CN116625686A - On-line diagnosis method for bearing faults of aero-engine - Google Patents
On-line diagnosis method for bearing faults of aero-engine Download PDFInfo
- Publication number
- CN116625686A CN116625686A CN202310490570.2A CN202310490570A CN116625686A CN 116625686 A CN116625686 A CN 116625686A CN 202310490570 A CN202310490570 A CN 202310490570A CN 116625686 A CN116625686 A CN 116625686A
- Authority
- CN
- China
- Prior art keywords
- data
- bearing
- data set
- training
- vibration characteristic
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 58
- 230000009467 reduction Effects 0.000 claims abstract description 34
- 238000013210 evaluation model Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 6
- 238000012216 screening Methods 0.000 claims abstract description 6
- 238000013461 design Methods 0.000 claims abstract description 5
- 230000001133 acceleration Effects 0.000 claims description 24
- 101100290354 Arabidopsis thaliana AMC1 gene Proteins 0.000 claims description 6
- 101100456241 Arabidopsis thaliana AMC2 gene Proteins 0.000 claims description 6
- 201000004009 Neurogenic arthrogryposis multiplex congenita Diseases 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000002405 diagnostic procedure Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 14
- 230000007246 mechanism Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The application belongs to the field of aeroengine design, and relates to an on-line diagnosis method for bearing faults of an aeroengine, which comprises the steps of firstly establishing a vibration characteristic data set and an off-line training set, performing dimension reduction processing on the off-line data set by using a PCA algorithm, and training the dimension reduction data set to obtain a first Softmax classifier; inputting the dimensionality reduction data set into an ACGAN algorithm model with a self-attention model, training to obtain a corresponding generator and a discriminator, and evaluating lambda by using a bearing fault evaluation model mean The value and screening out the effective synthetic data, and combining the effective synthetic data with the offline training set after dimension reduction to form an enhanced data set; fusing the enhanced data set and the reduced data set to obtain an extended data set, and training to obtain a second Softmax classifier; and inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and carrying out bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis result. Accurate and quick judgment can be made on line, and bearing faults can be effectively diagnosed.
Description
Technical Field
The application belongs to the field of aero-engine design, and particularly relates to an on-line diagnosis method for aero-engine bearing faults.
Background
The bearing is used as an important part of an aeroengine rotor system, and is in a working condition of severe high-speed fluctuation, large load and obvious change for a long time, so that performance degradation and even various faults are caused. Therefore, in order to ensure that the aeroengine safely and efficiently operates and save maintenance cost, the operation state and change rule of the aeroengine must be mastered, and the automatic and accurate diagnosis of the faults of the high-speed bearing of the aeroengine is beneficial to improving the operation safety and maintenance economy of the rotor system of the aeroengine.
The existing diagnostic methods have the following disadvantages:
1. weak signals of early failure of the bearing cannot be identified;
2. weak fault characteristics are difficult to extract;
3. the actual bearing failure samples are few.
The traditional fault diagnosis method is difficult to adapt to the current bearing diagnosis application by depending on a large number of fault diagnosis algorithms and expert engineering experience. Aiming at the current massive data, the fault information is complex and changeable, comprises internal and external excitation and the coupling of a plurality of faults, and the fault diagnosis method based on data driving is needed to directly utilize the acquired parameters to mine the data and identify the key characteristic information of the engine bearing to identify the faults.
Disclosure of Invention
The application aims to provide an on-line diagnosis method for bearing faults of an aeroengine, which aims to solve the problem of low diagnosis accuracy caused by fewer bearing fault samples and weak signals.
The technical scheme of the application is as follows: an on-line diagnosis method for bearing faults of an aeroengine, comprising the following steps:
establishing a vibration characteristic data set, and acquiring vibration characteristic values of vibration data construction working conditions of the bearing under different operation working conditions; extracting time domain indexes of different vibration data of the bearing to obtain a time domain vibration characteristic value; extracting frequency domain indexes of different vibration data of the bearing to obtain frequency domain vibration characteristic values; inputting the working condition vibration characteristic value, the time domain vibration characteristic value and the frequency domain vibration characteristic value into a vibration characteristic data set;
based on the vibration characteristic data set, extracting vibration characteristic values of the case library data again, inputting the vibration characteristic data set to form an offline training set, performing dimension reduction processing on the offline data set by using a PCA algorithm, and reducing the dimension of data in the offline training set to m dimension to form a dimension reduction data set;
training the dimension reduction data set to obtain a first Softmax classifier;
establishing a self-attention model, inputting a reduced-dimension data set into an ACGAN algorithm model with the self-attention model, training to obtain a corresponding generator and a corresponding discriminator, and simultaneously inputting synthetic data generated by training into a first Softmax classifier to form a bearing fault evaluation model;
estimating lambda using bearing failure estimation model mean The value is used for screening out the synthetic data which is judged to be effective, and the synthetic data is combined with the offline training set after the dimension reduction to form an enhanced data set lambda mean The value calculation formula is:
wherein a is AMC1 The proportion of the normal correct classification of the bearing to the total classification number is predicted; a, a AMC2 The proportion of the bearing fault classification to the total classification number is predicted; f (f) 11 The number of faults predicted as faulty bearings; f (f) 00 A number predicted to be normal for a normal bearing; f (f) 10 The number of bearings predicted to be normal for failure; f (f) 01 Is pre-arranged for normal bearingsMeasuring the number of faults; lambda (lambda) mean Is a as AMC1 And a AMC2 Is a geometric average of (2);
fusing the enhanced data set and the reduced data set to obtain an extended data set, and training the extended data set to obtain a second Softmax classifier for performing on-line fault diagnosis;
and acquiring online data in real time, obtaining a data set by extracting vibration characteristics, performing PCA dimension reduction on the acquired data to form an online diagnosis set, inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and performing bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis result.
Preferably, the generating method of the enhanced data set includes:
random noise is generated by random sampling in Gaussian distribution, the random noise is classified, and the random noise and an auxiliary classification label corresponding to the random noise are input into a generator to obtain generated data;
inputting the generated data and the real data of the vibration characteristics of the bearing through acquisition into a discriminator to form mixed data, discriminating the mixed data to obtain real/false data and different fault type data, and completing one-time training;
and respectively and sequentially updating network parameters of the first Softmax classifier and the discriminator, acquiring the generated data and the real data again for cyclic discrimination, and inputting the synthetic data with effective discrimination into the enhanced data set until the set cycle number is completed or the data volume in the enhanced data set meets the design requirement.
Preferably, the working condition vibration characteristic value comprises an average value of the rotation speed of the sample input shaft, a variance of the rotation speed of the sample input shaft, an average value of the sample torque and a variance of the sample torque; the time domain vibration characteristic value comprises an acceleration effective value, an acceleration waveform index, an acceleration peak value index, an acceleration pulse index, an acceleration margin index, an acceleration kurtosis index, a speed effective value, a speed waveform index, a speed peak value index, a speed pulse index, a speed margin index and a speed kurtosis index; the frequency domain vibration characteristic value comprises the frequency values of the 1, 2 and 3-order amplitude values of the shaft rotation frequency where the bearing is located and the front 100 amplitude values in the frequency spectrum.
Preferably, the method for establishing the self-attention model comprises the following steps:
three channel nodes F (x), G (x) and H (x) are arranged, and different weight matrixes W are respectively arranged on different channel nodes K 、W Q And W is V Multiplying the input x by different weight matrices to obtain:
calculating an incidence matrix S between F (x) and G (x):
S＝F T ·G
calculating the degree of correlation between F (x) and G (x):
wherein: s is(s) ij Is an element of the incidence matrix S;
output features of the self-attention model are obtained:
adding a weight coefficient gamma, and integrating the field information and the remote features to obtain:
y i ＝γo i +x i 。
preferably, the method for training the reduced-dimension data set by using the ACGAN algorithm model comprises the following steps:
define dimension reduction dataset x= { X 1 ,x 2 ,…,x N X, where x i For the ith sample, N is the total number of samples; the loss function of generator G is set to:
adding the auxiliary classification label c into a generator to obtain a loss function L for representing whether the data is real or not S The method comprises the following steps:
loss function L of discriminator D D The method comprises the following steps:
setting an auxiliary classifier C and introducing the auxiliary classifier C into a discriminator D to obtain a loss function L for representing the classification accuracy of the data C The method comprises the following steps:
introducing Wo Sesi metric, constructing an objective function of overall training, measuring the difference between a generated sample and a real sample, and constructing an ACGAN model based on self-attention, wherein the objective function of the ACGAN overall training based on self-attention is as follows:
preferably, the gradient penalty term is constructed while the objective function is being constructed:
wherein: I.I p Represents the p-norm; lambda is a regularized term coefficient;wherein epsilon-U (0, 1), U is uniformly distributed.
The application relates to an aeroengine bearingThe fault online diagnosis method comprises the steps of firstly collecting vibration data of different types, establishing a vibration characteristic data set, then extracting vibration characteristic values of case library data again, inputting the vibration characteristic values into the data set to form an offline training set, performing dimension reduction processing on the offline data set by using a PCA algorithm, and training the dimension reduction data set to obtain a first Softmax classifier; inputting the dimensionality reduction data set into an ACGAN algorithm model with a self-attention model, training to obtain a corresponding generator and a discriminator, and evaluating lambda by using a bearing fault evaluation model mean The value and screening out the effective synthetic data, and combining the effective synthetic data with the offline training set after dimension reduction to form an enhanced data set; fusing the enhanced data set and the reduced data set to obtain an extended data set, and training to obtain a second Softmax classifier; and inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and carrying out bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis result. The method can accurately and quickly judge and diagnose the bearing faults on line, has higher recognition precision and stability, and is suitable for on-site real-time diagnosis application.
Drawings
In order to more clearly illustrate the technical solution provided by the present application, the following description will briefly refer to the accompanying drawings. It will be apparent that the figures described below are merely some embodiments of the application.
FIG. 1 is a schematic diagram of the overall flow of the present application;
FIG. 2 is a schematic diagram of a self-attention model structure according to the present application;
FIG. 3 is a schematic diagram of an enhanced data set generation method according to the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application become more apparent, the technical solutions in the embodiments of the present application will be described in more detail below with reference to the accompanying drawings in the embodiments of the present application.
An on-line diagnosis method for bearing faults of an aeroengine, as shown in figure 1, comprises the following steps:
step S100, selecting vibration characteristic value and inputting vibration characteristic data set
(1) Because the running states of the bearing of the aero-engine under different working conditions have certain differences, the working condition vibration characteristic value is constructed by introducing the running state differences of the running states of the bearing and is input into the vibration characteristic data set. Thus, choose: the average value of the sample input shaft rotating speed, the variance of the sample input shaft rotating speed, the average value of the sample torque and the variance of the sample torque are combined into 4 working condition vibration characteristic values.
(2) Extracting time domain indexes of the vibration condition of the bearing, and reflecting different running states representing the bearing through the vibration time domain indexes. Thus, choose: an acceleration effective value, an acceleration waveform index, an acceleration peak value index, an acceleration pulse index, an acceleration margin index, an acceleration kurtosis index, a speed effective value, a speed waveform index, a speed peak value index, a speed pulse index, a speed margin index and a speed kurtosis index, wherein the total of the acceleration effective value, the acceleration waveform index, the acceleration pulse index, the acceleration margin index, the acceleration kurtosis index and the acceleration kurtosis index are 12 time domain vibration characteristic values.
(3) And extracting frequency domain indexes of the vibration condition of the bearing, and reflecting different health states representing the bearing through the vibration frequency domain indexes. Thus, choose: the shaft where the bearing is located rotates frequency domain vibration characteristic values (103 characteristic values in total) of 1, 2 and 3 orders of amplitude and the front 100 of amplitude in the frequency spectrum. Thus, the 119-dimensional features are co-constructed to form the vibration feature value input vibration feature dataset.
Step S200, PCA dimension reduction
Based on the vibration characteristic data set, extracting vibration characteristic values of the case library data again, inputting the vibration characteristic data set to form an offline training set, performing dimension reduction processing on the offline data set by using a PCA algorithm, reducing the dimension of data in the offline training set to m dimensions, and enabling the m values to automatically select the numerical values according to the effect to form a dimension reduction data set.
The PCA dimension reduction is packaged by a function, the dimension reduction is carried out by transmitting a data set and setting dimension reduction dimension, the PCA dimension reduction can eliminate factors which affect each other between original data, the calculated amount is reduced, and the PCA algorithm adopts the existing algorithm and is not described in detail.
Step S300, training the reduced-dimension data set, and obtaining a first Softmax classifier by setting a corresponding validity standard or threshold value for vibration feature data in the reduced-dimension data set, where the classifier is used to evaluate validity of synthetic data generated in the training process based on self-attention ACGAN, and a training method of the Softmax classifier is in the prior art and is not described in detail.
Step S400, a self-attention model is established, a reduced-dimension data set is input into an ACGAN algorithm model with the self-attention model, a corresponding generator and a corresponding discriminator are obtained through training, and meanwhile, synthetic data generated through training are input into a first Softmax classifier to form a bearing fault evaluation model;
the traditional GAN is unsupervised learning, consisting of two networks, namely a generator and a arbiter. But its mode is too free to make the training process controllable.
The CGAN adds an auxiliary classification label c into the generator and the discriminator on the basis of GAN, and uses the auxiliary classification label c to guide the direction of data generation so as to realize supervised learning.
ACGAN is improved on the basis of CGAN, tag information only acts on the generator, and an auxiliary classifier C is introduced into the discriminator for discriminating the class of the sample. The generation direction of the sample can be controlled by the auxiliary classifier in the generation process so as to generate a high-quality result; however, due to the limited size of the convolution kernel, only the relationship of the local areas of the sample can be learned, the learning efficiency of the model is low, and details may be lost.
The ACGAN based on self-attention adds a self-attention mechanism into G and D on the basis of the ACGAN, so that the limitation that the ACGAN can only calculate data characteristics in a specific neighborhood is solved, and the model is helped to capture the relation between long-distance characteristics of samples.
Preferably, as shown in fig. 2, the method for establishing the self-attention model includes:
three channel nodes F (x), G (x) and H (x) are arranged, and different weight matrixes W are respectively arranged on different channel nodes K 、W Q And W is V Multiplying the input x by different weight matrices to obtain:
calculating an incidence matrix S between F (x) and G (x):
S＝F T ·G
calculating the degree of correlation between F (x) and G (x):
wherein: s is(s) ij Is an element of the incidence matrix S;
output features of the self-attention model are obtained:
adding a weight coefficient gamma, and integrating the field information and the remote features to obtain:
y i ＝γo i +x i 。
preferably, the method for training the reduced-dimension data set based on the self-attention ACGAN algorithm model is as follows:
in this context, the subscript "G" indicates generator-related parameters and the subscript "D" indicates arbiter-related parameters.
Define dimension reduction dataset x= { X 1 ,x 2 ,…,x N X, where x i For the ith sample, N is the total number of samples; the generator G generates realistic data close to the real distribution for realizing the real distribution P from the data set X r (x) Is learned to generate the distribution P g (z). The generator G is composed of a multi-layer neural network, and after the attention mechanism is added, a loss function of the generator G is set as follows:
adding an auxiliary classification label c into the generator, and guiding the data generation direction by using the auxiliary classification label c to obtain a loss function L representing whether the data is true or not S The method comprises the following steps:
the input is judged to be the actual quantity or the generated quantity by a discriminator D, the discriminator D is also formed by a multi-layer neural network, and a loss function L of the discriminator D is obtained after a self-attention mechanism is added D The method comprises the following steps:
setting an auxiliary classifier C for discriminating the class of the sample, and introducing the auxiliary classifier C into the discriminator D to obtain a loss function L for characterizing the classification accuracy of the data C The method comprises the following steps:
introducing Wo Sesi tame measurement (Wasserstein) to construct an objective function of overall training, measuring the difference between a generated sample and a real sample, constructing an ACGAN model based on self-attention, and generating a high-quality fault sample to be used for aeroengine bearing fault diagnosis. The objective function of the self-attention based ACGAN ensemble training is:
by establishing a self-attention model, the characteristics of important channels are enhanced by dynamically weighting and fusing the weights of the three channels, the characteristics of non-important channels are weakened, and the correlation of global information is realized, so that the calculated amount is reduced, and the model training speed is improved.
The generated samples are closer to actual fault samples based on the game theory, so that the problem that bearing fault samples are fewer in actual diagnosis is solved, and the number of the fault samples is effectively increased.
Preferably, for the gradient explosion or gradient vanishing problem, a gradient penalty term is constructed:
wherein: I.I p Represents the p-norm; lambda is a regularized term coefficient;wherein epsilon-U (0, 1), U is uniformly distributed.
Step S500, estimating lambda by using bearing fault estimation model mean The value is used for screening out the synthetic data which is judged to be effective, and the synthetic data is combined with the offline training set after the dimension reduction to form an enhanced data set lambda mean The value calculation formula is:
wherein a is AMC1 The proportion of the normal correct classification of the bearing to the total classification number is predicted; a, a AMC2 The proportion of the bearing fault classification to the total classification number is predicted; f (f) 11 The number of faults predicted as faulty bearings; f (f) 00 A number predicted to be normal for a normal bearing; f (f) 10 The number of bearings predicted to be normal for failure; f (f) 01 The number of faults predicted as normal bearings; lambda (lambda) mean Is a as AMC1 And a AMC2 Is a geometric average of (2);
preferably, as shown in fig. 3, the generating method of the enhanced data set includes:
1) Random noise is generated by random sampling in Gaussian distribution, the random noise is classified, and the random noise and an auxiliary classification label corresponding to the random noise are input into a generator to obtain generated data;
2) Inputting the generated data and the real data of the vibration characteristics of the bearing through acquisition into a discriminator to form mixed data, discriminating the mixed data to obtain real/false data and different fault type data, and completing one-time training;
3) And respectively and sequentially updating network parameters of the first Softmax classifier and the discriminator, acquiring the generated data and the real data again for cyclic discrimination, and inputting the synthetic data with effective discrimination into the enhanced data set until the set cycle number is completed or the data volume in the enhanced data set meets the design requirement.
Step S600, fusing the enhanced data set and the reduced data set to obtain an extended data set, and training the extended data set to obtain a second Softmax classifier for performing on-line fault diagnosis; the second Softmax classifier is internally provided with fault diagnosis standard parameters to carry out fault diagnosis on vibration data, and the accuracy of fault diagnosis can be continuously improved by continuously collecting and carrying out on-line diagnosis on the data in the vibration characteristic data set.
And S700, acquiring online data in real time, obtaining a data set by extracting vibration characteristics, performing PCA dimension reduction on the acquired data in a mode of S200 to form an online diagnosis set, inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and performing bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis result.
The method comprises the steps of firstly collecting vibration data of different types, establishing a vibration characteristic data set, then extracting vibration characteristic values of case library data again, inputting the vibration characteristic values into the data set to form an offline training set, performing dimension reduction processing on the offline data set by using a PCA algorithm to reduce calculated amount, and training the dimension reduction data set to obtain a first Softmax classifier; inputting the dimensionality reduction data set into an ACGAN algorithm model with a self-attention model, training to obtain a corresponding generator and a corresponding discriminator, simultaneously increasing the sample number, and evaluating lambda by using a bearing fault evaluation model mean The value and screening out the effective synthetic data, and combining the effective synthetic data with the offline training set after dimension reduction to form an enhanced data set; fusing the enhanced data set and the reduced data set to obtain an extended data set, and training the extended data set to obtain a second Softmax classifier for performing on-line fault diagnosis; inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and carrying out bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis junctionAnd (5) fruits. By introducing a self-attention mechanism, the failure characteristic distribution characteristics which are not existed in the original failure can be learned, and the generation quality and the learning efficiency of the bearing failure sample are improved; the established bearing fault online diagnosis model has the characteristics of high diagnosis accuracy, can accurately identify the bearing faults of the aero-engine, can accurately and rapidly judge on line, has higher identification accuracy and stability, and is suitable for on-site real-time diagnosis application.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present application should be included in the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (6)
1. An on-line diagnosis method for bearing faults of an aeroengine is characterized by comprising the following steps:
establishing a vibration characteristic data set, and acquiring vibration characteristic values of vibration data construction working conditions of the bearing under different operation working conditions; extracting time domain indexes of different vibration data of the bearing to obtain a time domain vibration characteristic value; extracting frequency domain indexes of different vibration data of the bearing to obtain frequency domain vibration characteristic values; inputting the working condition vibration characteristic value, the time domain vibration characteristic value and the frequency domain vibration characteristic value into a vibration characteristic data set;
based on the vibration characteristic data set, extracting vibration characteristic values of the case library data again, inputting the vibration characteristic data set to form an offline training set, performing dimension reduction processing on the offline data set by using a PCA algorithm, and reducing the dimension of data in the offline training set to m dimension to form a dimension reduction data set;
training the dimension reduction data set to obtain a first Softmax classifier;
establishing a self-attention model, inputting a reduced-dimension data set into an ACGAN algorithm model with the self-attention model, training to obtain a corresponding generator and a corresponding discriminator, and simultaneously inputting synthetic data generated by training into a first Softmax classifier to form a bearing fault evaluation model;
estimating lambda using bearing failure estimation model mean The value is used for screening out the synthetic data which is judged to be effective, and the synthetic data is combined with the offline training set after the dimension reduction to form an enhanced data set lambda mean The value calculation formula is:
wherein a is AMC1 The proportion of the normal correct classification of the bearing to the total classification number is predicted; a, a AMC2 The proportion of the bearing fault classification to the total classification number is predicted; f (f) 11 The number of faults predicted as faulty bearings; f (f) 00 A number predicted to be normal for a normal bearing; f (f) 10 The number of bearings predicted to be normal for failure; f (f) 01 The number of faults predicted as normal bearings; lambda (lambda) mean Is a as AMC1 And a AMC2 Is a geometric average of (2);
fusing the enhanced data set and the reduced data set to obtain an extended data set, and training the extended data set to obtain a second Softmax classifier for performing on-line fault diagnosis;
and acquiring online data in real time, obtaining a data set by extracting vibration characteristics, performing PCA dimension reduction on the acquired data to form an online diagnosis set, inputting the online training set into a second Softmax classifier to obtain a bearing fault diagnosis model, and performing bearing online diagnosis through the bearing fault diagnosis model to obtain a bearing online diagnosis result.
2. The on-line diagnostic method for an aircraft engine bearing failure of claim 1, wherein the generating method for the enhanced data set comprises:
random noise is generated by random sampling in Gaussian distribution, the random noise is classified, and the random noise and an auxiliary classification label corresponding to the random noise are input into a generator to obtain generated data;
inputting the generated data and the real data of the vibration characteristics of the bearing through acquisition into a discriminator to form mixed data, discriminating the mixed data to obtain real/false data and different fault type data, and completing one-time training;
and respectively and sequentially updating network parameters of the first Softmax classifier and the discriminator, acquiring the generated data and the real data again for cyclic discrimination, and inputting the synthetic data with effective discrimination into the enhanced data set until the set cycle number is completed or the data volume in the enhanced data set meets the design requirement.
3. The on-line diagnosis method for bearing faults of an aeroengine as claimed in claim 1, wherein: the working condition vibration characteristic values comprise an average value of sample input shaft rotating speeds, a variance of the sample input shaft rotating speeds, an average value of sample torques and a variance of the sample torques; the time domain vibration characteristic value comprises an acceleration effective value, an acceleration waveform index, an acceleration peak value index, an acceleration pulse index, an acceleration margin index, an acceleration kurtosis index, a speed effective value, a speed waveform index, a speed peak value index, a speed pulse index, a speed margin index and a speed kurtosis index; the frequency domain vibration characteristic value comprises the frequency values of the 1, 2 and 3-order amplitude values of the shaft rotation frequency where the bearing is located and the front 100 amplitude values in the frequency spectrum.
4. The on-line diagnosis method for bearing faults of an aeroengine as claimed in claim 1, wherein the method for establishing the self-attention model comprises the following steps:
three channel nodes F (x), G (x) and H (x) are arranged, and different weight matrixes W are respectively arranged on different channel nodes K 、W Q And W is V Multiplying the input x by different weight matrices to obtain:
calculating an incidence matrix S between F (x) and G (x):
S＝F T ·G
calculating the degree of correlation between F (x) and G (x):
wherein: s is(s) ij Is an element of the incidence matrix S;
output features of the self-attention model are obtained:
adding a weight coefficient gamma, and integrating the field information and the remote features to obtain:
y i ＝γo i +x i 。
5. the on-line diagnosis method for bearing faults of aeroengines according to claim 1, wherein the method for training a reduced-dimension data set by an ACGAN algorithm model is as follows:
define dimension reduction dataset x= { X 1 ,x 2 ,…,x N X, where x i For the ith sample, N is the total number of samples; the loss function of generator G is set to:
adding the auxiliary classification label c into a generator to obtain a loss function L for representing whether the data is real or not S The method comprises the following steps:
loss function L of discriminator D D The method comprises the following steps:
setting an auxiliary classifier C and introducing the auxiliary classifier C into a discriminator D to obtain a loss function L for representing the classification accuracy of the data C The method comprises the following steps:
introducing Wo Sesi metric, constructing an objective function of overall training, measuring the difference between a generated sample and a real sample, and constructing an ACGAN model based on self-attention, wherein the objective function of the ACGAN overall training based on self-attention is as follows:
6. the on-line diagnosis method for bearing faults of aero-engines according to claim 5, characterized in that, while establishing the objective function, a gradient penalty term is established:
wherein: I.I p Represents the p-norm; lambda is a regularized term coefficient;wherein epsilon-U (0, 1), U is uniformly distributed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310490570.2A CN116625686A (en) | 2023-05-04 | 2023-05-04 | On-line diagnosis method for bearing faults of aero-engine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310490570.2A CN116625686A (en) | 2023-05-04 | 2023-05-04 | On-line diagnosis method for bearing faults of aero-engine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116625686A true CN116625686A (en) | 2023-08-22 |
Family
ID=87616222
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310490570.2A Pending CN116625686A (en) | 2023-05-04 | 2023-05-04 | On-line diagnosis method for bearing faults of aero-engine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116625686A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117629637A (en) * | 2024-01-24 | 2024-03-01 | 哈尔滨师范大学 | Aeroengine bearing fault diagnosis method and diagnosis system |
| CN117629637B (en) * | 2024-01-24 | 2024-04-30 | 哈尔滨师范大学 | Aeroengine bearing fault diagnosis method and diagnosis system |
-
2023
- 2023-05-04 CN CN202310490570.2A patent/CN116625686A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117629637A (en) * | 2024-01-24 | 2024-03-01 | 哈尔滨师范大学 | Aeroengine bearing fault diagnosis method and diagnosis system |
| CN117629637B (en) * | 2024-01-24 | 2024-04-30 | 哈尔滨师范大学 | Aeroengine bearing fault diagnosis method and diagnosis system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108960303B (en) | Unmanned aerial vehicle flight data anomaly detection method based on LSTM | |
| CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| CN110443117B (en) | Wind turbine generator fault diagnosis method | |
| CN112163640A (en) | Planetary gearbox fault diagnosis method and system based on deep convolution migration learning | |
| CN113188794B (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
| CN115187832A (en) | Energy system fault diagnosis method based on deep learning and gram angular field image | |
| Zhang et al. | A novel data-driven method based on sample reliability assessment and improved CNN for machinery fault diagnosis with non-ideal data | |
| CN114619292A (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| Liang et al. | Multibranch and multiscale dynamic convolutional network for small sample fault diagnosis of rotating machinery | |
| CN114705432A (en) | Method and system for evaluating health state of explosion-proof motor bearing | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN115905976B (en) | Attention mechanism-based fault diagnosis method, system and equipment for Highway Bi-LSTM bearing | |
| CN115795397A (en) | Gearbox temperature prediction method based on 1DCNN-LSTM and BiLSTM parallel network | |
| CN116771610A (en) | Method for adjusting fault evaluation value of variable pitch system of wind turbine | |
| CN116625686A (en) | On-line diagnosis method for bearing faults of aero-engine | |
| CN114818116A (en) | Aircraft engine failure mode identification and service life prediction method based on joint learning | |
| CN110967184B (en) | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition | |
| Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
| CN112881017A (en) | Intelligent fault diagnosis method for aeroengine control system sensor based on mode gradient spectral entropy | |
| Qian et al. | A diagnosis method for diesel engine wear fault based on grey rough set and SOM neural network | |
| CN117688496B (en) | Abnormality diagnosis method, system and equipment for satellite telemetry multidimensional time sequence data | |
| CN114383846B (en) | Bearing composite fault diagnosis method based on fault label information vector |
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 |