CN109918999A - Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database - Google Patents
Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database Download PDFInfo
- Publication number
- CN109918999A CN109918999A CN201910058792.0A CN201910058792A CN109918999A CN 109918999 A CN109918999 A CN 109918999A CN 201910058792 A CN201910058792 A CN 201910058792A CN 109918999 A CN109918999 A CN 109918999A
- Authority
- CN
- China
- Prior art keywords
- data
- mechanical
- model
- mechanical equipment
- signal
- 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
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses, based on the mechanical equipment fault intelligent diagnosing method for generating model, carry out zero-mean standardization pretreatment to a small amount of mechanical signal got under a kind of Small Sample Database;Establish the composite network generated for mechanical signal;In conjunction with Wasserstein distance and gradient penalty method confrontation type, training generates confrontation network model;Establish the depth convolutional neural networks model classified using mechanical signal to mechanical equipment operating status;In conjunction with confrontation complex neural network model and depth convolutional neural networks model is generated, two networks of a small amount of true mechanical signal training, the final intelligent trouble diagnosis to mechanical equipment realized under Small Sample Database are used.It is good to the feature extraction effect of mechanical signal that the present invention has, the feature that state classification accuracy is high and mechanical signal data extending performance is good.
Description
Technical field
The present invention relates to mechanical fault diagnosis fields, and in particular to based on generating model under a kind of Small Sample Database
Mechanical equipment fault intelligent diagnosing method.
Background technique
During mechanical equipment is run, main parts size bearing, gear, rotor etc. are due to lasting receiving load
Effect, be easy to happen failure, in turn result in economic loss and casualties.In order to reduce due to mechanical equipment fault bring
Loss, it is necessary to carry out fault diagnosis and the status monitoring research for mechanical equipment.It is adopted on mechanical equipment under actual condition
All kinds of mechanical signals collected will receive the pollution of noise, it is difficult to which the validity feature for carrying out mechanical signal extracts and state is known
Not.Denoising carried out to mechanical signal and feature extraction be generally viewed as mechanical fault diagnosis groundwork and main weight,
Difficult point, a large amount of Intelligent Diagnosis for Mcchanical Devices algorithm, which is all focused on, for a long time carries out denoising and feature extraction to mechanical signal
On.
However, being difficult to obtain the fault-signal of mechanical equipment in actual condition, the fault-signal quantity got is few, kind
Class is also few.On the other hand, a large amount of fault sample is needed when being trained to Intelligent Diagnosis for Mcchanical Devices algorithm to be improved
Its generalization and engineering practicability.Small sample problem, which has been seriously affected, carries out fault diagnosis and status monitoring to mechanical equipment
Timeliness and accuracy, it is therefore necessary to carry out the research for the mechanical fault diagnosis under small sample problem.
Traditionally the mode of EDS extended data set is over-sampling in failure diagnostic process, but over-sampling is also only reusing
Only a small amount of fault sample information does not have generality.
Summary of the invention
The purpose of the present invention is to provide intelligently examined under a kind of Small Sample Database based on the mechanical equipment fault for generating model
Disconnected method, of the existing technology to overcome the problems, such as, the present invention carries out feature to mechanical signal using depth convolutional neural networks
Extraction and running state recognition, can effectively extract the sensitive features in mechanical signal, get rid of traditionally feature extraction
Dependence of the journey to artificial experience.
In order to achieve the above objectives, the present invention adopts the following technical scheme:
Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database, this method includes following
Step:
Step 1: using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, to a small amount of machine got
Tool signal is standardized pretreatment;
Step 2: establishing the generation generated for mechanical signal and fight complex neural network model, the complex neural network mould
Type includes generator and two network minor structures of arbiter with subsidiary classification device;
Step 3: the complex neural network model that step 2 is established, in conjunction with Wasserstein distance and gradient punishment side
The network parameter of generator and the arbiter with subsidiary classification device is trained and updated to method with carrying out confrontation type, to make to generate
Device realizes the function that tape label data are generated using Gaussian noise, and then can get the mechanical signal with operating status label;
Step 4: establishing the neural network model for carrying out Classification and Identification to mechanical equipment operating status using mechanical signal, mould
Type input data is the true mechanical signal less than total amount of data 5% and the band operating status that the generator by step 3 generates
The mechanical signal of label, model output are the probability value of operating status corresponding to each data;
Step 5: the convolutional neural networks state classification model established to step 4 uses Dropout and Batch
Normalization parameter regularization method prevents from training over-fitting, stablizes training process, to keep network more rapidly more stable
Completion status classification work;
Step 6: convolution designed by generation confrontation complex neural network model and step 5 in conjunction with designed by step 3
Neural network state classification model, using true mechanical signal two networks of training for being less than total amount of data 5%, to make to generate
Confrontation complex neural network model can generate the data for having same distribution with true mechanical signal, and make convolutional neural networks
State classification model can obtain 95% or more state classification accuracy, it is final realize under Small Sample Database to mechanical equipment
Intelligent trouble diagnosis.
Further, the pretreatment of data normalization described in step 1 is standardized using zero-mean, calculating formula are as follows:
In formula, n is the data point number of single input signal, xiFor i-th of data point in input signal,For input
The mean value of signal, s are the sample standard deviation of input signal, yiFor i-th of number in the new signal after zero-mean standardization processing
According to.
Further, generator described in step 2 is made of 4 layers of full articulamentum, is responsible for generating and true mechanical signal
Data with same distribution.Arbiter with subsidiary classification device is made of 5 layers of full articulamentum, is completed at the same time judgement and is generated number
According to it is true and false and generate classification two of data work.
Further, step 3 is using the loss function of Wasserstein distance optimization composite network model to stablize training
Process, Wasserstein is apart from calculating formula is defined as:
In formula, A1It is the distribution that truthful data is obeyed, A2It is the distribution for generating data and obeying, ∏ (A1,A2) it is A1And A2Point
The set for all Joint Distributions that cloth combines, γ are one of Joint Distributions, and (x, y) is a pair of sample in γ,
E(x, y)~γ[| | x-y | |] is the desired value of the sample distance.
Further, the punishment of gradient used in step 3 refers to for arbiter part, in generation sample concentrated area, very
It real sample concentrated area and is clipped on region among them and applies Lipschitz limitation.Specifically, first stochastical sampling is a pair of
True and false sample, there are one the random numbers of 0-1:
xr~Ar,xf~Af, ε~Uniform [0,1]
Then in xrAnd xfLine on random interpolation sampling:Note sampling obtainsMet
Distribution be denoted asThen by the additional bring penalty values calculating formula of Lipschitz limitation institute are as follows:
In formula, LlFor by the additional bring penalty values of Lipschitz limitation institute, xrIt is truthful data sample, ArReally to count
According to distribution, xfIt is the data sample generated, AfFor the data distribution of generation, ε is the random number of a 0-1, and D (x) is arbiter
Output valve,For the second norm of the derivative value of arbiter output valve,Indicate this
Two norms subtract one square desired value, λ is positive number arbitrarily less than 1.
Further, convolutional neural networks state classification model designed in step 4 is by 6 layers of convolutional layer, 6 layers of pond
The depth convolutional neural networks of layer composition, use first floor convolutional layer for the big step-length of big convolution kernel, and intermediate and end convolutional layer is small
The approach configuring parameters of the small step-length of convolution kernel, specifically, the convolution kernel size being arranged in first floor convolutional layer are intermediate and end
8 times of convolution kernel size in convolutional layer, and the step sizes being arranged in first floor convolutional layer are step-length in intermediate and end convolutional layer
4 times of size, to achieve the effect that better mechanical signal feature extraction and operating status classification.The loss letter of network model
Number uses cross entropy loss function, calculating formula are as follows:
In formula, L is penalty values, and y is the label information of desired output,The label information of real network output.
Further, prevent convolutional neural networks from over-fitting occurs using Dropout parameter regularization method in step 5.
Dropout method refers in a wheel training, and each neural unit node is made to be retained (Dropout loss ratio with Probability p first
For 1-p), remaining node is hidden, the process that the network training and parameter for then carrying out epicycle again update.In next round training
In, and each neural unit node is retained with Probability p, repeatedly, until training terminates.
Further, in step 5 using Batch normalization parameter regularization method to stablize training process.
Batch normalization method is divided into 4 steps:
For input data x={ x1,x2,...,xm, data mean value is calculated firstM is that each is defeated
Enter the number of data point in data, xiFor i-th of data point in input data;
Secondly data variance is calculated
Then carrying out i-th of data point that batch standardization obtains in new data isWherein η is to prevent
The small positive number being arranged except zero error occurs;
Finally carry out change of scale and offset:α and β is what network itself learnt in the training process
Parameter.
This method is suitable for the mechanical fault diagnosis under Small Sample Database, using only the number for being less than total amount of data 5%
Training dataset can effectively be expanded according to being trained to network, and then obtain 95% or more mechanical equipment operating status
Classification accuracy rate improves the accuracy rate under Small Sample Database to mechanical fault diagnosis.
Compared with prior art, the invention has the following beneficial technical effects:
The present invention carries out feature extraction and running state recognition, Neng Gouyou to mechanical signal using depth convolutional neural networks
Effect extracts sensitive features in mechanical signal, get rid of traditionally characteristic extraction procedure to the dependence of artificial experience;This hair
The bright confrontation type training by generating model, is generated a variety of the mechanical of different operating statuses simultaneously using a kind of network model and believed
Number, effectively expand fault signal of mechanical equipment data set;The present invention, which passes through to combine, generates model and depth convolutional Neural net
Network effectively can carry out fault diagnosis to mechanical equipment under Small Sample Database, improve and set under Small Sample Database to machinery
The accuracy rate of standby fault diagnosis.
The generation model that the present invention uses is capable of the distribution of learning data, and then generates with data-oriented with same distribution
New data can be generated different types of mechanical signal using this advantage for generating model, the signal of generation with it is actual
Mechanical signal data distribution having the same, therefore the signal generated can be used to training smart diagnosis and calculate with more generality
Method improves under Small Sample Database to the accuracy rate of mechanical fault diagnosis.
Detailed description of the invention
Fig. 1 is the process based on the mechanical equipment fault intelligent diagnosing method for generating model under Small Sample Database of the present invention
Figure;
It includes that the data set of four kinds of bearing operating statuses carries out the result figure of state classification to certain that Fig. 2, which is using the present invention,;
Fig. 3 is that feature when carrying out state classification to the data set that certain includes four kinds of bearing operating statuses using the present invention mentions
Take figure;
It includes that the data set of four kinds of bearing operating statuses carries out the result of mechanical signal generation to certain that Fig. 4, which is using the present invention,
Figure, wherein (a) is outer ring failure actual signal;(b) signal is generated for outer ring failure;It (c) is normal condition actual signal;(d)
Signal is generated for normal condition;It (e) is inner ring failure actual signal;(f) signal is generated for inner ring failure;It (g) is ball failure
Actual signal;(h) signal is generated for ball failure.
Specific embodiment
Present invention is further described in detail with reference to the accompanying drawing:
Referring to Fig. 1, in order to improve the accuracy rate under Small Sample Database to mechanical fault diagnosis, the present invention provides one
Based on the Intelligent Diagnosis for Mcchanical Devices method for generating model under kind Small Sample Database.This method includes two parts: fault-signal
Generating portion and state classification identification division.Wherein fault-signal generating portion is realized based on generation confrontation model, is generated
Generator in confrontation model can generate the fault-signal of corresponding fault type according to given label, with sentencing for subsidiary classification device
Other device may determine that the classification for generating the true and false of signal and differentiating generation signal.By the training of confrontation type, generator can be with
Good generative capacity is obtained, EDS extended data set is come with this.State classification identification division realized by depth convolutional neural networks, energy
Enough Classification and Identifications for effectively extracting characteristic information from mechanical signal and then carry out mechanical equipment operating status.
Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database, comprising the following steps:
Step 1: using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, to a small amount of machine got
Tool signal is standardized pretreatment;
Wherein data normalization pretreatment is standardized using zero-mean, calculating formula are as follows:
In formula, n is the data point number of single input signal, xiFor i-th of data in input signal,For input letter
Number mean value, s be input signal sample standard deviation, yiFor i-th of data in the new signal after zero-mean standardization processing.
Step 2: establishing the generation generated for mechanical signal and fight complex neural network model, the complex neural network mould
Type includes generator and two network minor structures of arbiter with subsidiary classification device.
Wherein, generator is made of 4 layers of full articulamentum, is responsible for generating the number for having same distribution with true mechanical signal
According to.Arbiter with subsidiary classification device is made of 5 layers of full articulamentum, is completed at the same time judgement and is generated the true and false of data and generate
Classification two work of data.
Step 3: to step 2 establish composite network structure, in conjunction with Wasserstein distance and gradient penalty method,
The network parameter of generator and the arbiter with subsidiary classification device is trained and updates with carrying out confrontation type, to keep generator real
The function of tape label data is now generated using random Gaussian, and then can get the mechanical signal with operating status label.
Using the loss function of Wasserstein distance optimization composite network model to stablize training process,
Wasserstein is apart from calculating formula is defined as:
In formula, A1It is the distribution that truthful data is obeyed, A2It is the distribution for generating data and obeying, ∏ (A1,A2) it is A1And A2Point
The set for all Joint Distributions that cloth combines, γ are one of Joint Distributions, and (x, y) is a pair of sample in γ,
E(x, y)~γ[| | x-y | |] is the desired value of the sample distance.
Gradient punishment used refers to for arbiter part, is generating sample concentrated area, authentic specimen concentrated area
And it is clipped on region among them and applies Lipschitz limitation.Specifically, a pair of true and false sample of first stochastical sampling, also
The random number of one 0-1:
xr~Ar,xf~Af, ε~Uniform [0,1]
Then in xrAnd xfLine on random interpolation sampling:Note sampling obtainsMet
Distribution be denoted asThen by the additional bring penalty values calculating formula of Lipschitz limitation institute are as follows:
In formula, LlFor by the additional bring penalty values of Lipschitz limitation institute, xrIt is truthful data sample, ArReally to count
According to distribution, xfIt is the data sample generated, AfFor the data distribution of generation, ε is the random number of a 0-1, and D (x) is arbiter
Output valve,For the second norm of the derivative value of arbiter output valve,Indicate this second
Norm subtract one square desired value, λ is positive number arbitrarily less than 1.
Step 4: establishing the neural network model classified using mechanical signal to mechanical equipment operating status, model is defeated
Enter the mechanical letter for the tape label that data are the true mechanical signal less than total amount of data 5% and the generator generation by step 3
Number, model output is the probability value of operating status corresponding to each data.
The depth convolution that designed convolutional neural networks state classification model is made of 6 layers of convolutional layer, 6 layers of pond layer
Neural network uses first floor convolutional layer for the big step-length of big convolution kernel, and intermediate and end convolutional layer is the ginseng of the small step-length of small convolution kernel
Number configuration strategy, specifically, it is that convolution kernel is big in intermediate and end convolutional layer that the convolution kernel size in first floor convolutional layer, which is arranged,
Small 8 times, and the step sizes being arranged in first floor convolutional layer are 4 times of step sizes in intermediate and end convolutional layer, to reach
The effect of better mechanical signal feature extraction and operating status classification.The loss function of network model is using intersection entropy loss
Function, calculating formula are as follows:
In formula, LcFor the penalty values of cross entropy loss function, y is the label information of desired output,Real network output
Label information.
Step 5: the convolutional neural networks state classification model established to step 4 uses Dropout and Batch
Normalization parameter regularization method prevents from training over-fitting, stablizes training process, to keep network more rapidly more stable
Completion status classification work.
Wherein, prevent convolutional neural networks from over-fitting occurs using Dropout parameter regularization method.Dropout method
Refer in a wheel training, each neural unit node is made to be retained (Dropout loss ratio 1-p) with Probability p first, remaining
Node is hidden, the process that the network training and parameter for then carrying out epicycle again update.In next round training, and by each mind
It is retained through cell node with Probability p, repeatedly, until training terminates.
Using Batch normalization parameter regularization method to stablize training process.Batch
Normalization method is divided into 4 steps:
For input data x={ x1,x2,...,xm, data mean value is calculated firstM is that each is defeated
Enter the number of data point in data, xiFor i-th of data point in input data;
Secondly data variance is calculated
Then carrying out i-th of data point that batch standardization obtains in new data isWherein η is to prevent
The small positive number being arranged except zero error occurs;
Finally carry out change of scale and offset:α and β is what network itself learnt in the training process
Parameter.
Step 6: convolutional Neural designed by generation confrontation composite network model and step 5 in conjunction with designed by step 3
Network state disaggregated model is fought using true mechanical signal two networks of training for being less than total amount of data 5% to make to generate
Composite network model can generate the data for having same distribution with true mechanical signal, and make convolutional neural networks state classification
Model can obtain 95% or more state classification accuracy, the final intelligence event to mechanical equipment realized under Small Sample Database
Barrier diagnosis.
The invention will be described in further detail combined with specific embodiments below:
It is used certain include four kinds of bearing operating statuses data set one is shared normal, inner ring failure, outer ring failure and
Four kinds of rolling bearing operating statuses of ball failure, every kind of operating status include 148 samples, in total include 592 samples.Take it
In 16 samples as training data, for remaining 576 samples as test data, training sample data amount only accounts for total number of samples
According to the 2.7% of amount.
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 a small amount of rolling bearing got, is made
Standardized with zero-mean, calculating formula are as follows:
In formula, n is the data point number of single input signal, xiFor i-th of data in input signal,For input letter
Number mean value, s be input signal sample standard deviation, yiFor i-th of data in the new signal after zero-mean standardization processing.
Step 2: establishing the generation generated for bearing vibration signal and fight complex neural network model, the compound mind
Two network minor structures of arbiter through network model comprising generator and with subsidiary classification device.Generator is connected entirely by 4 layers
Layer composition is connect, the work for generating mechanical oscillation data is completed.Arbiter with subsidiary classification device is made of 5 layers of full articulamentum,
Judgement is completed at the same time to generate the true and false of data and generate classification two work of data.
Step 3: the complex neural network model that step 2 is established, in conjunction with Wasserstein distance and gradient punishment side
Method, trains to confrontation type and updates the network parameter of generator and the arbiter with subsidiary classification device, to keep generator real
The function of tape label data is now generated using random Gaussian, and then be can get the rolling bearing with operating status label and shaken
Dynamic data.
Wasserstein is apart from calculating formula is defined as:
In formula, A1It is the distribution that truthful data is obeyed, A2It is the distribution for generating data and obeying, ∏ (A1,A2) it is A1And A2Point
The set for all Joint Distributions that cloth combines, γ are one of Joint Distributions, and (x, y) is a pair of sample in γ,
E(x, y)~γ[| | x-y | |] is the desired value of the sample distance.
Gradient punishment refers to for arbiter part, is generating sample concentrated area, authentic specimen concentrated area and folder
Apply Lipschitz limitation on region among them.Specifically, a pair of true and false sample of first stochastical sampling, there are one 0-1
Random number:
xr~Ar,xf~Af, ε~Uniform [0,1]
Then in xrAnd xfLine on random interpolation sampling:Note sampling obtainsMet
Distribution be denoted asThen by the additional bring penalty values calculating formula of Lipschitz limitation institute are as follows:
In formula, LlFor by the additional bring penalty values of Lipschitz limitation institute, xrIt is truthful data sample, ArReally to count
According to distribution, xfIt is the data sample generated, AfFor the data distribution of generation, ε is the random number of a 0-1, and D (x) is arbiter
Output valve,For the second norm of the derivative value of arbiter output valve,Indicate this second
Norm subtract one square desired value, λ is positive number arbitrarily less than 1.
Step 4: establishing the neural network model classified using vibration signal to rolling bearing operating status, the model
The depth convolutional neural networks being made of 6 layers of convolutional layer, 6 layers of pond layer, use first floor convolutional layer for the big step-length of big convolution kernel,
Intermediate and end convolutional layer is the approach configuring parameters of the small step-length of small convolution kernel, specifically, the volume in first floor convolutional layer is arranged
Product core size is 8 times of convolution kernel size in intermediate and end convolutional layer, and during the step sizes being arranged in first floor convolutional layer are
Between and end convolutional layer in 4 times of step sizes, to reach better vibration data feature extraction and operating status classification
Effect.The loss function of network model uses cross entropy loss function, definition are as follows:
In formula, LcFor the penalty values of cross entropy loss function, y is the label information of desired output,Real network output
Label information.
Step 5: the convolutional neural networks state classification model established to step 4 uses Dropout and Batch
Normalization parameter regularization method prevents from training over-fitting, stablizes training process, to keep network more rapidly more stable
Completion status classification work.
Dropout method refers in a wheel training, is retained each neural unit node with Probability p
(Dropout loss ratio is 1-p), remaining node is hidden, the process that the network training and parameter for then carrying out epicycle again update.
In next round training, and each neural unit node is retained with Probability p, repeatedly, until training terminates.
And the Batch normalization method used is divided into 4 steps:
For input data x={ x1,x2,...,xm, data mean value is calculated firstM is that each is defeated
Enter the number of data point in data, xiFor i-th of data point in input data;
Secondly data variance is calculated
Then carrying out i-th of data point that batch standardization obtains in new data isWherein η is to prevent
The small positive number being arranged except zero error occurs;
Finally carry out change of scale and offset:α and β is what network itself learnt in the training process
Parameter.
Step 6: convolution designed by generation confrontation complex neural network model and step 5 in conjunction with designed by step 3
Neural network state classification model keeps generation confrontation compound using a small amount of rolling bearing indeed vibrations signal two networks of training
Neural network model can generate the data for having same distribution with indeed vibrations signal, and make convolutional neural networks state classification
Model can be using the indeed vibrations signal for being less than total amount of data 5% and the signal training for generating model generation
Obtain 95% or more state classification precision, the final intelligent trouble diagnosis to rolling bearing realized under Small Sample Database.
In Fig. 2, number 0 represents outer ring malfunction, and number 1 represents normal condition, and number 2 represents inner ring failure shape
State, number 3 represent ball malfunction.As shown in Fig. 2, it includes four kinds of bearings that the present invention, which is realized under Small Sample Database to certain,
The correct classification of the data set of operating status, and the present invention completes the spy to bearing vibration signal well as seen from Figure 3
Sign is extracted.The present invention has efficiently generated different operating statuses and has descended bearing vibration signal as seen from Figure 4 simultaneously, successfully logarithm
Expanded according to collection.
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. based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database, which is characterized in that including
Following steps:
Step 1: using the mechanical signal under the various operating statuses of mechanical equipment as data acquisition system, to the mechanical signal got into
Row standardization pretreatment;
Step 2: establishing the generation generated for mechanical signal and fight complex neural network model, the complex neural network model packet
Two network minor structures of arbiter containing generator and with subsidiary classification device;
Step 3: to step 2 establish complex neural network model, in conjunction with Wasserstein distance and gradient penalty method,
The network parameter of generator and the arbiter with subsidiary classification device is trained and updates with carrying out confrontation type, to keep generator real
The function of tape label data is now generated using Gaussian noise, and then obtains the mechanical signal for having operating status label;
Step 4: establishing the neural network model for carrying out Classification and Identification to mechanical equipment operating status using mechanical signal, model is defeated
Enter the band operating status label that data are the true mechanical signal less than total amount of data 5% and the generator by step 3 generates
Mechanical signal, model output be each data corresponding to operating status probability value;
Step 5: the convolutional neural networks state classification model established to step 4 uses Dropout and Batch
Normalization parameter regularization method prevents from training over-fitting and stablizes training process, to keep network more rapidly more steady
Determine completion status classification work;
Step 6: convolutional Neural designed by generation confrontation complex neural network model and step 5 in conjunction with designed by step 3
Network state disaggregated model is fought using true mechanical signal two networks of training for being less than total amount of data 5% to make to generate
Complex neural network model can generate the data for having same distribution with true mechanical signal, and make convolutional neural networks state
Disaggregated model can obtain 95% or more state classification accuracy, the final intelligence to mechanical equipment realized under Small Sample Database
It can fault diagnosis.
2. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that the pretreatment of data normalization described in step 1 is standardized using zero-mean, calculating formula are as follows:
In formula, n is the data point number of single input signal, xiFor i-th of data in input signal,For input signal
Mean value, s are the sample standard deviation of input signal, yiFor i-th of data in the new signal after zero-mean standardization processing.
3. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that generator described in step 2 is made of 4 layers of full articulamentum, is had for generating with true mechanical signal
The data of same distribution;Arbiter with subsidiary classification device is made of 5 layers of full articulamentum, generates number for being completed at the same time judgement
According to it is true and false and generate data classification.
4. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that step 3 was trained using the loss function of Wasserstein distance optimization composite network model with stablizing
Journey, Wasserstein is apart from calculating formula is defined as:
In formula, A1It is the distribution that truthful data is obeyed, A2It is the distribution for generating data and obeying, ∏ (A1,A2) it is A1And A2Distribution group
The set of all Joint Distributions altogether, γ are one of Joint Distributions, and (x, y) is a pair of sample in γ,
E(x, y)~γ[| | x-y | |] is the desired value of the sample distance.
5. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that the punishment of gradient used in step 3 refers to for arbiter part, is generating sample concentrated area, true sample
It this concentrated area and is clipped on region among them and applies Lipschitz limitation, specifically, first stochastical sampling a pair is true and false
Sample, there are one the random numbers of 0-1:
xr~Ar,xf~Af, ε~Uniform [0,1]
Then in xrAnd xfLine on random interpolation sampling:Note sampling obtainsThe distribution met
It is denoted asThen by the additional bring penalty values calculating formula of Lipschitz limitation institute are as follows:
In formula, LlFor by the additional bring penalty values of Lipschitz limitation institute, xrIt is truthful data sample, ArFor truthful data point
Cloth, xfIt is the data sample generated, AfFor the data distribution of generation, ε is the random number of a 0-1, and D (x) is the output of arbiter
Value,For the second norm of the derivative value of arbiter output valve,Indicate second model
Number subtract one square desired value, λ is positive number arbitrarily less than 1.
6. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that the neural network model established in step 4 is rolled up by the depth that 6 layers of convolutional layer and 6 layers of pond layer form
Product neural network, uses first floor convolutional layer for the big step-length of big convolution kernel, and intermediate and end convolutional layer is the small step-length of small convolution kernel
The loss function of approach configuring parameters, neural network model uses cross entropy loss function, calculating formula are as follows:
In formula, LcFor the penalty values of cross entropy loss function, y is the label information of desired output,The label of real network output
Information.
7. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 6
Method, which is characterized in that use first floor convolutional layer for the big step-length of big convolution kernel, intermediate and end convolutional layer is the small step-length of small convolution kernel
Approach configuring parameters specifically: the convolution kernel size in setting first floor convolutional layer is that convolution kernel is big in intermediate and end convolutional layer
Small 8 times, and the step sizes being arranged in first floor convolutional layer are 4 times of step sizes in intermediate and end convolutional layer.
8. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that prevent convolutional neural networks from over-fitting occurs using Dropout parameter regularization method in step 5,
Dropout method refers in a wheel training, is retained each neural unit node with Probability p, remaining node is hidden
Hiding, the process that the network training and parameter for then carrying out epicycle again update, in next round training, and by each neural unit section
Point is retained with Probability p, repeatedly, until training terminates.
9. based on the mechanical equipment fault intelligent diagnostics side for generating model under a kind of Small Sample Database according to claim 1
Method, which is characterized in that using Batch normalization parameter regularization method to stablize training process in step 5;
Batch normalization method is divided into 4 steps:
For input data x={ x1,x2,...,xm, data mean value is calculated firstM is each input data
The number of middle data point, xiFor i-th of data point in input data;
Secondly data variance is calculated
Then carrying out i-th of data point that batch standardization obtains in new data isWherein η is to prevent to remove
Zero error and the small positive number being arranged;
Finally carry out change of scale and offset:α and β is the parameter that network itself learns in the training process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910058792.0A CN109918999A (en) | 2019-01-22 | 2019-01-22 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910058792.0A CN109918999A (en) | 2019-01-22 | 2019-01-22 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109918999A true CN109918999A (en) | 2019-06-21 |
Family
ID=66960547
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910058792.0A Pending CN109918999A (en) | 2019-01-22 | 2019-01-22 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109918999A (en) |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363163A (en) * | 2019-07-18 | 2019-10-22 | 电子科技大学 | A kind of SAR target image generation method that azimuth is controllable |
CN110414604A (en) * | 2019-07-30 | 2019-11-05 | 重庆邮电大学 | Perception confrontation generates the lithium battery fault data generation method of network-driven |
CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep learning |
CN110458039A (en) * | 2019-07-19 | 2019-11-15 | 华中科技大学 | A kind of construction method of industrial process fault diagnosis model and its application |
CN110516305A (en) * | 2019-07-26 | 2019-11-29 | 西安交通大学 | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model |
CN110533070A (en) * | 2019-07-26 | 2019-12-03 | 西安交通大学 | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample |
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 |
CN110619386A (en) * | 2019-09-09 | 2019-12-27 | 国家电网有限公司 | TMR operation monitoring and fault intelligent research and judgment method and system |
CN110647923A (en) * | 2019-09-04 | 2020-01-03 | 西安交通大学 | Variable working condition mechanical fault intelligent diagnosis method based on self-learning under small sample |
CN111006865A (en) * | 2019-11-15 | 2020-04-14 | 上海电机学院 | Motor bearing fault diagnosis method |
CN111008692A (en) * | 2019-11-08 | 2020-04-14 | 国网天津市电力公司 | Method and device for generating multi-energy metering characteristic data based on improved generation countermeasure network |
CN111026058A (en) * | 2019-12-16 | 2020-04-17 | 浙江大学 | Semi-supervised deep learning fault diagnosis method based on Watherstein distance and self-encoder |
CN111310791A (en) * | 2020-01-17 | 2020-06-19 | 电子科技大学 | Dynamic progressive automatic target identification method based on small sample number set |
CN111337243A (en) * | 2020-02-27 | 2020-06-26 | 上海电力大学 | ACGAN-based wind turbine generator planet wheel gearbox fault diagnosis method |
CN111586728A (en) * | 2020-04-30 | 2020-08-25 | 南京邮电大学 | Small sample characteristic-oriented heterogeneous wireless network fault detection and diagnosis method |
CN111624977A (en) * | 2020-05-09 | 2020-09-04 | 黄河水利职业技术学院 | Mechanical control device and method of capacitive touch screen equipment |
CN111832428A (en) * | 2020-06-23 | 2020-10-27 | 北京科技大学 | Data enhancement method applied to strip breakage fault diagnosis of cold rolling mill |
CN112329579A (en) * | 2020-10-28 | 2021-02-05 | 厦门大学 | Small sample wireless equipment RF fingerprint identification method and device based on ACGAN-GP |
CN112328588A (en) * | 2020-11-27 | 2021-02-05 | 哈尔滨工程大学 | Industrial fault diagnosis unbalanced time sequence data expansion method |
CN112649198A (en) * | 2021-01-05 | 2021-04-13 | 西交思创智能科技研究院(西安)有限公司 | Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application |
CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN113256443A (en) * | 2021-04-29 | 2021-08-13 | 西安交通大学 | Nuclear power water pump guide bearing fault detection method, system and equipment and readable storage medium |
CN113505876A (en) * | 2021-06-11 | 2021-10-15 | 国网浙江省电力有限公司嘉兴供电公司 | High-voltage circuit breaker fault diagnosis method based on generation type countermeasure network |
CN114021285A (en) * | 2021-11-17 | 2022-02-08 | 上海大学 | Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning |
CN114176549A (en) * | 2021-12-23 | 2022-03-15 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generative countermeasure network |
CN114611233A (en) * | 2022-03-08 | 2022-06-10 | 湖南第一师范学院 | Rotating machinery fault unbalance data generation method and computer equipment |
CN116821697A (en) * | 2023-08-30 | 2023-09-29 | 聊城莱柯智能机器人有限公司 | Mechanical equipment fault diagnosis method based on small sample learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107341546A (en) * | 2016-04-29 | 2017-11-10 | 北京中科寒武纪科技有限公司 | A kind of apparatus and method for performing batch normalization computings |
US20180286034A1 (en) * | 2017-04-03 | 2018-10-04 | General Electric Company | Equipment damage prediction system |
CN108680807A (en) * | 2018-05-17 | 2018-10-19 | 国网山东省电力公司青岛供电公司 | The Diagnosis Method of Transformer Faults and system of network are fought based on condition production |
CN108711138A (en) * | 2018-06-06 | 2018-10-26 | 北京印刷学院 | A kind of gray scale picture colorization method based on generation confrontation network |
CN108830334A (en) * | 2018-06-25 | 2018-11-16 | 江西师范大学 | A kind of fine granularity target-recognition method based on confrontation type transfer learning |
CN109120652A (en) * | 2018-11-09 | 2019-01-01 | 重庆邮电大学 | It is predicted based on difference WGAN network safety situation |
-
2019
- 2019-01-22 CN CN201910058792.0A patent/CN109918999A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107341546A (en) * | 2016-04-29 | 2017-11-10 | 北京中科寒武纪科技有限公司 | A kind of apparatus and method for performing batch normalization computings |
US20180286034A1 (en) * | 2017-04-03 | 2018-10-04 | General Electric Company | Equipment damage prediction system |
CN108680807A (en) * | 2018-05-17 | 2018-10-19 | 国网山东省电力公司青岛供电公司 | The Diagnosis Method of Transformer Faults and system of network are fought based on condition production |
CN108711138A (en) * | 2018-06-06 | 2018-10-26 | 北京印刷学院 | A kind of gray scale picture colorization method based on generation confrontation network |
CN108830334A (en) * | 2018-06-25 | 2018-11-16 | 江西师范大学 | A kind of fine granularity target-recognition method based on confrontation type transfer learning |
CN109120652A (en) * | 2018-11-09 | 2019-01-01 | 重庆邮电大学 | It is predicted based on difference WGAN network safety situation |
Non-Patent Citations (2)
Title |
---|
SIYU SHAO等: "Generative adversarial networks for data augmentation in machine fault diagnosis", 《COMPUTERS IN INDUSTRY》 * |
汪晋宽等: "《电子商务实用技术》", 31 December 2003 * |
Cited By (37)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363163A (en) * | 2019-07-18 | 2019-10-22 | 电子科技大学 | A kind of SAR target image generation method that azimuth is controllable |
CN110363163B (en) * | 2019-07-18 | 2021-07-13 | 电子科技大学 | SAR target image generation method with controllable azimuth angle |
CN110458039A (en) * | 2019-07-19 | 2019-11-15 | 华中科技大学 | A kind of construction method of industrial process fault diagnosis model and its application |
CN110516305A (en) * | 2019-07-26 | 2019-11-29 | 西安交通大学 | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model |
CN110533070A (en) * | 2019-07-26 | 2019-12-03 | 西安交通大学 | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample |
CN110414604A (en) * | 2019-07-30 | 2019-11-05 | 重庆邮电大学 | Perception confrontation generates the lithium battery fault data generation method of network-driven |
CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep 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 |
CN110647923A (en) * | 2019-09-04 | 2020-01-03 | 西安交通大学 | Variable working condition mechanical fault intelligent diagnosis method based on self-learning under small sample |
CN110619386B (en) * | 2019-09-09 | 2022-12-16 | 国家电网有限公司 | TMR operation monitoring and fault intelligent research and judgment method and system |
CN110619386A (en) * | 2019-09-09 | 2019-12-27 | 国家电网有限公司 | TMR operation monitoring and fault intelligent research and judgment method and system |
CN111008692A (en) * | 2019-11-08 | 2020-04-14 | 国网天津市电力公司 | Method and device for generating multi-energy metering characteristic data based on improved generation countermeasure network |
CN111006865A (en) * | 2019-11-15 | 2020-04-14 | 上海电机学院 | Motor bearing fault diagnosis method |
CN111026058A (en) * | 2019-12-16 | 2020-04-17 | 浙江大学 | Semi-supervised deep learning fault diagnosis method based on Watherstein distance and self-encoder |
CN111310791A (en) * | 2020-01-17 | 2020-06-19 | 电子科技大学 | Dynamic progressive automatic target identification method based on small sample number set |
CN111337243B (en) * | 2020-02-27 | 2022-04-01 | 上海电力大学 | ACGAN-based wind turbine generator planet wheel gearbox fault diagnosis method |
CN111337243A (en) * | 2020-02-27 | 2020-06-26 | 上海电力大学 | ACGAN-based wind turbine generator planet wheel gearbox fault diagnosis method |
CN111586728A (en) * | 2020-04-30 | 2020-08-25 | 南京邮电大学 | Small sample characteristic-oriented heterogeneous wireless network fault detection and diagnosis method |
CN111624977A (en) * | 2020-05-09 | 2020-09-04 | 黄河水利职业技术学院 | Mechanical control device and method of capacitive touch screen equipment |
CN111832428B (en) * | 2020-06-23 | 2024-02-23 | 北京科技大学 | Data enhancement method applied to cold rolling mill broken belt fault diagnosis |
CN111832428A (en) * | 2020-06-23 | 2020-10-27 | 北京科技大学 | Data enhancement method applied to strip breakage fault diagnosis of cold rolling mill |
CN112329579A (en) * | 2020-10-28 | 2021-02-05 | 厦门大学 | Small sample wireless equipment RF fingerprint identification method and device based on ACGAN-GP |
CN112328588A (en) * | 2020-11-27 | 2021-02-05 | 哈尔滨工程大学 | Industrial fault diagnosis unbalanced time sequence data expansion method |
CN112649198A (en) * | 2021-01-05 | 2021-04-13 | 西交思创智能科技研究院(西安)有限公司 | Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application |
CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN112784930B (en) * | 2021-03-17 | 2022-03-04 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN113256443B (en) * | 2021-04-29 | 2023-06-20 | 西安交通大学 | Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium |
CN113256443A (en) * | 2021-04-29 | 2021-08-13 | 西安交通大学 | Nuclear power water pump guide bearing fault detection method, system and equipment and readable storage medium |
CN113505876A (en) * | 2021-06-11 | 2021-10-15 | 国网浙江省电力有限公司嘉兴供电公司 | High-voltage circuit breaker fault diagnosis method based on generation type countermeasure network |
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 |
CN114176549A (en) * | 2021-12-23 | 2022-03-15 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generative countermeasure network |
CN114176549B (en) * | 2021-12-23 | 2024-04-16 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network |
CN114611233A (en) * | 2022-03-08 | 2022-06-10 | 湖南第一师范学院 | Rotating machinery fault unbalance data generation method and computer equipment |
CN114611233B (en) * | 2022-03-08 | 2022-11-11 | 湖南第一师范学院 | Rotating machinery fault imbalance data generation method and computer equipment |
CN116821697A (en) * | 2023-08-30 | 2023-09-29 | 聊城莱柯智能机器人有限公司 | Mechanical equipment fault diagnosis method based on small sample learning |
CN116821697B (en) * | 2023-08-30 | 2024-05-28 | 聊城莱柯智能机器人有限公司 | Mechanical equipment fault diagnosis method based on small sample learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109918999A (en) | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database | |
CN110533070A (en) | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample | |
CN112418277B (en) | Method, system, medium and equipment for predicting residual life of rotating machine parts | |
CN109782603A (en) | The detection method and monitoring system of rotating machinery coupling fault | |
CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
CN111046945A (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
CN102879677A (en) | Intelligent fault diagnosis method based on rough Bayesian network classifier | |
CN112257767B (en) | Product key part state classification method aiming at class imbalance data | |
CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
CN106198000A (en) | A kind of rocker arm of coal mining machine gear failure diagnosing method | |
CN110376522A (en) | A kind of Method of Motor Fault Diagnosis of the deep learning network of data fusion | |
CN113538353B (en) | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network | |
CN106271881A (en) | A kind of Condition Monitoring of Tool Breakage method based on SAEs and K means | |
EP2392120A1 (en) | Method and sensor network for attribute selection for an event recognition | |
CN106768933A (en) | A kind of wind power generation unit blade method for diagnosing faults based on genetic algorithm | |
CN114462480A (en) | Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN111458145A (en) | Cable car rolling bearing fault diagnosis method based on road map characteristics | |
CN115545070A (en) | Intelligent diagnosis method for unbalance-like bearing based on comprehensive balance network | |
CN113869451B (en) | Rolling bearing fault diagnosis method under variable working condition based on improved JGSA algorithm | |
CN113324758B (en) | Bearing fault diagnosis method based on semi-supervised countermeasure network | |
CN113985156A (en) | Intelligent fault identification method based on transformer voiceprint big data | |
CN114964783B (en) | Gearbox fault detection model based on VMD-SSA-LSSVM | |
CN116561692A (en) | Dynamic update real-time measurement data detection method | |
CN113537010B (en) | Fifteen-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
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: 20190621 |
|
RJ01 | Rejection of invention patent application after publication |