CN113947099A - ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method - Google Patents
ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method Download PDFInfo
- Publication number
- CN113947099A CN113947099A CN202110759206.2A CN202110759206A CN113947099A CN 113947099 A CN113947099 A CN 113947099A CN 202110759206 A CN202110759206 A CN 202110759206A CN 113947099 A CN113947099 A CN 113947099A
- Authority
- CN
- China
- Prior art keywords
- phase
- lgbm
- amplitude
- reactive power
- current
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000003745 diagnosis Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000010845 search algorithm Methods 0.000 claims abstract description 5
- 238000002790 cross-validation Methods 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims description 16
- 238000004458 analytical method Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- 238000002474 experimental method Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 5
- 230000001052 transient effect Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000010183 spectrum analysis Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 2
- 230000009897 systematic effect Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- 238000003066 decision tree Methods 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Tests Of Circuit Breakers, Generators, And Electric Motors (AREA)
Abstract
The invention relates to a high-precision diagnosis method for the number of broken rotor bars of a five-phase asynchronous motor based on ESPRIT-PSA and LGBM. It uses ESPRIT (rotation invariant signal parameter estimation technique) and PSA (pattern search algorithm) to obtain the accurate amplitude A of characteristic component in the short-time (only 2 seconds) sampled instantaneous reactive power signalq(ii) a Then 31 characteristics such as corresponding voltage amplitude, current amplitude, average active power P and average reactive power Q are obtained; then, the LGBM (light gradient lifter) determines AqAnd the feature A with the weight being 5 in the 31 featuresqP, Q, phase 1 voltage amplitude UmAmplitude of phase 1 current ImThereby forming a data set; then training the LGBM model by the model, saving the LGBM model and carrying out high-precision diagnosis on the number of rotor broken bars(the training precision and the testing precision are both 100%, and the accuracy of 5-fold cross validation is 99.38%).
Description
Technical Field
The invention relates to a method capable of diagnosing the number of rotor broken bars of a five-phase asynchronous motor, and belongs to the technical field of fault diagnosis.
Background
Due to the advantages of high reliability, fault-tolerant operation and the like, the five-phase asynchronous motor is applied to special fields of ships, submarines and the like. The rotor broken bar is a typical fault mode of the five-phase asynchronous motor, so the fault diagnosis of the rotor broken bar plays an important role in improving the operation reliability of the five-phase asynchronous motor.
Rotor break of five-phase asynchronous motorIn the event of a bar fault, a frequency of (1 + -2 s) f will occur in its stator current1(s is slip, f)1Is the supply frequency). The method for diagnosing the broken rotor bar by examining the side frequency component is called a motor stator current signal analysis method. Research on such methods has matured, but at low slip rates, the sidefrequency component of the stator current may be represented by f1Frequency components are swamped, which poses a serious challenge for this type of approach.
Therefore, a method for carrying out spectrum analysis on the instantaneous reactive power signal and further realizing fault diagnosis of broken rotor bars is gradually developed and formed, and the essence is as follows: when the rotor has a broken bar fault, the frequency of the instantaneous reactive power of the rotor is 2sf1The characteristic component of (a). The method is generally called as a rotor broken bar fault diagnosis method of a motor instantaneous reactive power signal analysis class, and has the advantages that: even under the condition of low slip rate, the method can still accurately judge whether the rotor broken bar fault occurs.
After the occurrence of the rotor broken bar fault is accurately judged, the number of the rotor broken bars needs to be further diagnosed. This is because: the rotor broken bars are progressive faults, 1 guide bar is usually broken at the initial stage, other guide bars adjacent to the broken guide bar are continuously broken, and the output of the five-phase asynchronous motor is greatly reduced and even stops. If the number of the rotor broken bars can be diagnosed, the severity of the rotor broken bar fault can be naturally grasped, so that the maintenance can be arranged in time. Therefore, the diagnosis of the number of rotor bars is of great significance.
At present, a diagnosis formula of the number of rotor broken bars is provided by a motor instantaneous reactive power signal analysis method, but in practical application, a diagnosis result has larger deviation with the actual number of the rotor broken bars. In view of the above, the present invention adopts a method of combining ESPRIT (rotation invariant signal parameter estimation), PSA (pattern search algorithm) and LGBM (light weight gradient elevator) to diagnose the number of rotor breaks of the five-phase asynchronous motor.
Disclosure of Invention
The invention aims to provide a method for diagnosing the number of broken rotor bars of a five-phase asynchronous motor, which is based on an ESPRIT (rotation invariant signal parameter estimation technology), a PSA (pattern search algorithm) and an LGBM (light gradient hoisting machine) and takes an instantaneous reactive power signal sampled in a short time (only 2 seconds) as an analysis medium, has high diagnosis precision and is suitable for a low slip rate condition; in addition, the method is particularly suitable for severe interference conditions such as load fluctuation and noise because only short-time sampling is needed.
The problem is realized by the following technical scheme:
a method for diagnosing the number of rotor broken bars of five-phase asynchronous motor features that the instantaneous reactive power signal of five-phase asynchronous motor sampled in short time (only 2 seconds) is calculated by using ESPRIT to obtain its frequency of 2sf1The exact frequency value of the characteristic component of (a) and the rough amplitude and initial phase angle (s is the slip, f)1At the frequency of the power supply); then substituting the result obtained by ESPRIT calculation as initial value into PSA to calculate the accurate amplitude A of the characteristic componentqAnd an initial phase angle, and AqAs a first feature variable, placing in a data set X; and then, carrying out fine Fourier spectrum analysis on instantaneous signals of the five-phase voltage and the five-phase current of the five-phase asynchronous motor to obtain the amplitude of the five-phase voltage, the amplitude of the five-phase current, the phase of the five-phase voltage and the phase of the five-phase current of the stator, wherein 20 instantaneous signals and A are calculatedqA corresponding feature; then, according to the 20 characteristics, the effective values of the positive sequence voltage, the negative sequence voltage and the zero sequence current, the module values of the positive sequence impedance, the negative sequence impedance and the zero sequence impedance, and the average active power and the average reactive power are obtained, and the total of 11 and A are obtainedqA corresponding derivative feature; then the 31 pieces are connected with AqCorresponding features are put into the data set X to form a 32-dimensional data set X1(ii) a Further combine X1Inputting the signal into LGBM classifier to calculate the feature weight, determining and selecting the feature with the weight accounting for the first 5, namely the amplitude A of the instantaneous reactive power feature componentqAverage active power P, average reactive power Q and 1 st phase voltage amplitude UmAmplitude of phase 1 current ImForm a new data set X2(ii) a Then, X is added2According to the ratio of 82, dividing the proportion into a training set and a test set, inputting the training set and the test set into an LGBM model, and performing classification diagnosis on the number of broken bars of the rotor of the five-phase asynchronous motor, namely normal (0 broken bar), 1 broken bar and 2 broken bars; and finally, searching and selecting the optimal hyper-parameters of the LGBM model by using a GridSearchCV grid built in a scimit-learn library to ensure that the training precision of the model is highest, wherein the scimit-learn library is a free machine learning toolkit aiming at Python language, the scimit-learn library has various classification, regression and clustering algorithms and covers almost all mainstream machine learning algorithms including the LGBM, and the GridSearchCV is a parameter automatic adjusting module in the scirt-learn library, systematically traverses various parameter combinations by using an exhaustion method, and determines the parameters with the optimal effect through cross verification.
Attention is paid to: although the LGBM uses only the magnitude of the instantaneous reactive power characteristic component and not its frequency and initial phase angle in diagnosing the number of rotor bars, both the magnitude and initial phase angle must be calculated during application of the PSA.
The invention has the advantages that the training precision is 100 percent, the testing precision is 100 percent, the 5-fold cross validation accuracy is 99.38 percent, and the application of the LGBM in the field of five-phase asynchronous motor rotor broken bar fault diagnosis is successfully realized. Therefore, the number of the broken bars of the rotor of the five-phase asynchronous motor can be diagnosed, and meanwhile, the LGBM model obtained by training is saved for subsequent diagnosis.
The method for diagnosing the number of broken rotor bars of the five-phase asynchronous motor comprises the following steps of:
a. carrying out systematic and large amount of experiments to measure the stator five-phase voltage instantaneous signal usnAnd stator five-phase current instantaneous signal isn(n represents phase, n is 1, 2, … …, 5);
this work was carried out for 3 states of five-phase asynchronous motor being normal, 1 rotor broken bar fault occurring, 2 rotor broken bar faults occurring one by one, and the experiment in each state included load change (3 cases of full load, half load and no load).
The above-mentioned 3 states (normal, occurrence of 1 rotor breaking fault, occurrence of 2 rotor breaking faults) of the five-phase asynchronous motor are respectively and sequentially marked as states 0, 1 and 2.
b. Calculating the five-phase voltage and current instantaneous signals of the stator according to the formula (1) and the formula (2) to obtain instantaneous reactive power signals, and filtering the direct current components according to the formula (3) to obtain qA(instantaneous reactive power signal after filtering out the direct current component);
qA＝q0-mean(q0) (3)
in the formulae (1), (2) and (3),
representing the stator voltage transient u of the n-th phasesnThe Hilbert transform of (1); t represents time; τ represents the delay; q. q.s0Representing the instantaneous reactive power; mean (q)0) Denotes q0Average value (i.e., dc component).
c. To q isACarrying out ESPRIT analysis, and calculating to obtain an accurate frequency value, a rough amplitude value and an initial phase angle of the characteristic component in the instantaneous reactive power signal;
d. the amplitude and the initial phase angle of the instantaneous reactive power characteristic component calculated by ESPRIT are used as initial reference values, and the accurate amplitude A of the characteristic component is calculated by PSAqAnd an initial phase angle, then AqAs the first characteristic variable, into data set X (X is deposit A)qA data set of values);
e. performing refined Fourier analysis on instantaneous signals of the five-phase voltage and the five-phase current to obtain five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase of the stator, wherein the total number of the five-phase voltage amplitude, the five-phase current amplitude, the five-phase voltage phase and the five-phase current phase is 20 and AqCorresponding characteristics, which are then processed to obtain positive, negative and zero sequence voltage significance values, positive, negative and zeroThe effective value of sequence current, the positive sequence, negative sequence and zero sequence impedance module values, the average active power and the average reactive power are 11 and A in totalqCorresponding derivative features, then the above 31 are compared with AqCorresponding features are put into the data set X to form a 32-dimensional data set X1(X1For storing AqAnd 31 eigenvalues corresponding thereto) and then on dataset X using the LGBM classifier1Performing weight calculation, determining and selecting AqThe top weight accounts for the top 5 characteristic (amplitude A of the instantaneous reactive power characteristic component)qAverage active power P, average reactive power Q and 1 st phase voltage amplitude UmAmplitude of phase 1 current Im) As the characteristics of LGBM model training and learning, 1 new 5-dimensional data set X is formed2(X2A 5-dimensional data set composed of the characteristics of which the weight accounts for the first 5 obtained by LGBM calculation);
amplitude A of the characteristic component of instantaneous reactive powerqAverage active power P, average reactive power Q and 1 st phase voltage amplitude UmAmplitude of phase 1 current ImAll are obtained by processing the sampling signal by a sliding window method (the window comprises data with the duration of 2 seconds). Specifically, in the state 0, 750 sets of sample data can be obtained through the above operation under each load condition of full load, half load and no load, the state 0 includes 2250 sets of sample data, while in the states 1 and 2, 1250 sets of sample data can be obtained through the above operation under each load condition of full load, half load and no load, the states 1 and 2 each include 3750 sets of sample data, and 3 states together include 9750 sets of sample data.
f. For the above data set X2Marking (0, 1, 2) according to the corresponding states, and respectively corresponding to the state 0, the state 1 and the state 2;
g. data set X2Dividing the model into a training set and a test set according to the ratio of 8: 2, substituting the training set and the test set into an LGBM model for training, and carrying out model hyper-parameter tuning through GridSearchCV to obtain a model with the best effect;
h. and calling a jobb package in the scinit-spare library, storing the trained LGBM model into an executable code file (with the extension name of m), and reading the LGBM model through the jobb package when the LGBM model needs to be used (the jobb package can store the trained model and can be directly called when the LGBM model needs to be used, so that the method has the advantages of high efficiency and high reading speed).
Further, ESPRIT is explained as follows.
The application of ESPRIT makes it possible to calculate the exact frequency values of the characteristic components and the coarse amplitude and initial phase angles of the instantaneous reactive power signals with as short a duration as possible, briefly described below.
The rotation invariant Signal parameter Estimation Technique (ESPRIT, Estimation of Signal Parameters via Rotational Estimation Technique) is proposed and developed by r.roy, a.paulraj, t.kalith, and has become an effective tool for positive (remaining) chord Signal parameter (number and frequency) Estimation.
The sampled signal x (n) can be expressed as a combination of a series of cosine harmonic components, as shown in equation (4).
Wherein, TSRepresents a sampling period; n represents the number of sampling points; p represents the number of harmonics; a. thei、fi、φiRespectively showing the amplitude, frequency and initial phase angle of the ith harmonic.
Define y (N) ═ x (N +1), introduce the following m × N order matrix (guarantee m > p):
X(n)＝[x(n) x(n+1)…x(n+m-1)]T (5)
Y(n)＝[y(n) y(n+1)…y(n+m-1)]T (6)
in the expressions (5) and (6), T represents transposition.
Then the autocorrelation matrix of X (n) is
RXX＝E{X(n)XH(n)} (7)
And the cross-correlation matrix of X (n) and Y (n) is
RXY＝E{X(n)YH(n)} (8)
In the expressions (7) and (8), E represents mathematical expectation, and H represents conjugate transpose.
The ESPRIT procedure is as follows:
(a) constructing a correlation matrix R according to equations (7) and (8)XX、RXY；
(b) To RXXPerforming eigenvalue decomposition to determine its minimum eigenvalue sigma2；
(c) Calculation of R1＝RXX-σ2I, I represents an m-order unit array;
(d) calculation of R2＝RXY-σ2Z and Z are an m-order matrix,
(here, I represents an m-1 order unit matrix);
(e) to R1Performing singular value decomposition R1＝U∑VHWhere U is [ U ═ U1 U2]，
(U1、U2、∑1、∑2、V1、V2Are all to R1The results obtained by performing singular value decomposition, such as: sigma1Is a diagonal matrix composed of p main singular values);
(f) computing matrices
(g) To pair
Carrying out generalized eigenvalue decomposition to determine p generalized eigenvalues lambdai(i ═ 1, 2, … p) (the remaining m-p generalized eigenvalues are identical to 0);
(h) determining the frequency of each component of the sampled signal according to the generalized eigenvalue
Im(λi)、 Re(λi) Respectively representing the characteristic values lambdaiThe imaginary part and the real part of (c);
(i) computing matrices
(j) The calculation matrix c ═ λHλ)-1λHX, where c is a column vector c ═ c1 c2…cp]TAnd X is a column vector [ X (1) X (2) … X (N)]T；
(k) Determining the amplitude and initial phase angle A of each component of the sampled signali＝2|ci|、
After the five-phase asynchronous motor has a rotor broken bar fault, the instantaneous reactive power signal q after the direct-current component is filteredAThe ESPRIT performance can be analyzed using the simulation of equation (9) and the results are shown in Table 1. The slip s is 0.2% to reflect the low slip condition in engineering practice, and f1＝50Hz、Ts0.001 second, 2000N, 200 m.
qA＝A1cos[2π(2sf1)t+φ1]+A2cos[2π(4sf1)t+φ2] (9)
TABLE 1 ESPRIT calculation results
The data in table 1 show that: for a short-time sampling signal (only 2 seconds), the ESPRIT can accurately calculate the frequency of each frequency component (even if only the difference is 0.2 Hz); however, the calculation error is large for the amplitude and initial phase angle of each frequency component. In table 1, the calculation error means: absolute value/true value of (calculated value-true value) × 100%.
According to the formula (9), s, f are transformed randomly and combinatorially1、A1、φ1、A2、φ2The value of (2) is calculated in a large amount, and the result is consistent with the result.
From this it can be concluded that: applying ESPRIT to instantaneous reactive power signal analysis to implement rotor broken bar fault diagnosis is feasible and suitable for low slip rate conditions; in addition, the method is particularly suitable for serious interference conditions such as load fluctuation, noise and the like because only short-time sampling is needed; however, ESPRIT will not provide accurate results for the amplitude, the initial phase angle, of the rotor fault signature component.
Based on the calculation result of ESPRIT, the amplitude and initial phase angle of the characteristic component of the rotor fault can be further accurately calculated by using PSA, which is briefly described as follows.
The Pattern Search Algorithm (PSA) is a direct Search optimization method, which consists of 'Search movement' and 'Pattern movement', can perform optimization iteration at the same time by multivariable, and is suitable for multivariable Search. The exploration movement is to explore along the axial direction by a certain step length so as to reveal the change rule of the target function and detect the descending direction of the function; and the mode movement is directly searched along the favorable direction, so that a better iteration point is found by utilizing the found function change rule.
Considering optimization problems
min[E(α)]，α＝[α1 α1…αn]T
Wherein, E (alpha) is an objective function, alpha is an undetermined state enabling E (alpha) to take the minimum value, and min represents the minimum value. For this problem, the PSA basic steps are as follows:
(a) given an initial state α0In the axial direction e1，e1，…enStep delta, reduction ratio beta epsilon (0, 1), termination parameter epsilon, let y0＝α0。
(b) (exploration moving) pair
The following axial searches were performed in order:
order to
If it is
Then let y0＝y0+δ*ei(ii) a Otherwise, it orders
If it is
Then let y0＝y0-δ*ei。
(c) (Pattern shift) if E (y)0)＜E(α0) Then let alpha1＝y0+(y0-α0) At α1Turning to (b) for a new initial state to obtain a new iteration point y1If E (y)1)＜E(α1) Then let alpha1＝y1(ii) a Otherwise, let δ be β δ.
(d) If delta is less than or equal to epsilon, stopping; otherwise, go to (b).
For the sampling signal x (n) shown in the formula (1), ESPRIT is firstly applied to determine the frequency f of each frequency componentiAmplitude AiInitial phase angle phiiI is 1, 2, … p. From the above, fiIs accurate, and Ai、φiAnd the PSA treatment is still needed.
With PSA, it is critical to construct a feasible objective function, as follows.
The sampling signal x (n) shown in the formula (1) can be expressed as
Generating a p N matrix y1(n)、y2(n), specifically as follows:
y1(n)＝[cos(2πf1nTS) cos(2πf2nTS)…cos(2πfpnTS)]T，n＝1，2，…，N (11)
y2(n)＝[sin(2πf1nTS) sin(2πf2nTS)…sin(2πfpnTS)]T，n＝1，2，…，N (12)
let state α ═ α1 α2]In which α is1、α2Are respectively as
α1＝[A1cosφ1 A2cosφ2…Apcosφp] (13)
α2＝[A1sinφ1 A2sinφ2…Apsinφp] (14)
And an initial state alpha0May be set according to the calculation result of ESPRIT.
Constructing an objective function
E(α)＝(α1y1(n)-α2y2(n)-X)2 (15)
Here, X is a column vector [ X (1) X (2) … X (N)]T。
To this end, the PSA may be used to determine the amplitude A of each frequency component of the sampled signal x (n)iInitial phase angle phii，i＝1，2，…p。
For the instantaneous reactive power signal in case of a rotor bar failure of the five-phase asynchronous motor shown in equation (9), PSA was applied, and the results are shown in table 2.
The data in table 2 show that: for short-time sampled signals (only 2 seconds), the PSA can accurately calculate the amplitude and initial phase angle of each frequency component based on the calculation result of ESPRIT.
TABLE 2 PSA results
According to the formula (9), s, f are transformed randomly and combinatorially1、A1、φ1、A2、φ2The value of (2) is calculated in a large amount, and the result is consistent with the result.
From this it can be concluded that: the ESPRIT and the PSA are combined and applied to instantaneous reactive power signal analysis to diagnose the fault of the broken rotor bars, so that the method is feasible and suitable for the condition of low slip ratio, and is particularly suitable for the serious interference conditions such as load fluctuation, noise and the like because only a short-time signal sampling is needed.
The invention has two remarkable characteristics:
first, even for short-time signals (only 2 seconds), combining ESPRIT, PSA can still accurately estimate the characteristic component of rotor breaking fault-2 sf in instantaneous reactive power1The amplitude of the component is used as the 1 st reliable classification characteristic of machine learning (LGBM is selected by the invention); on the basis of ESPRIT and PSA analysis, a refined Fourier transform is applied to analyze transient signals of five-phase voltage and five-phase current in a short time (only 2 seconds), so that the average active power P, the average reactive power Q and the amplitude U of the phase voltage No. 1 of the stator are obtainedmThe amplitude I of the 1 st phase current of the statormIt is used as the 2 nd to 5 th reliable classification features of machine learning (LGBM is selected for the invention). Note that: p, Q, Um、ImThe equal variables are f corresponding to five-phase voltage and five-phase current instantaneous signals1Of the (dominant) component, so that a refined fourier transform can also yield accurate results using a short duration of 2 seconds.
Second, LGBM is trained and saved, resulting in LGBM-based diagnostic models.
Just because of the two significant features, the present invention has two distinct advantages:
first, because the transient reactive power signal sampled in a short time (only 2 seconds) is used as an analysis medium and ESPRIT and PSA are introduced, the method is suitable for low slip rate and serious interference situations such as load fluctuation and noise.
Second, the method has high accuracy because of the introduction of the LGBM.
Experiments prove that the method can still accurately diagnose the number of rotor broken bars even under the condition of low slip ratio, the model training precision is 100%, the test precision is 100%, and the 5-fold cross validation accuracy is 99.38%.
The invention will be further explained with reference to the drawings.
Drawings
FIG. 1 is an experimental wiring diagram;
FIG. 2 is a diagram of a decision tree growth strategy for LGBM.
Detailed Description
The invention provides a method for diagnosing the number of broken bars of a rotor of a five-phase asynchronous motor based on ESPRIT-PSA and LGBM, and the diagnosis accuracy of the method is as high as 99.38%. The invention is characterized in that ESPRIT-PSA is applied to the instantaneous reactive power fault component amplitude AqAs described in more detail below, and LGBM-based model training.
FIG. 1 is an experimental wiring diagram. The experimental motor is a five-phase asynchronous motor, and has a rated voltage of 380V, a rated power of 5.5kW and a rated frequency of 50 Hz. In order to carry out the rotor broken bar experiment, besides a normal rotor, two fault rotors (a hole is drilled on the conducting bar at a position 10mm away from an end ring, the depth is 15mm, and the diameter is 10mm) are additionally arranged for simulating broken bar faults. The two fault rotors respectively have one broken conducting bar and two continuous broken conducting bars. The data acquisition system acquires stator five-phase voltage instantaneous signals and stator five-phase current instantaneous signals through the current converter and the voltage converter. The load adopts a direct current dynamometer, and the five-phase asynchronous motor is respectively in a full load state, a half load state and an idle load state through adjustment of the direct current dynamometer.
A large number of experiments are carried out, and instantaneous signals of the five-phase voltage and the current of the stator are measured. The operation is carried out one by one aiming at 3 states of normal five-phase asynchronous motors, 1 rotor broken bar fault and 2 rotor broken bar faults, and the experiment in each state comprises the load change (full load, half load and no load) of the motor. The above-described 3 states of the motor are respectively and sequentially labeled as states 0, 1, and 2. Through this work, a large amount of motor sample data is acquired. Under the condition of the state 0, 750 groups of sample data can be obtained through the operation under each load condition of full load, half load and no load, the state 0 comprises 2250 groups of sample data, under the condition of the state 1 and the state 2, 1250 groups of sample data can be obtained through the operation under each load condition of full load, half load and no load, the state 1 and the state 2 respectively comprise 3750 groups of sample data, and then the 3 states of the motor totally comprise 9750 groups of sample data.
The LGBM used in the present invention is a new member of the Boosting algorithm, a machine learning method developed by Microsoft corporation. The method is an efficient implementation of a GBDT (gradient Boosting Decision Tree) algorithm, is similar to the GBDT algorithm in principle, and adopts the negative gradient of a loss function as a residual error approximate value of the current Decision tree to fit a new Decision tree. However, compared to the conventional machine learning algorithm, LGBM has significant advantages: the training efficiency is higher, the occupied memory is lower, the accuracy is higher, the parallel learning is supported, and large-scale data can be processed.
FIG. 2 is a diagram of a decision tree growth strategy for an LGBM that generates a decision tree by a leaf-wise (best-first) strategy. The LGBM picks the leaf node with the largest loss to grow a new leaf. The leaf-wise algorithm may reduce losses over the level-wise algorithm when growing the same number of leaves. But when the amount of data is small, the leaf-wise may cause an overfitting. Therefore, the LightGBM can limit the depth of the tree and avoid overfitting with an extra parameter max _ depth (the present invention sets the max _ depth value to 3). In addition, the method self-defines parameters such as min _ data _ in _ leaf, max _ bin, bagging _ fraction and the like through GridSearchCV to process the over-fitting problem of the model and improve the training speed of the model. The LGBM over-parameter values selected by the present invention are shown in table 3. For the meaning of the parameters in table 3, reference may be made to LGBM official documents, which are not described herein.
TABLE 3 LGBM hyper-parameter selection
To test the practical effect of the method of the present invention, 45 additional sets of data were measured for the normal state (state 0), 1 fault (state 1) and 2 faults (state 2) (15 sets per state, with the load condition randomly set to no load or half load or full load). The 45 sets of data were "blind tested" using the method of the present invention, and the results are shown in table 4. As can be seen from Table 4, the method of the present invention has high accuracy.
TABLE 4 LGBM model "Blind test" evaluation
Claims (3)
1. A high-precision diagnosis method for the number of broken rotor bars of a five-phase asynchronous motor based on ESPRIT-PSA and LGBM is characterized in that for an instantaneous reactive power signal of the five-phase asynchronous motor sampled in a short time (only 2 seconds), the ESPRIT (rotation invariant signal parameter estimation technology) is used for calculating the frequency of the instantaneous reactive power signal to be 2sf1The exact frequency value of the characteristic component of (a) and the rough amplitude and initial phase angle (s is slip, f)1At the frequency of the power supply); then substituting the result of ESPRIT calculation as an initial value into PSA (pattern search algorithm), thereby calculating the accurate amplitude A of the characteristic componentqAnd an initial phase angle, and AqAs a first feature variable, placing in a data set X; and then, carrying out refined Fourier spectrum analysis on instantaneous signals of the five-phase voltage and the five-phase current of the five-phase asynchronous motor to obtain the amplitude of the five-phase voltage, the amplitude of the five-phase current, the phase of the five-phase voltage and the phase of the five-phase current of the stator, wherein the total number of the signals is 20 and AqA corresponding feature; then, according to the 20 characteristics, the effective values of the positive sequence voltage, the negative sequence voltage and the zero sequence voltage, the effective values of the positive sequence current, the negative sequence current and the zero sequence current, the module values of the positive sequence impedance, the negative sequence impedance and the zero sequence impedance, and the average active power and the average reactive power are obtained, and 11 and A are calculated in totalqA corresponding derivative feature; then the 31 pieces are connected with AqCorresponding features are put into the data set X to form a 32-dimensional data set X1(ii) a Further combine X1Inputting into LGBM (light gradient hoist) classifier for feature weight calculation, determining and selecting feature with weight of 5 at the top, amplitude A of instantaneous reactive power feature componentqAverage active power P, average reactive power Q and 1 st phase voltage amplitude UmAmplitude of phase 1 current ImForm a new data set X2(ii) a Then, X is added2Dividing the test set into a training set and a test set according to the ratio of 8: 2 and inputting the training set and the test set into a pair of five-phase asynchronous power in an LGBM modelClassifying and diagnosing the number of broken bars of the motor rotor, namely normal (0 broken bar), 1 broken bar and 2 broken bars; finally, searching and selecting the optimal hyper-parameter of the LGBM model by using a GridSearchCV grid built in the scimit-learn library so as to ensure that the training precision of the model is highest; therefore, the number of the rotor broken bars of the five-phase asynchronous motor can be diagnosed with high precision (the training precision is 100%, the testing precision is 100%, and the 5-fold cross validation accuracy is 99.38%), and the LGBM model obtained by training is stored for subsequent diagnosis.
2. The ESPRIT-PSA and LGBM based high precision diagnosis method of rotor bar breakage in five phase asynchronous motors according to claim 1, comprising the steps of:
a. carrying out systematic and large amount of experiments to measure the stator five-phase voltage instantaneous signal usnAnd stator five-phase current instantaneous signal isn(n denotes phase, n ═ 1, 2, … …, 5) -this work was carried out one by one for 3 states of five-phase asynchronous motor normal (labeled as state 0), 1 rotor bar fault (labeled as state 1), 2 rotor bar fault (labeled as state 2), and the experiment in each state included load changes (3 cases of full load, half load and no load);
b. according to the formula (1) and the formula (2), instantaneous reactive power signals are obtained through calculation of stator pentaphase voltage and current instantaneous signals, and direct current components of the instantaneous reactive power signals are filtered according to the formula (3), so that q is obtainedA(instantaneous reactive power signal after filtering DC component) — A
qA＝q0-mean(q0) (3)
In the formulae (1), (2) and (3),
representing the stator voltage transient u of the n-th phasesnThe Hilbert transform of (1); t represents time; τ represents the delay; q. q.s0Representing the instantaneous reactive power; mean (q)0) Denotes q0Average value (i.e., direct current component) of (d);
c. to q isACarrying out ESPRIT analysis, and calculating to obtain an accurate frequency value, a rough amplitude value and an initial phase angle of the characteristic component in the instantaneous reactive power signal;
d. the amplitude and the initial phase angle of the instantaneous reactive power characteristic component calculated by ESPRIT are used as initial reference values, and the accurate amplitude A of the characteristic component is calculated by PSAqAnd an initial phase angle, then AqAs the first characteristic variable, into data set X (X is deposit A)qA data set of values);
e. performing refined Fourier analysis on instantaneous signals of the five-phase voltage and the five-phase current to obtain five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase of the stator, wherein the total number of the five-phase voltage amplitude, the five-phase current amplitude, the five-phase voltage phase and the five-phase current phase is 20 and AqCorresponding characteristics, these 20 characteristics are then processed to obtain positive, negative and zero sequence voltage, current, impedance and average active and reactive power, for a total of 11 and aqCorresponding derivative features, then the above 31 are compared with AqCorresponding features are put into the data set X to form a 32-dimensional data set X1(X1For storing AqAnd 31 eigenvalues corresponding thereto) and then on dataset X using the LGBM classifier1Performing weight calculation, determining and selecting AqThe top weight accounts for the top 5 characteristic (amplitude A of the instantaneous reactive power characteristic component)qAverage active power P, average reactive power Q and 1 st phase voltage amplitude UmAmplitude of phase 1 current Im) As a feature of LGBM model learning to form 1 new 5-dimensional dataset X2(X2A 5-dimensional dataset composed of the top 5 features for the LGBM calculated weights) — Aq、P、Q、Um、ImThe sampling signals are processed by adopting a sliding window method (a window comprises data with the time length of 2 seconds) to obtain the sampling signals (specifically, in a state 0, 750 groups of sample data can be obtained through the work under each load condition of full load, half load and no load, the state 0 comprises 2250 groups of sample data, in states 1 and 2, 1250 groups of sample data can be obtained through the work under each load condition of full load, half load and no load, the states 1 and 2 respectively comprise 3750 groups of sample data, and 3 types of sample data together comprise 9750 groups);
f. for the above data set X2Marking (0, 1, 2) according to the corresponding states, and respectively corresponding to the state 0, the state 1 and the state 2;
g. data set X2Dividing the model into a training set and a test set according to the ratio of 8: 2, substituting the training set and the test set into an LGBM model for training, and carrying out model hyper-parameter tuning through GridSearchCV to obtain a model with the best effect;
h. and calling a jobb package in the scinit-spare library, storing the trained LGBM model into an executable code file (with the extension name of m), and reading the LGBM model through the jobb package when the LGBM model needs to be used (the jobb package can store the trained model and can be directly called when the LGBM model needs to be used, so that the method has the advantages of high efficiency and high reading speed).
3. The method for diagnosing the rotor number of the five-phase asynchronous motor with high precision based on the ESPRIT-PSA and the LGBM as claimed in claim 2, wherein the light gradient elevator is used for classifying the states of the five-phase asynchronous motor (normal state-0, 1 rotor fault-state 1, 2 rotor fault-state 2), and the invention determines the super parameters of the light gradient elevator with the best effect, as shown in the following table.
Light gradient elevator over-parameter selection
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110759206.2A CN113947099A (en) | 2021-07-05 | 2021-07-05 | ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110759206.2A CN113947099A (en) | 2021-07-05 | 2021-07-05 | ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113947099A true CN113947099A (en) | 2022-01-18 |
Family
ID=79327282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110759206.2A Pending CN113947099A (en) | 2021-07-05 | 2021-07-05 | ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113947099A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115248377A (en) * | 2022-09-21 | 2022-10-28 | 北京科锐特科技有限公司 | Asynchronous motor rotor broken bar fault detection method, computer device and medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102279341A (en) * | 2011-07-23 | 2011-12-14 | 华北电力大学（保定） | Cage asynchronous motor rotor broken-bar fault detection method based on electronic stability program rotation invariant technology (ESPRIT) and pattern search algorithm (PSA) |
| CN109856538A (en) * | 2019-03-04 | 2019-06-07 | 中国矿业大学 | A kind of induction machine strip-broken failure detecting method based on correction FFT in short-term |
| CN110688836A (en) * | 2019-09-30 | 2020-01-14 | 湖南大学 | Automatic domain dictionary construction method based on supervised learning |
| CN111157894A (en) * | 2020-01-14 | 2020-05-15 | 许昌中科森尼瑞技术有限公司 | Motor fault diagnosis method, device and medium based on convolutional neural network |
| CN111506862A (en) * | 2020-05-01 | 2020-08-07 | 西北工业大学 | Rolling bearing fault diagnosis method based on multi-source weighted integration transfer learning |
| CN112731137A (en) * | 2020-09-15 | 2021-04-30 | 华北电力大学（保定） | Cage type asynchronous motor stator and rotor fault joint diagnosis method based on stack type self-coding and light gradient elevator algorithm |
| CN112861984A (en) * | 2021-02-25 | 2021-05-28 | 西华大学 | Speech emotion classification method based on feature fusion and ensemble learning |
-
2021
- 2021-07-05 CN CN202110759206.2A patent/CN113947099A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102279341A (en) * | 2011-07-23 | 2011-12-14 | 华北电力大学（保定） | Cage asynchronous motor rotor broken-bar fault detection method based on electronic stability program rotation invariant technology (ESPRIT) and pattern search algorithm (PSA) |
| CN109856538A (en) * | 2019-03-04 | 2019-06-07 | 中国矿业大学 | A kind of induction machine strip-broken failure detecting method based on correction FFT in short-term |
| CN110688836A (en) * | 2019-09-30 | 2020-01-14 | 湖南大学 | Automatic domain dictionary construction method based on supervised learning |
| CN111157894A (en) * | 2020-01-14 | 2020-05-15 | 许昌中科森尼瑞技术有限公司 | Motor fault diagnosis method, device and medium based on convolutional neural network |
| CN111506862A (en) * | 2020-05-01 | 2020-08-07 | 西北工业大学 | Rolling bearing fault diagnosis method based on multi-source weighted integration transfer learning |
| CN112731137A (en) * | 2020-09-15 | 2021-04-30 | 华北电力大学（保定） | Cage type asynchronous motor stator and rotor fault joint diagnosis method based on stack type self-coding and light gradient elevator algorithm |
| CN112861984A (en) * | 2021-02-25 | 2021-05-28 | 西华大学 | Speech emotion classification method based on feature fusion and ensemble learning |
Non-Patent Citations (7)
| Title |
|---|
| "lgbm.fi.plot: LightGBM Feature Importance Plotting", pages 1 - 2, Retrieved from the Internet <URL:http://rdrr.io/github/Laurae/man/lgbm.fi.plot.html> * |
| (美)布奇·昆托: "基于Spark的下一代机器学习：XGBoost、LightGBM、Spark NLP与Keras分布式深度学习实例", 31 May 2021, 机械工业出版社, pages: 148 * |
| IGNACIO MARTIN-DIAZ, AT EL.: "Early Fault Detection in Induction Motors Using AdaBoost With Imbalanced Small Data and Optimized Sampling", IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, 31 December 2017 (2017-12-31) * |
| 孙丽玲;许伯强;李志远;: "基于旋转不变信号参数估计技术与模式搜索算法的异步电动机转子故障检测新方法", 机械工程学报, no. 13, 5 July 2012 (2012-07-05) * |
| 汪雷鸣: "基于模型设计的车载异步电机参数辨识研究", 万方数据, 11 December 2019 (2019-12-11) * |
| 王曦;李兴源;王渝红;赵睿;: "基于TLS-ESPRIT辨识的多直流控制敏感点研究", 电力系统保护与控制, no. 19, 1 October 2012 (2012-10-01) * |
| 许伯强: "基于 SAE 与改进 LightGBM 算法的笼型异步电动机故障诊断方法", 电机与控制学报, 9 April 2021 (2021-04-09) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115248377A (en) * | 2022-09-21 | 2022-10-28 | 北京科锐特科技有限公司 | Asynchronous motor rotor broken bar fault detection method, computer device and medium |
| CN115248377B (en) * | 2022-09-21 | 2022-12-13 | 北京科锐特科技有限公司 | Asynchronous motor rotor broken bar fault detection method, computer equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xu et al. | Application of neural network algorithm in fault diagnosis of mechanical intelligence | |
| Xiao et al. | Fault diagnosis of asynchronous motors based on LSTM neural network | |
| Sadeghian et al. | Online detection of broken rotor bars in induction motors by wavelet packet decomposition and artificial neural networks | |
| Chen et al. | A meta-learning method for electric machine bearing fault diagnosis under varying working conditions with limited data | |
| Goumas et al. | Classification of washing machines vibration signals using discrete wavelet analysis for feature extraction | |
| CN112101457B (en) | PMSM demagnetizing fault diagnosis method based on torque signal fuzzy intelligent learning | |
| Huang et al. | Application of wavelet neural networks on vibration fault diagnosis for wind turbine gearbox | |
| Hamdani et al. | Neural network technique for induction motor rotor faults classification-dynamic eccentricity and broken bar faults | |
| Pietrzak et al. | PMSM stator winding fault detection and classification based on bispectrum analysis and convolutional neural network | |
| CN110174270A (en) | Multi-source time-frequency crestal line extracting method | |
| CN113947099A (en) | ESPRIT-PSA and LGBM-based five-phase asynchronous motor rotor broken number high-precision diagnosis method | |
| CN107992843A (en) | A kind of sparse features extracting method of rotating machinery fault signal | |
| Muduli et al. | Ann based on-board fault diagnostic for induction motor drive in low-cost electric vehicles | |
| Rinanto et al. | Rotor bars fault detection by DFT spectral analysis and Extreme Learning Machine | |
| CN113390631A (en) | Fault diagnosis method for gearbox of diesel engine | |
| Parvin et al. | A comprehensive interturn fault severity diagnosis method for permanent magnet synchronous motors based on transformer neural networks | |
| CN116956215A (en) | Fault diagnosis method and system for transmission system | |
| Yang et al. | A planetary gear reducer backlash identification based on servo motor current signal and optimized fisher discriminant analysis | |
| CN113947125A (en) | High-precision diagnosis method for rotor broken number of fifteen-phase asynchronous motor based on ESPRIT-PSA and LGBM | |
| CN112986821A (en) | Fault diagnosis method for broken blade of variable pitch motor rotor of offshore wind turbine generator | |
| Zamudio-Ramirez et al. | Detection of uniform gearbox wear in induction motors based on the analysis of stray flux signals through statistical time-domain features and dimensionality reduction techniques | |
| Zhiwu et al. | A rolling bearing fault diagnosis method based on fastDTW and an AGBDBN | |
| Ali et al. | Experimental investigation of machine learning based fault diagnosis for induction motors | |
| Kaminski et al. | General regression neural networks as rotor fault detectors of the induction motor | |
| CN110320033A (en) | A kind of rolling bearing health degenerate state discrimination method |
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 |