learning |
-
2019
- 2019-04-11 CN CN201910289486.8A patent/CN109947086B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009205615A (en) * | 2008-02-29 | 2009-09-10 | Internatl Business Mach Corp <Ibm> | Change analysis system, method, and program |
| CN105628383A (en) * | 2016-02-01 | 2016-06-01 | 东南大学 | Bearing fault diagnosis method and system based on improved LSSVM transfer learning |
| CN108009633A (en) * | 2017-12-15 | 2018-05-08 | 清华大学 | A kind of Multi net voting towards cross-cutting intellectual analysis resists learning method and system |
| CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
| CN108875918A (en) * | 2018-08-14 | 2018-11-23 | 西安交通大学 | It is a kind of that diagnostic method is migrated based on the mechanical breakdown for being adapted to shared depth residual error network |
| CN109165695A (en) * | 2018-09-17 | 2019-01-08 | 重庆交通大学 | Piler method for diagnosing faults based on fault tree and transfer learning |
Cited By (80)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414383A (en) * | 2019-07-11 | 2019-11-05 | 华中科技大学 | Convolutional neural networks based on Wasserstein distance fight transfer learning method and its application |
| CN110567720A (en) * | 2019-08-07 | 2019-12-13 | 东北电力大学 | method for diagnosing depth confrontation of fault of fan bearing under unbalanced small sample scene |
| CN110555273A (en) * | 2019-09-05 | 2019-12-10 | 苏州大学 | bearing life prediction method based on hidden Markov model and transfer learning |
| CN110738107A (en) * | 2019-09-06 | 2020-01-31 | 上海衡道医学病理诊断中心有限公司 | microscopic image recognition and segmentation method with model migration function |
| CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
| CN110907176B (en) * | 2019-09-30 | 2021-02-02 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
| CN110751207A (en) * | 2019-10-18 | 2020-02-04 | 四川大学 | Fault diagnosis method for anti-migration learning based on deep convolution domain |
| CN110751207B (en) * | 2019-10-18 | 2022-08-05 | 四川大学 | Fault diagnosis method for anti-migration learning based on deep convolution domain |
| CN110728377A (en) * | 2019-10-21 | 2020-01-24 | 山东大学 | Intelligent fault diagnosis method and system for electromechanical equipment |
| CN110728377B (en) * | 2019-10-21 | 2020-06-09 | 山东大学 | Intelligent fault diagnosis method and system for electromechanical equipment |
| CN110796199A (en) * | 2019-10-30 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Image processing method and device and electronic medical equipment |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN111158964B (en) * | 2019-11-26 | 2021-06-08 | 北京邮电大学 | Disk failure prediction method, system, device and storage medium |
| CN111158964A (en) * | 2019-11-26 | 2020-05-15 | 北京邮电大学 | Disk failure prediction method, system, device and storage medium |
| CN111027678A (en) * | 2019-12-04 | 2020-04-17 | 湃方科技(北京)有限责任公司 | Data migration method and device |
| CN111027678B (en) * | 2019-12-04 | 2023-08-04 | 湃方科技(北京)有限责任公司 | Data migration method and device |
| CN111046581A (en) * | 2019-12-27 | 2020-04-21 | 国网江苏省电力有限公司电力科学研究院 | Power transmission line fault type identification method and system |
| CN111046581B (en) * | 2019-12-27 | 2022-10-04 | 国网江苏省电力有限公司电力科学研究院 | Power transmission line fault type identification method and system |
| CN111060318B (en) * | 2020-01-09 | 2021-12-28 | 山东科技大学 | Bearing fault diagnosis method based on deep countermeasure migration network |
| CN111060318A (en) * | 2020-01-09 | 2020-04-24 | 山东科技大学 | Bearing fault diagnosis method based on deep countermeasure migration network |
| CN111442926A (en) * | 2020-01-11 | 2020-07-24 | 哈尔滨理工大学 | Fault diagnosis method for rolling bearings of different models under variable load based on deep characteristic migration |
| CN111314113A (en) * | 2020-01-19 | 2020-06-19 | 赣江新区智慧物联研究院有限公司 | Internet of things node fault detection method and device, storage medium and computer equipment |
| CN111444780A (en) * | 2020-03-06 | 2020-07-24 | 同济大学 | Bearing fault diagnosis method based on deep sparse noise reduction self-coding network |
| CN111444952B (en) * | 2020-03-24 | 2024-02-20 | 腾讯科技（深圳）有限公司 | Sample recognition model generation method, device, computer equipment and storage medium |
| CN111444952A (en) * | 2020-03-24 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Method and device for generating sample identification model, computer equipment and storage medium |
| CN115349111A (en) * | 2020-03-30 | 2022-11-15 | 三菱电机株式会社 | Diagnostic system, diagnostic method, and program |
| CN111428803A (en) * | 2020-03-31 | 2020-07-17 | 山东大学 | Wasserstein distance-based depth domain adaptive image classification method |
| CN111612035A (en) * | 2020-04-18 | 2020-09-01 | 华为技术有限公司 | Method for training migration model, method and device for detecting fault |
| CN111614215A (en) * | 2020-05-11 | 2020-09-01 | 东南大学 | Method for designing driving motor for electric vehicle based on generation of countermeasure network |
| CN111614215B (en) * | 2020-05-11 | 2021-11-12 | 东南大学 | Method for designing driving motor for electric vehicle based on generation of countermeasure network |
| CN111598161A (en) * | 2020-05-14 | 2020-08-28 | 哈尔滨工业大学（威海） | Engine gas circuit state diagnosis system based on CNN transfer learning |
| CN111626345A (en) * | 2020-05-15 | 2020-09-04 | 北京航空航天大学 | Multi-stage deep convolution transfer learning fault diagnosis method between different bearing devices |
| CN111651937A (en) * | 2020-06-03 | 2020-09-11 | 苏州大学 | Method for diagnosing similar self-adaptive bearing fault under variable working conditions |
| US11886311B2 (en) | 2020-06-03 | 2024-01-30 | Soochow University | Intra-class adaptation fault diagnosis method for bearing under variable working conditions |
| WO2021243838A1 (en) * | 2020-06-03 | 2021-12-09 | 苏州大学 | Fault diagnosis method for intra-class self-adaptive bearing under variable working conditions |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN111814396B (en) * | 2020-07-02 | 2024-02-20 | 重庆大学 | Centrifugal fan fault early warning method based on transfer learning |
| CN111814396A (en) * | 2020-07-02 | 2020-10-23 | 重庆大学 | Centrifugal fan fault early warning method based on transfer learning |
| CN111949459B (en) * | 2020-08-10 | 2022-02-01 | 南京航空航天大学 | Hard disk failure prediction method and system based on transfer learning and active learning |
| CN111949459A (en) * | 2020-08-10 | 2020-11-17 | 南京航空航天大学 | Hard disk failure prediction method and system based on transfer learning and active learning |
| CN111949796A (en) * | 2020-08-24 | 2020-11-17 | 云知声智能科技股份有限公司 | Resource-limited language speech synthesis front-end text analysis method and system |
| CN111949796B (en) * | 2020-08-24 | 2023-10-20 | 云知声智能科技股份有限公司 | Method and system for analyzing front-end text of voice synthesis of resource-limited language |
| CN111998936A (en) * | 2020-08-25 | 2020-11-27 | 四川长虹电器股份有限公司 | Equipment abnormal sound detection method and system based on transfer learning |
| CN112161784A (en) * | 2020-09-07 | 2021-01-01 | 华南理工大学 | Mechanical fault diagnosis method based on multi-sensor information fusion migration network |
| CN112149726B (en) * | 2020-09-21 | 2024-02-09 | 浙江工业大学 | Totally-enclosed compressor fault diagnosis method based on knowledge sharing and model migration |
| CN112149350B (en) * | 2020-09-21 | 2024-04-26 | 浙江工业大学 | Model migration method for solving data dependency problem in heat pump system fault diagnosis |
| CN112149726A (en) * | 2020-09-21 | 2020-12-29 | 浙江工业大学 | Totally-enclosed compressor fault diagnosis method based on knowledge sharing and model migration |
| CN112149350A (en) * | 2020-09-21 | 2020-12-29 | 浙江工业大学 | Model migration method for solving data dependence problem in fault diagnosis of heat pump system |
| CN112329329B (en) * | 2020-09-22 | 2024-02-20 | 东北大学 | Simulation data driven rotary machine depth semi-supervised migration diagnosis method |
| CN112329329A (en) * | 2020-09-22 | 2021-02-05 | 东北大学 | Simulation data driven rotary machine depth semi-supervised migration diagnosis method |
| CN112257851A (en) * | 2020-10-29 | 2021-01-22 | 重庆紫光华山智安科技有限公司 | Model confrontation training method, medium and terminal |
| CN112330063B (en) * | 2020-11-25 | 2024-03-26 | 新奥新智科技有限公司 | Equipment fault prediction method, equipment fault prediction device and computer readable storage medium |
| CN112330063A (en) * | 2020-11-25 | 2021-02-05 | 新智数字科技有限公司 | Equipment failure prediction method, device and computer readable storage medium |
| CN112668633B (en) * | 2020-12-25 | 2022-10-14 | 浙江大学 | Adaptive graph migration learning method based on fine granularity field |
| CN112668633A (en) * | 2020-12-25 | 2021-04-16 | 浙江大学 | Adaptive graph migration learning method based on fine granularity field |
| WO2022141669A1 (en) * | 2020-12-31 | 2022-07-07 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
| CN112733943A (en) * | 2021-01-13 | 2021-04-30 | 浙江工业大学 | Heat pump fault diagnosis model migration method based on data mixed shearing technology |
| CN112733943B (en) * | 2021-01-13 | 2024-03-22 | 浙江工业大学 | Heat pump fault diagnosis model migration method based on data mixed shearing technology |
| CN113569887A (en) * | 2021-01-18 | 2021-10-29 | 腾讯科技（深圳）有限公司 | Picture recognition model training and picture recognition method, device and storage medium |
| CN112836896A (en) * | 2021-03-03 | 2021-05-25 | 西门子工厂自动化工程有限公司 | Method for maintaining equipment and system for maintaining equipment |
| CN113076920A (en) * | 2021-04-20 | 2021-07-06 | 同济大学 | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model |
| CN113204280A (en) * | 2021-05-08 | 2021-08-03 | 山东英信计算机技术有限公司 | Method, system, equipment and medium for diagnosing power failure |
| CN113204280B (en) * | 2021-05-08 | 2023-09-26 | 山东英信计算机技术有限公司 | Method, system, equipment and medium for diagnosing power failure |
| CN113392881B (en) * | 2021-05-27 | 2023-04-18 | 重庆大学 | Rotary machine fault diagnosis method based on transfer learning |
| CN113392881A (en) * | 2021-05-27 | 2021-09-14 | 重庆大学 | Rotary machine fault diagnosis method based on transfer learning |
| CN113538353B (en) * | 2021-07-05 | 2023-09-01 | 华北电力大学（保定） | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
| CN113538353A (en) * | 2021-07-05 | 2021-10-22 | 华北电力大学（保定） | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
| CN114136622A (en) * | 2021-08-10 | 2022-03-04 | 南京航空航天大学 | DBN-DTL-based aeroengine gas circuit fault diagnosis method |
| CN114021610A (en) * | 2021-09-10 | 2022-02-08 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | Fan fault recognition model training method and system based on transfer learning |
| CN114021285A (en) * | 2021-11-17 | 2022-02-08 | 上海大学 | Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning |
| CN114021285B (en) * | 2021-11-17 | 2024-04-12 | 上海大学 | Rotary machine fault diagnosis method based on mutual local countermeasure migration learning |
| CN114167837B (en) * | 2021-12-02 | 2023-09-15 | 中国路桥工程有限责任公司 | Intelligent fault diagnosis method and system for railway signal system |
| CN114167837A (en) * | 2021-12-02 | 2022-03-11 | 中国路桥工程有限责任公司 | Intelligent fault diagnosis method and system for railway signal system |
| CN114305446A (en) * | 2021-12-25 | 2022-04-12 | 肇庆星网医疗科技有限公司 | Atrial fibrillation detection method and system based on artificial intelligence |
| CN114239859A (en) * | 2022-02-25 | 2022-03-25 | 杭州海康威视数字技术股份有限公司 | Time sequence data prediction method and device based on transfer learning and storage medium |
| CN114239859B (en) * | 2022-02-25 | 2022-07-08 | 杭州海康威视数字技术股份有限公司 | Power consumption data prediction method and device based on transfer learning and storage medium |
| CN115562029A (en) * | 2022-10-17 | 2023-01-03 | 杭州天然气有限公司 | Intelligent control method and system for natural gas turbine expansion generator set |
| CN116009480A (en) * | 2023-03-24 | 2023-04-25 | 中科航迈数控软件(深圳)有限公司 | Fault monitoring method, device and equipment of numerical control machine tool and storage medium |
| CN116992953A (en) * | 2023-09-27 | 2023-11-03 | 苏州捷杰传感技术有限公司 | Model training method, fault diagnosis method and device |
| CN116992953B (en) * | 2023-09-27 | 2024-04-19 | 苏州捷杰传感技术有限公司 | Model training method, fault diagnosis method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109947086B (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109947086A (en) | Mechanical breakdown migration diagnostic method and system based on confrontation study | |
| CN111539515B (en) | Complex equipment maintenance decision method based on fault prediction | |
| CN110361176A (en) | A kind of intelligent failure diagnosis method for sharing neural network based on multitask feature | |
| CN111274737A (en) | Method and system for predicting remaining service life of mechanical equipment | |
| CN105096053B (en) | A kind of health control decision-making technique suitable for complicated technology system | |
| CN107430715A (en) | Cascade identification in building automation | |
| CN107505837A (en) | A kind of semi-supervised neural network model and the soft-measuring modeling method based on the model | |
| CN109117883A (en) | SAR image sea ice classification method and system based on long memory network in short-term | |
| CN108304661A (en) | Diagnosis prediction method based on TDP models | |
| CN108764601A (en) | A kind of monitoring structural health conditions abnormal data diagnostic method based on computer vision and depth learning technology | |
| CN108053052B (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
| CN107742168A (en) | A kind of workshop bottleneck Forecasting Methodology based on Internet of Things technology | |
| CN106485594A (en) | A kind of main distribution integration incident response decision method | |
| CN106530082A (en) | Stock predication method and stock predication system based on multi-machine learning | |
| CN104634265B (en) | A kind of mineral floating froth bed soft measurement method of thickness based on multiplex images Fusion Features | |
| CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
| CN110895729A (en) | Prediction method for construction period of power transmission line engineering | |
| CN115860286B (en) | Air quality prediction method and system based on time sequence gate mechanism | |
| CN105868115A (en) | Building method and system for software test model of software intensive system | |
| CN107038450A (en) | Unmanned plane policing system based on deep learning | |
| Apeagyei et al. | Evaluation of deep learning models for classification of asphalt pavement distresses | |
| Zhao | A method of power supply health state estimation based on grey clustering and fuzzy comprehensive evaluation | |
| Liu et al. | Deep learning approach for sensor data prediction and sensor fault diagnosis in wind turbine blade | |
| CN110020737A (en) | Cereal temperature prediction technique and device | |
| CN107016440A (en) | The multiresolution deep neural network intelligent diagnosing method of machine driving failure |
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 |