CN113256443B - Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium - Google Patents
Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium Download PDFInfo
- Publication number
- CN113256443B CN113256443B CN202110474137.0A CN202110474137A CN113256443B CN 113256443 B CN113256443 B CN 113256443B CN 202110474137 A CN202110474137 A CN 202110474137A CN 113256443 B CN113256443 B CN 113256443B
- Authority
- CN
- China
- Prior art keywords
- sample
- fault
- vibration acceleration
- real
- generated
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 61
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 80
- 230000001133 acceleration Effects 0.000 claims abstract description 66
- 238000000034 method Methods 0.000 claims abstract description 56
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 22
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 22
- 238000001914 filtration Methods 0.000 claims abstract description 14
- 238000011425 standardization method Methods 0.000 claims abstract description 11
- 230000000087 stabilizing effect Effects 0.000 claims abstract description 4
- 238000009826 distribution Methods 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 16
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 238000001507 sample dispersion Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 239000006185 dispersion Substances 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 5
- 230000009977 dual effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 238000000844 transformation Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 abstract description 8
- 238000003745 diagnosis Methods 0.000 abstract description 7
- 238000012216 screening Methods 0.000 abstract 1
- 238000012423 maintenance Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 239000000498 cooling water Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
- Y02E30/30—Nuclear fission reactors
Abstract
The invention discloses a nuclear power water pump guide bearing fault detection method, a system, equipment and a readable storage medium, which are characterized in that an acquired vibration acceleration original signal is subjected to noise reduction through a threshold filtering method based on empirical mode decomposition, nonlinear and non-stationarity information of the signal can be fully reserved, the signal after noise reduction is ensured not to be distorted, then a bidirectional generation countermeasure network model containing a gradient punishment item is constructed, a fault sample generated by the vibration acceleration original signal in a fault state in the preprocessed vibration acceleration original signal is used for alternately training the bidirectional generation countermeasure network model, a single sample deviation standardization method is used for stabilizing a training process, a similarity index is constructed for screening BiGAN to generate a fault sample, a fault diagnosis model training dataset can be effectively expanded, the problem of unbalance of the training sample class is solved, the single sample deviation standardization method is used for stabilizing the countermeasure network training process, and the similarity index is constructed for improving the accuracy of a diagnosis model.
Description
Technical Field
The invention relates to the technical field of intelligent operation and maintenance, in particular to a method, a system and equipment for detecting faults of a guide bearing of a nuclear power water pump and a readable storage medium.
Background
Nuclear power is one of energy guarantees of economic and social development in China, and a circulating water pump is used as lifting equipment of a circulating water system of a nuclear power plant, and has the functions of providing cooling water for a turbine condenser of a conventional island and an auxiliary cooling system of the conventional island, so that once a fault affects the operation safety of a turbine, the nuclear power plant is stopped and shut down when serious, and huge economic loss is caused.
At present, the maintenance modes of the circulating water pump of the nuclear power plant comprise two modes of post maintenance and periodic maintenance. The existing two maintenance modes have the problems that potential defects are difficult to find, equipment is excessively maintained, and the like, namely: many devices with extremely high design reliability fail unexpectedly well below the expected lifetime, while others suffer from maintenance or even replacement while still operating reasonably well. As a key component of the circulating water pump, the guide shaft has high bearing capacity and is easy to damage, and intelligent monitoring and detection are carried out on the guide shaft, so that the normal and stable operation of the nuclear power circulating water pump can be effectively ensured, and serious accidents are prevented.
The deep learning obtains a detection model by utilizing the existing data, performs reasonable fault classification and prediction, provides necessary technical means for nuclear power intelligent operation and maintenance operation, and has wide research prospect. However, regular maintenance makes it difficult to obtain field failure data because the guide bearings are maintained or replaced before failure occurs. The data class imbalance caused by the problem can seriously reduce the accuracy of a data-driven fault detection model, namely: a limited training sample will result in incomplete model training and an inability to fit the distribution completely.
In addition, the current nuclear power unit mainly depends on engineering experience in maintenance scheme decision, the operation and maintenance management efficiency is not high, and the related invention aiming at a nuclear power water pump guide bearing fault detection model is not reported yet.
Disclosure of Invention
The invention aims to provide a nuclear power water pump guide bearing fault detection method, a system, equipment and a readable storage medium, so as to overcome the defects of the prior art.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a nuclear power water pump guide bearing fault detection method comprises the following steps:
s1, respectively acquiring vibration acceleration original signals of a guide bearing in a normal state and a fault state, denoising the acquired vibration acceleration original signals by a threshold filtering method based on empirical mode decomposition, and preprocessing the denoised vibration acceleration original signals;
s2, constructing a bidirectional generation countermeasure network model containing gradient punishment items, alternately training a fault sample generated by a vibration acceleration original signal in a fault state in the preprocessed vibration acceleration original signal to generate the bidirectional generation countermeasure network model, stabilizing a training process by adopting a single sample dispersion standardization method, and constructing a similarity index to screen BiGAN to generate the fault sample;
and S3, constructing a fault detection model of the guide bearing convolutional neural network, training the fault detection model by using the training set generated by the preprocessed vibration acceleration original signal and the generated fault sample obtained in the step S2 as the training set of the fault detection model, and detecting the monitored input data by using the trained fault detection model to realize the guide bearing fault detection of the nuclear power water pump.
Further, the threshold filtering method based on empirical mode decomposition specifically comprises the following steps:
firstly, EMD decomposition is carried out on the vibration acceleration original signal, and the method can be obtained:
wherein s is the original signal of vibration acceleration,
for the ith base pattern component, x k For the ith basic mode pseudo component, n and m are integers.
The index of the cross-correlation coefficient between each decomposed basic mode component and the original signal is as follows:
in the formula,
cross-correlating each fundamental mode component with the vibration acceleration raw signal; r is R s And (tau) is the autocorrelation of the vibration acceleration raw signal.
Further, the preprocessing comprises non-overlapping sampling with length 1024, fast Fourier transformation and dispersion normalization.
Further, the bidirectional generation countermeasure network model comprises a generator G, a discriminator D and an encoder E, wherein the encoder E is used for extracting the vibration acceleration original signal x real In (a) the hidden variable z, i.e. z=e (x real ) The generator G generates a true generated fault sample x by sampling from the random noise vector o generated =g (o); the discriminator D samples the true fault and its latent variables (z, x real ) And generating a fault sample x generate Noise (x) generate O) simultaneously taking the two types of problems as input of a network to train the two types of problems; the encoder E and generator G combine to attempt to confuse the arbiter when the arbiter fails to determine (z, x real )，(x generate When derived from (o), the description (z, x) real )，(x generate O) the joint data distribution is similar, when training reaches the optimal solution, G=E can be obtained -1 Thereby ensuring that the sample x is generated generate And true sample x real The data distribution is similar.
Further, the generation of the countermeasure network model uses a Wasserstein-1 distance and gradient penalty method to stabilize the training process, and the Wasserstein-1 distance calculation formula is defined as:
in the formula,A1 Is true data distribution, A 2 Is to generate data distribution, pi (A 1 ,A 2 ) Is A 1 and A2 A set of all joint distributions that are combined, γ being one of the joint distributions, (x, y) is a pair of samples in γ, E (x, y) γ [ ||x-y||]Is the expected value of the sample distance.
Applying the Wasserstein-1 distance and Kantorovich-Rubistein dual principle, the model training process can be expressed as:
where Ω is a set of 1-Lipschitz functions, D is a arbiter, G is a generator, E is an encoder,
representing the distribution of data from the real world->
Is>
Representing the distribution of data from noise->
Is a desired value of (2);
lipschitz constraint D ε Ω is implemented by limiting the gradient norms of the arbiter output relative to its inputs, and the bi-directional generation countermeasure network model loss function can be expressed as:
wherein ,
after the slave (x real,z) and (xgenerate O) on the sampled pair-point line, < ->
For its corresponding data distribution λ=10, +.>
Is a gradient penalty term.
Further, the single sample dispersion normalization method is used for sample x 1 ，x 2 ，x 3 ，…，x n The following transformations were performed:
wherein ,x1 ,x 2 ,x 3 ,…,x j For the jth data in a certain sample,
in order to correspond to the sample minimum data value,
for the most corresponding sampleLarge data values.
Further, similarity index C in The calculation formula is as follows:
C in ＝diag{FC,FSD}
wherein, FC is the center frequency, FV frequency variance, FSD is the standard deviation of frequency. The calculation formula is as follows:
wherein ,A(ft ) Is the amplitude, f t For frequency
Calculating Euclidean distance of real sample and generated sample index
If the calculation result
The generated fault sample has a small phase difference and can be considered to have high correlation with the real fault sample, and the generated fault sample can be used as the supplementary data of the real fault sample.
The nuclear power water pump guide bearing fault detection system comprises a preprocessing module, a fault sample generation module and a fault detection module;
the preprocessing module is used for denoising the obtained vibration acceleration original signal according to the vibration acceleration original signal under the normal state and the fault state of the guide bearing by a threshold filtering method based on empirical mode decomposition, and preprocessing the vibration acceleration original signal after denoising;
the fault sample generation module alternately trains the fault samples generated by the vibration acceleration original signals in the fault state in the preprocessed vibration acceleration original signals to generate a countering network model in a bidirectional manner, adopts a single sample dispersion standardization method to stabilize the training process, and builds a similarity index to screen BiGAN to generate the fault samples;
the fault detection module is used for performing self-training according to a training set generated by the preprocessed vibration acceleration original signal and a generated fault sample serving as the training set, detecting and detecting the monitoring input data by using a fault detection model after training is completed, and outputting a detection result.
The terminal equipment comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the nuclear power water pump guide bearing fault detection method when executing the computer program.
A computer readable storage medium storing a computer program which when executed by a processor performs the steps of the above-described method for detecting a failure of a guide bearing of a nuclear power water pump.
Compared with the prior art, the invention has the following beneficial technical effects:
according to the nuclear power water pump guide bearing fault detection method, vibration acceleration original signals under a normal state and a fault state of a guide bearing are respectively collected, the obtained vibration acceleration original signals are subjected to noise reduction through a threshold filtering method based on empirical mode decomposition, nonlinear and non-stationarity information of the signals can be fully reserved, the noise-reduced signals are ensured to be undistorted, then the noise-reduced vibration acceleration original signals are preprocessed, a bidirectional generation countermeasure network model containing gradient penalty items is constructed, fault samples generated by the vibration acceleration original signals under the fault state in the preprocessed vibration acceleration original signals are alternately trained to generate the countermeasure network model, a single sample deviation standardization method is adopted to stabilize a training process, similarity indexes are constructed to screen BiGAN to generate fault samples, a fault diagnosis model training data set can be effectively expanded, the problem of unbalance of training sample types is solved, the single sample deviation standardization method is adopted to stabilize the countermeasure network training process, and similarity indexes are constructed to improve the accuracy of a diagnosis model.
The invention realizes the fault diagnosis of the guide bearing of the nuclear power circulating water pump under the condition of few samples by combining the threshold filtering technology based on empirical mode decomposition, the technology of generating an antagonism network model, a depth convolution network model and the like. Technical support is provided for the transition of nuclear power from periodic maintenance to an optionally maintained operation mode.
Furthermore, a bidirectional generation countermeasure network model trained by combining the Wasserstein distance and gradient punishment method is established, so that a fault diagnosis model training data set can be effectively expanded, the problem of unbalance of training sample types is solved, and further, the precision of the fault diagnosis model is improved.
Drawings
FIG. 1 is a flow chart of a few sample intelligent fault detection method according to an embodiment of the present invention.
FIG. 2 shows the empirical mode decomposition of the original signal in an embodiment of the present invention.
FIG. 3 is a diagram showing the comparison of the original signal and the reconstructed signal in the embodiment of the present invention.
FIG. 4 is a schematic diagram of non-overlapping sampling in an embodiment of the present invention.
FIG. 5 is a schematic diagram of a model structure of an countermeasure network in an embodiment of the invention.
Fig. 6 is a comparison diagram of a dispersion normalization method according to an embodiment of the present invention, fig. 6a is a schematic diagram of a bi-directional generation countermeasure network encoder E, fig. 6b is a schematic diagram of a bi-directional generation countermeasure network encoder G, and fig. 6c is a schematic diagram of a bi-directional generation countermeasure network encoder D.
FIG. 7 is a diagram showing the comparison of an original sample and a generated sample in an embodiment of the present invention.
FIG. 8 is a comparison of fault samples and generation of fault samples in an embodiment of the present invention.
FIG. 9 is a graph showing the accuracy of the intelligent fault detection method with few samples according to the embodiment of the invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
as shown in fig. 1, in order to solve the problems of low detection model precision and difficult conventional generation countermeasure network training caused by insufficient guide bearing fault samples, the invention provides a nuclear power water pump guide bearing fault detection method. In model training, firstly, signal noise reduction is realized through a threshold filtering technology based on empirical mode decomposition, and FFT (fast Fourier transform) and random sample division are carried out; secondly, establishing a bidirectional generation countermeasure network model to realize fault sample expansion, designing a single sample dispersion standardization method to stabilize a training process, and constructing a similarity index to screen a fault sample generated by BiGAN; finally, training a convolutional neural network fault detection model by using an enhanced training set (a training set normal sample, a training set fault sample and a screened fault generation sample), so as to realize intelligent fault detection of few samples of the guide bearing of the circulating water pump; in the engineering application stage, monitoring data are input into a convolutional neural network fault detection model, the model outputs a fault detection result, and the updating interval of the output result is set to be 3s.
The specific nuclear power water pump guide bearing fault detection method comprises the following steps:
s1, respectively acquiring vibration acceleration original signals of a guide bearing in a normal state and a fault state, and carrying out noise reduction on the acquired vibration acceleration original signals by a threshold filtering method based on empirical mode decomposition (Empirical Mode Decomposition, EMD);
specifically, the threshold filtering method based on empirical mode decomposition (Empirical Mode Decomposition, EMD) specifically includes the following steps:
firstly, EMD decomposition is carried out on the vibration acceleration original signal, and the method can be obtained:
wherein s is the original signal of vibration acceleration,
for the ith base pattern component, x k For the ith basic mode pseudo component, n and m are integers.
The index of the cross-correlation coefficient between each decomposed basic mode component (IMF) and the vibration acceleration original signal is as follows:
in the formula,
the cross correlation between the IMF and the vibration acceleration original signal is realized; r is R s (τ) is the autocorrelation of the vibration acceleration raw signal, and the higher the index, the greater the correlation with the vibration acceleration raw signal.
Then, a kurtosis index for each basic mode component (IMF) is calculated:
wherein, mu and sigma are the mean value and standard deviation of the signal x respectively; e (t) represents an expected value.
When there are larger IMFs, it is shown that there are more impact components, i.e. more fault impacts remain in these IMFs after the original signal is decomposed. Through reasonable design kurtosis and cross-correlation coefficient threshold, IMF with high kurtosis value and cross-correlation coefficient index is selected for reconstruction, and signal noise reduction is achieved.
The results of the guide bearing wear failure sample correlation coefficient and kurtosis value calculation are shown in table 1:
table 1 plain bearing wear failure samples
The EMD decomposition result is shown in FIG. 2, and as can be seen from FIG. 2 and Table 1, the IMF1 and IMF2 kurtosis values and correlation coefficients are larger, which indicates that the signal component contains more impact components and has higher similarity with the original signal of the vibration acceleration, so that the first two IMFs are selected for signal reconstruction. As shown in fig. 3, the impact component of the time domain waveform of the reconstructed signal is more obvious, and the signal noise reduction is realized.
S2, preprocessing the original signals after noise reduction, wherein the preprocessing comprises non-overlapping sampling with the length of 1024, fast Fourier transform (Fast Fourier Transform, FFT) and dispersion standardization processing, and dividing the signals into a training set sample and a test set sample, wherein the vibration acceleration original signals in a fault state in the preprocessed vibration acceleration original signals generate fault samples, and the vibration acceleration original signals in a normal state in the preprocessed vibration acceleration original signals generate normal state samples;
as shown in fig. 4, let every 2048 time domain data points be one sample, and perform non-overlapping sampling; based on np.fft.fft (x) in python, 1024 points of fast fourier transform are completed, and finally, dispersion standardized preprocessing is performed, and the calculation formula is as follows:
where max is the maximum value of all sample data and min is the minimum value of all sample data.
According to the method, the original signals are randomly divided into a training set and a testing set according to the ratio of 7:3, wherein the training set is used for training the convolutional neural network fault detection model, and the testing set tests the trained convolutional neural network fault detection model and does not participate in the model training process.
S3, constructing a bidirectional generation countermeasure network (Bidirectional Generative Adversarial Network, biGAN) model containing gradient penalty items, alternately training the bidirectional generation countermeasure network (Bi-directional Generative Adversarial Network, biGAN) model by adopting the preprocessed fault samples, designing a single sample dispersion standardization method to stabilize a training process, and constructing a similarity index to screen the BiGAN to generate the fault samples;
as shown in fig. 5, the bidirectional generation countermeasure network model includes a generator G, a discriminator D, and an encoder E that extracts an original signal x real In (a) the hidden variable z, i.e. z=e (x real ) The generator G generates a true generated fault sample x by sampling from the random noise vector o generated =g (o). The discriminator D samples the true fault and its latent variables (z, x real ) And generating a fault sample x generate Noise (x) generate O) is used as the input of the network at the same time to train the classification problem so that the discriminator D can accurately recognize (z, x) real )，(x generate O) from the encoder E or the generator G. The encoder E and generator G combine to attempt to confuse the arbiter when the arbiter fails to determine (z, x real )，(x generate When derived from (o), the description (z, x) real )，(x generate O) the federated data distribution is similar. When training reaches the optimal solution, g=e can be obtained -1 The method comprises the following steps: x is x real ＝G(E(x real ) O=e (G (o)), thereby ensuring that samples x are generated generate And true sample x real Is sufficiently similar. Compared with the traditional generation countermeasure network, the bidirectional generation countermeasure network realizes the mapping of the potential semantic features of the generated data and the real data by introducing the encoder E, has the capability of representing more abstract features, and is more suitable for vibration signal generation.
However, the bidirectional generation countermeasure network also has the problems of difficult training, pattern collapse and the like, in order to solve the problems, martin Arjovsky proposes a WGAN, and the Wasserstein-1 distance is used for replacing the original JS divergence to measure the difference between two data distributions, so that the problems are basically solved. In order to further reduce the gradient disappearance or gradient explosion problem caused by WGAN, the invention adopts a method combining the Wasserstein-1 distance and gradient penalty to stabilize the training process, wherein the Wasserstein-1 distance calculation formula is defined as:
in the formula,A1 Is true data distribution, A 2 Is to generate data distribution, pi (A 1 ,A 2 ) Is A 1 and A2 A set of all joint distributions that are combined, γ being one of the joint distributions, (x, y) is a pair of samples in γ, E (x, y) γ [ ||x-y||]Is the expected value of the sample distance.
Applying the Wasserstein-1 distance and Kantorovich-Rubistein dual principle, the model training process can be expressed as:
where Ω is a set of 1-Lipschitz functions, D is a arbiter, G is a generator, E is an encoder,
representing the distribution of data from the real world->
Is>
Representing the distribution of data from noise->
Is a desired value of (2).
Lipschitz constraint D ε Ω is implemented by limiting the gradient norms of the arbiter output relative to its inputs, and the bi-directional generation countermeasure network model loss function can be expressed as:
wherein ,
after the slave (x real,z) and (xgenerate O) on the sampled pair-point line, < ->
For its corresponding data distribution λ=10, +.>
Is a gradient penalty term.Compared with the traditional generation of the countermeasure network model, the model has higher training stability.
The bidirectional generation countermeasure network structure is shown in fig. 5, wherein the activation function after the convolution layer adopts a ReLU; in order to stabilize the training process, a small number of fault samples in the original signals of the countermeasure network are generated in the training, and single sample dispersion normalization should be performed again, specifically as follows:
for sample x 1 ,x 2 ,x 3 ,…,x n The following transformations were performed:
wherein ,x1 ,x 2 ,x 3 ,…,x j For the jth data in a certain sample,
in order to correspond to the minimum value of the sample,
is the maximum value in the corresponding sample. Compared with the traditional dispersion normalization, the method has the advantages that the signal characteristics (frequency spectrum peak value and energy distribution of frequency) are not changed, and the signal characteristics are respectively measured in the intervals [0,1]The normalization process is performed to reduce the extremum among samples, so as to solve the problems of difficult convergence and poor stability of the generated counternetwork model, as shown in fig. 6, the stability of the generated model by adopting the single-sample dispersion normalization technology is higher.
Fig. 7 is a comparison of the generated fault sample and the actual fault sample, and it can be seen that the generated fault sample is similar to the spectrum peak value of the actual fault sample in both the low frequency band and the high frequency band, and the energy distribution of the frequency is similar to the actual fault sample in each frequency band. But the energy in the spectrum of the generated signal is not as concentrated as the real signal, and the partial side band peaks are slightly different, which may mean that the generated signal has more noise, and the generated vibration signal segment may be considered to be close to the real vibration signal segment with more noise.
In conclusion, the similarity between the generated signal and the real signal is high, and the generated fault sample can be used as the supplementary data of the real fault sample.
In order to further quantify the similarity between the generated fault sample and the real fault sample, the invention designs a similarity index C in The calculation formula is as follows:
C in ＝diag{FC,FSD}
wherein, FC is the center frequency, FV frequency variance, FSD is the standard deviation of frequency. The calculation formula is as follows:
wherein ,A(ft ) Is the amplitude, f t For frequency
Taking a generated sample as an example, the indexes are shown in table 2:
TABLE 2 failure sample accuracy index
Calculating Euclidean distance of real sample and generated sample index
(82.9 in this example), if the calculation result is equal to +.>
The difference between the generated sample (106.3 in this example) and the original sample is smaller (the first 20% of the generated sample is taken), and the generated failure sample can be considered to be the original sampleWith a high degree of correlation, the generated fault samples can be used as supplementary data to the actual fault samples.
And S4, constructing a sliding bearing convolutional neural network fault detection model by adopting a batch normalization and maximum pooling method.
In the step 4, the fault detection model at least comprises 4 convolution layers to realize the weight sharing of the neural network; and adding a batch normalization layer after each convolution layer to improve training speed and model generalization capability, setting an activation function as a ReLU and a loss function as a cross entropy loss function.
The specific structure is shown in table 3:
TABLE 3 structural parameters of convolutional neural network fault detection model
The input samples are convolved by utilizing a plurality of convolution cores, after the offset term is added, the corresponding feature map of the image is obtained through an activation function, and the mathematical expression of the convolution is as follows:
wherein ,
is the j element of the first layer; m is M j A j-th convolution region of the l-1 layer feature map; />
Is an element therein; />
Is a corresponding weight matrix; />
Is a bias term; f (·) is the activation function; convolutional neural network model by training->
Weight matrix values +.>
Bias term numerical value realizing classification task
Performing maximum value taking operation on the feature map output by the convolution layer in each non-overlapping region with the size of n multiplied by n by adopting a maximum value pooling method;
expanding the feature map into one-dimensional feature vectors, weighted summation and obtainable by activating the function:
y k ＝f(w k x k-1 +b k )
wherein k is the sequence number of the network layer; y is k The output of the full connection layer; x is x k-1 Is a one-dimensional feature vector; w (w) k Is a weight coefficient; b k Is a bias term;
training a fault detection model by adopting a back propagation algorithm, calculating the gradient of a loss function on each weight by utilizing a chain derivative, updating the weight according to a gradient descent algorithm, and solving a cross entropy function as a cost function used by a convolutional neural network, wherein the formula is as follows:
wherein, C represents cost, x represents samples, n represents total number of samples, a represents model output value, and y represents sample actual value.
The predicted fault type is output through a softmax function, wherein the softmax is a normalized exponential function, and is a popularization of a logic function and is defined as follows:
wherein ,Vi Is the output of the classifier front-stage output unit. i represents a category index, and the total number of categories is C. S is S i RepresentingThe ratio of the index of the current element to the sum of all element indices. It can "compress" a K-dimensional vector containing arbitrary real numbers into another K-dimensional real vector so that the range of each element is in [0,1 ]]And the sum of all elements is 1. Namely: by softmax index, the multi-class output values can be converted to relative probabilities.
S5, the training set generated after the pretreatment in the step S2 and the generated fault sample are combined to be used as a fault detection model training set for training, and the fault detection model after the training is used for detecting and detecting the monitoring input data so as to realize the fault detection of the guide bearing of the nuclear power water pump.
Different numbers of real fault samples and generated fault samples are set as training data sets, and detailed information is shown in table 4 and fig. 8, wherein the generated samples are not added to normal state samples, and are equal to 237.
Table 4 test set classification accuracy under different training samples
As can be seen from table 4 and fig. 9, as the number of training samples increases, the classification accuracy of the fault detection model increases. When using the enhanced data set, the model accuracy is improved higher than if only the real data were used. After enough training iteration, the generated data is used as a training sample, so that the fault detection model can achieve high-precision class prediction. The countermeasure generation network can generate more reasonable fault samples, so that the expansion of a training set is realized, and the problem of poor model precision caused by unbalanced data types is solved.
And inputting the monitoring data into a convolutional neural network fault detection model, outputting a fault detection result by the model, and setting the updating interval of the output result to be 3s.
In one embodiment of the present invention, there is provided a terminal device including a processor and a memory for storing a computer program including program instructions, the processor for executing the program instructions stored by the computer storage medium. The processor adopts a Central Processing Unit (CPU), or adopts other general purpose processors, digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), ready-made programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components and the like, which are a computation core and a control core of the terminal, and are suitable for realizing one or more instructions, in particular for loading and executing one or more instructions so as to realize corresponding method flows or corresponding functions; the processor provided by the embodiment of the invention can be used for the operation of the nuclear power water pump guide bearing fault detection method.
The nuclear power water pump guide bearing fault detection system can be used for realizing the nuclear power water pump guide bearing fault detection method in the embodiment, and specifically comprises a preprocessing module, a fault sample generation module and a fault detection module;
the preprocessing module is used for denoising the obtained vibration acceleration original signal according to the vibration acceleration original signal under the normal state and the fault state of the guide bearing by a threshold filtering method based on empirical mode decomposition, and preprocessing the vibration acceleration original signal after denoising;
the fault sample generation module alternately trains the fault samples generated by the vibration acceleration original signals in the fault state in the preprocessed vibration acceleration original signals to generate a countering network model in a bidirectional manner, adopts a single sample dispersion standardization method to stabilize the training process, and builds a similarity index to screen BiGAN to generate the fault samples;
the fault detection module is used for performing self-training according to a training set generated by the preprocessed vibration acceleration original signal and a generated fault sample serving as the training set, detecting and detecting the monitoring input data by using a fault detection model after training is completed, and outputting a detection result.
In still another embodiment of the present invention, a storage medium, specifically a computer readable storage medium (Memory), is a Memory device in a terminal device, for storing programs and data. The computer readable storage medium includes a built-in storage medium in the terminal device, provides a storage space, stores an operating system of the terminal, and may also include an extended storage medium supported by the terminal device. Also stored in the memory space are one or more instructions, which may be one or more computer programs (including program code), adapted to be loaded and executed by the processor. The computer readable storage medium may be a high-speed RAM memory or a Non-volatile memory (Non-volatile memory), such as at least one magnetic disk memory. One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the corresponding steps of the method for detecting a fault of a guide bearing of a nuclear power water pump in the above embodiments.
The above specific embodiments are used for further detailed description of the objects, technical solutions and advantageous effects of the present invention. It should be understood that the foregoing description is only of specific embodiments of the present invention and is not intended to limit the invention, but rather should be construed to cover all modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention.
Claims (6)
1. The nuclear power water pump guide bearing fault detection method is characterized by comprising the following steps of:
s1, respectively acquiring vibration acceleration original signals of a guide bearing in a normal state and a fault state, denoising the acquired vibration acceleration original signals by a threshold filtering method based on empirical mode decomposition, and preprocessing the denoised vibration acceleration original signals;
the threshold filtering method based on empirical mode decomposition specifically comprises the following steps:
firstly, EMD decomposition is carried out on the vibration acceleration original signal, and the method can be obtained:
wherein s is vibrationThe original signal of the acceleration is obtained,
for the ith base pattern component, x k N and m are integers which are the i-th basic mode pseudo components;
the index of the cross-correlation coefficient between each decomposed basic mode component and the original signal is as follows:
in the formula,
cross-correlating each fundamental mode component with the vibration acceleration raw signal; r is R s (τ) is the autocorrelation of the vibration acceleration raw signal;
s2, constructing a bidirectional generation countermeasure network model containing gradient punishment items, alternately training a fault sample generated by a vibration acceleration original signal in a fault state in the preprocessed vibration acceleration original signal to generate the bidirectional generation countermeasure network model, stabilizing a training process by adopting a single sample dispersion standardization method, and constructing a similarity index to screen BiGAN to generate the fault sample;
the bidirectional generation countermeasure network model comprises a generator G, a discriminator D and an encoder E, wherein the encoder E is used for extracting a vibration acceleration original signal x real In (a) the hidden variable z, i.e. z=e (x real ) The generator G generates a true generated fault sample x by sampling from the random noise vector o generated =g (o); the discriminator D samples the true fault and its latent variables (z, x real ) And generating a fault sample x generate Noise (x) generate O) simultaneously taking the two types of problems as input of a network to train the two types of problems; the encoder E and generator G combine to attempt to confuse the arbiter when the arbiter fails to determine (z, x real )，(x generate When derived from (o), the description (z, x) real )，(x generate O) the joint data distribution is similar, when training reaches the optimal solution, G=E can be obtained -1 Thereby ensuring that the sample x is generated generate And true sample x real The data distribution is similar;
generation of the challenge network model the training process was stabilized using the Wasserstein-1 distance and gradient penalty method, the Wasserstein-1 distance calculation being defined as:
in the formula,A1 Is true data distribution, A 2 Is to generate data distribution, pi (A 1 ,A 2 ) Is A 1 and A2 A set of all joint distributions that are combined, γ being one of the joint distributions, (x, y) is a pair of samples in γ, E (x, y) γ [ ||x-y||]Is the expected value of the sample distance;
applying the Wasserstein-1 distance and Kantorovich-Rubistein dual principle, the model training process can be expressed as:
where Ω is a set of 1-Lipschitz functions, D is a arbiter, G is a generator, E is an encoder,
representing the distribution of data from the real world->
Is>
Representing the distribution of data from noise->
Is a desired value of (2);
lipschitz constraint D ε Ω is implemented by limiting the gradient norms of the arbiter output relative to its inputs, and the bi-directional generation countermeasure network model loss function can be expressed as:
wherein ,
after the slave (x real,z) and (xgenerate O) on the sampled pair-point line, < ->
For its corresponding data distribution λ=10, +.>
A gradient penalty term;
similarity index C in The calculation formula is as follows:
C in ＝diag{FC,FSD}
wherein, FC is the center frequency, FV frequency variance and FSD is the frequency standard deviation; the calculation formula is as follows:
wherein ,A(ft ) Is the amplitude, f t Is frequency;
calculating Euclidean distance of real sample and generated sample index
If the calculation result
The phase difference is small, the generated fault sample and the real fault sample can be considered to have high correlation, and the generated fault sample can be used as the supplementary data of the real fault sample;
and S3, constructing a fault detection model of the guide bearing convolutional neural network, training the fault detection model by using the training set generated by the preprocessed vibration acceleration original signal and the generated fault sample obtained in the step S2 as the training set of the fault detection model, and detecting the monitored input data by using the trained fault detection model to realize the guide bearing fault detection of the nuclear power water pump.
2. The method for detecting the faults of the guide bearing of the nuclear power water pump according to claim 1, wherein the preprocessing comprises non-overlapping sampling, fast Fourier transformation and dispersion standardization processing with the length of 1024.
3. The method for detecting faults of guide bearings of nuclear power water pumps according to claim 1, characterized in that a single sample dispersion normalization method is adopted for samples x 1 ,x 2 ,x 3 ,…,x n The following transformations were performed:
wherein ,x1 ,x 2 ,x 3 ,…,x j For the jth data in a certain sample,
in order to correspond to the sample minimum data value,
is the maximum data value in the corresponding sample.
4. The nuclear power water pump guide bearing fault detection system is characterized by comprising a preprocessing module, a fault sample generation module and a fault detection module;
the preprocessing module is used for denoising the obtained vibration acceleration original signal according to the vibration acceleration original signal under the normal state and the fault state of the guide bearing by a threshold filtering method based on empirical mode decomposition, and preprocessing the vibration acceleration original signal after denoising;
the threshold filtering method based on empirical mode decomposition specifically comprises the following steps:
firstly, EMD decomposition is carried out on the vibration acceleration original signal, and the method can be obtained:
wherein s is the original signal of vibration acceleration,
for the ith base pattern component, x k N and m are integers which are the i-th basic mode pseudo components;
the index of the cross-correlation coefficient between each decomposed basic mode component and the original signal is as follows:
in the formula,
cross-correlating each fundamental mode component with the vibration acceleration raw signal; r is R s (τ) is the autocorrelation of the vibration acceleration raw signal;
the fault sample generation module alternately trains the fault samples generated by the vibration acceleration original signals in the fault state in the preprocessed vibration acceleration original signals to generate a countering network model in a bidirectional manner, adopts a single sample dispersion standardization method to stabilize the training process, and builds a similarity index to screen BiGAN to generate the fault samples;
the fault detection module is used for performing self-training according to a training set generated by the preprocessed vibration acceleration original signal and a generated fault sample serving as the training set, detecting and detecting the monitored input data by using a fault detection model after training is completed, and outputting a detection result;
the bidirectional generation countermeasure network model comprises a generator G, a discriminator D and an encoder E, wherein the encoder E is used for extracting a vibration acceleration original signal x real In (a) the hidden variable z, i.e. z=e (x real ) The generator G generates a true generated fault sample x by sampling from the random noise vector o generated =g (o); the discriminator D samples the true fault and its latent variables (z, x real ) And generating a fault sample x generate Noise (x) generate O) simultaneously taking the two types of problems as input of a network to train the two types of problems; the encoder E and generator G combine to attempt to confuse the arbiter when the arbiter fails to determine (z, x real )，(x generate When derived from (o), the description (z, x) real )，(x generate O) the joint data distribution is similar, when training reaches the optimal solution, G=E can be obtained -1 Thereby ensuring that the sample x is generated generate And true sample x real The data distribution is similar;
generation of the challenge network model the training process was stabilized using the Wasserstein-1 distance and gradient penalty method, the Wasserstein-1 distance calculation being defined as:
in the formula,A1 Is true data distribution, A 2 Is to generate data distribution, pi (A 1 ,A 2 ) Is A 1 and A2 A set of all joint distributions that are combined, γ being one of the joint distributions, (x, y) is a pair of samples in γ, E (x, y) γ [ ||x-y||]Is the expected value of the sample distance;
applying the Wasserstein-1 distance and Kantorovich-Rubistein dual principle, the model training process can be expressed as:
where Ω is a set of 1-Lipschitz functions, D is a arbiter, G is a generator, E is an encoder,
representing the distribution of data from the real world->
Is>
Representing the distribution of data from noise->
Is a desired value of (2);
lipschitz constraint D ε Ω is implemented by limiting the gradient norms of the arbiter output relative to its inputs, and the bi-directional generation countermeasure network model loss function can be expressed as:
wherein ,
after the slave (x real,z) and (xgenerate O) on the sampled pair-point line, < ->
For its corresponding data distribution λ=10, +.>
A gradient penalty term;
similarity index C in The calculation formula is as follows:
C in ＝diag{FC,FSD}
wherein, FC is the center frequency, FV frequency variance and FSD is the frequency standard deviation; the calculation formula is as follows:
wherein ,A(ft ) Is the amplitude, f t Is frequency;
calculating Euclidean distance of real sample and generated sample index
If the calculation result
The generated fault sample has a small phase difference and can be considered to have high correlation with the real fault sample, and the generated fault sample can be used as the supplementary data of the real fault sample.
5. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 3 when the computer program is executed by the processor.
6. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110474137.0A CN113256443B (en) | 2021-04-29 | 2021-04-29 | Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110474137.0A CN113256443B (en) | 2021-04-29 | 2021-04-29 | Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113256443A CN113256443A (en) | 2021-08-13 |
| CN113256443B true CN113256443B (en) | 2023-06-20 |
Family
ID=77223664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110474137.0A Active CN113256443B (en) | 2021-04-29 | 2021-04-29 | Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113256443B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113714766A (en) | 2021-09-27 | 2021-11-30 | 烟台杰瑞石油装备技术有限公司 | Automatic disassembling and assembling system and method for plunger pump |
| CN114417699A (en) * | 2021-12-10 | 2022-04-29 | 烟台杰瑞石油服务集团股份有限公司 | Pump valve fault detection method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN111458142A (en) * | 2020-04-02 | 2020-07-28 | 苏州智传新自动化科技有限公司 | Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10753997B2 (en) * | 2017-08-10 | 2020-08-25 | Siemens Healthcare Gmbh | Image standardization using generative adversarial networks |
| CN109933881A (en) * | 2019-03-06 | 2019-06-25 | 武汉大学 | A kind of Fault Diagnosis of Power Electronic Circuits method based on optimization deepness belief network |
-
2021
- 2021-04-29 CN CN202110474137.0A patent/CN113256443B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN111458142A (en) * | 2020-04-02 | 2020-07-28 | 苏州智传新自动化科技有限公司 | Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| 不均衡数据集下基于生成对抗网络的改进深度模型故障识别研究;包萍;刘运节;;电子测量与仪器学报(第03期);182-189 * |
| 基于自组织神经网络的滚动轴承状态评估方法;张全德;陈果;林桐;欧阳文理;滕春禹;王洪伟;;中国机械工程(第05期);50-58 * |
| 采用非平衡小样本数据的风机主轴承故障深度对抗诊断;黄南天;杨学航;蔡国伟;宋星;陈庆珠;赵文广;;中国电机工程学报(第02期);215-226 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113256443A (en) | 2021-08-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | A novel feature enhancement method based on improved constraint model of online dictionary learning | |
| Zhao et al. | A weighted multi-scale dictionary learning model and its applications on bearing fault diagnosis | |
| Han et al. | Intelligent fault diagnosis method for rotating machinery via dictionary learning and sparse representation-based classification | |
| Jiang et al. | An efficient fault diagnostic method for three-phase induction motors based on incremental broad learning and non-negative matrix factorization | |
| CN113256443B (en) | Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium | |
| Han et al. | Fault diagnosis system of induction motors based on neural network and genetic algorithm using stator current signals | |
| Gao et al. | ASM1D-GAN: An intelligent fault diagnosis method based on assembled 1D convolutional neural network and generative adversarial networks | |
| Zhang et al. | An FSK-MBCNN based method for compound fault diagnosis in wind turbine gearboxes | |
| Liu et al. | Multi-feature fusion for fault diagnosis of rotating machinery based on convolutional neural network | |
| Zhang et al. | A multi-module generative adversarial network augmented with adaptive decoupling strategy for intelligent fault diagnosis of machines with small sample | |
| Liu et al. | Rolling bearing fault diagnosis based on EEMD sample entropy and PNN | |
| Saravanakumar et al. | Hierarchical symbolic analysis and particle swarm optimization based fault diagnosis model for rotating machineries with deep neural networks | |
| Zhang et al. | An intelligent fault diagnosis method of rolling bearings based on short-time Fourier transform and convolutional neural network | |
| CN107590604B (en) | Coherent unit grouping method and system combining S transformation and 2DPCA | |
| CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| Zheng et al. | Recognition method of voltage sag causes based on two‐dimensional transform and deep learning hybrid model | |
| Cheng et al. | A novel planetary gearbox fault diagnosis method for nuclear circulating water pump with class imbalance and data distribution shift | |
| Ge et al. | A deep condition feature learning approach for rotating machinery based on MMSDE and optimized SAEs | |
| Zhang et al. | Bearing fault diagnosis with kernel sparse representation classification based on adaptive local iterative filtering-enhanced multiscale entropy features | |
| Zhao et al. | Fault diagnosis of rolling bearings with noise signal based on modified kernel principal component analysis and DC‐ResNet | |
| CN116186586A (en) | Rolling bearing fault diagnosis method based on improved empirical mode decomposition algorithm and optimized deep confidence network | |
| Zhou et al. | Bearing fault diagnosis based on RF-PCA-LSTM model | |
| Su et al. | PSR-LSTM model for weak pulse signal detection | |
| Qi et al. | A new deep fusion network for automatic mechanical fault feature learning | |
| Hao et al. | New fusion features convolutional neural network with high generalization ability on rolling bearing 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 |