CN106197999B - A kind of planetary gear method for diagnosing faults - Google Patents
A kind of planetary gear method for diagnosing faults Download PDFInfo
- Publication number
- CN106197999B CN106197999B CN201610537475.3A CN201610537475A CN106197999B CN 106197999 B CN106197999 B CN 106197999B CN 201610537475 A CN201610537475 A CN 201610537475A CN 106197999 B CN106197999 B CN 106197999B
- Authority
- CN
- China
- Prior art keywords
- planetary gear
- neural network
- neuron
- imf
- function
- 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
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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/021—Gearings
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/028—Acoustic or vibration analysis
Abstract
The planetary gear method for diagnosing faults of the invention discloses a kind of complete overall experience mode decomposition and fuzzy entropy based on adaptive noise.First original signal decompose using the complete overall experience mode decomposition method of adaptive noise and obtains the complete intrinsic mode function of high quality, for each intrinsic mode function, a kind of fault signature quantization parameter-fuzzy entropy for capableing of more acurrate description signal stationarity and complexity is proposed.Using the fuzzy entropy of each intrinsic mode function as fault signature, and as the input of MLP neural network, using mean square deviation as the training standard of MLP neural network, MLP neural network is trained using training sample, the identification and classification of planetary gear failure mode can be realized using the MLP neural network that training finishes.This method adaptive ability is strong, accuracy is high, can recognize that failure mode is perfect, improves the completeness of original algorithm, and accurate and effective extracts various faults characteristic information, realizes planetary gear fault identification and diagnosis.
Description
Technical field
The invention belongs to planetary gear fault diagnosis technology fields, relate to a kind of complete totality based on adaptive noise
The planetary gear method for diagnosing faults of empirical mode decomposition and fuzzy entropy.
Background technique
Planetary Gear Transmission has the advantages that small in size, light-weight, transmission ratio is big, is widely used in for large complicated
The key component of equipment Transmission system.Often working environment is severe for planetary gear, and receiving load is excessive, so as to cause its failure warp
Often occur, seriously affects the reliability of mechanical equipment, even result in major accident.Due to labyrinth, installation error and operating condition
Condition etc. influences, and leads to the amplitude modulation and frequency modulation phenomenon that occur more, shows stronger non-linear, non-stationary property.Therefore,
How hot spot that planet gear distress be current research is diagnosed.Pass through research planetary gear discovery, fault diagnosis side described herein
Method is suitable for processing Non-stationary vibration signal wheel, and is examined by extracting planetary gear failure in conjunction with advanced classification method
Disconnected feature.
Traditional fault signature extracting method is time-domain analysis and frequency-domain analysis, and time domain and frequency are extracted in fault diagnosis
Some indexs in domain.But these traditional index only have statistical property and global sense, therefore they are not suitable for processing non-stationary
Non-stationary signal caused by the nonlinear system of signal.EMD is a kind of adaptive Time Frequency Analysis method, and decomposable process is base
In the data of itself, Non-stationary vibration signal can be decomposed into a series of IMF.But EMD decomposition have the shortcomings that 2 it is main i.e.
End effect and modal overlap, end effect can seriously affect the quality of IMF component, and modal overlap will make IMF lose itself
Physical meaning, or even generate chaff component, at present solves the problems, such as main method first is that be based on white Gaussian noise frequency it is equal
The EEMD of one characteristic is decomposed.But the remnants that the signal rebuild after limited time ensemble average still contains certain amplitude make an uproar
Sound, although it can reduce reconstruction error by increasing average, this is by increased calculation scale.Simultaneously as increasing
White Gaussian noise each time unlike, this will lead to the IMF decomposited and residue signal is different.Therefore, this is not
One complete decomposable process.How to obtain complete intrinsic mode function is current urgent problem.
It is to be applied to calculation of thermodynamics that entropy, which calculates first, and personnel promoted after study later, in digital processing field
In be applied, it can reflect the complexity and stability of signal, suitable for handle nonlinear system generation non-stationary letter
Number.Currently, researcher has been presented for 10 kinds of methods such as singular spectrum entropy, Power Spectral Entropy and Sample Entropy.Wherein Sample Entropy reflects
The complexity of signal, and be applied in the fault signature extraction of mechanical equipment.But in the calculating process of Sample Entropy, signal
Similitude is dichotomy, it may be assumed that similitude and diversity.But in the comparison procedure of actual signal similitude, similarity degree is to connect
Continuous, fuzzy, it is difficult accurately to distinguish fault signature by the setting of simple threshold value.Currently, entropy calculates and neural network phase
In conjunction with how extensive use in feature information extraction and intelligent classification is selected suitable characteristic parameter and established perfect
Nerve identification network, is the big hot spot studied now.
Summary of the invention
Goal of the invention: in order to overcome the deficiencies in the prior art, the present invention provides one kind
Technical solution: to achieve the above object, planetary gear method for diagnosing faults of the invention the following steps are included:
1) the vibrating sensor measurement planetary gear vibration signal being arranged in planetary gear box enclosure is utilized；
2) believed using the complete overall experience mode decomposition method vibration obtained to step 1) based on adaptive noise
Number decomposed, extract include fault characteristic information complete intrinsic mode function；
3) constitution step 2) in extracted each intrinsic mode function space vector matrix, and calculate each group space vector
The distance between, introducing ambiguity function describes similar between two groups of space vectors according to the distance between every two groups of space vectors
Property, further defined to obtain the index of fault signature according to Sample Entropy --- fuzzy entropy；
4) MLP neural network is used to carry out planetary gear state recognition: using the fuzzy entropy of each intrinsic mode function as MLP
The input of neural network determines the input layer of MLP neural network, the network structure of hidden layer and output layer, utilizes training sample
MLP neural network is trained, using the mean square deviation of MLP neural network output valve and the difference of standard value as MLP nerve net
The training standard of network, completes the weight parameter adjustment of hidden layer and output layer, and the final MLP neural network finished using training is real
The identification and classification of existing star gear distress type.
Further, vibration signal described in step 1) is divided into Gear Planet Transmission sun gear normal condition, broken conditions, lacks
Five seed type of dentation state, gear wear and tooth root crackle.
Further, the decomposition of the complete overall experience mode decomposition method based on adaptive noise in the step 2)
Process is as follows:
A) the white noise number M that selection is added, and determine the amplitude of institute's plus noise；
B) obtain m white noise of addition adds vibration signal of making an uproar；
C) empirical mode decomposition is carried out to the vibration signal that white noise is added, obtains 1 intrinsic mode function IMF1；
If d) m < M, m=m+1, step b) and c) is carried out again, until m=M；
E) it calculates and each intrinsic mode function IMF that M noise decomposes is added1Population mean, and obtain remaining letter
Number r1(t), calculation formula is
r1(t)=x (t)-IMF1(t)
Wherein x (t) is collected vibration signal, ωmIt (t) is noise signal that unit variance mean value is zero, ε0It is to make an uproar
The amplitude of sound；
F) second intrinsic mode function IMF is calculated2, calculating process is as follows:
G) other intrinsic mode functions IMF is calculated, defines k=2,3.....K ,+1 IMF of kth can be expressed from the next
H) step g) is repeated, until the extreme point of residual signal is no more than two, stopping is decomposed.
Further, the calculating process of the fuzzy entropy in the step 3) is as follows:
A) the Vector Groups A being made of intrinsic mode function is definedi
Ai=z (i), z (i+1) ..., and z (i+m-1) }-u0(i) i=1,2 ..., n-m+1
Wherein u0(i) it is the mean value of each vector, m indicates the length of vector, and representative function is as follows:
B) the distance between every two groups of vectors are calculated
C) similitude between every group of vector is described with exponent obfuscation function
Wherein n is the boundary gradient of exponential function, and r is the similar limit, is set by the standard deviation of signal；
Seek fuzzy entropy representative function Bm:
D) m=m+1 is enabled, step (1) is repeated and arrives (4), obtain Bm+1
E) by seeking adjacent function BmThe logarithm of ratio you can get it fuzzy entropy
Further, the MLP neural network training process in the step 4) is as follows:
A) connection weight of MLP neural network is initialized, and the fuzzy entropy that the step 3) is extracted is transmitted to input layer；
Then the input feature vector of input layer is transmitted to hidden layer, calculates each hidden neuron s of hidden layerh, calculation formula is as follows:
Wherein: xijIt is input feature vector；
WjhIt is the connection weight inputted between neuron and hidden neuron；
θhIt is deviation；
F () is the activation primitive of hidden neuron；
B) calculated result of hidden layer neuron output layer is transmitted to carry out that output neuron yo is calculatedk, calculate
Formula is as follows:
Wherein: shIt is h-th of hidden layer output；
WhkIt is the connection weight between hidden layer neuron and output layer neuron；
ηkIt is deviation；
G () is the activation primitive of hidden layer neuron；
C) the corresponding input pattern t of each output neuronkThere is a target pattern, wherein the error of output neuron
Information is
δk=tk-yok
The control information of hidden layer is
The weight of hidden neuron more new formula is expressed as follows:
Wjh(t+1)=Wjh(t)+αδhxij+μ[Wjh(t)-Wjh(t-1)]
The weight update of output neuron is expressed as follows:
Whk(t+1)=Whk(t)+αδksh+μ[Whk(t)-Whk(t-1)]
α is learning rate in formula, and μ is factor of momentum；By continuously adjusting amendment, reaches final state, complete this
The training of a MLP neural network.
The utility model has the advantages that the method proposed of the invention can effectively extract the fault characteristic information of planetary gear generation, and
Planetary gear fault diagnosis can be realized according to the fault characteristic information of extraction, be a kind of effective planetary gear fault characteristic information
Extracting method, this method adaptive ability is strong, accuracy is high, can recognize that failure mode is perfect, is suitable for Planetary Gear Transmission mistake
It is influenced in journey by the interference and outside noise of the generations such as foozle, installation error, multiple tooth transmission, improves the complete of original algorithm
Property, accurate and effective extracts various faults characteristic information, realizes planetary gear fault identification and diagnosis.
Detailed description of the invention
Fig. 1 is a kind of planet of complete overall experience mode decomposition and fuzzy entropy based on adaptive noise of the present invention
The flow chart of gear failure diagnosing method；
Fig. 2 is the vibration of normal condition, broken conditions, few five kinds of dentation state, gear wear and tooth root crackle malfunctions
Time domain plethysmographic signal figure；
Fig. 3 is the 12 IMF and 1 remaining letter that planetary gear sun gear broken teeth fault-signal is decomposed by EEMD
Number；
Fig. 4 is 12 IMF and 1 remnants that planetary gear sun gear broken teeth fault-signal passes through that CEEMDAN is decomposed
Signal；
Fig. 5 is the reconstructed error figure after IMF component is reconstructed after EEMD and CEEMDAN are decomposed.
Fig. 6 is the Sample Entropy and fuzzy entropy comparison diagram of each IMF component after CEEMDAN is decomposed.
Fig. 7 is the Sample Entropy of five kinds of gear distress states and the case figure of fuzzy entropy.
Fig. 8 is influence of the different implicit number of plies of neural network to mean square deviation.
Fig. 9 is influence of the different implicit number of plies of neural network to overall failure discrimination.
Specific embodiment
With reference to the accompanying drawings and detailed description, the present invention is described in further detail.
As shown in Figure 1, a kind of complete overall experience mode decomposition (Complete based on adaptive noise that invention is described
Ensemble Empirical Mode DecompositionWithAdaptive Noise, hereinafter referred to as CEEMDAN) and
The planetary gear method for diagnosing faults of fuzzy entropy, comprising the following steps:
1) vibration signal of the vibration signals collecting using vibrating sensor measurement planetary gear tank shell, obtained vibration letter
It number include Gear Planet Transmission sun gear normal condition, broken conditions, few five seed type of dentation state, gear wear and tooth root crackle；
Planetary gear malfunction test is on the DDS mechanical breakdown integrated simulation experiment bench of Spectra Quest company, the U.S.
It carries out.This experiment measures planetary gear sun gear normal condition, broken conditions, few dentation state, gear wear and tooth root crackle altogether
Five kinds of states, by analyzing acquired vibration signal, the complete overall warp based on adaptive noise of inspection institute's foundation
Test the planetary gear method for diagnosing faults of mode decomposition and fuzzy entropy.
The emulation platform is by motor, planetary transmission, fixing axle gear-box, load system, acceleration transducer and data
Acquisition system composition.Planet transmission has 2 grades of planetary gear structures, the vibrating sensor installed on epicyclic gearbox
For collecting the vibration signal generated by planetary gear.During the experiment, the revolving speed of motor is set as 40Hz, and load is set as
13.5Nm.Since the sun gear failure of secondary planetary gear is it is easy to appear in Planetary Gear Transmission, so grinding at this
Study carefully it is middle select its as sample.Sample frequency is set as 12800Hz, is responsible for acquiring the Gearbox vibration signal of this five seed type.Base
In the planetary gear method for diagnosing faults of CEEMDAN and fuzzy entropy, to the carry out measuring and calculation of vibration signal.
2) planetary gear original vibration signal decompose first with CEEMDAN and be extracted comprising fault characteristic information
Complete intrinsic mode function (Intrinsic Mode Functions, hereinafter referred to as IMF), by the original vibration of planetary gear
Dynamic signal successively adds adaptive white noise and decomposes, and overcomes overall experience mode decomposition (Ensemble Empirical
Mode Decomposition, hereinafter referred to as EEMD) modal overlap and completeness the problems such as lacking, it is complete to obtain high quality
IMF component.CEEMDAN decomposable process is as follows:
A) the white noise number M that selection is added is set as 100, and noise amplitude is 0.2 times of original signal standard deviation, is added
Enter m white noise adds vibration signal of making an uproar to be expressed as follows
xm(t)=x (t)+nm(t)
B) EMD decomposition is carried out to the vibration signal that white noise is added, obtains I IMF1；Its EMD decomposable process is as follows:
1. time series signal xm(t), upper and lower envelope is respectively u (t) and v (t), the average out to w of upper and lower envelope
(t), x is usedm(t) w (t) is subtracted, remaining part is h1(t)
h1(t)=xm(t)-w(t)
2. using h1(t) x is replacedm(t), with h1(t) corresponding envelope up and down is respectively u1(t) and v1(t), it repeats to be moved through
Journey, i.e.,
w1(t)={ u1(t)+v1(t)}/2
h2(t)=h1(t)-w1(t)
……
wk-1(t)={ uk-1(t)-vk-1(t)}/2
hk(t)=hk-1(t)-wk-1(t)
Until resulting hk(t) meet two conditions of IMF, thus decompose and obtain first intrinsic mode functions c1(t) and
The remainder r of signal1(t)。
r1(t)=x (t)-c1(t)
r1(t) still include the frequency information of original signal in, repeat EMD decomposable process as new signal, until
When gained signal is lower than previously given value, decomposition is finished.Original signal is represented by the sum of all IMF and surplus.
If c) m < M, m=m+1, step 2) and 3) r are carried out again1(t), until m=M；
D) it calculates M times and each IMF that noise decomposes is added1Population mean, and obtain residual signal r1(t)
r1(t)=x (t)-IMF1(t)
E) second intrinsic mode function IMF is calculated2, calculating process is as follows
F) for other IMF, k=1 is defined, 2,3.....K ,+1 IMF of kth can be expressed from the next
G) step f) is repeated, until the extreme point of residual signal is no more than two, stopping is decomposed, and complete sheet is finally obtained
Levy mode function.
By taking the analysis of planetary gear broken teeth as an example, from figs. 3 and 4 it can be seen that vibration signal is broken down into 12 components and 1
A residue signal, in order to indicate convenient, residue signal is expressed as IMF13。IMF1-IMF13It is that frequency arranges from high to low.Due to vibration
Dynamic signal be it is non-smoothly, EEMD decomposition result still has modal overlap phenomenon, such as IMF6, IMF8And IMF9.Meanwhile
There is the ingredient of some falsenesses, such as IMF7And IMF9.And CEEMDAN decomposes to obtain result, it can be seen that IMF9The quality of component
It greatly improves, and further suppresses modal overlap phenomenon.Meanwhile reducing false ingredient.It is complete in order to verify CEEMDAN
Property, EEMD and CEEMDAN reconstructed error is as shown in figure 5, can prove that CEEMDAN method has more preferably by observing reconstructed error
Completeness, decomposition result and original signal error are smaller, the IMF component comprising more accurate characteristic information.
3) fault signature based on fuzzy entropy theory extracts, and defines the Vector Groups being made of intrinsic mode function IMF first
Ai
Ai=z (i), z (i+1) ..., and z (i+m-1) }-u0(i) i=1,2 ..., n-m+1
Wherein u0It (i) is the mean value of each vector, m indicates the length of vector, and representative function is as follows
Calculate the distance between every group of vector
It selects ambiguity function and describes similitude between every group of vector, type function selects exponential function
Wherein r is the similar limit, is chosen generally according to the standard deviation of signal, usually selection 0.1-0.2sd, we
Here selecting r=0.15sd, n is the convenient gradient of standard deviation, in order to capture details as much as possible, it is proposed that take one
A lesser integer value, such as 2 or 3, select n=2 here.The value of m is too small to will lead to that compare the information content that vector includes few, and mistake
Big value will lead to the number very little for comparing vector.After some experiments, determine that the parameter value of m is 7000.
Seek representative function Bm,
M=m+1 is enabled, step (1) is repeated and arrives (4), obtain Bm+1
Fuzzy entropy is by seeking adjacent function BmYou can get it for the logarithm of ratio:
In order to prove that fuzzy entropy has better performance compared with Sample Entropy, carry out each IMF's of CEEMDAN decomposition
Sample Entropy and fuzzy entropy are as shown in Figure 6.Although the overlapping phenomenon of the fuzzy entropy of each IMF component also occurs in Fig. 6, and it is each
The Sample Entropy of IMF component is compared, and overlapping phenomenon is smaller.The distribution relative distribution of the Sample Entropy of the identical gear of Fig. 7, it means that use
There is biggish undulating value when the different gears of Sample Entropy processing.It follows that the similitude for describing signal with Sample Entropy is not
It is accurate and incomplete, and such case will increase the difficulty of gear condition identification.And signal is described with ambiguity function
Similitude, the distribution Relatively centralized and value of identical gear fluctuate smaller.Therefore, feature calculation process can be solved using fuzzy entropy
The problems such as similitude of middle signal describes is better than Sample Entropy in terms of feature extraction, can more easily discriminate the tooth of five seed types
Wheel, and there is better robustness.
4) multi-layered perception neural networks are carried out using the fuzzy entropy of each IMF as the input parameter of multi-layered perception neural networks
(Multilayer Perceptron Neural Network, hereinafter referred to as MLP neural network) training.
A) connection weight of MLP neural network is initialized, and the feature of extraction is transmitted to input layer.Then it will be inputted
Feature be transmitted to hidden layer, calculate the output of each neuron of hidden layer:
Wherein: xijIt is input feature vector；
WjhIt is the connection weight inputted between neuron and hidden neuron；
θhIt is deviation；
F () is the activation primitive of hidden neuron；
B) calculated result of hidden layer neuron output layer is transmitted to carry out that output neuron is calculated:
Wherein: shIt is h-th of hidden layer output；
WhkIt is the connection weight between hidden layer neuron and output layer neuron；
ηkIt is deviation；
G () is the activation primitive of hidden layer neuron；
C) the corresponding input pattern t of each output neuronkThere is a target pattern, wherein the error of output neuron
Information are as follows:
δk=tk-yok
The control information of hidden layer are as follows:
The weight update of implicit neuron is expressed as follows:
Wjh(t+1)=Wjh(t)+αδhxij+μ[Wjh(t)-Wjh(t-1)]
The weight update of output neuron is expressed as follows:
Whk(t+1)=Whk(t)+αδksh+μ[Whk(t)-Whk(t-1)]
By constantly adjusting, corrects, reach final state, complete the training of this neural network.
In experiment, we initially set up training sample set, and each planetary gear state has 30 samples, totally 150 samples.
Using the fuzzy entropy of each IMF of CEEMDAN decomposition as the input of MLP neural network, so the input layer of MLP neural network
There are 13 input neurons, sample is finally divided into five seed types, so MLP neural network output layer there are 5 output neurons.
In order to train MLP neural network, for different planetary gear states with different tag representations in output layer.Normal gear by
(10000) it indicates, broken conditions are indicated by (01000), and hypodontia state is indicated by (00100) roll flute state by (00010), tooth root
Fracture is indicated by (00001).The number of hidden layer neuron is usually determined by test error, sets the study of MLP neural network
Rate is 0.5, and train epochs are set as 80, and the training effect of MLP neural network is evaluated with mean square error, completes MLP neural network
Training.The variation of mean square error is as shown in Figure 8 in hidden layer difference neuron number training process.Establish different planetary gears
The test sample of status data, each planetary gear state have 40 samples, totally 200 samples.Finally test sample is passed through
Above-mentioned calculating process completes the extraction of fault signature, and the identity of trained MLP neural network is verified using test sample
Energy.Ratio between the correct identification number and training sample set total quantity of test sample collection is defined as the fault diagnosis system to know
Not rate shows the planetary gear totality discrimination of hidden layer difference neuron number in Fig. 9.
As can be seen from Figure 8, mean square deviation declines with the increase of training pace, and training pace becomes between 65 and 80
It is completed in the training process of balance, MLP neural network.Hidden layer has different neuron numbers, final mean square deviation minimum value
Also not identical, when hidden layer has 11 neurons, mean square deviation is minimum, value 0.0035.It can be seen in figure 9 that with
The increase of training pace, the whole discrimination of hidden layer with different neuron numbers all increasing.It can by data
With discovery, when MLP neural network hidden layer takes 11 neurons, when the maximum value of planetary gear entirety discrimination, reach
To 91%.So hidden layer takes the MLP neural network of 11 neurons best to planetary gear performance of fault diagnosis, it is directed to
The detailed discrimination of five kinds of planetary gear states is as shown in table 1.
As it can be seen from table 1 the whole discrimination for the MLP neural network that hidden layer contains 11 neurons reaches 91%.
The discrimination of abrased gear is minimum, reaches 87.5%.Normally the discrimination of gear, teeth-missing gear and tooth root fracture gear is respectively
92.5%, 90% and 90%.It is demonstrated experimentally that the complete overall experience mode decomposition based on adaptive noise that is proposed and fuzzy
The planetary gear method for diagnosing faults of entropy is effective.
Fault recognition rate of the 1 different faults situation of table under 11 hidden layer neurons
The above is only a preferred embodiment of the present invention, it should be pointed out that: for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (5)
1. a kind of planetary gear method for diagnosing faults, it is characterised in that: the following steps are included:
1) the vibrating sensor measurement planetary gear vibration signal being arranged in planetary gear box enclosure is utilized；
2) using the complete overall experience mode decomposition method based on adaptive noise to the obtained vibration signal of step 1) into
Row decomposes, and extracts the complete intrinsic mode function comprising fault characteristic information；
3) constitution step 2) in extracted each intrinsic mode function space vector matrix, and calculate between each group space vector
Distance, introduce ambiguity function and the similitude between two groups of space vectors described according to the distance between every two groups of space vectors,
It is further defined to obtain this fault signature index of fuzzy entropy according to Sample Entropy；
4) MLP neural network is used to carry out planetary gear state recognition: neural using the fuzzy entropy of each intrinsic mode function as MLP
The input of network determines the input layer of MLP neural network, the network structure of hidden layer and output layer, using training sample to MLP
Neural network is trained, using the mean square deviation of MLP neural network output valve and the difference of standard value as the instruction of MLP neural network
Practice standard, completes the weight parameter adjustment of hidden layer and output layer, the final MLP neural fusion planet finished using training
The identification and classification of gear distress type.
2. a kind of planetary gear method for diagnosing faults according to claim 1, it is characterised in that: vibration described in step 1)
Dynamic signal is divided into Gear Planet Transmission sun gear normal condition, broken conditions, few five type of dentation state, gear wear and tooth root crackle
Type.
3. a kind of planetary gear method for diagnosing faults according to claim 1, it is characterised in that: the base in the step 2)
It is as follows in the decomposable process of the complete overall experience mode decomposition method of adaptive noise:
A) the white noise number M that selection is added, and determine the amplitude of added white noise；
B) obtain m white noise of addition adds vibration signal of making an uproar；
C) empirical mode decomposition is carried out to the vibration signal that white noise is added, obtains 1 first intrinsic mode function IMF1；
If d) m < M, m=m+1, step b) and c) is carried out again, until m=M；
E) it calculates and each first intrinsic mode function IMF that M noise decomposes is added1Population mean, and obtain remaining letter
Number r1(t), calculation formula are as follows:
r1(t)=x (t)-IMF1(t)
F) second intrinsic mode function IMF is calculated2, calculating process is as follows:
G) it calculates other intrinsic mode functions IMF, defines k=1,2,3....K ,+1 IMF of kth can be expressed from the next:
H) step g) is repeated, until the extreme point of residual signal is no more than two, stopping is decomposed.
4. a kind of planetary gear method for diagnosing faults according to claim 1, it is characterised in that: the mould in the step 3)
The calculating process for pasting entropy is as follows:
A) the Vector Groups A being made of intrinsic mode function is definedi
Ai=z (i), z (i+1) ..., and z (i+m-1) } 0-u0(i), i=1,2 ..., n-m+1
Wherein u0(i) it is the mean value of each vector, m indicates the length of vector, and representative function is as follows:
B) the distance between every two groups of vectors are calculated
C) similitude between every group of vector is described with exponent obfuscation function
Wherein n is the boundary gradient of exponential function, and r is the similar limit, is set by the standard deviation of signal；Seek fuzzy entropy
Representative function Bm:
D) it enables m=m+1 repeat step a) to d), obtains Bm+1
E) by seeking adjacent function BmThe logarithm of ratio you can get it fuzzy entropy
FuzzyEn=ln (Bm/Bm+1)。
5. a kind of planetary gear method for diagnosing faults according to claim 1, it is characterised in that: in the step 4)
MLP neural network training process is as follows:
A) connection weight of MLP neural network is initialized, and the fuzzy entropy that the step 3) is extracted is transmitted to input layer；Then
The input feature vector of input layer is transmitted to hidden layer, calculates each hidden neuron s of hidden layerh, calculation formula is as follows:
Wherein: xijIt is input feature vector；
WjhIt is the connection weight inputted between neuron and hidden neuron；
θhIt is deviation；
F () is the activation primitive of hidden neuron；
B) calculated result of hidden layer neuron output layer is transmitted to carry out that output neuron yo is calculatedk, calculation formula is such as
Under:
Wherein: shIt is h-th of hidden layer output；
WhkIt is the connection weight between hidden layer neuron and output layer neuron；
ηkIt is deviation；
G () is the activation primitive of hidden layer neuron；
C) the corresponding input pattern t of each output neuronkThere is a target pattern, wherein the control information of output neuron
For
δk=tk-yok
The control information of hidden layer is
The weight of hidden neuron more new formula is expressed as follows:
Wjh(t+1)=Wjh(t)+αδhxij+μ[Wjh(t)-Wjh(t-1)]
The weight update of output neuron is expressed as follows:
Whk(t+1)=Whk(t)+αδksh+μ[Whk(t)-Whk(t-1)]
α is learning rate in formula, and μ is factor of momentum；By continuously adjusting modification, reaches final state, complete this MLP
The training of neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610537475.3A CN106197999B (en) | 2016-07-08 | 2016-07-08 | A kind of planetary gear method for diagnosing faults |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610537475.3A CN106197999B (en) | 2016-07-08 | 2016-07-08 | A kind of planetary gear method for diagnosing faults |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106197999A CN106197999A (en) | 2016-12-07 |
| CN106197999B true CN106197999B (en) | 2019-07-09 |
Family
ID=57473326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610537475.3A Active CN106197999B (en) | 2016-07-08 | 2016-07-08 | A kind of planetary gear method for diagnosing faults |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106197999B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108225762B (en) * | 2016-12-15 | 2019-09-20 | 唐智科技湖南发展有限公司 | A kind of single gear tooth crackle broken teeth fault identification diagnostic method |
| CN107491606A (en) * | 2017-08-17 | 2017-12-19 | 安徽工业大学 | Variable working condition epicyclic gearbox sun gear method for diagnosing faults based on more attribute convolutional neural networks |
| US11188065B2 (en) | 2017-09-23 | 2021-11-30 | Nanoprecise Sci Corp. | System and method for automated fault diagnosis and prognosis for rotating equipment |
| CN108171263B (en) * | 2017-12-26 | 2019-08-30 | 合肥工业大学 | Based on the Fault Diagnosis of Roller Bearings for improving variation mode decomposition and extreme learning machine |
| CN108375971A (en) * | 2018-03-18 | 2018-08-07 | 哈尔滨工程大学 | Integrated Electronic System health control module and health control method for moonlet |
| CN110398362B (en) * | 2018-04-19 | 2021-06-11 | 中国科学院沈阳自动化研究所 | Robot RV reducer fault diagnosis and positioning method |
| CN109443768A (en) * | 2018-11-14 | 2019-03-08 | 中国直升机设计研究所 | A kind of Helicopter Main Reducer planetary gear vibration signal separation method |
| CN109632291A (en) * | 2018-12-04 | 2019-04-16 | 河北科技大学 | A kind of Fault Diagnosis of Gear Case method based on polynary mode decomposition-transfer entropy |
| CN110044621A (en) * | 2019-03-25 | 2019-07-23 | 西安交通大学 | The epicyclic gearbox oscillation power of gear distress composes prediction technique |
| CN110555226A (en) * | 2019-04-03 | 2019-12-10 | 太原理工大学 | method for predicting residual life of lithium iron phosphate battery based on EMD and MLP |
| CN109948597B (en) * | 2019-04-26 | 2022-06-07 | 福州大学 | High-voltage circuit breaker mechanical fault diagnosis method |
| CN112855922A (en) * | 2021-02-11 | 2021-05-28 | 中国人民解放军陆军装甲兵学院 | Planetary gear crack depth evaluation method |
| CN116061006B (en) * | 2023-04-03 | 2023-07-21 | 成都飞机工业（集团）有限责任公司 | Cutter monitoring method, device, equipment and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104006961A (en) * | 2014-04-29 | 2014-08-27 | 北京工业大学 | Cycloid bevel gear fault diagnosis method based on empirical mode decomposition and cepstrum |
| CN104390781A (en) * | 2014-11-26 | 2015-03-04 | 中国矿业大学 | Gear fault diagnosis method based on LMD and BP neural network |
| CN104483127A (en) * | 2014-10-22 | 2015-04-01 | 徐州隆安光电科技有限公司 | Method for extracting weak fault characteristic information of planetary gear |
| CN104748961A (en) * | 2015-03-30 | 2015-07-01 | 中国矿业大学 | Gear fault diagnosis method based on SVD decomposition and noise reduction and correlation EEMD entropy features |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4018092B2 (en) * | 2004-08-04 | 2007-12-05 | 本田技研工業株式会社 | A device that automatically determines the cause of a given surface condition of an inspection object |
-
2016
- 2016-07-08 CN CN201610537475.3A patent/CN106197999B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104006961A (en) * | 2014-04-29 | 2014-08-27 | 北京工业大学 | Cycloid bevel gear fault diagnosis method based on empirical mode decomposition and cepstrum |
| CN104483127A (en) * | 2014-10-22 | 2015-04-01 | 徐州隆安光电科技有限公司 | Method for extracting weak fault characteristic information of planetary gear |
| CN104390781A (en) * | 2014-11-26 | 2015-03-04 | 中国矿业大学 | Gear fault diagnosis method based on LMD and BP neural network |
| CN104748961A (en) * | 2015-03-30 | 2015-07-01 | 中国矿业大学 | Gear fault diagnosis method based on SVD decomposition and noise reduction and correlation EEMD entropy features |
Non-Patent Citations (2)
| Title |
|---|
| BP神经网络数据预测模型的建立及应用;宫唤春等;《起重运输机械》;20091231;第1卷;第95-98页 |
| 基于EEMD、模糊熵和SVM的齿轮故障诊断方法;李林峰等;《机械传动》;20141231;第38卷(第2期);第147-151页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106197999A (en) | 2016-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106197999B (en) | A kind of planetary gear method for diagnosing faults | |
| CN109460618B (en) | Rolling bearing residual life online prediction method and system | |
| CN100485342C (en) | Integrated supporting vector machine mixed intelligent diagnosing method for mechanical fault | |
| CN106596116A (en) | Vibration fault diagnosis method of wind generating set | |
| CN105354587B (en) | A kind of method for diagnosing faults of wind-driven generator group wheel box | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN110543860B (en) | Mechanical fault diagnosis method and system based on TJM (machine learning model) transfer learning | |
| CN103728551B (en) | A kind of analog-circuit fault diagnosis method based on cascade integrated classifier | |
| CN112036547B (en) | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM | |
| CN105760839A (en) | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine | |
| CN106096562B (en) | Gearbox of wind turbine method for diagnosing faults based on vibration signal blind sources separation and sparse component analysis | |
| CN108197648A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models | |
| CN105629958B (en) | A kind of batch process method for diagnosing faults based on sub-period MPCA SVM | |
| CN110702418A (en) | Aircraft engine fault prediction method | |
| CN108827605A (en) | A kind of mechanical breakdown characteristic automatic extraction method based on improvement sparseness filtering | |
| CN108398268A (en) | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps | |
| CN104792522A (en) | Intelligent gear defect analysis method based on fractional wavelet transform and BP neutral network | |
| CN105628425A (en) | Rotation machinery early stage fault diagnosis method based on heredity annealing optimization multi-core support vector machine | |
| CN107909118A (en) | A kind of power distribution network operating mode recording sorting technique based on deep neural network | |
| CN104198184A (en) | Bearing fault diagnostic method based on second generation wavelet transform and BP neural network | |
| CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
| CN109000930A (en) | A kind of turbogenerator performance degradation assessment method based on stacking denoising self-encoding encoder | |
| CN106198000A (en) | A kind of rocker arm of coal mining machine gear failure diagnosing method | |
| CN103115789A (en) | Second generation small-wave support vector machine assessment method for damage and remaining life of metal structure | |
| Di et al. | Ensemble deep transfer learning driven by multisensor signals for the fault diagnosis of bevel-gear cross-operation conditions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20190603 Address after: 277000 Room 309, Building 1, Zaozhuang Keshun Digital Co., Ltd., East of Fuyuan Fourth Road, Zaozhuang High-tech Zone, Shandong Province Applicant after: Shandong Zhongheng Photoelectric Technology Co., Ltd. Address before: Room 107, Kechuang Building, Zhongbang Pioneer Park, Huaibei City, Anhui Province Applicant before: Anhui Deyan Intelligent Control Technology Co., Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |