CN113705096A - Small sample deep learning-based impact fault diagnosis - Google Patents

Small sample deep learning-based impact fault diagnosis Download PDF

Info

Publication number
CN113705096A
CN113705096A CN202111000344.9A CN202111000344A CN113705096A CN 113705096 A CN113705096 A CN 113705096A CN 202111000344 A CN202111000344 A CN 202111000344A CN 113705096 A CN113705096 A CN 113705096A
Authority
CN
China
Prior art keywords
equipment
fault
fault diagnosis
frequency domain
migration
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
Application number
CN202111000344.9A
Other languages
Chinese (zh)
Inventor
高晖
赵大力
刘锦南
王牮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Bohua Xinzhi Technology Co ltd
Original Assignee
Beijing Bohua Xinzhi Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Bohua Xinzhi Technology Co ltd filed Critical Beijing Bohua Xinzhi Technology Co ltd
Priority to CN202111000344.9A priority Critical patent/CN113705096A/en
Publication of CN113705096A publication Critical patent/CN113705096A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M7/00—Vibration-testing of structures; Shock-testing of structures
    • G01M7/08—Shock-testing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The application discloses a class-I impact fault diagnosis method based on small sample deep learning, which comprises the following steps: acquiring migration characteristics corresponding to the equipment monitoring data to obtain a migration characteristic set; reconstructing the migration feature set to obtain a training set of the equipment, wherein the training set comprises real frequency domain data and virtual frequency domain data of the equipment; training the training set based on machine learning, and constructing a fault diagnosis model of the equipment, wherein the fault diagnosis model is used for identifying whether the equipment has faults or not. According to the method and the device, the proper migration characteristic is selected according to the fault mechanism of the equipment, the migration characteristic set is reconstructed to generate a rich training set, the generated training set is trained, and the fault diagnosis model of the equipment is constructed, so that whether the equipment has faults or not can be accurately diagnosed by utilizing the constructed fault diagnosis model, the accuracy and the efficiency of fault diagnosis of the equipment are improved, and the method and the device have good variable-load working condition migration capability.

Description

Small sample deep learning-based impact fault diagnosis
Technical Field
The application relates to the technical field of computers, in particular to a class-I impact fault diagnosis method based on small sample deep learning.
Background
In the using process of the equipment, due to the action of friction, external force, stress and chemical reaction, parts are gradually abraded, corroded and broken, so that the equipment is stopped due to failure. In order to prevent economic loss caused by fault shutdown, equipment faults need to be detected in time so as to enhance equipment maintenance, for example, the parts can be repaired and replaced before the parts enter a severe abrasion stage if the abrasion condition of the parts is mastered.
At present, in the process of detecting equipment faults, whether the equipment faults occur or not is mainly diagnosed through manual investigation, such as human eye observation of engineers or experience of the engineers, so that the diagnosis accuracy is low and the cost is high.
Disclosure of Invention
In view of the above defects or shortcomings in the prior art, it is desirable to provide a class-one impact fault diagnosis method based on small sample deep learning, which constructs an equipment fault diagnosis model based on machine learning, realizes intelligent diagnosis of equipment faults, and improves recognition accuracy and efficiency.
In a first aspect, an embodiment of the present application provides a class-one impact fault diagnosis method based on deep learning of a small sample, where the method includes:
acquiring migration characteristics corresponding to the equipment monitoring data to obtain a migration characteristic set;
reconstructing the migration feature set to obtain a training set of the equipment, wherein the training set comprises real frequency domain data and virtual frequency domain data of the equipment;
and training the training set based on machine learning, and constructing a fault diagnosis model of the equipment, wherein the fault diagnosis model is used for identifying whether the equipment has faults or not.
In summary, according to the class-I impact fault diagnosis method for the equipment based on the small sample deep learning, the proper migration characteristic is selected according to the fault mechanism of the equipment, the migration characteristic set is reconstructed based on the obtained real migration characteristic to generate a rich training set, and finally the model of the generated training set is trained to construct the fault diagnosis model of the equipment, so that whether the equipment has a fault or not can be accurately diagnosed by using the constructed fault diagnosis model, the accuracy and the efficiency of the fault diagnosis of the equipment are improved, and the fault diagnosis method has good variable-load working condition migration capability.
Furthermore, the acquired real monitoring data of the small sample is reconstructed by utilizing the synthetic cycle generation countermeasure network to generate vivid virtual frequency domain data, so that the enhancement and enrichment of the training sample are realized.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings in which:
FIG. 1 is a schematic flow chart diagram illustrating a method for constructing a device fault diagnosis model according to an embodiment of the present application;
FIG. 2 is a schematic flow chart of monitoring data analysis according to an embodiment of the present application;
FIG. 3 is a schematic structural diagram of a simulation test stand according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a training set reconstruction process according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a training set reconstruction process according to an embodiment of the present application;
FIG. 6 is a schematic diagram of a training set reconstruction process according to an embodiment of the present application;
FIG. 7 is a schematic diagram of a training set reconstruction process according to an embodiment of the present application;
FIG. 8 is a schematic representation of the reconstruction results of an embodiment of the present application;
FIG. 9 is a schematic flow chart of a further configuration of a fault diagnosis model according to an embodiment of the present application;
FIG. 10 is a schematic diagram illustrating the results of device fault diagnosis according to an embodiment of the present application;
FIG. 11 is a schematic diagram illustrating the results of device fault diagnosis according to an embodiment of the present application;
FIG. 12 is a schematic diagram illustrating the results of device fault diagnosis according to an embodiment of the present application;
FIG. 13 is a schematic diagram of an apparatus fault diagnosis model construction device according to an embodiment of the present application;
fig. 14 is a schematic structural diagram of a computer system of a server according to an embodiment of the present application.
Detailed Description
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant application and are not limiting of the application. It should be noted that, for the convenience of description, only the portions relevant to the application are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
It can be understood that, in the scene of diagnosing faulty equipment, the equipment fault needs to be detected in time to enhance the maintenance of the equipment, for example, the fault condition of a type of impact is grasped to ensure the normal operation of the equipment.
In the process of diagnosing equipment faults, training can be carried out by taking various kinds of operation condition information of the equipment and fault information generated under various kinds of operation conditions as sample data based on a machine learning algorithm, and a diagnosis model capable of intelligently identifying the equipment faults is constructed.
It can be understood that the actual operation conditions of the equipment are complex and changeable, and the condition information is adopted as the condition to adapt to different conditions.
A large amount of realistic device operation information is required as a training sample in the diagnostic model building process. While in practice the sample data available for model training is limited.
In order to improve the accuracy and efficiency of equipment fault diagnosis, under the scene of small sample type impact fault diagnosis, a sample data generation model (namely a generator) of equipment is obtained by generating a network through cyclic confrontation, and then rich sample data is provided for construction of an identification model by utilizing the generator. And finally, constructing an equipment fault diagnosis model by using the sample data generated by the generator.
For convenience of understanding and explanation, the device fault diagnosis model construction method, the identification method, the device and the medium according to the embodiment of the present application will be described in detail below with reference to fig. 1 to 14.
Fig. 1 is a schematic flowchart of a class of impact fault diagnosis method based on small sample deep learning according to an embodiment of the present application, and as shown in fig. 1, the method includes:
s110, obtaining migration characteristics corresponding to the equipment monitoring data to obtain a migration characteristic set.
Specifically, in the embodiment of the application, firstly, a failure mechanism of the device to be researched can be analyzed, and a change rule of migration characteristics under various failures is mastered so as to determine the migration characteristics and establish a migration characteristic set based on the rule.
For example, the valid feature vector applicable to diagnosing the equipment fault may be determined as the migration feature according to the related knowledge, experience, and the like provided by the fault mechanism, the diagnosis engineer, and the like, and may be expressed as:
Fea＝{fea_11,fea_12,...,fea_1j,...,fea_1m}
