CN115270845A - Bearing fault diagnosis method based on EMD and multichannel convolutional neural network - Google Patents

Bearing fault diagnosis method based on EMD and multichannel convolutional neural network Download PDF

Info

Publication number
CN115270845A
CN115270845A CN202210637458.2A CN202210637458A CN115270845A CN 115270845 A CN115270845 A CN 115270845A CN 202210637458 A CN202210637458 A CN 202210637458A CN 115270845 A CN115270845 A CN 115270845A
Authority
CN
China
Prior art keywords
emd
fault diagnosis
neural network
convolutional neural
imf
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
Application number
CN202210637458.2A
Other languages
Chinese (zh)
Inventor
刘晓昂
赵福凯
岳卓
丁晋
甄冬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hebei University of Technology
Original Assignee
Hebei University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202210637458.2A priority Critical patent/CN115270845A/en
Publication of CN115270845A publication Critical patent/CN115270845A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Acoustics & Sound (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention provides a bearing fault diagnosis method based on an EMD and a multi-channel convolutional neural network, which belongs to the technical field of bearing fault diagnosis of a gearbox and comprises the following steps: decomposing the nonlinear vibration signal into a series of IMF signals and a remainder using EMD; calculating kurtosis values of the IMF components, and performing weighting operation on the kurtosis values and the correlation coefficients of the IMF components; removing the decomposed remainder, and selecting IMF component with larger kurtosis value; selecting IMF components with obvious characteristics by using the weighted value; and performing convolution operation. The deep learning method is applied to the field of equipment fault diagnosis, compared with the traditional method, the deep learning method can generally better extract the characteristics of a fault signal, and the mechanical fault diagnosis efficiency is improved.

Description

Bearing fault diagnosis method based on EMD and multichannel convolutional neural network
Technical Field
The invention belongs to the technical field of fault diagnosis of bearings of gearboxes, and particularly relates to a bearing fault diagnosis method based on an EMD (empirical mode decomposition) and a multichannel convolutional neural network.
Background
Gearboxes are widely applied in the fields of machinery, electricity and the like, wherein bearings are important parts in the gearboxes. In actual use, more tasks are often required to be completed within a short construction period, and the gearbox needs to be overloaded for a long time, so that faults often occur, and unnecessary shutdown is caused. The bearing is a relatively fragile component in the gearbox, so early failure diagnosis of the bearing is very necessary. However, the bearing has various and complex faults, and when the traditional fault diagnosis method is used, the signals are single when the fault signals acquired under the complex condition are processed, and the ideal effect is difficult to achieve by the processing method.
Therefore, the deep learning method is applied to the field of equipment fault diagnosis, compared with the traditional method, the deep learning method can generally better extract the characteristics of a fault signal, and the mechanical fault diagnosis efficiency is improved.
Disclosure of Invention
In view of the above, the present invention is directed to a bearing fault diagnosis method based on EMD and a multi-channel convolutional neural network to alleviate the above technical problems.
In order to achieve the purpose, the technical scheme of the invention is realized as follows:
a bearing fault diagnosis method based on EMD and a multi-channel convolutional neural network comprises the following steps:
(1) The vibration signal is decomposed using EMD: an arbitrary nonlinear signal is decomposed into a series of IMF signals and a remainder.
Figure BDA0003681086700000021
In the formula, n is the number of IMF components obtained by decomposition; c. Ci(t) is the ith IMF component; r isn(t) is the final residual component.
Assuming that the signal to be decomposed is x (t), the specific algorithm flow is as follows:
the first step is as follows: and fitting by using a spline interpolation method to obtain an upper envelope line and a lower envelope line.
The second step: subtracting the upper and lower envelopes by x (t)The mean value m1 (t) of the line yields a component k1(t)
h1(t)＝x(t)-m1 (2)
The third step: judgment h1(t) whether or not IMF condition is satisfied, and if so, h1(t) is denoted as c1 as the first IMF. Otherwise, repeating the steps, and obtaining the result on the assumption that the kth time meets the condition
c1＝h1(k-1)(t)-m1 (3)
The fourth step: subtracting c from x (t)1(t) obtaining a residual component r1(t) mixing r1(t) repeating the steps (1) and (2), judging whether the obtained new residual component needs to be decomposed, if so, repeating the steps, otherwise, stopping the decomposition.
(2) Carrying out weighting operation on the kurtosis value and the correlation coefficient of the IMF component: and after EMD decomposition is carried out on the vibration signals, the kurtosis value of each IMF component is calculated.
The correlation coefficient ρ reflects the closeness of the correlation between the noisy signal x (t) and the IMFs of the respective orders.
Figure BDA0003681086700000022
In the formula: c is a matrix
Figure BDA0003681086700000023
The covariance matrix of (a); and N is the number of sampling points of the signal.
(3) Removing residual error items after decomposition, and selecting IMF components with larger kurtosis values: by analyzing rho of IMF and x (t) of each order and combining the EMD method and the characteristics of noise, IMF components with larger correlation coefficients are selected to achieve the purpose of denoising. Since each order IMF is derived from x (t) decomposition, 0 < ρ < 1 should be used in most cases. However, it is found through experiments that when the signal-to-noise ratio of the noisy signal x (t) is large, a correlation coefficient between a certain high-order IMF obtained through decomposition and itself is smaller than 0. Since the relative variation trend of rub-impact AE signal energy and noise energy in each stage of IMF is reflected by the correlation coefficient, the absolute value of the correlation coefficient is obtained when the correlation coefficient is negative.
(4) Selecting IMF components with obvious characteristics by using the weighted values, wherein the calculation flow is as follows:
Figure BDA0003681086700000031
wherein Q represents the number of IMF components, ku{i}Representing the i-th IMF component kurtosis value.
T(i)＝0.8*ω(i)+0.2* (6)
Wherein, P(i)The correlation coefficient, T, representing the ith IMF component(i)A weight value representing the ith IMF component.
(5) Carrying out convolution operation: the first convolution layer uses a single large convolution kernel to carry out convolution operation, the robustness of the network model can be improved, and a multi-channel convolution module is added in a subsequent network structure.
(6) Designing network structure parameters: the MC-CNN comprehensively uses a single-channel convolution and a multi-channel convolution module, and can effectively extract fault characteristics.
(7) Selecting a model loss function and a training method: in the training process, an Adam self-adaptive optimizer is adopted to update network training parameters until an optimal solution is obtained, the size of a training batch is set to be 128, the iteration times are 160, and finally a Softmax classifier is used for achieving fault classification.
(8) Initializing the multichannel convolutional neural network and starting training: after the accuracy of the method is determined, a low signal-to-noise ratio signal is constructed to test the network structure and is transversely compared with other network structures.
(9) And (3) carrying out robustness test after completing the test by using the test set: on the basis of the original signal, signals corresponding to noise construction signal-to-noise ratios of-6, -4, -2 and 0 are added. An image dataset is derived using a weighted value method, with which different network models are trained. Experimental results show that the MC-CNN is higher than other network models under different signal-to-noise ratios, the superiority of a network structure is proved, meanwhile, the accuracy rate of 82.34% is still obtained under a-6 signal-to-noise ratio, and the method is proved to have certain robustness.
Compared with the prior art, the bearing fault diagnosis method based on the EMD and the multichannel convolutional neural network has the following advantages:
1. the deep learning method is applied to the field of equipment fault diagnosis, compared with the traditional method, the deep learning method can generally better extract the characteristics of a fault signal, and the mechanical fault diagnosis efficiency is improved.
2. Compared with a data set obtained by a kurtosis value method, the bearing fault diagnosis accuracy rate of the IMF data set obtained by the weighting method reaches 99.30%.
3. A new deep neural network model, namely a multi-channel convolutional neural network (MC-CNN), is provided, and is applied to fault diagnosis of a gear box vibration signal, so that the fault diagnosis capability is greatly improved.
4. The noise with different signal-to-noise ratios of the original signal structure is tested, and the MC-CNN is found to be obviously superior to other network structures, thereby proving that the method has good generalization capability.
Detailed Description
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
In the description of the present invention, it is to be understood that the terms "central," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like are used merely for convenience in describing and simplifying the description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, are not to be considered as limiting the invention. Furthermore, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or to implicitly indicate a number of the indicated technical features. Thus, a feature defined as "first," "second," etc. may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless otherwise specified.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; either directly or indirectly through intervening media, or through elements within both elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art through specific situations.
The present invention will be described in detail with reference to examples.
A bearing fault diagnosis method based on an EMD and a multi-channel convolution neural network comprises the following steps:
(1) The vibration signal is decomposed using EMD:
an arbitrary nonlinear signal is decomposed into a series of IMF signals and a remainder,
Figure BDA0003681086700000051
in the formula, n is the number of IMF components obtained by decomposition; c. Ci(t) is the ith IMF component; r is a radical of hydrogenn(t) is the final residual component.
Assuming that the signal to be decomposed is x (t), the specific algorithm flow is as follows:
the first step is as follows: and fitting by utilizing a spline interpolation method to obtain an upper envelope line and a lower envelope line.
The second step: subtracting the mean value m1 (t) of the upper envelope and the lower envelope by x (t) to obtain a component k1(t)
h1(t)＝x(t)-m1 (2)
The third step: judgment h1(t) whether IMF conditions are satisfied, if so, h1(t) is denoted as c1 as the first IMF. Otherwise, repeating the steps, and obtaining the result on the assumption that the kth time meets the condition
c1＝h1(k-1)(t)-m1 (3)
The fourth step: subtracting c from x (t)1(t) obtaining a residual component r1(t) mixing r1(t) repeating the steps (1) and (2), judging whether the obtained new residual component needs to be decomposed, if so, repeating the steps, otherwise, stopping the decomposition.
(2) Carrying out weighting operation on the kurtosis value and the correlation coefficient of the IMF component:
and after EMD decomposition is carried out on the vibration signals, the kurtosis value of each IMF component is calculated.
The correlation coefficient ρ reflects the closeness of the correlation between the noisy signal x (t) and the IMFs of the respective orders.
Figure BDA0003681086700000061
In the formula: c is a matrix
Figure BDA0003681086700000063
The covariance matrix of (a); and N is the number of sampling points of the signal.
(3) Removing residual error items after decomposition, and selecting IMF components with larger kurtosis values:
by analyzing rho of IMF and x (t) of each order and combining the EMD method and the characteristics of noise, IMF components with larger correlation coefficients are selected to achieve the purpose of denoising. Since each order of IMF is derived from x (t) decomposition, in most cases 0 < p < 1 should be used. However, it is found through experiments that when the signal-to-noise ratio of the noisy signal x (t) is large, a correlation coefficient between a certain high-order IMF obtained through decomposition and itself is smaller than 0. Since the relative variation trend of rub-impact AE signal energy and noise energy in each stage of IMF is reflected by the correlation coefficient, the absolute value of the correlation coefficient is obtained when the correlation coefficient is negative.
(4) Selecting IMF components with obvious characteristics by using the weighted values, wherein the calculation flow is as follows:
Figure BDA0003681086700000062
wherein, Q is shown in the tableIndicating the number of IMF components, ku{i}Representing the i-th IMF component kurtosis value.
T(i)＝0.8*ω(i)+0.2* (6)
Where ρ is(i)The correlation coefficient, T, representing the ith IMF component(i)A weight value representing the ith IMF component.
In the specific operation process, 2048 sampling points are taken as one section of signal, the same section of bearing sampling signal can be decomposed into multiple sections of vibration signals, each section of segmented vibration signal is subjected to EMD decomposition, vibration frequencies are sequentially arranged from high to low, and the selection of IMF components with obvious characteristics and containing fault characteristics is of great importance.
(5) Carrying out convolution operation:
the first convolution layer uses a single convolution kernel to carry out convolution operation, and a multi-channel convolution module is used in a subsequent network structure. The MC-CNN network model comprises two multi-channel convolution modules, wherein a global average pooling layer is added in front of a full connection layer, and a dropout layer and a classifier layer are connected finally, so that the characteristic extraction and fault classification of bearing vibration signals are realized.
(6) Designing network structure parameters:
the MC-CNN comprehensively uses a single-channel convolution and multi-channel convolution module, so that the robustness of a network model is improved while fault features are effectively extracted.
(7) Selecting a model loss function and a training method:
in the training process, an Adam self-adaptive optimizer is adopted to update network training parameters until an optimal solution is obtained, the size of a training batch is set to be 128, the iteration times are 160, and finally a Softmax classifier is used for achieving fault classification.
(8) Initializing a multichannel convolutional neural network and starting training by using a training set:
after the accuracy of the method is determined, a low signal-to-noise ratio signal is constructed to test the network structure and is transversely compared with other network structures
(9) And (3) performing robustness testing after the testing is completed by using the test set:
the whole technical route flow adds signals corresponding to noise structure signal-to-noise ratios of-6, -4, -2 and 0 on the basis of the original signals. An image dataset is derived using a weighted value method, with which different network models are trained. The experimental result shows that MC-CNN is higher than other network models under different signal-to-noise ratios, the superiority of a network structure is proved, and meanwhile, the accuracy of 82.34% is still achieved under-6 signal-to-noise ratios, and the method is proved to have certain robustness.
(10) Data visualization:
to further explore the process of MC-CNN feature extraction. Inputting 10 types of data into a CNN network structure, and performing dimension reduction visual presentation on each layer of extracted features through t-SNE. The first convolutional layer, softmax layer, can observe that the classification of the raw data is gradually clear after passing through the network model.
The present invention is not limited to the above preferred embodiments, and any modifications, equivalent substitutions, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (6)

1. A bearing fault diagnosis method based on EMD and a multi-channel convolutional neural network is characterized by comprising the following steps:
decomposing the nonlinear vibration signal into a series of IMF signals and a remainder using EMD;
calculating kurtosis values of the IMF components, and performing weighted operation on the kurtosis values and the correlation coefficients of the IMF components;
removing decomposed remainder, and selecting IMF component with larger kurtosis value;
selecting IMF components with obvious characteristics by using the weighted value;
and performing convolution operation.
2. The EMD and multichannel convolutional neural network-based bearing fault diagnosis method of claim 1, wherein the step of performing convolution operations comprises: the first convolution layer uses a single convolution kernel to carry out convolution operation, and a multi-channel convolution module is used in a subsequent network structure.
3. The EMD and multichannel convolutional neural network-based bearing fault diagnosis method of claim 2, wherein the step of performing convolution operations is followed by: and designing network structure parameters.
4. The EMD and multichannel convolutional neural network-based bearing fault diagnosis method of claim 3, wherein the step of designing the network structure parameters comprises: and selecting a model loss function and a training method.
5. The EMD and multi-channel convolutional neural network-based bearing fault diagnosis method of claim 4, wherein said selecting a model loss function and training method step is followed by: the multi-channel convolutional neural network is initialized and training begins using the training set.
6. The EMD and multi-channel convolutional neural network-based bearing fault diagnosis method of claim 5, wherein after the initializing the multi-channel convolutional neural network and starting the training step using the training set, comprising: and carrying out robustness testing after the testing is finished by utilizing the test set.
CN202210637458.2A 2022-06-07 2022-06-07 Bearing fault diagnosis method based on EMD and multichannel convolutional neural network Pending CN115270845A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210637458.2A CN115270845A (en) 2022-06-07 2022-06-07 Bearing fault diagnosis method based on EMD and multichannel convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210637458.2A CN115270845A (en) 2022-06-07 2022-06-07 Bearing fault diagnosis method based on EMD and multichannel convolutional neural network

