CN112329626A - Modulation and deep learning fused equipment fault diagnosis method, system and medium - Google Patents
Modulation and deep learning fused equipment fault diagnosis method, system and medium Download PDFInfo
- Publication number
- CN112329626A CN112329626A CN202011223732.9A CN202011223732A CN112329626A CN 112329626 A CN112329626 A CN 112329626A CN 202011223732 A CN202011223732 A CN 202011223732A CN 112329626 A CN112329626 A CN 112329626A
- Authority
- CN
- China
- Prior art keywords
- frequency
- fault diagnosis
- time
- signal
- neural network
- 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 67
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000013135 deep learning Methods 0.000 title claims abstract description 13
- 238000010586 diagram Methods 0.000 claims abstract description 63
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 11
- 238000001228 spectrum Methods 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 239000002245 particle Substances 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 5
- 238000012360 testing method Methods 0.000 description 13
- 238000012545 processing Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 238000011176 pooling Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000005299 abrasion Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 208000037408 Device failure Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
- G06F2218/14—Classification; Matching by matching peak patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a modulation and deep learning fused equipment fault diagnosis method, system and medium, which comprises the following steps: decomposing a vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component, and drawing a time-frequency graph; preprocessing the time-frequency diagram to be used as the input of a convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training; the convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network. The method can well extract the characteristics in the time-frequency diagram, effectively improve the fault diagnosis accuracy, and can be widely applied to the technical field of gas turbine fault diagnosis.
Description
Technical Field
The invention relates to the technical field of fault diagnosis of gas turbines, in particular to a fault diagnosis method of a gas turbine integrating modulation and deep learning.
Background
The gas turbine is widely applied to the fields of ships, special vehicles, power generation equipment and the like as power equipment. The gas turbine has an extremely complicated structure, and the gas turbine rotor system is subjected to high temperature and high pressure and also to strong stress and abrasion, so that the gas turbine rotor system becomes a failure-prone part. When the rotor of the gas turbine runs, the working condition is complex, the frequency conversion change is fast, and the vibration signal of the gas turbine often shows a non-stable frequency modulation characteristic. The time-frequency analysis method such as short-time Fourier transform and wavelet analysis has the problem that the time-frequency expression is not clear and visual enough. The recently developed signal decomposition methods such as EMD and VMD have problems of mode aliasing and poor expression in the time-frequency domain.
The Convolutional Neural Network (CNN) is a typical deep learning method, and automatically extracts deep features of an image through continuous Convolutional layers and pooling layers, so that the method has good image recognition and classification effects, and is widely applied to the field of image recognition. Although many researchers have combined CNN and fault diagnosis, the convolutional neural network has achieved some results in the field of fault diagnosis, the corresponding convolutional neural network-based fault diagnosis model is still lacking for the vibration signals of the gas turbine, and there is still room for further optimization of the time-frequency diagram as input.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a modulation and deep learning integrated equipment fault diagnosis method, system and medium, which can well extract features in a time-frequency diagram and effectively improve fault diagnosis accuracy.
In order to achieve the purpose, the invention adopts the following technical scheme: a modulation and deep learning fused equipment fault diagnosis method comprises the following steps: 1) decomposing a vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component, and drawing a time-frequency graph; 2) preprocessing the time-frequency diagram to be used as the input of a convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training; 3) the convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network.
Further, in the step 1), the vibration signal is decomposed into a plurality of frequency components by using a nonlinear frequency modulation component decomposition method.
Further, the nonlinear frequency modulation component decomposition method comprises the following steps:
1.1) expressing a vibration signal to be decomposed by a polynomial phase signal, constructing a demodulation operator by adopting an estimated demodulation parameter, demodulating the vibration signal to be decomposed into a stable signal, and enabling the demodulated signal to have the most concentrated frequency band expression;
1.2) extracting the corresponding most concentrated frequency band component through a band-pass filter, and performing inverse demodulation on the filtered signal through an inverse transformation operator to obtain a meaningful frequency component;
1.3) repeatedly executing the step 1.2), gradually extracting all frequency components from the original vibration signal until all components are decomposed.
Further, the polynomial phase signal model for the non-chirp signal is defined as:
wherein a (t) is the instantaneous amplitude of the signal, ci(i-0, 1 …, k) is the phase parameter, k is the order,is the initial phase of the signal, t represents time, g (t) represents the polynomial phase signal, ti+1Representing the (i + 1) th order time instant, j representing the imaginary part.
Further, using the estimated demodulation parameters to define the demodulation operator Φ-(t, C) and inverse transform operator Φ+(t, C) are respectively:
wherein the content of the first and second substances,for each of the stages of the demodulation parameters,indicating the demodulation parameters of order k.
gd(t,C)=a(t)exp(j(2πc0t+φ0));
then g isd(t, C) has the most concentrated spectrum, and the signal energy will be concentrated at frequency C0Nearby.
Further, in the step 1.2), defining a spectrum concentration index sci (c):
SCI(C)=E(|F(gd(t,C))|4)
wherein E (-) is the expectation and F (-) is the Fourier transform;
Using the maximum peak value of the frequency spectrum of the demodulation signal as the initial estimation parameterAnd adopts a particle swarm optimization method toAnd estimating parameters for optimizing.
A modulation and deep learning fused device fault diagnosis system comprising: the system comprises a time-frequency diagram drawing module, a fault diagnosis model obtaining module and a fault diagnosis module;
the time-frequency diagram drawing module is used for decomposing the vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component and drawing a time-frequency diagram;
the fault diagnosis model acquisition module is used for preprocessing the time-frequency diagram to be used as the input of the convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training;
the fault diagnosis module convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network.
A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the above methods.
A computing device, comprising: one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the above-described methods.
Due to the adoption of the technical scheme, the invention has the following advantages: 1. the time-frequency diagram obtained by the nonlinear frequency modulation component decomposition method has better time-frequency expression capability than short-time Fourier transform and wavelet time-frequency diagrams. 2. The convolutional neural network fault diagnosis model based on the nonlinear frequency modulation component decomposition time-frequency diagram can well diagnose the gas turbine rotor system.
Drawings
Fig. 1 is a schematic diagram of a typical structure of a CNN model.
Fig. 2 is a time domain diagram of a vibration signal in an embodiment.
Fig. 3 is a time domain diagram and a magnitude-frequency diagram of each component in the embodiment.
Fig. 4 is a time-frequency diagram of a vibration signal in the embodiment.
Fig. 5 is a time-frequency diagram of the improved vibration signal in the embodiment.
Fig. 6 is a CNN network structure in the embodiment.
FIG. 7 is a schematic view of a gas turbine rotor system configuration.
Fig. 8 is a time-frequency diagram after the gray processing in three states.
FIG. 9 fault diagnosis test results.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the drawings of the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the invention, are within the scope of the invention.
In a first embodiment of the invention, a modulation and deep learning fused gas turbine fault diagnosis method is provided, which uses a nonlinear frequency modulation component decomposition method to perform time-frequency analysis on a gas turbine casing vibration signal and uses a convolutional neural network to perform fault diagnosis on a gas turbine rotor, wherein the gas turbine fault signal is a variable speed non-stationary signal. Specifically, the method comprises the following steps:
1) decomposing a vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component, and drawing a time-frequency graph;
in the present embodiment, the vibration signal is decomposed into a plurality of frequency components by a nonlinear frequency modulation component decomposition method.
2) Preprocessing the time-frequency diagram to be used as the input of a convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training;
in the present embodiment, the preprocessing includes gradation processing, size compression, and the like.
3) The convolutional neural network fault diagnosis model realizes effective diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network, can obtain a good diagnosis effect on the gas turbine rotor fault, and has the test accuracy rate of about 99%.
In the step 1), the nonlinear frequency modulation component decomposition method includes the following steps:
1.1) expressing the vibration signal to be decomposed by a polynomial phase signal, adopting the estimated demodulation parameters to construct a demodulation operator, demodulating the vibration signal to be decomposed into a stable signal, and enabling the demodulated signal to have the most concentrated frequency band expression.
1.2) extracting the corresponding most concentrated frequency band component through a band-pass filter, and performing inverse demodulation on the filtered signal through an inverse transformation operator to obtain a meaningful frequency component;
1.3) repeatedly executing the step 1.2), gradually extracting all frequency components from the original vibration signal until all components are decomposed.
In the step 1.1), the polynomial phase signal model for the non-linear frequency modulation signal is defined as:
where a (t) is the instantaneous amplitude of the signal,representing the instantaneous frequency, ci(i-0, 1 …, k) is the phase parameter, k is the order,is the initial phase of the signal. t represents time, g (t) represents polynomial phase signal, ti+1Representing the (i + 1) th order time instant, j representing the imaginary part.
Using estimated demodulation parameters to define the demodulation operator phi-(t, C) and inverse transform operator Φ+(t, C) are respectively:
wherein the content of the first and second substances,the demodulation parameters of each order are rough estimation values.Indicating the demodulation parameters of order k.
Then the demodulation function gd(t, C) is:
gd(t,C)=g(t)Φ-(t,C)。 (4)
gd(t,C)=a(t)exp(j(2πc0t+φ0))。 (5)
at this time, gd(t, C) has the most concentrated spectrum, and the signal energy will be concentrated at frequency C0Nearby.
In the step 1.2), the spectrum concentration index sci (c) is defined:
SCI(C)=E(|F(gd(t,C))|4) (6)
where E (-) is the expectation and F (-) is the Fourier transform.
Using the maximum peak value of the frequency spectrum of the demodulation signal as the initial estimation parameterAnd optimizing the estimation parameters by adopting a particle swarm optimization method.
All phase parameters (c) of the instantaneous frequency can be estimated by equations (6) and (7)0…,ck) The instantaneous frequency f (t) of the signal is the first derivative of the phase parameter,the time-frequency diagram of the signal can be drawn after the instantaneous frequency of the signal is obtained.
In order to facilitate observation of the amplitude and the energy ratio of each frequency component, the line width of each frequency component is defined as the integer of the product of the energy ratio of each component and the line width factor, the change and the energy ratio of each frequency component can be clearly shown in a time-frequency diagram, and the line width factor is set to be 5-10 in order to avoid too wide or too narrow line width.
In the step 2), the Convolutional Neural Network (CNN) is a deep learning Convolutional Neural Network, which is a typical deep Network, and reduces the number of weights through weight sharing, thereby improving the efficiency of model training. The network has the advantages that the network model can be trained by taking the image as input, and the feature extraction process is omitted. The convolutional neural network consists of input layers, alternating convolutional and pooling layers, and fully-connected layers and output layers, as shown in fig. 1. The convolutional neural network automatically extracts deep features of the image through continuous convolutional layers and pooling layers.
The convolutional layer realizes feature extraction through convolution calculation, and obtains a feature image from an input image. Assuming that the gray value of the point (x, y) in the input image P with size M × N is f (x, y), and the convolution kernel with size a × b is K (x, y), the convolution calculation can be expressed as:
wherein C (s, t) is a convolution operation matrix of the image P and the convolution kernel K, s is more than or equal to 1 and less than or equal to M-a + 1, and t is more than or equal to 1 and less than or equal to N-b + 1. The convolution operation is to form a new image matrix by continuously moving a convolution kernel on the image matrix and performing convolution operation with the corresponding area. The convolution layer mainly has the function of feature extraction, and can enhance signal features and reduce noise through convolution operation.
The pooling layer compresses images obtained by the convolution layer through down-sampling calculation, reduces data dimensionality, reduces calculated amount, and avoids overfitting. The purpose of the fully connected layer is to integrate features into a classification role. The average square error is usually used as a loss function of the output layer, and the parameters of the whole network are updated by adopting an inverse error propagation algorithm of the minimum mean square error principle. The loss function is as follows:
where loss is the mean squared error of forward propagation,Xirespectively, the output layer output and the desired output.
In a second embodiment of the present invention, there is provided a modulation and deep learning fused device failure diagnosis system, including: the system comprises a time-frequency diagram drawing module, a fault diagnosis model obtaining module and a fault diagnosis module;
the time-frequency diagram drawing module decomposes the vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component, and draws a time-frequency diagram;
the fault diagnosis model acquisition module is used for preprocessing the time-frequency diagram to be used as the input of the convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training;
the fault diagnosis module convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network.
In a third embodiment of the invention, there is provided a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of the first embodiment described above.
In a fourth embodiment of the present invention, there is provided a computing device comprising: one or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of the first embodiment described above.
Example (b):
the gas turbine rotor is easy to cause faults due to the reasons of installation and positioning accuracy, abrasion in operation and the like, so that the vibration is overlarge, the vibration signal of the gas turbine rotor system contains a lot of state characteristic information in the operation process of the gas turbine, and the operation state of the gas turbine can be found by analyzing the vibration signal.
The method includes the steps that vibration speed signals in the operation process of a certain type of gas turbine are measured through experiments, a section with overlarge vibration is intercepted and analyzed, a time domain diagram of the intercepted signals is shown in figure 2, nonlinear frequency modulation component decomposition is conducted on the signals to obtain 5 components, the time domain diagram and a frequency domain diagram of the 5 components are respectively drawn and shown in figure 3, frequency components with physical significance in the signals can be clearly extracted through the nonlinear frequency modulation component decomposition method, and the problem of mode aliasing is well solved.
The time-frequency diagram drawn by the nonlinear frequency modulation component decomposition method is shown in fig. 4, each line in the time-frequency diagram represents a frequency component, the change of the frequency component can be clearly observed through the time-frequency diagram, and the time-frequency diagram obtained by the method can be found to be clear and visual.
The time-frequency diagram obtained by the nonlinear frequency modulation component decomposition method can obtain a clear and visual time-frequency diagram. However, the amplitude and the energy ratio of each frequency component cannot be observed, so that on the basis of a time-frequency diagram drawn by a nonlinear frequency modulation component decomposition method, the integral of the product of the energy ratio of each component and a line width factor is used as the line width of each frequency component, so that the change and the energy ratio of each frequency component can be clearly observed from the time-frequency diagram, and in order to avoid the line width being too wide or too narrow, the line width factor is set to be 5-10. The improved time-frequency diagram is shown in fig. 5.
And (3) constructing a CNN network for input based on a time-frequency diagram generated by the nonlinear frequency modulation component decomposition method, and converting the time-frequency diagram generated by the nonlinear frequency modulation component decomposition method into a gray-scale diagram with the size of 28 multiplied by 28 pixels in order to reduce the calculated amount. The constructed CNN network structure is shown in fig. 6, and includes an input layer with an image size of 28 × 28, 2 convolutional layers with a convolutional kernel of 5 × 5, 2 pooling layers with a maximum value of 2 × 2, 1 full-link layer, and 1 output layer.
The method comprises the following steps of constructing a non-linear frequency modulation component decomposition time-frequency diagram-convolution neural network fault diagnosis model:
1) carrying out nonlinear frequency modulation component decomposition on the data sample set to obtain a time-frequency diagram;
2) carrying out gray level processing on the obtained time-frequency image to obtain a gray level image, converting the size of the gray level image into a 28 x 28 pixel image, carrying out range normalization on the obtained image, and normalizing the elements to be 0-1;
3) and taking the normalized picture set as input to train a convolutional neural network, and carrying out fault diagnosis by using the trained model.
The embodiments are verified, the principle of the rotor structure of a certain type of dual-rotor gas turbine is shown in fig. 7, and during vibration testing, a speed sensor is respectively installed at the radial positions of the casings of the low-pressure compressor and the high-pressure compressor. In the experiment, the rotating speed of the high-pressure rotor is controlled between 6000r/min and 7800r/min, the sampling frequency is 6000Hz, and the sampling time lasts 8000 s.
In the experiment, the vibration of the low-pressure compressor casing is obvious, so that the data measured by the low-pressure compressor casing is used as a data sample for fault diagnosis. Through later analysis, the gas turbine generates airflow shock excitation fault in the experimental process, and further causes the unbalance fault of the rotor. The obtained vibration data are divided into three types, namely a normal state, an air flow excitation state and an unbalanced state, the signal length of each type of state is 1024 points, 500 training samples and 100 testing samples are selected for each type of state during the test, and the data sample set is shown in table 1.
TABLE 1 data sample set
And decomposing the data sample set by the nonlinear frequency modulation component to obtain a time-frequency image, carrying out gray level processing on the obtained time-frequency image subjected to the nonlinear frequency modulation component decomposition, and converting the size of the obtained gray level image into 28 multiplied by 28 pixels, wherein typical gray level images in three states are shown in fig. 8.
And normalizing the obtained picture set to be used as a convolutional neural network constructed by input training, wherein the iteration times are set to be 50 times, and the learning rate is 0.1. The trained convolutional neural network fault diagnosis model is tested by using a test set, the result is shown in fig. 9, the test finds that a very good fault diagnosis effect can be achieved for the CNN network based on the nonlinear frequency modulation component decomposition time-frequency diagram, only three samples are wrongly divided, and the test accuracy is about 99%.
In order to verify the superiority of the proposed method, a short-time Fourier time-frequency graph and a wavelet transform time-frequency graph drawn by the same data set are respectively used as input training CNN networks, test results of a test set are shown in Table 2, and the CNN networks trained on the nonlinear frequency modulation component decomposition time-frequency graphs have the highest test accuracy. The time-frequency diagram obtained by the nonlinear frequency modulation component decomposition method has better time-frequency expression capability, and the fault model based on the convolutional neural network can well extract the characteristics in the time-frequency diagram to realize fault diagnosis.
TABLE 2 test accuracy for different diagnostic models
In summary, the invention uses a nonlinear frequency modulation component decomposition method to process signals and construct a time-frequency diagram aiming at the problem of the fault of the gas turbine rotor system, and uses the time-frequency diagram as an input training model of a convolutional neural network, and verifies the effectiveness of the model through testing. The result shows that the time-frequency diagram obtained by the nonlinear frequency modulation component decomposition method has better time-frequency expression capability than the short-time Fourier transform and the wavelet time-frequency diagram; the convolutional neural network fault diagnosis model based on the nonlinear frequency modulation component decomposition time-frequency diagram can well diagnose the gas turbine rotor system.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Claims (10)
1. A modulation and deep learning fused equipment fault diagnosis method is characterized by comprising the following steps:
1) decomposing a vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component, and drawing a time-frequency graph;
2) preprocessing the time-frequency diagram to be used as the input of a convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training;
3) the convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network.
2. The fault diagnosis method according to claim 1, wherein in step 1), the vibration signal is decomposed into a plurality of frequency components by a nonlinear frequency modulation component decomposition method.
3. The fault diagnosis method according to claim 2, wherein the nonlinear chirp component decomposition method comprises the steps of:
1.1) expressing a vibration signal to be decomposed by a polynomial phase signal, constructing a demodulation operator by adopting an estimated demodulation parameter, demodulating the vibration signal to be decomposed into a stable signal, and enabling the demodulated signal to have the most concentrated frequency band expression;
1.2) extracting the corresponding most concentrated frequency band component through a band-pass filter, and performing inverse demodulation on the filtered signal through an inverse transformation operator to obtain a meaningful frequency component;
1.3) repeatedly executing the step 1.2), gradually extracting all frequency components from the original vibration signal until all components are decomposed.
4. A fault diagnosis method as claimed in claim 3, characterized in that the nonlinear chirp signal is defined by a polynomial phase signal model as:
wherein a (t) is the instantaneous amplitude of the signal, ci(i-0, 1 …, k) is the phase parameter, k is the order,is the initial phase of the signal, t represents time, g (t) represents the polynomial phase signal, ti+1Representing the (i + 1) th order time instant, j representing the imaginary part.
5. A method of fault diagnosis as claimed in claim 3, characterized in that the demodulation operator Φ is defined by estimating the demodulation parameters-(t, C) and inverse transform operator Φ+(t, C) are respectively:
7. The method for diagnosing faults according to claim 5, wherein in the step 1.2), the spectral concentration index SCI (C) is defined:
SCI(C)=E(|F(gd(t,C))|4)
wherein E (-) is the expectation and F (-) is the Fourier transform;
8. A modulation and deep learning fused equipment fault diagnosis system is characterized by comprising: the system comprises a time-frequency diagram drawing module, a fault diagnosis model obtaining module and a fault diagnosis module;
the time-frequency diagram drawing module is used for decomposing the vibration signal of the gas turbine casing into a plurality of frequency components to obtain the instantaneous frequency of each frequency component and drawing a time-frequency diagram;
the fault diagnosis model acquisition module is used for preprocessing the time-frequency diagram to be used as the input of the convolutional neural network, and obtaining a convolutional neural network fault diagnosis model through training;
the fault diagnosis module convolutional neural network fault diagnosis model realizes the diagnosis of the gas turbine rotor fault by utilizing the characteristic extraction capability of the convolutional neural network.
9. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-7.
10. A computing device, comprising: one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011223732.9A CN112329626A (en) | 2020-11-05 | 2020-11-05 | Modulation and deep learning fused equipment fault diagnosis method, system and medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011223732.9A CN112329626A (en) | 2020-11-05 | 2020-11-05 | Modulation and deep learning fused equipment fault diagnosis method, system and medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112329626A true CN112329626A (en) | 2021-02-05 |
Family
ID=74315956
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011223732.9A Pending CN112329626A (en) | 2020-11-05 | 2020-11-05 | Modulation and deep learning fused equipment fault diagnosis method, system and medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112329626A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113627375A (en) * | 2021-08-16 | 2021-11-09 | 北京信息科技大学 | Planetary gear fault diagnosis method and system, storage medium and computing device |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109977920A (en) * | 2019-04-11 | 2019-07-05 | 福州大学 | Fault Diagnosis of Hydro-generator Set method based on time-frequency spectrum and convolutional neural networks |
-
2020
- 2020-11-05 CN CN202011223732.9A patent/CN112329626A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109977920A (en) * | 2019-04-11 | 2019-07-05 | 福州大学 | Fault Diagnosis of Hydro-generator Set method based on time-frequency spectrum and convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
蒋龙陈,王红军,张顺利,左云波,陈晓: "基于卷积神经网络的燃气轮机故障诊断研究" * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113627375A (en) * | 2021-08-16 | 2021-11-09 | 北京信息科技大学 | Planetary gear fault diagnosis method and system, storage medium and computing device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Borghesani et al. | CS2 analysis in presence of non-Gaussian background noise–Effect on traditional estimators and resilience of log-envelope indicators | |
Zhang et al. | Kurtosis based weighted sparse model with convex optimization technique for bearing fault diagnosis | |
Wang et al. | Fault diagnosis of diesel engine based on adaptive wavelet packets and EEMD-fractal dimension | |
Gryllias et al. | A Support Vector Machine approach based on physical model training for rolling element bearing fault detection in industrial environments | |
Wang et al. | Synthesis versus analysis priors via generalized minimax-concave penalty for sparsity-assisted machinery fault diagnosis | |
Chen et al. | Fault identification of gearbox degradation with optimized wavelet neural network | |
CN111238843B (en) | Fan health evaluation method based on rapid spectrum kurtosis analysis | |
CN108731945B (en) | Method for extracting fault signal characteristic information of aircraft engine rotor system | |
CN108072517B (en) | Method for detecting weak fault signal of rotating machine | |
Shuuji et al. | Low-speed bearing fault diagnosis based on improved statistical filtering and convolutional neural network | |
Zheng et al. | Faults diagnosis of rolling bearings based on shift invariant K-singular value decomposition with sensitive atom nonlocal means enhancement | |
Chen et al. | Fault identification method for planetary gear based on DT-CWT threshold denoising and LE | |
CN113627375A (en) | Planetary gear fault diagnosis method and system, storage medium and computing device | |
Zhang et al. | Improved local cepstrum and its applications for gearbox and rolling bearing fault detection | |
CN112098093A (en) | Bearing fault feature identification method and system | |
CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
Toma et al. | Comparative analysis of continuous wavelet transforms on vibration signal in bearing fault diagnosis of induction motor | |
Zhou et al. | Sparse dictionary analysis via structure frequency response spectrum model for weak bearing fault diagnosis | |
CN112329626A (en) | Modulation and deep learning fused equipment fault diagnosis method, system and medium | |
Gong et al. | Application of optimized multiscale mathematical morphology for bearing fault diagnosis | |
CN114263621A (en) | Test method and system for diagnosing and simulating cavitation fault of centrifugal pump | |
CN107490477B (en) | The Fault Diagnosis of Gear Case method compared based on frequency spectrum kernel density function correlation | |
CN113988125A (en) | Torsional vibration signal instantaneous frequency extraction method based on improved synchronous compression transformation | |
CN110160642B (en) | Propeller noise line spectrum reconstruction method under small sample condition | |
CN113639985B (en) | Mechanical fault diagnosis and state monitoring method based on optimized fault characteristic frequency spectrum |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210205 |