CN114459760B - Rolling bearing fault diagnosis method and system in strong noise environment - Google Patents

Rolling bearing fault diagnosis method and system in strong noise environment Download PDF

Info

Publication number
CN114459760B
CN114459760B CN202111672334.XA CN202111672334A CN114459760B CN 114459760 B CN114459760 B CN 114459760B CN 202111672334 A CN202111672334 A CN 202111672334A CN 114459760 B CN114459760 B CN 114459760B
Authority
CN
China
Prior art keywords
layer
cnn
feature extractor
senet
fault
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.)
Active
Application number
CN202111672334.XA
Other languages
Chinese (zh)
Other versions
CN114459760A (en
Inventor
陆宝春
吴连申
翁朝阳
叶邵鹏
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.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
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 Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202111672334.XA priority Critical patent/CN114459760B/en
Publication of CN114459760A publication Critical patent/CN114459760A/en
Application granted granted Critical
Publication of CN114459760B publication Critical patent/CN114459760B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Acoustics & Sound (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a rolling bearing fault diagnosis method and a system under a strong noise environment, which adopt deep learning to carry out bearing fault diagnosis, and a neural network model consists of a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a softmax layer. The normalized training data is input into a neural network model, the random sampling layer is used for randomly sampling the training data to increase the anti-interference performance of the model, a CNN global type feature extractor combined with SENet is used for extracting global type fault features, a CNN-based high-dimensional feature extractor is used for abstracting the fault features into higher-dimensional features, a GRU-based feature classifier is used for enhancing feature expression capability, and a full-connection layer and a Softmax layer are used for obtaining fault category diagnosis results. The method can overcome noise interference and diagnose rolling bearing signals under various loads.

Description

Rolling bearing fault diagnosis method and system in strong noise environment
Technical Field
The invention belongs to the technical field of rolling bearing fault diagnosis, and particularly relates to a rolling bearing fault diagnosis method and system in a strong noise environment.
Background
Rolling bearings, which are one of the most core components of rotary machines, produce a significant economic loss in the event of serious failure. With the rapid development of the sensing technology, people can record the running condition of the bearing well, but how to mine fault information in the mass data becomes a research hotspot in the field of bearing fault diagnosis.
In recent years, with the rapid development of deep learning technology, the characteristic representation and learning capability are greatly improved, and even an end-to-end use mode appears, namely, an output result can be obtained by inputting an original signal to a neural network model or only adopting an original signal processed by a simple preprocessing and data expansion method. In addition to convenience, and more importantly, this end-to-end approach exhibits very strong performance far more practical than most manual feature extraction methods.
Although most of the current bearing fault diagnosis methods based on deep learning have high accuracy, most of them do not consider the complex factors of strong noise and multiple loads. Although some factors of strong noise are considered, bearing fault diagnosis is performed only under a certain load; the invention aims to provide a rolling bearing fault diagnosis method and a system suitable for a high-noise environment, which can overcome noise interference and diagnose rolling bearing signals under various loads.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a method and a system for diagnosing faults of a rolling bearing in a strong noise environment, which can improve the fault diagnosis performance of the rolling bearing in a strong noise and multi-load composite environment.
The technical solution for realizing the purpose of the invention is as follows:
a rolling bearing fault diagnosis method under a strong noise environment comprises the following steps:
S1, collecting vibration acceleration signals x i,j of the bearing without faults and with different fault types and different fault degrees under different loads at a sampling frequency f s;
S2: normalizing the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal Setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
S3: constructing a neural network model, comprising: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer; the normalized sample sequentially passes through a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier and a full-connection layer, and is finally output through a Softmax layer to obtain a result label; the random sampling layer is used for adding training data interfered by analog noise; the CNN global type feature extractor combined with SENet is used for extracting global type fault features; the CNN-based high-dimensional feature extractor is used for abstracting high-dimensional fault features from the extracted global fault features; the GRU-based feature classifier is used for clustering the features extracted by the CNN-based high-dimensional feature extractor;
S4: the normalized vibration acceleration signal processed in the step S2 And inputting the corresponding training label into the neural network model in the step S3 for training;
s5: collecting vibration data of a rolling bearing of the current equipment by adopting the same sampling frequency f s in the step S1 to obtain a vibration acceleration signal X i,j to be detected;
S6: according to step S2, normalizing the vibration acceleration signal X i,j to be tested to obtain normalized test data Normalized test data/>And (3) inputting the current bearing fault state into the trained model in the step S4.
A rolling bearing fault diagnosis system in a high noise environment, comprising:
The signal acquisition module is used for acquiring vibration acceleration signals x i,j of the bearing without faults, with different fault types and with different fault degrees under different loads at the sampling frequency f s;
The normalization processing module is used for performing normalization processing on the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal
The training label setting module is used for setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
A neural network training module, comprising: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer; the normalized sample sequentially passes through a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier and a full-connection layer, and is finally output through a Softmax layer to obtain a result label; the random sampling layer is used for adding training data interfered by analog noise; the CNN global type feature extractor combined with SENet is used for extracting global type fault features; the CNN-based high-dimensional feature extractor is used for abstracting high-dimensional fault features from the extracted global fault features; the GRU-based feature classifier is used for clustering features extracted by the CNN-based high-dimensional feature extractor.
Compared with the prior art, the invention has the remarkable advantages that:
1. The invention can automatically learn fault information from the original vibration signal without complex denoising and artificial feature selection processes.
2. According to the invention, the training data is randomly sampled, so that the anti-noise interference capability can be greatly improved.
3. The invention can efficiently obtain important global features by adopting the CNN global feature extractor combined with SENet.
4. The invention adopts the GRU-based feature classifier to process the features extracted by the CNN feature extractor, so that the loss of the associated information of the original vibration signal can be prevented.
5. The invention can realize stable and accurate fault diagnosis of the rolling bearing under the strong noise and multi-load composite environment by automatically adjusting the distribution of the test data.
Drawings
FIG. 1 is a flowchart of an embodiment of a method and system for diagnosing a rolling bearing failure in a high noise environment of the present invention.
Fig. 2 is a diagram of a neural network model structure of the present invention.
Fig. 3 is a manifold representation of a random sampling layer of the present invention.
Fig. 4 is a specific structural diagram of SENet in fig. 2.
Fig. 5 is an external view of the rolling bearing test stand in the present embodiment.
Fig. 6 is a graph showing the comparison of the diagnostic results of the present invention and five comparison methods in a strong noise and multi-load composite environment in this example.
Fig. 7 is a graph showing the results of the visualization experiments of the present example, which shows the mapping of the original signal, CNN1 layer, SENet layer, CNN3 layer, GRU layer, and FC layer in order from 7 (a-f).
Detailed Description
The following description of the embodiments of the invention is presented in conjunction with the accompanying drawings to provide a better understanding of the invention to those skilled in the art. It is to be expressly noted that in the description below, detailed descriptions of known functions and designs are omitted here as perhaps obscuring the present invention.
FIG. 1 is a flowchart of an embodiment of a method and system for diagnosing a rolling bearing failure in a high noise environment according to the present invention. As shown in FIG. 1, the method for diagnosing the faults of the rolling bearing in the high-noise and multi-load composite environment comprises the following specific steps:
S1, collecting vibration acceleration signals x i,j of the bearing without faults and with different fault types and different fault degrees under different loads at a sampling frequency f s;
S2: normalizing the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal Setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
s3: constructing a neural network model;
The neural network model structure diagram of the invention is shown in fig. 2, and comprises six parts: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer, and the following components are described.
1. The random sampling layer is a random sampling that employs a bernoulli distribution subject to a probability p, where p takes a random value between 0.5 and 1 per training period. To maintain the data length, the data that is not sampled is set to 0 in place. The step is based on two aspects, on one hand, part of information is actively lost during training, so that the effect of simulating noise interference is achieved, and the noise interference resistance of the whole network is greatly improved. Manifold theory, on the other hand, considers that if low-dimensional data can map out high-dimensional data through a manifold, it can represent the characteristics of the data itself. Similarly, bearing fault information is abstractly represented by a low-dimensional manifold as shown in fig. 3, a point x on a real line is original training data, and the neural network can completely predict test data x'. But if it is desired to predict the test data to which noise is addedIt is necessary to widen the range of the original training data x so as to cover the test data/>, to which noise is addedBy this disruption of the random sampling of the raw data, new training data/>, mapped from the raw training data x, can be obtainedThe range of the mapping is in a dashed circle as shown in fig. 3. Continuously trained through neural networks, noise-added test data/>, within manifold range represented by dashed linesCan be predicted, thereby greatly improving the robustness of the whole neural network.
2. The enhanced CNN feature extractor in combination with SENet consists of one CNN feature extractor and one SENet in sequence.
2.1: The CNN feature extractor consists of a convolutional layer, a AdaBN layer, and an ELU activation layer in that order. A large-sized convolution kernel is employed to increase the field of view of the network, thereby extracting periodic fault characteristics and suppressing high frequency noise. The method AdaBN is adopted to process the characteristics extracted by the convolution layer so as to reduce the difference between the distribution of test data and training data, accelerate the convergence rate of the model and reduce the training time of the model. Finally, the ELU activation function is used, so that gradient disappearance is avoided, and meanwhile, the input negative half-axis information is saved, and further, more comprehensive information can be obtained in the subsequent steps.
2.1.1: The AdaBN layer can perform standardization operation on the convolution layer result, and the standardization formula is as follows:
Wherein: x (k) is AdaBN layer input, gamma (k)、β(k) is AdaBN layer scaling and biasing parameters, and y (k) is AdaBN layer output, wherein gamma (k)、β(k) is an autonomous learning training parameter in training mode, and the parameter is used to update test data in test mode So that test data/>Even due to noise effects, it is associated with training data/>In different distributions, good prediction results can be obtained.
2.1.2: The ELU function is:
wherein: x is the input of the ELU function and a is a positive super parameter, the saturation of the negative input value can be adjusted, so that the network has stronger robustness. The gradient vanishing problem can be solved to a certain extent in the non-negative interval of the input like the ReLU function, but the gradient vanishing problem is more excellent than the ReLU function, the gradient vanishing problem can utilize the information of the negative half axis of the input x of the ELU function, and the whole output of the function tends to be 0, so that the convergence speed of the network is increased.
2.2: SENet as shown in FIG. 4, which is composed of a global average pooling layer, a full connection layer, an ELU activation layer, a full connection layer and a Sigmoid activation layer in this order. The SENet blocks are adopted to carry out channel feature enhancement on the features extracted by the CNN feature extractor, thereby automatically enhancing beneficial features and inhibiting useless features so as to achieve the effect of global feature extraction. The input to SENet is the result of the CNN feature extractor output, and the output of the enhanced CNN feature extractor in combination with SENet is the result of the multiplication of the output of SENet with the CNN feature extractor output.
3. The CNN-based high-dimensional feature extractor consists of a plurality of CNN feature extractors, two of which are employed in the examples below. Each CNN feature extractor consists of a convolutional layer, a AdaBN layer, an ELU activation layer, and a max pooling layer in sequence.
4. The GRU-based feature classifier is formed by stacking a plurality of GRU layers, and the last time step of the last GRU layer is used as the output of the GRU-based feature classifier. Because the GRU can extract the information related to each other, the GRU is used for processing the characteristics generated by the CNN, and the related information of the original data can be better captured. Compared with a network consisting of a single CNN, the network combined by the CNN and the GRU can work stably under the interference of strong noise better.
5. The fully connected layer is a linear neuron layer that maps high-dimensional information to probabilities for each failure state.
6. The Softmax layer takes the maximum value of the fault state probabilities output by the full connection layer out as a fault diagnosis result.
S4: the normalized vibration acceleration signal processed in the step S2And inputting the corresponding training label into the neural network model in the step S3 for training;
s5: collecting vibration data of a rolling bearing of the current equipment by adopting the same sampling frequency f s in the step S1 to obtain a vibration acceleration signal X i,j to be detected;
S6: according to step S2, normalizing the vibration acceleration signal X i,j to be tested to obtain normalized test data Normalized test data/>And (3) inputting the current bearing fault state into the trained model in the step S4.
Based on the above method, the embodiment further provides a rolling bearing fault diagnosis system under a strong noise environment, including:
The signal acquisition module is used for acquiring vibration acceleration signals x i,j of the bearing without faults, with different fault types and with different fault degrees under different loads at the sampling frequency f s;
The normalization processing module is used for performing normalization processing on the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal
The training label setting module is used for setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
A neural network training module, comprising: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer; the normalized sample sequentially passes through a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier and a full-connection layer, and is finally output through a Softmax layer to obtain a result label; the random sampling layer is used for adding training data interfered by analog noise; the CNN global type feature extractor combined with SENet is used for extracting global type fault features; the CNN-based high-dimensional feature extractor is used for abstracting high-dimensional fault features from the extracted global fault features; the GRU-based feature classifier is used for clustering features extracted by the CNN-based high-dimensional feature extractor.
The steps of setting and method of each layer of the neural network are described above, and are not repeated here.
In order to better illustrate the technical effects of the invention, the invention is experimentally verified by adopting a specific embodiment. This experiment verifies that using the bearing test data of kesi Chu Da (CWRU), electrical discharge machining was used to provide a failure for the motor bearing, a failure of 0.007 to 0.040 inch diameter was introduced into the inner race, the rolling elements (i.e., balls) and the outer race, respectively, the failed bearing was reinstalled into the test motor, and vibration data for motor loads of 0 to 3 horsepower (motor speed 1797 to 1720 RPM) were recorded. Fig. 5 is an external view of the rolling bearing test stand of the present embodiment, including a motor of 1.5KW (left side of the figure), a torque sensor/encoder (middle joint of the figure) and a power tester (right side of the figure).
This example uses data acquired at a frequency of 12Hz, with one sample at every 2048 points. In order to make the training sample more sufficient, the training data is expanded by adopting a data enhancement technology in the embodiment, so that deep learning can learn more fault characteristics, and the overall performance is improved. The data enhancement method adopted by the embodiment is an overlap sampling method, so that training samples are increased, and fault information of the edges of the samples can be better learned. It should be noted that the data set must be divided into training data and test data before data enhancement can be performed in the training data, otherwise, there is partially repeated data in the training data and the test data. To verify the noise immunity of the proposed model, additive gaussian white noise was added to the test data to simulate random noise in an industrial environment. In order to simulate the real environment as much as possible, the present example mixes the data under different loads of 1-3 hp with each other, in addition to adding-10 dB of white gaussian noise to the data set. The final training data is the data of the multi-load working condition without noise, and the test data is the data of the multi-load working condition with Gaussian white noise of-10 dB. The number of training data and test data and the type of failure are shown in table 1.
TABLE 1
Adam is selected as a training optimizer in the experiment, and the initial learning rate is 0.005. The learning rate step-down strategy is adopted for rapidly improving the training precision of the model: the learning rate was reduced by 20% every 50 cycles for a total of 500 cycles. The specific structural parameters of AnNet are shown in table 2.
TABLE 2
TABLE 3 Table 3
This example demonstrates the superiority of the proposed method and will be compared with existing methods that are somewhat similar to the present invention. The methods are as follows: AAnNet based on CNN-GRU (see document Jin G,Zhu T,Akram M W,et al.An Adaptive Anti-Noise Neural Network for Bearing Fault Diagnosis Under Noise and Varying Load Conditions[J].IEEE Access,2020,8:74793-74807.), based on WDCNN of AdaBN (see document Wei Z,Peng G,Li C,et al.A New Deep Learning Model for Fault Diagnosis with Good Anti-Noise and Domain Adaptation Ability on Raw Vibration Signals[J].Sensors,2017,17(3):425.) and CAT-GRU based on channel attention (see document Z.Wang,H.Wang,Z.Liu and J.Liu,"Rolling Bearing Fault Diagnosis Using CNN-based Attention Modules and Gated Recurrent Unit,"2020Global Reliability and Prognostics and Health Management(PHM-Shanghai),2020,pp.1-6). experiment run 10 times independently, calculate the mean and variance of the 10 experimental results).
The experimental result of the neural network model prediction provided by the invention under the mixed environment of strong noise and multiple loads is shown as figure 6, the accuracy of the neural network model prediction is highest, meanwhile, the prediction with high accuracy can be carried out by most methods for tiny Gaussian white noise interference, and along with the increase of Gaussian white noise, only the slowest attenuation of the method provided by the invention is achieved until the Gaussian white noise is at-6 dB, the accuracy is 94.0%, and the variance of ten independent tests is 0.020, so that the stability is also very good.
In order to intuitively understand the processing effect of each network layer, the processing results of each layer of the bearing fault signal can be mapped by the method through the t-SNE dimension reduction technology. For rapid processing, the embodiment adopts a principal component analysis method to reduce the dimension of the high-dimensional feature from the high-dimensional feature to 100 dimensions, and then adopts a t-SNE algorithm to map the 100-dimensional feature to a two-dimensional plane so as to fully show the feature processing effect of each network layer.
The output of the layers of the neural network was visualized using the t-SNE dimension reduction technique with the test data processed by the-4 dB additive Gaussian white noise as input to the network, the results of which are shown in FIGS. 7 (a-f). There are 10 different data points in the graph, and the values of the data points are in one-to-one correspondence with table 1. It can be seen from the figure that the original signals are mixed together in the feature space, and then as the network layers of AnNET are processed, the output features of each layer start to be similar and close to each other, and the dissimilarities are separated from each other. This means that the proposed network is able to extract useful features from the original signal, and to distinguish between different types of bearing faults under high noise and variable load conditions.
While the foregoing describes illustrative embodiments of the present invention to facilitate an understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, but is to be construed as protected by the accompanying claims insofar as various changes are within the spirit and scope of the present invention as defined and defined by the appended claims.

Claims (8)

1. The rolling bearing fault diagnosis method in the strong noise environment is characterized by comprising the following steps:
S1, collecting vibration acceleration signals x i,j of the bearing without faults and with different fault types and different fault degrees under different loads at a sampling frequency f s;
S2: normalizing the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal Setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
S3: constructing a neural network model, comprising: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer; the normalized sample sequentially passes through a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier and a full-connection layer, and is finally output through a Softmax layer to obtain a result label; the random sampling layer is used for adding training data interfered by analog noise; the CNN global type feature extractor combined with SENet is used for extracting global type fault features; the CNN-based high-dimensional feature extractor is used for abstracting high-dimensional fault features from the extracted global fault features; the GRU-based feature classifier is used for clustering the features extracted by the CNN-based high-dimensional feature extractor;
The CNN global feature extractor combined with SENet consists of one CNN feature extractor and one SENet; the CNN extractor is composed of a convolution layer, a AdaBN layer and an ELU activation layer in sequence, wherein AdaBN layers can perform standardization operation on the convolution layer result, and a standardization formula is as follows:
Wherein: x (k) is AdaBN layer input, gamma (k)、β(k) is AdaBN layer scaling and biasing parameters, and y (k) is AdaBN layer output, wherein gamma (k)、β(k) is an autonomous learning training parameter in training mode, and the parameter is used to update test data in test mode Is a distribution of (3); SENet is composed of a global average pooling layer, a full connection layer, an ELU activation layer, a full connection layer and a Sigmoid activation layer in sequence; the input to SENet is the output of the CNN feature extractor, and the output of the CNN global feature extractor in combination with SENet is the product of SENet and the output of the CNN feature extractor;
S4: the normalized vibration acceleration signal processed in the step S2 And inputting the corresponding training label into the neural network model in the step S3 for training;
s5: collecting vibration data of a rolling bearing of the current equipment by adopting the same sampling frequency f s in the step S1 to obtain a vibration acceleration signal X i,j to be detected;
S6: according to step S2, normalizing the vibration acceleration signal X i,j to be tested to obtain normalized test data Normalized test data/>And (3) inputting the current bearing fault state into the trained model in the step S4.
2. The method according to claim 1, wherein the random sampling layer in step S3 is a random sampling using a bernoulli distribution with probability p, where p takes a random value between 0.5 and 1 in each training period, and the data that is not sampled is set to 0 in place for maintaining the data length.
3. The rolling bearing fault diagnosis method according to claim 1, wherein the CNN-based high-dimensional feature extractor in step S3 is composed of a plurality of CNN feature extractors, each CNN feature extractor being composed of a convolution layer, a AdaBN layer, an ELU activation layer, and a max pooling layer in sequence.
4. The rolling bearing fault diagnosis method according to claim 1, wherein the GRU-based feature classifier in step S3 is formed by stacking a plurality of GRU layers, and a last time step of a last GRU layer is used as an output of the GRU-based feature classifier.
5. A rolling bearing failure diagnosis system in a high noise environment, comprising:
The signal acquisition module is used for acquiring vibration acceleration signals x i,j of the bearing without faults, with different fault types and with different fault degrees under different loads at the sampling frequency f s;
The normalization processing module is used for performing normalization processing on the collected vibration acceleration signal x i,j to obtain a normalized vibration acceleration signal
The training label setting module is used for setting a training label for deep learning according to the fault type and the fault degree as the normalized vibration acceleration signal;
A neural network training module, comprising: the system comprises a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier, a full-connection layer and a Softmax layer; the normalized sample sequentially passes through a random sampling layer, a CNN global feature extractor combined with SENet, a CNN-based high-dimensional feature extractor, a GRU-based feature classifier and a full-connection layer, and is finally output through a Softmax layer to obtain a result label; the random sampling layer is used for adding training data interfered by analog noise; the CNN global type feature extractor combined with SENet is used for extracting global type fault features; the CNN-based high-dimensional feature extractor is used for abstracting high-dimensional fault features from the extracted global fault features; the GRU-based feature classifier is used for clustering the features extracted by the CNN-based high-dimensional feature extractor;
The CNN global feature extractor combined with SENet consists of one CNN feature extractor and one SENet; the CNN extractor is composed of a convolution layer, a AdaBN layer and an ELU activation layer in sequence, wherein AdaBN layers can perform standardization operation on the convolution layer result, and a standardization formula is as follows:
Wherein: x (k) is AdaBN layer input, gamma (k)、β(k) is AdaBN layer scaling and biasing parameters, and y (k) is AdaBN layer output, wherein gamma (k)、β(k) is an autonomous learning training parameter in training mode, and the parameter is used to update test data in test mode Is a distribution of (3); SENet is composed of a global average pooling layer, a full connection layer, an ELU activation layer, a full connection layer and a Sigmoid activation layer in sequence; the input to SENet is the output of the CNN feature extractor, and the output of the CNN global type feature extractor in combination with SENet is the product of SENet and the output of the CNN feature extractor.
6. The system of claim 5, wherein the random sampling layer is a random sampling using a bernoulli distribution with probability p, where p takes a random value between 0.5 and 1 per training period, and the non-sampled data is set to 0 in place for maintaining the data length.
7. The rolling bearing fault diagnosis system in a high noise environment of claim 5, wherein the CNN-based high-dimensional feature extractor is composed of a plurality of CNN feature extractors, each CNN feature extractor being composed of a convolution layer, a AdaBN layer, an ELU activation layer and a max pooling layer in sequence.
8. The rolling bearing fault diagnosis system in a high noise environment according to claim 5, wherein the GRU-based feature classifier is formed by stacking a plurality of GRU layers, and the last time step of the last GRU layer is used as the output of the GRU-based feature classifier.
CN202111672334.XA 2021-12-31 2021-12-31 Rolling bearing fault diagnosis method and system in strong noise environment Active CN114459760B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111672334.XA CN114459760B (en) 2021-12-31 2021-12-31 Rolling bearing fault diagnosis method and system in strong noise environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111672334.XA CN114459760B (en) 2021-12-31 2021-12-31 Rolling bearing fault diagnosis method and system in strong noise environment

Publications (2)

Publication Number Publication Date
CN114459760A CN114459760A (en) 2022-05-10
CN114459760B true CN114459760B (en) 2024-05-24

Family

ID=81407593

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111672334.XA Active CN114459760B (en) 2021-12-31 2021-12-31 Rolling bearing fault diagnosis method and system in strong noise environment

Country Status (1)

Country Link
CN (1) CN114459760B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115420490A (en) * 2022-09-01 2022-12-02 重庆大学 Mechanical fault intelligent diagnosis method based on relationship migration domain generalization network

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109029999A (en) * 2018-09-19 2018-12-18 河北工业大学 Fault Diagnosis of Roller Bearings based on enhanced modulation double-spectrum analysis
CN110595775A (en) * 2019-09-19 2019-12-20 电子科技大学 Rolling bearing fault diagnosis method based on multi-branch multi-scale convolutional neural network
CN110647867A (en) * 2019-10-09 2020-01-03 中国科学技术大学 Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network
CN110896381A (en) * 2019-11-25 2020-03-20 中国科学院深圳先进技术研究院 Deep neural network-based traffic classification method and system and electronic equipment
CN112001314A (en) * 2020-08-25 2020-11-27 江苏师范大学 Early fault detection method for variable speed hoist
KR102189269B1 (en) * 2019-10-22 2020-12-09 경북대학교 산학협력단 Fault Diagnosis method and system for induction motor using convolutional neural network
CN112284735A (en) * 2020-10-21 2021-01-29 兰州理工大学 Multi-sensor rolling bearing fault diagnosis based on one-dimensional convolution and dynamic routing
CN112577748A (en) * 2020-12-07 2021-03-30 东南大学 Reinforced lightweight multi-scale CNN-based rolling bearing fault diagnosis method
CN113776834A (en) * 2021-10-11 2021-12-10 山东大学 Rolling bearing fault diagnosis method and system based on discrete cosine cycle spectrum coherence

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109029999A (en) * 2018-09-19 2018-12-18 河北工业大学 Fault Diagnosis of Roller Bearings based on enhanced modulation double-spectrum analysis
CN110595775A (en) * 2019-09-19 2019-12-20 电子科技大学 Rolling bearing fault diagnosis method based on multi-branch multi-scale convolutional neural network
CN110647867A (en) * 2019-10-09 2020-01-03 中国科学技术大学 Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network
KR102189269B1 (en) * 2019-10-22 2020-12-09 경북대학교 산학협력단 Fault Diagnosis method and system for induction motor using convolutional neural network
CN110896381A (en) * 2019-11-25 2020-03-20 中国科学院深圳先进技术研究院 Deep neural network-based traffic classification method and system and electronic equipment
CN112001314A (en) * 2020-08-25 2020-11-27 江苏师范大学 Early fault detection method for variable speed hoist
CN112284735A (en) * 2020-10-21 2021-01-29 兰州理工大学 Multi-sensor rolling bearing fault diagnosis based on one-dimensional convolution and dynamic routing
CN112577748A (en) * 2020-12-07 2021-03-30 东南大学 Reinforced lightweight multi-scale CNN-based rolling bearing fault diagnosis method
CN113776834A (en) * 2021-10-11 2021-12-10 山东大学 Rolling bearing fault diagnosis method and system based on discrete cosine cycle spectrum coherence

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于卷积神经网络和双向长短期记忆的稳定抗 噪声滚动轴承故障诊断;陈晓雷等;吉林大学学报（ 工学版）;20220228;第52卷(第2期);296-309 *
基于自适应变尺度频移带通随机共振降噪的EMD 多频微弱信号检测;韩东颖等;机械工程学报;20130430;第49卷(第8期);10-18 *

Also Published As

Publication number Publication date
CN114459760A (en) 2022-05-10

Similar Documents

Publication Publication Date Title
CN110849626B (en) Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system
CN111797567B (en) Bearing fault classification method and system based on deep learning network
CN110595775A (en) Rolling bearing fault diagnosis method based on multi-branch multi-scale convolutional neural network
CN113008559B (en) Bearing fault diagnosis method and system based on sparse self-encoder and Softmax
CN113723489A (en) Rolling bearing fault identification method for improved relation network
CN113112000B (en) Construction method and application of multilayer convolutional neural network model
CN116304861A (en) Time-frequency characteristic fusion fault diagnosis method based on self-attention
Xu et al. Hierarchical multiscale dense networks for intelligent fault diagnosis of electromechanical systems
CN114459760B (en) Rolling bearing fault diagnosis method and system in strong noise environment
Shaowu et al. Study on the health condition monitoring method of hydraulic pump based on convolutional neural network
Peng et al. Research on fault diagnosis method of rolling bearing based on 2DCNN
Chen et al. An adversarial learning framework for zero-shot fault recognition of mechanical systems
CN111783941B (en) Mechanical equipment diagnosis classification method based on probability confidence convolutional neural network
CN117628005A (en) Signal-fused hydraulic motor fault diagnosis method and system
Lu et al. An automated bearing fault diagnosis using a self-normalizing convolutional neural network
CN116256174A (en) Noise robust bearing fault diagnosis method based on interpretable multilayer wavelet drive
CN114235409B (en) Rolling bearing multi-user cooperative intelligent fault diagnosis method for light weight communication
Cheng et al. A rolling bearing fault diagnosis model based on WCNN-BiGRU
CN115587289A (en) Multi-source domain adaptive network-based aeroengine fault identification method
CN115905976A (en) Method, system and equipment for diagnosing high way Bi-LSTM bearing fault based on attention mechanism
CN115979660A (en) Filter fault diagnosis method for internal combustion engine based on MPA optimization
He et al. Fault Identification of Rotating Machinery Based on Dynamic Feature Reconstruction Signal Graph
Li et al. Intelligent CNN-based Fault Diagnosis of Rotating Machinery with Small Fault Samples
Jiang et al. Image classification method of bearing fault based on BOA optimization ResNet-18
CN116625689B (en) Rolling bearing fault diagnosis method and system based on SMDER

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