Publications (1)

Publication Number Publication Date
CN115270845A true CN115270845A (en) 2022-11-01

Family

ID=83759622

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210637458.2A Pending CN115270845A (en) 2022-06-07 2022-06-07 Bearing fault diagnosis method based on EMD and multichannel convolutional neural network

Country Status (1)

Country Link
CN (1) CN115270845A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118094477A (en) * 2024-04-25 2024-05-28 山东大学 Signal processing method and system based on multi-sensor information fusion

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118094477A (en) * 2024-04-25 2024-05-28 山东大学 Signal processing method and system based on multi-sensor information fusion
CN118094477B (en) * 2024-04-25 2024-07-26 山东大学 Signal processing method and system based on multi-sensor information fusion

Similar Documents

Publication Publication Date Title
CN109946389B (en) Structural damage identification method based on ensemble empirical mode decomposition and convolutional neural network
US11630034B2 (en) Method for diagnosing and predicting operation conditions of large-scale equipment based on feature fusion and conversion
CN110516339B (en) Adaboost algorithm-based method for evaluating reliability of sealing structure in multiple failure modes
CN111340238A (en) Fault diagnosis method, device, equipment and storage medium of industrial system
CN111931983B (en) Precipitation prediction method and system
CN109655266B (en) Wind turbine generator bearing fault diagnosis method based on AVMD and spectrum correlation analysis
CN113505655A (en) Bearing fault intelligent diagnosis method for digital twin system
CN113723171B (en) Electroencephalogram signal denoising method based on residual error generation countermeasure network
CN111914705A (en) Signal generation method and device for improving health state evaluation accuracy of reactor
CN113076920B (en) Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model
CN115270845A (en) Bearing fault diagnosis method based on EMD and multichannel convolutional neural network
CN106502850B (en) A kind of the multiple faults discrimination method and device of sparse tight frame dictionary learning model
CN115204272A (en) Industrial system fault diagnosis method and equipment based on multi-sampling rate data
CN112414715B (en) Bearing fault diagnosis method based on mixed feature and improved gray level symbiosis algorithm
CN115114841A (en) U-net frame-based subsurface temperature field reconstruction method for high spatial-temporal resolution of edge sea
CN115046766A (en) Small sample bearing fault diagnosis method based on two-dimensional gray image self-adaptive subspace
CN112885368A (en) Multi-band spectral subtraction vibration signal denoising method based on improved capsule network
CN115034263A (en) Variable-scale evolution self-adaptive noise elimination method, noise elimination system and fault diagnosis system
CN115600088A (en) Distribution transformer fault diagnosis method based on vibration signals
CN114386452B (en) Nuclear power circulating water pump sun gear fault detection method
CN113703422B (en) Gas turbine pneumatic actuator fault diagnosis method based on feature analysis processing
CN116594061B (en) Seismic data denoising method based on multi-scale U-shaped attention network
CN117171544B (en) Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network
CN112906591A (en) Radar radiation source identification method based on multi-stage jumper residual error network
CN113657268B (en) Signal automatic decomposition method applied to wind turbine generator gearbox fault diagnosis

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