CN108535648A - Method of Motor Fault Diagnosis and system - Google Patents
Method of Motor Fault Diagnosis and system Download PDFInfo
- Publication number
- CN108535648A CN108535648A CN201810268354.2A CN201810268354A CN108535648A CN 108535648 A CN108535648 A CN 108535648A CN 201810268354 A CN201810268354 A CN 201810268354A CN 108535648 A CN108535648 A CN 108535648A
- Authority
- CN
- China
- Prior art keywords
- regression models
- vibration signal
- softmax regression
- alternating current
- motor
- 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
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/34—Testing dynamo-electric machines
- G01R31/343—Testing dynamo-electric machines in operation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The present invention provides a kind of Method of Motor Fault Diagnosis and system, this method extracts the feature vector of the vibration signal by obtaining the vibration signal of alternating current generator；Using the feature vector of the vibration signal as the input of target Softmax regression models, the fault diagnosis of motor is carried out by the target Softmax regression models.To realize the purpose for carrying out exact classification diagnosis to alternating current generator, it is weak to compensate for conventional motors method for diagnosing faults independent learning ability；The disadvantage of robustness, generalization difference.
Description
Technical field
The present invention relates to motor diagnostic technical fields, and in particular, to Method of Motor Fault Diagnosis and system.
Background technology
With the progress of science and technology, motor driving is used as social production and the main energy of daily life to supply already
Answer part.With industrial large-scale, production equipment also constantly develops towards the direction of enlargement, automation.It is close several
Over 10 years, the accident caused by motor device failure happens occasionally, and causes serious economic loss.Therefore, it is necessary to motor
Fault diagnosis is carried out, with the generation prevented accident, and reference can be provided for the manufacture and maintenance of motor.
The method for diagnosing faults of motor is mainly the following at present：
Method 1：Method of Motor Fault Diagnosis based on expert system.Mainly according to the experience of brainstrust, failure is believed for it
Breath, which is concluded, becomes a kind of rule, establishes knowledge base.When faulty generation, is reached using the empirical analysis reasoning in knowledge base and examined
Disconnected purpose.But the diagnosis of method 1 is difficult to set up than more complete knowledge base.The system does not have autonomous learning simultaneously
Ability, and poor robustness.
Method 2：Method of Motor Fault Diagnosis based on rough set theory.Rough set theory is at data directly analysis
Reason, finds tacit knowledge from uncertainty.But the reasoning of method 2 is caused to training sample according to knowledge logic reasoning
This has very strong dependence, generalization ability weak.
Method 3：The Method of Motor Fault Diagnosis of Bayesian network based on fault tree.Bayesian network is a kind of oriented nothing
The network structure of ring figure.According to prior distribution, relationship between likelihood function and Posterior distrbutionp in a probabilistic manner to motor into
Row fault diagnosis.Method 3 carries out Construction of A Model when carrying out net structure using logic gate；This is just to analysis personnel's
It is required that it is very high, once analysis is inaccurate to become the problem of will appear error diagnosis or delay diagnosis.
Invention content
For the defects in the prior art, the object of the present invention is to provide a kind of Method of Motor Fault Diagnosis and systems.
In a first aspect, the present invention provides a kind of Method of Motor Fault Diagnosis, including：
The vibration signal of alternating current generator is obtained, and extracts the feature vector of the vibration signal；
Using the feature vector of the vibration signal as the input of target Softmax regression models, pass through the target
Softmax regression models carry out the fault diagnosis of motor.
Optionally, using the feature vector of the vibration signal as the input of target Softmax regression models, pass through institute
Before stating the fault diagnosis that target Softmax regression models carry out motor, further include：
The alternating current generator of 4 types is chosen, the alternating current generator of 4 type includes：Normal motor, rotor unbalance event
Barrier motor built-in misaligns faulty motor and bearing fault motor；
The vibration signal of the alternating current generator of 4 type is acquired by NI cRIO-9030 performance detection control systems；
The feature vector of the vibration signal of the alternating current generator of 4 type is extracted with reconstruct by WAVELET PACKET DECOMPOSITION；
Different labels is added respectively for the feature vector of the vibration signal of the alternating current generator of 4 type, and will be added
After feature vector after tagging carries out cross validation, it is divided into：Training set data, test set data, training set label and survey
Examination collection label；
Initial Softmax regression models are trained by the training set data and the training set label, are obtained
Trained Softmax regression models；
The trained Softmax regression models are surveyed by the test set data and the test set label
Examination, to obtain the target Softmax regression models by test.
Optionally, described to pass through WAVELET PACKET DECOMPOSITION and the vibration signal of the alternating current generator of reconstruct extraction 4 type
Feature vector, including：
Choose the initial number of plies of WAVELET PACKET DECOMPOSITION；
Noise reduction process is carried out to the vibration signal of the alternating current generator of 4 type by wavelet packet, obtains de-noising signal；
Draw the power spectrum chart and third-octave spectrogram of the de-noising signal；
The end layer of WAVELET PACKET DECOMPOSITION is determined according to the Energy distribution of the power spectrum of the de-noising signal and third-octave spectrum
Number, and carry out the signal after being reconstructed after wavelet package reconstruction processing；
Wavelet-packet energy spectrogram is drawn according to the signal after the reconstruct；Wherein, the wavelet-packet energy spectrogram is institute
State the feature vector of the vibration signal of the alternating current generator of 4 types.
Optionally, it is described by the training set data and the training set label to initial Softmax regression models into
Row training, obtains trained Softmax regression models, including：
The basic parameter of Softmax regression models is set, to build initial Softmax regression models；Wherein, described basic
Parameter includes：Input layer dimension, output layer dimension, hidden layer node number and the hidden layer number of plies, iterations, learning rate, just
Then change parameter, initialization weight, non-linearization excitation function；
Using training set label as accuracy rate evaluation index, returned the training set data as the initial Softmax
The input for returning model adjusts the weight of the initial Softmax regression models by repetitive exercise, when described initial
When the accuracy rate of the training set label of Softmax regression models output is more than predetermined threshold value, preserves the initial Softmax and return
The weight of model obtains trained Softmax regression models.
Optionally, described that the trained Softmax is returned by the test set data and the test set label
Model is returned to be tested, to obtain the target Softmax regression models by test, including：
Using test set label as evaluation index, by test set data to the trained Softmax regression models
It is tested；If the label of the maximum probability of the trained Softmax regression models output is consistent with test set label,
Then test passes through, otherwise, test crash；
By test by the trained Softmax regression models be used as target Softmax regression models.
Optionally, described to obtain the vibration signal of alternating current generator, and the feature vector of the vibration signal is extracted, including：
The vibration signal of the alternating current generator is acquired by NI cRIO-9030 performance detection control systems；
The feature vector of the vibration signal of the alternating current generator is extracted with reconstruct by WAVELET PACKET DECOMPOSITION.
Optionally, the feature vector of the vibration signal for extracting the alternating current generator with reconstruct by WAVELET PACKET DECOMPOSITION,
Including：
Choose the initial number of plies of WAVELET PACKET DECOMPOSITION；
Noise reduction process is carried out to the vibration signal of the alternating current generator by wavelet packet, obtains de-noising signal；
Draw the power spectrum chart and third-octave spectrogram of the de-noising signal；
The end layer of WAVELET PACKET DECOMPOSITION is determined according to the Energy distribution of the power spectrum of the de-noising signal and third-octave spectrum
Number, and carry out the signal after being reconstructed after wavelet package reconstruction processing；
Wavelet-packet energy spectrogram is drawn according to the signal after the reconstruct；Wherein, the wavelet-packet energy spectrogram is institute
State the feature vector of the vibration signal of alternating current generator.
Second aspect, the present invention provide a kind of Diagnostic system of motor fault, including：Processor and memory, the storage
Have program stored therein instruction in device, and the processor is for transferring described program instruction to execute described in any one of first aspect
Method of Motor Fault Diagnosis.
Compared with prior art according to the Method of Motor Fault Diagnosis of Softmax regression models provided by the invention, this hair
It is bright that there is following advantageous effect：
Method of Motor Fault Diagnosis provided by the invention and system obtain the vibration signal for waiting for diagnosing motor, pass through small echo
Packet decomposition and reconstruction, finds fault characteristic frequency, determines fault type；Extract the feature vector of the vibration signal；It will be described
Feature vector carries out Diagnosing Faults of Electrical as input, by Softmax regression models.It realizes through frequency-domain analysis and algorithm
Analysis is carried out at the same time alternating current generator the purpose of diagnostic classification, solves conventional motors method for diagnosing faults independent learning ability
It is weak, robustness, the disadvantage of generalization ability difference.
Description of the drawings
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon：
Fig. 1 is the principle configuration diagram for the Method of Motor Fault Diagnosis that the embodiment of the present invention one provides；
Fig. 2 is the flow chart of Method of Motor Fault Diagnosis provided by Embodiment 2 of the present invention；
Fig. 3 is the flow chart for the Method of Motor Fault Diagnosis that the embodiment of the present invention three provides；
Fig. 4 is the flow chart for the Method of Motor Fault Diagnosis that the embodiment of the present invention four provides.
Specific implementation mode
With reference to specific embodiment, the present invention is described in detail.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, several changes and improvements can also be made.These belong to the present invention
Protection domain.
Fig. 1 is the principle Organization Chart for the Method of Motor Fault Diagnosis that the embodiment of the present invention one provides, flow packet shown in FIG. 1
It includes：Choosing 4 type motor types first is：Normal motor, rotor unbalance faulty motor, it is built-in misalign faulty motor and
Bearing fault motor.Basic parameter is tested by NI cRIO-9030 performance detection set-up of control system；Acquisition module is set
Sample frequency is 10.24ks/s；Motor rotation frequency is 15Hz；The acquisition time of vibration signal is 7 minutes.Due to sample frequency
Intrinsic frequency is rotated much larger than motor, 4 class motors of setting choose 1 number of seconds strong point as analysis respectively.Mould is acquired by NI-9234
Vibration signal at block acquisition experiment electric motor end cap, at this end cap sensor vibration signal be divided into end cap direction tangential, along end cap
Radial direction.The direction that the present embodiment acquires vibration signal is radial direction, outputting radial vibration signal.
Fig. 2 is the flow chart of Method of Motor Fault Diagnosis provided by Embodiment 2 of the present invention, and method shown in Fig. 2 can wrap
It includes：
S101, the vibration signal for obtaining alternating current generator, and extract the feature vector of the vibration signal.
In the present embodiment, the vibration of the alternating current generator can be acquired by NI cRIO-9030 performance detection control systems
Signal；Then the feature vector of the vibration signal of the alternating current generator is extracted with reconstruct by WAVELET PACKET DECOMPOSITION.Due to extraction
Include noise in the vibration signal of the alternating current generator, therefore vibration signal does wavelet packet denoising.First, wavelet packet is chosen
The initial number of plies decomposed；Noise reduction process is carried out to the vibration signal of the alternating current generator by wavelet packet, obtains de-noising signal；It paints
Make the power spectrum chart and third-octave spectrogram of the de-noising signal；It is composed according to the power spectrum of the de-noising signal and third-octave
Energy distribution determine the final number of plies of WAVELET PACKET DECOMPOSITION, and carry out the signal after being reconstructed after wavelet package reconstruction processing；Root
Wavelet-packet energy spectrogram is drawn according to the signal after the reconstruct；Wherein, the wavelet-packet energy spectrogram is the alternating current generator
Vibration signal feature vector.
Since wavelet packet is during noise reduction, analog signal finer decomposition is all subjected in low-and high-frequency, has been made up
Multiresolution analysis is not finely divided in high frequency section, causes the frequency that some are useful in high frequency section to be lost, to making letter
The deficiency of larger distortion number occurs after reconstitution.In order to make signal in canonical part smooth out noise, the sharp change part of signal is thin
Section is preferably retained, while Mutational part and noise can be come by efficiently differentiating, and final the present embodiment determines choosing
Use the obtained adaptive threshold of unbiased possibility predication principle as noise-removed threshold value.
S102, using the feature vector of the vibration signal as the input of target Softmax regression models, pass through the mesh
Mark the fault diagnosis that Softmax regression models carry out motor.
In the present embodiment, target Softmax regression models are instructed by the initial Softmax regression models to structure
It is obtained after white silk and test, has the model of fault diagnosis classification.
Fig. 3 is the flow chart for the Method of Motor Fault Diagnosis that the embodiment of the present invention three provides, and method shown in Fig. 3 is in Fig. 2
On the basis of shown method, using the feature vector of the vibration signal as the input of target Softmax regression models, pass through
Before the target Softmax regression models carry out the fault diagnosis of motor, it can also include the following steps：
S201, the alternating current generator for choosing 4 types.
In the present embodiment, the alternating current generator of 4 type includes：It is normal motor, rotor unbalance faulty motor, built-in
Misalign faulty motor and bearing fault motor；
S202, acquired by NI cRIO-9030 performance detection control systems 4 type alternating current generator vibration
Signal.
S203, extracted with reconstruct by WAVELET PACKET DECOMPOSITION 4 type alternating current generator vibration signal feature to
Amount.
In the present embodiment, the initial number of plies of WAVELET PACKET DECOMPOSITION can be chosen first；By wavelet packet to 4 type
The vibration signal of alternating current generator carries out noise reduction process, obtains de-noising signal；Draw the de-noising signal power spectrum chart and 1/3 times
Sound interval spectrogram；The final of WAVELET PACKET DECOMPOSITION is determined according to the power spectrum of the de-noising signal and the Energy distribution that third-octave is composed
The number of plies, and carry out the signal after being reconstructed after wavelet package reconstruction processing；Wavelet packet energy is drawn according to the signal after the reconstruct
Measure spectrogram；Wherein, the wavelet-packet energy spectrogram is the feature vector of the vibration signal of the alternating current generator of 4 type.
Specifically, the WAVELET PACKET DECOMPOSITION number of plies is primarily determined to above-mentioned 4 type motor, carries out denoising and obtains noise reduction letter
Number；The time domain waveform of de-noising signal and the time domain waveform of the vibration signal of non-noise reduction process are drawn, the vibration signal is observed
Noise reduction degree.Draw power spectrum chart of the de-noising signal in frequency domain and third-octave spectrogram；According to the power spectrum of de-noising signal
Figure and the Energy distribution of third-octave spectrogram are compared by the power corresponding to centre frequency between 4 type alternating current generators
Difference redefines the WAVELET PACKET DECOMPOSITION number of plies, carries out wavelet package reconstruction, and Wavelet Packet Energy Spectrum is drawn to the signal after reconstruct；Root
Spectrogram is drawn according to the maximum corresponding reconstruct node of energy in reconstruction signal Wavelet Packet Energy Spectrum, according to the fault signature of spectrogram
Frequency determines electrical fault type；Wherein Wavelet Packet Energy Spectrum is the feature vector of 4 type motors.
Specifically, the frequency content that vibration signal is included when breaking down by motor can change, and compare 4 classes
Power of motor is composed redefines the WAVELET PACKET DECOMPOSITION number of plies with the energy collection Mid Frequency of third-octave spectrogram.It is adopted according to Nyquist
Sample theorem, analysis upper frequency limit are the 1/2 of sample frequency, are determined according to power spectrum and are all generally concentrated at per class motor power
0-160Hz frequency ranges.So determining that the WAVELET PACKET DECOMPOSITION number of plies should be able to distinguish the difference of 0-160Hz band energies, that is, determine small
Wave packet Decomposition order n meets：It is thus determined that the WAVELET PACKET DECOMPOSITION number of plies is 5.
When the WAVELET PACKET DECOMPOSITION number of plies is 5 layers, the reconstruction coefficients of [5,0]-[5,31] are obtained, [5,0] indicate 0 knot of reconstruct
Point, [5,31] indicate 31 nodes of reconstruct；Signal after reconstruct is：
S=S5,0+S5,i+…+S5,31
In formula：S indicates the signal after reconstruct, S5,0Indicate the reconstruction signal of node [5,0], S5,iIndicate node [5, i]
Reconstruction signal, wherein i=0,1,2 ..., 31；
Obtain the corresponding energy of signal after the reconstruct；If S5,iCorresponding energy is E5,i；Then：
S in formula5,i(t) reconstruction signal of the node [5, i] about time t, x are indicatedi,kIndicate reconstruction signal S5,iDiscrete point
Amplitude；I=0,1,2 ..., 31；K=1,2 ..., n；N counts for signal sampling；Feature vector is expressed as at this time：[E5,0,
E5,1…E5,31]。
The characteristic energy of 4 type alternating current generators is normalized, Wavelet Packet Energy Spectrum scalogram, the small echo are obtained
Packet energy spectrum scalogram is the feature vector of different type motor oscillating signal.Wherein, the cross of Wavelet Packet Energy Spectrum scalogram
Coordinate is reconstruct node；Ordinate value represents each included energy proportion shared in gross energy； SkIndicate that k-th of reconstruct node energy proportion shared in gross energy, E indicate gross energy, E5,iTable
Show the energy contained by i-th of reconstruct node.Feature vector T after extraction normalization at this time is expressed as：
S204, different labels is added respectively for the feature vector of the vibration signal of the alternating current generator of 4 type, and
After feature vector after addition label is carried out cross validation, it is divided into：Training set data, test set data, training set label
With test set label.
S205, initial Softmax regression models are trained by the training set data and the training set label,
Obtain trained Softmax regression models.
In the present embodiment, the basic parameter of Softmax regression models can be set first, returned with building initial Softmax
Return model；Wherein, the basic parameter includes：Input layer dimension, output layer dimension, hidden layer node number with it is implicit layer by layer
Number, iterations, learning rate, regularization parameter, initialization weight, non-linearization excitation function；Using training set label as standard
True rate evaluation index, using the training set data as the input of the initial Softmax regression models, by repetitive exercise come
The weight for adjusting the initial Softmax regression models, when the training set label that the initial Softmax regression models export
When accuracy rate is more than predetermined threshold value, the weight of the initial Softmax regression models is preserved, trained Softmax is obtained and returns
Return model.
S206, by the test set data and the test set label to the trained Softmax regression models
It is tested, to obtain the target Softmax regression models by test.
It, first can be using test set label as evaluation index, by test set data to the training in the present embodiment
Good Softmax regression models are tested；If the label of the maximum probability of the trained Softmax regression models output
When consistent with test set label, then test passes through, otherwise, test crash；The trained Softmax that test is passed through is returned
Return model as target Softmax regression models.
The present embodiment carries out fault diagnosis based on WAVELET PACKET DECOMPOSITION, reconstruct by Softmax regression models.It will be special
Sign vector is trained by neural network, is tested as input.It realizes through frequency-domain analysis and Algorithm Analysis to alternating current
Machine is carried out at the same time the purpose of diagnostic classification, and it is weak to solve conventional motors method for diagnosing faults independent learning ability, robustness, extensive
The disadvantage of energy force difference.
It should be noted that the quantity of the unlimited random sample sheet of the present embodiment, theoretically the quantity of sample is more, then training result
Also more accurate, but the increase of sample size will increase computation complexity.Specifically, above-mentioned 4 kinds can be obtained in the present embodiment
Type motor is per class 400 feature vectors of motor, totally 1600 feature vector, that is, sample points.Label is added to 4 type motors,
Respectively 1,2,3,4.It obtains the feature vector after normalization and carries out 5 folding cross validations, data and label are obtained to divide.
Specifically, training set data is standardized with test set data, so that the data of each sample is distributed in identical
In the range of.Sample after standardization is built into Softmax regression models by loss function to intersect entropy loss；It determines initial
Parameter.
Wherein cross entropy loss function meets：
Each iteration updates weight with gradient descent method.Wherein m is sample number；I is i-th of sample；K is classification；J is
The true tag of j classification；θ is weight vectors；γ is penalty coefficient；J (θ) is the intersection entropy loss about weight, y(i)It is
The prediction label of i sample,For the transposition of j-th of class weight, x(i)For i-th of sample to be predicted,For i-th of weight
Transposition,For belong to j-th of classification, i-th sample weight inner product.
It is trained by training set data in the present embodiment, to obtain the optimal weight of neural network.It is taken turns in iteration 50
In, reach 0.022 in the iteration loss of the 16th wheel, accuracy 98.6719%.It is tested by test set data, with
Accuracy is index, and accuracy 99.0625% meets the requirement of diagnostic classification.
More detailed explanation is done to the method in the present invention with reference to specific embodiment.
Fig. 4 is the flow chart for the Method of Motor Fault Diagnosis that the embodiment of the present invention four provides.Applied to multistage gear case event
Hinder simulating table, pastes sensor in electric motor end cap, while with radial direction, then the wiring of direction of twist installs motor
In fault simulation testing stand designated position, NI-9234 is total to be realized by NI-9234 acquisition modules to the acquisition of vibration signal
There are four connectivity port, respectively 0,1,2,3；The connection of four road synchronized sampling analog input channels can be provided；It simultaneously will acquisition
Signal be digitized.Two wiring of sensor are connected to 0,1 port of NI-9234 acquisition modules；Wherein distortion side
0 port is connected to wiring, radial direction wiring connects 1 port；The distortion of acquisition motor and the vibration of radial both direction are believed respectively
Number.
By NI cRIO-9030 performance detections control systems to motor rotation frequency, in the sampling time, sample rate is set
It is fixed, set motor rotation frequency as 15hz, sampling time 420s, sample rate 10.24ks/s.
Since what the present invention analyzed is the vibration signal of motor radial direction, therefore 4 type motor, 1 port is obtained respectively
Data carry out Analysis on Fault Diagnosis.
Specific implementation mode is as follows：
1) 4 type motor oscillating signals are acquired.
2) wavelet packet threshold denoising is carried out.The WAVELET PACKET DECOMPOSITION number of plies is tentatively chosen to four class motor datas, threshold value makes
With mode, threshold denoising mode.
3) parameter determined according to step 1), step 2) carries out wavelet-packet noise reduction.The vibration signal waveforms after noise reduction are drawn,
Compare the variation of waveform before and after noise reduction, observes noise reduction degree.
4) draw noise reduction after 4 type motors power spectrum chart, observe vibration signal Energy distribution.
5) 4 type motor third-octave spectrograms after drafting noise reduction, determine vibration signal energy collection Mid Frequency.
6) power spectrum and third-octave spectrum according to step 4), 5) determines spectrum energy collection Mid Frequency, redefines small echo
Packet Decomposition order.
7) determine that the WAVELET PACKET DECOMPOSITION number of plies is 5；Above-mentioned steps are re-started, wavelet packet tree is obtained.
8) wavelet package reconstruction is carried out to the WAVELET PACKET DECOMPOSITION number of plies that above-mentioned steps determine, obtains 5 layers of wavelet package reconstruction node
Coefficient.
9) the wavelet package reconstruction node coefficient obtained according to step 8) seeks each frequency band energy and is normalized to obtain
Feature vector.
10) the normalization characteristic vector obtained according to step 9), draws wavelet-packet energy spectrogram.
11) power spectrum chart is drawn according to the maximum corresponding reconstruct node of energy in Wavelet Packet Energy Spectrum.Find fault signature
Frequency compares with fault-signal theoretic frequency point, determines electrical fault type.
12) above step is repeated, each 400 feature vectors of 4 type motors are obtained.
13) feature vector after normalizing 4 type motor of step 12) is merged by row；Add respectively for 4 type motors
It tags and is classified as：1、2、3、4.
14) cross validation is carried out to the feature vector that step 13) merges, respectively obtains training set data；Training set label；
Test set number and test set label；Wherein test set data account for 20%.
15) to training set data, test set data are normalized.
16) Softmax regression models carry out diagnostic classification.Determine model initial parameter：Input layer dimension is 32；Output layer
Dimension is 4；The hidden layer number of plies chooses 1 layer；Hidden layer node number is 96；Excitation function is relu；Iterations are 50 times；It learns
Habit rate is 0.01；Regularization parameter is 0.01；Initialization weight meets standardized normal distribution.
17) it is inputted, is examined as Softmax regression models according to the training set data after step 15), step 16) normalization
For disconnected result compared with training set label, loss function is to intersect entropy loss；After each iteration of chain type Rule for derivation update
Weight and offset, w here1For input weight vector；b1For input offset vector；w2For hidden layer weight vectors；b2It is hidden
Offset vector containing layer.
18) step 17) is repeated to obtain the intersection entropy loss of each iteration and update weight；Iteration terminates to acquire optimal weights
Vector and offset；The as optimized parameter of model.
19) according to the optimal w of the acquisition of step 18)1、b1、w2、b2Parameter to the normalization test set data of step 15) into
Row test；Compared with test set label, accuracy 99.0625% meets the requirement of diagnostic classification.
In the present embodiment, fault characteristic frequency is found by WAVELET PACKET DECOMPOSITION and reconstruct, obtains electrical fault type；Then
The size that decision probability is exported by Softmax regression models carries out diagnostic classification, realizes through frequency-domain analysis and algorithm point
Analysis is carried out at the same time alternating current generator the purpose of diagnostic classification, with certain extensive while improving the accuracy of diagnostic classification
Ability.Compared with the method 3 of background technology, the method in the present embodiment all seems in diagnostic accuracy and order of accuarcy more may be used
It takes, while not being restricted by sample size, is applicable under big-sample data and Small Sample Database situation.
The embodiment of the present invention also provides a kind of Diagnostic system of motor fault, including：Processor and memory, the memory
In have program stored therein instruction, the processor is examined for transferring described program instruction with executing the electrical fault described in Fig. 2, Fig. 3
Disconnected method.
It should be noted that the step in the Method of Motor Fault Diagnosis provided by the invention, can utilize the electricity
Corresponding module, device, unit etc. are achieved in machine fault diagnosis system, and those skilled in the art are referred to the system
Technical solution realize that the step flow of the method, i.e., the embodiment in the described system can be regarded as realizing the excellent of the method
Example is selected, it will not be described here.
One skilled in the art will appreciate that in addition to realizing system provided by the invention in a manner of pure computer readable program code
It, completely can be by the way that method and step be carried out programming in logic come so that system provided by the invention and its each and its outside each device
A device is come in the form of logic gate, switch, application-specific integrated circuit, programmable logic controller (PLC) and embedded microcontroller etc.
Realize identical function.So system provided by the invention and its every device are considered a kind of hardware component, and to it
The device for realizing various functions for including inside can also be considered as the structure in hardware component；It can also will be for realizing various
The device of function is considered as either the software module of implementation method can be the structure in hardware component again.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or change within the scope of the claims, this not shadow
Ring the substantive content of the present invention.In the absence of conflict, the feature in embodiments herein and embodiment can arbitrary phase
Mutually combination.
Claims (8)
1. a kind of Method of Motor Fault Diagnosis, which is characterized in that including：
The vibration signal of alternating current generator is obtained, and extracts the feature vector of the vibration signal；
Using the feature vector of the vibration signal as the input of target Softmax regression models, pass through the target Softmax
Regression model carries out the fault diagnosis of motor.
2. Method of Motor Fault Diagnosis according to claim 1, which is characterized in that by the feature of the vibration signal to
The input as target Softmax regression models is measured, the fault diagnosis of motor is carried out by the target Softmax regression models
Before, further include：
The alternating current generator of 4 types is chosen, the alternating current generator of 4 type includes：Normal motor, rotor unbalance failure electricity
Machine built-in misaligns faulty motor and bearing fault motor；
The vibration signal of the alternating current generator of 4 type is acquired by NI cRIO-9030 performance detection control systems；
The feature vector of the vibration signal of the alternating current generator of 4 type is extracted with reconstruct by WAVELET PACKET DECOMPOSITION；
Different labels is added respectively for the feature vector of the vibration signal of the alternating current generator of 4 type, and addition is marked
After feature vector after label carries out cross validation, it is divided into：Training set data, test set data, training set label and test set
Label；
Initial Softmax regression models are trained by the training set data and the training set label, are trained
Good Softmax regression models；
The trained Softmax regression models are tested by the test set data and the test set label,
To obtain the target Softmax regression models by test.
3. Method of Motor Fault Diagnosis according to claim 2, which is characterized in that described to pass through WAVELET PACKET DECOMPOSITION and reconstruct
The feature vector of the vibration signal of the alternating current generator of 4 type is extracted, including：
Choose the initial number of plies of WAVELET PACKET DECOMPOSITION；
Noise reduction process is carried out to the vibration signal of the alternating current generator of 4 type by wavelet packet, obtains de-noising signal；
Draw the power spectrum chart and third-octave spectrogram of the de-noising signal；
The final number of plies of WAVELET PACKET DECOMPOSITION is determined according to the Energy distribution of the power spectrum of the de-noising signal and third-octave spectrum,
And carry out the signal after being reconstructed after wavelet package reconstruction processing；
Wavelet-packet energy spectrogram is drawn according to the signal after the reconstruct；Wherein, the wavelet-packet energy spectrogram is described 4 kinds
The feature vector of the vibration signal of the alternating current generator of type.
4. Method of Motor Fault Diagnosis according to claim 2, which is characterized in that it is described by the training set data and
The training set label is trained initial Softmax regression models, obtains trained Softmax regression models, including：
The basic parameter of Softmax regression models is set, to build initial Softmax regression models；Wherein, the basic parameter
Including：Input layer dimension, output layer dimension, hidden layer node number and the hidden layer number of plies, iterations, learning rate, regularization
Parameter, initialization weight, non-linearization excitation function；
Using training set label as accuracy rate evaluation index, mould is returned using the training set data as the initial Softmax
The input of type adjusts the weight of the initial Softmax regression models by repetitive exercise, when the initial Softmax is returned
When the accuracy rate of the training set label of model output being returned to be more than predetermined threshold value, the power of the initial Softmax regression models is preserved
Weight, obtains trained Softmax regression models.
5. Method of Motor Fault Diagnosis according to claim 2, which is characterized in that it is described by the test set data and
The test set label tests the trained Softmax regression models, to obtain the target by test
Softmax regression models, including：
Using test set label as evaluation index, the trained Softmax regression models are carried out by test set data
Test；If the label of the maximum probability of the trained Softmax regression models output is consistent with test set label, survey
It pinged, otherwise, test crash；
By test by the trained Softmax regression models be used as target Softmax regression models.
6. Method of Motor Fault Diagnosis according to any one of claims 1-5, which is characterized in that the acquisition alternating current
The vibration signal of machine, and the feature vector of the vibration signal is extracted, including：
The vibration signal of the alternating current generator is acquired by NI cRIO-9030 performance detection control systems；
The feature vector of the vibration signal of the alternating current generator is extracted with reconstruct by WAVELET PACKET DECOMPOSITION.
7. Method of Motor Fault Diagnosis according to claim 6, which is characterized in that described to pass through WAVELET PACKET DECOMPOSITION and reconstruct
The feature vector of the vibration signal of the alternating current generator is extracted, including：
Choose the initial number of plies of WAVELET PACKET DECOMPOSITION；
Noise reduction process is carried out to the vibration signal of the alternating current generator by wavelet packet, obtains de-noising signal；
Draw the power spectrum chart and third-octave spectrogram of the de-noising signal；
The final number of plies of WAVELET PACKET DECOMPOSITION is determined according to the Energy distribution of the power spectrum of the de-noising signal and third-octave spectrum,
And carry out the signal after being reconstructed after wavelet package reconstruction processing；
Wavelet-packet energy spectrogram is drawn according to the signal after the reconstruct；Wherein, the wavelet-packet energy spectrogram is the friendship
The feature vector of the vibration signal of galvanic electricity machine.
8. a kind of Diagnostic system of motor fault, which is characterized in that including：Processor and memory are stored in the memory
Program instruction, the processor are used to transfer described program instruction with the motor event described in any one of perform claim requirement 1-7
Hinder diagnostic method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810268354.2A CN108535648A (en) | 2018-03-29 | 2018-03-29 | Method of Motor Fault Diagnosis and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810268354.2A CN108535648A (en) | 2018-03-29 | 2018-03-29 | Method of Motor Fault Diagnosis and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108535648A true CN108535648A (en) | 2018-09-14 |
Family
ID=63481668
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810268354.2A Pending CN108535648A (en) | 2018-03-29 | 2018-03-29 | Method of Motor Fault Diagnosis and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108535648A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109726736A (en) * | 2018-11-27 | 2019-05-07 | 西人马(西安)测控科技有限公司 | The method and system of training elevator internal contracting brake fault diagnosis model, band-type brake fault diagnosis |
| CN109726505A (en) * | 2019-01-14 | 2019-05-07 | 哈尔滨工程大学 | A kind of forging machine tool main drive gear fault diagnosis system based on intelligent trouble tree |
| CN109917287A (en) * | 2019-03-20 | 2019-06-21 | 华南理工大学 | Decelerating motor product examine method based on empirical mode decomposition and octave spectrum analysis |
| CN110081966A (en) * | 2019-05-27 | 2019-08-02 | 苏州工业职业技术学院 | The fault diagnosis system and method for fan vibration sensor |
| CN110956268A (en) * | 2019-10-16 | 2020-04-03 | 中国石化青岛液化天然气有限责任公司 | Compressor fault diagnosis method based on big data |
| CN110988677A (en) * | 2019-11-25 | 2020-04-10 | 北京昊鹏智能技术有限公司 | Fault detection method and device for direct current motor and mechanical equipment driven by direct current motor |
| CN111272427A (en) * | 2020-02-14 | 2020-06-12 | 西安交通大学 | Bearing fault detection method based on weighted sparse regularization |
| CN112395959A (en) * | 2020-10-30 | 2021-02-23 | 天合云能源互联网技术(杭州)有限公司 | Power transformer fault prediction and diagnosis method and system based on audio features |
| CN112433176A (en) * | 2020-11-17 | 2021-03-02 | 上海大学 | Power electronic converter fault diagnosis system and method |
| CN113009334A (en) * | 2021-02-18 | 2021-06-22 | 华中科技大学 | Motor fault detection method and system based on wavelet packet energy analysis |
| CN113325314A (en) * | 2021-07-20 | 2021-08-31 | 安徽巨森电器有限公司 | Motor fault diagnosis method |
| CN115993511A (en) * | 2023-03-22 | 2023-04-21 | 武汉纺织大学 | Partial discharge type high-precision detection and identification device, method and equipment |
| CN116381491A (en) * | 2023-06-05 | 2023-07-04 | 南京苏尚电气科技有限公司 | New energy automobile motor fault positioning system and method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011073613A1 (en) * | 2009-12-15 | 2011-06-23 | University Of Portsmouth | A method, apparatus and computer program for diagnosing a mode of operation of a machine |
| CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
-
2018
- 2018-03-29 CN CN201810268354.2A patent/CN108535648A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011073613A1 (en) * | 2009-12-15 | 2011-06-23 | University Of Portsmouth | A method, apparatus and computer program for diagnosing a mode of operation of a machine |
| CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
Non-Patent Citations (8)
| Title |
|---|
| JUNWEI WANG等: "Feature ensemble learning using stacked denoising autoencoders for induction motor fault diagnosis", 《2017 PROGNOSTICS AND SYSTEM HEALTH MANAGEMENT CONFERENCE (PHM-HARBIN)》 * |
| YAGUO LEI等: "An Intelligent Fault Diagnosis Method Using Unsupervised Feature Learning Towards Mechanical Big Data", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》 * |
| 刘亚军: "基于小波分析和神经网络的电机故障诊断研究", 《中国优秀硕士学位论文全文数据库工程科技II辑》 * |
| 刘遥行等: "机器学习中常常提到的正则化到底是什么意思", 《知乎》 * |
| 王丽华等: "采用深度学习的异步电机故障诊断方法", 《西安交通大学学报》 * |
| 薛超等: "基于卷积神经网络的级联人脸检测", 《中国安防》 * |
| 郁可人等: "基于神经网络语言模型的分布式词向量研究进展", 《华东师范大学学报（自然科学版）》 * |
| 郁松等: "基于卷积神经网络的自然背景字符识别", 《计算机应用与软件》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109726736A (en) * | 2018-11-27 | 2019-05-07 | 西人马(西安)测控科技有限公司 | The method and system of training elevator internal contracting brake fault diagnosis model, band-type brake fault diagnosis |
| CN109726505A (en) * | 2019-01-14 | 2019-05-07 | 哈尔滨工程大学 | A kind of forging machine tool main drive gear fault diagnosis system based on intelligent trouble tree |
| CN109917287A (en) * | 2019-03-20 | 2019-06-21 | 华南理工大学 | Decelerating motor product examine method based on empirical mode decomposition and octave spectrum analysis |
| CN109917287B (en) * | 2019-03-20 | 2021-06-08 | 华南理工大学 | Speed reduction motor quality inspection method based on empirical mode decomposition and octave spectrum analysis |
| CN110081966A (en) * | 2019-05-27 | 2019-08-02 | 苏州工业职业技术学院 | The fault diagnosis system and method for fan vibration sensor |
| CN110956268A (en) * | 2019-10-16 | 2020-04-03 | 中国石化青岛液化天然气有限责任公司 | Compressor fault diagnosis method based on big data |
| CN110988677B (en) * | 2019-11-25 | 2021-11-09 | 北京昊鹏智能技术有限公司 | Fault detection method and device for direct current motor and mechanical equipment driven by direct current motor |
| CN110988677A (en) * | 2019-11-25 | 2020-04-10 | 北京昊鹏智能技术有限公司 | Fault detection method and device for direct current motor and mechanical equipment driven by direct current motor |
| CN111272427A (en) * | 2020-02-14 | 2020-06-12 | 西安交通大学 | Bearing fault detection method based on weighted sparse regularization |
| CN111272427B (en) * | 2020-02-14 | 2021-04-13 | 西安交通大学 | Bearing fault detection method based on weighted sparse regularization |
| CN112395959A (en) * | 2020-10-30 | 2021-02-23 | 天合云能源互联网技术(杭州)有限公司 | Power transformer fault prediction and diagnosis method and system based on audio features |
| CN112395959B (en) * | 2020-10-30 | 2024-01-02 | 天合光能股份有限公司 | Power transformer fault prediction and diagnosis method and system based on audio frequency characteristics |
| CN112433176A (en) * | 2020-11-17 | 2021-03-02 | 上海大学 | Power electronic converter fault diagnosis system and method |
| CN113009334B (en) * | 2021-02-18 | 2022-07-01 | 华中科技大学 | Motor fault detection method and system based on wavelet packet energy analysis |
| CN113009334A (en) * | 2021-02-18 | 2021-06-22 | 华中科技大学 | Motor fault detection method and system based on wavelet packet energy analysis |
| CN113325314A (en) * | 2021-07-20 | 2021-08-31 | 安徽巨森电器有限公司 | Motor fault diagnosis method |
| CN115993511A (en) * | 2023-03-22 | 2023-04-21 | 武汉纺织大学 | Partial discharge type high-precision detection and identification device, method and equipment |
| CN116381491A (en) * | 2023-06-05 | 2023-07-04 | 南京苏尚电气科技有限公司 | New energy automobile motor fault positioning system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108535648A (en) | Method of Motor Fault Diagnosis and system | |
| Ma et al. | High-voltage circuit breaker fault diagnosis using a hybrid feature transformation approach based on random forest and stacked autoencoder | |
| Grezmak et al. | Interpretable convolutional neural network through layer-wise relevance propagation for machine fault diagnosis | |
| Zhang et al. | Ensemble deep contractive auto-encoders for intelligent fault diagnosis of machines under noisy environment | |
| Zhang et al. | Fault diagnosis of planetary gearbox using a novel semi-supervised method of multiple association layers networks | |
| Dai et al. | Dissolved gas analysis of insulating oil for power transformer fault diagnosis with deep belief network | |
| Gu et al. | Bearing fault diagnosis with varying conditions using angular domain resampling technology, SDP and DCNN | |
| Zhao et al. | A weighted multi-scale dictionary learning model and its applications on bearing fault diagnosis | |
| Li et al. | Mechanical fault diagnosis based on redundant second generation wavelet packet transform, neighborhood rough set and support vector machine | |
| CN107817106B (en) | Fault Diagnosis of Roller Bearings based on Bayes's residual transform-singular value decomposition-Gaussian Mixture Hidden Markov Model frame | |
| CN109145706A (en) | A kind of sensitive features selection and dimension reduction method for analysis of vibration signal | |
| CN106197999B (en) | A kind of planetary gear method for diagnosing faults | |
| CN108197648A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models | |
| CN103728551B (en) | A kind of analog-circuit fault diagnosis method based on cascade integrated classifier | |
| CN105760839A (en) | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine | |
| CN109145949A (en) | Non-intrusive electrical load monitoring and decomposition method and system based on integrated study | |
| CN107766816A (en) | A kind of Mechanical Failure of HV Circuit Breaker recognition methods based on LVQ neutral nets | |
| Chen et al. | Hierarchical deep convolution neural networks based on transfer learning for transformer rectifier unit fault diagnosis | |
| CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
| CN107909118A (en) | A kind of power distribution network operating mode recording sorting technique based on deep neural network | |
| CN108535635A (en) | A kind of analog circuit intermittent fault diagnostic method based on EEMD and HMM | |
| CN108562853A (en) | Method of Motor Fault Diagnosis based on error correcting output codes support vector machines and system | |
| Wang et al. | Gear fault diagnosis based on variational modal decomposition and wide+ narrow visual field neural networks | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| Huang et al. | Deep residual networks-based intelligent fault diagnosis method of planetary gearboxes in cloud environments |
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: 20180914 |