Dynamic eccentricity fault diagnosis method of permanent magnet synchronous motor based on Elman neural network

Abstract

The present invention relates to the technical field of motor eccentricity fault diagnosis, and in particular to a method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on an Elman neural network. The method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on an Elman neural network comprises the following steps: S1, establishing a characteristic database of dynamic eccentricity fault of a permanent magnet synchronous motor; S2, establishing a dynamic eccentricity fault diagnosis model of a permanent magnet synchronous motor based on an Elman network; S3, collecting the stray magnetic field of the permanent magnet synchronous motor to be detected, and extracting fault characteristic values; S4, inputting the experimental data of step S3 into the model of step S2 to diagnose the dynamic eccentricity fault of the permanent magnet synchronous motor. The present invention can accurately diagnose the presence or absence of a dynamic eccentricity fault of a permanent magnet synchronous motor and the dynamic eccentricity ratio when a dynamic eccentricity fault exists; the non-invasive diagnostic method does not require the motor to be modified, and will not affect the normal operation of the motor; and the versatility is high.

Description

Technical Field

[0001] The invention relates to the technical field of motor eccentricity fault diagnosis, in particular to a permanent magnet synchronous motor dynamic eccentricity fault diagnosis method based on Elman neural network. Background Art

[0002] Permanent magnet synchronous motors have no brushes, slip rings, or excitation systems, and are simple in structure, high in power density, and high in efficiency. They are widely used in aerospace, national defense, CNC machine tools, electric vehicles, and other fields. In order to improve the performance of motors, extend the service life of motors, reduce the cost of motors, and avoid serious irreversible accidents, it is of great significance to adopt effective methods to diagnose motor failures.

[0003] Motor eccentricity refers to the misalignment of the axis of the rotor and the axis of the motor. It is a common motor fault. Defects in the motor manufacturing process, errors in the assembly process, impact and wear during operation, and rotor mass imbalance can all cause motor eccentricity. Motor eccentricity faults can be divided into: static eccentricity, dynamic eccentricity, and mixed eccentricity. Dynamic eccentricity refers to the misalignment of the center of the motor rotor C R With the motor stator center C S No overlap, the motor rotor rotates around its own center C R In addition to rotating, it also revolves around the stator center C S Rotation, such as Figure 1 As shown. The degree of dynamic eccentricity is usually expressed by the dynamic eccentricity ratio E = e / l0, where e is the eccentricity and l0 is the air gap length when the motor does not have an eccentricity fault. After the motor has dynamic eccentricity, the working condition of the motor bearing deteriorates sharply, the motor vibration intensifies, and the torque fluctuation is obvious. In severe cases, the motor may stop or even suffer permanent damage. In the early stage of a dynamic eccentricity fault in a permanent magnet synchronous motor, if the eccentricity fault is diagnosed in time, the motor can be maintained as early as possible to avoid the deterioration of the motor fault.

[0004] The existing motor dynamic eccentricity fault detection methods can be roughly divided into the following categories:

[0005] The first category: eccentricity fault diagnosis based on vibration signals. This method requires a high-precision acceleration sensor to measure the vibration acceleration on the surface of the casing, and to determine the type of eccentricity based on the characteristic frequency that appears in the vibration acceleration spectrum of the motor. Patent CN105698740A is based on the additional characteristic frequency components in the vibration acceleration spectrum to determine whether the motor has a dynamic eccentricity fault or a static eccentricity fault. However, the causes of abnormal vibration of the motor are complex, and it is difficult to exclude other interference signals. The diagnostic effect is not ideal, and the degree of eccentricity cannot be accurately diagnosed.

[0006] The second category: eccentricity fault diagnosis based on current / voltage signals in stator windings. This method can diagnose the eccentricity fault of the motor based on the characteristic frequency presented in the stator current / voltage. Patent CN107091986A decomposes the stator current under the motor eccentricity fault by wavelet, and diagnoses the dynamic eccentricity, static eccentricity and mixed eccentricity faults of the motor by extracting the energy value of the characteristic frequency band in the spectrum diagram. Patent CN109814030A diagnoses whether the motor has a dynamic eccentricity fault by detecting whether the output voltage of the stator winding of the synchronous generator has even harmonics. However, the fault characteristics of this type of method are weak, greatly affected by the load, easily submerged by noise, difficult to extract, and difficult to detect low-degree eccentricity faults.

[0007] The third category: eccentricity fault diagnosis based on magnetic flux signals. Patent CN103713261A arranges magnetic field detection rings at different axial positions on the same circumferential surface inside the stator core, and determines the type of eccentricity fault by comparing the magnetic field signal characteristics. Patent CN108614212A arranges Hall sensors equidistantly in the axial direction to obtain the axial magnetic induction intensity of the motor, and determines the fault type based on the fault characteristic value. However, this type of detection method requires a built-in magnetic field detection ring or the installation of a Hall sensor, which is an invasive detection method. Its process is complex and requires major changes to the motor. In addition, it is difficult to structurally avoid the risk of collision between the Hall sensor and the rotor for motors with a small air gap.

[0008] The fourth category: eccentricity fault diagnosis based on detection coil voltage signal. This method generally requires pre-embedded detection coils. Patent CN107192947A determines eccentricity faults by winding detection coils on each stator tooth and calculating fault characteristic values ​​based on the induced voltage signals of each coil over a period of time. However, this method requires pre-embedded multiple detection coils on the stator teeth, which is an intrusive detection method, with significant changes to the motor, complex process, and poor versatility.

[0009] The fifth category: eccentricity fault diagnosis based on magnetic flux leakage signal. Patent CN 103713261A sets a detection coil on the stator core yoke, measures the induced voltage and performs spectrum analysis on the detection coil voltage, and determines the eccentricity type and the minimum air gap position by the characteristic frequency and amplitude. However, this method requires setting a detection coil on the stator core yoke, which requires major changes to the motor, is difficult to operate, and cannot diagnose the size of the eccentricity. Summary of the invention

[0010] The purpose of the present invention is to provide a permanent magnet synchronous motor dynamic eccentricity fault diagnosis method based on Elman neural network, which overcomes the shortcomings of the above-mentioned prior art and adopts a non-invasive diagnosis method. The method is easy to operate and can quickly and accurately diagnose whether a permanent magnet synchronous motor has a dynamic eccentricity fault and its dynamic eccentricity, providing theoretical support for motor maintenance and repair.

[0011] The technical solution adopted by the present invention to solve the technical problem is:

[0012] A method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on Elman neural network comprises the following steps:

[0013] S1. Establish a characteristic database of dynamic eccentricity fault of permanent magnet synchronous motor:

[0014] 1.1 Establish an electromagnetic simulation model of the dynamic eccentricity fault of the permanent magnet synchronous motor: establish a finite element model with different dynamic eccentricities in the electromagnetic finite element software, and set the dynamic eccentricity to 0, a, 2a, 3a, ..., (n-1)a, where a≤2%, (n-1)a≤1;

[0015] 1.2 Obtain the stray magnetic field time history of the electromagnetic simulation model under different dynamic eccentricity: At a certain speed n r Under the condition, n electromagnetic simulation models with different dynamic eccentricities are calculated respectively to obtain the no-load stray magnetic field time history of each finite element model;

[0016] 1.3 Obtain the amplitude-frequency diagram of the stray magnetic field under different dynamic eccentricity: When the motor has a dynamic eccentricity fault, f c ±f r The left and right sideband harmonics of Figure 3 As shown, the amplitude of the sideband harmonics increases with the increase of dynamic eccentricity;

[0017] 1.4 Establish a permanent magnet synchronous motor dynamic eccentricity fault feature library: Select the fundamental amplitude B of the stray magnetic field of the i-th electromagnetic simulation model i_fc , Left side with harmonic amplitude B i_fc-fr and the right side harmonic amplitude B i_fc+fr is the eigenvalue, where i = 1, 2, 3…n. The dynamic eccentricity fault eigenvalue can be expressed by the following matrix:

[0018]

[0019] The corresponding dynamic eccentricity is expressed as:

[0020]

[0021] S2. Establish a dynamic eccentricity fault diagnosis model for permanent magnet synchronous motor based on Elman network:

[0022] 2.1 Construction of Elman neural network: Construct an Elman neural network with a structure of 3-8-8-1. Figure 6As shown, it mainly includes input layer, hidden layer, connection layer and output layer; the number of input layer nodes is 3, the number of hidden layer nodes is 8, the number of connection layer nodes is 8, and the number of output layer nodes is 1;

[0023] 2.2 Elman neural network training: The fault characteristic value B composed of the fundamental amplitude of the stray magnetic field, the left harmonic amplitude, and the right harmonic amplitude obtained in step 1.4 is n×3 and the corresponding dynamic eccentricity Y n×1 After normalization, the data is input into the neural network model for training. The training function is 'traindx', the maximum number of iterations is 2000, the error tolerance is 0.00001, and multiple iterations are performed to obtain the dynamic eccentricity fault diagnosis model.

[0024] S3. Collect the stray magnetic field of the permanent magnet synchronous motor to be tested and extract the fault characteristic value:

[0025] 3.1 Connect the permanent magnet synchronous motor no-load stray magnetic field acquisition instrument, install the magnetic density sensor, and measure the stray magnetic field on the surface of the casing: The connection method between the permanent magnet synchronous motor to be tested and the permanent magnet synchronous motor no-load stray magnetic field acquisition instrument is as follows: Figure 2 As shown, the magnetic density sensor is installed on the surface of the casing, such as Figure 4 As shown, it is ensured that the collected data are all radial magnetic flux density of stray magnetic field;

[0026] 3.2 Measure the time history of the stray magnetic field and extract the fault characteristics: Drag the motor to be tested back to a fixed speed n r Under the condition of t to t+T, the radial magnetic flux density of the stray magnetic field is measured and recorded, where T is the synchronous electrical period. After the measured experimental data is subjected to fast Fourier transform, the fundamental amplitude B' is extracted. i_fc , left side with harmonic amplitude B' i_fc-fr and the right side harmonic amplitude B' i_fc+fr , forming the fault feature matrix B' 1×3 :

[0027] B' 1×3 =[B' 1_fc B' 1_fc-fr B' 1_fc+fr ]

[0028] S4, input the experimental data of step S3 into the model of step S2 to diagnose the dynamic eccentricity fault of the permanent magnet synchronous motor:

[0029] B' 1×3After normalization, the result is used as the input of the neural network, and the permanent magnet synchronous motor dynamic eccentricity fault diagnosis model based on the Elman neural network established in step S2 is called to calculate the output value of the neural network. The calculation result is denormalized to obtain the dynamic eccentricity output by the diagnosis model.

[0030] Furthermore, in step 3.1, the permanent magnet synchronous motor no-load stray magnetic field collection instrument includes a test bench, a fixture, a servo motor, a torque and speed sensor, a Tesla meter, a data acquisition device, a computer and a motor control cabinet. At least two fixtures are arranged on the test bench, one of which clamps and fixes the servo motor, and the other clamps and fixes the permanent magnet synchronous motor to be tested. The servo motor is electrically connected to the motor control cabinet arranged on one side of the test bench, and the servo motor is connected to the permanent magnet synchronous motor to be tested through a coupling and a torque and speed sensor. The Tesla meter for collecting stray magnetic fields is installed on the stator housing of the permanent magnet synchronous motor to be tested, the Tesla meter is electrically connected to the data acquisition device, and the data acquisition device is electrically connected to the computer.

[0031] The beneficial effects of the present invention are as follows: compared with the prior art, the permanent magnet synchronous motor dynamic eccentricity fault diagnosis method based on Elman neural network of the present invention has the following advantages:

[0032] (1) It can accurately diagnose the presence or absence of dynamic eccentricity fault of permanent magnet synchronous motor and the dynamic eccentricity ratio when dynamic eccentricity fault exists;

[0033] (2) The present invention is a non-intrusive diagnostic method for dynamic eccentricity fault of a permanent magnet synchronous motor, which does not require modification of the motor, is easy to operate, and does not affect the normal operation of the motor;

[0034] (3) The present invention has high versatility and is suitable for all types of permanent magnet synchronous motors. According to different types of motors, it is only necessary to establish a dynamic eccentricity fault diagnosis feature library for the type of motor to diagnose the dynamic eccentricity fault of the motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of dynamic eccentricity;

[0036] Figure 2 is a schematic diagram of a stray magnetic field collection device;

[0037] Figure 3 It is a schematic diagram of fundamental wave and sideband harmonics under 50% dynamic eccentricity;

[0038] Figure 4 This is a schematic diagram of the installation of the magnetic field position sensor;

[0039] Figure 5 It is a flow chart of dynamic eccentricity fault diagnosis;

[0040] Figure 6 It is a schematic diagram of the Elman neural network;

[0041] Figure 7 It is a schematic diagram of the dynamic eccentricity fault diagnosis result;

[0042] in, Figure 2 Among them, 1 motor control cabinet, 2 first fixture, 3 servo motor, 4 coupling, 5 torque speed sensor, 6 second fixture, 7 Tesla meter, 8 permanent magnet synchronous motor to be tested, 9 data acquisition device, 10 computer. DETAILED DESCRIPTION

[0043] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0044] Example 1

[0045] For a certain 8-pole 48-slot permanent magnet synchronous motor, the dynamic eccentricity fault of the motor is detected in the whole process according to the method of the present invention, including the following steps:

[0046] Step S1: Establishing a characteristic database of dynamic eccentricity fault of permanent magnet synchronous motor

[0047] 1.1 Establish an electromagnetic simulation model of dynamic eccentricity fault of permanent magnet synchronous motor:

[0048] Finite element models with different dynamic eccentricities were established in electromagnetic finite element software, and the dynamic eccentricities were set to 0, 0.02, 0.04, 0.06, ... 0.98 in sequence, with a total of 50 sets of data;

[0049] 1.2 Obtain the stray magnetic field time history of the electromagnetic simulation model under different dynamic eccentricities:

[0050] At a rotation speed of 5 rpm, 50 electromagnetic simulation models with different dynamic eccentricities were calculated to obtain the no-load stray magnetic field time history of each finite element model.

[0051] 1.3 Perform fast Fourier decomposition on the stray magnetic field of each simulation model to obtain the fundamental amplitude B of the stray magnetic field under different dynamic eccentricities i_fc , Left side with harmonic amplitude B i_fc-fr and the right side harmonic amplitude B i_fc+fr , where i = 1, 2, 3, ... 50. Construct the dynamic eccentricity fault eigenvalue matrix B 50×3 for:

[0052]

[0053] The corresponding dynamic eccentricity is expressed as:

[0054]

[0055] Step S2: Establish a dynamic eccentricity fault diagnosis model for permanent magnet synchronous motor based on Elman network

[0056] Structure Figure 6 As shown in Figure 1, the model mainly includes input layer, hidden layer, connection layer and output layer. The input layer has 3 nodes, the number of nodes in the hidden layer and the connection layer is 8, and the number of nodes in the output layer is 1. The fault characteristic value B composed of the fundamental amplitude of the stray magnetic field, the harmonic amplitude of the left band and the harmonic amplitude of the right band obtained by simulation is 50×3 and the normalized dynamic eccentricity Y 50×1 , respectively as input and output, substituted into the neural network model for training, the training function is 'traindx', the maximum number of iterations is 2000 times, the error tolerance is 0.00001, and the dynamic eccentricity fault diagnosis model is obtained;

[0057] Step S3: Collect the stray magnetic field of the permanent magnet synchronous motor to be tested and extract the fault characteristic value

[0058] 3.1 Connect the data acquisition instrument, install the magnetic density sensor, and measure the stray magnetic field on the surface of the casing:

[0059] Permanent magnet synchronous motor no-load stray magnetic field acquisition device Figure 2 As shown, the servo motor 3 is fixed on the test bench 11 through the first fixture 2, the servo motor 3 is controlled by the motor control cabinet 1, and the left end is connected to the permanent magnet synchronous motor 8 to be tested through the coupling 4 and the torque speed sensor 5. The permanent magnet synchronous motor is fixed on the test bench through the second fixture 6, and the Tesla meter 7 for collecting stray magnetic fields is installed on the stator housing. The stray magnetic field is collected by the data acquisition device 9 and then input into the computer 10. The computer 10 is used to record and analyze the collected experimental data;

[0060] 3.2 Press Figure 5 As shown in the flow chart, the following process is performed for the dynamic eccentricity fault diagnosis of the permanent magnet synchronous motor to be tested:

[0061] A magnetic field sensor is installed on the stator housing of the permanent magnet synchronous motor to collect stray magnetic fields; Figure 2 The servo motor in the test reversely drags the permanent magnet synchronous motor to a speed of 5rpm, and collects the time history of the stray magnetic field within a synchronous electrical cycle; the measured experimental data is processed by fast Fourier transform to extract the fundamental wave amplitude B' i_fc , left side with harmonic amplitude B' i_fc-fr and the right side harmonic amplitude B' i_fc+fr , forming the matrix B' 1×3 :

[0062] B'1×3 =[B' 1_fc B' 1_fc-fr B' 1_fc+fr ]

[0063] Step S4: Input the experimental data into the model established in step S2 to diagnose the dynamic eccentricity fault of the permanent magnet synchronous motor:

[0064] B' 1×3 After normalization, the result is used as the input of the neural network, and the permanent magnet synchronous motor dynamic eccentricity fault diagnosis model based on the Elman neural network established in step S2 is called to calculate the output value of the neural network, and the dynamic eccentricity is obtained by denormalizing the calculation result.

[0065] Figure 7 The figure is a schematic diagram of the dynamic eccentricity fault diagnosis results. There are 40 training samples and 10 test samples. It can be seen from the figure that the dynamic eccentricity prediction value of the fault diagnosis model is consistent with the actual value, the fault diagnosis accuracy is 98.18%, and the mean square error of the test data is 0.000144.

[0066] The above-mentioned specific implementations are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the product form and style of the above-mentioned specific implementations. Any appropriate changes or modifications made to them by ordinary technicians in the technical field that conform to the claims of the present invention shall fall within the patent protection scope of the present invention.

Claims

1. A method for diagnosing dynamic eccentricity fault of permanent magnet synchronous motor based on Elman neural network, characterized in that: The steps include: S1. Establish a characteristic database of dynamic eccentricity fault of permanent magnet synchronous motor: 1.1 Establish an electromagnetic simulation model of the dynamic eccentricity fault of the permanent magnet synchronous motor, establish a finite element model with different dynamic eccentricities in the electromagnetic finite element software, and set the dynamic eccentricity to 0, a, 2a, 3a, ..., (n-1)a, where a≤2%, (n-1)a≤1; 1.2 Obtain the stray magnetic field time history of the electromagnetic simulation model under different dynamic eccentricities, at a certain speed n r Under the condition, n electromagnetic simulation models with different dynamic eccentricities are calculated respectively to obtain the no-load stray magnetic field time history of each finite element model; 1.3 Obtain the amplitude-frequency diagram of the stray magnetic field under different dynamic eccentricity. When the motor has a dynamic eccentricity fault, f c ±f r The left and right sideband harmonics of the dynamic eccentricity increase, and the amplitude of the sideband harmonics also increases with the increase of dynamic eccentricity; 1.4 Establish a dynamic eccentricity fault feature library for permanent magnet synchronous motors and select the fundamental amplitude B of the stray magnetic field of the i-th electromagnetic simulation model i_fc , Left side with harmonic amplitude B i_fc-fr and the right side harmonic amplitude B i_fc+fr is the eigenvalue, where i = 1, 2, 3…n. The dynamic eccentricity fault eigenvalue can be expressed by the following matrix: The corresponding dynamic eccentricity is expressed as: S2. Establish a dynamic eccentricity fault diagnosis model for permanent magnet synchronous motor based on Elman network; 2.1 Construction of Elman neural network: Construct an Elman neural network with a structure of 3-8-8-1, which mainly includes input layer, hidden layer, connection layer and output layer; the number of input layer nodes is 3, the number of hidden layer nodes is 8, the number of connection layer nodes is 8, and the number of output layer nodes is 1; 2.2 Elman neural network training: The fault characteristic value B composed of the fundamental amplitude of the stray magnetic field, the left harmonic amplitude, and the right harmonic amplitude obtained in step 1.4 is n×3 and the corresponding dynamic eccentricity Y n×1 After normalization, the data is input into the neural network model for training. The training function is 'traindx', the maximum number of iterations is 2000, the error tolerance is 0.00001, and multiple iterations are performed to obtain the dynamic eccentricity fault diagnosis model. S3. Connect the permanent magnet synchronous motor no-load stray magnetic field collection instrument, install a magnetic density sensor on the surface of the casing, measure the stray magnetic field on the surface of the casing, and extract the fault characteristic values, namely the fundamental wave amplitude, the left-band harmonic amplitude, and the right-band harmonic amplitude; S4. Input the experimental data of step S3 into the model of step S2, diagnose the dynamic eccentricity fault of the permanent magnet synchronous motor, and obtain the dynamic eccentricity output by the diagnosis model.

2. The method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on Elman neural network according to claim 1 is characterized in that: The step S3 comprises the following steps: 3.1 Connect the permanent magnet synchronous motor no-load stray magnetic field collection instrument, install the magnetic density sensor, and measure the stray magnetic field on the surface of the casing: the magnetic density sensor is installed on the surface of the casing, and ensure that the collected data are all radial magnetic density of the stray magnetic field; 3.2 Measure the time history of the stray magnetic field and extract the fault characteristics: Drag the motor to be tested back to a fixed speed n r Under the condition of t to t+T, the radial magnetic flux density of the stray magnetic field is measured and recorded, where T is the synchronous electrical period. After the measured experimental data is subjected to fast Fourier transform, the fundamental amplitude B' is extracted. i_fc , left side with harmonic amplitude B' i_fc-fr and the right side harmonic amplitude B' i_fc+fr , forming the fault feature matrix B' 1×3 : B' 1×3 =[B' 1_fc B' 1_fc-fr B' 1_fc+fr ]。 3. The method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on Elman neural network according to claim 2 is characterized in that: The step S4 comprises the following steps: 1×3 After normalization, the result is used as the input of the neural network, and the permanent magnet synchronous motor dynamic eccentricity fault diagnosis model based on the Elman neural network established in step S2 is called to calculate the output value of the neural network. The calculation result is denormalized to obtain the dynamic eccentricity output by the diagnosis model.

4. The method for diagnosing dynamic eccentricity fault of a permanent magnet synchronous motor based on Elman neural network according to claim 2 is characterized in that: The permanent magnet synchronous motor no-load stray magnetic field collection instrument includes a test bench, a fixture, a servo motor, a torque and speed sensor, a Tesla meter, a data acquisition device, a computer and a motor control cabinet. At least two fixtures are arranged on the test bench, one of which clamps and fixes the servo motor, and the other clamps and fixes the permanent magnet synchronous motor to be tested. The servo motor is electrically connected to the motor control cabinet arranged on one side of the test bench. The servo motor is connected to the permanent magnet synchronous motor to be tested through a coupling and a torque and speed sensor. The Tesla meter for collecting stray magnetic fields is installed on the stator housing of the permanent magnet synchronous motor to be tested, the Tesla meter is electrically connected to the data acquisition device, and the data acquisition device is electrically connected to the computer.

Images