CN115905853A - Aero-engine rotor system fault diagnosis method and device based on deep learning - Google Patents
Aero-engine rotor system fault diagnosis method and device based on deep learning Download PDFInfo
- Publication number
- CN115905853A CN115905853A CN202211079626.7A CN202211079626A CN115905853A CN 115905853 A CN115905853 A CN 115905853A CN 202211079626 A CN202211079626 A CN 202211079626A CN 115905853 A CN115905853 A CN 115905853A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- frequency domain
- feature extraction
- extraction network
- vibration signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000003745 diagnosis Methods 0.000 title claims abstract description 34
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 238000000605 extraction Methods 0.000 claims abstract description 103
- 230000000007 visual effect Effects 0.000 claims abstract description 20
- 238000004458 analytical method Methods 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 17
- 238000013526 transfer learning Methods 0.000 claims abstract description 9
- 238000010586 diagram Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 10
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 230000015654 memory Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 230000036541 health Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- 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
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a method and a device for diagnosing faults of an aircraft engine rotor system based on deep learning, and relates to the technical field of aircraft engine fault diagnosis. The method comprises the following steps: acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed; preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image; and inputting the two-dimensional time-frequency domain image into the trained feature extraction network to obtain a fault classification result of the aeroengine bearing to be diagnosed. According to the invention, the two-dimensional time-frequency image with physical significance is generated by performing time-frequency analysis on the vibration signal data of the one-dimensional bearing, the frequency domain characteristics of the data can be fully excavated, and the method has the advantages of high robustness and noise resistance. The feature extraction network based on the visual multi-layer perceptron can obtain high-dimensional fault features with global dependency, and further improves the accuracy of diagnosis. The transfer learning is helpful for overcoming the limitation of insufficient fault sample quantity and accelerating the practical application of the fault diagnosis algorithm.
Description
Technical Field
The invention relates to the technical field of aircraft engine fault diagnosis, in particular to a method and a device for diagnosing faults of an aircraft engine rotor system based on deep learning.
Background
The rotor system of the aircraft engine is an important component of the aircraft engine, the operation process of the rotor system is influenced by factors such as variable environment, variable load, variable working condition, large disturbance, strong impact and the like, the performance can also be inevitably degraded along with the increase of the operation time, and once the rotor system finally fails due to the degradation of the equipment performance, huge casualties and property loss can be caused. The rapid development of artificial intelligence and computer technology greatly improves the capability of health management of an aircraft engine rotor system. The fault diagnosis is a link and a key for connecting the information perception of the running state of the aircraft engine and realizing individual and accurate health management based on the running state, abnormality is timely found and diagnosis and prediction are carried out on the engine fault according to the monitoring information of the rotor system of the aircraft engine, accordingly, the health management is carried out on equipment, and the fault diagnosis method has important significance for practically guaranteeing the running safety, reliability and economy of the aircraft engine.
In recent years, with the development of deep learning, many researchers use deep learning algorithms with characteristics of high efficiency, strong generalization and the like in the fault diagnosis work of the rotor system of their aircraft engine. The fault diagnosis method based on deep learning is characterized in that the original data are input into a deep neural network model after being preprocessed, and then a fault diagnosis result is obtained. However, most of the existing intelligent fault diagnosis methods for the aircraft engine rotor system based on deep learning directly start from a training data set using a one-dimensional vibration signal of a bearing as a model, but information of the reaction of the one-dimensional vibration signal has limitations and is not enough to ensure the accuracy and robustness of diagnosis. Secondly, most of the existing fault diagnosis feature extraction networks based on deep learning are based on a cyclic neural network and a convolutional neural network. In addition, fault data used for training a deep neural network model are mostly small sample data, however, the traditional fault diagnosis method based on deep learning has a great limitation in processing a small sample data set.
Disclosure of Invention
The invention provides a fault diagnosis method aiming at the problems of low precision and poor robustness of the fault diagnosis method in the prior art.
In order to solve the technical problems, the invention provides the following technical scheme:
in one aspect, the invention provides a method for diagnosing faults of an aircraft engine rotor system based on deep learning, which is implemented by electronic equipment and comprises the following steps:
s1, acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed.
And S2, preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image.
And S3, inputting the two-dimensional time-frequency domain image into the trained feature extraction network.
And S4, obtaining a fault classification result of the aeroengine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network.
Optionally, the preprocessing the one-dimensional vibration signal data in S2 to obtain a two-dimensional time-frequency domain image includes:
and S21, performing missing data completion on the one-dimensional vibration signal data based on the generated countermeasure network to obtain complete one-dimensional vibration signal data.
And S22, performing time-frequency domain analysis on the complete one-dimensional vibration signal data based on Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data.
Optionally, the performing, in S22, time-frequency domain analysis on the complete one-dimensional vibration signal data based on hilbert yellow transform to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data includes:
and S221, decomposing the complete one-dimensional vibration signal data into a plurality of inherent modal components based on empirical mode decomposition.
S222, performing Hilbert transform on the plurality of natural modal components to obtain instantaneous frequency, instantaneous phase and instantaneous amplitude of each natural modal component in the plurality of natural modal components, and further obtaining a complete two-dimensional time-frequency domain image of the one-dimensional vibration signal data.
Optionally, the training process of the feature extraction network in S3 includes:
s31, acquiring a real sample data set of the aeroengine bearing; the real sample data set comprises a plurality of original one-dimensional vibration signal data with missing data.
And S32, performing missing data completion on the real sample data set based on the generated countermeasure network to obtain a complete sample data set.
And S33, performing time-frequency domain analysis on the complete sample data set through Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image sample data set.
And S34, acquiring a pre-trained feature extraction network.
And S35, obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network.
Optionally, the acquiring of the set of true sample data of the aircraft engine bearing in S31 comprises:
the method comprises the steps of obtaining original one-dimensional vibration signal data with missing data through a sensor arranged on an aircraft engine bearing, and obtaining a real sample data set of the aircraft engine bearing.
Optionally, the obtaining a pre-trained feature extraction network in S34 includes:
and S341, constructing a feature extraction network based on the visual multi-layer perceptron.
And S342, pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set to obtain the pre-trained feature extraction network.
Optionally, the pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set in S342, and obtaining the pre-trained feature extraction network includes:
s3421, acquiring a two-dimensional time-frequency domain image in the ImageNet data set.
S3422, inputting the two-dimensional time-frequency domain image into the two-dimensional convolution layer to obtain a characteristic diagram with the increased channel number.
And S3423, inputting the feature map into the full connection layer to obtain a reconstructed 2D image block.
And S3424, inputting the 2D image block into the image block mixed layer to obtain an image block mixed layer and outputting the image block mixed layer.
S3425, outputting and inputting the mixed layer of the image blocks to the mixed layer of the channel to obtain high-dimensional fault features with global dependency relationship, and further obtaining a pre-trained feature extraction network.
Optionally, the obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network in S35 includes:
and inputting the two-dimensional time-frequency domain image sample data set into a pre-trained feature extraction network for transfer learning, and adjusting the pre-trained feature extraction network to obtain a trained feature extraction network.
Optionally, the step of obtaining a fault classification result of the aero-engine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network in the step S4 includes:
and extracting a global average pooling layer, a full connection layer and a Softmax classifier of the network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aeroengine bearing to be diagnosed.
On the other hand, the invention provides a device for diagnosing faults of an aircraft engine rotor system based on deep learning, which is applied to a method for diagnosing faults of the aircraft engine rotor system based on the deep learning, and comprises the following steps:
the acquisition module is used for acquiring one-dimensional vibration signal data of the aeroengine bearing to be diagnosed.
And the data preprocessing module is used for preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image.
And the characteristic extraction module is used for inputting the two-dimensional time-frequency domain image into the trained characteristic extraction network.
And the fault classification module is used for extracting a network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aeroengine bearing to be diagnosed.
Optionally, the data preprocessing module is further configured to:
and S21, performing missing data completion on the one-dimensional vibration signal data based on the generated countermeasure network to obtain complete one-dimensional vibration signal data.
And S22, performing time-frequency domain analysis on the complete one-dimensional vibration signal data based on Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data.
Optionally, the data preprocessing module is further configured to:
and S221, decomposing the complete one-dimensional vibration signal data into a plurality of inherent modal components based on empirical mode decomposition.
S222, performing Hilbert transform on the plurality of natural modal components to obtain instantaneous frequency, instantaneous phase and instantaneous amplitude of each natural modal component in the plurality of natural modal components, and further obtaining a complete two-dimensional time-frequency domain image of the one-dimensional vibration signal data.
