A method and system for characterizing motor rotor eccentricity

Abstract

The present invention discloses a method and system for characterizing motor rotor eccentricity, which belongs to the field of quantitative characterization of eccentric faults in rotating machinery. In the data acquisition stage, a mathematical model of the motor rotor eccentricity is first established using known physical information to obtain a model data set consisting of motor data generated by the mathematical model and corresponding rotor eccentricity information; then, the actual data set is sampled from the actual motor. In the training stage, the generator is first pre-trained using the model data set; then, adversarial training is performed on the generator and the discriminator in combination with the actual data set to achieve correction of the generator; finally, the corrected generator is used to generate motor data for training the characterizer. Thus, the problem of difficulty in training the network in the absence of data with different eccentricities is solved, and the motor data generated by the trained generator is close to the actual data, so that high-precision eccentricity characterization can be obtained.

Description

Technical Field

[0001] The present invention belongs to the field of quantitative characterization of rotating machinery eccentricity faults, and more specifically, relates to a method and system for characterizing motor rotor eccentricity. Background Art

[0002] With the development of the new energy industry, permanent magnet synchronous motors (PMSMs) are widely used in various fields, including renewable energy power generation, electric vehicles, transportation, and propulsion systems. Therefore, to ensure the high reliability of PMSM operation, timely maintenance and fault diagnosis are essential. Due to manufacturing tolerances, nearly all motors experience rotor eccentricity. This eccentricity generates harmonic components in the motor's current and voltage, which in turn lead to increased torque ripple and additional losses. Furthermore, eccentricity disrupts the motor's balance, causing unbalanced magnetic pull within the motor. This magnetic pull can exacerbate eccentricity faults and, in severe cases, lead to motor scavenging and damage. Therefore, to mitigate the negative effects of eccentricity, accurate prediction of the motor's eccentricity is necessary.

[0003] The degree of motor eccentricity is difficult to measure directly, so it is necessary to predict the degree of motor eccentricity with the help of relevant physical quantities in the motor. However, due to the nonlinearity and high complexity of the motor system, it is very difficult to obtain an analytical expression for the motor's eccentricity through motor modeling. In the current mechanical fault diagnosis, neural networks play a very important role. However, as a data-driven model, neural networks require a large amount of labeled data to train the neural network. For the problem of predicting the degree of motor eccentricity, the eccentricity of the actual motor is difficult to obtain directly, and the motors are all running under normal conditions, and there is a lack of data under eccentric fault conditions. Therefore, there is a lack of eccentricity labels to train the neural network, and it is impossible to effectively predict the degree of motor eccentricity. Summary of the Invention

[0004] In response to the defects of the existing technology and the need for improvement, the present invention provides a method and system for characterizing motor rotor eccentricity, aiming to solve the technical problem that it is difficult to characterize the motor rotor eccentricity in the absence of data on different motor eccentricities.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for characterizing motor rotor eccentricity, comprising:

[0006] Establish a mathematical model for the motor rotor eccentricity to obtain the motor data X1 and the corresponding rotor eccentricity information Y1 generated by the mathematical model; sample the data of the actual motor to obtain the motor data X1 under normal conditions. 2h ;

[0007] Construct a neural network model including a generator, a discriminator, and a characterizer; wherein the generator inputs rotor eccentricity information and outputs motor data; the discriminator inputs motor data and outputs a quantity that characterizes whether the input motor data comes from a mathematical model or an actual motor; the characterizer inputs motor data and outputs an eccentricity characterization quantity;

[0008] The generator is pre-trained using the rotor eccentricity information Y1 and motor data X1; then the generator is pre-trained using the rotor eccentricity information Y 1h and motor data X 2h , conduct adversarial training on the discriminator and the pre-trained generator, where Y 1h ∈Y1 represents the rotor eccentricity information of the motor under normal conditions; finally, the rotor eccentricity information Y1 and the motor data generated by the generator after adversarial training are used to train the characterizer, and the trained characterizer is used to characterize the rotor eccentricity.

[0009] Furthermore, the generator is pre-trained using the rotor eccentricity information Y1 and the motor data X1, including:

[0010] Taking the rotor eccentricity information Y1 as input, fitting the motor data X1, and pre-training the generator with the goal of minimizing the generator's generation loss loss1.

[0011] Furthermore, the rotor eccentricity information Y is used 1h and motor data X 2h , conduct adversarial training on the discriminator and the pre-trained generator, including:

[0012] Fixed the parameters of the generator and the rotor eccentricity information Y 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h Input the discriminator and update the discriminator parameters with the goal of minimizing the discriminant loss loss2;

[0013] Fixed the discriminator parameters and the rotor eccentricity information Y 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h Input into the discriminator, and update the parameters of the generator with the goal of minimizing the comprehensive loss loss3; the comprehensive loss loss3 = λ1·loss1-λ2·loss2, where loss1 represents the generation loss of the generator, loss2 represents the discrimination loss of the discriminator, and λ1 and λ2 are positive weight coefficients;

[0014] Repeat the above steps until the training iteration threshold is reached and the adversarial training is completed.

[0015] Furthermore, the method of training the characterizer using the rotor eccentricity information Y1 and the motor data generated by the generator after adversarial training includes:

[0016] Fix the parameters of the generator, input the rotor eccentricity information Y1 into the generator, and generate multiple motor data; then input the generated multiple motor data into the characterizer, fit the rotor eccentricity information Y1, and update the parameters of the characterizer with the goal of minimizing the characterization loss loss4 of the characterizer.

[0017] Furthermore, the generator, discriminator and characterizer are all composed of fully connected networks.

[0018] Furthermore, the motor data is the no-load induced electromotive force, and the mathematical model is:

[0019]

[0020] Where B(·) is the air gap flux density, θ w+ and θ w- are the spatial positions of the positive and negative poles of the winding, N is the number of winding turns, l is the depth of the motor, ω r is the rotor mechanical angular velocity, R is the stator inner diameter, h is the permanent magnet height, δ0 is the air gap length of the motor under normal conditions, B0 is the air gap flux amplitude under normal conditions, ɑ is the rotor pole position, p is the number of rotor pole pairs, is the eccentricity, O s and O r are the geometric centers of the stator and rotor, θ e is the eccentric angle.

[0021] In a second aspect, the present invention provides a motor rotor eccentricity characterization system, comprising: a computer-readable storage medium and a processor;

[0022] The computer-readable storage medium is used to store executable instructions;

[0023] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the motor rotor eccentricity characterization method as described in the first aspect.

[0024] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0025] During the data acquisition phase, the present invention first uses known physical information to establish a mathematical model for motor rotor eccentricity, obtaining a model dataset consisting of motor data generated by the mathematical model and corresponding rotor eccentricity information. The actual motor data is then sampled to obtain the actual dataset. During the training phase, the model dataset is first used to pre-train the generator. The generator and discriminator are then trained adversarially using the model dataset and the actual dataset to correct the generator. Finally, the corrected generator is used to generate motor data for training the characterizer. This solves the problem of network training difficulties in the absence of data with varying eccentricities. The motor data generated by the trained generator is close to the actual data, enabling high-precision eccentricity characterization. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flow chart of a method for characterizing motor rotor eccentricity provided by an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of the structure of a neural network provided by an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of the neural network training process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0030] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0031] Example 1

[0032] See Figure 1 , combined with Figure 2 and Figure 3 The present invention provides a method for characterizing motor rotor eccentricity. The method models the motor and uses the model data and actual data sampled from the motor under normal conditions to perform adversarial training on a neural network. This method generates a large amount of data for training using the neural network, thereby achieving quantitative characterization of the motor rotor eccentricity. The method includes operations S1 to S3.

[0033] Operation S1: Establish a mathematical model under the condition of motor rotor eccentricity to obtain motor data X1 and corresponding rotor eccentricity information Y1 generated by the mathematical model; sample the data of the actual motor to obtain the motor data X1 under normal conditions. 2h .

[0034] Operation S1 includes sub-operation S11 and sub-operation S12.

[0035] In sub-operation S11, firstly, a mathematical model of the motor rotor eccentricity is established using known physical information to obtain motor data generated by the mathematical model. Corresponding rotor eccentricity information Constructing a labeled model dataset where X 1h ∈X1 represents the data of the motor under normal conditions in the generated data, Y 1h ∈Y1 represents the rotor eccentricity information of the motor under normal conditions. Generally, the normal condition of the motor means that the eccentricity is less than 10%.

[0036] It should be noted that the rotor eccentricity information includes the eccentricity angle and eccentricity; the motor data refers to the physical quantity that can be directly or indirectly obtained during the operation of the motor and is related to the rotor eccentricity. This embodiment takes the no-load induced electromotive force as the motor data as an example to illustrate the process of constructing the motor mathematical model.

[0037] When the rotor is eccentric, the air gap length at θ on the motor circumference is:

[0038]

[0039] Where R is the inner diameter of the stator, r is the outer diameter of the rotor, θ e is the eccentric angle, O s and O r are the geometric centers of the stator and rotor respectively, δ0 is the air gap length in the absence of eccentricity, is the eccentricity.

[0040] When the rotor is eccentric, the air gap magnetic flux density at θ on the circumference of the motor is:

[0041]

[0042] Where h is the height of the permanent magnet, B0 is the air gap flux amplitude under normal conditions, p is the number of rotor pole pairs, and ɑ is the rotor pole position.

[0043] According to the law of electromagnetic induction and the direction of the induced electromotive force, the no-load induced electromotive force in the winding is:

[0044]

[0045] Where N is the number of winding turns, l is the motor depth, ω r is the rotor mechanical angular velocity, θ w+ and θ w- are the spatial positions of the positive and negative poles of the winding respectively.

[0046] In sub-operation S12, the sensor is used to sample the actual motor data to obtain the motor data under normal conditions. Accordingly, if the no-load induced electromotive force data obtained by the mathematical model in sub-operation S11 is no-load induced electromotive force data, then in sub-operation S12, the no-load induced electromotive force data of the motor under normal conditions is obtained through actual sampling.

[0047] In addition to sampling data from actual motors, finite element simulation data can also be used as the actual data set.

[0048] Operation S2 constructs a neural network model including a generator, a discriminator and a characterizer; wherein the input of the generator is rotor eccentricity information, and the output is motor data; the input of the discriminator is motor data, and the output is a quantity that characterizes the input motor data from a mathematical model or an actual motor; the input of the characterizer is motor data, and the output is a rotor eccentricity characterization quantity.

[0049] like Figure 2 As shown in the figure, the neural network consists of three parts. The generator (network 1), discriminator (network 2), and characterizer (network 3) are all composed of fully connected networks, and network 1 is fully connected with network 2 and network 3 respectively.

[0050] Operation S3, using the rotor eccentricity information Y1 and the motor data X1, pre-training the generator; then using the rotor eccentricity information Y 1h and motor data X 2h , conduct adversarial training on the discriminator and the pre-trained generator, where Y 1h ∈Y1 represents the rotor eccentricity information of the motor under normal conditions; finally, the rotor eccentricity information Y1 and the generator after adversarial training are used to train the characterizer, and the trained characterizer is used to characterize the rotor eccentricity.

[0051] like Figure 3 As shown, operation S3 includes sub-operations S31 to S34.

[0052] In sub-operation S31, the rotor eccentricity information Y1 is used as input to fit the motor data X1, and the generator is pre-trained with the goal of minimizing the generator's generation loss loss1. The pre-trained generator can output corresponding motor data based on the input rotor eccentricity information.

[0053] in, For network 1 pair The output value of .

[0054] In sub-operation S32, the parameters of the generator are fixed and the rotor eccentricity information Y is converted to 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h Input into the discriminator, and update the discriminator parameters with the goal of minimizing the discriminator loss2. Labels during training m is the number of motor data generated.

[0055] in, is the output of network 2.

[0056] In sub-operation S33, the parameters of the discriminator are fixed, and the rotor eccentricity information Y 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h The input is input into the discriminator, and the parameters of the generator are updated with the goal of minimizing the comprehensive loss loss3; the comprehensive loss loss3 = λ1·loss1-λ2·loss2, where loss1 represents the generation loss of the generator, loss2 represents the discrimination loss of the discriminator, and λ1 and λ2 are positive weight coefficients.

[0057] Repeat sub-operations S32 and S33 until the training iteration number threshold is reached, and the adversarial training is completed.

[0058] In sub-operation S34, the parameters of the generator are fixed, the rotor eccentricity information Y1 is input into the generator, and a plurality of motor data are generated; the generated plurality of motor data are then input into the characterizer, the rotor eccentricity information Y1 is fitted, and the parameters of the characterizer are updated with the goal of minimizing the characterization loss loss4 of the characterizer.

[0059] in, is the output value of the characterizer.

[0060] In order to better illustrate the present invention, the following examples are given:

[0061] This example uses Ansys finite element simulation to illustrate a specific implementation. The established mathematical model data is used as model data X1, and the finite element simulation data is used as actual data X2. The actual data only includes the back EMF data under normal motor conditions. For quantitative characterization of static eccentricity, the input to the characterizer and the output data from the generator are the back EMF amplitudes of each phase winding.

[0062] The specific implementation of this example is as follows:

[0063] Step 1: Obtain the motor back EMF data of different eccentricities from the model and simulation respectively;

[0064] Step 2: Construct a neural network. All networks are fully connected neural networks. The number of neurons in each layer of network 1 is 20, 20, and 12, and the activation function is leak_relu function. The number of neurons in each layer of network 2 is 10 and 1, and the activation functions of the hidden layer and output layer are leak_relu function and sigmoid function respectively. The number of neurons in each layer of network 3 is 20, 20, 20, and 1, and the activation function of the hidden layer is leak_relu function, and the activation function of the output layer is sigmoid function.

[0065] Step 3: Use the model data to train Network 1. Input the rotor eccentricity information in the model into the generator. The mean square error function between the generator output and the model back EMF is used as the loss function loss1. The Adam algorithm is used to optimize Network 1 to obtain a pre-trained generator.

[0066] Step 4: Fix the parameters of network 1, input the generated data and simulated data into network 2, and train network 2 to distinguish the generated data from the simulated data by minimizing loss2;

[0067] Step 5: Fix the parameters of network 2, input the generated data and simulated data into network 2, and train network 1 by minimizing loss 3;

[0068] Step 6: Repeat steps 4 and 5 until the number of training iterations is reached, stop training, and obtain the trained generator;

[0069] In step 7, the generated data and the corresponding eccentricity information are input into the characterizer for fitting. The trained neural network can realize eccentricity characterization.

[0070] Example 2

[0071] A motor rotor eccentricity characterization system includes: a computer-readable storage medium and a processor; the computer-readable storage medium is used to store executable instructions; the processor is used to read the executable instructions stored in the computer-readable storage medium and execute the motor rotor eccentricity characterization method as described in Example 1.

[0072] The relevant technical solutions are the same as those in Example 1 and will not be described again here.

[0073] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for characterizing motor rotor eccentricity, characterized in that: include: Establish a mathematical model for the motor rotor eccentricity to obtain the motor data X1 and the corresponding rotor eccentricity information Y1 generated by the mathematical model; sample the data of the actual motor to obtain the motor data X1 under normal conditions. 2h ; Construct a neural network model including a generator, a discriminator, and a characterizer; wherein the generator inputs rotor eccentricity information and outputs motor data; the discriminator inputs motor data and outputs a quantity that characterizes whether the input motor data comes from a mathematical model or an actual motor; the characterizer inputs motor data and outputs an eccentricity characterization quantity; The generator is pre-trained using the rotor eccentricity information Y1 and motor data X1; then the generator is pre-trained using the rotor eccentricity information Y 1h and motor data X 2h , conduct adversarial training on the discriminator and the pre-trained generator, where Y 1h ∈Y1 represents the rotor eccentricity information of the motor under normal conditions; finally, the rotor eccentricity information Y1 and the motor data generated by the generator after adversarial training are used to train the characterizer, and the trained characterizer is used to characterize the rotor eccentricity; Wherein, the motor data is the no-load induced electromotive force, and the mathematical model is: Where B(·) is the air gap flux density, θ w+ and θ w- are the spatial positions of the positive and negative poles of the winding, N is the number of winding turns, l is the depth of the motor, ω r is the rotor mechanical angular velocity, R is the stator inner diameter, h is the permanent magnet height, δ0 is the air gap length of the motor under normal conditions, B0 is the air gap flux amplitude under normal conditions, ɑ is the rotor pole position, p is the number of rotor pole pairs, is the eccentricity, O s and O r are the geometric centers of the stator and rotor, θ e is the eccentric angle.

2. The motor rotor eccentricity characterization method according to claim 1, characterized in that: The generator is pre-trained using the rotor eccentricity information Y1 and the motor data X1, including: Taking the rotor eccentricity information Y1 as input, fitting the motor data X1, and pre-training the generator with the goal of minimizing the generator's generation loss loss1.

3. The motor rotor eccentricity characterization method according to claim 1, characterized in that: The utilization of the rotor eccentricity information Y 1h and motor data X 2h , conduct adversarial training on the discriminator and the pre-trained generator, including: Fixed the parameters of the generator and the rotor eccentricity information Y 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h Input the discriminator and update the discriminator parameters with the goal of minimizing the discriminant loss loss2; Fixed the discriminator parameters and the rotor eccentricity information Y 1h Input the generator to generate multiple motor data, and then combine the generated multiple motor data with the motor data X 2h Input into the discriminator, and update the parameters of the generator with the goal of minimizing the comprehensive loss loss3; the comprehensive loss loss3 = λ1·loss1-λ2·loss2, where loss1 represents the generation loss of the generator, loss2 represents the discrimination loss of the discriminator, and λ1 and λ2 are positive weight coefficients; Repeat the above steps until the training iteration threshold is reached and the adversarial training is completed.

4. The motor rotor eccentricity characterization method according to claim 1, characterized in that: The method of training the characterizer using the rotor eccentricity information Y1 and the motor data generated by the generator after adversarial training includes: Fix the parameters of the generator, input the rotor eccentricity information Y1 into the generator, and generate multiple motor data; then input the generated multiple motor data into the characterizer, fit the rotor eccentricity information Y1, and update the parameters of the characterizer with the goal of minimizing the characterization loss loss4 of the characterizer.

5. The motor rotor eccentricity characterization method according to claim 1, characterized in that: The generator, discriminator and characterizer are all composed of fully connected networks.

6. A motor rotor eccentricity characterization system, characterized in that: include: Computer-readable storage medium and processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the motor rotor eccentricity characterization method according to any one of claims 1 to 5.

Images