High-precision diagnosis method for the number of broken rotor bars of five-phase asynchronous motors based on ESPRIT-PSA and LGBM

Abstract

The present invention is 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 (rotationally invariant signal parameter estimation technology) and PSA (pattern search algorithm) to obtain the accurate amplitude A of the characteristic component of the instantaneous reactive power signal sampled for a short time (only 2 seconds). q ; Then get the corresponding voltage amplitude, current amplitude, average active power P and average reactive power Q and other 31 features; Then use LGBM (light gradient boosting machine) to determine A q And the top 5 features among the above 31 features - A q , P, Q, first phase voltage amplitude U m , the first phase current amplitude I m , thus forming a data set; then the LGBM model was trained and saved to perform high-precision diagnosis of the number of broken rotor bars (the training and testing accuracy were both 100%, and the 5-fold cross-validation accuracy was 99.38%).

Description

Technical Field

[0001] The invention relates to a method capable of diagnosing the number of broken bars of a five-phase asynchronous motor rotor, belonging to the technical field of fault diagnosis. Background Art

[0002] Due to its many advantages such as high reliability and fault-tolerant operation, five-phase asynchronous motors have been used in special fields such as ships and submarines. Broken rotor bars are a typical failure mode of five-phase asynchronous motors, so rotor broken bar fault diagnosis plays an important role in improving the reliability of five-phase asynchronous motor operation.

[0003] When a five-phase asynchronous motor has a broken rotor bar fault, a side frequency component with a frequency of (1±2s)f1 (s is the slip rate, f1 is the power supply frequency) will appear in its stator current. The method of achieving the purpose of rotor bar fault diagnosis by examining this side frequency component is called the motor stator current signal analysis method. The research on this type of method is mature, but in the case of low slip rate, the side frequency component of the stator current may be submerged by the f1 frequency component, which poses a severe challenge to this type of method.

[0004] To this end, a method for spectral analysis of instantaneous reactive power signals has been gradually developed and formed to realize rotor bar fault diagnosis. Its essence is that when a rotor bar fault occurs, a characteristic component with a frequency of 2sf1 will appear in its instantaneous reactive power. This type of method is usually called a rotor bar fault diagnosis method of the instantaneous reactive power signal analysis type of motor. Its advantage is that even in the case of low slip rate, this type of method can still accurately determine whether a rotor bar fault occurs.

[0005] After accurately determining that a broken rotor bar fault has occurred, it is necessary to further diagnose the number of broken rotor bars. This is because: a broken rotor bar is a progressive fault. Usually one conductor bar breaks in the early stage, and then other conductor bars adjacent to the broken conductor bar continue to break, and the output of the five-phase asynchronous motor will drop significantly or even cause shutdown. If the number of broken rotor bars can be diagnosed, the severity of the broken rotor bar fault can naturally be grasped, so that maintenance can be arranged in time. Therefore, the diagnosis of the number of broken rotor bars is of great significance.

[0006] At present, the instantaneous reactive power signal analysis method of the motor has provided a diagnostic formula for the number of broken rotor bars, but in practical applications, the diagnostic result is significantly different from the actual number of broken rotor bars. In view of this, the present invention adopts a method combining ESPRIT (rotationally invariant signal parameter estimation technology), PSA (pattern search algorithm) and LGBM (light gradient boosting machine) to diagnose the number of broken rotor bars of a five-phase asynchronous motor. Summary of the invention

[0007] The purpose of the present invention is to provide a method for diagnosing the number of broken rotor bars of a five-phase asynchronous motor based on ESPRIT (rotationally invariant signal parameter estimation technology), PSA (pattern search algorithm) and LGBM (light gradient boosting machine), using a short-time (only 2 seconds) sampled instantaneous reactive power signal as an analysis medium. The method has high diagnostic accuracy and is suitable for low slip conditions. In addition, since only a short-time sampling signal is required, the method is particularly suitable for conditions with severe interference such as load fluctuations and noise.

[0008] The problem addressed by the present invention is achieved by the following technical solution:

[0009] A diagnostic method for the number of broken rotor bars of a five-phase asynchronous motor is proposed. The method is based on the instantaneous reactive power signal of the five-phase asynchronous motor sampled for a short time (only 2 seconds). First, ESPRIT is used to calculate the accurate frequency value, rough amplitude and initial phase angle (s is the slip rate and f1 is the power supply frequency) of the characteristic component with a frequency of 2sf1; then the result calculated by ESPRIT is substituted into PSA as the initial value to calculate the accurate amplitude A of the characteristic component. q and initial phase angle, and A q As the first characteristic variable, it is placed in the data set X; then the instantaneous signals of the five-phase voltage and five-phase current of the five-phase asynchronous motor are refined by Fourier spectrum analysis to obtain the stator five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase, a total of 20 q Then, according to these 20 features, we can get the effective values ​​of positive-sequence, negative-sequence and zero-sequence voltages, the effective values ​​of positive-sequence, negative-sequence and zero-sequence currents, the modulus values ​​of positive-sequence, negative-sequence and zero-sequence impedances, and the average active power and average reactive power, a total of 11 features related to A q Corresponding derivative features; then the above 31 and A q The corresponding features are placed into the data set X to form a 32-dimensional data set X 1 ; Then X 1 The feature weights are calculated by inputting into the LGBM classifier, and the top 5 features with the highest weights are determined and selected, namely, the amplitude A of the instantaneous reactive power feature component. q , average active power P, average reactive power Q, first phase voltage amplitude U m , the first phase current amplitude I m , forming a new data set X 2 ; Then, X 2The data were divided into training set and test set in a ratio of 8:2 and input into the LGBM model to classify and diagnose the number of broken bars on the rotor of the five-phase asynchronous motor - normal (0 broken bars), 1 broken bar, and 2 broken bars; finally, the optimal hyperparameters of the LGBM model were selected by using the built-in GridSearchCV grid search of the scikit-learn library to achieve the highest training accuracy of the model. The scikit-learn library is a free machine learning toolkit for the Python language. It has various classification, regression, and clustering algorithms, covering almost all mainstream machine learning algorithms including LGBM. GridSearchCV is a parameter automatic adjustment module in the scikit-learn library. It uses the exhaustive method to systematically traverse various parameter combinations and determine the optimal parameters through cross-validation.

[0010] It should be noted that although LGBM only uses the amplitude of the instantaneous reactive power characteristic component without using its frequency and initial phase angle when diagnosing the number of broken rotor bars, in the process of applying PSA, the amplitude and initial phase angle must be calculated simultaneously.

[0011] The training accuracy of the invention is 100%, the test accuracy is 100%, and the 5-fold cross validation accuracy is 99.38%, which successfully realizes the application of LGBM in the field of five-phase asynchronous motor rotor broken bar fault diagnosis. Based on this, the number of five-phase asynchronous motor rotor broken bars can be diagnosed, and the trained LGBM model can be saved for subsequent diagnosis.

[0012] The above-mentioned method for diagnosing the number of broken rotor bars of a five-phase asynchronous motor comprises the following steps:

[0013] a. Carry out systematic and extensive experiments to measure the stator five-phase voltage instantaneous signal u sn And the stator five-phase current instantaneous signal i sn (n represents phase difference, n=1, 2, ..., 5);

[0014] This work was carried out one by one for three states of the five-phase asynchronous motor: normal, one rotor bar broken, and two rotor bar broken. The experiment in each state included load changes (full load, half load, and no load).

[0015] The above three states of the five-phase asynchronous motor (normal, one rotor bar broken fault, two rotor bar broken faults) are marked as states 0, 1, and 2 respectively.

[0016] b. According to equations (1) and (2), the instantaneous signals of the stator five-phase voltage and current are calculated to obtain the instantaneous reactive power signal and its DC component is filtered out according to equation (3), thereby obtaining q A(Instantaneous reactive power signal after filtering out DC component);

[0017]

[0018]

[0019] q A =q0-mean(q0) (3)

[0020] In formula (1), formula (2) and formula (3), Indicates the instantaneous signal u of the n-th phase stator voltage sn Hilbert transform; t represents time; τ represents delay; q0 represents instantaneous reactive power; mean(q0) represents the average value of q0 (i.e., the DC component).

[0021] c. for q A Perform ESPRIT analysis to calculate the accurate frequency value, rough amplitude and initial phase angle of the characteristic component in the instantaneous reactive power signal;

[0022] d. Using the amplitude and initial phase angle of the instantaneous reactive power characteristic component calculated by ESPRIT as the initial reference value, use PSA to calculate the accurate amplitude A of the characteristic component q and initial phase angle, then A q As the first feature variable, it is placed in the data set X (X is the storage of A q value dataset);

[0023] e. Perform a detailed Fourier analysis on the instantaneous signals of the five-phase voltage and five-phase current to obtain the stator five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase, a total of 20 q The corresponding features are then processed to obtain the positive, negative and zero sequence voltage effective values, positive, negative and zero sequence current effective values, positive, negative and zero sequence impedance moduli, and average active power and average reactive power, a total of 11 features corresponding to A q The corresponding derivative features are then combined with the above 31 q The corresponding features are placed into the data set X to form a 32-dimensional data set X 1 (X 1 To store A q and its corresponding 31 eigenvalues), and then use the LGBM classifier to classify the dataset X 1 Calculate the weights and determine and select A q The top 5 features with the highest weight (the amplitude of the instantaneous reactive power characteristic component A q , average active power P, average reactive power Q, first phase voltage amplitude Um , the first phase current amplitude I m ) as the features of LGBM model training and learning, thus forming a new 5-dimensional data set X 2 (X 2 A 5-dimensional dataset consisting of the top 5 features with the highest weights calculated by LGBM);

[0024] The amplitude of the instantaneous reactive power characteristic component A q And the average active power P, average reactive power Q, the first phase voltage amplitude U m , the first phase current amplitude I m All of them are obtained by processing the sampled signals through the sliding window method (the window includes data of 2 seconds in length). Specifically, in state 0, 750 sets of sample data can be obtained through the above work under each load condition of full load, half load and no load, and state 0 includes 2250 sets of sample data, while in state 1 and state 2, 1250 sets of sample data can be obtained through the above work under each load condition of full load, half load and no load, and state 1 and state 2 each include 3750 sets of sample data, so the three states contain a total of 9750 sets of sample data.

[0025] f. For the above dataset X 2 , marked according to their corresponding states (0, 1, 2), corresponding to state 0, state 1, and state 2 respectively;

[0026] g. Dataset X 2 The training set and the test set are divided into two sets in a ratio of 8:2, and the sets are put into the LGBM model for training. The model hyperparameters are tuned through GridSearchCV to obtain the best model.

[0027] h. Call the joblib package in the scikit-learn library to save the trained LGBM model into an executable code file (with the extension .m). When the LGBM model is needed, it can be read through the joblib package (the joblib package can save the trained model and call it directly when needed, with the advantages of high efficiency and fast reading speed).

[0028] Furthermore, ESPRIT is described as follows.

[0029] ESPRIT can be used to calculate the exact frequency value, rough amplitude, and initial phase angle of the characteristic component of an instantaneous reactive power signal with the shortest possible duration. A brief introduction is as follows.

[0030] The Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT) was proposed and developed by R. Roy, A. Paulraj, and T. Kailath, and has now become an effective tool for estimating sine (cosine) signal parameters (number and frequency).

[0031] The sampled signal x(n) can be expressed as a combination of a series of cosine harmonic components, as shown in equation (4).

[0032]

[0033] Among them, T S represents the sampling period; N represents the number of sampling points; p represents the number of harmonics; A i 、f i ,φ i Respectively represent the amplitude, frequency, and initial phase angle of the i-th harmonic.

[0034] Define y(n)=x(n+1), and introduce the following m×N matrix (ensuring m>>p):

[0035] X(n)=[x(n) x(n+1)…x(n+m-1)] T (5)

[0036] Y(n)=[y(n) y(n+1)…y(n+m-1)] T (6)

[0037] In formulas (5) and (6), T represents transposition.

[0038] Then the autocorrelation matrix of X(n) is

[0039] R XX =E{X(n)X H (n)} (7)

[0040] The cross-correlation matrix of X(n) and Y(n) is

[0041] R XY =E{X(n)Y H (n)} (8)

[0042] In formulas (7) and (8), E represents mathematical expectation and H represents conjugate transpose.

[0043] The ESPRIT steps are as follows:

[0044] (a) Construct the correlation matrix R according to equations (7) and (8) XX , RXY ;

[0045] (b) For R XX Perform eigenvalue decomposition and determine its minimum eigenvalue σ 2 ;

[0046] (c) Calculate R1 = R XX -σ 2 I, I represents the m-order unit matrix;

[0047] (d) Calculate R2 = R XY -σ 2 Z, Z is an m-order matrix, (Here, I represents the m-1 order unit matrix);

[0048] (e) Perform singular value decomposition on R1: R1 = U∑V H , where U = [U1 U2], (U1, U2, ∑1, ∑2, V1, V2 are all the results obtained by performing singular value decomposition on R1, such as: ∑1 is a diagonal matrix composed of p main singular values);

[0049] (f) Calculation matrix

[0050] (g) Yes Perform generalized eigenvalue decomposition and determine p generalized eigenvalues ​​λ i (i=1,2,…p)(the remaining mp generalized eigenvalues ​​are always equal to 0);

[0051] (h) Determine the frequency of each component of the sampled signal based on the generalized eigenvalue Im(λ i )、Re(λ i ) represent the eigenvalues ​​λ i The imaginary part and real part of

[0052] (i) Calculation matrix

[0053] (j) Calculate the matrix c = (λ H λ) -1 λ H X, where c is a column vector c = [c1 c2…c p ] T , and X is a column vector [x(1) x(2)…x(N)] T ;

[0054] (k) Determine the amplitude and initial phase angle A of each component of the sampling signal i =2|c i |、

[0055] After a five-phase asynchronous motor has a broken rotor bar fault, its instantaneous reactive power signal q after the DC component is filtered out is A The performance of ESPRIT can be analyzed by simulation using equation (9), and the results are shown in Table 1. Here, the slip rate s = 0.2% is selected to reflect the low slip rate in engineering practice, and f1 = 50 Hz, T s =0.001 seconds, N=2000, m=200.

[0056] q A =A1cos[2π(2sf1)t+φ1]+A2cos[2π(4sf1)t+φ2] (9)

[0057] Table 1 ESPRIT calculation results

[0058]

[0059] The data in Table 1 show that for a short-time sampling signal (only 2 seconds), ESPRIT can accurately calculate the frequency of each frequency component (even if the difference is only 0.2Hz); however, the calculation error of the amplitude and initial phase angle of each frequency component is quite large. In Table 1, the calculation error refers to: the absolute value of (calculated value - true value) / true value × 100%.

[0060] According to formula (9), the values ​​of s, f1, A1, φ1, A2, and φ2 are randomly and combinatorially changed, and a large number of calculations are performed, and the results are consistent with the above.

[0061] Based on this, it can be inferred that applying ESPRIT to instantaneous reactive power signal analysis to implement rotor bar broken fault diagnosis is feasible and applicable to low slip conditions; in addition, since only short-term signal sampling is required, it is particularly suitable for situations with severe interference such as load fluctuations and noise; however, ESPRIT will not be able to provide accurate results for the amplitude and initial phase angle of the characteristic components of the rotor bar broken fault.

[0062] Based on the calculation results of ESPRIT, PSA can be used to further accurately calculate the amplitude and initial phase angle of the characteristic component of the broken rotor bar fault, as briefly described below.

[0063] Pattern Search Algorithm (PSA) is a type of direct search optimization method, which consists of "exploratory movement" and "pattern movement". It can perform optimization iterations for multiple variables at the same time and is suitable for multi-variable search. Exploratory movement is to explore along the axis with a certain step length, the purpose is to reveal the change law of the target function and detect the descending direction of the function; while pattern movement is to search directly along the favorable direction, the purpose is to use the discovered function change law to find a better iteration point.

[0064] Consider the optimization problem

[0065] min[E(α)], α=[α1 α1…α n ] T

[0066] Among them, E(α) is the objective function, α is the state to be determined that makes E(α) take the minimum value, and min means finding the minimum value. For this problem, the basic steps of PSA are as follows:

[0067] (a) Given an initial state α 0 , axial direction e1, e1, …e n , step size δ, reduction rate β∈(0,1), termination parameter ε, let y 0 =α 0 .

[0068] (b) (Exploring Mobile) Perform the following axial searches in sequence:

[0069] make like Then let y 0 =y 0 +δ*e i Otherwise, let like Then let y 0 =y 0 -δ*e i .

[0070] (c) (Mode shift) If E(y 0 )<E(α 0 ), then let α 1 =y 0 +(y 0 -α 0 ), with α 1 is the new initial state, go to (b), and get the new iteration point y 1 ——If E(y 1 )<E(α 1 ), then let α 1 =y 1 ; Otherwise, let δ = β*δ.

[0071] (d) If δ≤ε, stop; otherwise, go to (b).

[0072] For the sampled signal x(n) shown in equation (1), ESPRIT is first used to determine the frequency f of each frequency component. i , Amplitude A i 、 Initial phase angle φ i , i = 1, 2, ... p. From the above, we can see that f iis accurate, and A i ,φ i Pending PSA processing.

[0073] The key to applying PSA is to construct a feasible objective function, as follows.

[0074] The sampling signal x(n) shown in formula (1) can be expressed as

[0075]

[0076] Generate p×N matrices y1(n) and y2(n) as follows:

[0077] y1(n)=[cos(2πf1nT S ) cos(2πf2nT S )…cos(2πf p nT S )] T , n=1,2,…,N (11)

[0078] y2(n)=[sin(2πf1nT S ) sin(2πf2nT S )…sin(2πf p nT S )] T , n=1,2,…,N (12)

[0079] Let state α = [α1 α2], where α1 and α2 are

[0080] α1=[A1cosφ1 A2cosφ2…A p cosφ p ] (13)

[0081] α2=[A1sinφ1 A2sinφ2…A p sinφ p ] (14)

[0082] The initial state α0 can be set according to the calculation results of ESPRIT.

[0083] Constructing the objective function

[0084] E(α)=(α1y1(n)-α2y2(n)-X) 2 (15)

[0085] Here, X is a column vector [x(1) x(2)…x(N)] T .

[0086] At this point, PSA can be used to determine the amplitude A of each frequency component of the sampled signal x(n) i 、Initial phase angle φ i , i=1,2,…p.

[0087] For the instantaneous reactive power signal of the five-phase asynchronous motor under the condition of broken rotor bar fault shown in equation (9), PSA is applied and the results are shown in Table 2.

[0088] The data in Table 2 show that for short-time sampling signals (only 2 seconds), based on the calculation results of ESPRIT, PSA can accurately calculate the amplitude and initial phase angle of each frequency component.

[0089] Table 2 PSA results

[0090]

[0091] According to formula (9), the values ​​of s, f1, A1, φ1, A2, and φ2 are randomly and combinatorially changed, and a large number of calculations are performed, and the results are consistent with the above.

[0092] Based on this, it can be inferred that the combination of ESPRIT and PSA for instantaneous reactive power signal analysis to implement rotor bar broken fault diagnosis is feasible and applicable to low slip conditions. Therefore, only short-term signal sampling is required and it is particularly suitable for severe interference conditions such as load fluctuations and noise.

[0093] The present invention has two significant characteristics:

[0094] First, even for short-term signals (only 2 seconds), the combination of ESPRIT and PSA can still accurately estimate the characteristic component of the rotor broken bar fault - the amplitude of the 2sf1 component in the instantaneous reactive power, which is used as the first reliable classification feature of machine learning (LGBM is selected in the present invention); based on the analysis of ESPRIT and PSA, the refined Fourier transform is applied to analyze the short-term (only 2 seconds) five-phase voltage and five-phase current instantaneous signals, so as to obtain the average active power P, the average reactive power Q, the stator first phase voltage amplitude U m , stator first phase current amplitude I m , which are used as the 2nd to 5th reliable classification features for machine learning (LGBM is used in this invention). Note: P, Q, U m ,I m The variables all correspond to the f1 (main frequency) component of the five-phase voltage and five-phase current instantaneous signals. Therefore, the refined Fourier transform can also obtain accurate results using 2 seconds of short-term data.

[0095] Second, train and save LGBM to obtain a diagnostic model based on LGBM.

[0096] Due to the above two significant characteristics, the present invention has two unique advantages:

[0097] First, because the instantaneous reactive power signal sampled in a short time (only 2 seconds) is used as the analysis medium and ESPRIT and PSA are introduced, this method is suitable for low slip rate and serious interference conditions such as load fluctuation and noise.

[0098] Second, due to the introduction of LGBM, the method has high accuracy.

[0099] Experimental verification shows that this method can accurately diagnose the number of broken rotor bars even in the case of low slip, with a model training accuracy of 100%, a test accuracy of 100%, and a 5-fold cross-validation accuracy of 99.38%.

[0100] The present invention will be further described below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 This is the wiring diagram for the experiment;

[0102] Figure 2 It is the decision tree growth strategy diagram of LGBM. DETAILED DESCRIPTION

[0103] The present invention proposes a method for diagnosing the number of broken bars of a five-phase asynchronous motor rotor based on ESPRIT-PSA and LGBM, and the diagnostic accuracy of the method is as high as 99.38%. The present invention is characterized in that ESPRIT-PSA is used to diagnose the amplitude of the instantaneous reactive power fault component A q The extraction and model training based on LGBM are described in detail below.

[0104] Figure 1 The following is the wiring diagram of the experiment. The experimental motor is a five-phase asynchronous motor with a rated voltage of 380V, a rated power of 5.5kW, and a rated frequency of 50Hz. In order to conduct the rotor broken bar experiment, in addition to the normal rotor, two faulty rotors are equipped (holes are drilled on the conductor bar 10mm away from the end ring, with a depth of 15mm and a diameter of 10mm) to simulate the broken bar fault. The two faulty rotors have one broken conductor bar and two consecutive broken conductor bars respectively. The data acquisition system collects the stator five-phase voltage instantaneous signal and the stator five-phase current instantaneous signal through the current converter and the voltage converter. The load uses a DC dynamometer, and the DC dynamometer is adjusted to make the five-phase asynchronous motor in full load, half load and no load states respectively.

[0105] A large number of experiments were conducted to measure the instantaneous signals of the five-phase voltage and current of the stator. This work was carried out one by one for the three states of the five-phase asynchronous motor: normal, one rotor broken bar fault, and two rotor broken bar faults, and the experiments in each state included the load changes of the motor (full load, half load, and no-load). The above three states of the motor are marked as states 0, 1, and 2 respectively. Through this work, a large amount of motor sample data was obtained. When the motor is in state 0, 750 sets of sample data can be obtained through the above work under each load condition of full load, half load, and no-load, and state 0 includes 2250 sets of sample data. When the motor is in state 1 and state 2, 1250 sets of sample data can be obtained through the above work under each load condition of full load, half load, and no-load, and state 1 and state 2 each include 3750 sets of sample data, so the three states of the motor contain a total of 9750 sets of sample data.

[0106] The LGBM used in the present invention is a new member of the Boosting algorithm and is a machine learning method developed by Microsoft. It is an efficient implementation of the GBDT (Gradient Boosting Decision Tree) algorithm. In principle, it is similar to the GBDT algorithm, and both use the negative gradient of the loss function as the residual approximation of the current decision tree to fit a new decision tree. However, compared with traditional machine learning algorithms, LGBM has obvious advantages: higher training efficiency, lower memory usage, higher accuracy, support for parallel learning, and can process large-scale data.

[0107] Figure 2 It is the decision tree growth strategy diagram of LGBM. LGBM generates a decision tree through the leaf-wise (best-first) strategy. LGBM selects the leaf node with the largest loss to grow new leaves. When growing the same number of leaves, the leaf-wise algorithm can reduce the loss than the level-wise algorithm. However, when the amount of data is small, leaf-wise may cause overfitting. Therefore, LightGBM can use the additional parameter max_depth to limit the depth of the tree and avoid overfitting (the max_depth value set in the present invention is 3). In addition, the present invention customizes the min_data_in_leaf, max_bin, bagging_fraction and other parameters through GridSearchCV to deal with the overfitting problem of the model and improve the training speed of the model. The values ​​of the LGBM hyperparameters selected by the present invention are shown in Table 3. For the meaning of the parameters in Table 3, please refer to the official document of LGBM, which will not be repeated here.

[0108] Table 3 LGBM hyperparameter selection

[0109]

[0110] In order to test the actual effect of the method of the present invention, 45 sets of data were measured for the normal state (state 0), 1 broken bar fault (state 1) and 2 broken bar faults (state 2) (15 sets for each state, with the load conditions randomly set to no load, half load or full load). The method of the present invention was used to perform a "blind test" on the above 45 sets of data, and the results are shown in Table 4. As can be seen from Table 4, the method of the present invention has high accuracy.

[0111] Table 4 LGBM model “blind test” evaluation

[0112]

Claims

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, characterized in that: The following steps are involved: For the instantaneous reactive power signal of the five-phase asynchronous motor sampled for 2 seconds, the rotationally invariant signal parameter estimation technology ESPRIT is first used to calculate the accurate frequency value, rough amplitude and initial phase angle of the characteristic component with a frequency of 2sf1, where s is the slip rate and f1 is the power supply frequency; Then the result calculated by ESPRIT is substituted into the pattern search algorithm PSA as the initial value to calculate the accurate amplitude A of the characteristic component. q and initial phase angle, and A q Place it in the data set X as the first feature variable; Then, the instantaneous signals of the five-phase voltage and five-phase current of the five-phase asynchronous motor are refined by Fourier spectrum analysis to obtain the five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase of the stator, a total of 20 q Corresponding features; Then, based on these 20 characteristics, we can obtain the effective values ​​of positive, negative and zero sequence voltages, the effective values ​​of positive, negative and zero sequence currents, the modulus values ​​of positive, negative and zero sequence impedances, and the average active power and average reactive power, a total of 11 characteristics related to A q The corresponding derived features; Then the above 31 and A q The corresponding features are placed into the data set X to form a 32-dimensional data set X 1 ; Then X 1 The feature weights are calculated by inputting into the light gradient boosting machine LGBM classifier, and the amplitude A of the instantaneous reactive power feature component with the top 5 weights is determined and selected. q , average active power P, average reactive power Q, first phase voltage amplitude U m , the first phase current amplitude I m , forming a new data set X 2 ; Then, X 2 The training set and the test set were divided into two sets in a ratio of 8:2 and input into the LGBM model to classify and diagnose the number of broken bars in the rotor of the five-phase asynchronous motor. 0 broken bars = normal, 1 broken bar, and 2 broken bars. Finally, the optimal hyperparameters of the LGBM model were selected by using the built-in GridSearchCV grid search of the scikit-learn library to achieve the highest training accuracy of the model. Based on this, the number of broken rotor bars of a five-phase asynchronous motor can be diagnosed with high accuracy, and the trained LGBM model can be saved for subsequent diagnosis. Specifically include: a. Carry out the experiment and measure the instantaneous signal u of the stator five-phase voltage sn And the stator five-phase current instantaneous signal i sn , n represents the phase, n = 1, 2, ..., 5. This work is for the five-phase asynchronous motor. Normal, marked as state 0, one rotor bar fault occurs, marked as state 1, and two rotor bar faults occur, marked as state 2. These three states are carried out one by one, and the experiment in each state includes three load changes: full load, half load and no load. b. According to equations (1) and (2), the instantaneous reactive power signal is calculated from the instantaneous signals of the stator five-phase voltage and current, and its DC component is filtered out according to equation (3), thereby obtaining the instantaneous reactive power signal q after the DC component is filtered out. A q A =q0-mean(q0) (3) In formula (1), formula (2) and formula (3), Indicates the instantaneous signal u of the n-th phase stator voltage sn Hilbert transform; t represents time; τ represents delay; q0 represents instantaneous reactive power; mean(q0) represents the average value of q0, i.e., the DC component; c. for q A Perform ESPRIT analysis to calculate the accurate frequency value, rough amplitude and initial phase angle of the characteristic component in the instantaneous reactive power signal; d. Using the amplitude and initial phase angle of the instantaneous reactive power characteristic component calculated by ESPRIT as the initial reference value, use PSA to calculate the accurate amplitude A of the characteristic component q and initial phase angle, then A q As the first feature variable, it is placed in the data set X. X is the storage of A q Dataset of values; e. Perform a detailed Fourier analysis on the instantaneous signals of the five-phase voltage and five-phase current to obtain the stator five-phase voltage amplitude, five-phase current amplitude, five-phase voltage phase and five-phase current phase, a total of 20 q The corresponding features are then processed to obtain the positive, negative and zero sequence voltage effective values, positive, negative and zero sequence current effective values, positive, negative and zero sequence impedance moduli, and average active power and average reactive power, a total of 11 features corresponding to A q The corresponding derivative features are then combined with the above 31 q The corresponding features are placed into the data set X to form a 32-dimensional data set X 1 , X 1 To store A q And the 32-dimensional data set with its corresponding 31 eigenvalues, and then use the LGBM classifier to classify the data set X 1 Calculate the weights and determine and select A q The features with the highest weights are A, the amplitude of the instantaneous reactive power characteristic component is q , average active power P, average reactive power Q, first phase voltage amplitude U m , the first phase current amplitude I m , as the feature learned by the LGBM model, thus forming a new 5-dimensional data set X 2 , X 2 A is a 5-dimensional dataset composed of the top 5 features calculated by LGBM. q , P, Q, U m ,I m They are all obtained by processing the sampling signals using the sliding window method, and the window includes data of 2 seconds in length. Specifically, in state 0, 750 sets of sample data can be obtained through the above work under each load condition of full load, half load, and no load, and state 0 includes 2250 sets of sample data; in state 1 and state 2, 1250 sets of sample data can be obtained through the above work under each load condition of full load, half load, and no load, and state 1 and state 2 each include 3750 sets of sample data, so the three states contain a total of 9750 sets of sample data; f. For the above dataset X 2 , marked 0, 1, 2 according to their corresponding states, corresponding to state 0, state 1, and state 2 respectively; g. Dataset X 2 The training set and the test set are split into two sets in a ratio of 8:2, and the sets are put into the LGBM model for training. The model hyperparameters are tuned through GridSearchCV to obtain the best model. h. Call the joblib package in the scikit-learn library to save the trained LGBM model into an executable code file with a .m extension. When the LGBM model is needed, it can be read through the joblib package. The joblib package can save the trained model and can be directly called when needed.

2. According to claim 1, the 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: The light gradient lifting machine is used to classify the states of the five-phase asynchronous motor, specifically: normal is state 0, one rotor bar fault is state 1, and two rotor bar faults are state 2. The best hyperparameters of the light gradient lifting machine are determined as follows:

Images