CN110533070A - A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample - Google Patents
A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample Download PDFInfo
- Publication number
- CN110533070A CN110533070A CN201910684544.7A CN201910684544A CN110533070A CN 110533070 A CN110533070 A CN 110533070A CN 201910684544 A CN201910684544 A CN 201910684544A CN 110533070 A CN110533070 A CN 110533070A
- Authority
- CN
- China
- Prior art keywords
- original shape
- migration
- mechanical
- signal
- small sample
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses the mechanical breakdown intelligent diagnosing methods based on migration original shape network under a kind of small sample, the present invention carries out feature extraction and running state recognition to mechanical signal using depth convolutional neural networks, the sensitive features in mechanical signal can effectively be extracted, get rid of traditionally characteristic extraction procedure to the dependence of artificial experience；By the present invention in that the principle clustered with original shape, realizes under conditions of retrievable sample number is few, obtains the validity feature of signal, get rid of conventional machines learning method to the dependence of huge data volume；The present invention crosses the principle using transfer learning, by source domain data related but with different characteristic distribution, further increases the generalization ability of network；The present invention effectively can carry out fault diagnosis to mechanical equipment under Small Sample Database, improve the accuracy rate under Small Sample Database to mechanical fault diagnosis by combining depth convolutional neural networks, original shape network and transfer learning thought.
Description
Technical field
The invention belongs to mechanical fault diagnosis fields, and in particular to based on migration original shape network under a kind of small sample
Mechanical breakdown intelligent diagnosing method.
Background technique
Currently, conventional mechanical devices intelligent fault diagnosis technology is still relied on from great amount of samples Learning Samples feature.So
And in actual condition, the complexity of security consideration and mechanical equipment working environment for mechanical equipment is difficult to obtain
The fault-signal of mechanical equipment, the fault-signal quantity got is few, and type is also few.In the fewer feelings of available number of samples
Under condition, intelligent diagnostics algorithm is difficult to sufficiently learn to arrive the validity feature of data sample, and small sample problem seriously affects intelligent diagnostics
Accuracy of the algorithm to the condition monitoring and fault diagnosis of mechanical equipment.Therefore, it is necessary to mechanical equipment under Small Sample Size
The new technology of intelligent fault diagnosis is studied with new method.
Summary of the invention
The purpose of the present invention is to provide the mechanical breakdown intelligent diagnostics sides based on migration original shape network under a kind of small sample
Method, it is not necessary to depend on a large amount of tape label data, overcome due to fault signal of mechanical equipment it is few caused by intelligent diagnostics algorithm without
The problem of method sufficiently learns is used to train network, and enables state-estimator using only the mechanical signal for being less than total amount of data 1%
Enough obtain 95% or more state classification accuracy.
In order to achieve the above object, the present invention the following steps are included:
Step 1, using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, to the machinery got
Signal is standardized pretreatment；
Step 2 establishes the migration original shape network model for being used for mechanical signal state recognition, migrates original shape network model packet
Three network minor structures of module and condition identifier are adapted to containing feature extractor, domain；
Step 3, the migration original shape network model combination Euclidean distance that step 2 is established and Largest Mean diversity factor
Amount, is trained and updates the network parameter of feature extractor and condition identifier, to keep feature extractor and domain suitable
It answers module to realize and obtains the characteristic information of each state using minute quantity tape label data, and then obtain and represented respectively in target domain space
The original shape of state；
Step 4, to the migration original shape network model input data that step 2 is established, input data is less than total data
Measure 1% aiming field mechanical signal and by under other different working conditions, with target numeric field data there is different characteristic to be distributed
Source domain mechanical signal with operating status label, migration original shape network model output is operating status corresponding to each data
Probability value；
Step 5, the migration original shape network model established to step 2 prevent from training over-fitting using Dropout method
And stablize training process, to make network more rapidly more stable completion status classification work；
Step 6, will be Step 3: the migration original shape network model after step 4 and step 5 training combines, using being less than
Total amount of data 1% aiming field mechanical signal training network, and enable state-estimator obtain 95% or more state classification
Accuracy, the final intelligent trouble diagnosis to mechanical equipment realized under Small Sample Database.
