CN107657250A - Bearing fault detection and localization method and detection location model realize system and method - Google Patents
Bearing fault detection and localization method and detection location model realize system and method Download PDFInfo
- Publication number
- CN107657250A CN107657250A CN201711032275.3A CN201711032275A CN107657250A CN 107657250 A CN107657250 A CN 107657250A CN 201711032275 A CN201711032275 A CN 201711032275A CN 107657250 A CN107657250 A CN 107657250A
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- fault
- detection
- rolling bearing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention provides a kind of bearing fault detection and localization method and detection location model to realize system and method.After data prediction being carried out to rolling bearing without labeling data, it is input to the feature learning and detection model trained, solve quick detection and orientation problem of the rolling bearing under multiple fault modes, the probabilistic polling occurred by minimizing loss function to each classification results counts;If certain fault signature number of votes obtained is most, that is, determine that the failure is the fault mode currently estimated and positions trouble location.Whole feature learning process does not need any artificial characteristic extraction procedure, feature learning algorithm is used as input using initial data, and learning process uses unsupervised feature learning process, data extending and projection by depth, the bearing fault characteristics extracted can be realized efficient from expression, it is difficult to have solved the problems, such as that label data obtains, and there is very high detection and positioning precision feature.
Description
Technical field
It is applied to bearing fault detection and localization method the present invention relates to one kind and detection location model realizes system and side
Method, realized more particularly to a kind of fault detect suitable for rolling bearing detection positioning and localization method and detection location model
System and method.
Background technology
In recent years, with the trend that plant equipment develops towards maximization, complication, serialization, automation and centralization,
The structure and composition of equipment becomes increasingly complex, and the contact between each subsystem also more and more closely, run common by these equipment
Point is can not to fully rely on conventional method and establish accurate physical model to be monitored, and in operation due to non-linear factor
(Such as damping, rigidity, frictional force gap, external applied load)Influence, and the moment produce a large amount of reflection process operation states and operation
The Non-stationary Data of mechanism, even if the system of normal operation, because the interference of system noise and ambient noise etc. can also make operation
Tables of data reveals very strong nonlinear characteristic, and typically multi-frequency composition is superimposed(Such as when rolling bearing breaks down, vibration letter
Corresponding frequency multiplication and frequency dividing number can be also produced in addition to the frequency corresponding to original rotating speed, and different failure strengths produces
Frequency multiplication and frequency dividing etc. fault message also have nothing in common with each other), or even there is continuous Spectral structure.Thus, non linear mechanical equipment fault
Diagnose extremely difficult, critical mechanical equipment once breaks down, and very serious economic loss and casualties can be caused, because of machine
Tool equipment fault and caused catastrophic failure at home and abroad repeatedly occurs.
Largely rotating machinery in various plant equipment, and rolling bearing be in various rotating machineries using it is most universal,
Most flimsy auxiliary equipment and basic part, it is referred to as the joint of machinery.According to statistics.It is by bearing that the failure of rotating machinery, which has 30%,
It is caused.Therefore effectively plant equipment rolling bearing fault is detected, not only facilitate the safe handling for improving mechanical system
Rate, reduce major accident harm, moreover it is possible to obtain potential social and economic benefit.
Rolling bearing operating state signal is that it most intuitively reflects, working condition signal can use a variety of methods to obtain number
According to mainly having infrared measurement of temperature method, vibration monitoring method, sound emission method, oil analysis method and lossless detection method etc..It is general
All over using vibration monitoring method, because the theory and measuring method of vibration monitoring be simple and comparative maturity, it is easy to automate
And inline diagnosis, vibration measurement can promptly show the running status and mechanism of bearing, be a kind of non-destructive testing technology.It is related
Statistics shows, the failure of bearing 90% can be detected from vibration signal.Therefore, examined in state monitoring of rolling bearing and failure
It is method the most frequently used at present using vibrating working condition signal and its working state signal being monitored and diagnosed in disconnected research.Rolling
Dynamic bearing fault diagnosis seeks to the collection by the signal to that can reflect bearing working state, analysis and processing, to identify
The state of bearing.The general process of bearing failure diagnosis mainly includes:Signal acquisition(Including vibration, temperature, noise, fluid and
Picture signal), information processing(Feature extraction and feature selecting)With state recognition and decision-making(Fault diagnosis, abnormality detection are with determining
Plan intervention etc.).
The traditional analysis method of rolling bearing fault diagnosis has a Shock Pulse Method, resonance and demodulation method, cepstrum analysis technology,
The Intelligent Diagnosis Technology of rolling bearing fault is exactly the technologies such as neutral net, expert system, fuzzy theory and rolling bearing
Characteristic parameter organically combines the fault diagnosis technology for carrying out comprehensive analysis.Pass through the signal from reflection Rolling Bearing Status
Middle to extract fault signature (sign) to recognize bearing running status, these parameters can reflect the changing features of the system failure, therefore
It is called fault characteristic value.And same failure can be reflected on several characteristic quantities, different failures may have identical
Feature influences each other, interpenetrate be fault diagnosis difficult point, to rolling bearing working condition signal fault signature extract and identify close
It is the reliability and accuracy to fault diagnosis, is the key issue in bearing failure diagnosis research.According to process signal feature
Difference, Rolling Bearing Fault Character extracting method is divided into stationary signal analysis method(Time domain and frequency domain method)And non-stationary signal
Analysis method(T/F technology and adaptive Time Frequency Analysis).
But a common feature of these rolling bearing feature extracting methods is all artificial selection design feature, then
With reference to corresponding failure modes diagnosis of technique failure.Because bearing floor data is a kind of multiple dimensioned multi-level integrated information,
Engineer's feature does not have unified understanding, and this method wastes time and energy, it is necessary to heuristic professional knowledge, largely by warp
Test and fortune so that a limited number of failure symptoms are quite faint for the interpretability of all previous operating mode of bearing.Therefore, only from low
The manual extraction and selection complex fault feature of level, which set out diagnose, is difficult to meet the needs of intelligent diagnostics.So use for reference
Human brain is in extracting object feature by different level, and the characteristics of each layer is individually handled signal, to reflection bearing operation shape
The signal of state carries out deep learning, is formed by Automatic Combined low-level feature and more abstract high-rise represents the new of attribute or feature
Thought, can significantly more efficient monitoring rolling bearing performance to find that the distributed nature of data represents.
Further, since rolling bearing operating mode has label data to be difficult to obtain, therefore, deep learning is made full use of to learn automatically
The characteristics of practising rolling bearing working condition signal profound level feature, with reference to the advantages of different depth learning neural network model, design nothing
The deep neural network mixed model with multiple hidden layers of supervised learning, excavate the high-level letter implied in bearing floor data
Breath feature is judged to the failure structure detector that may occur, and failure is positioned, and fault detect effect will be excellent
In with the deep-neural-network of a certain deep learning model buildings, being a kind of advanced and effective technology merely.
The content of the invention
The invention solves a technical problem be to provide one kind and overcome in existing bearing failure diagnosis engineer special
The deficiency of sign and feature representation, there is label data to obtain the bearing operating mode data characteristics that automatically extracts of the problems such as difficult and carry out failure
Detection and the system and method for positioning.
The invention solves another technical problem be to provide and a kind of can be applied to bearing fault detection and positioning
Deep neural network learns and detection model.
It is a kind of to realize system suitable for bearing fault detection and the detection model of positioning, it is characterised in that:Including,
Data preprocessing module, to rolling bearing without label non-stationary initial data, carry out successively at trend processing and pollution
Reason;
Detection module, receives the data of data preprocessing module output, and the data are carried out before reverting to data prediction
Data sample, realize expressing certainly to initial data;It is made up of 5 layer depth feature learnings and detection model, layers 1 and 2
Using denoising self-encoding encoder(denoising autoencoder,DA)Realize, the 3rd layer and the limited Boltzmann machine of the 4th layer of use
(Restricted boltzmann machines, RBM)Realize;5th layer using logic, this special recurrence carries out ballot detection and determined
Position bearing fault;
The detection module includes,
Data extending module, including the layers 1 and 2, the 1st layer of the input exported as the 2nd layer;1st layer of section is set
Points are input dimension N times, and input data is mapped into a N times of higher dimensional space, realizes the work(of data extending and separability
Can, and second the number of hidden nodes to extracting more abstract characteristics is compressed;
The N is more than or equal to 2 and is less than or equal to 5;
Data projection module, including described 3rd layer and the 4th layer, cascade to obtain two layer depth confidences using limited Boltzmann machine
Output of the network to denoising self-encoding encoder projects.
Data after being handled using pollution as deep neural network detection model input, by being laminated neutral net
(deep belief network,DBN)Contaminated preceding sample data is recovered, is realized to initial data from expression, so that
The abstracting power of model is improved, obtains more generally expressing initial data, improves the anti-noise ability of fault diagnosis.
