CN112254964A - Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network - Google Patents
Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network Download PDFInfo
- Publication number
- CN112254964A CN112254964A CN202010917253.0A CN202010917253A CN112254964A CN 112254964 A CN112254964 A CN 112254964A CN 202010917253 A CN202010917253 A CN 202010917253A CN 112254964 A CN112254964 A CN 112254964A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- convolution
- rolling bearing
- scale
- 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
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000005096 rolling process Methods 0.000 title claims abstract description 53
- 238000003745 diagnosis Methods 0.000 title claims abstract description 51
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 41
- 238000012360 testing method Methods 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000012544 monitoring process Methods 0.000 claims abstract description 23
- 238000003062 neural network model Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 14
- 230000001133 acceleration Effects 0.000 claims abstract description 10
- 210000002569 neuron Anatomy 0.000 claims description 32
- 230000004913 activation Effects 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 20
- 238000005070 sampling Methods 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 16
- 238000010586 diagram Methods 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 8
- 230000036541 health Effects 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 230000006835 compression Effects 0.000 claims description 4
- 238000007906 compression Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 3
- 230000006872 improvement Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 16
- 230000008569 process Effects 0.000 description 6
- 239000000284 extract Substances 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 238000012800 visualization Methods 0.000 description 5
- 238000013136 deep learning model Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000007943 implant Substances 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 239000000314 lubricant Substances 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Images
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
- 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
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a rolling bearing fault diagnosis method based on a rapid multi-scale convolutional neural network, belonging to the technical field of rolling bearing fault diagnosis methods based on the rapid multi-scale convolutional neural network; the technical problem to be solved is as follows: the improvement of the rolling bearing fault diagnosis method based on the rapid multi-scale convolution neural network is provided; the technical scheme for solving the technical problems is as follows: the method comprises the following steps: collecting a one-dimensional vibration signal of a rolling bearing driving end; standardizing each monitoring vibration signal acquired in the first step; building a rapid multi-scale convolution neural network model; training a rapid multi-scale convolution neural network model; acquiring an acceleration vibration signal of a current rolling bearing and processing the acceleration vibration signal to obtain a test set of a sample; inputting the sample test set in the fifth step into the trained rapid multi-scale convolution neural network model in the fourth step, and outputting a fault diagnosis result of the rolling bearing; the invention is applied to bearing fault diagnosis.
Description
Technical Field
The invention discloses a rolling bearing fault diagnosis method based on a rapid multi-scale convolutional neural network, and belongs to the technical field of rolling bearing fault diagnosis methods based on the rapid multi-scale convolutional neural network.
Background
Rolling bearings are a precision mechanical element known as lubricants for modern industry. Its working state is directly related to the efficiency and stability of the whole industrial production process. The health of the rolling bearings must be monitored in real time in order to maintain safe operation of the equipment.
At present, most fault diagnosis methods generally adopt signal processing methods such as short-time Fourier transform, wavelet transform, empirical mode decomposition and the like to extract fault characteristics of bearing vibration signals, and then the characteristics are automatically classified through mainstream machine learning methods such as a support vector machine, a k-proximity algorithm, an artificial neural network and the like, so that the error probability depending on expert experience is reduced. However, the accuracy still depends on whether the manually extracted features can accurately express the fault information, and the limitation is large.
Compared with a shallow network structure of a machine learning model, the deep learning model has deeper network layer, richer functions and stronger capability of self-adaptive learning characteristics, can integrate characteristic extraction and characteristic classification into a whole, and realizes an end-to-end process from data to a diagnosis result. However, the one-dimensional convolutional neural network and the two-dimensional convolutional neural network method applied to bearing fault diagnosis at present have the following defects: (1) most of the fault information on a single time scale of the vibration signal is considered, namely, only one fault feature of the time scale determined by the size of a single convolution kernel can be extracted each time; (2) how the convolutional neural network learns the obvious characteristics of the fault still looks like a black box, and the deep learning model is prevented from being continuously studied deeply in the field of fault diagnosis to a certain extent.
Therefore, the rolling bearing fault diagnosis method based on the rapid multi-scale convolution neural network, which can aim at the time domain multi-scale characteristics of the bearing vibration signal, is required to be provided, the multi-scale convolution module is made on the time domain multi-scale characteristics of the bearing vibration signal to extract multi-scale time information, the convolution-pooling alternating module is used for self-adapting overcoming the time dependence characteristics, and the feature extraction-feature classification two-in-one intelligent rolling bearing fault diagnosis method is constructed by combining the softmax classification layer.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to solve the technical problems that: the improvement of the rolling bearing fault diagnosis method based on the rapid multi-scale convolution neural network is provided.
In order to solve the technical problems, the invention adopts the technical scheme that: a rolling bearing fault diagnosis method based on a rapid multi-scale convolution neural network comprises the following steps:
the method comprises the following steps: data acquisition: collecting a one-dimensional vibration signal at the drive end of a rolling bearing, collecting a monitoring vibration signal X [ n ] of the rolling bearing with fault and without fault in different running states at fs sampling frequency, and setting corresponding fault labels for different fault conditions according to the collected monitoring vibration signal X [ n ];
step two: data processing: comprises collecting each monitored vibration signal X [ n ] collected in step one]Carrying out standardization processing, and recording the processed monitoring vibration signal as
Step three: building a rapid multi-scale convolution neural network model;
step four: monitoring vibration signals processed in the step two
As the input of the fast multi-scale convolution neural network model in the third step, the monitored vibration signals are correspondedTaking the fault state label as expected output of the rapid multi-scale convolution neural network model in the step three, and training the rapid multi-scale convolution neural network model;
step five: acquiring acceleration vibration signal x of current rolling bearing through same sampling frequency fstest[n]And standardizing the acceleration vibration signal to obtain a test set of samples
Step six: test set of samples in step five
Inputting the fast multi-scale convolutional neural network model trained in the fourth step, thereby outputting the fault diagnosis result of the rolling bearing.
The step of processing data in the step two further comprises: the data volume acquired in the first step is enlarged through a data enhancement method, and the specific steps are as follows:
setting the total length of a section of fault signal as L, the size of a selected sampling window as M, the sliding step length as S, selecting X samples on an original input signal, and obtaining a data set of X capacity when L is larger than or equal to M X-S (X-1).
The rapid multi-scale convolution neural network model in the third step comprises a large-scale convolution module, a multi-scale convolution module, a conventional convolution unit and a classification module.
The calculation formula of the standardization processing in the second step and the fifth step is as follows:
in the above formula: mu is X [ n ]]、xtest[n]The mean value of all the sampling point data in (a) is X [ n ]]、xtest[n]The standard deviation of the data of all the sampling points.