wherein f isea_1jRepresenting m-dimensional migration features Fea_1The jth feature component of (a), the migration feature may also be Fea_iAnd the like.
It can be understood that, in practice, when the migration features are extracted from the monitoring data on the device, feature extraction and analysis methods such as time domain analysis, frequency domain analysis, time-frequency domain analysis and the like are generally adopted to extract features.
For example, the flow shown in fig. 2 is used to preprocess the monitoring data of the device to obtain frequency domain data as the migration feature.
For example, suppose a device may have n types of failures:
Fau＝{fau_1,fau_2,...,fau_i,...,fau_n}
determining F under various faults according to the mechanism of each faulteaThe change rule of (2) and further converting the change condition into a mathematical model description, expressed as:
Figure BDA0003234468350000041
wherein h isi-jIndicating the occurrence of a fault fau_iTime characteristic component fea_1jMathematical expression of the variation.
It will be appreciated that the application of the device failure diagnostic algorithm is in the major bearing device failures.
In the embodiments of the present application, for better understanding, the acquisition of monitoring data of a bearing device and the construction of a bearing fault diagnosis model are taken as examples for explanation.
The method is characterized in that a migration feature set of the bearing equipment is constructed, vibration data of an actual bearing can be collected, fault data simulated by a built test bed can also be collected, and the two kinds of test data are used as real sample data in fault machine learning of the bearing equipment.
The 6205 deep groove ball bearing produced by SKF is adopted near the driving end of the motor in the bearing experiment, the working conditions of the bearing at the driving end under different states are simulated respectively, each state is subjected to 0-3 horsepower 4 load tests respectively, vibration signals are collected through an acceleration sensor arranged on a bearing seat, the sampling frequency is 12kHz, parameters such as theoretical fault frequency and the like and data serial numbers are shown in table 1, and the fault size 0 is healthy frequency domain data.
TABLE 1 bearing data parameters and numbering
Figure BDA0003234468350000051
In addition, a test bench as shown in fig. 3 can be set up to perform a simulated bearing fault test. The experiment table comprises speed regulating motor, drive division and support bearing etc. and the antifriction bearing model is N205EM, and 2 acceleration sensor magnetism are inhaled and are fixed in on the bearing frame, sampling frequency 25 kHz. At a certain rotating speed, faults of an inner ring and an outer ring of the bearing are artificially manufactured, data under three states of health, faults of the inner ring and faults of the outer ring are respectively collected, application verification is carried out by adopting data of a No. 2 sensor, and more experimental data parameters are listed in a table 2.
TABLE 2 bench simulation Fault data parameters
Figure BDA0003234468350000052
After the monitoring data of the bearing equipment are collected, the migration feature set can be obtained through spectrum analysis.
It can be understood that the spectrum analysis is the most effective diagnostic method for the rotating equipment, i.e. the equipment fault can be inferred through observing the spectrum according to experience, the principle is that each rotation of the rotating equipment generates impact on the fault part to form obvious periodic impact response, and based on the obvious periodic impact response, the impact can be added to the spectrum according to the response period of the fault part, so that the fault data can be obtained.
Namely, the bearing failure mechanism analysis shows that the outer ring failure has obvious outer ring failure frequency FOR(ii) a Inner ring fault has obvious inner ring fault frequency FIRAlso receives the power frequency FrModulation of (2), F may occurIR±(t)Fr(ii) a Rolling element failure with obvious rolling element failure frequency FBWhile F isBIs also subjected to FrAnd cage failure frequency FFTModulation of (2), F may occurB±(t)Fr，FB±(t)FFT. In addition, there are frequency doubling and sidebands of the above major fault frequencies.
Then, according to experience, each frequency multiplication and sideband intensity is sequentially decreased, the attenuation coefficient α in the embodiment of the present application may be 0.65, and the sideband order t is {1,2,3 }; the theoretical fault frequency in the real fault data is not consistent with the actual fault frequency, a value which floats up and down on the theoretical fault frequency is used as the predicted fault frequency, and the value is the reference frequency resolution and the difference between the fault frequencies.
It can also be understood that mechanical vibration is often accompanied by modulation phenomena, which require selection of appropriate frequency bands for bandpass filtering compared to conventional signal processing methods. In the embodiment of the application, 2kHz high-pass filtering can be selected empirically, and the flow shown in fig. 2 is adopted to preprocess the monitoring data of the vibration signal, so that the obtained frequency domain data can be used as the migration feature, that is, the acquisition of the migration feature set of the bearing device is realized.
Specifically, the expression of the preprocessed vibration signal is as follows:
Figure BDA0003234468350000061
where N is the number of sampling points, f0To frequency resolution, CiIs frequency i x f0Is a function of the excitation, fresThe residual vibration of the equipment.
In the formula, N, f0Delta can be obtained directly from the pretreatment process, and Ci、fresIs a key parameter reflecting the state of the apparatus, CiFor the size of each migration feature component, fresRelated to device personality characteristics. For the device individual characteristic space required to be obtained in the method, namely the health state of the device, C is obtained through GAN learningi、fres。
For example, for a bearing device, then CiIncluding outer ring fault frequency component coefficient
Figure BDA0003234468350000062
Inner ring fault frequency component coefficient
Figure BDA0003234468350000063
And rolling element fault frequency component coefficient
Figure BDA0003234468350000064
As can be seen from tables 1 and 2, the bearing device has an inner ring fault, an outer ring fault and a rolling element fault, i.e., the fault type n is 3; the sampling point N can be 12288, the characteristic dimension m is (N/2+1), and the fault feature set is as follows:
H＝[HOR HIR HB]T (3)
for outer ring faults, the fault response period is
Figure BDA0003234468350000065
Then
Figure BDA0003234468350000066
Figure BDA0003234468350000067
Figure BDA0003234468350000071
In the formula (I), the compound is shown in the specification,
Figure BDA0003234468350000072
respectively doubling the frequency of the outer ring fault frequency by 1-3; p is the multiplication factor, expressed as FreThe expected amplitude value is taken as a reference and multiplied by a coefficient beta, wherein the beta is used for reducing the influence of various random factors in practice, and the value of the beta is sampled in Gaussian distribution with the mean value of 1 and the standard deviation of 0.33 between 0 and 2; and q is the intensity of measuring the impact response, and is a sampling value in uniform distribution at 3-100 according to the 3 sigma principle and magnitude difference for highlighting the fault characteristics.
It should be noted that: for F possibly occurring as mentioned in the bearing failure mechanismrAnd FFTThe component P is multiplied by a coefficient uniformly distributed between 0 and 1.
It can be understood that for the test bed simulation fault data, there are two kinds of fault and health frequency domain data of the inner circle and the outer circle, the number of sampling points is 16384, and the same method can be adopted to obtain the fault feature set.
It will also be appreciated that the actual true feature quantities are small, i.e., insufficient as a training set for machine learning.
In this embodiment of the present application, the obtained migration feature set may be reconstructed to learn distribution features of the device monitoring data, and generate rich training sample data, that is, execute S120.
S120, reconstructing the migration feature set to obtain a training set of the equipment, wherein the training set comprises real frequency domain data and virtual frequency domain data of the equipment.
Specifically, the embodiment of the present application may perform reconstruction of sample data by using a round robin generation countermeasure network (GAN).
This reconstruction process may be implemented by a method as shown in fig. 4. The method specifically comprises the following steps:
and S121, generating a loop generation countermeasure network.
S122, training the circularly generated countermeasure network based on the migration feature set to obtain a target generator, wherein the target generator is used for generating virtual frequency domain data of the equipment;
and S123, generating the training set based on the reconstructed model and the acquired feature set, wherein the training set comprises real samples and generated virtual samples.
Specifically, an initial loop-generated countermeasure network may be set up, that is, the network includes two generators and two discriminators, such as a first generator, a second generator, a first discriminator, and a second discriminator.
It will be appreciated that the first generator may comprise a mapping algorithm of virtual frequency domain data of the device to real characteristic frequency domain data, the input of which is the virtual frequency domain data of the device, so that after the virtual frequency domain data is input, the characteristics of the real frequency domain data can be learned due to the mapping algorithm to output real frequency domain data that is as realistic as possible. The second generator may include a mapping algorithm of real frequency domain data to virtual frequency domain data, so that after the real frequency domain data is input, the real frequency domain data is made to learn the characteristics of the virtual frequency domain data, and device frequency domain data similar to the virtual frequency domain data is output.
The first discriminator may be used to identify the output of the first generator and the second discriminator may be used to identify the output of the second generator. It will be appreciated that in generating this first discriminator, the device frequency domain data in the training set may be utilized such that it outputs a high score for the real frequency domain data and a low score for the virtual state features. Similarly, for the second discriminator, the frequency domain data in the training set may be used for preliminary training, so that when the real state features are output, the low score is output, and when the virtual state features are input, the high score is output.
It can be understood that the mapping table established by the preliminary learning in the embodiment of the present application includes a small amount of information, so that the generated frequency domain data has a large difference from the actual state feature. Therefore, through continuous training of the generator, the mapping relation table can be gradually enriched to obtain an optimized generator model, so that the generator model outputs vivid virtual frequency domain data after inputting the frequency domain data.
After the cyclic generation confrontation network is generated, the acquired data in the training set can be input into the network, and the two generators which are preliminarily constructed are trained to obtain a plurality of optimized first generators, namely target generators, so that the frequency domain data output by the target generators are as vivid as possible, namely the target generators are used for generating training sample data for training the equipment fault diagnosis model.
For example, as shown in fig. 5, 6 and 7, the device personality characteristic space generator G may be first configured based on the real data x of the state of the specific device and using the operating condition information c as the condition of the c-DCGANcharAnd reconstructing, namely constructing the target generator. Then cascading the random variable z and the working condition parameter c as GcharVia the deconvolution and activation function output Gchar(z + c); finally G ischar(z + c) and x are randomly input into D, and whether the input data is x or G is judged through a plurality of layers of convolution layers, pooling layers and BN operationcharThe generated false data to train an accurate discriminator.
The generator G is trained cyclically using the objective function of equation (1) abovecharAnd a discriminator D for discriminating every trainingDevice D, update k times generator Gchar。
It can be understood that in the GAN training, the gradient vanishing problem exists in the generator, because the discriminant and the generator training are not balanced, so that the discriminant performance is good and the generator cannot generate vivid data. Therefore, in the training process, the discriminator is updated once, k secondary generators are trained, and in the generation of the one-dimensional vibration data, k is preferably 4-10. Table 1 shows the network parameters in the recurrent generated countermeasure network.
TABLE 3 c DCGAN network parameters
Figure BDA0003234468350000091
Further, after the goal generator, i.e., the objective function, for generating the sample data is constructed through the learning of the countermeasure network, the sample data can be sampled by the goal generator GcharThe virtual healthy frequency domain data x and the real healthy frequency domain data x are mixed, namely, the migration characteristic set operation is carried out, the migration characteristic set operation is used as the healthy frequency domain data in the training sample, and G is carried outfautSampling to obtain n fault frequency domain data samples of the equipment, thereby forming n +1 frequency domain data as training samples for GfautAnd GcharShould be consistent with the number of samples for each state. I.e. sampling at GcharAnd calculating the virtual health frequency domain data samples with the feature set to obtain data samples of various virtual faults of the equipment, and combining the data samples with x to form training samples.
It is understood that, in the generated training samples, the characteristic data includes two types, one type is real frequency domain data, such as healthy frequency domain data sample data that can be acquired in a large amount, a small amount of fault data sample data during a commissioning, and the other type is generated virtual frequency domain data, where the generation of the virtual frequency domain data is represented as:
Gfaut＝H×Gchar(z+c) (5)
as shown in fig. 8, the reconstructed virtual frequency domain data is very close to the real frequency domain data, and can be completely used as training sample data of the device for constructing a diagnostic model.
It is understood that after the reconstruction of the training sample data is completed through the network, the reconstructed training sample may be used to generate a fault diagnosis model of the device, i.e., the method may execute S130.
And S130, training the training set based on machine learning, and constructing a fault diagnosis model of the equipment, wherein the fault diagnosis model is used for identifying whether the equipment has faults or not.
Specifically, the preprocessed monitoring data is used as a fault classifier, i.e., an input of a fault diagnosis model, so as to train the fault diagnosis model.
As shown in fig. 9, the fault classifier is a one-dimensional CNN model based on Lenet-5 structure, and includes 4 convolution layers and pooling layers alternately connected, and 2 fully-connected layers, where the 2 nd fully-connected layer has a neuron number, the probability that the input is classified into each state is predicted, and the state with the highest probability is finally output by the fault classifier via Softmax as a diagnosis conclusion, and its detailed network parameters are shown in table 4.
TABLE 4 Fault classifier parameters
Figure BDA0003234468350000101
In the embodiment of the application, Tensorflow 1.8 can be adopted for building and training the diagnosis model, cross entropy loss is taken as a target function, the initial learning rate is set to be 0.001, and the batch size is 32 and taken as a training parameter to train the fault classifier.
Further, the present application also provides an apparatus fault diagnosis method, that is, after the apparatus fault diagnosis model is constructed by the above embodiments, the constructed fault diagnosis model can be used to diagnose and identify an apparatus fault.
Specifically, monitoring data of the device to be diagnosed, such as operation parameters of various aspects, may be acquired, and then the detection data may be preprocessed to extract migration features. Finally, the extracted migration features of the device may be input into the fault diagnosis model constructed in the above embodiment to diagnose whether the device has a fault.
For example, for the data of the bearing device, the method of the above embodiment is used to perform sampling, and the N _0 and N _3 data are taken as known data, and a migration feature set is combined to obtain training samples, each type of state includes 1024 samples, so as to train the diagnostic model shown in fig. 5. Test samples were obtained for 36 fault data and 2 health data (N _1, N _2) in table 3 at 64 resampling intervals.
For the simulation fault data of the test bed, 138 health state samples are taken as known data, training sample training diagnosis models with 512 samples in three states are established, and the rest data of 3 states listed in table 4 are tested.
The test results are shown in tables 5 and 6, respectively.
TABLE 5 diagnostic accuracy for actual bearing data
Figure BDA0003234468350000111
TABLE 6 diagnostic accuracy for bench simulation fault data
Figure BDA0003234468350000112
As can be seen from tables 5 and 6, the average diagnosis accuracy rate of the method in the embodiment of the application to the bearing data exceeds 84 percent, the diagnosis accuracy rate to the simulation fault data of the test bed reaches 100 percent,
the detailed analysis is as follows: by observing the envelope frequency spectrum of a large amount of experimental data, the diagnosis capability of the method is superior to that of manual work, for example, as shown in fig. 10, the envelope frequency spectrum obtained by adopting the flow of fig. 2 is adopted for B07_0 fault data, if the judgment of human eyes is easily interfered by the fault frequency of the 166Hz inner ring, the method obtains an accurate diagnosis conclusion, and the identification time is within milliseconds; the diagnosis accuracy rate of the OR14 fault is obviously lower than that of other experimental data, for example, the OR14_0 fault data envelope frequency spectrum in fig. 11 shows that the frequency band is more, the characteristic is often seen in rolling element faults, and meanwhile, the obvious 165Hz inner ring fault frequency and frequency multiplication exist, it is presumed that the rolling element faults and the inner ring faults caused by the outer ring moderate faults are caused, at this time, the composite fault state is already existed, but the faults are only marked as outer ring faults when the faults are marked, and the reason that the B21 fault diagnosis accuracy rate is low is similar to this.
And (3) variable working condition migration performance analysis:
by comparing and analyzing the method, as shown in fig. 12, only healthy frequency domain data and 3 fault data at 0.007 inches in the bearing data were tested. The diagnosis accuracy of the comparison method on 2 experimental data is shown in tables 7 and 8, in table 7, oblique line lattice data is used as a training sample of the method in the table 7, and in table 8, 138 groups of data are used as the training sample in each working condition.
TABLE 7 diagnostic accuracy of comparison method on partial bearing data
Figure BDA0003234468350000121
TABLE 8 diagnostic accuracy of the comparative method on bench simulation fault data
Figure BDA0003234468350000122
Comparing the italic parts of table 7 and table 5, the diagnosis accuracy of the embodiment of the present application is high under the variable load working condition, and the embodiment of the present application uses two healthy frequency domain data to realize the effective diagnosis of the rest 76.32% of the working conditions in table 3 (the effective diagnosis is that the accuracy is more than 95%).
The above results show that the embodiment of the application has good variable load working condition migration capability. The method selects proper migration characteristics according to the fault mechanism and generates fault data according to the distribution rule of the migration characteristics, so that the diagnosis model trained based on the data can distinguish the type of state.
According to the method for diagnosing the impact fault of the equipment based on the small sample deep learning, the proper migration characteristic is selected according to the fault mechanism of the equipment, the migration characteristic set is reconstructed based on the obtained real migration characteristic to generate a rich training set, and finally the model of the generated training set is trained to construct the fault diagnosis model of the equipment, so that whether the equipment has a fault or not can be accurately diagnosed by utilizing the constructed fault diagnosis model, the accuracy and the efficiency of fault diagnosis of the equipment are improved, and the method has good variable-load working condition migration capability.
On the other hand, the embodiment of the present application further provides a class of impact fault diagnosis device based on small sample deep learning, as shown in the figure, the device 100 may include:
the acquisition module 101 is configured to acquire migration features corresponding to device monitoring data to obtain a migration feature set;
a reconstruction module 102, configured to reconstruct the migration feature set to obtain a training set of the device, where the training set includes real frequency domain data and virtual frequency domain data of the device;
and the training module 103 is configured to train the training set based on machine learning to obtain a fault diagnosis model of the device, where the fault diagnosis model is used to diagnose whether the device has a fault.
Optionally, in the impact fault diagnosis device based on deep learning of a small sample provided in the embodiment of the present application, the obtaining module is specifically configured to:
and carrying out spectrum analysis on the monitoring data to obtain corresponding frequency domain data, wherein the frequency domain data is used as the migration characteristic of the equipment.
Optionally, in the impact fault diagnosis apparatus based on deep learning with small samples provided in this embodiment of the present application, the reconstruction module 102 includes:
a generating unit 1021, configured to generate a loop generation countermeasure network.
A training unit 1022, configured to train the cycle generation countermeasure network based on the migration feature set, so as to obtain a target generator, where the target generator is configured to generate virtual frequency domain data of the device;
the reconstruction unit 1023 generates the training set based on the reconstructed model and the acquired feature set, wherein the training set includes real samples and generated virtual samples.
Optionally, in the apparatus for diagnosing a class of impact faults based on small-sample deep learning provided in this embodiment of the present application, the cyclic generation countermeasure network includes a first generator, a second generator, a first discriminator and a second discriminator, where the first generator includes a mapping algorithm of virtual frequency domain data characteristics to real frequency domain data, the second generator includes a mapping algorithm of real frequency domain data to virtual frequency domain data, the first discriminator is configured to identify an output result of the first generator, the second discriminator is configured to identify an output result of the second generator,
wherein the target generator is one of the first generators.
Optionally, in the impact fault diagnosis device based on deep learning of a small sample provided in the embodiment of the present application, the fault diagnosis model is a one-dimensional CNN model, and the one-dimensional CNN model includes a plurality of convolution layers, a pooling layer, and a full-link layer.
Optionally, in the device fault diagnosis model construction apparatus provided in the embodiment of the present application, the device is a bearing, and the migration characteristic is frequency domain data obtained by analyzing the monitored vibration signal, and is represented as:
Figure BDA0003234468350000141
where N is the number of sampling points, f0To frequency resolution, CiIs frequency i x f0Is a function of the excitation, fresThe residual vibration of the equipment.
Optionally, in the device fault diagnosis model construction apparatus provided in the embodiment of the present application, the fault types of the bearing include an inner ring fault, an outer ring fault, and a rolling element fault, and then CiIncluding outer ring fault frequency component coefficient
Figure BDA0003234468350000142
Inner ring fault frequency component coefficient
Figure BDA0003234468350000143
And rolling element fault frequency component coefficient
Figure BDA0003234468350000144
The migration feature set is represented as:
H＝[HOR HIR HB]T，
wherein the content of the first and second substances,
Figure BDA0003234468350000145
Figure BDA0003234468350000146
Figure BDA0003234468350000151
wherein the content of the first and second substances,
Figure BDA0003234468350000152
the frequency of the outer ring fault is 1-3 times, P is a frequency multiplication coefficient, beta is used for reducing the influence of various random factors in practice, and q is used for measuring the strength of the impact response.
Optionally, the one-class impact fault diagnosis based on the small-sample deep learning provided in the embodiment of the present application further includes a diagnosis module 104, configured to obtain monitoring data of a device to be diagnosed;
preprocessing the monitoring data to obtain a corresponding migration feature set;
and inputting the device migration feature set into the fault diagnosis model, and outputting a fault diagnosis result of the device.
On the other hand, the embodiment of the present application further provides a server, where the server includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the method for building the device fault diagnosis model as described above.
Referring now to FIG. 8, shown is a block diagram of a computer system 800 suitable for use in implementing a server according to embodiments of the present application.
As shown in fig. 8, the computer system 800 includes a Central Processing Unit (CPU)801 that can perform various appropriate actions and processes in accordance with a program stored in a Read Only Memory (ROM)802 or a program loaded from a storage section 803 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data necessary for the operation of the system 800 are also stored. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input/output (I/O) interface 805 is also connected to bus 804.
The following components are connected to the I/O interface 805: an input portion 806 including a keyboard, a mouse, and the like; an output section 807 including a signal such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 808 including a hard disk and the like; and a communication section 809 including a network interface card such as a LAN card, a modem, or the like. The communication section 809 performs communication processing via a network such as the internet. A drive 810 is also connected to the I/O interface 805 as necessary. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as necessary, so that a computer program read out therefrom is mounted on the storage section 808 as necessary.
In particular, the process described above with reference to fig. 1 may be implemented as a computer software program according to embodiments of the device fault diagnosis model building disclosed herein. For example, embodiments of the device fault diagnosis model construction disclosed herein include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for performing the method of FIG. 1. In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 809 and/or installed from the removable medium 811.
It should be noted that the computer readable medium shown in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various fault diagnosis embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The units or modules described in the embodiments of the present application may be implemented by software or hardware. The described units or modules may also be provided in a processor, and may be described as: a processor includes an acquisition module, a reconstruction module, and a training module. The names of these units or modules do not in some cases form a limitation on the units or modules themselves, for example, a training module may also be described as "reconstructing the migration feature set to obtain a training set of the device, where the training set includes real frequency domain data and virtual frequency domain data of the device".
As another aspect, the present application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the foregoing device in the foregoing embodiment; or it may be a separate computer readable storage medium not incorporated into the device. The computer readable storage medium stores one or more programs, which are used by one or more processors to execute a class of impact fault diagnosis methods based on small sample deep learning described in the present application, and specifically executes:
acquiring migration characteristics corresponding to the equipment monitoring data to obtain a migration characteristic set;
reconstructing the migration feature set to obtain a training set of the equipment, wherein the training set comprises real frequency domain data and virtual frequency domain data of the equipment;
and training the training set based on machine learning to obtain a fault diagnosis model of the equipment, wherein the fault diagnosis model is used for diagnosing whether the equipment has faults or not.
In summary, according to the class-one impact fault diagnosis method for the equipment based on the small sample deep learning, the appropriate migration characteristic is selected according to the fault mechanism of the equipment, then the migration characteristic set is reconstructed based on the acquired real migration characteristic to generate a rich training set, finally the model of the generated training set is trained, and the fault diagnosis model of the equipment is constructed, so that whether the equipment has a fault or not can be accurately diagnosed by using the constructed fault diagnosis model, the accuracy and the efficiency of the fault diagnosis of the equipment are improved, and the fault diagnosis method has good variable-load working condition migration capability.
The above description is only a preferred embodiment of the application and is illustrative of the principles of the technology employed. It will be appreciated by persons skilled in the art that the scope of the application referred to in the present application is not limited to the particular combination of features described above, but also covers other arrangements formed by any combination of the above features or their equivalents without departing from the spirit of the application. For example, the above features may be replaced with (but not limited to) features having similar functions disclosed in the present application.

Claims (8)

1. A class-I impact fault diagnosis method based on small sample deep learning is characterized by comprising the following steps:
acquiring migration characteristics corresponding to the equipment monitoring data to obtain a migration characteristic set;
reconstructing the migration feature set to obtain a training set of the equipment, wherein the training set comprises real frequency domain data and virtual frequency domain data of the equipment;
and training the training set based on machine learning, and constructing a fault diagnosis model of the equipment, wherein the fault diagnosis model is used for identifying whether the equipment has faults or not.
2. The method for diagnosing the impact fault based on the small sample deep learning as claimed in claim 1, wherein the obtaining of the migration feature set of the device comprises:
and carrying out spectrum analysis on the monitoring data to obtain the corresponding frequency domain data, wherein the frequency domain data is used as the migration characteristic of the equipment.
3. The method for diagnosing a class of impact faults based on small sample deep learning of claim 1, wherein the reconstructing the migration feature set to obtain the training set of the device comprises:
and generating a loop to generate the countermeasure network.
Training the circularly generated countermeasure network based on the migration feature set to obtain a target generator, wherein the target generator is used for generating virtual frequency domain data of equipment;
and generating the training set based on the reconstruction model and the acquired feature set, wherein the training set comprises real samples and generated virtual samples.
4. The small-sample deep learning-based one-class impact fault diagnosis method according to claim 3, wherein the cyclic generation countermeasure network comprises a first generator, a second generator, a first discriminator and a second discriminator, the first generator comprises a mapping algorithm of virtual frequency domain data characteristics to real frequency domain data, the second generator comprises a mapping algorithm of real frequency domain data to virtual frequency domain data, the first discriminator is used for identifying the output result of the first generator, the second discriminator is used for identifying the output result of the second generator,
wherein the target generator is one of the first generators.
5. The small-sample deep learning-based impact fault diagnosis method according to any one of claims 1-4, wherein the fault diagnosis model is a one-dimensional CNN model, and the one-dimensional CNN model comprises a plurality of convolutional layers, pooling layers and full-link layers.
6. The small-sample deep learning-based impact fault diagnosis method according to any one of claims 1-4, wherein the device is a bearing, and the migration characteristic is frequency domain data obtained by analyzing the monitored vibration signal, and is represented as:
Figure FDA0003234468340000021
where N is the number of sampling points, f0To frequency resolution, CiIs frequency i x f0Is a function of the excitation, fresThe residual vibration of the equipment.
7. The small-sample deep learning-based impact fault diagnosis method as claimed in claim 6, wherein the fault types of the bearing comprise inner ring fault, outer ring fault and rolling body fault, then CiIncluding outer ring fault frequency component coefficient
Figure FDA0003234468340000022
Inner ring fault frequency component coefficient
Figure FDA0003234468340000023
And rolling element fault frequency component coefficient
Figure FDA0003234468340000024
The migration feature set is represented as:
H＝[HOR HIR HB]T，
wherein the content of the first and second substances,
Figure FDA0003234468340000025
Figure FDA0003234468340000026
Figure FDA0003234468340000031
wherein the content of the first and second substances,
Figure FDA0003234468340000032
the frequency of the outer ring fault is 1-3 times, P is a frequency multiplication coefficient, beta is used for reducing the influence of various random factors in practice, and q is used for measuring the strength of the impact response.
8. The small sample deep learning-based impact fault diagnosis method according to any one of claims 1-7, characterized in that the method further comprises:
acquiring monitoring data of equipment to be diagnosed;
preprocessing the monitoring data to obtain a corresponding migration feature set;
and inputting the device migration feature set into the fault diagnosis model, and outputting a fault diagnosis result of the device.
CN202111000344.9A 2021-08-27 2021-08-27 Small sample deep learning-based impact fault diagnosis Pending CN113705096A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111000344.9A CN113705096A (en) 2021-08-27 2021-08-27 Small sample deep learning-based impact fault diagnosis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111000344.9A CN113705096A (en) 2021-08-27 2021-08-27 Small sample deep learning-based impact fault diagnosis

Publications (1)

Publication Number Publication Date
CN113705096A true CN113705096A (en) 2021-11-26

Family

ID=78656385

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111000344.9A Pending CN113705096A (en) 2021-08-27 2021-08-27 Small sample deep learning-based impact fault diagnosis

Country Status (1)

Country Link
CN (1) CN113705096A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114429009A (en) * 2022-04-07 2022-05-03 中国石油大学(华东) Small sample sucker-rod pump well working condition diagnosis method based on meta-migration learning
CN114491823A (en) * 2022-03-28 2022-05-13 西南交通大学 Train bearing fault diagnosis method based on improved generation countermeasure network
CN114757305A (en) * 2022-06-13 2022-07-15 华中科技大学 Voltage transformer insulation fault identification method and system based on ensemble learning

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109710636A (en) * 2018-11-13 2019-05-03 广东工业大学 A kind of unsupervised industrial system method for detecting abnormality based on depth migration study
CN109753998A (en) * 2018-12-20 2019-05-14 山东科技大学 The fault detection method and system, computer program of network are generated based on confrontation type
WO2020079860A1 (en) * 2018-10-18 2020-04-23 株式会社日立製作所 Equipment failure diagnosis support system and equipment failure diagnosis support method
CN111094927A (en) * 2017-09-26 2020-05-01 舍弗勒技术股份两合公司 Bearing fault diagnosis method and device, readable storage medium and electronic equipment
CN111811819A (en) * 2020-06-30 2020-10-23 佛山科学技术学院 Bearing fault diagnosis method and device based on machine learning
CN112649198A (en) * 2021-01-05 2021-04-13 西交思创智能科技研究院(西安)有限公司 Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111094927A (en) * 2017-09-26 2020-05-01 舍弗勒技术股份两合公司 Bearing fault diagnosis method and device, readable storage medium and electronic equipment
WO2020079860A1 (en) * 2018-10-18 2020-04-23 株式会社日立製作所 Equipment failure diagnosis support system and equipment failure diagnosis support method
CN109710636A (en) * 2018-11-13 2019-05-03 广东工业大学 A kind of unsupervised industrial system method for detecting abnormality based on depth migration study
CN109753998A (en) * 2018-12-20 2019-05-14 山东科技大学 The fault detection method and system, computer program of network are generated based on confrontation type
CN111811819A (en) * 2020-06-30 2020-10-23 佛山科学技术学院 Bearing fault diagnosis method and device based on machine learning
CN112649198A (en) * 2021-01-05 2021-04-13 西交思创智能科技研究院(西安)有限公司 Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李智 等: "《基于循环生成对抗网络的道路场景语义分割》", 《武汉大学学报(理学版)》 *
马波 等: "《基于GAN样本生成技术的智能诊断方法》", 《振动与冲击》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114491823A (en) * 2022-03-28 2022-05-13 西南交通大学 Train bearing fault diagnosis method based on improved generation countermeasure network
CN114491823B (en) * 2022-03-28 2022-07-12 西南交通大学 Train bearing fault diagnosis method based on improved generation countermeasure network
CN114429009A (en) * 2022-04-07 2022-05-03 中国石油大学(华东) Small sample sucker-rod pump well working condition diagnosis method based on meta-migration learning
CN114429009B (en) * 2022-04-07 2022-07-29 中国石油大学(华东) Small sample sucker-rod pump well working condition diagnosis method based on meta-migration learning
CN114757305A (en) * 2022-06-13 2022-07-15 华中科技大学 Voltage transformer insulation fault identification method and system based on ensemble learning
CN114757305B (en) * 2022-06-13 2022-09-20 华中科技大学 Voltage transformer insulation fault identification method and system based on ensemble learning

Similar Documents

Publication Publication Date Title
CN113705096A (en) Small sample deep learning-based impact fault diagnosis
CN107560849B (en) Wind turbine generator bearing fault diagnosis method of multichannel deep convolutional neural network
Li et al. Extracting repetitive transients for rotating machinery diagnosis using multiscale clustered grey infogram
CN109858352B (en) Fault diagnosis method based on compressed sensing and improved multi-scale network
Cong et al. Short-time matrix series based singular value decomposition for rolling bearing fault diagnosis
Lin et al. Hyper-spherical distance discrimination: A novel data description method for aero-engine rolling bearing fault detection
JP7462400B2 (en) Artificial intelligence device and method for pre-processing noise data to identify problem noise sources
CN107003665B (en) For assessing the method for being connected to the state of rotating machinery of motor
CN111238815B (en) Bearing fault identification method based on data enhancement under sample imbalance
CN109187021B (en) Multi-source Wind turbines Method for Bearing Fault Diagnosis based on entropy
CN104198184A (en) Bearing fault diagnostic method based on second generation wavelet transform and BP neural network
CN104215456B (en) Plane clustering and frequency-domain compressed sensing reconstruction based mechanical fault diagnosis method
CN111914883A (en) Spindle bearing state evaluation method and device based on deep fusion network
CN112179691A (en) Mechanical equipment running state abnormity detection system and method based on counterstudy strategy
CN109932179A (en) A kind of rolling bearing fault testing method based on the reconstruct of DS Adaptive spectra
CN110060368A (en) Mechanical method for detecting abnormality based on potential feature coding
CN111458144B (en) Wind driven generator fault diagnosis method based on convolutional neural network
Zhao et al. Rolling bearing composite fault diagnosis method based on EEMD fusion feature
CN114034481A (en) Fault diagnosis system and method for rolling mill gearbox
CN114048787B (en) Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model
CN114705432B (en) Method and system for evaluating health state of explosion-proof motor bearing
CN113916535B (en) Bearing diagnosis method, system, equipment and medium based on time frequency and CNN
CN114235405A (en) Feature extraction method and device of vibration signal, and equipment analysis method and device
Minervini et al. Transfer learning technique for automatic bearing fault diagnosis in induction motors
CN116541771A (en) Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion

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