It can ensure to throw because RBM pre-training process is unrelated with dimension, and by the restorability of data in second step
Shadow result is stablized relatively, data can be carried out using this model effectively projecting and expressing.So, according to each layer
After type carries out unsupervised pre-training successively to network, the preferable initial weight of network availability, that is, obtain
The stronger feature of ability to express.Data extending obtains with the weight vector obtained after projection, as fault characteristic value, thus pre-training
Feature learning model is arrived.
The denoising self-encoding encoder be stack denoising self-encoding encoder (stacked denoising autoencoder,
SDA).The same matrix of SDA methods codes and decodes to input data, is originally inputted and decoding data by minimizing
Error carrys out Level by level learning this matrix(Certain expression of input data), and by first hidden layer(1st layer)Output as
Two hidden layers(2nd layer)Input, realize the effect of initial data restorability.
The N is equal to 3, and reconstructed error and separability now is best.
Also include training fine setting module, there is label non-stationary initial data to carry out the data prediction to rolling bearing
Afterwards, the input using the weight vector that 5 layer depth feature learnings and detection model learn as Softmax algorithms, returned with Softmax
Reduction method is counted in rolling bearing as rolling bearing multiple-fault classifier and positioning algorithm by minimizing loss function
Circle, outer ring, rolling element failure and normal condition the ballot probability that totally 5 kinds of classification results occur, if certain fault signature number of votes obtained
At most, then detect that fault mode corresponding to the Fault characteristic parameters occurs.Due to pre-training have found one it is relatively reasonable
Initial weight, therefore, whole 5 layer model that DBN and Softmax is returned is laminated to final SDA using back-propagation algorithm and is weighed
Value and biasing are finely adjusted, and so far, have obtained final feature learning model.And using have label data to testing result carry out
Test, obtains fault detect and location model.
A kind of detection model implementation method for being applied to bearing fault detection and positioning, it is characterised in that:Specific method walks
Suddenly it is:
First, trend processing and pollution processing are carried out successively without label non-stationary initial data to rolling bearing;
2nd, the data of data preprocessing module output are received, and the data revert to the data sample before data prediction
This, realizes expressing certainly to initial data;
The realization of the step 2 is made up of 5 layer depth feature learnings and detection model, and layers 1 and 2 is self-editing using denoising
Code device realization, the 3rd layer and the limited Boltzmann machine realization of the 4th layer of use;5th layer using logic, this spy's recurrence carries out ballot detection
And positioning bearing fault;
The 5 layer depth feature learning and detection model include data extending and data projection, wherein,
Data extending includes the layers 1 and 2, the 1st layer of the input exported as the 2nd layer;Setting the 1st node layer number is
N times of dimension is inputted, input data is mapped to a N times of higher dimensional space, realizes the function of data extending and separability, and it is right
2nd node layer number of extraction more abstract characteristics is compressed;
The N is more than or equal to 2 and is less than or equal to 5;
Data projection includes described 3rd layer and the 4th layer, cascades to obtain two layer depth confidence networks pair using limited Boltzmann machine
The output of denoising self-encoding encoder is projected.
The denoising self-encoding encoder is stack denoising self-encoding encoder.
The N is equal to 3.
Methods described also includes, using there is label data to be trained and finely tune feature learning and detection model, specifically
Method is:After thering is label non-stationary initial data to carry out the data prediction to rolling bearing, with 5 layer depth feature learnings and
Input of the weight vector that detection model learns as Softmax algorithms, rolling bearing event more is used as by the use of Softmax regression algorithms
Barrier detection and positioning algorithm, rolling bearing inner ring, outer ring, rolling element failure and normal are counted by minimizing loss function
The state ballot probability that totally 5 kinds of classification results occur, if certain fault signature number of votes obtained is most, detect that the fault signature is joined
Fault mode corresponding to number occurs.
It is described to go in trend processing, retain the spike and Mutational part in vibration signal.
In the pollution processing, random noise is added in sample after going trend, to be uniformly distributed the part of input layer
Node is arranged to random value, obtains new pollution input sample.
The data projection also includes, and improving degree of approximation to sdpecific dispersion algorithm using holding carries out unsupervised pre-training.
The bearing event of system or implementation method is realized based on the above-mentioned detection model suitable for bearing fault detection and positioning
Barrier detection and localization method, specific method are:It is defeated after carrying out the data prediction without labeling data to rolling bearing
Enter to the feature learning and detection model trained, solve quick detection and positioning of the rolling bearing under multiple fault modes and ask
Topic, the probabilistic polling occurred by minimizing loss function to each classification results count;If certain fault signature obtains
Poll is most, that is, determines that the failure is the fault mode currently estimated and positions trouble location.
The deep neural network learns and detection model is stack denoising self-encoding encoder (stacked denoising
Autoencoder, SDA) stacking depth confidence network(deep belief network,DBN)5 layer depths returned with Softmax
Spend feature learning and detection model;In the 5 layer depth feature learning and detection model, 4 layers of feature learning model by denoising from
Encoder(denoising autoencoder,DA)With limited Boltzmann machine(restricted boltzmann
Machines, RBM)Two kinds of different basic modules are built.Layers 1 and 2 using denoising self-encoding encoder realize, the 3rd layer and
4th layer of use is limited Boltzmann machine and realized;5th layer using logic, this spy's recurrence carries out vote detection and positioning bearing fault.
Compared with prior art, the beneficial effects of the invention are as follows:Whole feature learning process does not need any artificial spy
Extraction process is levied, feature learning algorithm is using initial data as input, and learning process uses unsupervised feature learning mistake
Journey, data extending and projection by depth, the bearing fault characteristics extracted can realize efficient expression certainly, solve
Label data obtains the problem of difficult, and has very high detection and positioning precision feature.
Brief description of the drawings
Fig. 1 is the schematic network structure of bearing fault detection of the present invention and location model.
Fig. 2 is fault signature study and the detection model pre-training process schematic of a wherein embodiment of the invention.
Fig. 3 is the wherein fault detect of an embodiment and position fixing process schematic diagram of the invention.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, not
For limiting the present invention.
This specification(Including summary and accompanying drawing)Disclosed in any feature, unless specifically stated otherwise, can be equivalent by other
Or the alternative features with similar purpose are replaced.I.e., unless specifically stated otherwise, each feature is a series of equivalent or class
Like an example in feature.
The bearing fault detection and the tool of location technology that a kind of SDA stacking DBN and Softmax of this patent presented below are returned
Body embodiment.Embodiment gives the specific embodiment that network node and network parameter are set, but is not limited to the embodiment.
1st, the embodiment of depth S DA2-DBN2 networks is built:
As shown in figure 1, in 5 layer depth feature learnings and detection model structure that a kind of SDA stacking DBN and Softmax are returned, 4
Layer feature learning model is built by two kinds of different basic modules of denoising self-encoding encoder and limited Boltzmann machine.Preceding two layer data
Expansion is built with denoising self-encoding encoder, wherein, the first hidden layer is set(1st layer)Nodes be to input three times of dimension, by number
According to the higher dimensional space for being mapped to a three times input dimension, the function of preferable data extending is realized, and spy is more abstracted to extracting
Second hidden layer of sign(2nd layer)Nodes are compressed(It is general to take the integer less than input dimension 1/3).Middle two layer datas
Migration using limited Boltzmann machine cascade to obtain output of the depth confidence network to denoising self-encoding encoder projected to obtain it is whole
Third and fourth hidden layer of individual network(3rd layer and the 4th layer), also, set third layer nodes to be equal to second layer nodes, the
Four node layer numbers take the half of third layer nodes.Finally carry out vote detection and position bearing fault logic this spy return layer
Nodes are arranged to 4(Rolling bearing normal condition, 4 kinds of inner ring, outer ring and rolling element failure states).Except first layer nodes are set
Three times of input dimension or so are set to, the second layer is equal with the nodes of third layer, the specific nodes of further feature learning layer
It is not limited to the embodiment.
2nd, SDA is laminated fault detect and the positioning three phases embodiment that DBN and Softmax is returned:
First stage:Obtain feature learning model with pre-training is carried out without label data, as shown in figure 3, the stage be divided into again as
Under several steps:
The first step:Data prediction:Rolling bearing different frequency, different parts collection without label non-stationary primary signal conduct
Input sample x, first using going trend to handle, i.e., using mean square error is normal to rolling bearing and some minizones of fault-signal
More lines fitting(This patent is using second order polynomial)Eliminate its trend, spike and Mutational part in stick signal.
Then handled using pollution, i.e., random noise is added in sample after going trend, to be uniformly distributed the part of nodes of input layer
Random value is arranged to, 20% element is arranged to random value at random in the present embodiment, obtains new pollution input sample x';So
The input of sample x' after being handled afterwards using pollution as SDA2-DBN2 neural network models.
Second step:Data extending:Before pre-training, the initial weight of SDA2-DBN2-Softmax networks and biasing take with
Machine number.The same matrix of SDA methods codes and decodes to input data, is originally inputted and decoding data by minimizing
Error carrys out Level by level learning this matrix(Certain expression of input data), during two layers of SDA of pre-training, every 1 layer of iterations is
20 times, and the input using the output weights of first hidden layer and biasing as second hidden layer.
3rd step:Data projection:Output of middle two layers of the 2 layer depth confidence networks to denoising self-encoding encoder projects,
Degree of approximation is improved to sdpecific dispersion algorithm using holding and carries out unsupervised pre-training, when pre-training is limited Boltzmann machine, every time
We run 1 gibbs sampler to undated parameter, i.e., the step number to sdpecific dispersion is 1.Due to being limited the pre- instruction of Boltzmann machine
White silk process is unrelated with dimension, and can ensure the relatively stable of projection result by the restorability of data in second step.So,
According to each layer of type to network carry out unsupervised pre-training successively after, the preferable weight of network availability and partially
Put, the weight of four-layer network network is the stronger fault signature of ability to express.So far, the feature after unsupervised pre-training has been obtained
Learning model.
Second stage:With having label data to feature learning model training and finely tuning, the detection model stage is obtained after test.
As shown in Fig. 2 after thering is label non-stationary initial data to carry out identical pretreatment with the first stage to rolling bearing,
The weight vector learnt using first stage SDA stackings DBN model and biasing pass through minimum as the input of Softmax algorithms
Each classification results of whole loss function pair carry out probabilistic polling.According to first stage pre-training obtain it is relatively reasonable just
Beginning weights, final SDA2-DBN2-Softmax whole 5 layer model weights and biasing are carried out using back-propagation algorithm
Fine setting, the upper limit of back-propagation algorithm iterations is 200 times during fine setting, while weights constraint factor during backpropagation is
0.1.When back-propagation algorithm fine setting is embodied, there is the phenomenon of study in each model, can be by introducing 1 in fine setting
Norm and 2 norm constraints carry out the ability to express of limited model.So far, the feature learning after supervision pre-training and detection have been obtained
Model.
Phase III:To being detected and being determined with detection model using feature learning model without label bearing data to be sorted
The position stage.
As shown in figure 3, after the stage without label data to be sorted and first stage to carrying out identical pretreatment, Ran Houying
Obtained feature learning model is trained with earlier stage, learns the feature of data to be sorted, the data characteristics that will finally learn
As the input of detection model, Fast Classification and positioning are carried out to rolling bearing, by minimizing loss function to each point
Class result carries out probabilistic polling.If certain fault signature number of votes obtained is most, that is, it is the fault mode currently estimated to determine the failure
And position trouble location.
Claims (10)
1. a kind of realize system suitable for bearing fault detection and the detection model of positioning, it is characterised in that:Including,
Data preprocessing module, to rolling bearing without label non-stationary initial data, carry out successively at trend processing and pollution
Reason;
Detection module, receives the data of data preprocessing module output, and the data are carried out before reverting to data prediction
Data sample, realize expressing certainly to initial data;It is made up of 5 layer depth feature learnings and detection model, layers 1 and 2
Realized using denoising self-encoding encoder, the 3rd layer and the limited Boltzmann machine realization of the 4th layer of use;5th layer using logic, this spy returns
Carry out vote detection and positioning bearing fault;
The detection module includes,
Data extending module, including the layers 1 and 2, the 1st layer of the input exported as the 2nd layer;1st layer of section is set
Points are input dimension N times, and input data is mapped into a N times of higher dimensional space, realizes the work(of data extending and separability
Can, and second the number of hidden nodes to extracting more abstract characteristics is compressed;
The N is more than or equal to 2 and is less than or equal to 5;
Data projection module, including described 3rd layer and the 4th layer, cascade to obtain two layer depth confidences using limited Boltzmann machine
Output of the network to denoising self-encoding encoder projects.
2. according to claim 1 realize system, it is characterised in that;The denoising self-encoding encoder is stack denoising own coding
Device;The N is equal to 3.
3. according to claim 1 or 2 realize system, it is characterised in that;Also include training fine setting module, to rolling bearing
After thering is label non-stationary initial data to carry out the data prediction, learnt with 5 layer depth feature learnings and detection model
Input of the weight vector as Softmax algorithms, it is quick by the use of Softmax regression algorithms as rolling bearing multiple-fault classifier and positioning
Algorithm, rolling bearing inner ring, outer ring, rolling element failure and normal condition totally 5 kinds of classification knots are counted by minimizing loss function
The ballot probability that fruit occurs, if certain fault signature number of votes obtained is most, detects failure mould corresponding to the Fault characteristic parameters
Formula occurs.
A kind of 4. detection model implementation method for being applied to bearing fault detection and positioning, it is characterised in that:Specific method step
For:
First, trend processing and pollution processing are carried out successively without label non-stationary initial data to rolling bearing;
2nd, the data of data preprocessing module output are received, and the data revert to the data sample before data prediction
This, realizes expressing certainly to initial data;
The realization of the step 2 is made up of 5 layer depth feature learnings and detection model, and layers 1 and 2 is self-editing using denoising
Code device realization, the 3rd layer and the limited Boltzmann machine realization of the 4th layer of use;5th layer using logic, this spy's recurrence carries out ballot detection
And positioning bearing fault;
The 5 layer depth feature learning and detection model include data extending and data projection, wherein,
Data extending includes the layers 1 and 2, the 1st layer of the input exported as the 2nd layer;Setting the 1st node layer number is
N times of dimension is inputted, input data is mapped to a N times of higher dimensional space, realizes the function of data extending and separability, and it is right
2nd node layer number of extraction more abstract characteristics is compressed;
The N is more than or equal to 2 and is less than or equal to 5;
Data projection includes described 3rd layer and the 4th layer, cascades to obtain two layer depth confidence networks pair using limited Boltzmann machine
The output of denoising self-encoding encoder is projected.
5. implementation method according to claim 4, it is characterised in that:The denoising self-encoding encoder is stack denoising own coding
Device;The N is equal to 3.
6. the implementation method according to claim 4 or 5, it is characterised in that:Methods described also includes, using there is label data
Feature learning and detection model are trained and finely tuned, specific method is:There is label non-stationary initial data to rolling bearing
After carrying out the data prediction, Softmax algorithms are used as using the weight vector that 5 layer depth feature learnings and detection model learn
Input, by the use of Softmax regression algorithms as rolling bearing multiple-fault classifier and positioning algorithm, pass through and minimize loss letter
Number statistics rolling bearing inner ring, outer ring, rolling element failure and the normal condition ballot probability that totally 5 kinds of classification results occur, if
Certain fault signature number of votes obtained is most, then detects that fault mode corresponding to the Fault characteristic parameters occurs.
7. implementation method according to claim 6, it is characterised in that:It is described to go in trend processing, retain in vibration signal
Spike and Mutational part.
8. implementation method according to claim 6, it is characterised in that:In the pollution processing, in sample after going trend
Random noise is added, random value is arranged to the part of nodes being uniformly distributed input layer, obtains new pollution input sample.
9. implementation method according to claim 6, it is characterised in that:The data projection also includes, and is contrasted using holding
Divergence algorithm improves degree of approximation and carries out unsupervised pre-training.
10. based on the bearing fault detection for realizing system or the implementation method described in claim 4 described in claim 1 and fixed
Position method, specific method are:After carrying out the data prediction without labeling data to rolling bearing, it is input to and trains
Feature learning and detection model, solve quick detection and orientation problem of the rolling bearing under multiple fault modes, by most
The probabilistic polling that smallization loss function occurs to each classification results counts;If certain fault signature number of votes obtained is most,
Determine that the failure is the fault mode currently estimated and positions trouble location.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711032275.3A CN107657250B (en) | 2017-10-30 | 2017-10-30 | Bearing fault detection and positioning method and detection and positioning model implementation system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711032275.3A CN107657250B (en) | 2017-10-30 | 2017-10-30 | Bearing fault detection and positioning method and detection and positioning model implementation system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107657250A true CN107657250A (en) | 2018-02-02 |
CN107657250B CN107657250B (en) | 2020-11-24 |
Family
ID=61095285
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711032275.3A Active CN107657250B (en) | 2017-10-30 | 2017-10-30 | Bearing fault detection and positioning method and detection and positioning model implementation system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107657250B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108592812A (en) * | 2018-05-10 | 2018-09-28 | 电子科技大学 | Fan blade optical fiber load strain characteristics extract and crack monitoring method |
CN109214356A (en) * | 2018-09-29 | 2019-01-15 | 南京东振测控技术有限公司 | A kind of fan transmission system intelligent fault diagnosis method based on DCNN model |
CN109491816A (en) * | 2018-10-19 | 2019-03-19 | 中国船舶重工集团公司第七六研究所 | Knowledge based engineering method for diagnosing faults |
CN109519340A (en) * | 2018-12-27 | 2019-03-26 | 中国船舶重工集团海装风电股份有限公司 | A kind of wind turbine generator drive system method for diagnosing faults |
CN112345872A (en) * | 2020-12-04 | 2021-02-09 | 中国南方电网有限责任公司超高压输电公司昆明局 | Converter station secondary circuit abnormity identification and early warning method based on deep confidence network |
CN112949097A (en) * | 2021-04-19 | 2021-06-11 | 合肥工业大学 | Bearing residual life prediction model and method based on deep migration learning |
CN113554602A (en) * | 2021-06-30 | 2021-10-26 | 西安翔迅科技有限责任公司 | Fan variable-pitch bearing monitoring method |
CN113588266A (en) * | 2021-07-19 | 2021-11-02 | 西安交通大学 | Rolling bearing composite fault diagnosis method with embedded fault semantic space |
CN113740667A (en) * | 2021-08-30 | 2021-12-03 | 华北电力大学 | Power grid fault diagnosis method integrating self-encoder and convolutional neural network |
CN114484732A (en) * | 2022-01-14 | 2022-05-13 | 南京信息工程大学 | Air conditioning unit sensor fault diagnosis method based on novel voting network |
CN115031357A (en) * | 2022-05-10 | 2022-09-09 | 南京信息工程大学 | Novel voting strategy-based fault diagnosis method suitable for different types of fault characteristics |
CN115587294A (en) * | 2022-12-08 | 2023-01-10 | 中国空气动力研究与发展中心设备设计与测试技术研究所 | Compressor fault diagnosis method based on blade tip clearance and bearing vibration |
CN116756675A (en) * | 2023-08-14 | 2023-09-15 | 湘江实验室 | Bearing fault diagnosis and classification method based on impulse neural network |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102254177A (en) * | 2011-04-22 | 2011-11-23 | 哈尔滨工程大学 | Bearing fault detection method for unbalanced data SVM (support vector machine) |
CN104777418A (en) * | 2015-05-11 | 2015-07-15 | 重庆大学 | Analog circuit fault diagnosis method based on depth Boltzman machine |
CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
CN106127230A (en) * | 2016-06-16 | 2016-11-16 | 上海海事大学 | Image-recognizing method based on human visual perception |
CN106203625A (en) * | 2016-06-29 | 2016-12-07 | 中国电子科技集团公司第二十八研究所 | A kind of deep-neural-network training method based on multiple pre-training |
CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
CN107169535A (en) * | 2017-07-06 | 2017-09-15 | 谈宜勇 | The deep learning sorting technique and device of biological multispectral image |
CN206504869U (en) * | 2017-02-27 | 2017-09-19 | 苏州大学 | A kind of rolling bearing fault diagnosis device |
-
2017
- 2017-10-30 CN CN201711032275.3A patent/CN107657250B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102254177A (en) * | 2011-04-22 | 2011-11-23 | 哈尔滨工程大学 | Bearing fault detection method for unbalanced data SVM (support vector machine) |
CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
CN104777418A (en) * | 2015-05-11 | 2015-07-15 | 重庆大学 | Analog circuit fault diagnosis method based on depth Boltzman machine |
CN106127230A (en) * | 2016-06-16 | 2016-11-16 | 上海海事大学 | Image-recognizing method based on human visual perception |
CN106203625A (en) * | 2016-06-29 | 2016-12-07 | 中国电子科技集团公司第二十八研究所 | A kind of deep-neural-network training method based on multiple pre-training |
CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
CN206504869U (en) * | 2017-02-27 | 2017-09-19 | 苏州大学 | A kind of rolling bearing fault diagnosis device |
CN107169535A (en) * | 2017-07-06 | 2017-09-15 | 谈宜勇 | The deep learning sorting technique and device of biological multispectral image |
Non-Patent Citations (2)
Title |
---|
RUI ZHAO 等: ""Deep Learning and Its Applications to Machine Health Monitoring: A Survey"", 《ARXIV:1612.07640V1 [CS.LG]》 * |
李兆飞 等: ""多重分形去趋势波动分析的振动信号故障诊断"", 《华中科技大学学报》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108592812B (en) * | 2018-05-10 | 2019-12-31 | 电子科技大学 | Method for extracting load strain characteristics and monitoring cracks of optical fiber of fan blade |
CN108592812A (en) * | 2018-05-10 | 2018-09-28 | 电子科技大学 | Fan blade optical fiber load strain characteristics extract and crack monitoring method |
CN109214356A (en) * | 2018-09-29 | 2019-01-15 | 南京东振测控技术有限公司 | A kind of fan transmission system intelligent fault diagnosis method based on DCNN model |
CN109491816B (en) * | 2018-10-19 | 2022-05-27 | 中国船舶重工集团公司第七一六研究所 | Knowledge-based fault diagnosis method |
CN109491816A (en) * | 2018-10-19 | 2019-03-19 | 中国船舶重工集团公司第七六研究所 | Knowledge based engineering method for diagnosing faults |
CN109519340A (en) * | 2018-12-27 | 2019-03-26 | 中国船舶重工集团海装风电股份有限公司 | A kind of wind turbine generator drive system method for diagnosing faults |
CN112345872A (en) * | 2020-12-04 | 2021-02-09 | 中国南方电网有限责任公司超高压输电公司昆明局 | Converter station secondary circuit abnormity identification and early warning method based on deep confidence network |
CN112949097A (en) * | 2021-04-19 | 2021-06-11 | 合肥工业大学 | Bearing residual life prediction model and method based on deep migration learning |
CN112949097B (en) * | 2021-04-19 | 2022-09-16 | 合肥工业大学 | Bearing residual life prediction model and method based on deep migration learning |
CN113554602A (en) * | 2021-06-30 | 2021-10-26 | 西安翔迅科技有限责任公司 | Fan variable-pitch bearing monitoring method |
CN113588266A (en) * | 2021-07-19 | 2021-11-02 | 西安交通大学 | Rolling bearing composite fault diagnosis method with embedded fault semantic space |
CN113740667B (en) * | 2021-08-30 | 2022-06-14 | 华北电力大学 | Power grid fault diagnosis method integrating self-encoder and convolutional neural network |
CN113740667A (en) * | 2021-08-30 | 2021-12-03 | 华北电力大学 | Power grid fault diagnosis method integrating self-encoder and convolutional neural network |
CN114484732A (en) * | 2022-01-14 | 2022-05-13 | 南京信息工程大学 | Air conditioning unit sensor fault diagnosis method based on novel voting network |
CN115031357A (en) * | 2022-05-10 | 2022-09-09 | 南京信息工程大学 | Novel voting strategy-based fault diagnosis method suitable for different types of fault characteristics |
CN115587294A (en) * | 2022-12-08 | 2023-01-10 | 中国空气动力研究与发展中心设备设计与测试技术研究所 | Compressor fault diagnosis method based on blade tip clearance and bearing vibration |
CN116756675A (en) * | 2023-08-14 | 2023-09-15 | 湘江实验室 | Bearing fault diagnosis and classification method based on impulse neural network |
CN116756675B (en) * | 2023-08-14 | 2023-11-07 | 湘江实验室 | Bearing fault diagnosis and classification method based on impulse neural network |
Also Published As
Publication number | Publication date |
---|---|
CN107657250B (en) | 2020-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107657250A (en) | Bearing fault detection and localization method and detection location model realize system and method | |
CN110132598B (en) | Fault noise diagnosis algorithm for rolling bearing of rotating equipment | |
CN106682688A (en) | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization | |
CN108646725A (en) | Dual model method for diagnosing faults based on dynamic weighting | |
CN108537259A (en) | Train control on board equipment failure modes and recognition methods based on Rough Sets Neural Networks model | |
CN106485594A (en) | A kind of main distribution integration incident response decision method | |
Yang et al. | Compound fault diagnosis of harmonic drives using deep capsule graph convolutional network | |
Yao et al. | Improving rolling bearing fault diagnosis by DS evidence theory based fusion model | |
Oh et al. | Fault detection for lubricant bearing with CNN | |
Heidari | Fault detection of bearings using a rule-based classifier ensemble and genetic algorithm | |
Chen et al. | A novel Bayesian-optimization-based adversarial TCN for RUL prediction of bearings | |
Wu et al. | Ss-InfoGAN for class-imbalance classification of bearing faults | |
Zhong et al. | CAE‐WANN: A novel anomaly detection method for gas turbines via search space extension | |
Li et al. | Source-free domain adaptation framework for fault diagnosis of rotation machinery under data privacy | |
Gougam et al. | Bearing faults classification using a new approach of signal processing combined with machine learning algorithms | |
CN111488711A (en) | Network robustness assessment method and system | |
CN110490218A (en) | A kind of rolling bearing fault self-learning method based on two-stage DBN | |
Du et al. | Graph neural network-based early bearing fault detection | |
Zhao | [Retracted] Fault Diagnosis Method for Wind Power Equipment Based on Hidden Markov Model | |
Yang et al. | Machinery fault detection using autoencoder and online sequential extreme learning machine | |
Guerroum et al. | Machine learning for the predictive maintenance of a Jaw Crusher in the mining industry | |
Moghaddam et al. | A Comparative Analisis of Artificial Intelligence-Based Methods for Fault Diagnosis of Mechanical Systems | |
You et al. | A new generative neural network for bearing fault diagnosis with imbalanced data | |
Nguyen | Feature Engineering and Health Indicator Construction for Fault Detection and Diagnostic | |
Cai et al. | Ensemble dilated convolutional neural network and its application in rotating machinery fault diagnosis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |