A method for detecting static eccentricity fault of a wound brushless doubly-fed machine

Abstract

The application discloses a winding type brushless doubly-fed motor static eccentric fault detection method, and belongs to the motor fault detection field. A finite element model of a prototype under different static eccentricity rates is established; original data sets of stator power winding currents are obtained; power winding current characteristic frequency signals are obtained; an improved EMD algorithm is adopted to decompose effective intrinsic mode functions, to form characteristic data sets, and a random forest diagnosis model based on the marine predator algorithm optimization is constructed; the static eccentric fault severity is evaluated; and the static eccentric fault and the fault severity are on-line diagnosed. The improved random forest diagnosis model is integrated on the basis of the winding type brushless doubly-fed motor, so that the on-line monitoring and real-time diagnosis functions of the motor can be realized, the diagnosis precision is high, the operability is strong, and the accuracy of the random forest classification prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of motor fault detection technology, specifically to a method for detecting static eccentricity faults in a wound-rotor brushless doubly fed motor. Background Technology

[0002] Wound-rotor brushless doubly-fed motors feature adjustable power factor, multifunctionality, and safety and reliability. Furthermore, their specially designed rotor winding structure improves winding utilization and reduces harmonic content, facilitating their practical application. However, rotor eccentricity faults are highly susceptible to occur due to manufacturing and assembly errors, as well as bearing wear. These faults can lead to distortion of the air gap magnetic field and a decline in motor performance. If fault detection is not timely, the unbalanced magnetic pull at both ends of the bearing will further increase, the air gap will further decrease, and serious consequences such as rotor-stator "sweeping" can occur, threatening normal industrial production and the safety of personnel. Therefore, early warning of eccentricity faults is crucial.

[0003] A comprehensive review of existing literature reveals the following key aspects of current motor eccentricity fault detection technologies:

[0004] 1) Unlike traditional induction motors, brushless doubly-fed motors have unique stator and rotor winding structures and complex air gap magnetic fields. Furthermore, eccentricity faults can further distort the air gap magnetic field, increasing the difficulty of fault diagnosis. For example, in the study of air gap eccentricity faults in large submersible motors based on tooth magnetic field analysis, non-invasive detection methods such as stator current analysis and vibration signal analysis are affected by the inherent high harmonic content of wound-rotor brushless doubly-fed motors. In this case, determining the severity of the eccentricity fault by the amplitude of characteristic frequency points in the spectrum will result in amplitude fluctuations, which will affect the accuracy of the diagnostic results.

[0005] 2) Traditional invasive test coil methods often perform spectrum analysis on the induced voltage collected by a single test coil. However, static eccentricity only changes the number of spatial harmonic pole pairs without altering the spatial harmonic frequency. Therefore, this method is more suitable for diagnosing faults caused by dynamic eccentricity. Moreover, patent CN 104965175A improves upon the traditional testing method by arranging four detection coils in the motor air gap. By comparing the voltages of each detection coil, the eccentricity fault and its severity can be determined. However, this requires modifications to the motor structure early in the manufacturing process, increasing manufacturing costs and making it unsuitable for practical applications with motors that have small air gaps. Furthermore, it is not applicable to some pre-manufactured motors, hindering mass production testing.

[0006] In summary, there is relatively little research literature on eccentricity faults in wound-rotor brushless doubly-fed motors, both domestically and internationally. Furthermore, it is difficult to achieve convenient, accurate, and reliable online detection and early warning of eccentricity faults in this type of motor using existing technologies. Summary of the Invention

[0007] The present invention proposes a method for detecting static eccentricity faults in a wound-rotor brushless doubly fed motor, which can at least solve one of the above-mentioned technical problems.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for detecting static eccentricity faults in a wound-rotor brushless doubly-fed motor includes the following steps:

[0010] (1) Establishing a finite element model: Establishing a finite element simulation model of a wound-rotor brushless doubly fed motor under different eccentricities; the different eccentricities are set according to the principle of taking data at equal intervals. In order to ensure that the test effect is obvious under different intervals, the static eccentricities are set to a wide range of α%, 2α%, 3α%, 4α%... etc.; where α can be selected according to actual needs.

[0011] (2) Obtaining the original dataset: The stator current signals with different eccentricities are collected by a high-precision current acquisition device to obtain the original current datasets under different eccentricities; the stator current signal is the current signal of the motor power winding; the winding current signal is less affected by harmonics and glitches.

[0012] (3) Current signal decomposition and spectrum analysis: Fast Fourier decomposition is performed on the original current dataset, and spectrum analysis is used to determine whether the dataset contains fault characteristic frequency signals; the fault characteristic frequency signals are:

[0013]

[0014] In the formula: f ph f is the characteristic frequency of the power winding current; ch To control the characteristic frequency of the winding current; f p f is the fundamental frequency of the power winding current; c To control the fundamental frequency of the winding current; s p S is the slip of the power winding; c To control winding slip; m r This represents the number of phases of the rotor winding;

[0015] (4) Improved EMD algorithm decomposition: The original current dataset under different eccentricities is decomposed by improving the EMD algorithm and the effective intrinsic mode function (IMF) is extracted;

[0016] The effective intrinsic mode function Y i The original current signal X should satisfy the following:

[0017]

[0018] In the formula: r xy X is the correlation function; Y is the acquired raw current signal; i Let EY be the i-th IMF component, i = 0, 1, 2, 3...; EX is the mean of the original current signal; EY is the mean of the IMF component. i Let be the mean of the i-th intrinsic mode function; DX be the variance of the original current signal; DY be the mean of the eigenmode function. i Let be the variance of the i-th intrinsic mode function.

[0019] (5) Construct an improved random forest diagnostic model: Collect effective intrinsic mode functions (IMFs) under different eccentricities to form a feature dataset, randomly select 70% of the data in the feature dataset as the training set and 30% of the data as the test set; train a random forest based on the improved ocean predator algorithm through the training set to form a fault diagnosis system.

[0020] (6) Test the random forest diagnostic model: Import the remaining 30% of the test set data in the feature dataset into the improved random forest fault diagnosis system and evaluate the corresponding static eccentricity.

[0021] Specifically, step (3) includes the following steps:

[0022] (3a) The stator power winding of the motor is connected to the power grid, and the stator control winding is connected to the power grid after being connected to the frequency converter. At this time, the resultant magnetomotive force f(θ, t) generated by the three-phase symmetrical winding of the stator is:

[0023] f(θ, t) = F p cos[w p t-(1±6k p )p p θ-θ p ]+F c cos[w c t+(1±6k c )p c θ-θ c (2)

[0024] In the formula, F p F c The fundamental magnetomotive forces of the stator power winding and control winding are respectively, w p w c The angular frequencies of the power winding and control winding are p, respectively. p p c These represent the number of pole pairs for the power winding and the control winding, respectively; and p p≠p c θ p θ c These are the initial position angles.

[0025] (3b) When the motor experiences an eccentricity fault, the air gap length changes, and the air gap permeability is expressed as:

[0026]

[0027] In the formula, μ0 is the vacuum permeability, δ0 is the air gap length in the healthy state, and ε s The static eccentricity is given. Because the number of pole pairs in the stator power winding and control winding of a wound-rotor brushless doubly-fed motor is inconsistent, the fundamental magnetic field will not be directly coupled. Therefore, the influence of the fundamental magnetomotive force of the power winding and control winding is studied separately, and then the two are coupled together.

[0028] (3c) Air gap magnetic flux density under the influence of the fundamental magnetic field of the power winding under static eccentricity fault:

[0029] B(θ, t) = F p C0cos(w p tp p θ)+F p C1cos[w p t-(p p ±1)θ] / 2 (4)

[0030] Furthermore, the slip ratio of the power winding and control winding of this type of motor can be expressed as:

[0031]

[0032] In the formula f p f c These are the fundamental frequencies of the power winding and the control winding, respectively.

[0033] (3d) The rotor magnetomotive force is multiplied by equation (3), and the result is expressed as:

[0034]

[0035] In the formula, F rp The amplitude of the rotor magnetomotive force after coupling between the fundamental magnetic field of the power winding and the rotor winding is m. r φ is the number of phases of the rotor winding. m This represents the corresponding initial position angle.

[0036] (3f) Similarly, the rotor magnetomotive force after coupling between the fundamental magnetic field of the control winding and the rotor winding can also be obtained, which will not be elaborated upon due to space limitations. Moreover, a significant characteristic frequency signal will only be induced in the winding when the number of pole pairs of the air gap magnetic field is related to the number of pole pairs of the winding. In summary, unlike traditional AC motors, wound-rotor brushless doubly-fed motors need to comprehensively consider the influence of the fundamental magnetomotive force of the power winding and the fundamental magnetomotive force of the control winding on the power winding current spectrum after coupling with the rotor winding. Therefore, the characteristic frequency signals of the healthy state and static eccentricity fault can be summarized.

[0037] In a healthy state:

[0038]

[0039] Since static eccentricity only changes the number of pole pairs of the spatial magnetic field without changing the characteristic frequency, this type of motor can comprehensively consider the influence of the fundamental magnetomotive force of the control winding and the rotor winding coupling on the power winding spectrum.

[0040] Under static eccentricity fault:

[0041]

[0042] Therefore, it can be determined whether the characteristic frequency point appears in the stator current spectrum according to Equation 8. The characteristic frequency point can be directly used to determine whether a static eccentricity fault exists in a wound-rotor brushless doubly fed motor.

[0043] Specifically, step (4) includes the following steps:

[0044] (4a) Determine the position and value of the first extreme value (maximum) and the last extreme value (maximum) of each original dataset;

[0045] (4b) If the first value in the curve formed by the dataset is a maximum (minimum), then by comparing the first time point with the first minimum (maximum), if the value at the first time point is larger (smaller), then the two maximum (minimum) points after the first maximum (minimum) value and the two minimum (maximum) points after that point are successively extended to the left; otherwise, the two maximum (minimum) values ​​after that time point and one minimum (maximum) value are shifted to the left.

[0046] (4c) The rightward extension of the last maximum (minimum) value of this dataset follows the same pattern as described above; the purpose is to extend the upper and lower envelopes to reach every signal point, thereby reducing the impact of the divergence effect of the data at both ends on the accuracy of EMD decomposition.

[0047] (4d) Calculate the correlation r between the i-th intrinsic mode function of each raw data and the raw current signal. xy ;

[0048] (4e) Extract the correlation r respectivelyxy Effective intrinsic mode functions greater than 0.

[0049] Step (5) specifically includes the following steps:

[0050] (5a) Constructing a feature dataset: After the current datasets under different eccentricities are processed by the improved EMD, the effective IMF components are combined together and labeled with the health status as 0, and labeled with 1, 2, 3... according to different static eccentricities to form a feature signal set; the current datasets under different eccentricities are health status, α% static eccentricity, 2α% static eccentricity...nα% static eccentricity data.

[0051] (5b) Test set and training set division: Data is extracted by random replacement, with 70% of the data in the feature dataset designated as P_train and T_train and 30% designated as P_test and T_test; the data in the feature dataset needs to be normalized to adapt to the new fault diagnosis model.

[0052] (5c) Population initialization definition: Population size pop, maximum number of iterations Max_iter, improved dimension dim=2 (improved the number of leaf nodes and decision trees), upper boundary lb and lower boundary ub, FADs=0.2, P=0.5.

[0053] (5d) Generation of fitness function fobj: First, P_train, T_train and the two parameters to be improved, n_layers and n_trees, are used to generate the model model through the random forest classification function; then, P_train and model are used to generate the corresponding test results through the random forest prediction function; finally, the corresponding fitness function fobj is generated according to the improvement target.

[0054] (5f) MPA algorithm process: First, initialize the predator matrix, convergence curve matrix, fitness function, etc.; second, start the loop iteration, detect the top predator, and update the fitness function fobj through ocean memory to form the elite matrix; third, under three different conditions, predators and prey choose Brownian motion or Levy motion respectively to form the prey matrix; finally, update the fitness function fobj again through ocean memory, and use FADs to escape the local optimum to complete the parameter optimization processing of the number of decision trees and the number of leaf nodes.

[0055] (5g) Improved construction of random forest diagnostic model: Extract the number of improved decision trees and leaf nodes to improve the random forest algorithm, and train the improved random forest algorithm through the training set to form a fault diagnosis model.

[0056] As can be seen from the above technical solution, the method for detecting static eccentricity faults in wound-rotor brushless doubly-fed motors of the present invention aims to diagnose static eccentricity faults in wound-rotor brushless doubly-fed motors. In terms of overall architecture, it utilizes a wound-rotor brushless doubly-fed motor and a random forest fault detection model optimized based on the Ocean Searcher algorithm to propose a novel method for detecting static eccentricity faults in wound-rotor brushless doubly-fed motors. This method enables online monitoring and real-time diagnosis of this type of motor, offering advantages such as ease of implementation, rapid application, and batch detection. It also improves the accuracy and precision of random forest classification, making it suitable for engineering fields such as static eccentricity fault detection in wound-rotor brushless doubly-fed motors.

[0057] Because eccentric faults alter the air gap length between the stator and rotor, leading to changes in air gap magnetic flux density, the distortion of the air gap magnetic field causes an imbalance in the originally symmetrical magnetic field. Although traditional motor theory states that static eccentric faults only change the number of pole pairs of spatial harmonics without altering the time harmonic frequency, wound-rotor brushless doubly-fed motors have two sets of windings with different numbers of pole pairs. While the time harmonic frequency of the power winding current remains unchanged, the coupling between the fundamental magnetomotive force of the control winding and the rotor winding affects the spectrum of the power winding current. When the number of air gap pole pairs and the number of winding pole pairs are correlated, characteristic frequency harmonics are generated in the power winding current spectrum. Therefore, this can serve as a theoretical basis for diagnosing static eccentric faults in wound-rotor brushless doubly-fed motors.

[0058] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0059] 1. Based on this invention, a fault detection method for static eccentricity of a wound-rotor brushless doubly-fed motor is proposed. By analyzing the air gap magnetic field of this type of motor, a novel characteristic frequency point for static eccentricity of the wound-rotor brushless doubly-fed motor is proposed. Furthermore, this characteristic frequency point can be used to intuitively and accurately achieve early warning of static eccentricity faults, while avoiding the need for early modification of the motor and the increased cost of invasive testing methods.

[0060] 2. Based on the proposed random forest detection method optimized by the marine predator algorithm, this invention improves the accuracy and precision of classification by optimizing the number of decision trees and leaf nodes in traditional random forests. Furthermore, it addresses the impact of the inherently high harmonic content of this type of motor on the amplitude fluctuations of characteristic frequency points, thereby enabling the identification and classification of the severity of static eccentricity faults. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the fault detection principle of an embodiment of the present invention;

[0062] Figure 2a This is a spectrum analysis diagram of the health status according to an embodiment of the present invention;

[0063] Figure 2b This is a spectrum analysis diagram of the static eccentric fault current according to an embodiment of the present invention;

[0064] Figure 3a This is a graph showing the accuracy analysis of the random forest algorithm on the test set according to an embodiment of the present invention.

[0065] Figure 3b This is a graph showing the accuracy analysis of the test set of the random forest algorithm based on MPA improvement according to an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0067] Combination Figure 1 The flowchart illustrating the fault detection principle of this invention, and the method for detecting static eccentricity faults in a wound-rotor brushless doubly-fed motor proposed in this invention, mainly includes the following steps:

[0068] (1) Establishing a finite element model: Establishing a finite element simulation model of a wound-rotor brushless doubly fed motor under different eccentricities; the different eccentricities are set according to the principle of taking data at equal intervals. In order to ensure that the test effect is obvious under different intervals, the static eccentricities are set to a wide range of α%, 2α%, 3α%, 4α%... etc.; where α can be selected according to actual needs.

[0069] (2) Obtaining the original dataset: The stator current signals with different eccentricities are collected by a high-precision current acquisition device to obtain the original current datasets under different eccentricities; the stator current signal is the current signal of the motor power winding; the winding current signal is less affected by harmonics and glitches.

[0070] (3) Current signal decomposition and spectrum analysis: Fast Fourier decomposition is performed on the original current dataset, and spectrum analysis is used to determine whether the dataset contains fault characteristic frequency signals; the fault characteristic frequency signals are:

[0071]

[0072] In the formula: f ph f is the characteristic frequency of the power winding current; ch To control the characteristic frequency of the winding current; f p f is the fundamental frequency of the power winding current; c To control the fundamental frequency of the winding current; s p S is the slip of the power winding; cTo control winding slip; m r This represents the number of phases of the rotor winding;

[0073] (4) Improved EMD algorithm decomposition: The original current dataset under different eccentricities is decomposed by improving the EMD algorithm and the effective intrinsic mode function (IMF) is extracted;

[0074] The effective intrinsic mode function Y i The original current signal X should satisfy the following:

[0075]

[0076] In the formula: r xy X is the correlation function; Y is the acquired raw current signal; i Let EY be the i-th IMF component, i = 0, 1, 2, 3...; EX is the mean of the original current signal; EY is the mean of the IMF component. i Let be the mean of the i-th intrinsic mode function; DX be the variance of the original current signal; DY be the mean of the eigenmode function. i Let be the variance of the i-th intrinsic mode function.

[0077] (5) Construct an improved random forest diagnostic model: Collect effective intrinsic mode functions (IMFs) under different eccentricities to form a feature dataset, randomly select 70% of the data in the feature dataset as the training set and 30% of the data as the test set; train a random forest based on the improved ocean predator algorithm through the training set to form a fault diagnosis system.

[0078] (6) Test the random forest diagnostic model: Import the remaining 30% of the test set data in the feature dataset into the improved random forest fault diagnosis system and evaluate the corresponding static eccentricity.

[0079] Step (3) specifically includes the following steps:

[0080] (3a) The stator power winding of the motor is connected to the power grid, and the stator control winding is connected to the power grid after being connected to the frequency converter. At this time, the resultant magnetomotive force f(θ,t) generated by the three-phase symmetrical winding of the stator is:

[0081] f(θ, t) = F p cos[w p t-(1±6k p )p p θ-θ p ]+F c cos[w c t+(1±6k c )p c θ-θ c (2)

[0082] In the formula, F p Fc The fundamental magnetomotive forces of the stator power winding and control winding are respectively, w p w c The angular frequencies of the power winding and control winding are p, respectively. p p c These represent the number of pole pairs for the power winding and the control winding, respectively; and p p ≠p c θ p θ c These are the initial position angles.

[0083] (3b) When the motor experiences an eccentricity fault, the air gap length changes, and the air gap permeability is expressed as:

[0084]

[0085] In the formula, μ0 is the vacuum permeability, δ0 is the air gap length in the healthy state, and ε s The static eccentricity is given. Because the number of pole pairs in the stator power winding and control winding of a wound-rotor brushless doubly-fed motor is inconsistent, the fundamental magnetic field will not be directly coupled. Therefore, the influence of the fundamental magnetomotive force of the power winding and control winding is studied separately, and then the two are coupled together.

[0086] (3c) Air gap magnetic flux density under the influence of the fundamental magnetic field of the power winding under static eccentricity fault:

[0087] B(θ, t) = F p C0cos(w p tp p θ)+F p C1cos[w p t-(p p ±1)θ] / 2 (4)

[0088] Furthermore, the slip ratio of the power winding and control winding of this type of motor can be expressed as:

[0089]

[0090] In the formula f p f c These are the fundamental frequencies of the power winding and the control winding, respectively.

[0091] (3d) The rotor magnetomotive force is multiplied by equation (3), and the result is expressed as:

[0092]

[0093] In the formula, F rp The amplitude of the rotor magnetomotive force after coupling between the fundamental magnetic field of the power winding and the rotor winding is m. r φ is the number of phases of the rotor winding. m This represents the corresponding initial position angle.

[0094] (3f) Similarly, the rotor magnetomotive force after coupling between the fundamental magnetic field of the control winding and the rotor winding can also be obtained, which will not be elaborated here due to space limitations. Moreover, a significant characteristic frequency signal will only be induced in the winding when the number of pole pairs of the air gap magnetic field is related to the number of pole pairs of the winding.

[0095] In summary, unlike traditional AC motors, wound-rotor brushless doubly fed motors need to comprehensively consider the influence of the fundamental magnetomotive force of the power winding and the fundamental magnetomotive force of the control winding on the power winding current spectrum after coupling with the rotor winding. Therefore, the characteristic frequency signals of the healthy state and static eccentricity fault can be summarized.

[0096] In a healthy state:

[0097]

[0098] Since static eccentricity only changes the number of pole pairs of the spatial magnetic field without changing the characteristic frequency, this type of motor can comprehensively consider the influence of the fundamental magnetomotive force of the control winding and the rotor winding coupling on the power winding spectrum.

[0099] Under static eccentricity fault:

[0100]

[0101] Therefore, it can be determined whether the characteristic frequency point appears in the stator current spectrum according to Equation 8. The characteristic frequency point can be directly used to determine whether a static eccentricity fault exists in a wound-rotor brushless doubly fed motor.

[0102] Step (4) specifically includes the following steps:

[0103] (4a) Determine the position and value of the first extreme value (maximum) and the last extreme value (maximum) of each original dataset;

[0104] (4b) If the first value in the curve formed by the dataset is a maximum (minimum), then by comparing the first time point with the first minimum (maximum), if the value at the first time point is larger (smaller), then the two maximum (minimum) points after the first maximum (minimum) value and the two minimum (maximum) points after that point are successively extended to the left; otherwise, the two maximum (minimum) values ​​after that time point and one minimum (maximum) value are shifted to the left.

[0105] (4c) The rightward extension of the last maximum (minimum) value of this dataset follows the same pattern as described above; the purpose is to extend the upper and lower envelopes to reach every signal point, thereby reducing the impact of the divergence effect of the data at both ends on the accuracy of EMD decomposition.

[0106] (4d) Calculate the correlation r between the i-th intrinsic mode function of each raw data and the raw current signal.xy ;

[0107] (4e) Extract the correlation r respectively xy Effective intrinsic mode functions greater than 0.

[0108] Step (5) specifically includes the following steps:

[0109] (5a) Constructing a feature dataset: After the current datasets under different eccentricities are processed by the improved EMD, the effective IMF components are combined together and labeled with the health status as 0, and labeled with 1, 2, 3... according to different static eccentricities to form a feature signal set; the current datasets under different eccentricities are health status, α% static eccentricity, 2α% static eccentricity...nα% static eccentricity data.

[0110] (5b) Test set and training set division: Data is extracted by random replacement, with 70% of the data in the feature dataset designated as P_train and T_train and 30% designated as P_test and T_test; the data in the feature dataset needs to be normalized to adapt to the new fault diagnosis model.

[0111] (5c) Population initialization definition: Population size pop, maximum number of iterations Max_iter, improved dimension dim=2 (improved the number of leaf nodes and decision trees), upper boundary lb and lower boundary ub, FADs=0.2, P=0.5.

[0112] (5d) Generation of fitness function fobj: First, P_train, T_train and the two parameters to be improved, n_layers and n_trees, are used to generate the model model through the random forest classification function; then, P_train and model are used to generate the corresponding test results through the random forest prediction function; finally, the corresponding fitness function fobj is generated according to the improvement target.

[0113] (5f) MPA algorithm process: First, initialize the predator matrix, convergence curve matrix, fitness function, etc.; second, start the loop iteration, detect the top predator, and update the fitness function fobj through ocean memory to form the elite matrix; third, under three different conditions, predators and prey choose Brownian motion or Levy motion respectively to form the prey matrix; finally, update the fitness function fobj again through ocean memory, and use FADs to escape the local optimum to complete the parameter optimization processing of the number of decision trees and the number of leaf nodes.

[0114] (5g) Improved construction of random forest diagnostic model: Extract the number of improved decision trees and leaf nodes to improve the random forest algorithm, and train the improved random forest algorithm through the training set to form a fault diagnosis model;

[0115] The remaining test set data is imported into the improved random forest fault diagnosis system to assess the severity of static eccentricity faults and to perform classification prediction of static eccentricity rate.

[0116] Figure 2 illustrates the spectrum analysis of the health status and static eccentric fault current of the present invention. Taking a 3kW, 2 / 4-pole wound-rotor brushless doubly-fed motor as an example, Table 1 shows the parameters of the wound-rotor brushless doubly-fed motor. As can be seen from Table 1, the motor achieves its maximum efficiency when the power winding fundamental frequency is 110Hz and the control winding fundamental frequency is 40Hz, allowing it to be rectified and used in marine shaft-driven power generation systems.

[0117] Table 1

[0118]

[0119] Table 2 shows the time harmonic frequencies in the air gap magnetic field. As can be seen from Table 2, the air gap magnetic field of a wound-rotor brushless doubly-fed induction generator contains harmonic components resulting from the indirect coupling of the power winding fundamental magnetomotive force and the rotor winding, as well as harmonic components resulting from the indirect coupling of the control winding fundamental magnetomotive force and the rotor winding. Therefore, under the condition that the number of pole pairs in the air gap magnetic field is related to the number of pole pairs in the windings, the influence of the air gap magnetic field formed after the coupling of the control winding fundamental magnetomotive force and the rotor winding on the power winding current spectrum needs to be taken into account.

[0120] Table 2

[0121]

[0122] Figure 2(a) shows the spectrum analysis of the power winding current under healthy conditions. When the motor is in a healthy state, the characteristic frequency points satisfy Equation 1. Combining the parameters in Table 2, when the power winding k1 is even and the control winding k2 is odd, the time harmonic frequencies 110Hz, 190Hz, 410Hz, 490Hz, 710Hz, and 790Hz satisfy kp. p The condition will appear in the current spectrum of the power winding; however, when k2 is odd, the time harmonic frequencies of 40Hz, 260Hz, 340Hz, 560Hz, and 640Hz do not satisfy kp. p If the conditions are not met, the condition will not occur.

[0123] Figure 2(b) shows the spectrum analysis of the power winding current under static eccentricity fault. When the motor experiences a static eccentricity fault, the characteristic frequency points under static eccentricity fault conditions satisfy Equation 2. Combining the parameters in Table 2, the power winding current spectrum is not affected by the static eccentricity fault. When k1 is even, the time harmonic frequencies of 110Hz, 190Hz, 410Hz, 490Hz, 710Hz, and 790Hz satisfy kp. p The condition will appear in the current spectrum of the power winding; and due to the influence of static eccentricity, the number of air gap pole pairs of the control winding is P. c + k2P c m r + 2. When k2 is odd, satisfying the condition that the number of air gap magnetic field pole pairs is related to the number of power winding pole pairs, then its time harmonic frequencies of 40Hz, 260Hz, 340Hz, 560Hz, and 640Hz will appear in the power winding current spectrum. Therefore, the characteristic frequency points of 40Hz, 260Hz, 340Hz, 560Hz, and 640Hz that appear due to static eccentricity faults can be used as the basis for fault detection.

[0124] Specific Implementation Method 3: Figure 3 illustrates a comparison of the accuracy of the improved random forest algorithm based on the marine predator algorithm and the traditional random forest algorithm. The specific implementation steps are as follows:

[0125] Figure 3(a) shows the accuracy analysis of the traditional random forest algorithm on the test set, and Figure 3(b) shows the accuracy analysis of the traditional random forest algorithm on the test set improved based on the marine predator algorithm. In this example, finite element models of prototypes with eccentricities of 0%, 20%, 40%, 60%, and 80% were established, and their stator current signals were collected as the original dataset. Feature datasets were obtained through improved EMD decomposition, and then defined as labeled 0, 1, 2, 3, and 4, respectively. The accuracy of the test set data was compared under the premise that the accuracy of the test set data was 100%. The comparison of the two figures shows that the random forest algorithm improved based on the marine predator algorithm, by optimizing the number of leaf nodes and the number of decision trees, achieved an improvement in the accuracy of random forest classification prediction.

[0126] In summary, this invention proposes, on the one hand, fault characteristic frequency points for static eccentricity of wound-rotor brushless doubly fed motors, effectively solving the problem of online monitoring of static eccentricity faults in this type of motor based on the stator current method; on the other hand, it proposes a random forest algorithm based on the ocean predator algorithm optimization, which improves the accuracy of the diagnostic model by optimizing the number of leaf nodes and the number of decision trees.

[0127] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of any of the methods described above.

[0128] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of any of the methods described above.

[0129] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of any of the methods described in the above embodiments.

[0130] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0131] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0132] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting static eccentricity fault in a wound rotor brushless doubly-fed machine, characterized in that, The method comprises the following steps, S1, establishing a finite element simulation model of the motor under different eccentricity rates; S2, based on the finite element simulation model and deriving the original stator current data under the corresponding conditions; S3, performing fast Fourier decomposition on the original current data set respectively, judging whether the fault characteristic frequency signal exists in the data set through spectrum analysis, and if the fault characteristic frequency signal exists, entering the next step; S4, decomposing the original current data set under different eccentricity rates through the improved EMD algorithm respectively, and extracting the effective intrinsic mode function IMF; S5, training the random forest algorithm based on the marine predator algorithm optimization through the training set to form a fault diagnosis system; S6, importing the remaining test set data into the improved random forest fault diagnosis system to evaluate the static eccentricity rate; The fault characteristic frequency signal in step S3 is: ； wherein: is the power winding current characteristic frequency; is the control winding current characteristic frequency; is the power winding current fundamental frequency; is the control winding current fundamental frequency; is the power winding slip; is the control winding slip; is the number of rotor winding phases; The effective eigenmode function in the step S4 Y i The original current signal X should satisfy: （1） wherein: r xy is the correlation function; X is the original current signal collected; Y i is the i-th IMF component, i = 0, 1, 2, 3, ···; EX is the mean of the original current signal; EY i is the mean of the i-th intrinsic mode function; DX is the variance of the original current signal; DY i is the variance of the i-th intrinsic mode function.

2. The method for detection of static eccentricity fault in a wound brushless doubly-fed machine according to claim 1, characterized in that: Step S3 specifically includes: 3a, the motor stator power winding is connected to the power grid, and the stator control winding is connected to the frequency converter and then connected to the power grid, at this time the synthesized magnetic potential generated by the three-phase symmetrical winding of the stator f (θ,t) is: （2） wherein F p , F c are the fundamental magnetic potentials of the stator power winding and control winding, respectively, w p , w c are the angular frequencies of the power winding and control winding, respectively, p p , p c are the pole pair numbers of the power winding and control winding, respectively; and p p ≠ p c , θ p , θ c are the initial position angles, respectively. 3b, when the motor is eccentric, the air gap length changes, and the air gap permeance is represented as: （3） wherein μ 0 is the vacuum permeability, δ 0 is the healthy state air gap length, ε s is the static eccentricity; 3c, under the static eccentricity fault, the air gap flux density under the influence of the power winding fundamental magnetic field is: （4） The slip rate of the power winding and the control winding of such a motor is represented as: （5） wherein f p , f c are the fundamental frequencies of the power winding and the control winding, respectively; 3d, the rotor magnetic potential is multiplied by formula (3), and the result is represented as: （6） wherein F rp is the rotor magnetic potential amplitude after coupling of the power winding fundamental magnetic field with the rotor winding, m r is the number of phases of the rotor winding, is the corresponding initial position angle; 3f, the characteristic frequency signals of the healthy state and the static eccentricity fault are as follows: Under the healthy state: （7） Under the static eccentricity fault: （8） According to formula (8), it is judged whether the characteristic frequency point exists in the stator current spectrum, and the characteristic frequency point is directly used to judge whether the static eccentricity fault of the brushless doubly-fed motor exists.

3. The method of claim 1, wherein: Step S4 specifically includes: 4a, determine the position and value of the first extreme value point and the last extreme value point of each original data set; 4b, if the first point of the curve formed by the data set is a maximum or minimum value, compare the first time point and the first minimum or maximum value, if the value at the time point is larger or smaller, then the two maximum or minimum value points after the first maximum or minimum value and the two minimum or maximum value points after the point are extended to the left one by one; otherwise, the two maximum or minimum values after the time point and a minimum or maximum value point are moved to the left; 4c, the right extension rule of the last maximum or minimum value of the data set is consistent with the above rule; the purpose is to extend the upper and lower envelope lines to contact each signal point, so that the influence of the divergence effect of the two ends on the accuracy of EMD decomposition is smaller; 4d. calculating a correlation of each original data eigenmode function i with the original current signal r xy ; 4e, extracting correlation r xy effective eigenmode functions greater than 0.

4. The method of claim 1, wherein: Step S5 specifically includes: 5a, form a feature data set: after the current data set under different eccentricity rates is processed by the improved EMD, the effective IMF components are synthesized together, and the health state label is 0, and according to different static eccentricity rates, the labels are 1, 2, 3…… to form a feature signal set; the current data set under different eccentricity rates is the health state, α% static eccentricity rate, 2α% static eccentricity rate……nα% static eccentricity rate data; 5b, test set, training set division: through the random replacement extraction data, 70% of the data in the feature data set as P_train, T_train and 30% of the data as P_test, T_test; the feature data in the data need to be normalized to adapt to the new fault diagnosis model; 5c, population initialization definition: population quantity pop, maximum iteration number Max_iter, improved dimension dim=2, improved leaf node number and decision tree number, upper bound lb and lower bound ub, FADs=0.2, P=0.5; 5d, fitness function fobj generation: first, P_train, T_train and the two parameters to be improved n_layers, n_trees are generated through the random forest classification function model; second, P_train and model are generated through the random forest prediction function to generate the corresponding test results; finally, the corresponding fitness function fobj is generated according to the improvement target; 5f, MPA algorithm process: first, initialize the predator matrix, convergence curve matrix and fitness function; second, start loop iteration, detect top predators, and update the fitness function fobj through marine memory to form an elite matrix; third, through three different conditions, the predator and prey respectively select Brown motion or levy motion to form a prey matrix; finally, the fitness function fobj is updated again through marine memory, and the local optimal solution is jumped out through FADs to complete the parameter optimization processing of the decision tree number and the leaf node number; 5g, improved random forest diagnosis model construction: extract the improved decision tree number and leaf node number, and improve the random forest algorithm to form a fault diagnosis model through the training set.

5. A computer readable storage medium storing a computer program, the computer program being executed by a processor to make the processor execute the steps of the method according to any one of claims 1 to 4.

Images