CN114091553A - Diagnosis method for rolling bearing fault - Google Patents
Diagnosis method for rolling bearing fault Download PDFInfo
- Publication number
- CN114091553A CN114091553A CN202010641351.6A CN202010641351A CN114091553A CN 114091553 A CN114091553 A CN 114091553A CN 202010641351 A CN202010641351 A CN 202010641351A CN 114091553 A CN114091553 A CN 114091553A
- Authority
- CN
- China
- Prior art keywords
- neural network
- rolling bearing
- network
- training
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
-
- 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
- G06N3/048—Activation functions
-
- 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
Abstract
The invention discloses a rolling bearing fault diagnosis method based on a twin neural network, which comprises the following steps: collecting vibration signals of various bearings with faults to be identified; dividing the signals into data sequences with equal length to form a basic data sample library containing various faults or normal working states; establishing a twin neural network based on a one-dimensional convolution neural network and training the network, and outputting by taking the similarity of the characteristic sequences output by the two sub-networks as a model; and calculating the similarity of the data sequence to be detected relative to the basic data sample base, and classifying the signal to be detected into the type of the sample with the maximum similarity by using the nearest neighbor classification principle so as to obtain the fault diagnosis type. The invention combines the convolution neural network and the twin network structure, so that the fault diagnosis accuracy of the rolling bearing is obviously improved, and the rolling bearing has good adaptability to a small sample database and small-range working condition changes.
Description
Technical Field
The invention relates to the field of fault detection based on vibration, in particular to a fault diagnosis and identification method based on a rolling bearing.
Background
Rolling bearings are one of the most common parts in rotary machines, and are widely used in various mechanical devices in various industries. Whether precision machining or large field mobile equipment is adopted, the fault of the rolling bearing always occurs at any time, and fault investigation shows that the fault related to the bearing is about 40 percent in the most common faults in the induction motor, and the serious fault of the bearing not only brings economic loss, but also more possibly causes serious production accidents and even casualties. Therefore, early and effective bearing fault diagnosis is an important measure to ensure the normal operation and routine maintenance of the machine.
The common signal processing methods for mechanical fault diagnosis based on machine learning are roughly divided into two categories, one is to extract certain frequency domain or time domain characteristic indexes of vibration signals, and then construct a fault classification model through the machine learning method. However, such characteristic indexes are generally sensitive to working condition change conditions, for example, small changes of load or rotating speed can cause large differences in the characteristics of the test sample to cause model failure, and the characteristics are often selected manually and have large randomness; the other type is that the oscillogram of the signal is used for deep learning by a CNN algorithm of a convolutional neural network to complete fault diagnosis, such as CN 107421741A. However, the neural network based on deep learning usually needs a large number of samples in different states to perform network training, and the fault is not normal, so that it is difficult to obtain or calibrate a large number of data in different fault states. Therefore, an important problem faced by the fault diagnosis of the rolling bearing is how to establish a model applicable to a certain degree of working condition change by using a small number of labeled training samples.
Disclosure of Invention
The invention aims to provide a rolling bearing fault diagnosis method which is applicable to small sample size and has good robustness.
The invention provides a model and a method for establishing rolling bearing fault diagnosis by combining a twin neural network and a convolution neural network.
A rolling bearing fault diagnosis method based on a twin neural network is to train a neural network SCNN model and judge the working state or fault condition of a bearing by using the SCNN model, and the model establishment steps are as follows:
and 3, establishing a twin neural network SCNN model structure based on the one-dimensional convolutional neural network and initializing. The twin Network comprises two one-dimensional convolutional neural networks CNN (sub-Network in figure 2) with the same structure and shared parameters and a similarity measurement function DWAnd (4) forming. The one-dimensional convolutional neural network CNN comprises 3 convolutional layers, 3 maximum pooling layers, 1 Flatten layer and 1 full-connection activation layer; the selection range of the convolution kernel size of the convolution layer and the pooling window size of the maximum pooling layer is 2 x 1-5 x 1, the step length is 1-3, and the number of convolution kernels is 16-128; and the Flatten layer connects the characteristic diagrams output by the previous layer, the number of the neurons of the full-connection activation layer is 32-64, a Dropout layer is added between the Flatten layer and the full-connection activation layer, and the neurons of the full-connection activation layer use a Sigmoid activation function. Model weight parameter initialization is interval [ -0.1, 0.1 [ -0.1 [ ]]The inner part is uniformly distributed, and the bias is initialized to 0; the model weight parameters were optimized using Adam's algorithm, with the learning rate set to 0.0001.
And 4, forming a sample pair by using the samples in the training set to carry out SCNN model training: let vector XiAnd XjFor training samples with labeled classes in the set, a pair of samples (X) of same type or different types is usedi,Xj) Respectively input into two sub-networks for forward propagation, XiNonlinear mapping to new eigenvector space through a sub-network, outputting eigenvector Net (w, X)i),XjThe non-linear mapping to a new eigenvector space output eigenvector Net (w, X) through another subnetworkj) Where w denotes two sub-networks in commonA shared weight parameter matrix. Obtaining Net (w, X) according to formula (1)i) And Net (w, X)j) Between as a function of similarity DWAnd as the output of the SCNN:
defining a loss function L (w, (Y, X)i,Xj) Is represented by the formula (2):
in order to achieve the goal of minimizing the loss function, a random gradient descent algorithm is used for reversely and layer-by-layer propagating the error to each node, the weight and the parameters of the network are updated, and the loss function is minimized, so that the purposes of minimum interval of the same-type samples and maximum interval of the heterogeneous samples are achieved.
Traversing all sample pairs in the training set to carry out iterative training of the network, recording a loss function curve, if the loss function after iteration is less than a set maximum threshold value of 0.02 and the iteration is kept stable, converging the model, finishing the network training, and executing the step 5; otherwise, continuously and randomly traversing the sample pairs of the training set to perform network training, and executing the step 4;
step 5, utilizing the trained twin neural network model SCNN to diagnose the faults of the rolling bearing: for the rolling bearing to be diagnosed, the vibration signal is sampled by the same sampling method and frequency as the modeling, and continuous data with the same length is intercepted to form one-dimensional time sequence data which is set as Xk。
Let the training set contain n samples { X1,X2…XnBelongs to the state type C1,C2…CmAnd, each sample Xi in the training set is (x)1,x2,…,xd) (i-1 … n) to obtain Net (w, X)i) (ii) a Data sequence X to be predictedkInputting another sub-network of the SCNN model to obtain the corresponding output Net (w, X)k)。
Calculating each Net (w, X) separately as in equation (1)i) (i-1 … n) and Net (w, X)k) Similarity between them DWiIf the following conditions are met: dWi=min({DW1,...,DWn}) and X)i∈Cj(j 1 … m), and outputting a classification decision X according to the nearest neighbor classification principlek∈CjThat is, the rolling bearing is in the jth state, and the failure diagnosis is completed.
Advantageous effects
According to the invention, the fault diagnosis model of the rolling bearing is directly established according to the one-dimensional time sequence data acquired by the machine vibration signal acquired by the sensor, signal processing in other modes is not required, the requirements on relevant knowledge and actual experience of signal processing are reduced, and the time for manually processing the signal data is also reduced, so that the fault diagnosis and prediction of the rolling bearing are more intelligent and efficient; compared with other methods based on deep learning or neural network learning, the method provided by the invention can still establish an efficient model for accurate fault diagnosis under the condition of less training samples, and has stronger robustness to the change of working conditions.
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 process of the method of the present invention;
FIG. 2 is a diagram of the structural hierarchy and unit number of a one-dimensional convolutional neural network CNN;
FIG. 3 is a diagram of a twin neural network SCNN model;
FIG. 4 is an embedded t-SNE dimension reduction visualization of an original set of test sets in an embodiment example, showing fuzzy data set class boundaries;
FIG. 5 is a characteristic set Net (w, X) output by the SCNN model transformation of the test set in the embodiment examplei) The embedded t-SNE dimension reduction visualization graph shows that the data clustering effect is obvious after model processing, and the improvement of the fault diagnosis accuracy is facilitated.
The specific implementation mode is as follows:
the embodiments of the present invention will be described in further detail below with reference to the drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention. The specific implementation process of the method of the invention is as follows:
a1.5 KW motor is used for experiments, a bearing to be detected is an SKF6205 deep groove bearing, electric spark single-point damage is used for simulating bearing faults, an acceleration sensor is adopted on a bearing seat at the driving end of the motor to detect vibration acceleration signals of the fault bearing, the vibration acceleration signals are collected by a 16-channel data recorder, the rotating speed of the running state of the bearing is 1797rpm, and the sampling frequency is 12 kHz. The acquired vibration signals of the rolling bearing comprise four different working states of the bearing, and are divided into 4 types, namely a ball fault, an inner ring fault, an outer ring fault and a normal state, wherein the respective state labels are respectively 1,2,3 and 4. The failure diameter was 0.1778mm and the load was 0 hp.
Dividing signals in each state to form data samples, setting the length of each sample data to be 512 points, obtaining 236 samples in each signal state to obtain a sample database, dividing the sample database into a training set/a testing set, and respectively setting the number of the samples contained in each state to be 72/166; the training set samples are used for training the network model, and the test set is used for evaluating the accuracy of the model.
Establishing a one-dimensional convolutional neural network CNN and initializing network parameters as shown in the attached figure 2: the input layer comprises 512 units, the network sequentially comprises a convolutional layer, a maximum pooling layer, a Flatten layer and a full-connection activation layer, wherein the sizes of convolutional kernels of the C1 convolutional layer, the C2 convolutional layer and the C3 convolutional layer are set to be 3 multiplied by 1, the step length is 1, model weight parameters are initialized to be uniformly distributed in an interval of (-0.1, 0.1), bias is initialized to be 0, and the number of convolutional kernels is (16, 32, 64) sequentially. Setting the size of the pooling windows of the maximum pooling layers M1, M2 and M3 to be 3 multiplied by 1, taking the step length to be 2, connecting the feature maps output by the previous layer by the Flatten layer, finally setting the number of neurons of the full-connection activation layer to be 64, adding a Dropout layer between the Flatten layer and the full-connection activation layer to improve the generalization performance of the model, using a Sigmoid activation function for the neurons of the full-connection activation layer, and initializing the weight parameters of the model into uniform distribution in the interval of-0.1 and 0.1. The model weight parameters were optimized using Adam's algorithm, with the learning rate set to 0.0001.
And (3) constructing the twin neural network SCNN model by the 2 one-dimensional convolution neural networks CNN according to the mode of the attached figure 3. Traverse all samples X in the training setiAnd XjForm a sample pair, and input two sub-networks, XiNonlinear mapping to new eigenvector spatial output Net (w, X) through a subnetworki) Vector, XjThen non-linearly mapped to the new eigenvector spatial output Net (w, X) through another subnetworkj) Vector, where w represents the weight parameter matrix shared by the two subnetworks.
Let sub-network k layer have p(k)Each neuron (K ═ 1,2, …, K), inputs Xi=(x1,x2,…,xd) Output of the k-th layer z(k)=s(w(k)z(k-1)+b(k)) Wherein b is(k)Is a length p(k)Offset vector of w(k)Is a p(k)D, s is a nonlinear activation function sigmoid, and the output of the top layer of the subnetwork is as follows:
Net(w,Xi)=z(8)=s(w(8)z(7)+b(8)) (4)
calculating the sequence pairs (X) according to (1)i,Xj) Similarity of (D)W(Xi,Xj) And (3) carrying out the training of the SCNN model according to the loss function definitions (1) and (2).
In the training stage, a loss function is solved through data forward propagation, network parameters are optimized through back propagation adjustment, and parameters { w ] are optimized according to a random gradient descent algorithm(k),b(k)And reaching the aim of minimizing the loss function. The specific method comprises the following steps:
when X is inputtediAnd XjWhen Y is 0 for the same data pair, the loss function of equation (2) is:
parameter { w(k),b(k)The updating process is as follows:
when inputting XiAnd XjWhen Y is 1 for different data pairs, the loss function is:
in the formula (8), when the meta-parameter m is 0.2, D isW>m is, max (0, m-D)W) Is equal to 0, LDIs 0, parameter w(k),b(k)No need of adjustment, when DW<m is as follows:
thus, the loss function L (w, (Y, X) is minimizedi,Xj) Make homogeneous data pairs (X)i,Xj) Similarity measure function D at inputWMonotonically decreasing, heterogeneous data pairs (X)i,Xj) D at the time of inputWThe value monotonically increases.
And setting a maximum threshold value of an iterative loss function to be 0.02, stopping training if the loss function value is smaller than the maximum threshold value and keeps stable according to a loss function descending curve, wherein the SCNN of the embodiment achieves stability after approximately randomly traversing 50 times of training, and obtaining a network model.
In order to diagnose the bearing fault by using the model, the vibration signal is sampled by the rolling bearing to be diagnosed by the same method and the sampling frequency of 12kHz, 512 continuous data points are intercepted to form a one-dimensional data sequence, and the one-dimensional data sequence is set as Xk。
Training set { X1,X2,…,XnSample X of }i=(x1,x2,…,xd) (i-1 … n) belonging to the respective fault type {1,2,3,4}, each input a sub-network of the model, resulting in all nets (w, X)i) (ii) a Mixing XkEnter another sub-network of the model, get Net (w, X)k) (ii) a All Net (w, X) are calculated from equation (1)i) And Net (w, X)k) Similarity between them Dw={Dw1,…,Dwn}。
If the following conditions are met: dWi==min({DW1,...,DWn}) then X is indicatedkThe current rolling bearing is in the same state as the ith sample, which is the same type as the ith sample. 168 samples of the test set are tested, and the classification accuracy reaches 100%.
In order to test the robustness of the model in the case of load changes, the load and the bearing speed are varied with the same bearing fault state: the load increase was 1hp and the speed reduction 25 was 1772 rpm. By adopting the same signal acquisition equipment and the same signal acquisition method, the bearing states comprise 4 types of 'ball fault', 'inner ring fault', 'outer ring fault' and 'normal state', 236 samples are obtained in each signal state, and all the samples are used as a test set; each sample data length is again 512 points. All samples are input into the model for testing, and 29 samples in the test results are misjudged, and the accuracy reaches 96.6 percent, which shows that the method still has good adaptability to small changes of the working condition.
In order to verify the generalization ability of the model under the training of fewer samples, 236 data of each state are randomly divided into a training set/a verification set/a test set, and the number of the data is set to 20/20/196 respectively, so that the training set only contains 80 samples; the verification set is set for inspecting whether the training result has an overfitting condition on the verification set after the training is finished so as to judge whether the model needs to adjust parameters for continuous training. The test result shows that 196 samples in the test set are misjudged as inner ring faults except 3 outer ring faults, the accuracy rate of ball faults, inner ring faults and normal state vibration signal identification reaches 100%, the accuracy rate reaches 99.6%, and the fault diagnosis model with high accuracy rate can be obtained under the condition of small sample set training.
The above embodiments are merely illustrative of the present invention and are not to be construed as limiting the invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that various combinations, modifications or equivalents may be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention, and the technical solution of the present invention is covered by the claims of the present invention.
Claims (4)
1. A rolling bearing fault diagnosis method based on a twin neural network is characterized by comprising the following steps:
step 1, collecting vibration signals of a rolling bearing, wherein the sampling frequency of the signals is not less than 5 times of the rotation frequency of the bearing. Each operating state should be sampled continuously for a time sufficient to ensure that there are a substantial number (greater than 50) of samples;
step 2, intercepting the acquired vibration signals in equal length to obtain continuous one-dimensional time sequence data, forming a data sample library containing various fault states and normal working states, and obtaining a training sample set;
step 3, establishing a twin neural network SCNN model based on a one-dimensional convolution neural network;
step 4, traversing sample pairs formed by samples in the training set to carry out SCNN model training;
and 5, utilizing the trained twin one-dimensional convolution neural network model SCNN to diagnose the faults of the rolling bearing.
2. The twin one-dimensional convolutional neural network rolling bearing fault diagnosis method of claim 1, wherein a twin neural network SCNN model structure based on a one-dimensional convolutional neural network is established and initialized. The twin network comprises two one-dimensional convolutional neural networks CNN (sub-network) shared by the same parameters with the same structure and a similarity measurement function DWAnd (4) forming. The one-dimensional convolutional neural network CNN comprises 3 convolutional layers, 3 maximum pooling layers, 1 Flatten layer and 1 full-connection activation layer; the selection range of the convolution kernel size of the convolution layer and the pooling window size of the maximum pooling layer is 2 x 1-5 x 1, the step length is 1-3, and the number of convolution kernels is 16-128; and the Flatten layer connects the characteristic diagrams output by the previous layer, the number of the neurons of the full-connection activation layer is 32-64, a Dropout layer is added between the Flatten layer and the full-connection activation layer, and the neurons of the full-connection activation layer use a Sigmoid activation function. Model weight parameter initialization is interval [ -0.1, 0.1 [ -0.1 [ ]]The inner part is uniformly distributed, and the bias is initialized to 0; the weight parameters are optimized using the adam (adaptive motion estimation) algorithm, and the learning rate is set to 0.0001.
3. The twin one-dimensional convolutional neural network rolling bearing fault diagnosis method of claim 1, wherein the sample composition sample pair in the training set is used for SCNN network model training: let vector XiAnd XjFor training samples with labeled classes in the set, a pair of samples (X) of same type or different types is usedi,Xj) Respectively input into two sub-networks for forward propagation, XiNonlinear mapping to new eigenvector space through a CNN sub-network, output vector Net (w, X)i),XjThen the non-linear mapping to a new eigenvector spatial output vector Net (w, X) through another subnetwork is performedj) Where w represents a weight parameter matrix shared by the two subnetworks. Obtaining Net (w, X) according to formula (1)i) And Net (w, X)j) The Euclidean distance therebetween is used as the similarity DWAnd as the output of the SCNN:
defining a loss function L (w, (Y, X)i,Xj) Is represented by the formula (2):
wherein Y is a class factor:
and (3) reversely and layer-by-layer propagating the error to each node by using an Adam optimization algorithm, updating the weight and parameters of the network, and minimizing the loss function value so as to achieve the purposes of minimum interval of the same-type samples and maximum interval of the heterogeneous samples.
Traversing all sample pairs in the training set to carry out iterative training of the network, recording a loss function curve, if the loss function after iteration is smaller than a set maximum threshold and the iteration is kept stable, converging the model, finishing the network training, and executing the step 5; otherwise, continuing to randomly traverse the sample pairs of the training set to perform network training, and executing the step 4.
4. The twin one-dimensional convolutional neural network rolling bearing fault diagnosis method according to claim 1, wherein the fault diagnosis of the rolling bearing is performed by using a twin neural network model SCNN completed by a training set: for the rolling bearing to be diagnosed, the vibration signal is sampled by the same sampling method and frequency as the modeling, and the continuous data with equal length is intercepted to form one-dimensional time sequence data which is set as Xk;
Let the training set contain n samples { X1,X2,…,XnBelong to the state types C respectively1,C2,…,CnWill train each in the setSample Xi=(x1,x2,…,xd) (i-1 … n) to obtain Net (w, X)i) (ii) a Data sequence X to be predictedkInputting another sub-network of the SCNN model to obtain the corresponding output Net (w, X)k)。
Calculating each Net (w, X) separately as in equation (1)i) (i-1 … n) and Net (w, X)k) The Euclidean distance therebetween is taken as the similarity DWIf the following conditions are met:and Xi∈Cj(j 1 … m), and outputting a classification decision X according to the nearest neighbor classification principlek∈CjThat is, the rolling bearing is in the jth state, and the failure diagnosis is completed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010641351.6A CN114091553A (en) | 2020-08-06 | 2020-08-06 | Diagnosis method for rolling bearing fault |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010641351.6A CN114091553A (en) | 2020-08-06 | 2020-08-06 | Diagnosis method for rolling bearing fault |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114091553A true CN114091553A (en) | 2022-02-25 |
Family
ID=80294789
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010641351.6A Pending CN114091553A (en) | 2020-08-06 | 2020-08-06 | Diagnosis method for rolling bearing fault |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114091553A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114593917A (en) * | 2022-03-08 | 2022-06-07 | 安徽理工大学 | Small sample bearing fault diagnosis method based on triple model |
CN115430815A (en) * | 2022-08-09 | 2022-12-06 | 衡阳镭目科技有限责任公司 | Crystallizer liquid level control method and device, electronic equipment and storage medium |
CN115659258A (en) * | 2022-11-10 | 2023-01-31 | 国网山东省电力公司德州供电公司 | Power distribution network fault detection method based on multi-scale graph convolution twin network |
CN115795292A (en) * | 2022-10-20 | 2023-03-14 | 南京工大数控科技有限公司 | Gear milling machine spindle box fault diagnosis system and method based on LabVIEW |
CN116578889A (en) * | 2023-06-30 | 2023-08-11 | 国恒能元(天津)电力科技发展有限公司 | Power generation fault diagnosis method |
CN116842402A (en) * | 2023-09-01 | 2023-10-03 | 北京科技大学 | Blast furnace abnormal furnace condition detection method based on stable characteristic extraction of twin neural network |
-
2020
- 2020-08-06 CN CN202010641351.6A patent/CN114091553A/en active Pending
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114593917A (en) * | 2022-03-08 | 2022-06-07 | 安徽理工大学 | Small sample bearing fault diagnosis method based on triple model |
CN115430815A (en) * | 2022-08-09 | 2022-12-06 | 衡阳镭目科技有限责任公司 | Crystallizer liquid level control method and device, electronic equipment and storage medium |
CN115795292A (en) * | 2022-10-20 | 2023-03-14 | 南京工大数控科技有限公司 | Gear milling machine spindle box fault diagnosis system and method based on LabVIEW |
CN115795292B (en) * | 2022-10-20 | 2023-10-17 | 南京工大数控科技有限公司 | Gear milling machine spindle box fault diagnosis system and method based on LabVIEW |
CN115659258A (en) * | 2022-11-10 | 2023-01-31 | 国网山东省电力公司德州供电公司 | Power distribution network fault detection method based on multi-scale graph convolution twin network |
CN115659258B (en) * | 2022-11-10 | 2024-04-30 | 国网山东省电力公司德州供电公司 | Power distribution network fault detection method based on multi-scale graph roll-up twin network |
CN116578889A (en) * | 2023-06-30 | 2023-08-11 | 国恒能元(天津)电力科技发展有限公司 | Power generation fault diagnosis method |
CN116578889B (en) * | 2023-06-30 | 2023-11-10 | 国网甘肃省电力公司经济技术研究院 | Power generation fault diagnosis method |
CN116842402A (en) * | 2023-09-01 | 2023-10-03 | 北京科技大学 | Blast furnace abnormal furnace condition detection method based on stable characteristic extraction of twin neural network |
CN116842402B (en) * | 2023-09-01 | 2024-02-13 | 北京科技大学 | Blast furnace abnormal furnace condition detection method based on stable characteristic extraction of twin neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114091553A (en) | Diagnosis method for rolling bearing fault | |
Yang et al. | SuperGraph: Spatial-temporal graph-based feature extraction for rotating machinery diagnosis | |
CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
Huang et al. | A robust weight-shared capsule network for intelligent machinery fault diagnosis | |
CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
CN110297479B (en) | Hydroelectric generating set fault diagnosis method based on convolutional neural network information fusion | |
CN111539132B (en) | Dynamic load time domain identification method based on convolutional neural network | |
CN111539152B (en) | Rolling bearing fault self-learning method based on two-stage twin convolutional neural network | |
CN110991424A (en) | Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder | |
CN111006865A (en) | Motor bearing fault diagnosis method | |
CN111753891B (en) | Rolling bearing fault diagnosis method based on unsupervised feature learning | |
CN113469219B (en) | Rotary machine fault diagnosis method under complex working condition based on element transfer learning | |
CN111397901A (en) | Rolling bearing fault diagnosis method based on wavelet and improved PSO-RBF neural network | |
CN113705424A (en) | Performance equipment fault diagnosis model construction method based on time convolution noise reduction network | |
CN115017945A (en) | Mechanical fault diagnosis method and system based on enhanced convolutional neural network | |
CN114969995A (en) | Rolling bearing early fault intelligent diagnosis method based on improved sparrow search and acoustic emission | |
CN113569990B (en) | Strong noise interference environment-oriented performance equipment fault diagnosis model construction method | |
CN111060316A (en) | Rolling bearing state monitoring method and system based on convolutional neural network model | |
CN114169377A (en) | G-MSCNN-based fault diagnosis method for rolling bearing in noisy environment | |
CN115358259A (en) | Self-learning-based unsupervised cross-working-condition bearing fault diagnosis method | |
CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
CN114118138A (en) | Bearing composite fault diagnosis method based on multi-label field self-adaptive model | |
Li et al. | An orthogonal wavelet transform-based K-nearest neighbor algorithm to detect faults in bearings | |
Peng et al. | Research on fault diagnosis method of rolling bearing based on 2DCNN | |
Hou et al. | Multiple sensors fault diagnosis for rolling bearing based on variational mode decomposition and convolutional neural networks |
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 |