The large scale convolution module comprises a 1 st layer, a 2 nd layer and a 3 rd layer:
the layer 1 is a large-scale convolution layer and is used for directly extracting features of an input original monitoring vibration signal;
the layer 2 is an activation layer and is used for obtaining nonlinear expression of input signals and further distinguishing learned features, and an activation function specifically adopts a non-limiting unit (Relu);
and the layer 3 is a maximum pooling layer and is used for performing global maximum pooling on the nonlinear expression compression network parameters output by the layer 2.
The multi-scale convolution module comprises a 4 th layer, a 5 th layer, two multi-scale convolution layers and a 6 th layer:
the layer 4 is a convolutional layer and comprises three small-size convolution kernels, the size of each small-size convolution kernel is 3 multiplied by 1, and the small-size convolution kernels can effectively control parameters of a network model;
the 5 th layer is a convolutional layer and comprises three convolutional kernels, the sizes of the convolutional kernels are respectively 3 multiplied by 1, 5 multiplied by 1 and 7 multiplied by 1, the convolutional kernels are used for extracting accurate fault information in a small range and respectively obtaining YT、YF、YSThree output characteristics;
the 6 th layer is a cascade layer and is used for extracting output characteristics Y of the multi-scale convolution layers of the 4 th layer and the 5 th layerT、YF、YSFusing and splicing to output characteristic YC＝[YT，YF，YS]And fusing the two images in a one-dimensional space to obtain a feature y;
in the above formula: y isTA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 3 × 1 in the 5 th convolutional layerFA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 5 × 1 in the 5 th convolutional layerSA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 7 × 1 in the 5 th convolutional layerCIs three feature vectors YT、YF、YSOne-dimensional feature map after stitching, YCIs equal to YT、YF、YSSum of the lengths of the three feature maps.
The conventional convolution unit comprises a bottom 7 layer, an 8 th layer and a 9 th layer:
the 7 th layer is a convolutional layer and is used for carrying out self-adaptive extraction on the features input into the 7 th layer from the 6 th layer;
the 8 th layer is an activation layer and is used for obtaining a nonlinear expression of the 7 th layer output, and the activation function is specifically a non-limiting unit (Relu);
and the layer 9 is a global maximum pooling layer and is used for further compressing network parameters for the features output by the eighth layer to perform maximum pooling.
The classification module comprises a 10 th layer and a 12 th layer which are full-connection layers with different neuron numbers, and a 11 th layer and a 12 th layer:
the 10 th layer is a full connection layer, all neurons of the 10 th layer are connected with each node of the output characteristic diagram of the 9 th layer, and adaptive calculation of weight and bias parameters is performed;
the 11 th layer is a dropout layer and is used for randomly discarding part of neurons to perform network training, so that the accuracy of the model can be improved;
the 12 th layer is a full connection layer, the number of the neurons of the 12 th layer is set in one-to-one correspondence with the number of types of the faults needing to be classified, all the neurons of the 12 th layer are connected with each node of the output characteristic diagram of the 10 th layer, and adaptive calculation of weight and bias parameters is carried out;
and the layer 13 is a softmax layer and is used for converting the logarithm of the neuron in the layer 12 in a higher dimensionality so as to accord with the probability distribution form of the health conditions of different types of bearings, and selecting the corresponding fault state with the maximum probability as a fault diagnosis result.
The calculation formula of the regression function used in the layer 13 is as follows:
in the above formula: zjLogarithmic form of the j-th neuron of the output layer, ZkIn the logarithmic form of the kth neuron of the output layer.
Compared with the prior art, the invention has the beneficial effects that: 1) according to the method, a large convolution kernel with the size of 64 multiplied by 1 is used in the first layer of the model, so that the receptive field is increased, the generalization characteristics of time signals are learned, the model operation is accelerated, and the high-frequency disturbance is inhibited;
2) according to the invention, two layers of multi-scale convolution layers are introduced behind a large-size convolution module, the sizes of the first layer are all 3 multiplied by 1, and the sizes of the second layer are respectively 3 multiplied by 1, 5 multiplied by 1 and 7 multiplied by 1; the small size of the first layer effectively controls parameters of a network model, and the small convolution kernels of the second layer with different sizes extract accurate fault information in a small range; then, the extracted multi-time scale information is fused on a one-dimensional space by utilizing a cascade layer, so that the multi-scale convolution in the real sense is realized;
3) the method can well aim at the characteristics of nonlinearity and discontinuity of the vibration signal of the bearing, not only ensures the integrity of the original vibration signal, but also can effectively extract the multi-scale fault characteristics of the vibration signal time sequence;
4) according to the method, the network parameters are reduced by using the large-size convolution kernel, the training of the model is accelerated, the fault information of the one-dimensional vibration signal among different scales is extracted and fused by combining the multi-scale convolution module, and the accuracy and efficiency of fault diagnosis of the rolling bearing are greatly improved.
Drawings
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a flowchart of a rolling bearing fault diagnosis algorithm based on a fast multi-scale one-dimensional convolutional neural network constructed in the preferred embodiment of the present invention;
FIG. 3 is a block diagram of a fast multi-scale one-dimensional convolutional neural network constructed in accordance with a preferred embodiment of the present invention;
FIG. 4 is a schematic view of the test stand of the present invention;
FIG. 5 is a schematic diagram of the training process of the fast multi-scale one-dimensional convolutional neural network model of the present invention;
FIG. 6 is a schematic diagram of a confusion matrix of the fast multi-scale one-dimensional convolutional neural network model of the present invention;
FIG. 7 is a comparison of different convolution kernel sizes taken at layer 1 of the fast multi-scale one-dimensional convolutional neural network model of the present invention versus experimental test set and average elapsed time;
FIG. 8 is a line graph of the average recognition rate for 1000 test samples each time the present invention is trained 10 times with other convolutional neural network models;
FIGS. 9-12 are t-SNE dimension reduction visualizations of different data samples of the present invention.
In the figure: the device comprises a motor 1, an encoder 2, a dynamometer 3 and a control center 4.
Detailed Description
As shown in fig. 1 to 12, in the rolling bearing fault diagnosis method based on the fast multi-scale one-dimensional convolutional neural network of the present invention, firstly, monitoring vibration signals of a rolling bearing without faults and with different faults in different operation states are collected, a fault state label is set according to a fault state corresponding to each monitoring vibration signal, each monitoring vibration signal is subjected to standardization processing, the monitoring vibration signals are used as training samples to train a fast multi-scale one-dimensional convolutional neural network model, the fast multi-scale one-dimensional convolutional neural network model includes a large-scale convolution module, a multi-scale convolution module and a classification module, then, current monitoring vibration signals of the rolling bearing are collected to form a test set, and the test set is sent to the fast multi-scale one-dimensional convolutional neural network model for fault diagnosis.
The invention provides a rapid multi-scale convolutional neural network (LMCNN), which provides a multi-scale convolutional module for extracting multi-scale time information aiming at the time domain multi-scale characteristics of bearing vibration signals, uses a convolutional-pooling alternating module for self-adaptively overcoming time dependence characteristics, and then combines a softmax classification layer to construct an intelligent fault diagnosis method integrating feature extraction and feature classification.
The invention relates to a rolling bearing fault diagnosis method based on a rapid multi-scale convolutional neural network, which comprises the following steps of:
s1: and data acquisition, namely acquiring samples x [ n ] of normal working conditions and different fault working conditions in a normal production process by using a sampling frequency fs at a rolling bearing driving end, and marking corresponding fault labels, wherein the health conditions of the rolling bearing are divided into four categories, namely normal rolling ball damage, inner ring damage and outer ring damage.
S2: data processing, namely expanding the data volume by using a data enhancement method, setting the total length L of a section of fault signal, selecting X samples from an original input signal when the selected sampling window size is M and the sliding step length is S, and satisfying the relation: when L is not less than M X-S (X-1), a data set of X capacity is obtained at the same time. Simultaneously collecting each monitoring vibration signal x [ n ]]Standardized and recorded as
S3: and constructing a rapid multi-scale convolution neural network model, wherein the rapid multi-scale convolution neural network model comprises a large-scale convolution module, a multi-scale convolution module, a conventional convolution unit and a classification module.
The large scale convolution module includes:
the layer 1 is a large-scale convolution layer and is used for directly extracting features of an input original vibration signal, improving the model operation speed and inhibiting high-frequency disturbance;
the layer 2 is an activation layer, and is used for obtaining a non-linear expression of an input signal to enable learned features to be distinguished more easily, the used activation function is a non-limiting unit (Relu), and the calculation formula of the activation function is as follows:
in the above formula:
for the output of the activation layer, f (-) is the activation function,
is the output of the convolutional layer; the equation of the activation function is the output value of the convolution neural network for solving 0 and the previous layer
The maximum value therebetween.
And the 3 rd layer is a maximum pooling layer and is used for performing global maximum pooling on the nonlinear expression compression network parameters output by the 2 nd layer.
The multi-scale convolution module comprises two layers of multi-scale convolution layers:
the 4 th convolution layer consists of three small convolution kernels with the size of 3 multiplied by 1, and the small convolution kernels effectively control the parameters of the network model; the parameters of the convolution kernel are a 3 x 1 vector, wherein three parameters are weight parameters which need to be updated continuously by the network layer;
the 5 th convolution layer is composed of convolution kernels with the sizes of 3 multiplied by 1, 5 multiplied by 1 and 7 multiplied by 1 respectively, and accurate fault information in a small range is extracted to obtain YT、YF、YSThree output characteristics;
the 6 th layer is a cascade layer and is used for extracting output characteristics Y of the multi-scale convolution layers of the 4 th layer and the 5 th layerT、YF、YSFusing and splicing to output characteristic YC＝[YT，YF，YS]And fusing the two images in a one-dimensional space to obtain a feature y;
the 7 th layer is a convolutional layer and is used for carrying out self-adaptive extraction on the features input into the 7 th layer from the 6 th layer;
the 8 th layer is an activation layer and is used for obtaining the nonlinear expression of the 7 th layer output, and the activation function is a non-restrictive unit (Relu);
and the layer 9 is a global maximum pooling layer and is used for further compressing network parameters for the features output by the eighth layer and performing maximum pooling.
The classification module comprises a full connection layer with two different neuron numbers and a softmax classification layer:
the 10 th layer is a full connection layer, all neurons of the full connection layer are connected with each node of the output characteristic diagram of the 9 th layer, and adaptive calculation of weight and bias parameters is carried out;
the 11 th layer is a dropout layer, and part of neurons are randomly discarded for network training, so that the accuracy of the model can be improved;
the 12 th layer is a full connection layer, the number of the neurons of the full connection layer corresponds to the number of types of faults needing to be classified, all the neurons of the full connection layer are connected with each node of the output characteristic diagram of the 10 th layer, and the self-adaptive calculation of the weight and the bias parameters is carried out;
and the layer 13 is a softmax layer and is used for converting the logarithm of the layer 12 neurons on a higher dimension so as to accord with the probability distribution form of different bearing health conditions, and the corresponding fault state with the maximum probability is selected as a fault diagnosis result.
S4: processing each signal obtained in step S2
And as the input of the fast multi-scale convolution neural network model, the corresponding fault state label is used as the expected output of the fast multi-scale convolution neural network model, and the fast multi-scale convolution neural network model is trained.
S5: acquiring acceleration vibration signal x of current rolling bearing at same sampling frequency fstest[n]Standardizing the test sample by the same method in step S2 to obtain a sample test set
S6: will signal
And inputting the data into the multi-branch multi-scale convolutional neural network model trained in the step S4 to obtain a fault diagnosis result of the rolling bearing.
The rolling bearing failure diagnosis method of the present invention will be described in detail below with reference to a preferred embodiment of the present invention shown in fig. 2 and 3.
Fig. 2 is a flowchart of a rolling bearing fault diagnosis algorithm based on a fast multi-scale one-dimensional convolutional neural network, fig. 3 is a schematic structural diagram of the fast multi-scale one-dimensional convolutional neural network, and according to the algorithm flowchart and the schematic structural diagram of the neural network shown in fig. 2 and fig. 3, the rolling bearing fault diagnosis method of the present invention specifically includes the steps of:
step 1: the method comprises the steps of collecting monitoring vibration signals Xm [ n ] of a rolling bearing without faults and with different faults in different running states by using sampling frequency fs, and marking corresponding fault labels according to different fault conditions corresponding to the collected monitoring vibration signals Xm [ n ].
Step 2: in practical application, in order to avoid the problem of overfitting of a network model caused by too small data set and too many network parameters needing to be trained, the data set enhancement method in fig. 3 is used for expanding the sample size of the data set, the principle of the data set enhancement method in this embodiment is to intercept and sample a section of vibration signal with hundreds of thousands of vibration points in a sliding window form with a fixed length, and the length of the sample is the length of the sliding window, namely 2048. The method is characterized in that a sampling window with the size of M is used for carrying out sliding sampling on a monitoring vibration signal with the length of S points in steps of L unit lengths, N equal-length sub-signals can be obtained, and the capacity of a data set is greatly increased.
And step 3: normalization of data samples:
monitoring each acquired vibration signal x [ n ]]Performing normalization to obtain signal
In the embodiment, each acceleration vibration signal Xm [ n ] is normalized by using a z-score normalization method, and the normalized calculation formula is as follows:
where μ is the average of the data for all sample points in x [ n ], and σ is the standard deviation of the data for all sample points in x [ n ].
And 4, step 4: dividing the data set:
and (3) constructing an experimental data set, referring to the data set enhancement method in the step 2 in the specific steps, then dividing the test set, the verification set and the test set into 7000, 2000 and 1000 samples, wherein the length of each sample is 2048, when the test set is acquired and expanded, the acquisition method in the step 1 and the data set enhancement method in the step 2 are also used, and when the test set is input into the rapid multi-scale one-dimensional convolutional neural network, a test set signal without a label is used.
And 5: constructing a rapid multi-scale convolutional neural network:
FIG. 3 is a diagram of a fast multi-scale one-dimensional convolutional neural network model in the present invention. As shown in fig. 3, the fast multi-scale one-dimensional convolutional neural network model constructed in the present invention includes three steps of feature extraction, feature classification and fault determination for fault diagnosis. The feature extraction part mainly comprises a large-scale convolution module, a multi-scale convolution module and a conventional convolution unit, and each module is explained in detail below.
A large scale convolution module:
the convolution kernel with larger size is used, the receptive field is increased, the generalization characteristic of a time signal is learned, the model operation is accelerated, and the high-frequency disturbance is restrained. The large scale convolution module in this embodiment includes a large scale convolution layer, an active layer, and a maximum pooling layer.
The layer 1 is a large-scale convolution layer, the convolution kernel size is set to 64 × 1, the step size is set to S16, the convolution kernel regularization L2 is set, the coefficient is 0.0001, and the matrix is thinned. The characteristics of the input original monitoring vibration signals are directly extracted, the model operation speed is improved, and high-frequency disturbance is restrained. The large-size convolution layer directly extracts the characteristics of the input monitoring vibration signal when the input signal is
The output characteristic after passing through a large-size convolution layer is
The calculation formula of (A) is as follows:
in the above formula: k represents the length of the l-th layer convolution kernel;
representing a bias parameter of an ith neuron of the l layer;
and representing the weight parameter of the ith neuron channel of the ith layer.
And the 3 rd layer is a maximum pooling layer and is used for performing global maximum pooling on the nonlinear expression compression network parameters output by the 2 nd layer.
A multi-scale convolution module:
according to the invention, two multi-scale convolution layers are introduced after a large-size convolution module, the sizes of the first layers are all 3 multiplied by 1, the sizes of the second layers are respectively 3 multiplied by 1, 5 multiplied by 1 and 7 multiplied by 1, and the step length is S-1. The small size of the first layer effectively controls the parameters of the network model, and the small convolution kernels of different sizes of the second layer extract accurate fault information in a small range. And then, the extracted multi-time scale information is fused on a one-dimensional space by utilizing the cascade layer, so that the multi-scale convolution in the real sense is realized.
The 4 th convolution layer consists of three small convolution kernels, all of which have a size of 3 × 1, and the step size is S ═ 1. The small-size convolution kernel effectively controls the parameters of the network model;
the 5 th convolution layer is composed of convolution kernels of sizes 3 × 1, 5 × 1, and 7 × 1, respectively, and the step size is S ═ 1. Extracting accurate fault information in a small range to respectively obtain YT、YF、YSThree output characteristics;
the 6 th layer is a cascade layer and is used for extracting output characteristics Y of the multi-scale convolution layers of the 4 th layer and the 5 th layerT、YF、YSFusing and splicing to output characteristic YC＝[YT，YF，YS]And fusing on a one-dimensional space to obtain the feature y.
A conventional convolution unit:
the conventional convolution unit increases the depth of a network model and comprises a convolution layer, an activation layer and a pooling layer;
the 7 th layer is a convolution layer, the convolution kernel size is 3 × 1, and the step S is 2. The system is used for adaptively extracting the features input into the 7 th layer from the 6 th layer;
the 8 th layer is an activation layer and is used for obtaining the nonlinear expression of the 7 th layer output, and the activation function is a non-restrictive unit (Relu);
and the layer 9 is a global maximum pooling layer and is used for further compressing network parameters for the features output by the eighth layer and performing maximum pooling.
A classification module:
the 10 th layer is a full connection layer and comprises 1024 neurons, all the neurons are connected with each node of the output characteristic diagram of the ninth layer, and adaptive calculation of weight and bias parameters is carried out;
the 11 th layer is a dropout layer, the coefficient is set to be 0.8, and part of neurons are randomly discarded for network training, so that the accuracy of the model can be improved;
the 12 th layer is a full connection layer and comprises 10 neurons, all the neurons are connected with each node of the output characteristic diagram of the 10 th layer, and adaptive calculation of weight and bias parameters is carried out;
the 13 th layer is a softmax layer and is used for converting the logarithm of the 12 th layer of neurons on a higher dimensionality so as to accord with the probability distribution form of different bearing health conditions, and selecting a corresponding fault state with the maximum probability as a fault diagnosis result;
the calculation formula of the regression function of the softmax layer is as follows:
in the above formula: zjLogarithmic form of the j-th neuron of the output layer, ZkIn the logarithmic form of the kth neuron of the output layer.
Step 6: training a rapid multi-scale one-dimensional convolution neural network:
inputting the training set prepared in the step 4 into a rapid multi-scale one-dimensional convolution neural network, activating each layer of the network model through a forward propagation algorithm, updating weight parameters for finding the optimal parameters with the minimum loss through a backward propagation algorithm, and training the network model.
And 7: fault diagnosis:
and inputting the test set sample data into the network model to obtain the current fault diagnosis result of the rolling bearing.
Fig. 4 is a schematic structural view of a test bed on which the present invention is based. The test bench comprises motor 1, encoder 2, dynamometer 3 and control center 4, and wherein motor 1's rotor links to each other with dynamometer 3, and encoder 2 sets up the bearing drive end at motor 1, and encoder 2 also can be replaced by torque sensor. To better illustrate the effects of the present invention, the present invention was experimentally verified using a specific example. Experiments were conducted in this experimental validation using a 2 horsepower place motor and acceleration data was measured at locations close to and away from the bearings of the machine 1. Electric Discharge Machining (EDM) is used to fault implant the motor bearings. Resulting in different locations and different levels of wear. The method is characterized in that a sampling frequency of 12kHZ is used for sample collection at a bearing driving end of a motor, the health conditions of the bearing are divided into four categories, namely normal, rolling ball damage, inner ring damage and outer ring damage, each fault is divided into three fault sizes, namely 0.18mm, 0.36mm and 0.54mm according to the damage degree, and 10 bearing fault states are obtained. Table 1 is the sample capacities for ten different bearing health states and corresponding training, validation and test sets in this example.
| Label (R) | Bearing damage location | Degree of damage/mm | Training set | Verification set | Test set | Sample length |
| (Normal) | Is normal | 0 | 700 | 200 | 100 | 2048 |
| (Ball007) | Rolling ball | 0.18 | 700 | 200 | 100 | 2048 |
| (Ball014) | 0.36 | 700 | 200 | 100 | 2048 | |
| (Ball021) | 0.54 | 700 | 200 | 100 | 2048 | |
| (InnerRace007) | Inner ring failure | 0.18 | 700 | 200 | 100 | 2048 |
| (InnerRace014) | 0.36 | 700 | 200 | 100 | 2048 | |
| (InnerRace021) | 0.54 | 700 | 200 | 100 | 2048 | |
| (OuterRace007) | Outer ring failure | 0.18 | 700 | 200 | 100 | 2048 |
| (OuterRace014) | 0.36 | 700 | 200 | 100 | 2048 | |
| (OuterRace021) | 0.54 | 700 | 200 | 100 | 2048 | |
| Total of | 7000 | 2000 | 1000 |
TABLE 1
In order to illustrate the experimental performance of the scheme of the invention, the accuracy of fault diagnosis, the loss function and the confusion matrix are tested and verified.
Fig. 5 and 6 are graphs of a training process of the fast multi-scale one-dimensional convolutional neural network proposed by the present invention, after 20 epochs of training are performed, the trained model is used to perform fault classification on 1000 randomly divided test set samples, and as a result, the accuracy rate reaches 99.8%, and the loss function is only 0.00065.
In the training process of the model, the selection of the convolution kernel size of the first layer convolution layer has a large influence on the experimental result. Table 2 shows that different convolution kernel sizes correspond to different evaluation indexes, the recognition rate of the model training set reaches 0.998 under the size of 64 × 1, and the loss function of the test set is 0.00065.
TABLE 2
FIG. 7 is a graph comparing the experimental test set and the average elapsed time for taking different convolution kernel sizes at layer 1 of the fast multi-scale convolutional neural network model in the present invention. The accuracy of the 10 test sets was averaged and the elapsed time per epoch of the training process was recorded, showing that a convolution kernel size of 64 x 1 has the best performance.
FIG. 8 is a line graph of the average recognition rate of 1000 test samples each time the invention is trained 10 times with 3 other models. Deep learning model as a comparison: one-dimensional convolutional neural networks (CNN-1d) and large-scale convolutional neural networks (WDCNN-1d) and machine-learned classical model SVM. As can be seen from fig. 8, WDCNN-1d, although having a faster model training speed, ignores the rich multi-scale fault information in the vibration signal, and thus the diagnosis accuracy is reduced. In addition, the accuracy of all deep learning models is generally higher than that of a machine learning model SVM, which shows that the deep learning deep-level network structure characteristics are very beneficial to carrying out sufficient feature extraction on a complex time sequence, and the method is more suitable for bearing fault diagnosis under the background of big data.
As shown in table 3 below, the average accuracy on the test set during 10 training sessions of 4 methods is: 99.65%, 97.90%, 93.92% and 66.82%, also substantially conform to the trend of the accuracy of the test set in line graph 8.
| Diagnostic method | Test set accuracy | Mean loss function |
| LMCNN-1d | 99.65％ | 0.065％ |
| WDCNN-1d | 97.90％ | 1.56％ |
| CNN-1d | 93.92％ | 2.4％ |
| SVM | 66.82％ | \ |
TABLE 3
FIGS. 9-12 are t-SNE dimension reduction visualizations of different data samples. From the prediction probability of the input fault sample to the prediction probability of the output fault sample label, the neural network is like a black box, and the specific process of extracting the features in the neural network cannot be intuitively ascertained. t-SNE is a machine learning tool for dimensionality reduction visualization, which can map high-dimensional features of sample data to low dimensions and keep the features close, and can be generally used for researching the feature learning capability of a network model on input sample data. Fig. 10, 11, 12 show sample scattergrams with very high gradeability. And exhibits different degrees of aggregation between samples at different test sample volumes.
In the embodiment, 1000 samples which are not input into the invention (LMCNN-1d) are firstly subjected to dimension reduction visualization by using t-SNE, the visualization result is shown in FIG. 9, ten types of fault samples are mixed together, and the gradeability is poor. Then 90 samples are input into the invention (LMCNN-1d), and the feature learned by the last layer of the full connection layer of the model is subjected to dimension reduction processing. As shown in fig. 10, after model feature learning, the failure samples in each state begin to have a significant degree of separation, 90 test samples are basically completely separated, only a few samples are mixed in the failure samples of other types, and even it can be clearly seen that 4 ball failure samples of 0.36mm are mixed in the failure samples in the healthy state, and one inner ring failure sample of 0.18mm is mixed in the ball failure sample of 0.54 mm. As shown in fig. 11, as the sample size increases, when the sample size becomes 1000, the separability of the samples is higher, and the distance between the samples increases, the separability is stronger, and the softmax classification layer can easily distinguish different types of failure samples. Until the sample volume increased to 7000, as shown in fig. 12, the pitch further expanded, and the ten failure samples were completely self-clustered individually. This reflects that as the sample capacity increases, the generalization capability of the model increases, the model becomes more accurate and stable, and the fault identification becomes more accurate. Meanwhile, the idea that the neural network needs large-capacity sample data for training is explained.
The invention discloses a rolling bearing fault diagnosis method based on a rapid multi-scale convolutional neural network (LMCNN), which combines a mature knowledge theory system in the field of deep learning with the practical application in industrial production. Firstly, vibration signals of healthy and fault rolling bearings in different damage states in actual work are collected, and corresponding fault labels are set for the collected fault vibration signals. The method comprises the steps of carrying out standardization processing on collected vibration signals, dividing a training set to carry out parameter fitting on the performance of the fast multi-scale convolution neural network model of the patent, wherein the fast multi-scale convolution neural network comprises a large-scale convolution module, a multi-scale convolution module and a mode identification module. And then inputting the acceleration vibration signals acquired in real time into a fast multi-scale convolution network for diagnosis. The invention reduces network parameters by using a large-size convolution kernel, accelerates the training of the model, extracts and fuses the fault information of the one-dimensional vibration signal among different scales by combining a multi-scale convolution module, and greatly improves the accuracy and efficiency of fault diagnosis.
It should be noted that, regarding the specific structure of the present invention, the connection relationship between the modules adopted in the present invention is determined and can be realized, except for the specific description in the embodiment, the specific connection relationship can bring the corresponding technical effect, and the technical problem proposed by the present invention is solved on the premise of not depending on the execution of the corresponding software program.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (9)
1. A rolling bearing fault diagnosis method based on a rapid multi-scale convolution neural network is characterized by comprising the following steps: the method comprises the following steps:
the method comprises the following steps: data acquisition: collecting a one-dimensional vibration signal at the drive end of a rolling bearing, collecting a monitoring vibration signal X [ n ] of the rolling bearing with fault and without fault in different running states at fs sampling frequency, and setting corresponding fault labels for different fault conditions according to the collected monitoring vibration signal X [ n ];
step two: data processing: comprises collecting each monitored vibration signal X [ n ] collected in step one]Carrying out standardization processing, and recording the processed monitoring vibration signal as
Step three: building a rapid multi-scale convolution neural network model;
step four: monitoring vibration signals processed in the step two
As the input of the rapid multi-scale convolution neural network model in the third step, taking the fault state label corresponding to the monitored vibration signal as the expected output of the rapid multi-scale convolution neural network model in the third step, and training the rapid multi-scale convolution neural network model;
step five: acquiring acceleration vibration signal x of current rolling bearing through same sampling frequency fstest[n]And standardizing the acceleration vibration signal to obtain a test set of samples
Step six: test set of samples in step five
Inputting the fast multi-scale convolutional neural network model trained in the fourth step, thereby outputting the fault diagnosis result of the rolling bearing.
2. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 1, characterized in that: the step of processing data in the step two further comprises: the data volume acquired in the first step is enlarged through a data enhancement method, and the specific steps are as follows:
setting the total length of a section of fault signal as L, the size of a selected sampling window as M, the sliding step length as S, selecting X samples on an original input signal, and obtaining a data set of X capacity when L is larger than or equal to M X-S (X-1).
3. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 2, characterized in that: the rapid multi-scale convolution neural network model in the third step comprises a large-scale convolution module, a multi-scale convolution module, a conventional convolution unit and a classification module.
4. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 3, characterized in that: the calculation formula of the standardization processing in the second step and the fifth step is as follows:
in the above formula: mu is X [ n ]]、xtest[n]The mean value of all the sampling point data in (a) is X [ n ]]、xtest[n]The standard deviation of the data of all the sampling points.
5. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 4, characterized in that: the large scale convolution module comprises a 1 st layer, a 2 nd layer and a 3 rd layer:
the layer 1 is a large-scale convolution layer and is used for directly extracting features of an input original monitoring vibration signal;
the layer 2 is an activation layer and is used for obtaining nonlinear expression of input signals and further distinguishing learned features, and an activation function specifically adopts a non-limiting unit (Relu);
and the layer 3 is a maximum pooling layer and is used for performing global maximum pooling on the nonlinear expression compression network parameters output by the layer 2.
6. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 5, characterized in that: the multi-scale convolution module comprises a 4 th layer, a 5 th layer, two multi-scale convolution layers and a 6 th layer:
the layer 4 is a convolutional layer and comprises three small-size convolution kernels, the size of each small-size convolution kernel is 3 multiplied by 1, and the small-size convolution kernels can effectively control parameters of a network model;
the 5 th layer is a convolutional layer and comprises three convolutional kernels, the sizes of the convolutional kernels are respectively 3 multiplied by 1, 5 multiplied by 1 and 7 multiplied by 1, the convolutional kernels are used for extracting accurate fault information in a small range and respectively obtaining YT、YF、YSThree output characteristics;
the 6 th layer is a cascade layer and is used for extracting output characteristics Y of the multi-scale convolution layers of the 4 th layer and the 5 th layerT、YF、YSFusing and splicing to output characteristic YC＝[YT，YF，YS]And fusing the two images in a one-dimensional space to obtain a feature y;
in the above formula: y isTA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 3 × 1 in the 5 th convolutional layerFA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 5 × 1 in the 5 th convolutional layerSA one-dimensional feature map, Y, output by concatenating the convolution block 3 × 1 in the 4 th convolutional layer and the convolution block 7 × 1 in the 5 th convolutional layerCIs three feature vectors YT、YF、YSOne-dimensional feature map after stitching, YCIs equal to YT、YF、YSSum of the lengths of the three feature maps.
7. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 6, characterized in that: the conventional convolution unit comprises a bottom 7 layer, an 8 th layer and a 9 th layer:
the 7 th layer is a convolutional layer and is used for carrying out self-adaptive extraction on the features input into the 7 th layer from the 6 th layer;
the 8 th layer is an activation layer and is used for obtaining a nonlinear expression of the 7 th layer output, and the activation function is specifically a non-limiting unit (Relu);
and the layer 9 is a global maximum pooling layer and is used for further compressing network parameters for the features output by the eighth layer to perform maximum pooling.
8. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 7, characterized in that: the classification module comprises a 10 th layer and a 12 th layer which are full-connection layers with different neuron numbers, and a 11 th layer and a 12 th layer:
the 10 th layer is a full connection layer, all neurons of the 10 th layer are connected with each node of the output characteristic diagram of the 9 th layer, and adaptive calculation of weight and bias parameters is performed;
the 11 th layer is a dropout layer and is used for randomly discarding part of neurons to perform network training, so that the accuracy of the model can be improved;
the 12 th layer is a full connection layer, the number of the neurons of the 12 th layer is set in one-to-one correspondence with the number of types of the faults needing to be classified, all the neurons of the 12 th layer are connected with each node of the output characteristic diagram of the 10 th layer, and adaptive calculation of weight and bias parameters is carried out;
and the layer 13 is a softmax layer and is used for converting the logarithm of the neuron in the layer 12 in a higher dimensionality so as to accord with the probability distribution form of the health conditions of different types of bearings, and selecting the corresponding fault state with the maximum probability as a fault diagnosis result.
9. The rolling bearing fault diagnosis method based on the rapid multiscale convolutional neural network as claimed in claim 8, characterized in that: the calculation formula of the regression function used in the layer 13 is as follows:
in the above formula: zjLogarithmic form of the j-th neuron of the output layer, ZkIn the logarithmic form of the kth neuron of the output layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010917253.0A CN112254964A (en) | 2020-09-03 | 2020-09-03 | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010917253.0A CN112254964A (en) | 2020-09-03 | 2020-09-03 | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112254964A true CN112254964A (en) | 2021-01-22 |
Family
ID=74223830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010917253.0A Pending CN112254964A (en) | 2020-09-03 | 2020-09-03 | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112254964A (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112697435A (en) * | 2021-01-26 | 2021-04-23 | 山西三友和智慧信息技术股份有限公司 | Rolling bearing fault diagnosis method based on improved SELD-TCN network |
| CN112949823A (en) * | 2021-02-05 | 2021-06-11 | 杭州电子科技大学 | Industrial process performance diagnosis method based on one-dimensional multi-scale depth convolution neural network |
| CN113033309A (en) * | 2021-02-25 | 2021-06-25 | 北京化工大学 | Fault diagnosis method based on signal downsampling and one-dimensional convolution neural network |
| CN113240022A (en) * | 2021-05-19 | 2021-08-10 | 燕山大学 | Wind power gear box fault detection method of multi-scale single-classification convolutional network |
| CN113281029A (en) * | 2021-06-09 | 2021-08-20 | 重庆大学 | Rotating machinery fault diagnosis method and system based on multi-scale network structure |
| CN113361042A (en) * | 2021-06-22 | 2021-09-07 | 石家庄铁道大学 | Method for evaluating service life of rolling bearing |
| CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
| CN113469060A (en) * | 2021-07-02 | 2021-10-01 | 浙大城市学院 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
| CN113611568A (en) * | 2021-09-06 | 2021-11-05 | 辽宁石油化工大学 | Vacuum circuit breaker based on genetic convolution depth network |
| CN113740064A (en) * | 2021-07-27 | 2021-12-03 | 西安交通大学 | Rolling bearing fault type diagnosis method, device, equipment and readable storage medium |
| CN113758709A (en) * | 2021-09-30 | 2021-12-07 | 河南科技大学 | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning |
| CN113962256A (en) * | 2021-09-29 | 2022-01-21 | 西安交通大学 | Intelligent fault diagnosis method and system for electromechanical actuator |
| CN113984389A (en) * | 2021-12-01 | 2022-01-28 | 广西大学 | Rolling bearing fault diagnosis method based on multi-receptive-field and improved capsule map neural network |
| CN114112399A (en) * | 2021-11-26 | 2022-03-01 | 江南大学 | Combined fault diagnosis system and fault diagnosis method for wind turbine generator bearing |
| CN114330439A (en) * | 2021-12-28 | 2022-04-12 | 盐城工学院 | Bearing diagnosis method based on convolutional neural network |
| CN114332531A (en) * | 2021-12-23 | 2022-04-12 | 南京航空航天大学 | Mechanical equipment rapid fault diagnosis method and device based on 2D-CNN and t-SNE |
| CN114383844A (en) * | 2021-12-15 | 2022-04-22 | 太原理工大学 | Rolling bearing fault diagnosis method based on distributed deep neural network |
| CN115049852A (en) * | 2022-08-15 | 2022-09-13 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN115062867A (en) * | 2022-07-26 | 2022-09-16 | 中车长春轨道客车股份有限公司 | Bearing fault prediction method based on Teager energy spectrum and improved CNN model |
| CN116010805A (en) * | 2023-03-24 | 2023-04-25 | 昆明理工大学 | Rolling bearing fault feature extraction method and device based on convolutional neural network |
| CN116026586A (en) * | 2023-01-31 | 2023-04-28 | 东华大学 | Harmonic reducer delivery qualification judging method and device |
| CN116484172A (en) * | 2023-01-19 | 2023-07-25 | 南京林业大学 | Mechanical fault diagnosis method and system based on robust multi-scale network |
| CN117171544A (en) * | 2023-09-06 | 2023-12-05 | 哈尔滨理工大学 | Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018000752A1 (en) * | 2016-06-27 | 2018-01-04 | 浙江工商大学 | Monocular image depth estimation method based on multi-scale cnn and continuous crf |
| CN109299705A (en) * | 2018-10-24 | 2019-02-01 | 电子科技大学 | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks |
| KR20190119261A (en) * | 2018-04-12 | 2019-10-22 | 가천대학교 산학협력단 | Apparatus and method for segmenting of semantic image using fully convolutional neural network based on multi scale image and multi scale dilated convolution |
| 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 |
| CN111046916A (en) * | 2019-11-20 | 2020-04-21 | 上海电机学院 | Motor fault diagnosis method and system based on void convolution capsule network |
| CN111413075A (en) * | 2020-04-02 | 2020-07-14 | 重庆交通大学 | Fan base bolt loosening diagnosis method of multi-scale one-dimensional convolution neural network |
| CN111444871A (en) * | 2020-04-01 | 2020-07-24 | 北京信息科技大学 | Fault diagnosis method for multi-scale deep convolution neural network planetary gearbox |
| CN111523509A (en) * | 2020-05-08 | 2020-08-11 | 江苏迪赛司自动化工程有限公司 | Equipment fault diagnosis and health monitoring method integrating physical and deep expression characteristics |
-
2020
- 2020-09-03 CN CN202010917253.0A patent/CN112254964A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018000752A1 (en) * | 2016-06-27 | 2018-01-04 | 浙江工商大学 | Monocular image depth estimation method based on multi-scale cnn and continuous crf |
| KR20190119261A (en) * | 2018-04-12 | 2019-10-22 | 가천대학교 산학협력단 | Apparatus and method for segmenting of semantic image using fully convolutional neural network based on multi scale image and multi scale dilated convolution |
| CN109299705A (en) * | 2018-10-24 | 2019-02-01 | 电子科技大学 | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks |
| 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 |
| CN111046916A (en) * | 2019-11-20 | 2020-04-21 | 上海电机学院 | Motor fault diagnosis method and system based on void convolution capsule network |
| CN111444871A (en) * | 2020-04-01 | 2020-07-24 | 北京信息科技大学 | Fault diagnosis method for multi-scale deep convolution neural network planetary gearbox |
| CN111413075A (en) * | 2020-04-02 | 2020-07-14 | 重庆交通大学 | Fan base bolt loosening diagnosis method of multi-scale one-dimensional convolution neural network |
| CN111523509A (en) * | 2020-05-08 | 2020-08-11 | 江苏迪赛司自动化工程有限公司 | Equipment fault diagnosis and health monitoring method integrating physical and deep expression characteristics |
Non-Patent Citations (4)
| Title |
|---|
| 彭丹丹: "基于HHT和CNN的高速列车轮对轴承故障诊断方法", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 李玉花: "基于卷积神经网络的脑电信号分类", 《科学技术与工程》 * |
| 祝道强等: "变负载轴承故障诊断卷积神经网络模型", 《科学技术与工程》 * |
| 黄文艺: "一维MC-CNN特征自主学习的恒定工况滚动轴承故障诊断", 《中国博士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112697435A (en) * | 2021-01-26 | 2021-04-23 | 山西三友和智慧信息技术股份有限公司 | Rolling bearing fault diagnosis method based on improved SELD-TCN network |
| CN112949823A (en) * | 2021-02-05 | 2021-06-11 | 杭州电子科技大学 | Industrial process performance diagnosis method based on one-dimensional multi-scale depth convolution neural network |
| CN113033309B (en) * | 2021-02-25 | 2023-12-19 | 北京化工大学 | Fault diagnosis method based on signal downsampling and one-dimensional convolutional neural network |
| CN113033309A (en) * | 2021-02-25 | 2021-06-25 | 北京化工大学 | Fault diagnosis method based on signal downsampling and one-dimensional convolution neural network |
| CN113240022A (en) * | 2021-05-19 | 2021-08-10 | 燕山大学 | Wind power gear box fault detection method of multi-scale single-classification convolutional network |
| CN113281029A (en) * | 2021-06-09 | 2021-08-20 | 重庆大学 | Rotating machinery fault diagnosis method and system based on multi-scale network structure |
| CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
| CN113361042A (en) * | 2021-06-22 | 2021-09-07 | 石家庄铁道大学 | Method for evaluating service life of rolling bearing |
| CN113469060A (en) * | 2021-07-02 | 2021-10-01 | 浙大城市学院 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
| CN113740064A (en) * | 2021-07-27 | 2021-12-03 | 西安交通大学 | Rolling bearing fault type diagnosis method, device, equipment and readable storage medium |
| CN113611568A (en) * | 2021-09-06 | 2021-11-05 | 辽宁石油化工大学 | Vacuum circuit breaker based on genetic convolution depth network |
| CN113962256A (en) * | 2021-09-29 | 2022-01-21 | 西安交通大学 | Intelligent fault diagnosis method and system for electromechanical actuator |
| CN113758709A (en) * | 2021-09-30 | 2021-12-07 | 河南科技大学 | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning |
| CN114112399A (en) * | 2021-11-26 | 2022-03-01 | 江南大学 | Combined fault diagnosis system and fault diagnosis method for wind turbine generator bearing |
| CN114112399B (en) * | 2021-11-26 | 2022-12-02 | 江南大学 | Combined fault diagnosis system and fault diagnosis method for wind turbine generator bearing |
| CN113984389A (en) * | 2021-12-01 | 2022-01-28 | 广西大学 | Rolling bearing fault diagnosis method based on multi-receptive-field and improved capsule map neural network |
| CN113984389B (en) * | 2021-12-01 | 2023-09-01 | 广西大学 | Rolling bearing fault diagnosis method based on multiple receptive fields and improved capsule graph neural network |
| CN114383844A (en) * | 2021-12-15 | 2022-04-22 | 太原理工大学 | Rolling bearing fault diagnosis method based on distributed deep neural network |
| 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 |
| CN114332531A (en) * | 2021-12-23 | 2022-04-12 | 南京航空航天大学 | Mechanical equipment rapid fault diagnosis method and device based on 2D-CNN and t-SNE |
| CN114332531B (en) * | 2021-12-23 | 2023-07-25 | 南京航空航天大学 | Mechanical equipment rapid fault diagnosis method and device based on 2D-CNN and t-SNE |
| CN114330439A (en) * | 2021-12-28 | 2022-04-12 | 盐城工学院 | Bearing diagnosis method based on convolutional neural network |
| CN115062867A (en) * | 2022-07-26 | 2022-09-16 | 中车长春轨道客车股份有限公司 | Bearing fault prediction method based on Teager energy spectrum and improved CNN model |
| CN115049852B (en) * | 2022-08-15 | 2022-12-02 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN115049852A (en) * | 2022-08-15 | 2022-09-13 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN116484172A (en) * | 2023-01-19 | 2023-07-25 | 南京林业大学 | Mechanical fault diagnosis method and system based on robust multi-scale network |
| CN116484172B (en) * | 2023-01-19 | 2024-01-30 | 南京林业大学 | Mechanical fault diagnosis method and system based on robust multi-scale network |
| CN116026586A (en) * | 2023-01-31 | 2023-04-28 | 东华大学 | Harmonic reducer delivery qualification judging method and device |
| CN116010805A (en) * | 2023-03-24 | 2023-04-25 | 昆明理工大学 | Rolling bearing fault feature extraction method and device based on convolutional neural network |
| CN116010805B (en) * | 2023-03-24 | 2023-06-16 | 昆明理工大学 | Rolling bearing fault feature extraction method and device based on convolutional neural network |
| CN117171544A (en) * | 2023-09-06 | 2023-12-05 | 哈尔滨理工大学 | Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network |
| CN117171544B (en) * | 2023-09-06 | 2024-04-12 | 哈尔滨理工大学 | Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112254964A (en) | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network | |
| CN111046945B (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
| CN109186973B (en) | Mechanical fault diagnosis method of unsupervised deep learning network | |
| CN111964908B (en) | Bearing fault diagnosis method under variable working condition based on multi-scale weight distribution convolutional neural network model | |
| CN113567130A (en) | Bearing fault diagnosis method based on multiple working conditions of equipment | |
| CN109858352B (en) | Fault diagnosis method based on compressed sensing and improved multi-scale network | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN110516305B (en) | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN111753891B (en) | Rolling bearing fault diagnosis method based on unsupervised feature learning | |
| CN111046916A (en) | Motor fault diagnosis method and system based on void convolution capsule network | |
| CN113111820B (en) | Rotary part fault diagnosis method and device based on improved CNN and relation module | |
| CN114358124B (en) | New fault diagnosis method for rotary machinery based on deep countermeasure convolutional neural network | |
| CN115034267A (en) | One-dimensional convolution neural network rolling bearing fault diagnosis method fusing channel attention | |
| CN112504682A (en) | Chassis engine fault diagnosis method and system based on particle swarm optimization algorithm | |
| CN114112398A (en) | Fault diagnosis method for rolling bearing under variable speed working condition | |
| CN112541510A (en) | Intelligent fault diagnosis method based on multi-channel time series data | |
| CN111678699B (en) | Early fault monitoring and diagnosing method and system for rolling bearing | |
| CN113255591A (en) | Bearing fault diagnosis method based on random forest and fusion characteristics | |
| CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN112729834A (en) | Bearing fault diagnosis method, device and system | |
| CN114048787B (en) | Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model | |
| CN116702076A (en) | Small sample migration learning fault diagnosis method, system, computer and storage medium based on CNN feature fusion | |
| CN105823634A (en) | Bearing damage identification method based on time frequency relevance vector convolution Boltzmann machine |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210122 |
|
| RJ01 | Rejection of invention patent application after publication |