CN114186593A - Bearing fault identification method, device, equipment and storage medium under noise condition - Google Patents
Bearing fault identification method, device, equipment and storage medium under noise condition Download PDFInfo
- Publication number
- CN114186593A CN114186593A CN202111524116.1A CN202111524116A CN114186593A CN 114186593 A CN114186593 A CN 114186593A CN 202111524116 A CN202111524116 A CN 202111524116A CN 114186593 A CN114186593 A CN 114186593A
- Authority
- CN
- China
- Prior art keywords
- graph
- node
- bearing
- neural network
- sample
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 238000001228 spectrum Methods 0.000 claims abstract description 31
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 230000001133 acceleration Effects 0.000 claims abstract description 15
- 238000010586 diagram Methods 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 4
- 238000010276 construction Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000005065 mining Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 2
- 238000005096 rolling process Methods 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000009467 reduction Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 238000012800 visualization Methods 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
- G06F2218/12—Classification; Matching
-
- 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
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
A method, apparatus, device and storage medium for identifying bearing failure under noisy conditions, the method comprising: s1: collecting a vibration acceleration signal of a bearing; s2: carrying out normalization processing on an original signal, and constructing a continuous preset number of sampling points into a sample to be measured; s3: and performing discrete wavelet transform on the sample to be detected, and performing signal reconstruction by taking a low-frequency coefficient of wavelet decomposition of one layer to obtain a reconstructed signal. S4: and performing fast Fourier transform on the reconstructed signal in the S3, and taking a half spectrum of the reconstructed signal as a node characteristic of a neural network of a subsequent graph. S5: constructing a characteristic diagram: taking the half-edge spectrum in the S4 as the node characteristics of the graph neural network, and calculating Euclidean distances among the characteristics of each node; and forming edge connection based on the K neighbor rule, wherein no edge connection exists between non-neighbor points. S6: and (3) performing feature learning by using graph convolution operation of the graph neural network, and constructing a Soft-max classifier at the last of the graph neural network to realize fault identification of the bearing.
Description
Technical Field
The invention belongs to the technical field of bearing fault diagnosis, and particularly relates to a bearing fault identification method, device, equipment and storage medium under a noise condition.
Background
The bearing is always an important equipment part in industrial production, and the state stability of the bearing is important because the bearing is in a high-speed rotation state for a long time. In the field of bearing fault diagnosis, many researchers have carried out related studies and made related progress. Among them, a failure diagnosis method based on a vibration signal is often used. The method is used for collecting vibration signals in the bearing rotation process, and further analyzing fault characteristics contained in the signals.
The bearing fault identification method based on the vibration signal can be divided into a traditional non-intelligent method and an intelligent identification method. The traditional non-intelligent method is usually established on the basis of expert experience, and partial steps need to be customized, so the popularization range is often limited. The intelligent method mainly takes the technologies of machine learning, deep learning and the like as the main technologies, and emphasizes the characteristics of trainable training and self-learning, so that the universality and the intelligence are stronger. The effectiveness of the intelligent approach depends largely on the effectiveness of the model training. While the effectiveness of model training depends largely on the reliability of the data samples. Most of the research based on intelligent methods at present focuses on the problem of identification under ideal data conditions, and neglects the characteristic of high noise in the background of actual diagnosis. Due to the universality of noise in industrial production, the acquired vibration signals contain a large amount of noise, so that a model trained under an ideal data condition cannot achieve an ideal effect.
The graph neural network can perform feature learning by means of the information of the adjacent edge nodes, so that certain robustness is provided for noise. In contrast, the invention provides a graph network construction mode by utilizing the characteristic of a graph neural network, discloses a method, a device, equipment and a storage medium for identifying the bearing fault under the noise condition, and can well solve the problem of noise interference under the real background.
Disclosure of Invention
In view of the above-identified deficiencies in the art or needs for improvement, the present invention provides a method, apparatus, device and storage medium for identifying bearing faults under noisy conditions, the objective of which is to accurately identify the current state of the bearing. Therefore, the problem that the data samples are interfered by noise to influence the diagnosis precision is solved.
To achieve the above object, according to one aspect of the present invention, there are provided a method, an apparatus, a device, and a storage medium for identifying a bearing failure.
A method of identifying a bearing fault, comprising:
s1: collecting a vibration acceleration signal of a bearing;
s2: carrying out normalization processing on an original signal, and constructing a continuous preset number of sampling points into a sample to be measured;
s3: and performing discrete wavelet transform on the sample to be detected, and performing signal reconstruction by taking a low-frequency coefficient of wavelet decomposition of one layer to obtain a reconstructed signal.
S4: and performing fast Fourier transform on the reconstructed signal in the S3 to obtain a spectrum signal which is equal to the reconstructed signal in length and is symmetrical about the center, and taking a half spectrum of the spectrum signal as the node characteristic of the neural network of the subsequent graph.
S5: constructing a characteristic diagram: taking the half-edge spectrum in the S4 as the node characteristics of the graph neural network, and calculating Euclidean distances among the characteristics of each node; based on the K neighbor rule, the nearest K points in the Euclidean distance of each node are selected as the neighbor points of each node to form edge connection, and edge connection does not exist between non-neighbor points.
S6: and (3) performing feature learning by using graph convolution operation of the graph neural network, and constructing a Soft-max classifier at the last of the graph neural network to realize fault identification of the bearing.
In one embodiment, the step S3 specifically includes:
a set of high frequency wavelet coefficients and a set of low frequency wavelet coefficients are obtained using a discrete wavelet transform. Because the noise has high-frequency characteristics, the high-frequency wavelet coefficients are abandoned, and the low-frequency wavelet coefficients are reserved.
And the low-frequency wavelet coefficient is utilized to reconstruct the signal, and the reconstruction mode adopts a famous Mallat algorithm, so that a reconstructed signal based on the low-frequency wavelet coefficient reconstruction is obtained.
In one embodiment, the step S4 specifically includes:
and performing fast Fourier transform on the reconstructed signal (such as 1024 sampling points) to obtain equal-length (1024 point composition) spectrograms which are symmetrical about the center.
As the spectrogram is centrosymmetric, in order to avoid information redundancy, a half spectrum (1 st to 512 th points) is taken as a node feature of the neural network of the subsequent graph.
In one embodiment, the step S5 specifically includes:
and constructing a characteristic graph G, which mainly comprises the construction of two elements of nodes and edges.
And constructing node elements, wherein the node characteristics are formed by the half-edge spectrum characteristics in the S4, and the total node number is the sample to be detected.
And (4) constructing edge elements, and calculating Euclidean distances among the node characteristics. And for each node, selecting K points closest to the node in Euclidean distance as the neighbor points of the node based on a K neighbor rule. Adjacent points mutually form edge connection, and non-adjacent points do not have edge connection.
And setting the connection weight w to be 1 between nodes connected by the edges, otherwise, setting the connection weight w to be 0. The edge connection is a non-directional edge connection, and the feature map in this step is a non-directional feature map.
In one embodiment, the step S6 specifically includes:
and taking the feature graph constructed in the step S5 as the input of the graph neural network, and performing feature mining on the node and edge connection information based on multi-layer graph convolution operation.
The graph convolution operation, generally set to 3-5 times, can be self-defined according to the dimension of the node attribute. The specific flow of the graph convolution operation can be referred to the following documents:
D.I.Shuman,S.K.Narang,P.Frossard,A.Ortega,and P.Vandergheynst,“The emerging field of signal processing on graphs,”IEEE Signal Processing Magazine,vol.30,no.3,pp.83–98,2013.
in the graph convolution operation, during specific operation, the Chebyshev polynomial is used as a convolution kernel, so that the operation is simplified.
And after the last graph convolution operation, obtaining a fault identification result of the bearing by using a Soft-max classifier.
In one embodiment, the step S6 is preceded by:
the graph neural network needs to be trained before it can be used for the recognition task. The training process needs a sample of known fault labels with certain data to construct a feature map and train a neural network model of the map. The samples used for training are also obtained under the noise condition, so that the characteristic distribution of the bearing fault under the noise samples can be well learned by the graph network.
An apparatus for identifying bearing failure, comprising:
the data acquisition module is used for acquiring a vibration acceleration original signal of the bearing;
the sample construction module is used for carrying out normalization processing on the vibration acceleration signal to obtain a data sample, and constructing a continuous preset number of data samples into a sample to be detected;
and the characteristic denoising module is used for performing discrete wavelet transform and reconstruction on the sample to be detected and performing fast Fourier transform to obtain a half spectrum of the sample to be detected. When a plurality of samples to be tested are input, the samples to be tested are mutually processed without mutual coherence;
and the graph building module builds the node attribute and the edge connection attribute based on the denoised features to form a feature graph.
The graph analysis module is used for carrying out multilayer graph convolution operation on the characteristic graph, and obtaining a Soft-max classification result after the last layer of graph convolution layer, namely the result of bearing fault identification;
an apparatus for identifying bearing faults comprises a memory having a computer program stored therein and a processor implementing the steps of the method described above when executing the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method.
In general, compared with the prior art, the technical scheme of the invention has the following beneficial effects:
the method carries out identification and analysis on the vibration signal of the bearing based on the noise reduction characteristics and the graph neural network, and can well avoid the influence of external noise on data analysis, thereby improving the accuracy of model fault identification. Specifically, discrete wavelet transform is performed on a sample to be detected, signal reconstruction is performed by using a low-frequency wavelet coefficient of the sample, and then fast Fourier transform is performed on a reconstructed signal. And (3) constructing node attributes by taking a half-edge spectrum of a fast Fourier transform result, and determining adjacent points by using the Euclidean distance between nodes so as to determine edge connection and form a characteristic diagram. And taking the feature graph as an input of the graph neural network, and mining graph features through multilayer graph convolution operation. And finally, obtaining a bearing fault identification result by using a Soft-max classifier. Under the high-noise industrial environment, the method provided by the invention can effectively process the influence of noise on fault diagnosis, fully excavate fault characteristics and discover the existing fault. Thereby improving the reliability of the equipment, reducing the operation and maintenance cost and improving the production efficiency.
Drawings
FIG. 1 is a flow diagram of a bearing fault identification method under noisy conditions in one embodiment;
FIG. 2 is a schematic diagram of the signature graph constructed in step S5 in one embodiment;
FIG. 3 is a diagram of a confusion matrix after fault identification of a high noise sample in one embodiment;
FIG. 4 is a block diagram of a bearing fault identification apparatus according to an embodiment;
fig. 5 is a schematic view of the internal structure of the bearing failure recognition apparatus of an embodiment.
The same reference numbers will be used throughout the drawings to refer to the same or like elements or structures, wherein:
a data acquisition module 401, a sample construction module 402, a feature noise reduction module 403, a graph construction module 404, and a graph analysis module 405.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
FIG. 1 is a flow chart of a method of identifying a bearing fault under noisy conditions in one embodiment, in this embodiment the bearing is a rolling bearing. As shown in fig. 1, the fault recognition of the rolling bearing includes: step S1 to step S6.
S1: and acquiring a vibration acceleration signal of the bearing, namely the rolling bearing.
Specifically, the processor in the fault identification device of the present invention may be utilized to transmit a control signal to the acceleration sensor, so that the vibration acceleration sensor collects an original vibration signal of the bearing and transmits the original vibration signal to the processor. Namely, a processor is used for acquiring vibration acceleration signals.
S2: and normalizing the vibration acceleration original signals acquired in the step S1, and combining a continuous preset number of sampling points to construct a sample to be detected, wherein the number of the samples to be detected is usually multiple.
Specifically, the collected original vibration acceleration signals are normalized, a min-max normalization method is adopted, so that the numerical value of each sampling point is kept in the range of [0, 1], and each preset number of continuous sampling points is constructed into a sample to be measured; the preset number may be set to 256, 512, 1024, etc., or may be set to other values, which is not limited herein. In the present embodiment, the preset number of samples is set to 1024. In this example, a rolling bearing (model 6205-2RS JEM SKF) was used for the verification of the method of the present invention, and the data source of this example is Kaiser university, USA. With the eleven types of fault status data for the rolling bearing, inner ring faults (including three types of fault sizes 0.007, 0.014, 0.021 inches, identified by labels 1, 2, 3, respectively), outer ring faults (including three types of fault sizes 0.007, 0.014, 0.021, 0.028 inches, identified by labels 4, 5, 6, 7, respectively), and rolling element faults (including four types of fault sizes 0.007, 0.014, 0.021, 0.028 inches, identified by labels 8, 9, 10, 11, respectively). In order to verify the effectiveness of the method provided by the invention, 100 samples to be tested are constructed for each type of fault samples (the length of a sample sampling point is 1 × 1024), wherein 70 samples are training set samples used for model training, and 30 samples are testing set samples used for testing the model and the method. Thus, a total of 1100 samples of the sample set, 770 samples of the training set and 330 samples of the test set were formed. In order to verify the effectiveness of the method for identifying the noise sample faults, Gaussian noise is added to the samples in the whole sample set respectively, so that the signal-to-noise ratio of each sample is kept at about 5 dB.
S3: and performing discrete wavelet transform on the sample to be detected, and performing signal reconstruction by taking a low-frequency coefficient of wavelet decomposition of one layer to obtain a reconstructed signal.
Specifically, a discrete wavelet transform is performed on a sample to be measured to obtain a group of high-frequency wavelet coefficients and a group of low-frequency wavelet coefficients. In this embodiment, the sample to be measured includes 1024 sampling points, so after one discrete wavelet transform, the sizes of the two sets of wavelet coefficients are 512 × 1.
The high frequency wavelet coefficients are discarded because the noise tends to be high frequency. The low frequency wavelet coefficients are retained.
And reconstructing the low-frequency wavelet coefficient by using a Mallat algorithm to obtain a reconstructed signal with the size of 1024 multiplied by 1.
S4: and performing fast Fourier transform on the reconstructed signal in the S3 to obtain a spectrum signal which is equal to the reconstructed signal in length and is symmetrical about the center, and taking a half spectrum of the spectrum signal as the node characteristic of the neural network of the subsequent graph.
Specifically, the reconstructed signal of size 1024 × 1 of S3 is subjected to fast fourier transform, and the obtained spectrum signal size is also 1024 × 1.
The spectrum signal is symmetrical about the center, so that the half spectrum can represent the characteristic, and in this embodiment, the 1 st to 512 th points of the spectrum are uniformly taken.
S5: constructing a characteristic graph G: taking the half-edge spectrum in the S4 as the node characteristics of the graph neural network, and calculating Euclidean distances among the characteristics of each node; based on the K neighbor rule, the nearest K points in the Euclidean distance of each node are selected as the neighbor points of each node to form edge connection, and edge connection does not exist between non-neighbor points.
Specifically, the construction of the feature graph G comprises the construction of two elements, namely nodes and edges.
And constructing node elements, wherein the node features are formed by the half-edge spectrum features in the S4, and the node feature size is 512 multiplied by 1. The total node number is the sample to be measured.
And (4) constructing edge elements, and calculating Euclidean distances among the node characteristics. And for each node, selecting K points closest to the node in Euclidean distance as the neighbor points of the node based on a K neighbor rule. Adjacent points mutually form edge connection, and non-adjacent points do not have edge connection. The value of k can be set according to specific situations, and in the embodiment, the value of k is set to be 5.
And setting the connection weight w to be 1 between nodes connected by the edges, otherwise, setting the connection weight w to be 0. The edge connection is a non-directional edge connection, and the feature map in this step is a non-directional feature map.
S6: and (3) performing feature learning by using graph convolution operation of the graph neural network, and constructing a Soft-max classifier at the last of the graph neural network to realize fault identification of the bearing.
For the graph convolution operation, reference may be made to the following references:
D.I.Shuman,S.K.Narang,P.Frossard,A.Ortega,and P.Vandergheynst,“The emerging field of signal processing on graphs,”IEEE Signal Processing Magazine,vol.30,no.3,pp.83–98,2013.
in order to improve the speed and efficiency of graph convolution, the graph convolution operation is performed three times in this embodiment, and in the specific operation, a chebyshev (Cheb) polynomial is used as a convolution kernel, and ReLu is an activation function.
Specifically, the structural form of the graph neural network can be expressed as:
output＝softmax(Cheb(σ(Cheb(σ(Cheb(input,W(0))),W(1))),W(2)))
wherein σ is activation function, in this embodiment, the three convolution layers of the neural network all adopt ReLU activation function, W(0)，W(1)，W(2)The weight matrices for the first convolution, the second convolution and the third convolution are respectively shown.
Specifically, in this embodiment, the graph neural network includes 1 input layer and 3 graph convolution layers in total, and the detailed structural parameters are as follows: 512-300-200-11. The number of convolution kernels per layer of map convolution layer is set to 3.
Specifically, after the last graph convolution operation, a Soft-max classifier is used to perform fault identification on the features obtained after the last convolution of each node, so as to obtain class labels between 1 and 11, which respectively represent 11 faults of the rolling bearing in the embodiment.
By using the method of the present invention, fig. 3 shows the confusion matrix after the 330 test set samples are subjected to fault identification. In the field of machine learning, a confusion matrix (also called a probability table or an error matrix). It is a specific matrix used to present the visualization effect of the performance of the algorithm. Each column represents a prediction value and each row represents the actual category. As can be seen from fig. 3, all of the 330 test samples were correctly identified, and the accuracy rate was as high as 100%.
Fig. 4 is a block diagram showing a structure of a bearing failure recognition apparatus according to an embodiment. As shown in fig. 4, the fault recognition apparatus of a bearing includes: a data acquisition module 401, a sample construction module 402, a feature noise reduction module 403, a graph construction module 404, and a graph analysis module 405.
The data acquisition module 401 is used for acquiring a vibration acceleration original signal of the bearing;
a sample construction module 402, configured to perform normalization processing on the vibration acceleration signal to obtain a data sample, and construct a to-be-detected sample from a continuous preset number of data samples;
and the feature denoising module 403 is configured to perform discrete wavelet transform and reconstruction on the sample to be detected, and perform fast fourier transform to obtain a half spectrum of the sample. When a plurality of samples to be tested are input, the samples to be tested are mutually processed without mutual coherence;
the graph construction module 404 constructs node attributes and edge connection attributes based on the denoised features to form a feature graph.
And the graph analysis module 405 is configured to perform multilayer graph convolution operation on the feature graph, and obtain a Soft-max classification result after the last layer of graph convolution layer, that is, a bearing fault identification result.
The bearing fault recognition device provided by the application is based on a signal processing technology and a graph neural network model, combines respective advantages of a spectrum analysis technology and a deep learning technology, can effectively recognize bearing fault samples under a noise condition, and has a fault recognition accuracy rate of 100%. The method can effectively identify the existing faults of the bearing under the noise condition, and has strong practical value and popularization significance.
In one embodiment, the feature denoising module is configured to perform discrete wavelet transform and reconstruction on the sample to be detected, and perform fast fourier transform to obtain a half-edge spectrum of the sample. And performing one-time discrete wavelet transform, reserving a low-frequency wavelet coefficient and performing signal reconstruction. And then, obtaining a spectrogram by utilizing fast Fourier transform, taking a half spectrum of the spectrogram, and outputting the half spectrum to a graph construction module as a node characteristic.
In one embodiment, the graph construction module is configured to perform node construction on the half-edge spectrum features output by the feature noise reduction module, so as to construct node elements. And determining K neighbor points of each node by using a K neighbor rule according to the Euclidean distance between the nodes, constructing edge connection between the neighbor points, forming a non-directional characteristic diagram, and inputting the characteristic diagram into a diagram analysis module.
In one embodiment, the graph analysis module is used for carrying out deep feature mining and fault identification on the input feature graph. And performing feature extraction on the node information and the side information of the input feature graph by using multilayer convolution operation, and identifying faults by using a Soft-max classifier after the last convolution operation so that the graph neural network can finally output bearing fault information.
The division of each module in the bearing fault recognition device under the noise condition is only used for illustration, and in other embodiments, the bearing fault recognition device may be divided into different modules as required to complete all or part of the functions of the bearing fault recognition device.
For the specific definition of the bearing fault identification device, reference may be made to the above definition of the bearing fault identification method, which is not described herein again. The modules in the bearing fault identification device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
Fig. 5 is a schematic view of the internal structure of the bearing failure recognition apparatus in one embodiment. As shown in fig. 5, the bearing failure recognition apparatus includes a processor and a memory connected by a system bus. Wherein the processor is used for providing calculation and control capability and supporting the operation of the whole bearing fault identification device. The memory may include a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The computer program can be executed by a processor for implementing a bearing fault identification method provided in the following embodiments. The internal memory provides a cached execution environment for the operating system computer programs in the non-volatile storage medium.
The implementation of each module in the bearing fault identification apparatus provided in the embodiment of the present application may be in the form of a computer program. The computer program may be run on a bearing fault identification device. The program modules of the computer program may be stored on a memory of the bearing fault identification device. Which when executed by a processor, performs the steps of the method described in the embodiments of the present application.
The embodiment of the application also provides a computer readable storage medium. One or more non-transitory computer-readable storage media containing computer-executable instructions that, when executed by one or more processors, cause the processors to perform the steps of the bearing fault identification method. A computer program product containing instructions which, when run on a computer, cause the computer to perform a bearing fault identification method.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (7)
1. A method of identifying a bearing fault, comprising:
s1: collecting a vibration acceleration signal of a bearing;
s2: carrying out normalization processing on an original signal, and constructing a continuous preset number of sampling points into a sample to be measured;
s3: performing discrete wavelet transform on a sample to be detected, and performing signal reconstruction by taking a low-frequency coefficient of wavelet decomposition to obtain a reconstructed signal;
s4: performing fast Fourier transform on the reconstructed signal in S3 to obtain a frequency spectrum signal which is equal to the reconstructed signal in length and is symmetrical about the center, and taking a half spectrum of the frequency spectrum signal as a node characteristic of a neural network of a subsequent graph;
s5: constructing a characteristic diagram: taking the half-edge spectrum in the S4 as the node characteristics of the graph neural network, and calculating Euclidean distances among the characteristics of each node; based on a K neighbor rule, selecting the K nearest points on the Euclidean distance of each node as neighbor points of each node to form edge connection, wherein edge connection does not exist between non-neighbor points;
s6: performing feature learning by using graph convolution operation of a graph neural network, and constructing a Soft-max classifier at the end of the graph neural network to realize fault identification of the bearing;
the step S3 specifically includes:
a group of high-frequency wavelet coefficients and a group of low-frequency wavelet coefficients can be obtained by utilizing one-time discrete wavelet transform; because the noise has high-frequency characteristics, high-frequency wavelet coefficients are abandoned, and low-frequency wavelet coefficients are reserved;
performing signal reconstruction by using the low-frequency wavelet coefficient, wherein a reconstruction mode adopts a Mallat algorithm, so that a reconstructed signal based on low-frequency wavelet coefficient reconstruction is obtained;
the step S4 specifically includes:
performing fast Fourier transform on the reconstructed signal to obtain a spectrogram with equal length, wherein the spectrogram is symmetrical about the center;
because the spectrogram is centrosymmetric, in order to avoid information redundancy, a half spectrum of the spectrogram is taken as a node feature of a neural network of a subsequent graph.
2. The method according to claim 1, wherein the step S5 specifically includes:
constructing a characteristic graph G, which mainly comprises the construction of two elements of nodes and edges;
constructing node elements, wherein the node characteristics are formed by the half-edge spectrum characteristics in S4, and the total node number is the sample to be detected;
constructing edge elements, and calculating Euclidean distances among the node characteristics; for each node, based on a K neighbor rule, selecting K points nearest to the node in Euclidean distance as neighbor points of the node; adjacent points mutually form edge connection, and non-adjacent points do not have edge connection;
setting the connection weight w as 1 between nodes with edge connection, otherwise, setting the weight w as 0; the edge connection is a non-directional edge connection, and the feature map in this step is a non-directional feature map.
3. The method according to claim 1, wherein the step S6 specifically includes:
taking the characteristic graph constructed in the S5 as the input of a graph neural network, and carrying out characteristic mining on node and edge connection information based on multi-layer graph convolution operation;
the graph convolution operation is set to be 3-5 times and can be defined according to the dimension of the node attribute;
in the graph convolution operation, during specific operation, a Chebyshev polynomial is used as a convolution kernel, so that the operation is simplified;
and after the last graph convolution operation, obtaining a fault identification result of the bearing by using a Soft-max classifier.
4. The method of claim 1, wherein the step S6 is preceded by:
the graph neural network needs to be trained before being used for the recognition task; in the training process, a sample of known fault labels with certain data is needed to construct a characteristic diagram, and a diagram neural network model is trained; the samples used for training are also obtained under noise conditions, so that the graph network is guaranteed to learn the characteristic distribution of the bearing fault under the noise samples.
5. An apparatus for identifying a bearing failure, comprising:
the data acquisition module is used for acquiring a vibration acceleration original signal of the bearing;
the sample construction module is used for carrying out normalization processing on the vibration acceleration signal to obtain a data sample, and constructing a continuous preset number of data samples into a sample to be detected;
the characteristic denoising module is used for carrying out discrete wavelet transform and reconstruction on the sample to be detected and carrying out fast Fourier transform to obtain a half spectrum of the sample to be detected; when a plurality of samples to be tested are input, the samples to be tested are mutually processed without mutual coherence;
the graph building module builds a node attribute and an edge connection attribute based on the denoised features to form a feature graph;
and the graph analysis module is used for carrying out multilayer graph convolution operation on the characteristic graph, and obtaining a Soft-max classification result after the last layer of graph convolution layer, namely the result of bearing fault identification.
6. An apparatus for identifying bearing faults, comprising a memory storing a computer program and a processor implementing the method of claims 1 to 4 when executing the computer program.
7. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111524116.1A CN114186593B (en) | 2021-12-14 | 2021-12-14 | Bearing fault identification method, device, equipment and storage medium under noise condition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111524116.1A CN114186593B (en) | 2021-12-14 | 2021-12-14 | Bearing fault identification method, device, equipment and storage medium under noise condition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114186593A true CN114186593A (en) | 2022-03-15 |
| CN114186593B CN114186593B (en) | 2024-09-24 |
Family
ID=80543582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111524116.1A Active CN114186593B (en) | 2021-12-14 | 2021-12-14 | Bearing fault identification method, device, equipment and storage medium under noise condition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114186593B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982573A (en) * | 2023-03-20 | 2023-04-18 | 东莞市杰达机械有限公司 | Multifunctional feeder and control method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108076058A (en) * | 2017-12-14 | 2018-05-25 | 北京博大光通物联科技股份有限公司 | Heterogeneous network communications protocol method for amalgamation processing based on self-adapting changeable long codes |
| CN109633289A (en) * | 2018-12-28 | 2019-04-16 | 集美大学 | A kind of red information detecting method of electromagnetism based on cepstrum and convolutional neural networks |
| WO2020228141A1 (en) * | 2019-05-13 | 2020-11-19 | 清华大学 | Electromagnetic signal identification method and device for constructing graph convolutional network on basis of implicit knowledge |
| CN112651167A (en) * | 2020-12-02 | 2021-04-13 | 杭州电子科技大学 | Semi-supervised rolling bearing fault diagnosis method based on graph neural network |
-
2021
- 2021-12-14 CN CN202111524116.1A patent/CN114186593B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108076058A (en) * | 2017-12-14 | 2018-05-25 | 北京博大光通物联科技股份有限公司 | Heterogeneous network communications protocol method for amalgamation processing based on self-adapting changeable long codes |
| CN109633289A (en) * | 2018-12-28 | 2019-04-16 | 集美大学 | A kind of red information detecting method of electromagnetism based on cepstrum and convolutional neural networks |
| WO2020228141A1 (en) * | 2019-05-13 | 2020-11-19 | 清华大学 | Electromagnetic signal identification method and device for constructing graph convolutional network on basis of implicit knowledge |
| CN112651167A (en) * | 2020-12-02 | 2021-04-13 | 杭州电子科技大学 | Semi-supervised rolling bearing fault diagnosis method based on graph neural network |
Non-Patent Citations (1)
| Title |
|---|
| 李恒;张氢;秦仙蓉;孙远韬;: "基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法", 振动与冲击, no. 19, 15 October 2018 (2018-10-15) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982573A (en) * | 2023-03-20 | 2023-04-18 | 东莞市杰达机械有限公司 | Multifunctional feeder and control method thereof |
| CN115982573B (en) * | 2023-03-20 | 2023-11-17 | 东莞市杰达机械有限公司 | Multifunctional feeder and control method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114186593B (en) | 2024-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110849627B (en) | Width migration learning network and rolling bearing fault diagnosis method based on same | |
| CN110059601B (en) | Intelligent fault diagnosis method for multi-feature extraction and fusion | |
| CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
| WO2023137807A1 (en) | Rolling bearing class imbalance fault diagnosis method and system | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN109946080B (en) | Mechanical equipment health state identification method based on embedded circulation network | |
| CN113111820B (en) | Rotary part fault diagnosis method and device based on improved CNN and relation module | |
| Tang et al. | A robust deep learning network for low-speed machinery fault diagnosis based on multikernel and RPCA | |
| CN112633098A (en) | Fault diagnosis method and system for rotary machine and storage medium | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN111753891A (en) | Rolling bearing fault diagnosis method based on unsupervised feature learning | |
| CN114091504B (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| CN114487129B (en) | Flexible material damage identification method based on acoustic emission technology | |
| CN110553839A (en) | Single and composite fault diagnosis method, equipment and system for gearbox | |
| CN113865872B (en) | Bearing fault diagnosis method based on wavelet packet reconstruction imaging and CNN | |
| CN112146882B (en) | Bearing fault diagnosis method based on transfer learning vibration signal image recognition | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN115905976B (en) | Attention mechanism-based fault diagnosis method, system and equipment for Highway Bi-LSTM bearing | |
| CN114462480A (en) | Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set | |
| CN116168720A (en) | Motor sound abnormality fault diagnosis method, system and storable medium | |
| CN114186593B (en) | Bearing fault identification method, device, equipment and storage medium under noise condition | |
| Sun et al. | Application of deep belief networks for precision mechanism quality inspection | |
| CN111382792A (en) | Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation | |
| GB2625165A (en) | Method and system for fault diagnosis of rolling bearing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |