CN110647867B - Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network - Google Patents
Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network Download PDFInfo
- Publication number
- CN110647867B CN110647867B CN201910957332.1A CN201910957332A CN110647867B CN 110647867 B CN110647867 B CN 110647867B CN 201910957332 A CN201910957332 A CN 201910957332A CN 110647867 B CN110647867 B CN 110647867B
- Authority
- CN
- China
- Prior art keywords
- neural network
- data set
- network model
- training
- noise
- 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
Links
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
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Computational Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Pure & Applied Mathematics (AREA)
- Molecular Biology (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Acoustics & Sound (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a bearing fault diagnosis method and a system based on a self-adaptive anti-noise neural network, wherein the method comprises the following steps: acquiring a data set, and dividing the data set into a training data set and a testing data set according to a preset proportion; constructing a neural network model, wherein the neural network model comprises the following steps: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; training the constructed neural network model based on training data in a noise-free training data set to obtain a trained neural network model; and adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result. The method can automatically extract the features from the original signals without manually selecting the features and denoising, can distinguish different fault types and fault severity degrees, and can adapt to different noise environments.
Description
Technical Field
The invention relates to the technical field of bearing faults, in particular to a bearing fault diagnosis method and system based on a self-adaptive anti-noise neural network.
Background
The diagnosis of mechanical equipment failure, which causes economic loss and casualties, is a widespread concern in modern industries. Rolling bearings are important targets for fault diagnosis of mechanical equipment, and especially in rotating mechanical equipment, rolling bearing faults account for a large proportion of faults. In the past decades, the diagnosis of faults in rolling bearings has been extensively studied. The data-driven-based method is a common method in bearing fault diagnosis and is mainly divided into a diagnosis method based on signal analysis and a diagnosis method based on deep learning.
The bearing fault diagnosis method based on signal analysis mainly comprises the following steps: fourier analysis, wavelet transform, cepstrum, empirical mode decomposition, and the like. These diagnostic methods require expert knowledge for signal preprocessing and feature selection, and are less effective in complex environmental conditions.
The traditional bearing fault diagnosis method based on deep learning comprises the following steps: the system comprises an artificial neural network, a support vector machine, a stacked self-encoder, a deep Boltzmann machine, a deep belief network and the like, and has poor diagnosis effect in a complex environment, particularly a noise environment.
In recent years, with the development of deep neural network training optimization algorithms, various deep neural networks are beginning to be widely researched. CNNs (Convolutional Neural Networks) and RNNs (Recurrent Neural Networks) are the most widely used deep learning Networks at present, and have been widely used in various tasks including bearing fault diagnosis. CNNs use convolution filters to extract local features, and RNNs are widely used in time series data or sequence modeling to capture time-related features from signals. However, when the bearing fault diagnosis is carried out by early experiments of CNNs or RNNs, complicated manual operation is required, so that the fault diagnosis of the bearing under the complex noise environment is still a challenging task, and particularly under the condition of high noise, the adaptability of a neural network to different noise levels still needs to be improved.
Therefore, how to effectively diagnose the bearing fault is an urgent problem to be solved.
Disclosure of Invention
In view of the above, the invention provides a bearing fault diagnosis method based on a self-adaptive anti-noise neural network, which can automatically extract features from an original signal without manually selecting the features and denoising, can distinguish different fault types and fault severity degrees, and can adapt to different noise environments.
The invention provides a bearing fault diagnosis method based on a self-adaptive anti-noise neural network, which comprises the following steps:
acquiring a data set, and dividing the data set into a training data set and a testing data set according to a preset proportion;
constructing a neural network model, wherein the neural network model comprises: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
training the constructed neural network model based on training data in a noise-free training data set to obtain a trained neural network model;
and adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
Preferably, the acquiring the data set and dividing the data set into a training data set and a testing data set according to a preset proportion includes:
acquiring a data set, and enabling the data set to be as follows: the scale of 2 is divided into a training data set and a test data set.
Preferably, two of the convolutional layers act as feature extractors using the ELU as an activation function.
Preferably, the training data in the noise-free training data set is used to train the constructed neural network model to obtain a trained neural network model, and the training includes:
inputting the signals of the training data in the noise-free training data set in a random sampling mode;
setting a hyper-parameter, a weight initialization mode and a loss function of a neural network model;
and training the neural network model by using a random gradient descent method and a back propagation algorithm to obtain the trained neural network model.
Preferably, the hyper-parameters comprise: learning rate, attenuation rate and training times of the neural network model.
A bearing fault diagnostic system based on an adaptive noise immune neural network, comprising:
the acquisition module is used for acquiring a data set and dividing the data set into a training data set and a testing data set according to a preset proportion;
a building module configured to build a neural network model, wherein the neural network model comprises: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
the training module is used for training the built neural network model based on training data in the noiseless training data set to obtain a trained neural network model;
and the diagnosis module is used for adding additive white Gaussian noise into the test data in the test data set and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
Preferably, when the obtaining module executes obtaining of a data set and divides the data set into a training data set and a testing data set according to a preset proportion, the obtaining module is specifically configured to:
acquiring a data set, and enabling the data set to be as follows: the scale of 2 is divided into a training data set and a test data set.
Preferably, the two convolutional layers act as feature extractors using the ELU as an activation function.
Preferably, when the training module executes training data in the noise-free training data set to train the constructed neural network model, so as to obtain a trained neural network model, the training module is specifically configured to:
inputting the signals of the training data in the noise-free training data set in a random sampling mode;
setting a hyper-parameter, a weight initialization mode and a loss function of a neural network model;
and training the neural network model by using a random gradient descent method and a back propagation algorithm to obtain the trained neural network model.
Preferably, the hyper-parameters comprise: learning rate, attenuation rate and training times of the neural network model.
In summary, the invention discloses a bearing fault diagnosis method based on a self-adaptive anti-noise neural network, when a bearing fault needs to be diagnosed, a data set is firstly obtained, the data set is divided into a training data set and a testing data set according to a preset proportion, and then a neural network model is constructed, wherein the neural network model comprises: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; training the constructed neural network model based on training data in the noise-free training data set to obtain a trained neural network model; and adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result. The neural network obtained through training can automatically extract features from the original signal without manually selecting the features and denoising, can distinguish different fault types and fault severity degrees, and can adapt to different noise environments.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flowchart of a bearing fault diagnosis method based on an adaptive anti-noise neural network according to embodiment 1 of the present disclosure;
FIG. 2 is a flowchart of a method of embodiment 2 of a bearing fault diagnosis method based on an adaptive anti-noise neural network disclosed in the present invention;
FIG. 3 is a schematic structural diagram of an embodiment 1 of a bearing fault diagnosis system based on an adaptive anti-noise neural network, which is disclosed by the invention;
FIG. 4 is a schematic structural diagram of an embodiment 2 of a bearing fault diagnosis system based on an adaptive anti-noise neural network disclosed in the present invention;
FIG. 5 is a schematic diagram of a CWRU data set selected by a neural network according to the present disclosure;
FIG. 6 is a schematic diagram of a neural network structure disclosed in the present invention;
FIG. 7 is a schematic diagram of an attention mechanism in a neural network according to the present disclosure;
fig. 8 is a schematic diagram of specific structural parameters of the neural network disclosed in the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, which is a flowchart of a method of embodiment 1 of a bearing fault diagnosis method based on an adaptive anti-noise neural network disclosed in the present invention, the method may include the following steps:
s101, acquiring a data set, and dividing the data set into a training data set and a testing data set according to a preset proportion;
when bearing faults need to be detected, firstly, a data set is obtained and divided into a training data set and a testing data set according to a preset proportion. And preprocessing and amplitude normalization are carried out on signals in the data set, and a data input strategy for improving the anti-noise performance of the network by adopting random sampling is adopted.
S102, constructing a neural network model, wherein the neural network model comprises the following steps: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
then, a neural network model is established, and the typical structure of the neural network model mainly comprises an input layer, two convolution layers, two GRU layers, an attention mechanism layer and a DNN layer. The neural network model takes a signal with a fixed length as an input, and realizes random sampling of input data by using dropout at an input layer. After the input layers, the two convolutional layers act as feature extractors using the ELU as an activation function. After each convolution layer, there is a batch normalization layer and a dropout layer, respectively, for stability and versatility. Then, the CNN part is followed by two GRU layers, and then another dropout layer. Note that the mechanism layer follows the dropout layer. Finally, the classification result is generated and stabilized by the two fully connected layers and the softmax layer.
S103, training the constructed neural network model based on training data in the noise-free training data set to obtain a trained neural network model;
then, training all neural network models on the noise-free original data, testing the data added with different noise levels, inputting the original signals of the training data in the noise-free training data set in a random sampling mode, and performing model training to obtain the trained neural network models.
And S104, adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
And finally, adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
In summary, in the above embodiment, when a bearing fault needs to be diagnosed, a data set is first obtained, and the data set is divided into a training data set and a testing data set according to a preset proportion, and then a neural network model is constructed, where the neural network model includes: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; training the constructed neural network model based on training data in the noise-free training data set to obtain a trained neural network model; and adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result. The neural network obtained through training can automatically extract features from the original signal without manually selecting the features and denoising, can distinguish different fault types and fault severity degrees, and can adapt to different noise environments.
As shown in fig. 2, which is a flowchart of a method of embodiment 2 of the bearing fault diagnosis method based on the adaptive anti-noise neural network disclosed in the present invention, the method may include the following steps:
s201, acquiring a data set, and enabling the data set to be as 8:2 into a training data set and a test data set;
this example uses two different data sets, one from the bearing data center at the university of Keiss Sige (CWRU) and the second from the bearing fault diagnosis test platform QPZZ-II. Through bearing fault diagnosis experiments, four types of bearing vibration signals are collected. The two data sets were divided into a training data set and a test data set on an 8:2 scale.
Laboratory table for CWRU data set as shown in fig. 5, the vibration signal used was collected from the b-side of the accelerometer driver at a 48000Hz sampling rate. The motor load is 3hp, and the motor rotating speed is changed between 1719 rpm and 1729rpm under different bearing health states. In the embodiment, the average rotating speed of the motor is 1724rpm, and the driving end bearing is a deep groove ball bearing (6205-2RS JEM SKF). The test bearings have four states: normal state, inner ring fault, rolling element fault and outer ring fault. Each fault condition has three types of fault diameters, 7mils, 14mils, and 21mils (1 mil-0.001 inch), which are produced by electrical discharge machining. Thus, there is a normal condition and three fault types, each with three severity levels.
The signal of the data set is preprocessed, since the input length of the algorithm is a fixed value, while the vibration signal is periodic, it is necessary to segment the signal according to the period and normalize the amplitude of the input signal to [ -1, +1 ].
S202, constructing a neural network model, wherein the neural network model comprises the following steps: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
as shown in fig. 6, the typical structure of the neural network model is mainly composed of one input layer, two convolutional layers, two GRU layers, an attention mechanism layer, and a DNN layer. The neural network model takes a signal with a fixed length as an input, and realizes random sampling of input data by using dropout at an input layer. After the input layers, the two convolutional layers act as feature extractors, using the ELU as an activation function, the formula is as follows:
after each convolution layer, there is a batch normalization layer and a dropout layer, respectively, for stability and versatility. Then, following the CNN part by two GRU layers, a Gated Recursion Unit (GRU) is a modified RNN that can capture the long-term dependence of the gate structure. These gates are used to delete or add information to the hidden state to decide whether to remember long term dependencies or to use short term information. The work flow of the GRU is as follows: reset gate r at the t-th time steptThe formula is as follows:
rt=σ(Wrxt+Urht-1+br);
wherein xtAs an input vector, WrAnd UrAs a weight matrix, brTo bias the weights, ht-1Is the previous activation. σ is a logistic sigmoid function. Reset door rtDetermining how many past activations to retain in computing candidate activations
Wherein |, indicates an element multiplication. Candidate activationFrom an input vector xtAnd a preceding step ht-1Calculated by resetting the gate rtAnd (5) modulating.
zt=σ(Wzxt+Uzht-1+bz);
Updating the door ztControlling the ratio of past activations and candidate activations to calculate a new activation:
the GRU layer is followed by another dropout layer. Note that the mechanism layer follows the dropout layer.
The attention mechanism obtains a final result by adopting a weighting method for each time step, and the utilization rate of information is improved. Note that the main calculation process of the institution is shown in FIG. 7, and that the weight vectors are calculated as follows: given all hidden source statesCurrent target state htCalculating an attention weight vector at:
Wherein the function score compares the current target hidden state htAnd per source hidden stateThe contribution of each time step to the final output is calculated, as:
wherein, WSIs a trainable weighting matrix. Then, by noticing the weight vector atAnd each stepComputing context vector ctI.e., weighted time step vector:
finally, the context vector ctWith the target hidden state htConnected to obtain an output attention-hiding stateWherein WcFor a trainable weighting matrix:
finally, the classification result is generated and stabilized by the two fully connected layers and the softmax layer. Specific structural parameters are shown in fig. 8.
The Dropout rate is fixed to 0.5. The convolution kernel size is 128 x 1 and the number of first and second layers of filters is 64 and 72, respectively. The step size of the convolutional layer is 1, and no padding is used for the convolutional layer. GRU first layer neurons 64, second layer neurons 128. There are 128 neurons in the hidden state vector of note. Then two fully connected layers, 64 and 32 neurons respectively. Finally, a softmax layer with 10 neurons.
S203, inputting a signal of training data in a noise-free training data set in a random sampling mode, setting a hyper-parameter, a weight initialization mode and a loss function of a neural network model, and training the neural network model by using a random gradient descent method and a back propagation algorithm to obtain a trained neural network model;
training all neural network models on noiseless original data, testing data added with different noise levels, inputting original signals of training data in the noiseless training data set in a random sampling mode, setting hyper-parameters such as learning rate, attenuation rate and training times of the neural network models, and using an Adam optimization algorithm as an optimizer. And (3) adopting a power attenuation learning rate, setting an initial value of 0.0001 and a weight of 0.9, setting a weight initialization mode, setting a loss function of the network model, training the model by using a back propagation algorithm, and storing the trained model weight.
And S204, adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
Preparing a test data set, adding additive white Gaussian noise to the test data in the test data set, inputting the test data into a trained neural network model, simulating different noise conditions, and finally measuring by a signal-to-noise ratio (SNR), wherein the SNR is defined as the ratio of the power of a meaningful signal to the power of background noise [63 ]. The decibel form (dB) of the signal-to-noise ratio is expressed as:
wherein, P is the average power of the periodic signal c (t) in the period t, and is defined as:
when the signal is in discrete form, the average power can be calculated as:
for a signal with zero mean and known variance, the power can be measured by the varianceIndicating that, therefore, for standard normally distributed noise, the power is 1. Thus, the power of the original signal is first calculated, and then the noise signal P generated at the desired signal-to-noise ratio is calculatednoiseOf the power of (c). Finally, an additive white gaussian noise is generated by the following formula and then added to the original signal to make the signal have a desired signal-to-noise ratio. The standard normally distributed noise can be generated by the following equation:
wherein randn represents a function of a random number or matrix that produces a standard normal distribution;
and finally, obtaining a bearing fault diagnosis result.
In summary, the neural network architecture of the present invention mainly comprises four parts: random sample data input, CNN-based enhanced feature extractor, GRU-based feature classifier, attention-based mechanism and DNN-based feature post-processing. The input signal adopts a simple random sampling strategy, so that the adaptability of the network to noise is improved. In the feature extraction part, the convolutional layer adopts an exponential linear unit activation function and a dropout to improve the adaptability of the neural network under the noise condition. The function of the feature extractor is similar to that of a time-domain frequency-domain converter, a noise reduction filter and a feature selector, the feature extractor is similar to that of the time-domain frequency-domain converter, the noise reduction filter, the feature selector and the like, features can be automatically extracted from noise signals without manual feature selection and denoising processes, and the feature extractor has strong adaptability to noise. The features conveyed by the CNN part are further learned and classified using the famous GRU as a decoder or classifier when processing time-varying sequences. Finally, using the attention mechanism and DNN component as a post processor, appropriate features are selected from the GRU output as the final classification result. The neural network can distinguish not only different fault types, but also severity of corresponding faults; the method has stronger processing adaptability to signals with large noise and better universality to noise signals with different grades.
As shown in fig. 3, which is a schematic structural diagram of an embodiment 1 of the bearing fault diagnosis system based on the adaptive anti-noise neural network disclosed in the present invention, the system may include:
an obtaining module 301, configured to obtain a data set, and divide the data set into a training data set and a testing data set according to a preset ratio;
when bearing faults need to be detected, firstly, a data set is obtained and divided into a training data set and a testing data set according to a preset proportion. And preprocessing and amplitude normalization are carried out on signals in the data set, and a data input strategy for improving the anti-noise performance of the network by adopting random sampling is adopted.
A building module 302, configured to build a neural network model, where the neural network model includes: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
then, a neural network model is established, and the typical structure of the neural network model mainly comprises an input layer, two convolution layers, two GRU layers, an attention mechanism layer and a DNN layer. The neural network model takes a signal with a fixed length as an input, and realizes random sampling of input data by using dropout at an input layer. After the input layers, the two convolutional layers act as feature extractors using the ELU as an activation function. After each convolution layer, there is a batch normalization layer and a dropout layer, respectively, for stability and versatility. Then, the CNN part is followed by two GRU layers, and then another dropout layer. Note that the mechanism layer follows the dropout layer. Finally, the classification result is generated and stabilized by the two fully connected layers and the softmax layer.
The training module 303 is configured to train the constructed neural network model based on training data in the noise-free training data set to obtain a trained neural network model;
then, training all neural network models on the noise-free original data, testing the data added with different noise levels, inputting the original signals of the training data in the noise-free training data set in a random sampling mode, and performing model training to obtain the trained neural network models.
And the diagnosis module 304 is used for adding additive white gaussian noise into the test data in the test data set, inputting the additive white gaussian noise into the trained neural network model, and obtaining a bearing fault diagnosis result.
And finally, adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result.
In summary, in the above embodiments, when a bearing fault needs to be diagnosed, a data set is first obtained, and the data set is divided into a training data set and a testing data set according to a preset proportion, and then a neural network model is constructed, where the neural network model includes: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; training the constructed neural network model based on training data in the noise-free training data set to obtain a trained neural network model; and adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result. The neural network obtained through training can automatically extract features from the original signal without manually selecting the features and denoising, can distinguish different fault types and fault severity degrees, and can adapt to different noise environments.
As shown in fig. 4, which is a schematic structural diagram of an embodiment 2 of the bearing fault diagnosis system based on the adaptive anti-noise neural network disclosed in the present invention, the system may include:
an obtaining module 401, configured to obtain a data set, and compare the data set with a data set of 8:2 into a training data set and a test data set;
this example uses two different data sets, one from the bearing data center at the university of Keiss Sige (CWRU) and the second from the bearing fault diagnosis test platform QPZZ-II. Through bearing fault diagnosis experiments, four types of bearing vibration signals are collected. The two data sets were divided into a training data set and a test data set on an 8:2 scale.
Laboratory table for CWRU data set as shown in fig. 5, the vibration signal used was collected from the b-side of the accelerometer driver at a 48000Hz sampling rate. The motor load is 3hp, and the motor rotating speed is changed between 1719 rpm and 1729rpm under different bearing health states. In the embodiment, the average rotating speed of the motor is 1724rpm, and the driving end bearing is a deep groove ball bearing (6205-2RS JEM SKF). The test bearings have four states: normal state, inner ring fault, rolling element fault and outer race fault. Each fault condition has three types of fault diameters, 7mils, 14mils, and 21mils (1 mil-0.001 inch), which are produced by electrical discharge machining. Thus, there is a normal condition and three fault types, each with three severity levels.
The signal of the data set is preprocessed, since the input length of the algorithm is a fixed value, while the vibration signal is periodic, it is necessary to segment the signal according to the period and normalize the amplitude of the input signal to [ -1, +1 ].
A building module 402, configured to build a neural network model, where the neural network model includes: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer;
as shown in fig. 6, the typical structure of the neural network model is mainly composed of one input layer, two convolutional layers, two GRU layers, an attention mechanism layer, and a DNN layer. The neural network model takes a signal with a fixed length as an input, and realizes random sampling of input data by using dropout at an input layer. After the input layers, the two convolutional layers act as feature extractors, using the ELU as an activation function, the formula is as follows:
after each convolution layer, there is a batch normalization layer and a dropout layer, respectively, for stability and versatility. However, the device is not suitable for use in a kitchenLater, two GRU layers follow the CNN part, and the Gated Recursion Unit (GRU) is a modified RNN that can capture the long-term dependence of the gate structure. These gates are used to delete or add information to the hidden state to decide whether to remember long term dependencies or to use short term information. The work flow of the GRU is as follows: reset gate r at the t-th time steptThe formula is as follows:
rt=σ(Wrxt+Urht-1+br);
wherein xtAs an input vector, WrAnd UrAs a weight matrix, brTo bias the weights, ht-1Is the previous activation. σ is a logistic sigmoid function. Reset door rtDetermining how many past activations to retain in computing candidate activations
Wherein |, indicates an element multiplication. Candidate activationFrom an input vector xtAnd a preceding step ht-1Calculated by resetting the gate rtAnd (5) modulating.
zt=σ(Wzxt+Uzht-1+bz);
Updating the door ztControlling the ratio of past activations and candidate activations to calculate a new activation:
the GRU layer is followed by another dropout layer. Note that the mechanism layer follows the dropout layer.
The attention mechanism obtains a final result by adopting a weighting method for each time step, and the utilization rate of information is improved. Note that the main calculation process of the institution is shown in FIG. 7, and that the weight vectors are calculated as follows: given all hidden source statesCurrent target state htCalculating an attention weight vector at:
Wherein the function score compares the current target hidden state htAnd per source hidden stateThe contribution of each time step to the final output is calculated, as:
wherein, WSIs a trainable weighting matrix. Then, by noticing the weight vector atAnd each stepComputing context vector ctI.e., weighted time step vector:
finally, the context vector ctWith the target hidden state htConnected to obtain an output attention-hiding stateWherein WcFor a trainable weighting matrix:
finally, the classification result is generated and stabilized by the two fully connected layers and the softmax layer. Specific structural parameters are shown in fig. 8.
The Dropout rate is fixed at 0.5. The convolution kernel size is 128 x 1 and the number of first and second layers of filters is 64 and 72, respectively. The step size of the convolutional layer is 1, and no padding is used for the convolutional layer. GRU first layer neurons 64, second layer neurons 128. There are 128 neurons in the hidden state vector of note. Then two fully connected layers, 64 and 32 neurons respectively. Finally, a softmax layer with 10 neurons.
A training module 403, configured to input a signal of training data in a noise-free training data set in a random sampling manner, set a hyper-parameter, a weight initialization manner, and a loss function of a neural network model, train the neural network model by using a stochastic gradient descent method and a back propagation algorithm, and obtain a trained neural network model;
training all neural network models on noiseless original data, testing data added with different noise levels, inputting original signals of training data in the noiseless training data set in a random sampling mode, setting hyper-parameters such as learning rate, attenuation rate and training times of the neural network models, and using an Adam optimization algorithm as an optimizer. And (3) adopting a power attenuation learning rate, setting an initial value of 0.0001 and a weight of 0.9, setting a weight initialization mode, setting a loss function of the network model, training the model by using a back propagation algorithm, and storing the trained model weight.
And the diagnosis module 404 is configured to add additive white gaussian noise to the test data in the test data set, and input the test data to the trained neural network model to obtain a bearing fault diagnosis result.
Preparing a test data set, adding additive white Gaussian noise to the test data in the test data set, inputting the test data into a trained neural network model, simulating different noise conditions, and finally measuring by a signal-to-noise ratio (SNR), wherein the SNR is defined as the ratio of the power of a meaningful signal to the power of background noise [63 ]. The decibel form (dB) of the signal-to-noise ratio is expressed as:
wherein, P is the average power of the periodic signal c (t) in the period t, and is defined as:
when the signal is in discrete form, the average power can be calculated as:
for a signal with zero mean and known variance, the power can be measured by the varianceIndicating that, therefore, for standard normally distributed noise, the power is 1. Thus, the power of the original signal is first calculated, and then the noise signal P generated at the desired signal-to-noise ratio is calculatednoiseOf the power of (c). Finally, an additive white gaussian noise is generated by the following formula and then added to the original signal to make the signal have a desired signal-to-noise ratio. The standard normally distributed noise can be generated by:
wherein randn represents a function of a random number or matrix that produces a standard normal distribution;
and finally, obtaining a bearing fault diagnosis result.
In summary, the neural network architecture of the present invention mainly comprises four parts: random sample data input, CNN-based enhanced feature extractor, GRU-based feature classifier, attention-based mechanism and DNN-based feature post-processing. The input signal adopts a simple random sampling strategy, so that the adaptability of the network to noise is improved. In the feature extraction part, the convolutional layer adopts an exponential linear unit activation function and a dropout to improve the adaptability of the neural network under the noise condition. The function of the feature extractor is similar to that of a time-domain frequency-domain converter, a noise reduction filter and a feature selector, the feature extractor is similar to that of the time-domain frequency-domain converter, the noise reduction filter, the feature selector and the like, features can be automatically extracted from noise signals without manual feature selection and denoising processes, and the feature extractor has strong adaptability to noise. The features conveyed by the CNN part are further learned and classified using the famous GRU as a decoder or classifier when processing time-varying sequences. Finally, using the attention mechanism and DNN component as a post processor, appropriate features are selected from the GRU output as the final classification result. The neural network can distinguish not only different fault types, but also severity of corresponding faults; the method has stronger processing adaptability to signals with large noise and better universality to noise signals with different grades.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A bearing fault diagnosis method based on an adaptive anti-noise neural network is characterized by comprising the following steps:
acquiring a data set, and dividing the data set into a training data set and a testing data set according to a preset proportion;
constructing a neural network model, wherein the neural network model comprises: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; the input layer realizes random sampling of input data by using dropout;
training the constructed neural network model based on training data in a noise-free training data set to obtain a trained neural network model;
adding additive white Gaussian noise into the test data in the test data set, and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result;
wherein, the attention mechanism obtains the final result by adopting a weighting method for each time step, and pays attention to the weight vector atThe calculation formula of (a) is as follows:
h istExpressed as a current target hidden state, saidRepresented as a hidden source state; the above-mentionedThe function score compares the current target hidden state htAnd each hidden source stateCalculating the contribution of each time step to the final output; the W isSExpressed as a trainable weight matrix;
weighted time step vector ctThe calculation formula of (a) is as follows:
2. The method of claim 1, wherein the obtaining the data set and the dividing the data set into a training data set and a testing data set according to a preset ratio comprises:
acquiring a data set, and enabling the data set to be as follows: the scale of 2 is divided into a training data set and a test data set.
3. The method of claim 1, wherein two of the convolutional layers act as feature extractors using ELUs as activation functions.
4. The method of claim 1, wherein training the constructed neural network model based on the training data in the noise-free training data set to obtain a trained neural network model comprises:
inputting the signals of the training data in the noise-free training data set in a random sampling mode;
setting a hyper-parameter, a weight initialization mode and a loss function of a neural network model;
and training the neural network model by using a random gradient descent method and a back propagation algorithm to obtain the trained neural network model.
5. The method of claim 4, wherein the hyper-parameters comprise: learning rate, attenuation rate and training times of the neural network model.
6. A bearing fault diagnostic system based on an adaptive noise immune neural network, comprising:
the acquisition module is used for acquiring a data set and dividing the data set into a training data set and a testing data set according to a preset proportion;
a building module configured to build a neural network model, wherein the neural network model comprises: the system comprises an input layer, two convolutional layers, two gating recursion unit layers, an attention mechanism layer and a deep neural network layer; the input layer realizes random sampling of input data by using dropout;
the training module is used for training the built neural network model based on training data in the noiseless training data set to obtain a trained neural network model;
the diagnosis module is used for adding additive white Gaussian noise into the test data in the test data set and inputting the test data into the trained neural network model to obtain a bearing fault diagnosis result;
wherein, the attention mechanism obtains the final result by adopting a weighting method for each time step, and pays attention to the weight vector atThe calculation formula of (a) is as follows:
h istExpressed as a current target hidden state, saidRepresented as a hidden source state; the above-mentionedThe function score compares the current target hidden state htAnd each hidden source stateCalculating the contribution of each time step to the final output; the W isSExpressed as a trainable weight matrix;
weighted time step vector ctThe calculation formula of (a) is as follows:
7. The system of claim 6, wherein the obtaining module, when executing obtaining the data set and dividing the data set into the training data set and the testing data set according to a preset ratio, is specifically configured to:
acquiring a data set, and enabling the data set to be as follows: the scale of 2 is divided into a training data set and a test data set.
8. The system of claim 6, wherein the two convolutional layers act as feature extractors using ELUs as activation functions.
9. The system according to claim 6, wherein the training module, when executing the training data in the noise-free training data set to train the constructed neural network model, is specifically configured to:
inputting the signals of the training data in the noise-free training data set in a random sampling mode;
setting a hyper-parameter, a weight initialization mode and a loss function of a neural network model;
and training the neural network model by using a random gradient descent method and a back propagation algorithm to obtain the trained neural network model.
10. The system of claim 9, wherein the hyper-parameters comprise: learning rate, attenuation rate and training times of the neural network model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910957332.1A CN110647867B (en) | 2019-10-09 | 2019-10-09 | Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910957332.1A CN110647867B (en) | 2019-10-09 | 2019-10-09 | Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110647867A CN110647867A (en) | 2020-01-03 |
CN110647867B true CN110647867B (en) | 2022-05-13 |
Family
ID=69012501
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910957332.1A Active CN110647867B (en) | 2019-10-09 | 2019-10-09 | Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110647867B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111579243B (en) * | 2020-06-17 | 2021-04-20 | 大连理工大学 | Rolling bearing intelligent diagnosis system based on deep migration learning |
CN112149730B (en) * | 2020-09-22 | 2022-06-10 | 中南大学 | High-speed train bearing fault prediction method based on oil sample vision |
CN112629854B (en) * | 2020-11-25 | 2022-08-05 | 西安交通大学 | Bearing fault classification method based on neural network attention mechanism |
CN112839349B (en) * | 2020-12-25 | 2022-10-25 | 朗坤智慧科技股份有限公司 | Angular domain stationary signal denoising method and device based on 5G technology and network server |
CN113203566B (en) * | 2021-04-06 | 2022-09-16 | 上海吞山智能科技有限公司 | Motor bearing fault diagnosis method based on one-dimensional data enhancement and CNN |
CN113343855B (en) * | 2021-06-09 | 2022-09-16 | 西南交通大学 | Rolling bearing fault diagnosis system and method based on guide type sub-field self-adaption |
CN113473514B (en) * | 2021-09-06 | 2021-11-16 | 国网江苏省电力有限公司信息通信分公司 | Electric power wireless private network fault diagnosis model training method, diagnosis method and apparatus |
CN113761225A (en) * | 2021-09-09 | 2021-12-07 | 昆明理工大学 | Automobile engine fault prediction method fusing knowledge graph and multivariate neural network model |
CN114383844B (en) * | 2021-12-15 | 2023-06-23 | 太原理工大学 | Rolling bearing fault diagnosis method based on distributed deep neural network |
CN114459760B (en) * | 2021-12-31 | 2024-05-24 | 南京理工大学 | Rolling bearing fault diagnosis method and system in strong noise environment |
CN115929495B (en) * | 2022-11-30 | 2024-05-14 | 昆明理工大学 | Engine valve fault diagnosis method based on Markov transition field and improved Gaussian prototype network |
CN115984801A (en) * | 2023-03-07 | 2023-04-18 | 安徽蔚来智驾科技有限公司 | Point cloud target detection method, computer equipment, storage medium and vehicle |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109446187A (en) * | 2018-10-16 | 2019-03-08 | 浙江大学 | Complex equipment health status monitoring method based on attention mechanism and neural network |
CN109522600A (en) * | 2018-10-16 | 2019-03-26 | 浙江大学 | Complex equipment remaining life prediction technique based on combined depth neural network |
CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11188065B2 (en) * | 2017-09-23 | 2021-11-30 | Nanoprecise Sci Corp. | System and method for automated fault diagnosis and prognosis for rotating equipment |
-
2019
- 2019-10-09 CN CN201910957332.1A patent/CN110647867B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109446187A (en) * | 2018-10-16 | 2019-03-08 | 浙江大学 | Complex equipment health status monitoring method based on attention mechanism and neural network |
CN109522600A (en) * | 2018-10-16 | 2019-03-26 | 浙江大学 | Complex equipment remaining life prediction technique based on combined depth neural network |
CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
Non-Patent Citations (2)
Title |
---|
Bearing Fault Diagnosis Using Structure Optimized Deep Convolutional Neural Network under Noisy Environment;Guoqiang Jin等;《IOP Conference Series: Materials Science and Engineering》;20190807;1-5 * |
深度学习在设备故障预测与健康管理中的应用;陈志强等;《仪器仪表学报》;20190930;206-226 * |
Also Published As
Publication number | Publication date |
---|---|
CN110647867A (en) | 2020-01-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110647867B (en) | Bearing fault diagnosis method and system based on self-adaptive anti-noise neural network | |
CN110516305B (en) | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model | |
CN110657985B (en) | Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis | |
CN114970605B (en) | Refrigerating equipment fault diagnosis method of multi-mode feature fusion neural network | |
CN110617966A (en) | Bearing fault diagnosis method based on semi-supervised generation countermeasure network | |
CN110795843B (en) | Method and device for identifying faults of rolling bearing | |
CN112906644A (en) | Mechanical fault intelligent diagnosis method based on deep migration learning | |
CN110991424A (en) | Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder | |
CN111595583B (en) | GAU neural network-based bearing residual life prediction method | |
CN115017945A (en) | Mechanical fault diagnosis method and system based on enhanced convolutional neural network | |
CN111581892A (en) | Method for predicting residual life of bearing based on GDAU neural network | |
CN113532829A (en) | Reciprocating compressor fault diagnosis method based on improved RCMDE | |
CN115901249B (en) | Rolling bearing performance degradation evaluation method combining feature optimization and multi-strategy optimization SVDD | |
CN114166509A (en) | Motor bearing fault prediction method | |
CN114169110A (en) | Motor bearing fault diagnosis method based on feature optimization and GWAA-XGboost | |
CN117851872B (en) | High-frequency vibration treatment diagnosis method and system for aero-engine | |
CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
CN113657244B (en) | Fan gearbox fault diagnosis method and system based on improved EEMD and speech spectrum analysis | |
CN117150882A (en) | Engine oil consumption prediction method, system, electronic equipment and storage medium | |
Li et al. | Research on rolling bearing fault diagnosis based on DRS frequency spectrum image and deep learning | |
Guo et al. | Research on bearing fault diagnosis based on novel MRSVD-CWT and improved CNN-LSTM | |
CN117974736B (en) | Underwater sensor output signal noise reduction method and system based on machine learning | |
CN117708574B (en) | CNN variable-speed rolling bearing fault diagnosis method embedded with physical information | |
CN114739673B (en) | Bearing vibration signal characteristic interpretability dimension reduction and fault diagnosis method | |
Wu et al. | Sound Visualization and Convolutional Neural Network in Fault Diagnosis of Electric Motorbike. |
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 |