Pretreatment is standardized using the method for zero-mean standardization, calculation formula to the mechanical signal got are as follows:
In formula, niFor the data point number of i-th of input signal, xijFor j-th of data of i-th of input signal,For
The mean value of i-th of input signal, siFor the sample standard deviation of i-th of input signal, XijFor i-th after zero-mean standardization processing
J-th of data in a new signal.
Feature extractor is made of four layers of convolutional layer and four layers of pond layer, and feature extractor uses first floor convolutional layer for big volume
The product big step-length of core, the approach configuring parameters that the convolution kernel of each convolutional layer and step-length gradually reduce later.Specifically, each layer is set
Convolution kernel size in convolutional layer is respectively 9,7,5,3, and the step sizes being arranged in each layer convolutional layer are respectively 2,2,1,1.
Condition identifier is made of two layers of full articulamentum, and the output vector of condition identifier represents input signal in aiming field
The position in space, and for judging the state that input signal is in.
Distance calculation formula in target domain space are as follows:
In formula, njIndicate that the dimension of target domain space, D () indicate the output vector and a certain state of input signal
The Euclidean distance of original shape, YiIndicate the output vector of i-th of input signal, YijIndicate j-th of output valve of output vector, PkTable
Show the original shape of k-th of signal condition, PkjIndicate j-th of data value of the original shape of k-th of signal condition.
In step 3, the renewal process of migration original shape network model combination Euclidean distance is calculated for accelerating training process
Formula are as follows:
P′k=Pk-lrp·(Pi-Pk)
In formula, P 'kIt is the original shape of updated k-th of signal condition, lrpIt is the learning rate of the step, PiIt is and P 'kEurope
Formula is apart from the smallest original shape.
In step 3, migration original shape network model combination Largest Mean difference measurement is by the data-optimized original shape of source domain
Renewal process, calculation formula are as follows:
In formula, XsIt is the input signal from source domain, XtIt is the input signal from aiming field, φ () is feature space
Mapping, ns、ntRespectively indicate the number of source domain signal Yu aiming field signal；To reduceFor update feature extractor and
One of foundation of the network parameter of condition identifier enhances the excavation to aiming field data characteristics by source domain data.
In step 4, the probability value of output is the original shape vector of the output vector and each state by migration original shape network model
Euclidean distance be calculated, calculation formula are as follows:
yi=[p (Xi,1),p(Xi,2),...,p(Xi,k),...,p(Xi,nk)]
In formula, p (Xi, k) and it is that i-th of input signal belongs to kth class shape probability of state, yiBelong to each shape for input signal
The loss function of the set of probability of state, the part uses cross entropy loss function, calculation formula are as follows:
In formula, LosscFor the penalty values of cross entropy loss function, y is the actual label information of input signal,Network is defeated
Prediction label out；The total loss function of neural network is the synthesis for intersecting entropy loss and MMD measurement, calculation formula are as follows:
In formula, Loss is the total loss function of neural network, and λ is the adjusting ginseng for adjusting two partial loss function specific gravity
Number.
Migration original shape network model prevents from training over-fitting and stablizes the specific method of training process using Dropout method
It is as follows:
It is retained the neural unit node in each layer with specified probability, remaining node is hidden, i.e., in training process
The parameter of only retained neuron is updated；
In next round training, and probability will be reassigned and retain partial nerve member at random, until training terminates.
In step 6, network is trained using the aiming field mechanical signal for being less than total amount of data 1%, and enable state-estimator
Enough obtain 95% or more state classification accuracy, it was demonstrated that this method can extract validity feature and be completed under condition of small sample
Classification task.
Compared with prior art, the present invention carries out feature extraction and operation to mechanical signal using depth convolutional neural networks
State recognition can effectively extract the sensitive features in mechanical signal, get rid of traditionally characteristic extraction procedure and pass through to artificial
The dependence tested；By the present invention in that the principle clustered with original shape, is realized under conditions of retrievable sample number is few, obtain
The validity feature of signal gets rid of conventional machines learning method to the dependence of huge data volume；The present invention is crossed to be learned using migration
The principle of habit further increases the generalization ability of network by source domain data related but with different characteristic distribution；The present invention
It, can be effectively right under Small Sample Database by combining depth convolutional neural networks, original shape network and transfer learning thought
Mechanical equipment carries out fault diagnosis, improves the accuracy rate under Small Sample Database to mechanical fault diagnosis.The present invention uses
Original shape migration models combination neural network adaptive learning feature ability and original shape clustering method data volume is required it is little
The advantages of, input signal can be mapped into target domain space, convert operating status for the Euclidean distance of target domain space
Probability value.It is aided with transfer learning thought, maximum difference in migration measurement is used for parameter and is updated, improves model by source domain data and exist
Study in target numeric field data further increases the generalization ability of model.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is network structure of the invention；
Fig. 3 is the accuracy rate comparison diagram of network and other methods of the invention.
Specific embodiment
The present invention will be further described with reference to the accompanying drawing.
The present invention carries out feature extraction to mechanical signal using depth convolutional neural networks, is learned based on original shape network and migration
Learning method enables model effectively to extract the validity feature in mechanical signal from minute quantity sample.Specifically include following step
It is rapid:
Step 1: using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, the machinery got being believed
Number it is standardized pretreatment；
Step 2: establishing the migration original shape network for being used for mechanical signal state recognition, which mentions comprising feature
Device, domain is taken to adapt to three network minor structures of module and state-estimator；
Step 3: the neural network model established to step 2 is instructed in conjunction with Euclidean distance and Largest Mean difference measurement
Practice and update the network parameter of feature extractor and condition identifier, is realized so that feature extractor and domain be made to adapt to module
The characteristic information of each state is obtained using minute quantity tape label data, and then obtains and represents the original of each state in target domain space
Shape；
Step 4: the neural network model established to step 2, mode input data are the aiming field less than total amount of data 1%
Mechanical signal and the band operating status label by having different characteristic distribution under other different working conditions, with target numeric field data
Source domain mechanical signal, model output be each data corresponding to operating status probability value；
Step 5: the neural network model established to step 2 prevents from training over-fitting and stabilization using Dropout method
Training process, to make network more rapidly more stable completion status classification work；
Step 6: the original in conjunction with designed by step 3,4,5 migrates original shape network model, uses the mesh for being less than total amount of data 1%
Domain mechanical signal training network is marked, to make feature extractor and the study of domain adaptation module to the original of aiming field mechanical signal state
Shape, and enable state-estimator obtain 95% or more state classification accuracy, it is final realize under Small Sample Database to machine
The intelligent trouble diagnosis of tool equipment.
The present invention carries out zero-mean standardization pretreatment to a small amount of mechanical signal got；It establishes and is used for mechanical signal shape
The neural network model of state identification；In conjunction with Euclidean distance and Largest Mean difference measurement, it is trained and updates feature extractor
And the network parameter of condition identifier；Input data is less than the aiming field mechanical signal of total amount of data 1% and by other
There is the source domain mechanical signal with operating status label of different characteristic distribution under different working condition, with source domain data；It uses
Dropout method prevents trained over-fitting and stablizes training process, the final intelligence event to mechanical equipment realized under small sample
Barrier diagnosis.Present invention saves the manpower and material resources for manually extracting feature, get rid of conventional mechanical devices intelligent fault diagnosis
Dependence of the algorithm to a large amount of tape label samples, further improves the generalization ability of network in conjunction with transfer learning thought, in sample
State classification accuracy rate with higher under the conditions of this.
Embodiment:
It is used certain include four kinds of bearing operating statuses data set one is shared normal, ball failure, inner ring failure and
Four kinds of rolling bearing operating statuses of outer ring failure, every kind of operating status include 800 samples, in total include 3200 samples.It takes
20 samples therein are as training data, and for remaining 3180 samples as test data, training sample data amount only accounts for gross sample
The 0.625% of notebook data amount.Taking other one includes certain data set of same four kinds of bearing operating statuses, data sample tool therein
There is different feature distributions, takes part of sample as source domain data for supplemental training.
As shown in Figure 1, the present invention the following steps are included:
Step 1: pretreatment being standardized to the vibration signal of the four kinds of operating statuses of rolling bearing got, uses zero
Mean value standardization, calculating formula are as follows:
In formula, niFor the data point number of i-th of input signal, xijFor j-th of data of i-th of input signal,For
The mean value of i-th of input signal, siFor the sample standard deviation of i-th of input signal, XijFor i-th after zero-mean standardization processing
J-th of data in a new signal.
Step 2: establishing the migration original shape network model for being used for mechanical signal state recognition, which includes spy
Levy extractor, domain adapts to three network minor structures of module and state-estimator.
Step 3: the neural network model established to step 2 is instructed in conjunction with Euclidean distance and Largest Mean difference measurement
Practice and update the network parameter of feature extractor and condition identifier, and then obtains and represent the original of each state in target domain space
Shape.
Target domain space Central Europe formula is apart from calculating formula are as follows:
In formula, njIndicate that the dimension of target domain space, D () indicate the output vector and a certain state of input signal
The Euclidean distance of original shape, YiIndicate the output vector of i-th of input signal, YijIndicate j-th of output valve of output vector, PkTable
Show the original shape of k-th of signal condition, PkjIndicate j-th of data value of the original shape of k-th of signal condition.Euclidean distance is used simultaneously
Optimize the renewal process of original shape to accelerate training process, calculating formula are as follows:
P′k=Pk-lrp·(Pi-Pk)
In formula, P 'kIt is the original shape of updated k-th of signal condition, lrpIt is the learning rate of the step, PiIt is and P 'kEurope
Formula is apart from the smallest original shape.Using Largest Mean difference measurement, by the renewal process of the data-optimized original shape of source domain, calculating formula
Are as follows:
In formula, XsIt is the input signal from source domain data set, XtIt is the input signal from aiming field data set, φ
() is feature space mapping, ns、ntRespectively indicate the number of source domain signal Yu aiming field signal.To reduceFor more
One of foundation of the network parameter of new feature extractor and condition identifier, it is special to target numeric field data by the enhancing of source domain data
The excavation of sign.
Step 4: the neural network model established to step 2, mode input data are four kinds of states totally 20 aiming field machines
Tool signal and by have under other different working conditions, with target numeric field data different characteristic distribution with operating status label
Source domain mechanical signal, model output are the probability value of operating status corresponding to each data.The probability value exported be by
The output vector of model and the Euclidean distance calculating of the original shape vector of each state are got, calculating formula are as follows:
yi=[p (Xi,1),p(Xi,2),...,p(Xi,k),...,p(Xi,nk)]
In formula, p (Xi, k) and it is that i-th of input signal belongs to kth class shape probability of state, yiBelong to each shape for input signal
The set of probability of state.The loss function of the part uses cross entropy loss function, calculating formula are as follows:
In formula, LosscFor the penalty values of cross entropy loss function, y is the actual label information of input signal,Network is defeated
Prediction label out.The total loss function of neural network is the synthesis for intersecting entropy loss and MMD measurement, calculating formula are as follows:
In formula, Loss is the total loss function of neural network, and λ is the adjusting ginseng for adjusting two partial loss function specific gravity
Number.
Step 5: the neural network model established to step 2 prevents from training over-fitting and stabilization using Dropout method
Training process, to make network more rapidly more stable completion status classification work.
Dropout method refers in a wheel training, makes the neural unit node in each layer first to specify probability quilt
Retain, remaining node is hidden, i.e., the parameter of neuron only retained is updated in training process.In next round training
In, and partial nerve member will be retained at random with specified probability again, until training terminates.
Step 6: the neural network model in conjunction with designed by step 3,4,5 uses the 20 aiming field mechanical signals selected
Training network to make feature extractor and the study of domain adaptation module to the original shape of aiming field mechanical signal state, and makes state
Judging device obtains 99.06% state classification accuracy, realizes the intelligent trouble diagnosis to mechanical equipment under Small Sample Database.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
It describes in detail bright.It should be understood that the above is only a specific embodiment of the present invention, it is not intended to restrict the invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent substitution, improvement and etc. done should be included in guarantor of the invention
Within the scope of shield.
Claims (9)
1. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample, which is characterized in that including following
Step:
Step 1, using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, to the mechanical signal got
It is standardized pretreatment；
Step 2, establishes the migration original shape network model for being used for mechanical signal state recognition, and migration original shape network model includes spy
Levy extractor, domain adapts to three network minor structures of module and condition identifier；
Step 3, the migration original shape network model combination Euclidean distance that step 2 is established and Largest Mean difference measurement, into
Row training and the network parameter for updating feature extractor and condition identifier, so that feature extractor and domain be made to adapt to module
It realizes and obtains the characteristic information of each state using minute quantity tape label data, and then obtain and represent each state in target domain space
Original shape；
Step 4, to the migration original shape network model input data that step 2 is established, input data is less than total amount of data 1%
Aiming field mechanical signal and by under other different working conditions, with target numeric field data have different characteristic distribution band run
The source domain mechanical signal of state tag, migration original shape network model output are the probability of operating status corresponding to each data
Value；
Step 5, the migration original shape network model established to step 2 prevent from training over-fitting and steady using Dropout method
Training process is determined, to make network more rapidly more stable completion status classification work；
Step 6, will be Step 3: the migration original shape network model after step 4 and step 5 training combines, using less than sum
According to amount 1% aiming field mechanical signal training network, and enable state-estimator obtain 95% or more state classification it is correct
Rate, the final intelligent trouble diagnosis to mechanical equipment realized under Small Sample Database.
2. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, being standardized pretreatment using the method for zero-mean standardization to the mechanical signal got.
3. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, feature extractor is made of four layers of convolutional layer and four layers of pond layer, feature extractor use first floor convolutional layer for
The big step-length of big convolution kernel, the approach configuring parameters that the convolution kernel of each convolutional layer and step-length gradually reduce later.
4. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, condition identifier is made of two layers of full articulamentum, the output vector of condition identifier represents input signal in mesh
The position of domain space is marked, and for judging the state that input signal is in.
5. the mechanical breakdown intelligent diagnostics side based on migration original shape network under a kind of small sample according to claim 1 or 4
Method, which is characterized in that distance calculation formula in target domain space are as follows:
In formula, njIndicate the dimension of target domain space, the output vector of D () expression input signal and a certain state original shape
Euclidean distance, YiIndicate the output vector of i-th of input signal, YijIndicate j-th of output valve of output vector, PkIndicate kth
The original shape of a signal condition, PkjIndicate j-th of data value of the original shape of k-th of signal condition.
6. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, the renewal process of migration original shape network model combination Euclidean distance is counted for accelerating training process in step 3
Calculate formula are as follows:
P′k=Pk-lrp·(Pi-Pk)
In formula, P 'kIt is the original shape of updated k-th of signal condition, lrpIt is the learning rate of the step, PiIt is and PkIt is European away from
From the smallest original shape.
7. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, migration original shape network model combination Largest Mean difference measurement is by the data-optimized original of source domain in step 3
The renewal process of shape, calculation formula are as follows:
In formula, XsIt is the input signal from source domain, XtIt is the input signal from aiming field, φ () is feature space mapping,
ns、ntRespectively indicate the number of source domain signal Yu aiming field signal；To reduceTo update feature extractor and state
One of foundation of the network parameter of arbiter enhances the excavation to aiming field data characteristics by source domain data.
8. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, the probability value of output is the original shape of the output vector and each state by migration original shape network model in step 4
The Euclidean distance of vector is calculated, calculation formula are as follows:
yi=[p (Xi,1),p(Xi,2),...,p(Xi,k),...,p(Xi,nk)]
In formula, p (Xi, k) and it is that i-th of input signal belongs to kth class shape probability of state, yiBelong to each state for input signal
The loss function of the set of probability, the part uses cross entropy loss function, calculation formula are as follows:
In formula, LosscFor the penalty values of cross entropy loss function, y is the actual label information of input signal,Network output
Prediction label；The total loss function of neural network is the synthesis for intersecting entropy loss and MMD measurement, calculation formula are as follows:
In formula, Loss is the total loss function of neural network, and λ is the adjustment parameter for adjusting two partial loss function specific gravity.
9. the mechanical breakdown intelligent diagnosing method based on migration original shape network under a kind of small sample according to claim 1,
It is characterized in that, migration original shape network model prevents from training over-fitting and stablizes the specific of training process using Dropout method
Method is as follows:
It is retained the neural unit node in each layer with specified probability, remaining node is hidden, i.e., in training process only
The parameter of retained neuron is updated；
In next round training, and probability will be reassigned and retain partial nerve member at random, until training terminates.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910684544.7A CN110533070A (en) | 2019-07-26 | 2019-07-26 | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910684544.7A CN110533070A (en) | 2019-07-26 | 2019-07-26 | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110533070A true CN110533070A (en) | 2019-12-03 |
Family
ID=68660546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910684544.7A Pending CN110533070A (en) | 2019-07-26 | 2019-07-26 | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110533070A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110780146A (en) * | 2019-12-10 | 2020-02-11 | 武汉大学 | Transformer fault identification and positioning diagnosis method based on multi-stage transfer learning |
| 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 |
| 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 |
| CN111401454A (en) * | 2020-03-19 | 2020-07-10 | 创新奇智(重庆)科技有限公司 | Few-sample target identification method based on transfer learning |
| CN111537207A (en) * | 2020-04-29 | 2020-08-14 | 西安交通大学 | Data enhancement method for intelligent diagnosis of mechanical fault under small sample |
| 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 |
| CN111738413A (en) * | 2020-06-04 | 2020-10-02 | 东华大学 | Spinning full-process energy consumption monitoring method based on feature self-matching transfer learning |
| CN111982514A (en) * | 2020-08-12 | 2020-11-24 | 河北工业大学 | Bearing fault diagnosis method based on semi-supervised deep belief network |
| CN112146879A (en) * | 2020-08-21 | 2020-12-29 | 江苏大学 | Rolling bearing fault intelligent diagnosis method and system |
| 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 |
| CN112288078A (en) * | 2020-11-10 | 2021-01-29 | 北京理工大学 | Self-learning, small sample learning and transfer learning method and system based on impulse neural network |
| CN112396088A (en) * | 2020-10-19 | 2021-02-23 | 西安交通大学 | Intelligent diagnosis method for mechanical fault of implicit excitation countertraining under small sample |
| CN112418065A (en) * | 2020-11-19 | 2021-02-26 | 上海至数企业发展有限公司 | Equipment operation state identification method, device, equipment and storage medium |
| CN113076834A (en) * | 2021-03-25 | 2021-07-06 | 华中科技大学 | Rotating machine fault information processing method, processing system, processing terminal, and medium |
| CN114104328A (en) * | 2020-08-31 | 2022-03-01 | 中国航天科工飞航技术研究院（中国航天海鹰机电技术研究院） | Aircraft state monitoring method based on deep migration learning |
| CN115409124A (en) * | 2022-09-19 | 2022-11-29 | 小语智能信息科技(云南)有限公司 | Small sample sensitive information identification method based on fine-tuning prototype network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109376620A (en) * | 2018-09-30 | 2019-02-22 | 华北电力大学 | A kind of migration diagnostic method of gearbox of wind turbine failure |
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
-
2019
- 2019-07-26 CN CN201910684544.7A patent/CN110533070A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109376620A (en) * | 2018-09-30 | 2019-02-22 | 华北电力大学 | A kind of migration diagnostic method of gearbox of wind turbine failure |
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
Non-Patent Citations (3)
| Title |
|---|
| DENGYU XIAO等: "Domain Adaptive Motor Fault Diagnosis Using Deep Transfer Learning", 《IEEE ACCESS》 * |
| 何振亚: "自组织特征映射模型", 《神经智能 认知科学中若干重大问题的研究》 * |
| 王健弘: "基于视频的人体动作识别关键技术研究", 《中国优秀博士学位论文全文数据库信息科技辑》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110780146A (en) * | 2019-12-10 | 2020-02-11 | 武汉大学 | Transformer fault identification and positioning diagnosis method based on multi-stage transfer learning |
| 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 |
| CN111401454A (en) * | 2020-03-19 | 2020-07-10 | 创新奇智(重庆)科技有限公司 | Few-sample target identification method based on transfer learning |
| CN111537207A (en) * | 2020-04-29 | 2020-08-14 | 西安交通大学 | Data enhancement method for intelligent diagnosis of mechanical fault under small sample |
| 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 |
| CN111738413A (en) * | 2020-06-04 | 2020-10-02 | 东华大学 | Spinning full-process energy consumption monitoring method based on feature self-matching transfer learning |
| CN111982514A (en) * | 2020-08-12 | 2020-11-24 | 河北工业大学 | Bearing fault diagnosis method based on semi-supervised deep belief network |
| CN112146879A (en) * | 2020-08-21 | 2020-12-29 | 江苏大学 | Rolling bearing fault intelligent diagnosis method and system |
| CN114104328B (en) * | 2020-08-31 | 2023-10-17 | 中国航天科工飞航技术研究院（中国航天海鹰机电技术研究院） | Aircraft state monitoring method based on deep migration learning |
| CN114104328A (en) * | 2020-08-31 | 2022-03-01 | 中国航天科工飞航技术研究院（中国航天海鹰机电技术研究院） | Aircraft state monitoring method based on deep migration 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 |
| CN112396088A (en) * | 2020-10-19 | 2021-02-23 | 西安交通大学 | Intelligent diagnosis method for mechanical fault of implicit excitation countertraining under small sample |
| CN112396088B (en) * | 2020-10-19 | 2023-05-12 | 西安交通大学 | Mechanical fault intelligent diagnosis method for implicit excitation countermeasure training under small sample |
| CN112288078A (en) * | 2020-11-10 | 2021-01-29 | 北京理工大学 | Self-learning, small sample learning and transfer learning method and system based on impulse neural network |
| CN112288078B (en) * | 2020-11-10 | 2023-05-26 | 北京理工大学 | Self-learning, small sample learning and migration learning method and system based on impulse neural network |
| CN112418065A (en) * | 2020-11-19 | 2021-02-26 | 上海至数企业发展有限公司 | Equipment operation state identification method, device, equipment and storage medium |
| CN113076834B (en) * | 2021-03-25 | 2022-05-13 | 华中科技大学 | Rotating machine fault information processing method, processing system, processing terminal, and medium |
| CN113076834A (en) * | 2021-03-25 | 2021-07-06 | 华中科技大学 | Rotating machine fault information processing method, processing system, processing terminal, and medium |
| CN115409124A (en) * | 2022-09-19 | 2022-11-29 | 小语智能信息科技(云南)有限公司 | Small sample sensitive information identification method based on fine-tuning prototype network |
| CN115409124B (en) * | 2022-09-19 | 2023-05-23 | 小语智能信息科技(云南)有限公司 | Small sample sensitive information identification method based on fine tuning prototype network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110533070A (en) | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample | |
| CN109857835A (en) | A kind of adaptive network security knowledge assessment method based on cognitive diagnosis theory | |
| CN109918999A (en) | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database | |
| CN104931960B (en) | Trend message and radar target situation information full flight path segment data correlating method | |
| CN107832835A (en) | The light weight method and device of a kind of convolutional neural networks | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN107247938A (en) | A kind of method of high-resolution remote sensing image City Building function classification | |
| CN108520301A (en) | A kind of circuit intermittent fault diagnostic method based on depth confidence network | |
| CN108009594B (en) | A kind of image-recognizing method based on change grouping convolution | |
| CN107657281A (en) | A kind of image-recognizing method based on improved convolutional neural networks | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN104777418B (en) | A kind of analog-circuit fault diagnosis method based on depth Boltzmann machine | |
| CN108537259A (en) | Train control on board equipment failure modes and recognition methods based on Rough Sets Neural Networks model | |
| CN104849650B (en) | One kind is based on improved analog-circuit fault diagnosis method | |
| CN110516305A (en) | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model | |
| CN114842208B (en) | Deep learning-based power grid harmful bird species target detection method | |
| CN108921285A (en) | Single-element classification method in sequence based on bidirectional valve controlled Recognition with Recurrent Neural Network | |
| CN107967487A (en) | A kind of colliding data fusion method based on evidence distance and uncertainty | |
| CN110009030A (en) | Sewage treatment method for diagnosing faults based on stacking meta learning strategy | |
| CN110363230A (en) | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier | |
| CN108694408A (en) | A kind of driving behavior recognition methods based on depth sparseness filtering convolutional neural networks | |
| CN107506350A (en) | A kind of method and apparatus of identification information | |
| CN109800785A (en) | One kind is based on the relevant data classification method of expression and device certainly | |
| CN108108716A (en) | A kind of winding detection method based on depth belief network | |
| CN110866775A (en) | User air-rail joint inter-city trip information processing method based on machine learning |
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: 20191203 |
|
| RJ01 | Rejection of invention patent application after publication |