Optionally, the feature extraction module is further configured to:
s31, acquiring a real sample data set of the aeroengine bearing; wherein the real sample data set comprises a plurality of original one-dimensional vibration signal data with missing data.
And S32, performing missing data completion on the real sample data set based on the generated countermeasure network to obtain a complete sample data set.
And S33, performing time-frequency domain analysis on the complete sample data set through Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image sample data set.
And S34, acquiring a pre-trained feature extraction network.
And S35, obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network.
Optionally, the feature extraction module is further configured to:
the method comprises the steps of obtaining original one-dimensional vibration signal data with missing data through a sensor arranged on an aircraft engine bearing, and obtaining a real sample data set of the aircraft engine bearing.
Optionally, the feature extraction module is further configured to:
and S341, constructing a feature extraction network based on the visual multi-layer perceptron.
And S342, pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set to obtain the pre-trained feature extraction network.
Optionally, the feature extraction module is further configured to:
s3421, acquiring a two-dimensional time-frequency domain image in the ImageNet data set.
S3422, inputting the two-dimensional time-frequency domain image into the two-dimensional convolution layer to obtain the characteristic diagram with the increased channel number.
And S3423, inputting the feature map into the full connection layer to obtain a reconstructed 2D image block.
And S3424, inputting the 2D image block into the image block mixed layer to obtain an image block mixed layer and outputting the image block mixed layer.
S3425, outputting and inputting the mixed layer of the image blocks to the mixed layer of the channel to obtain high-dimensional fault features with global dependency relationship, and further obtaining a pre-trained feature extraction network.
Optionally, the feature extraction module is further configured to:
and inputting the two-dimensional time-frequency domain image sample data set into a pre-trained feature extraction network for transfer learning, and adjusting the pre-trained feature extraction network to obtain a trained feature extraction network.
Optionally, the fault classification module is further configured to:
and extracting a global average pooling layer, a full connection layer and a Softmax classifier of the network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aero-engine bearing to be diagnosed.
In one aspect, an electronic device is provided, and the electronic device includes a processor and a memory, where at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the method for diagnosing the fault of the rotor system of the aircraft engine based on deep learning.
In one aspect, a computer-readable storage medium is provided, and at least one instruction is stored in the storage medium and loaded and executed by a processor to implement the method for diagnosing faults of an aircraft engine rotor system based on deep learning.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
in the scheme, the fault diagnosis model is provided, the two-dimensional time-frequency image with physical significance is generated by performing time-frequency analysis on the one-dimensional bearing vibration signal data, the frequency domain characteristics of the data can be fully excavated, and the fault diagnosis model has the advantages of high robustness and noise resistance.
The invention uses the feature extraction network based on the visual multilayer perceptron to extract the fault features, and can obtain the high-dimensional fault features with global dependency relationship.
The limitation of insufficient fault sample amount is overcome by using the transfer learning technology, so that the practical application of the fault diagnosis algorithm is accelerated.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic flow chart of a deep learning-based aircraft engine rotor system fault diagnosis method provided by an embodiment of the invention;
fig. 2 is a schematic structural diagram of a time-series data missing value completion model based on a generative countermeasure network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an overall structure of a feature extraction network based on a visual multi-layer perceptron according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a mixed layer structure for an image block according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a channel mixing layer provided in an embodiment of the present invention;
FIG. 6 is a block diagram of a device for diagnosing faults of an aircraft engine rotor system based on deep learning, provided by an embodiment of the invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, an embodiment of the present invention provides a method for diagnosing a fault of an aircraft engine rotor system based on deep learning, where the method may be implemented by an electronic device. The method for diagnosing the faults of the aircraft engine rotor system based on deep learning is a flow chart shown in FIG. 1, and the processing flow of the method can comprise the following steps:
s1, acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed.
In one possible embodiment, the one-dimensional vibration signal data of the aircraft engine bearing to be diagnosed can be acquired by a sensor mounted on the aircraft engine bearing.
And S2, preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image.
Optionally, the preprocessing the one-dimensional vibration signal data in S2 to obtain a two-dimensional time-frequency domain image includes:
and S21, performing missing data completion on the one-dimensional vibration signal data based on the generated countermeasure network to obtain complete one-dimensional vibration signal data.
In a possible implementation manner, as shown in fig. 2, the structural diagram of the model based on the missing value complementation of the time series data of the generation countermeasure network is implemented as follows:
the generation countermeasure network is composed of a generator and an arbiter. The input of the generator is a low-dimensional random vector, and the output is a generated new sample; the input of the discriminator is a new sample generated by the generator or a real sample with missing data, and the output is a probability value representing the possibility that the input sample is a real sample. The final goal of the generator is to generate false new samples that can fool the discriminator, and the final goal of the discriminator is to try to tell if the input samples are true or false. The optimization formula for generating the countermeasure network can be written as the following formula (1):
wherein P is data (x) Representing the distribution of the original data set, P z (z) represents the distribution of new samples generated by the generator, z represents the random input vector of the generator, D (x) represents the output probability of the discriminator, i.e. the probability that the input sample is a true sample, and G (z) represents the output of the generator.
And S22, performing time-frequency domain analysis on the complete one-dimensional vibration signal data based on Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data.
Optionally, the performing, in S22, time-frequency domain analysis on the complete one-dimensional vibration signal data based on hilbert yellow transform to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data includes:
and S221, decomposing the complete one-dimensional vibration signal data into a plurality of inherent modal components based on empirical mode decomposition.
In one possible implementation, the hilbert yellow transform includes empirical mode decomposition and hilbert transform. Decomposition of bearing vibration signals into multiple intrinsic modal components IMF by empirical mode decomposition 1 ,IMF 2 ,…,IMF n-1 And a residual term r n Thus, the signal can be reconstructed as the following equation (2):
s222, performing Hilbert transform on the plurality of natural modal components to obtain instantaneous frequency, instantaneous phase and instantaneous amplitude of each natural modal component in the plurality of natural modal components, and further obtaining a complete two-dimensional time-frequency domain image of the one-dimensional vibration signal data.
In one possible implementation, the hilbert spectrum may be generated by combining the frequency spectra of the components. Each IMF i (t) Hilbert transform is as follows (3):
wherein i (t) is an IMF component; i (τ) is the IMF component; t is a time variable of i (t); τ is the time variable of i (τ).
Resolving the signal z i (t) is represented by the following formula (4):
z(t)＝i(t)+jH[i(t)]＝a(t)e jθ(t) (4)
wherein j is an imaginary unit, a (t) and θ (t) are as follows (5) (6):
where c (t) is the IMF component.
The instantaneous frequency is defined by the following formula (7):
the original signal y (t) is reconstructed as follows (8):
wherein Re is a real part, and the influence of residual terms is ignored during calculation.
And S3, inputting the two-dimensional time-frequency domain image into the trained feature extraction network.
Optionally, the training process of the feature extraction network in S3 includes:
s31, acquiring a real sample data set of the aeroengine bearing; wherein the real sample data set comprises a plurality of original one-dimensional vibration signal data with missing data.
Optionally, the acquiring of the set of true sample data of the aircraft engine bearing in S31 comprises:
the method comprises the steps of obtaining original one-dimensional vibration signal data with missing data through a sensor arranged on an aircraft engine bearing, and obtaining a real sample data set of the aircraft engine bearing.
In one possible embodiment, raw vibration signal data of the bearing is acquired by sensors mounted on the bearing of the aircraft engine.
And S32, performing missing data completion on the real sample data set based on the generated countermeasure network to obtain a complete sample data set.
And S33, performing time-frequency domain analysis on the complete sample data set through Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image sample data set.
And S34, acquiring a pre-trained feature extraction network.
Optionally, the obtaining a pre-trained feature extraction network in S34 includes:
and S341, constructing a feature extraction network based on the visual multi-layer perceptron.
And S342, pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set to obtain the pre-trained feature extraction network.
Optionally, the obtaining of the ImageNet-based data set in S342 and the pre-training of the feature extraction network based on the visual multi-layer perceptron, and the obtaining of the pre-trained feature extraction network includes:
s3421, acquiring a two-dimensional time-frequency domain image in the ImageNet data set.
S3422, inputting the two-dimensional time-frequency domain image into the two-dimensional convolution layer to obtain the characteristic diagram with the increased channel number.
In a possible implementation manner, the overall structure of the feature extraction network based on the visual multi-layer perceptron is shown in fig. 3. For a given two-dimensional time-frequency domain image used for training
Where (H, W) is the resolution of the original image. First, the number of channels of the input image is increased by a two-dimensional convolution, which can be described by the following equation (9):
U＝F Conv2D (X) (9)
wherein the content of the first and second substances,
c is the number of channels of the feature map U, F Conv2D Representing a two-dimensional convolution operation with a convolution kernel size of 1 x 1 and a step size of 1.
And S3423, inputting the feature map into the full connection layer to obtain a reconstructed 2D image block.
In one possible embodiment, the characteristic diagram U is fed into a fully connected layer. To process a 2D image, the fully-connected layers reconstruct the image U as a series of 2D image blocks Z = [ Z ] 1 ,z 2 ,…,z n ]In which
(P, P) resolution of each image block, N = HW/P 2 This is also the length of the sequence to be processed by the visual multi-layer perceptron for the number of image blocks obtained.
And S3424, inputting the 2D image block into the image block mixed layer to obtain an image block mixed layer and outputting the image block mixed layer.
In a possible embodiment, to capture the spatial dependency between different image blocks, the image block 2D image block Z = [ Z ] is further processed 1 ,z 2 ,…,z n ]An image patch mixed layer is introduced, and as shown in fig. 4, the image patch mixed layer has a schematic structure, which can be expressed by the following equation (10):
V＝Z+σ(LayerNorm(Z)W 1 )W 2 (10)
where σ denotes the GELU activation function, layerNorm is once normalized, W 1 And W 2 Representing the learned parameters of the mixed layer of the image block.
S3425, outputting and inputting the mixed layer of the image blocks to the mixed layer of the channel to obtain high-dimensional fault features with global dependency relationship, and further obtaining a pre-trained feature extraction network.
In a possible embodiment, in order to obtain the dependency relationship between different channels of each tile, the output of the tile mixture layer is further fed into a channel mixture layer, as shown in fig. 5, which can be represented by the following formula (11):
Y＝V+W 4 σ(W 3 LayerNorm(V)) (11)
wherein, W 3 And W 4 Representing the parameters learned by the channel mixing layer.
And S35, obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network.
Optionally, the obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network in S35 includes:
and inputting the two-dimensional time-frequency domain image sample data set into a pre-trained feature extraction network for transfer learning, and adjusting the pre-trained feature extraction network to obtain a trained feature extraction network.
And S4, obtaining a fault classification result of the aeroengine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network.
Optionally, the step of obtaining a fault classification result of the aero-engine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network in S4 includes:
and extracting a global average pooling layer, a full connection layer and a Softmax classifier of the network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aeroengine bearing to be diagnosed.
In the embodiment of the invention, the fault diagnosis model is provided, the two-dimensional time-frequency image with physical significance is generated by performing time-frequency analysis on the one-dimensional bearing vibration signal data, the frequency domain characteristics of the data can be fully excavated, and the fault diagnosis model has the advantages of high robustness and noise resistance.
The invention uses the feature extraction network based on the visual multilayer perceptron to extract the fault features, and can obtain the high-dimensional fault features with global dependency relationship.
The limitation of insufficient fault sample amount is overcome by using the transfer learning technology, so that the practical application of the fault diagnosis algorithm is accelerated.
As shown in fig. 6, an embodiment of the present invention provides an aircraft engine rotor system fault diagnosis apparatus 600 based on deep learning, where the apparatus 600 is applied to implement an aircraft engine rotor system fault diagnosis method based on deep learning, and the apparatus 600 includes:
the obtaining module 610 is configured to obtain one-dimensional vibration signal data of an aircraft engine bearing to be diagnosed.
And the data preprocessing module 620 is configured to preprocess the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image.
And the feature extraction module 630 is configured to input the two-dimensional time-frequency domain image to the trained feature extraction network.
And the fault classification module 640 is used for extracting a network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aero-engine bearing to be diagnosed.
Optionally, the data preprocessing module 620 is further configured to:
and S21, performing missing data completion on the one-dimensional vibration signal data based on the generated countermeasure network to obtain complete one-dimensional vibration signal data.
And S22, performing time-frequency domain analysis on the complete one-dimensional vibration signal data based on Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data.
Optionally, the data preprocessing module 620 is further configured to:
and S221, decomposing the complete one-dimensional vibration signal data into a plurality of inherent modal components based on empirical mode decomposition.
S222, performing Hilbert transform on the plurality of natural modal components to obtain instantaneous frequency, instantaneous phase and instantaneous amplitude of each natural modal component in the plurality of natural modal components, and further obtaining a complete two-dimensional time-frequency domain image of the one-dimensional vibration signal data.
Optionally, the feature extraction module 630 is further configured to:
s31, acquiring a real sample data set of the aeroengine bearing; the real sample data set comprises a plurality of original one-dimensional vibration signal data with missing data.
And S32, performing missing data completion on the real sample data set based on the generated countermeasure network to obtain a complete sample data set.
And S33, performing time-frequency domain analysis on the complete sample data set through Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image sample data set.
And S34, acquiring a pre-trained feature extraction network.
And S35, obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network.
Optionally, the feature extraction module 630 is further configured to:
the method comprises the steps of obtaining original one-dimensional vibration signal data with missing data through a sensor arranged on an aeroengine bearing, and obtaining a real sample data set of the aeroengine bearing.
Optionally, the feature extraction module 630 is further configured to:
and S341, constructing a feature extraction network based on the visual multi-layer perceptron.
And S342, pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set to obtain the pre-trained feature extraction network.
Optionally, the feature extraction module 630 is further configured to:
s3421, acquiring a two-dimensional time-frequency domain image in the ImageNet data set.
S3422, inputting the two-dimensional time-frequency domain image into the two-dimensional convolution layer to obtain a characteristic diagram with the increased channel number.
And S3423, inputting the feature map into the full connection layer to obtain a reconstructed 2D image block.
And S3424, inputting the 2D image block into the image block mixed layer to obtain an image block mixed layer output.
S3425, outputting and inputting the mixed layer of the image blocks to the mixed layer of the channel to obtain high-dimensional fault features with global dependency relationship, and further obtaining a pre-trained feature extraction network.
Optionally, the feature extraction module 630 is further configured to:
and inputting the two-dimensional time-frequency domain image sample data set into a pre-trained feature extraction network for transfer learning, and adjusting the pre-trained feature extraction network to obtain a trained feature extraction network.
Optionally, the fault classification module 640 is further configured to:
and extracting a global average pooling layer, a full connection layer and a Softmax classifier of the network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aero-engine bearing to be diagnosed.
In the embodiment of the invention, the fault diagnosis model is provided, the two-dimensional time-frequency image with physical significance is generated by performing time-frequency analysis on the one-dimensional bearing vibration signal data, the frequency domain characteristics of the data can be fully excavated, and the fault diagnosis model has the advantages of high robustness and noise resistance.
The invention uses the feature extraction network based on the visual multilayer perceptron to extract the fault features, and can obtain the high-dimensional fault features with global dependency relationship.
The limitation of insufficient fault sample amount is overcome by using the transfer learning technology, so that the practical application of the fault diagnosis algorithm is accelerated.
Fig. 7 is a schematic structural diagram of an electronic device 300 according to an embodiment of the present invention, where the electronic device 300 may generate a relatively large difference due to different configurations or different performances, and may include one or more processors (CPUs) 301 and one or more memories 302, where at least one instruction is stored in the memory 302, and the at least one instruction is loaded and executed by the processor 301 to implement the following method for diagnosing a fault of an aircraft engine rotor system based on deep learning:
s1, acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed.
And S2, preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image.
And S3, inputting the two-dimensional time-frequency domain image into the trained feature extraction network.
And S4, obtaining a fault classification result of the aeroengine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network.
In an exemplary embodiment, a computer-readable storage medium, such as a memory, is also provided that includes instructions executable by a processor in a terminal to perform the above-described deep learning-based aircraft engine rotor system fault diagnosis method. For example, the computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. A method for diagnosing faults of an aircraft engine rotor system based on deep learning is characterized by comprising the following steps:
s1, acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed;
s2, preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image;
s3, inputting the two-dimensional time-frequency domain image into a trained feature extraction network;
and S4, obtaining a fault classification result of the aeroengine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network.
2. The method of claim 1, wherein the preprocessing the one-dimensional vibration signal data in S2 to obtain a two-dimensional time-frequency domain image comprises:
s21, performing missing data completion on the one-dimensional vibration signal data based on a generated countermeasure network to obtain complete one-dimensional vibration signal data;
and S22, performing time-frequency domain analysis on the complete one-dimensional vibration signal data based on Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data.
3. The method according to claim 2, wherein the performing, in S22, a time-frequency domain analysis on the complete one-dimensional vibration signal data based on the hilbert yellow transform to obtain a two-dimensional time-frequency domain image of the complete one-dimensional vibration signal data comprises:
s221, decomposing the complete one-dimensional vibration signal data into a plurality of inherent modal components based on empirical mode decomposition;
s222, performing Hilbert transform on the plurality of natural modal components to obtain instantaneous frequency, instantaneous phase and instantaneous amplitude of each natural modal component in the plurality of natural modal components, and further obtaining a complete two-dimensional time-frequency domain image of one-dimensional vibration signal data.
4. The method according to claim 1, wherein the training process of the feature extraction network in S3 comprises:
s31, acquiring a real sample data set of the aeroengine bearing; wherein the real sample data set comprises a plurality of original one-dimensional vibration signal data with missing data;
s32, performing missing data completion on the real sample data set based on the generated countermeasure network to obtain a complete sample data set;
s33, performing time-frequency domain analysis on the complete sample data set through Hilbert-Huang transformation to obtain a two-dimensional time-frequency domain image sample data set;
s34, acquiring a pre-trained feature extraction network;
s35, obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and the pre-trained feature extraction network.
5. The method of claim 4, wherein the acquiring of the set of true sample data of the aircraft engine bearing in S31 comprises:
the method comprises the steps of obtaining original one-dimensional vibration signal data with missing data through a sensor arranged on an aircraft engine bearing, and obtaining a real sample data set of the aircraft engine bearing.
6. The method of claim 4, wherein the obtaining the pre-trained feature extraction network in S34 comprises:
s341, constructing a feature extraction network based on a visual multilayer perceptron;
and S342, pre-training the feature extraction network based on the visual multi-layer perceptron based on the ImageNet data set to obtain the pre-trained feature extraction network.
7. The method according to claim 6, wherein the pre-training the visual multi-layered perceptron-based feature extraction network based on ImageNet data set in S342, and obtaining the pre-trained feature extraction network comprises:
s3421, acquiring a two-dimensional time-frequency domain image in the ImageNet data set;
s3422, inputting the two-dimensional time-frequency domain image into a two-dimensional convolution layer to obtain a characteristic diagram with the increased channel number;
s3423, inputting the feature map into a full connection layer to obtain a reconstructed 2D image block;
s3424, inputting the 2D image block into an image block mixed layer to obtain an image block mixed layer for outputting;
s3425, outputting the image block mixed layer to a channel mixed layer to obtain high-dimensional fault features with global dependency relationship, and further obtaining a pre-trained feature extraction network.
8. The method according to claim 4, wherein the obtaining a trained feature extraction network according to the two-dimensional time-frequency domain image sample data set and a pre-trained feature extraction network in S35 comprises:
and inputting the two-dimensional time-frequency domain image sample data set into a pre-trained feature extraction network for transfer learning, and adjusting the pre-trained feature extraction network to obtain a trained feature extraction network.
9. The method according to claim 1, wherein the step S4 of obtaining a fault classification result of the aero-engine bearing to be diagnosed according to the two-dimensional time-frequency domain image and the trained feature extraction network comprises:
and extracting a global average pooling layer, a full connection layer and a Softmax classifier of the network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aeroengine bearing to be diagnosed.
10. An aircraft engine rotor system fault diagnosis device based on deep learning, characterized in that the device comprises:
the system comprises an acquisition module, a diagnosis module and a diagnosis module, wherein the acquisition module is used for acquiring one-dimensional vibration signal data of an aeroengine bearing to be diagnosed;
the data preprocessing module is used for preprocessing the one-dimensional vibration signal data to obtain a two-dimensional time-frequency domain image;
the feature extraction module is used for inputting the two-dimensional time-frequency domain image into a trained feature extraction network;
and the fault classification module is used for extracting a network according to the two-dimensional time-frequency domain image and the trained features to obtain a fault classification result of the aeroengine bearing to be diagnosed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211079626.7A CN115905853A (en) | 2022-09-05 | 2022-09-05 | Aero-engine rotor system fault diagnosis method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211079626.7A CN115905853A (en) | 2022-09-05 | 2022-09-05 | Aero-engine rotor system fault diagnosis method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115905853A true CN115905853A (en) | 2023-04-04 |
Family
ID=86492210
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211079626.7A Pending CN115905853A (en) | 2022-09-05 | 2022-09-05 | Aero-engine rotor system fault diagnosis method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115905853A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114065919A (en) * | 2021-11-17 | 2022-02-18 | 南京晨光集团有限责任公司 | Deficiency value completion method and medium based on generation countermeasure network |
| CN114912481A (en) * | 2022-04-20 | 2022-08-16 | 上海电机学院 | Motor bearing fault diagnosis method based on multiple time-frequency analysis self-adaptive fusion |
-
2022
- 2022-09-05 CN CN202211079626.7A patent/CN115905853A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114065919A (en) * | 2021-11-17 | 2022-02-18 | 南京晨光集团有限责任公司 | Deficiency value completion method and medium based on generation countermeasure network |
| CN114912481A (en) * | 2022-04-20 | 2022-08-16 | 上海电机学院 | Motor bearing fault diagnosis method based on multiple time-frequency analysis self-adaptive fusion |
Non-Patent Citations (3)
| Title |
|---|
| ILYA TOLSTIKHIN ET AL.: "MLP-Mixer: An all-MLP Architecture for Vision", 《ARXIV》, 11 June 2021 (2021-06-11), pages 1 - 16 * |
| MOU-FA GUO ET AL.: "Deep-Learning-Based Fault Classification Using Hilbert–Huang Transform and Convolutional Neural Network in Power Distribution Systems", 《IEEE SENSORS JOURNAL》, 24 April 2019 (2019-04-24), pages 6905 - 6913 * |
| 周浩敏等: "《测试信号处理技术 第2版》", 31 May 2009, 北京航空航天大学出版社, pages: 561 - 566 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shao et al. | DCNN-based multi-signal induction motor fault diagnosis | |
| Ding et al. | A novel time–frequency Transformer based on self–attention mechanism and its application in fault diagnosis of rolling bearings | |
| Han et al. | An adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems | |
| Chen et al. | Intelligent fault diagnosis for rotary machinery using transferable convolutional neural network | |
| Zhu et al. | Estimation of bearing remaining useful life based on multiscale convolutional neural network | |
| Tao et al. | Bearing fault diagnosis method based on stacked autoencoder and softmax regression | |
| Cheng et al. | Autoencoder quasi-recurrent neural networks for remaining useful life prediction of engineering systems | |
| CN110705525A (en) | Method and device for diagnosing rolling bearing fault | |
| CN111538947B (en) | Method for constructing wind power generator bearing fault classification model | |
| CN110334764A (en) | Rotating machinery intelligent failure diagnosis method based on integrated depth self-encoding encoder | |
| CN111562612B (en) | Deep learning microseismic event identification method and system based on attention mechanism | |
| Wang et al. | Gear fault diagnosis based on variational modal decomposition and wide+ narrow visual field neural networks | |
| CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| Jiao et al. | Partly interpretable transformer through binary arborescent filter for intelligent bearing fault diagnosis | |
| Shahid et al. | Performance comparison of 1D and 2D convolutional neural networks for real-time classification of time series sensor data | |
| CN115114965A (en) | Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN111539381B (en) | Construction method of wind turbine bearing fault classification diagnosis model | |
| Zurita et al. | A review of vibration machine diagnostics by using artificial intelligence methods | |
| CN115905853A (en) | Aero-engine rotor system fault diagnosis method and device based on deep learning | |
| Zhou et al. | Bearing fault diagnosis based on RF-PCA-LSTM model | |
| CN116610990A (en) | Method and device for identifying hidden danger of breaker based on characteristic space differentiation | |
| CN115902620A (en) | Motor fault enhancement diagnosis method and device based on loop generation countermeasure network | |
| CN115356599A (en) | Multi-mode urban power grid fault diagnosis method and system | |
| CN114781427A (en) | Wind generating set rolling bearing fault diagnosis system based on information fusion technology |
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 |