CN109141884A - Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN - Google Patents
Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN Download PDFInfo
- Publication number
- CN109141884A CN109141884A CN201810939116.XA CN201810939116A CN109141884A CN 109141884 A CN109141884 A CN 109141884A CN 201810939116 A CN201810939116 A CN 201810939116A CN 109141884 A CN109141884 A CN 109141884A
- Authority
- CN
- China
- Prior art keywords
- layer
- dbn
- model
- eemd
- rbm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- 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
- G01M13/045—Acoustic or vibration analysis
Abstract
The present invention provides a kind of Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN, due to bearing vibration signal nonlinear change, it is first passed through polymerization empirical mode decomposition (abbreviation EEMD) method and is decomposed into multiple IMF components, several IMF components before selection wherein, establish autoregression (abbreviation AR) model, autoregressive coefficient and corresponding variance are calculated to each AR model, the input parameter as depth confidence network (abbreviation DBN network)；The low order feature of signal is extracted by EEMD and AR pretreatment；The Dividing Characteristics for recycling DBN network powerful extract and generalization ability advantage excavates representative high-order feature from the parameter of input, are used for fault diagnosis.The method that the present invention uses has higher accuracy rate than traditional support vector machines and artificial neural network method for diagnosing faults.
Description
Technical field
The present invention relates to technical field of wind power, especially a kind of bearing failure diagnosis based on EEMD-AR model and DBN
Method.
Background technique
Wind energy is one of current fastest-rising renewable energy, wind-powered electricity generation in global renewable energy power generation installed capacity
Occupy overwhelming dominance.With the construction, popularization and operation of large-scale wind electricity unit, the failure of wind power generating set will increase year by year
Add, economic loss resulting from is also increasingly severe.Therefore, the critical components such as Wind turbines bearing are carried out in time, effectively
Failure predication become improve running of wind generating set reliability, reduce maintenance cost important channel.
In view of the above-mentioned problems, document [1] using EMD (empirical mode decomposition) to high speed shaft of aerogenerator hold signal into
Row pretreatment, and amplitude on the basis of Hilbert envelope spectrum analysis at extraction failure-frequency uses divergence as characteristic quantity
Index describes fault type and severity.Document [2] proposes the method for fault diagnosis of wind turbines based on Decision fusion, first
Tentative diagnosis is carried out using gray relative analysis method, decision information is then carried out to different evidences according to evidence fusion theory and is melted
Close, thus obtain fault diagnosis as a result, but Decision fusion method be both needed to acquire a large amount of historical data and carry out training pattern, than
It is more complex.Document [3] uses the method for diagnosing faults of support vector machines, according to wind driven generator output power predicted value and reality
Residual error variation between actual value judges that wind generator set main shaft holds health status, but SVM cannot exclude other kinds of failure to wind
Interference caused by motor group axle bearing fault diagnosis.
[1] Guo Yanping, the face text person of outstanding talent new IMF criterion of one kind and its in wind-driven generator rolling bearing fault diagnosis
It is studied using [J] computer application, 2012,29 (9): 3362-3364.
[2] An Xueli, Jiang Dongxiang, Lee lack direct wind-driven generator group bearing failure diagnosis of the China based on Decision fusion
[J] electric power network technique, 2011,35 (7): 36-41.
[3] Huang Yuan ties up wind generator set main shaft bearing failure diagnosis [J] the instrument and meter user of based on support vector machines,
2016,23(11):88-92。
Summary of the invention
There is non-stationary, nonlinear characteristic for current traditional Wind turbines bearing fault signal, fault signature extracts
Difficult problem, the present invention provide a kind of Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN, realize Wind turbines
The high-precision fault diagnosis of bearing.The technical solution adopted by the present invention is that:
A kind of Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN, due to the non-linear change of bearing vibration signal
Change, is first passed through polymerization empirical mode decomposition (abbreviation EEMD) method and be decomposed into multiple IMF components, it is several before selection wherein
IMF component establishes autoregression (abbreviation AR) model, calculates autoregressive coefficient and corresponding variance to each AR model, makees
For the input parameter of depth confidence network (abbreviation DBN network)；The low order for being extracted signal by EEMD and AR pretreatment is special
Sign；The Dividing Characteristics for recycling DBN network powerful, which extract, and generalization ability advantage is excavated from the parameter of input has representative
Property high-order feature, be used for fault diagnosis.
The present invention has the advantages that the present invention is used based on polymerization empirical mode decomposition and autoregression (EEMD-AR) model
Fault signature is extracted, using depth confidence network (DBN) method, the high-precision fault diagnosis of Wind turbines bearing is realized, than passing
The support vector machines and artificial neural network method for diagnosing faults of system have higher accuracy rate.
Detailed description of the invention
Fig. 1 is an exemplary diagram of DBN network of the invention.
Fig. 2 is bearing different faults type signal waveform diagram of the invention.
Fig. 3 is FPE (P) error coefficient change curve of IMF component 1 of the invention.
Fig. 4 is FPE (P) error coefficient change curve of IMF component 2 of the invention.
Fig. 5 is FPE (P) error coefficient change curve of IMF component 3 of the invention.
Fig. 6 is FPE (P) error coefficient change curve of IMF component 4 of the invention.
Specific embodiment
Below with reference to specific drawings and examples, the invention will be further described.
The present invention proposes a kind of Wind turbines Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN, due to axis
Vibration signal nonlinear change is held, polymerization empirical mode decomposition (abbreviation EEMD) method is first passed through and is decomposed into multiple IMF points
Amount selects wherein first four IMF component, establishes autoregression (abbreviation AR) model, calculate autoregression system to each AR model
Several and corresponding variance, the input parameter as depth confidence network (abbreviation DBN network)；It is mentioned by EEMD and AR pretreatment
The low order feature of signal is taken；The Dividing Characteristics for recycling DBN network powerful extract the parameter with generalization ability advantage from input
In excavate representative high-order feature, be used for fault diagnosis.
1) the AR aspect of model extracting method based on EEMD；
1.1) it polymerize empirical mode decomposition (EEMD)；
The bearing vibration signal s (t) of nonlinear and nonstationary can be decomposed into multiple IMF components by EEMD, such as formula (1) institute
Show:
R in formula (1)IIt is residual error, cε(t) (ε=1,2 ... I) indicate that the IMF component of frequency from high to low, t indicate that signal is adopted
At sample time point, IMF component must satisfy following two o'clock when being decomposed using EEMD:
(1) IMF component extreme point number and zero passage points it is identical or it is most difference one；
(2) the upper and lower envelope of IMF component is about time shaft Local Symmetric；
1.2) autoregression (AR) aspect of model extracts；
AR model is suitable for linear prediction, and parameter reflects the characteristic of system；However, AR model is not suitable for analyzing
Non-stationary signal.Therefore, it is necessary so original signal is carried out EEMD decomposition, establishes IMF component after decomposing based on EEMD
AR model；
For each IMF component, AR model is established using formula (2):
Wherein c (t) represents selected IMF component, and P is order, and w (t) is white noise, mean value zero； ak(k=1,2 ...
P autoregressive coefficient) is indicated；
AR model extraction characterization method based on EEMD mainly have the following three steps:
(1) original bearing vibration signal is subjected to EEMD decomposition, obtains IMF component；
(2) the optimal order of AR model is selected；
An optimal AR model is established, the selection of order is very important；If order is too high to be will lead to
Degree fitting, can also make calculation amount excessive；If order is too small, AR model cannot be made to be optimal.Utilize minimum pre- evaluated error
Criterion (FPE) selects optimal order P, selects the P corresponding when error coefficient FPE (P) minimum as optimal rank
Number, such as formula (3), wherein N indicates length, that is, number of sampling points of signal；
(3) optimal AR model is established to each component IMF, calculates autoregressive coefficient ak(k=1,2 ... P) and side
Difference
Least square method meter can be passed through after the optimal order of corresponding A R model determines for each component IMF
Calculate autoregressive coefficient, on the basis of formula (2), w (t) and varianceIt can be calculated by formula (4) and (5)；
For the AR model of a P rank, as soon as including P autoregressive coefficient and variance, there is P+1 characteristic value, because
This, the input of DBN network is exactly the vector of 4 × (P+1) dimension；
2) DBN Principles of Network；
Depth confidence network (deep belief network, DBN) is limited Boltzmann machine (restricted by multilayer
Boltzmann machine, RBM) successively superposition form deep layer network structure；Can by it is unsupervised it is layer-by-layer in a manner of efficiently
It is trained；Each RBM is made of visible layer and hidden layer, neuron in each RBM visible layer or hidden layer (or be
Node) between connection be restricted；In this example, DBN network is made of three layers of RBM；As shown in Figure 1；
DBN network learning procedure includes two stages: first stage, with layer-by-layer greedy learning algorithm pre-training RBM's
Layer；Second stage is finely adjusted whole network using backpropagation (BP) algorithm；
First stage, training data are first inputted in the visible layer of first layer RBM, with unsupervised mode training the
One layer of RBM；It is finished when upper one layer of RBM is trained, the hidden layer of this layer of RBM is exported into the input as next layer of RBM visible layer,
This training process is repeated, until RBM all in DBM network all training are completed；Each RBM includes multiple neurons, nerve
Member, which has, inactivates and activates two states, is indicated respectively with binary value 0 and 1；The energy function such as formula of RBM interlayer neuron
(6):
In formula: θ={ bj,cy,wi,j, wi,jFor the power between i-th of neuron of j-th of neuron of visible layer and hidden layer
Weight；vjAnd hiRespectively indicate the state of i-th of neuron of j-th of neuron of visible layer and hidden layer； bjAnd ciIt is respectively visible
The bias term of layer and hidden layer；
The Joint Distribution of visible layer and hidden layer neuron can be indicated by the energy of neuron are as follows:
Z (θ)=∑v,hexp(-E(v,h)) (8)
Wherein, Z (θ) is normalization factor；Above formula gives the general of each input vector after energy function indicates
Rate value；Probability value can change the energy value in formula (6) by changing parameter θ to be adjusted；
The condition distribution of i-th of neuron of hidden layer is expressed as in RBM:
The condition distribution of j-th of neuron of visible layer is expressed as in RBM:
Firstly, can determine the conditional probability of hidden layer neuron by formula (9), then is determined and implied by Gibbs sampling
The state of layer neuron；The conditional probability of visible layer neuron can be determined also with formula (10), then is sampled by Gibbs
Determine the state of visible layer neuron；This process is exactly the reconstruct of input data.If this model can be weighed very well
Structure input data shows that hidden layer neuron remains input signal characteristic information abundant, then being obtained by training study
Weight and biasing be exactly input data characterization rules.RBM is trained using contrast divergence algorithm herein, parameter θ is more
Newly, such as formula (11), (12), (13) are shown:
Δwi,j=η (< vjhi>p(h|v)-<vjhi>recon) (11)
Δbj=η (< vj>p(h|v)-<vj>recon) (12)
Δci=η (< hi>p(h|v)-<hi>recon) (13)
In formula: η is learning rate,<>p(h|v)Expectation under the distribution of expression condition,<>reconIndicate the distribution after reconstructing
Under expectation；
Second stage, the last layer, that is, output layer of DBM network are BP network, BP network with the feature of RBM layer output to
Amount is input vector；The network weight (including weight and biasing) of every layer of RBM although opposite this layer of feature vector is optimal,
But the network weight that not can guarantee entire DBN network is optimal, it is therefore desirable to carry out parameter optimization；By will reversely miss
Difference is top-down to propagate to every layer of RBM, entire DBN network weight is finely tuned, so that overall network performance be made to be optimal；
After the completion of DBN network learning procedure, the feature vector that the last layer RBM is exported in DBN network is exactly to have representative
The high-order feature of property；
3) wind turbine bearing failure diagnosis；
3.1) sample data acquires；
The historical data that experimental data is provided from certain wind power plant.Have chosen bearing inner race failure respectively, outer ring failure,
Rolling element failure and normal four kinds of different health status, label is 1,2,3,4 respectively.Data acquisition is bear vibration letter
Number, sampled point is set as 1000 and is located at driving end with 12kHz sampling, fault point, failure size has chosen respectively
Tri- kinds of different sizes of 0.1778mm, 0.3556mm, 0.5334mm.Be as shown in Figure 2 failure having a size of 0.1778mm when four kinds
The waveform of different health status.Wherein (a) represents inner ring failure, (b) represents rolling element failure, (c) represents outer ring failure, (d)
Represent normal condition.
Each failure size acquires 600 samples under same fault type, and wherein 400 samples are made for random selection
For training sample, remaining 200 are used as test sample.So inner ring failure, outer ring failure and rolling element failure all include
1800 samples, wherein having 1200 training samples and 600 test samples.There are 600 samples under normal condition, randomly chooses
Wherein 400 samples are used as test sample as training sample, remaining 200.Table 1 gives trained and test sample
Details.
Table 1
3.2) data prediction；
Original bearing vibration signal is decomposed by EEMD, since the information of bearing is mainly reflected on high frequency, institute
Only to consider first four IMF component herein.Then, optimal order P is determined using FPE criterion.Fig. 3~Fig. 6 is first four
FPE (P) error coefficient change curve of IMF component.
3~Fig. 6 of complex chart can be seen that when order P is in 24, and FPE (P) error coefficient of first four IMF component is several
It remains unchanged, therefore, optimal factor P is determined as 24.First four IMF component is all established to the AR model of 24 ranks, then
25 (24 autoregressive coefficients and a variance) a AR parameters of each IMF component are obtained using least square method.Due to each
Sample only considers first four IMF component, finally obtained input of the 100 dimension AR parameter vectors as DBN network.
3.3) experimental result；
DBN fault diagnosis network proposed in this paper includes three hidden layers, each hidden layer respectively by 100,50 and
10 neuron compositions.Input layer has 100 neurons to represent 100 dimension AR parameters, and output layer has 4 neurons to represent 4 kinds not
Same health status.The learning rate of each hidden layer is respectively 0.1,0.1,0.1.Table 2 is the detail parameters of DBN network；
Table 2
Using method for diagnosing faults proposed in this paper, when having 5 sample diagnosis inside erroneous judgement to other fault types: having
One inner ring fault sample is judged by accident to outer ring failure.Outer ring failure is judged by accident there are two sample, and one is mistaken for inner ring event
Barrier, another is mistaken for rolling element failure.Rolling element failure is also judged by accident there are two sample, and inner ring event is both mistaken for
Barrier.The Average Accuracy of method presented here has reached 99.75%.
In order to further prove superiority of the proposed method in wind turbine bearing fault detection, using SVM and
Two kinds of ANN common method for diagnosing faults are compared.ANN takes the network structure of 10-9-4, and input is from original signal
10 temporal signatures values of middle extraction, learning rate 0.08, the number of iterations are 300 times.SVM chooses Radial basis kernel function, punishment
Coefficient is 2050, kernel functional parameter 0.02, and input is also by EEMD+AR method treated data.DBN algorithm is used
The input data as ANN, other parameters with herein.The fault diagnosis accuracy rate of every kind of method is as shown in table 3.
Table 3
EEMD+AR+DBN method proposed in this paper is substantially better than traditional SVM and ANN method as can be seen from Table 3,
It proves the high-order feature that there is the depth confidence network of deep layer framework can sufficiently excavate signal, keeps accuracy rate of diagnosis higher.And
And signal is pre-processed using EEMD+AR model, the low order feature of signal is extracted, can obtain preferably diagnosing effect
Fruit.
It should be noted last that the above specific embodiment is only used to illustrate the technical scheme of the present invention and not to limit it,
Although being described the invention in detail referring to example, those skilled in the art should understand that, it can be to the present invention
Technical solution be modified or replaced equivalently, without departing from the spirit and scope of the technical solution of the present invention, should all cover
In the scope of the claims of the present invention.
Claims (7)
1. a kind of Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN characterized by comprising
Step S1 acquires bearing vibration signal, is decomposed into multiple IMF components by polymerizeing empirical mode decomposition, that is, EEMD method,
Several IMF components before selection wherein；
Step S2 establishes autoregression, that is, AR model of selected IMF component after decomposing based on EEMD；Each AR model is calculated
Autoregressive coefficient and corresponding variance parameter out, as depth confidence network, that is, DBN network input parameter；
Step S3 recycles DBN network to excavate representative high-order feature from the parameter of input, examines for failure
It is disconnected.
2. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as described in claim 1, it is characterised in that:
In step S1, EEMD method is decomposed, as shown in formula (1):
S (t) is bearing vibration signal, r in formula (1)IIt is residual error, cε(t) (ε=1,2 ... I) indicates IMF from high to low points of frequency
Amount, t indicate signal sampling time point.
3. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as claimed in claim 2, it is characterised in that:
In step S2, specifically include:
For each IMF component, AR model is established using formula (2):
Wherein c (t) represents selected IMF component, and P is order, and w (t) is white noise, mean value zero；ak(k=1,2 ... P) is indicated
Autoregressive coefficient；
Select the optimal order of AR model；
Autoregression is calculated by least square method after the optimal order of corresponding A R model determines for each component IMF
Coefficient, on the basis of formula (2), w (t) and varianceIt can be calculated by formula (4) and (5)；
Length, that is, number of sampling points of N expression signal.
4. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as claimed in claim 3, it is characterised in that:
In step S2, optimal order P is selected using minimum pre- measurement error criterion, selection is minimum as error coefficient FPE (P)
When corresponding P as optimal order, such as formula (3), wherein N indicates length, that is, number of samples of signal；Variance
5. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as claimed in claim 3, it is characterised in that:
The DBN network includes several RBM being successively superimposed, and each RBM is made of visible layer and hidden layer；
DBN network learning procedure includes two stages: first stage, with the layer of layer-by-layer greedy learning algorithm pre-training RBM；Second
Stage is finely adjusted entire DBM network using back-propagation algorithm, so that DBN overall network performance is optimal.
6. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as claimed in claim 5, it is characterised in that:
In the first stage of DBN network learning procedure, training data is first inputted in the visible layer of first layer RBM, with no prison
The mode training first layer RBM superintended and directed；When upper one layer of RBM training finishes, by the hidden layer output of this layer of RBM as next layer of RBM
The input of visible layer repeats this training process, until RBM all in DBM network all training are completed；Each RBM includes more
A neuron, neuron, which has, inactivates and activates two states, is indicated respectively with binary value 0 and 1；RBM interlayer neuron
Energy function such as formula (6):
In formula: θ={ bj,cy,wi,j, wi,jFor the weight between i-th of neuron of j-th of neuron of visible layer and hidden layer；vj
And hiRespectively indicate the state of i-th of neuron of j-th of neuron of visible layer and hidden layer；bjAnd ciRespectively visible layer and hidden
Bias term containing layer；
The Joint Distribution of visible layer and hidden layer neuron is indicated by the energy of neuron are as follows:
Z (θ)=∑v,hexp(-E(v,h)) (8)
Wherein, Z (θ) is normalization factor；
The condition distribution of i-th of neuron of hidden layer is expressed as in RBM:
The condition distribution of j-th of neuron of visible layer is expressed as in RBM:
Firstly, can determine the conditional probability of hidden layer neuron by formula (9), then is sampled by Gibbs and determine hidden layer mind
State through member；The conditional probability of visible layer neuron can be determined also with formula (10), then sampling determination by Gibbs can
See the state of layer neuron；RBM is trained using contrast divergence algorithm, parameter θ updates, such as formula (11), (12), (13) institute
Show:
Δwi,j=η (< vjhi>p(h|v)-<vjhi>recon) (11)
Δbj=η (< vj>p(h|v)-<vj>recon) (12)
Δci=η (< hi>p(h|v)-<hi>recon) (13)
In formula: η is learning rate,<>p(h|v)Expectation under the distribution of expression condition,<>reconIndicate the phase after reconstructing under the distribution
It hopes.
7. the Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN as claimed in claim 5, it is characterised in that:
After the completion of DBN network learning procedure, the feature vector that the last layer RBM is exported in DBN network is exactly representative
High-order feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810939116.XA CN109141884A (en) | 2018-08-17 | 2018-08-17 | Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810939116.XA CN109141884A (en) | 2018-08-17 | 2018-08-17 | Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109141884A true CN109141884A (en) | 2019-01-04 |
Family
ID=64790146
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810939116.XA Pending CN109141884A (en) | 2018-08-17 | 2018-08-17 | Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109141884A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110135564A (en) * | 2019-05-15 | 2019-08-16 | 深圳朴生智能科技有限公司 | Coal mine gas sensing data method for detecting abnormality based on edge calculations |
| CN111136509A (en) * | 2020-01-10 | 2020-05-12 | 大连理工大学 | Cutter state monitoring method based on ELM-SDAE algorithm |
| CN113405799A (en) * | 2021-05-20 | 2021-09-17 | 新疆大学 | Bearing early fault detection method based on health state index construction and fault early warning limit self-learning |
| CN113435327A (en) * | 2021-06-25 | 2021-09-24 | 西安交通大学 | Electric spindle state evaluation method, system, equipment and readable storage medium |
| CN114469131A (en) * | 2021-12-13 | 2022-05-13 | 中国科学院深圳先进技术研究院 | Self-adaptive real-time electrocardiosignal quality evaluation method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106895975A (en) * | 2017-01-17 | 2017-06-27 | 苏州大学 | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks |
-
2018
- 2018-08-17 CN CN201810939116.XA patent/CN109141884A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106895975A (en) * | 2017-01-17 | 2017-06-27 | 苏州大学 | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 王春梅: "基于深度置信网络的风电机组主轴承故障诊断方法研究", 《自动化仪表》 * |
| 王玉静 等: "基于EEMD-Hilbert包络谱和DBN的变负载下滚动轴承状态识别方法", 《中国电机工程学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110135564A (en) * | 2019-05-15 | 2019-08-16 | 深圳朴生智能科技有限公司 | Coal mine gas sensing data method for detecting abnormality based on edge calculations |
| CN111136509A (en) * | 2020-01-10 | 2020-05-12 | 大连理工大学 | Cutter state monitoring method based on ELM-SDAE algorithm |
| CN113405799A (en) * | 2021-05-20 | 2021-09-17 | 新疆大学 | Bearing early fault detection method based on health state index construction and fault early warning limit self-learning |
| CN113435327A (en) * | 2021-06-25 | 2021-09-24 | 西安交通大学 | Electric spindle state evaluation method, system, equipment and readable storage medium |
| CN113435327B (en) * | 2021-06-25 | 2023-12-19 | 西安交通大学 | Electric spindle state evaluation method, system, equipment and readable storage medium |
| CN114469131A (en) * | 2021-12-13 | 2022-05-13 | 中国科学院深圳先进技术研究院 | Self-adaptive real-time electrocardiosignal quality evaluation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109141884A (en) | Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN | |
| Zhang et al. | The multiclass fault diagnosis of wind turbine bearing based on multisource signal fusion and deep learning generative model | |
| Jin et al. | Actual bearing compound fault diagnosis based on active learning and decoupling attentional residual network | |
| Zhao et al. | Intelligent fault diagnosis of multichannel motor–rotor system based on multimanifold deep extreme learning machine | |
| Li et al. | Wind turbine fault diagnosis based on transfer learning and convolutional autoencoder with small-scale data | |
| Bommidi et al. | Hybrid wind speed forecasting using ICEEMDAN and transformer model with novel loss function | |
| Hou et al. | Sparse oblique decision tree for power system security rules extraction and embedding | |
| CN108363382A (en) | A kind of complex equipment fault diagnosis method and system | |
| CN112651290A (en) | Hydroelectric generating set vibration trend prediction method and system | |
| Li et al. | Multiscale dynamic fusion prototypical cluster network for fault diagnosis of planetary gearbox under few labeled samples | |
| Wang et al. | A high-stability diagnosis model based on a multiscale feature fusion convolutional neural network | |
| Yu et al. | Application of small sample virtual expansion and spherical mapping model in wind turbine fault diagnosis | |
| Yin et al. | Deep learning based feature reduction for power system transient stability assessment | |
| Luo et al. | Fault diagnosis method based on two-stage GAN for data imbalance | |
| Xiuli et al. | Fault diagnosis method of wind turbine gearbox based on deep belief network and vibration signal | |
| Cheng et al. | A novel planetary gearbox fault diagnosis method for nuclear circulating water pump with class imbalance and data distribution shift | |
| Zhang et al. | Reliability assessment of wind power converter considering SCADA multistate parameters prediction using FP-growth, WPT, K-means and LSTM network | |
| Fu et al. | Powercog: A practical method for recognizing power quality disturbances accurately in a noisy environment | |
| Karamti et al. | A new deep stacked architecture for multi-fault machinery identification with imbalanced samples | |
| CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
| Du et al. | From anomaly detection to novel fault discrimination for wind turbine gearboxes with a sparse isolation encoding forest | |
| Shi et al. | A novel rotating machinery fault diagnosis method based on adaptive deep belief network structure and dynamic learning rate under variable working conditions | |
| Xue et al. | A data-driven diagnosis method of open-circuit switch faults for PMSG-based wind generation system | |
| CN112163474A (en) | Intelligent gearbox diagnosis method based on model fusion | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |