CN108388692B - Rolling bearing fault feature extraction method based on layered sparse coding - Google Patents

Rolling bearing fault feature extraction method based on layered sparse coding Download PDF

Info

Publication number
CN108388692B
CN108388692B CN201810045889.3A CN201810045889A CN108388692B CN 108388692 B CN108388692 B CN 108388692B CN 201810045889 A CN201810045889 A CN 201810045889A CN 108388692 B CN108388692 B CN 108388692B
Authority
CN
China
Prior art keywords
dictionary
fault
signal
time domain
sparse coding
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810045889.3A
Other languages
Chinese (zh)
Other versions
CN108388692A (en
Inventor
林京
梁凯旋
焦金阳
赵健
赵明
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.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN201810045889.3A priority Critical patent/CN108388692B/en
Publication of CN108388692A publication Critical patent/CN108388692A/en
Application granted granted Critical
Publication of CN108388692B publication Critical patent/CN108388692B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • 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

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The rolling bearing fault feature extraction method based on layered sparse coding comprises the steps of firstly carrying out high-frequency sampling on a bearing vibration signal, and intercepting a signal in a period of time as an original time domain signal; then constructing a fixed dictionary sparse coding model, setting relevant parameters, setting and applying the fixed dictionary sparse coding model, obtaining a sparse representation coefficient matrix through sparse coding, and obtaining a harmonic interference component irrelevant to the fault signal through multiplication of the fixed dictionary and the sparse coefficient matrix; filtering harmonic interference components from the original time domain signal to obtain an input time domain signal; then constructing a dictionary learning model based on K-SVD, inputting a time domain signal as the input of the model, and performing feature extraction by using the model according to parameter setting to obtain fault related features; performing Hilbert transform on the fault related characteristics, then performing fast Fourier transform to obtain a fault related characteristic envelope spectrum, and outputting a diagnosis result; the invention improves the signal-to-noise ratio and filters out irrelevant signals.

Description

Rolling bearing fault feature extraction method based on layered sparse coding
Technical Field
The invention relates to the technical field of rolling bearing fault diagnosis, in particular to a rolling bearing fault feature extraction method based on hierarchical sparse coding.
Background
The rolling bearing is the most common and universal rotating part in modern industry, the health condition of the rolling bearing has extremely important influence on the normal operation of whole equipment, and the rolling bearing has very important significance on effectively detecting and diagnosing early faults of the bearing. The rolling bearing vibration signal contains a large amount of bearing health state information, however, the information is usually interfered by some rotating parts (gears, shafts) and background noise, so it is important how to extract effective characteristic components from the signals to accurately judge whether the bearing is in fault and the fault degree line.
In recent years, a large number of signal processing methods are proposed for noise reduction and fault feature extraction of rolling bearing vibration signals, and a deep learning method (K-SVD) based on dictionary learning has attracted extensive attention and research in recent years.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a rolling bearing fault feature extraction method based on hierarchical sparse coding, which improves the signal-to-noise ratio and filters out irrelevant signals.
In order to achieve the purpose, the invention adopts the technical scheme that:
the rolling bearing fault feature extraction method based on layered sparse coding comprises the following steps:
adsorbing a vibration acceleration sensor on a bearing cover at the end part of a rolling bearing to be tested, carrying out high-frequency sampling on a bearing vibration signal to obtain vibration data, intercepting a signal within a period of time according to sampling frequency and rotating speed to directly serve as an original time domain signal, and recording as y (t);
constructing a fixed dictionary sparse coding model, and setting relevant parameters, wherein the relevant parameters comprise data sample segmentation length, dictionary atom number, iteration threshold, sparsity and a fixed dictionary;
thirdly, a fixed dictionary sparse coding model is applied according to the parameter setting in the second step, a sparse representation coefficient matrix is obtained through sparse coding, and a harmonic interference component delta (t) irrelevant to the fault signal is obtained through multiplication of the fixed dictionary and the sparse coefficient matrix;
step four: filtering harmonic interference component delta (t) obtained in the third step from original time domain signal y (t) to obtain time domain signal input in the next stage
Figure BDA0001550892670000023
Step five: constructing a dictionary learning model based on K-SVD, initializing a learning dictionary D, and setting the atom length and number of D, an iteration threshold, iteration times and sparsity;
step six: inputting the time domain signal obtained in the step four
Figure BDA0001550892670000024
As input of a dictionary learning model based on K-SVD, according to the parameter setting in the step five, the dictionary learning model based on K-SVD is used for feature extraction, firstly, a learning dictionary is fixed, a coefficient matrix is solved by using an orthogonal matching pursuit algorithm, then, the learning dictionary is updated column by using the SVD algorithm, and the step is repeated until an iteration condition is met; finally extracting fault related features
Figure BDA0001550892670000021
Step seven: for the fault-related characteristics in step six
Figure BDA0001550892670000022
And performing Hilbert transform and then performing fast Fourier transform to obtain a fault related characteristic envelope spectrum, and outputting a diagnosis result.
Compared with the prior art, the invention has the following beneficial effects:
a) the invention extracts the harmonic component irrelevant to the fault characteristic from the original signal by using the fixed dictionary and filters the harmonic component, overcomes the defect that the self-learning dictionary cannot distinguish the fault signal from the irrelevant harmonic component, and improves the accuracy of bearing fault diagnosis.
b) The invention utilizes the mode of combining the fixed dictionary and the self-learning dictionary in a layering way to extract the characteristic fault, and provides a new thought for the study of the dictionary learning and sparse coding technology in the field of fault diagnosis.
c) The invention does not need prior knowledge, has good robustness and is beneficial to the automation of the monitoring and diagnosing technology.
Drawings
FIG. 1 is a schematic structural diagram of a test bed according to an embodiment of the present invention.
Fig. 2 shows a failure of the inner ring of the rolling bearing according to the embodiment of the present invention.
FIG. 3 is a flow chart of the method of the present invention.
Fig. 4 is an original time domain signal according to an embodiment of the present invention.
Fig. 5 is an envelope spectrum of an original time-domain signal according to an embodiment of the present invention.
Fig. 6 shows the extracted fault-related features according to the embodiment of the present invention.
Fig. 7 is a fault-related feature envelope spectrum extracted according to an embodiment of the present invention.
FIG. 8 is a diagram of fault-related features extracted using a single K-SVD.
FIG. 9 is a fault-related feature envelope spectrum extracted using a single K-SVD.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
Taking a locomotive rolling bearing fault detection test bed of a certain vehicle section as an example, the rolling bearing test bed is composed of a driving motor 1, a driving wheel 2, a wheel pair 3, a rolling bearing 4 and a rolling bearing 5, as shown in fig. 1, the driving motor 1 drives the driving wheel 2 to rotate, the driving wheel 2 contacts with an outer ring of the rolling bearing 3 to be tested and drives the outer ring to rotate, and the rolling bearing 5 and the wheel pair 4 are fixed.
The specific parameters are as follows: 1) contact angle of rolling bearing 3: 9 degrees; 2) rolling element diameter of rolling bearing 3: 23.775 mm; 3) number of rolling elements of rolling bearing 3: 20, the number of the cells is 20; 4) the pitch diameter of the rolling bearing 3 is: 180 mm; 5) the failure type of the rolling bearing 3 is inner ring peeling failure, as shown in fig. 2; 6) the vibration acceleration sensor M is arranged on a bearing cover at the end part of the shaft; 7) the test system carries out high-frequency sampling and data storage on the bearing vibration signal, the frequency of the sampling process is 76800Hz, and the sampling time is 15 s.
The invention is applied to the analysis and determination of the bearing fault type of the original data and the comparison with the single dictionary learning method.
As shown in fig. 3, the rolling bearing fault feature extraction method based on hierarchical sparse coding includes the following steps:
firstly, a vibration acceleration sensor is attached to a bearing cover at the end part of a rolling bearing 3 to be tested, a bearing vibration signal is subjected to high-frequency sampling to obtain vibration data, 1s of data is cut as an original time domain signal and is recorded as y (t) according to the rotating speed and sampling frequency set during data acquisition, the original time domain signal is shown in a graph 4 in the embodiment, an envelope spectrum of the original time domain signal is shown in a graph 5, and obvious fault characteristics cannot be found in the two graphs;
step two, constructing a fixed dictionary sparse coding model, and setting relevant parameters, wherein the relevant parameters comprise a sample segmentation length n0Dictionary number of atoms m0Iteration threshold0Degree of sparsity T0Fixed dictionary D0；
Thirdly, a fixed dictionary sparse coding model is applied according to the parameter setting in the second step, and a sparse representation coefficient matrix X is obtained through sparse coding0By means of a fixed dictionary D0And sparse coefficient matrix X0Multiplying to obtain a harmonic interference component delta (t) irrelevant to the fault signal;
the fixed dictionary-based sparse coding model is used, and the following optimization problems are solved:
Figure BDA0001550892670000051
in the formula (1), Y is the sample division length n0A signal matrix obtained by shift-dividing the original time-domain signal y (t), D0For fixing dictionaries, X0In order to sparsely characterize the coefficient matrix,0to iterate threshold values, xjRepresents X0J th line of (1), T0Is the sparsity; the goal of equation (1) is to search for the optimal sparse representation coefficient matrix X0And solving the optimization problem of the formula (1) by using a matching pursuit optimization algorithm to obtain a sparse representation coefficient matrix X0；
Step four: inputting the harmonic interference component delta (A) obtained in the third stept) and filtering harmonic interference component delta (t) obtained in the third step from the original time domain signal y (t) to obtain the time domain signal input in the next stage
Figure BDA0001550892670000052
Step five: constructing a dictionary learning model based on K-SVD, initializing a learning dictionary D, and setting the atom length n and number m of D, an iteration threshold, iteration times Item and sparsity T;
step six: the time domain signal obtained in the step four
Figure BDA0001550892670000053
As input of a dictionary learning model based on K-SVD, according to the parameter setting in the step five, the dictionary learning model based on K-SVD is used for feature extraction, firstly, a learning dictionary is fixed, a coefficient matrix X is solved by using an orthogonal matching pursuit algorithm, then, the learning dictionary is updated column by using the SVD algorithm, and the step is repeated until an iteration condition is met; finally extracting fault related features
Figure BDA0001550892670000054
The dictionary learning model based on the K-SVD is used for solving the following optimization problems:
Figure BDA0001550892670000061
in the formula (2), the reaction mixture is,
Figure BDA0001550892670000062
taking sample segmentation length n to obtain
Figure BDA0001550892670000063
Signal matrix obtained by shift division, xiRepresents the ith row of X. Firstly, assuming that a learning dictionary D is fixed, converting the learning dictionary D into an optimal coefficient matrix X for searching, and solving the optimization problem of a formula (2) by using a matching pursuit optimization algorithm to obtain a sparse representation coefficient matrix:
Figure BDA0001550892670000064
wherein
Figure BDA0001550892670000065
Represents
Figure BDA0001550892670000066
I column of (2), xiRepresents the ith column of X. N represents
Figure BDA0001550892670000067
The number of columns;
then updating the learning dictionary atom d column by columnkAnd the coefficients of the corresponding row in X
Figure BDA0001550892670000068
The rewrite equation (2) is as follows:
Figure BDA0001550892670000069
wherein EkIs that the k atom is set to zero by the learning dictionary D and then the signal is reconstructed
Figure BDA00015508926700000616
An error of (2);
definition of
Figure BDA00015508926700000610
φkIs the size Nx | omegakMatrix of i, i.e. at (ω)k(i) I) has the value 1 and the remainder 0, so that the formula (4) can again be written as
Figure BDA00015508926700000611
Wherein
Figure BDA00015508926700000612
Is EkIs reducedSubtracting the matrix;
to pair
Figure BDA00015508926700000613
Performing singular value decomposition to obtain
Figure BDA00015508926700000614
Further updating the corresponding column of the learning dictionary atom d by using the first column of UkUpdating the coefficient by multiplying the first column of V by Δ (1,1)
Figure BDA00015508926700000615
Repeating the steps until all the learning dictionary atoms in the K columns are updated;
after obtaining D and X through the dictionary learning model based on K-SVD, equation (2) is simplified as follows:
Figure BDA0001550892670000074
λ is the Lagrange multiplier, RiIs a weight operator extracted from Z, ZiColumn i of Z;
obtaining fault correlation characteristics by optimized closed solving of quadratic problem
Figure BDA0001550892670000071
Figure BDA0001550892670000072
Step seven: for the fault related characteristics obtained in the sixth step
Figure BDA0001550892670000073
And performing Hilbert transform and then performing fast Fourier transform to obtain a fault related characteristic envelope spectrum, judging the fault type, and outputting a diagnosis result.
The fault-related characteristic and the fault-related characteristic envelope spectrum obtained in the embodiment are shown in fig. 6 and 7, and the fault-related characteristic envelope spectrum is a harmonic of the inner ring fault characteristic frequency and is matched with the fault characteristic, so that the fault can be accurately diagnosed. While the traditional single self-learning dictionary extracts rolling bearing fault-related features and fault-related feature envelope spectrums are shown in fig. 8 and 9, and faults occurring in the inner ring cannot be found.
The rolling bearing fault feature extraction method based on hierarchical sparse coding overcomes the defects of a single learning dictionary and a fixed dictionary, accurately extracts fault feature information, effectively diagnoses faults, and has good robustness.

Claims (1)

1. The rolling bearing fault feature extraction method based on layered sparse coding is characterized by comprising the following steps of:
adsorbing a vibration acceleration sensor on a bearing cover at the end part of a rolling bearing to be tested, carrying out high-frequency sampling on a bearing vibration signal to obtain vibration data, intercepting a signal within a period of time according to sampling frequency and rotating speed to directly serve as an original time domain signal, and recording as y (t);
constructing a fixed dictionary sparse coding model, and setting relevant parameters, wherein the relevant parameters comprise data sample segmentation length, dictionary atom number, iteration threshold, sparsity and a fixed dictionary;
thirdly, a fixed dictionary sparse coding model is applied according to the parameter setting in the second step, a sparse representation coefficient matrix is obtained through sparse coding, and a harmonic interference component delta (t) irrelevant to the fault signal is obtained through multiplication of the fixed dictionary and the sparse coefficient matrix;
step four: filtering harmonic interference component delta (t) obtained in the third step from original time domain signal y (t) to obtain time domain signal input in the next stage
Figure FDA0001550892660000011
Step five: constructing a dictionary learning model based on K-SVD, initializing a learning dictionary D, and setting the atom length and number of D, an iteration threshold, iteration times and sparsity;
step six: inputting the time domain signal obtained in the step four
Figure FDA0001550892660000012
As input of a dictionary learning model based on K-SVD, according to the parameter setting in the step five, the dictionary learning model based on K-SVD is used for feature extraction, firstly, a dictionary is fixed, a coefficient matrix is solved by using an orthogonal matching pursuit algorithm, then, the dictionary is updated column by using the SVD algorithm, and the step is repeated until an iteration condition is met; finally extracting fault related features
Figure FDA0001550892660000013
Step seven: for the fault-related characteristics in step six
Figure FDA0001550892660000014
And performing Hilbert transform and then performing fast Fourier transform to obtain a fault related characteristic envelope spectrum, and outputting a diagnosis result.
CN201810045889.3A 2018-01-17 2018-01-17 Rolling bearing fault feature extraction method based on layered sparse coding Active CN108388692B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810045889.3A CN108388692B (en) 2018-01-17 2018-01-17 Rolling bearing fault feature extraction method based on layered sparse coding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810045889.3A CN108388692B (en) 2018-01-17 2018-01-17 Rolling bearing fault feature extraction method based on layered sparse coding

Publications (2)

Publication Number Publication Date
CN108388692A CN108388692A (en) 2018-08-10
CN108388692B true CN108388692B (en) 2020-12-29

Family

ID=63077048

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810045889.3A Active CN108388692B (en) 2018-01-17 2018-01-17 Rolling bearing fault feature extraction method based on layered sparse coding

Country Status (1)

Country Link
CN (1) CN108388692B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109060350A (en) * 2018-09-04 2018-12-21 重庆大学 A kind of Rolling Bearing Fault Character extracting method dictionary-based learning
CN109117896B (en) * 2018-09-28 2020-07-24 西安交通大学 Rolling bearing fault feature extraction method based on KSVD dictionary learning
CN110580471B (en) * 2019-09-12 2021-11-02 北京航空航天大学 Mechanical equipment fault diagnosis method based on encoder signal transient characteristics
CN111382792B (en) * 2020-03-09 2022-06-14 兰州理工大学 Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation
CN112215196B (en) * 2020-10-26 2024-04-02 杭州电子科技大学 Electrocardiogram identity recognition method
CN113310684B (en) * 2021-04-20 2022-05-24 东南大学 Gearbox fault feature extraction method based on scale space and improved sparse representation
CN113740055B (en) * 2021-07-14 2022-08-09 西安交通大学 Method and device for separating and diagnosing composite fault components of gear box
CN116361727A (en) * 2023-03-28 2023-06-30 重庆大学 Audio feature and SRC-Adaboost-based battery power conversion system driving gear fault diagnosis method
CN116558828B (en) * 2023-07-10 2023-09-15 昆明理工大学 Rolling bearing health state assessment method based on autocorrelation coefficient sparsity characteristic

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS57194331A (en) * 1981-05-26 1982-11-29 Nippon Seiko Kk Method and device for detecting fault of machine element having tumbling body
WO2006106508A2 (en) * 2005-04-04 2006-10-12 Technion Research & Development Foundation Ltd. System and method for designing of dictionaries for sparse representation
CN105424359A (en) * 2015-11-25 2016-03-23 华南理工大学 Sparse-decomposition-based hybrid fault feature extraction method of gear wheel and bearing

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107543722A (en) * 2017-08-18 2018-01-05 西安交通大学 The Rolling Bearing Fault Character extracting method of dictionary learning is stacked based on depth

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS57194331A (en) * 1981-05-26 1982-11-29 Nippon Seiko Kk Method and device for detecting fault of machine element having tumbling body
WO2006106508A2 (en) * 2005-04-04 2006-10-12 Technion Research & Development Foundation Ltd. System and method for designing of dictionaries for sparse representation
CN105424359A (en) * 2015-11-25 2016-03-23 华南理工大学 Sparse-decomposition-based hybrid fault feature extraction method of gear wheel and bearing

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Harmonics-to-noise ratio guided deconvolution and its application for bearing fault detection;Liang kaixuan;《IEEE Xplore》;20171023;第1-7页 *
大数据下机械智能故障诊断的机遇与挑战;林京等;《机械工程学报》;20170822;第54卷(第5期);第94-100页 *
感知压缩中音频稀疏表示的研究;王月明;《中国优秀硕士学位论文全文数据库 信息科技辑》;20150615(第6期);第I136-38页 *

Also Published As

Publication number Publication date
CN108388692A (en) 2018-08-10

Similar Documents

Publication Publication Date Title
CN108388692B (en) Rolling bearing fault feature extraction method based on layered sparse coding
CN109460618B (en) Rolling bearing residual life online prediction method and system
CN109211546B (en) Rotary machine fault diagnosis method based on noise reduction automatic encoder and increment learning
CN113375939B (en) Mechanical part fault diagnosis method based on SVD and VMD
CN112906644A (en) Mechanical fault intelligent diagnosis method based on deep migration learning
CN104215456B (en) Plane clustering and frequency-domain compressed sensing reconstruction based mechanical fault diagnosis method
CN111024398B (en) Deconvolution method for maximum correlation kurtosis without period
CN109117896B (en) Rolling bearing fault feature extraction method based on KSVD dictionary learning
CN111665050B (en) Rolling bearing fault diagnosis method based on clustering K-SVD algorithm
Kong et al. Sparse learning based classification framework for planetary bearing health diagnostics
CN113776835A (en) Rolling bearing fault diagnosis method based on enhanced Mel linear frequency cepstrum coefficient
CN110646199B (en) Gear weak fault diagnosis method based on weighted derivative dynamic time warping
CN103926097B (en) Method for collecting and extracting fault feature information of low-speed and heavy-load device
CN111678691A (en) Gear fault detection method based on improved sparse decomposition algorithm
CN114263621A (en) Test method and system for diagnosing and simulating cavitation fault of centrifugal pump
CN113865872A (en) Bearing fault diagnosis method based on wavelet packet reconstruction imaging and CNN
CN110044619B (en) Multi-fault feature identification method based on sparse multi-cycle group lasso
CN113281047A (en) Bearing inner and outer ring fault quantitative trend diagnosis method based on variable-scale Lempel-Ziv
CN110231165B (en) Mechanical equipment fault diagnosis method based on expectation difference constraint confidence network
CN111649855A (en) Dynamometer signal acquisition method based on compressive sensing
CN111855192A (en) Singular value decomposition method for denoising encoder signal
Zhang et al. Weighted sparsity-based denoising for extracting incipient fault in rolling bearing
CN115146673A (en) Multi-dimensional data compression method for fault diagnosis of rolling bearing
CN115326396A (en) Bearing fault diagnosis method and device
CN114136604A (en) Rotary equipment fault diagnosis method and system based on improved sparse dictionary

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant