CN110160786A - A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer - Google Patents
A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer Download PDFInfo
- Publication number
- CN110160786A CN110160786A CN201910510895.6A CN201910510895A CN110160786A CN 110160786 A CN110160786 A CN 110160786A CN 201910510895 A CN201910510895 A CN 201910510895A CN 110160786 A CN110160786 A CN 110160786A
- Authority
- CN
- China
- Prior art keywords
- particle
- small echo
- bearing
- particle swarm
- swarm optimizer
- 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
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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of bearing fault classification methods based on small echo Mutation Particle Swarm Optimizer, include the following steps: S1, obtain bearing initial data, extract the energy feature of the bearing initial data；S2, energy feature is inputted into the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer；S3, failure modes result is obtained.The present invention can be improved the nicety of grading of detection bearing fault, provide convenience with detection for bearing operation.
Description
Technical field
The present invention relates to bearing fault classification field more particularly to a kind of bearing events based on small echo Mutation Particle Swarm Optimizer
Hinder classification method.
Background technique
With the development of the times with economic prosperity, the rapid development of China's bearing industry, the kind of bearing from less to more, produced
From low to high, industry size from small to large, has formd that product class is substantially complete, production distribution for quality and technical level
Relatively reasonable professional production system.Bearing is as important components a kind of in contemporary industry mechanical equipment, its main function
Can support mechanical rotary body, reduce the coefficient of friction in its motion process, and guarantee its rotating accuracy.But the mechanical axis in China
It holds manufacture view and remains many problems, the bearing industry production capacity in China is lower at present, most bearings manufacturer
From foreign countries, theoretical basis ability is weaker for bearing industry, and R & D Level is not high.
The design and fabrication technology basic source in current China in the imitation to foreign technology, and manufacturing technology level compared with
Low, China's bearing industry manufacturing process and equipment Technology develop slowly, and Vehicle Processing numerical control rate is low.These reasons cause to cause
Bearing process capability index is low, and consistency is poor, and product processing dimension dispersion is big, influences axis because product inherent quality is unstable
Precision, the performance, life and reliability held, but bearing plays essential effect in mechanical movement, so discovery in time
Failure in bearing distinguishes normal bearing and all kinds of faulty bearings, becomes an essential research.
In conclusion the problem of according to encountering in above-mentioned bearing use process, the invention discloses one kind to be become based on small echo
The bearing fault classification method of different particle swarm algorithm can be improved the nicety of grading of detection bearing fault, runs and examines for bearing
It surveys and convenience is provided.
Summary of the invention
In view of the above shortcomings of the prior art, the technical problem to be solved by the present invention is how to improve detection bearing
The nicety of grading of failure provides convenience with detection for bearing operation.
In order to solve the above technical problems, present invention employs the following technical solutions:
A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer, includes the following steps:
S1, bearing initial data is obtained, extracts the energy feature of the bearing initial data；
S2, energy feature is inputted into the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer；
S3, failure modes result is obtained.
Preferably, the training side of the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer
Method includes:
S101, training dataset is obtained, training dataset is inputted to be trained based on small echo Mutation Particle Swarm Optimizer
Least square method supporting vector machine disaggregated model；
S102, evolutionary generation, the Studying factors for initializing population, if the initial position of i-th of particle is xi, i-th
The initial velocity of particle is vi, the optimal historical position that i-th of particle reaches is pi, the original position that each particle is arranged is
Desired positions, adaptive optimal control degree correspond to the desired positions of each particle；
S103, the velocity information and location information for calculating each particle flight, and generate next-generation new population；
S104, the scale parameter a for finding out small echo variation function and wavelet function value σ, and according to wavelet function formula to working as
Mutation operation is carried out for optimal particle；
S105, the fitness for calculating particle is substituted into fitness function, and p is updated according to the fitness value of particlebestWith
gbest, while the velocity information and location information of more new particle；
S106, judge whether algorithm calculated result reaches optimizing condition, if reaching, save penalty coefficient C and Gauss diameter
To kernel function R, the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer after training is constructed, if not
Meet, returns to step 104.
In conclusion the invention discloses a kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer, packet
It includes following steps: S1, obtaining bearing initial data, extract the energy feature of the bearing initial data；It is S2, energy feature is defeated
Enter the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer；S3, failure modes result is obtained.This hair
The bright nicety of grading that can be improved detection bearing fault provides convenience with detection for bearing operation.
Detailed description of the invention
Fig. 1 is a kind of process of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer disclosed in the present application
Figure.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
The detailed description explanation of one step.
A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer, includes the following steps:
S1, bearing initial data is obtained, extracts the energy feature of the bearing initial data；
In the present invention, initial data included it is outer split, implosion, abrasion, the data such as speed under miss status.
S2, energy feature is inputted into the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer；
S3, failure modes result is obtained.
Compared with conventional art, the device have the advantages that as follows:
Wavelet function is added in this method in particle swarm algorithm, and the fine adjustment function of the function can effectively adjustment algorithm exist
Later period is easily trapped into the problem of local optimum, so can carry out better optimizing operation with the important parameter to disaggregated model.
Wavelet function disturbs each dimension for forming contemporary optimal particle, will disturb result as selected variation
The position of particle, so that algorithm computation complexity decreases.
In the operational process of entire algorithm, Riming time of algorithm is shortened, improves classification effectiveness.
On the basis of the present invention, the parameter of least square method supporting vector machine disaggregated model is optimized, according to reality
The a variety of fault data of the bearing on border, improves the nicety of grading of algorithm.
To advanced optimize above-mentioned technical proposal, the least square supporting vector based on small echo Mutation Particle Swarm Optimizer
The training method of machine disaggregated model includes:
S101, training dataset is obtained, training dataset is inputted to be trained based on small echo Mutation Particle Swarm Optimizer
Least square method supporting vector machine disaggregated model；
In the present invention, data set is divided into two parts of training dataset and test data set, training dataset is used to seek
Find out the boundary of classification, test data set be used to test the categorised demarcation line trained nicety of grading how.
S102, evolutionary generation, the Studying factors for initializing population, if the initial position of i-th of particle is xi, i-th
The initial velocity of particle is vi, the optimal historical position that i-th of particle reaches is pi, the original position that each particle is arranged is
Desired positions, adaptive optimal control degree correspond to the desired positions of each particle；
By the initial velocity v of particleiWith initial position xiIt is defined as in LSSVM disaggregated model need to optimize two respectively
Parameter penalty coefficient C and Gauss radial direction kernel function R.
S103, the velocity information and location information for calculating each particle flight, and generate next-generation new population；
S104, the scale parameter a for finding out small echo variation function and wavelet function value σ, and according to wavelet function formula to working as
Mutation operation is carried out for optimal particle；
S105, the fitness for calculating particle is substituted into fitness function, and p is updated according to the fitness value of particlebestWith
gbest, while the velocity information and location information of more new particle；
The fitness of particle is defined as training precision.The calculation method of training precision: the data of input are tape label number
According to, every one kind data will be input in disaggregated model to different labels (such as: 0.1.2.3.4) is taken as training dataset,
It is trained using the mathematical model of LSSVM " categorised demarcation line ", is that can't see data label when training this " categorised demarcation line "
, after training obtains categorised demarcation line, trained data substitution is tested, data are tape labels at this time, thus may be used
To check that how many data classification is correct, how many classification error, and trained nicety of grading is calculated, it is calculated as population
The particle fitness of method.
Using training precision as the fitness of particle, the higher the better for precision, then in the cyclic process of particle swarm algorithm,
It calculated fitness will be compared each time, and update and save best fitness (the namely highest training essence of precision
Degree), it is also just corresponding to save two parametric speed v for calculating degree of being preferably adapted toiAnd displacement xi(namely penalty coefficient C and Gauss
Radial kernel function R), ptsebAnd gbestTo be used to store speed v corresponding to best fitnessiAnd displacement xiValue, every circulation
Once, the fitness of all more current fitness and last time obtains best fitness and corresponding speed viAnd displacement xi, and
It is stored in pbestAnd gbestIn.
S106, judge whether algorithm calculated result reaches optimizing condition, if reaching, save penalty coefficient C and Gauss diameter
To kernel function R, the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer after training is constructed, if not
Meet, returns to step 104.
The training nicety of grading of defining classification model is the fitness in particle swarm algorithm, according to the classification of training result essence
Spend the output to determine calculated result.If optimal with the calculated fitness of this particle, current particle is saved as optimal
Particle, if this time fitness be not it is optimal, save last round-robin algorithm and export optimized parameter after circulation terminates, until circulation
Terminate output optimal particle.Penalty coefficient C and Gauss radial direction kernel function R is saved, LSSVM disaggregated model is constructed.
Particle swarm algorithm (Particle Swarm Optimization, PSO) is according in research flock of birds crawler behavior
On the basis of, it is saved and is shared using the information that each individual is attempted to explore in group, by comparing itself and other
The information of individual finds optimal solution.Particle swarm algorithm is also to find optimal solution by iteration, it is also logical from RANDOM SOLUTION
Fitness is crossed to evaluate the fitness size of solution.In the present invention, a particle rapidity and position are defined during analogue simulation
The bound for moving size can select an initial value using random function within the scope of the bound of parameter when analogue simulation, from
And complete the initialization of particle.Particle swarm algorithm can be expressed as follows: in D dimension space, there is m particle, the position of i-th of particle
Are as follows: xi=(xi1,xi2,……xid), the speed of i-th of particle are as follows: vi=(vi1,vi2,……vid), wherein 1≤i≤m, 1
≤ d≤D, the history desired positions that i-th of particle reaches are as follows: pi=(pi1,pi2,……pid), basic particle group algorithm can be by
Following formula indicates:
What k was indicated is current iteration number, in formula (1), c1, c2For Studying factors or acceleration factor, r1, r2For with
Machine number, value range are [0,1].In the present invention, it on the basis of standard particle group's algorithm, is carried out in contemporary particle by selection
When variation, the wavelet function of (3) formula is used to disturb each dimension of the composition present age optimal particle, will disturbance result as
The new position of selected variation particle, calculation formula are as follows:
WhereinFor the particle newly to make a variation selected,It is global optimum's particle in m generation, σ is small echo letter
Numerical value,
Wherein ψ (x) is wavelet function, and a is scale parameter, selects Morlet function as wavelet basis, concrete form is such as
Under:
In this way, the calculating of (4) formula wavelet function value σ may be expressed as:
Wavelet function be more than 99% energy be included in [- 2.5,2.5] between, so in (6) formula a value range be [-
2.5a, 2.5a] between pseudo random number.The calculation formula of scale parameter a are as follows:
ζwmFor the form parameter of monotonically increasing function, t is current iteration number, and T is maximum number of iterations, and g is the upper of a
Limit, the present invention improve mutation algorithm strategy, when certain variation selected for particle, use the wavelet function pair of (6) formula
Each dimension for forming contemporary optimal particle is disturbed, and will disturb result as the position of selected variation particle, (3) formula is only
Small wave disturbance is carried out to global extremum, computation complexity decreases.Global optimum's particle of every generation particle is all particles
The middle preferable particle of performance, searches for a part of particle in global optimum's particle periphery by control, on the one hand makes full use of most
The information of excellent particle guides other particles close to optimal particle, accelerates convergence rate；On the other hand pass through the fine tuning function of small echo
Can, it effectively prevent algorithm to fall into local optimum.
As soon as two parameters can all define a bound at the time of initialization, the parameter of initialization is sharp within this range
It is defined with random function, then carries out the parameter optimization of particle swarm algorithm again, optimizing when is equally also the model in parameter
Within enclosing, by the way that the number of iterations is arranged, the numerical value of two parameters is replaced ceaselessly to find optimal parameter.
Least square method supporting vector machine disaggregated model (least squares support vector machine,
LSSVM)
Assuming that the model of classification samples is { (x1,y1),(x2,y2),……(xl,yl) training set, then it is existing optimal
It is necessary to meet following condition for plane of classifying:
In formula: ω is the normal vector of hyperplane；β is amount of bias.Then categorised decision function are as follows:
f(xi)=sgn (ωTxi+β) (9)
Then the disaggregated model of least square SVM can be obtained by solving disaggregated model:
In formula: C is penalty coefficient；eiFor error variance.In order to admit of certain mistake point rate, by minimum error sample and
Maximum class interval compromise considers.Formula (10) needs to meet following equality constraint:
yi(ωTφ(xi)+β)=1-ei (11)
φ (x in formulai) it is Nonlinear Mapping, sample set is mapped to from the input space feature space of higher-dimension.Building
Lagrange equation are as follows:
α in formulaiFor Lagrange multiplier.Optimal condition are as follows:
Formula (12) and formula (13) simultaneous are obtained into following linear equation:
In formula: matrix Ω=yiyjφ(xi)φ(xy)=yiyjK(xi,xj), wherein K (xi,xj) it is support vector machines core letter
Number (being equal to R), j=1,2 ... ..l；yT=[y1,y2,……yl]；I is unit matrix；α=[α1, α2... ... αl], then most
Small two multiply support vector machines categorised decision function beX in formula is the data of input
Sample, y are tag along sort.
Finally, it is stated that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although passing through ginseng
According to the preferred embodiment of the present invention, invention has been described, it should be appreciated by those of ordinary skill in the art that can
To make various changes to it in the form and details, without departing from the present invention defined by the appended claims
Spirit and scope.
Claims (2)
1. a kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer, which comprises the steps of:
S1, bearing initial data is obtained, extracts the energy feature of the bearing initial data；
S2, energy feature is inputted into the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer；
S3, failure modes result is obtained.
2. as described in claim 1 based on the bearing fault classification method of small echo Mutation Particle Swarm Optimizer, which is characterized in that institute
The training method for stating the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer includes:
S101, training dataset is obtained, training dataset is inputted into the minimum to be trained based on small echo Mutation Particle Swarm Optimizer
Two multiply support vector cassification model；
S102, evolutionary generation, the Studying factors for initializing population, if the initial position of i-th of particle is xi, i-th particle
Initial velocity is vi, the optimal historical position that i-th of particle reaches is pi, the original position that each particle is arranged is best position
It sets, adaptive optimal control degree corresponds to the desired positions of each particle；
S103, the velocity information and location information for calculating each particle flight, and generate next-generation new population；
S104, the scale parameter a for finding out small echo variation function and wavelet function value σ, and according to wavelet function formula to the present age most
Excellent particle carries out mutation operation；
S105, the fitness for calculating particle is substituted into fitness function, and p is updated according to the fitness value of particlebestAnd gbest,
The velocity information and location information of more new particle simultaneously；
S106, judge whether algorithm calculated result reaches optimizing condition, if reaching, save penalty coefficient C and Gauss radial kernel
Function R, the least square method supporting vector machine disaggregated model based on small echo Mutation Particle Swarm Optimizer after building training, if discontented
Foot, returns to step 104.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910510895.6A CN110160786A (en) | 2019-06-13 | 2019-06-13 | A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910510895.6A CN110160786A (en) | 2019-06-13 | 2019-06-13 | A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110160786A true CN110160786A (en) | 2019-08-23 |
Family
ID=67628690
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910510895.6A Pending CN110160786A (en) | 2019-06-13 | 2019-06-13 | A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110160786A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103066595A (en) * | 2012-12-26 | 2013-04-24 | 中国电力科学研究院 | Optimization method of extra-high voltage transient stability control |
| CN105046057A (en) * | 2015-06-24 | 2015-11-11 | 上海大学 | LSSVM (Least Square Support Vector Machine) pulsation wind speed prediction method based on Morlet wavelet kernel |
| CN107449603A (en) * | 2016-05-31 | 2017-12-08 | 华北电力大学（保定） | Fault Diagnosis of Fan method based on SVMs |
| CN108875788A (en) * | 2018-05-23 | 2018-11-23 | 东南大学 | A kind of SVM classifier parameter optimization method based on modified particle swarm optiziation |
-
2019
- 2019-06-13 CN CN201910510895.6A patent/CN110160786A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103066595A (en) * | 2012-12-26 | 2013-04-24 | 中国电力科学研究院 | Optimization method of extra-high voltage transient stability control |
| CN105046057A (en) * | 2015-06-24 | 2015-11-11 | 上海大学 | LSSVM (Least Square Support Vector Machine) pulsation wind speed prediction method based on Morlet wavelet kernel |
| CN107449603A (en) * | 2016-05-31 | 2017-12-08 | 华北电力大学（保定） | Fault Diagnosis of Fan method based on SVMs |
| CN108875788A (en) * | 2018-05-23 | 2018-11-23 | 东南大学 | A kind of SVM classifier parameter optimization method based on modified particle swarm optiziation |
Non-Patent Citations (3)
| Title |
|---|
| S.H.LING ET AL.: "Hybrid Particle Swarm Optimization With Wavelet Mutation and Its Industrial Applications", 《IEEE TRANSACTIONSON SYSTEMS,MAN,AND CYBERNETICS-PART B:CYBERNETICS》 * |
| 吴迪: "基于小波理论和改进PSO-LSSVM的电机轴承故障诊断", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 张超: "基于Morlet小波变异的粒子群优化算法", 《江汉大学学报（自然科学版）》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111738512B (en) | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model | |
| WO2019067931A1 (en) | Gradient-based auto-tuning for machine learning and deep learning models | |
| CN105550374A (en) | Random forest parallelization machine studying method for big data in Spark cloud service environment | |
| CN109472321A (en) | A kind of prediction towards time series type surface water quality big data and assessment models construction method | |
| CN108051660A (en) | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method | |
| CN103258210B (en) | A kind of high-definition image classification method based on dictionary learning | |
| US11087344B2 (en) | Method and system for predicting and indexing real estate demand and pricing | |
| Tang et al. | Variable predictive model class discrimination using novel predictive models and adaptive feature selection for bearing fault identification | |
| CN109871809A (en) | A kind of machine learning process intelligence assemble method based on semantic net | |
| CN109961093A (en) | A kind of image classification method based on many intelligence integrated studies | |
| CN113743474B (en) | Digital picture classification method and system based on collaborative semi-supervised convolutional neural network | |
| CN109543693A (en) | Weak labeling data noise reduction method based on regularization label propagation | |
| CN110348287A (en) | A kind of unsupervised feature selection approach and device based on dictionary and sample similar diagram | |
| CN114897451B (en) | Double-layer clustering correction method and device considering key features of demand response user | |
| CN111191685A (en) | Method for dynamically weighting loss function | |
| CN114818809B (en) | Fault diagnosis method for SCA-SVM motor rolling bearing based on cross wavelet | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Gao et al. | Act: Empowering decision transformer with dynamic programming via advantage conditioning | |
| CN110160786A (en) | A kind of bearing fault classification method based on small echo Mutation Particle Swarm Optimizer | |
| US11468352B2 (en) | Method and system for predictive modeling of geographic income distribution | |
| Griffiths et al. | Finding your (3D) center: 3D object detection using a learned loss | |
| Zhou et al. | Effective vision transformer training: A data-centric perspective | |
| US20230004870A1 (en) | Machine learning model determination system and machine learning model determination method | |
| CN113642632B (en) | Power system customer classification method and device based on self-adaptive competition and equalization optimization | |
| Hussein et al. | The Patch‐Levy‐Based Bees Algorithm Applied to Dynamic Optimization Problems |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190823 |