A method for predicting rotor displacement of a magnetic levitation bearing, a self-sensing system and a medium

Abstract

This invention discloses a method, self-sensing system, and medium for predicting rotor displacement of a magnetic levitation bearing, belonging to the interdisciplinary field of artificial intelligence and magnetic levitation systems. The method includes: a training phase: collecting waveform data of the winding current and actual rotor displacement of the magnetic levitation bearing under different operating conditions; using the winding current values ​​and operating condition categories at times (t-1), (t-2), ..., (t-n) as input and the rotor displacement value at time t as output, training a rotor displacement prediction model; where n is an integer greater than or equal to 2; and an application phase: acquiring the winding current values ​​and operating condition categories from the previous n times in real time, inputting them into the trained rotor displacement prediction model, and predicting the rotor displacement value at the current time. This invention can predict rotor displacement through real-time winding current information of the magnetic bearing, thereby enabling fault response to displacement sensors in magnetic levitation systems, reducing dependence on displacement sensor hardware, and even replacing displacement sensors, thus reducing costs.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of artificial intelligence and magnetic levitation systems, and more specifically, relates to a method for predicting the displacement of a magnetic levitation bearing rotor, a self-sensing system, and a medium. Background Technology

[0002] Magnetic bearing (magnetic bearing) technology uses magnetic force to suspend the rotor in the center of the bearing, fundamentally changing the support method of traditional mechanical bearings. Magnetic bearings allow the stator and rotor to not contact each other, require low power consumption, have a long service life, and do not require lubrication, making them one of the best solutions for supporting high-speed rotating machinery.

[0003] In the control system of an active magnetic bearing, rotor displacement is a crucial control variable. By comparing the feedback rotor displacement with the displacement command, the current in the winding coils is adjusted, thereby controlling the electromagnetic force on the rotor and allowing for real-time adjustment of the rotor's position and offset. In traditional magnetic bearing systems, displacement sensors are essential components for measuring rotor offset and achieving stable levitation of the active magnetic bearing.

[0004] However, displacement sensors are too large to be installed inside the bearing, causing discrepancies between the measured displacement and the rotor displacement, and also hindering the integrated design of the motor and magnetic bearing. Furthermore, displacement sensors are typically expensive, impeding the widespread adoption and industrialization of electromagnetic bearings. Additionally, displacement sensors are prone to failure, reducing the reliability of magnetic levitation systems. Therefore, research on self-sensing electromagnetic bearings or sensorless electromagnetic bearings has received increasing attention in recent years.

[0005] Self-sensing methods allow magnetic bearings to eliminate the need for conventional displacement sensors. The original function is replaced by some form of signal processing, specifically extracting rotor position information from the voltage and current waveforms of the magnetic bearing's electromagnetic coil. There are generally two methods for achieving self-sensing operation of electromagnetic bearings: one is the state observation method based on modern control theory, and the other is the parameter estimation method based on inductance detection.

[0006] As early as the 1990s, a method for estimating rotor position information based on state observation theory was proposed by foreign scholars, and stable levitation of the system was achieved. However, the drawbacks of this method have been discussed in many papers; it is easily affected by changes in system parameters and has poor robustness. The parameter estimation method based on inductance detection utilizes the functional relationship between the equivalent inductance of the magnetic bearing winding and the stator-rotor air gap. By detecting the equivalent inductance parameter of the winding, the air gap length in that direction can be estimated. However, this method leads to the conclusion that the self-sensing system tends to work well when there is a large amount of high-frequency ripple in the coil current. This contradicts the pursuit of high frequency, high efficiency, and low noise in modern power electronic control technology, making the method based on switching ripple only achievable at the expense of efficiency. Summary of the Invention

[0007] To address the shortcomings and improvement needs of existing technologies, this invention provides a method for predicting the displacement of a magnetic levitation bearing rotor, a self-sensing system, and a medium. The aim is to replace traditional displacement sensors, reduce costs, and provide a model with high prediction accuracy, generalization ability, robustness, minimal impact on system control performance, and a simple model that is easy to calculate in real time.

[0008] To achieve the above objectives, according to a first aspect of the present invention, a method for predicting the displacement of a magnetic levitation bearing rotor is provided, comprising:

[0009] Training phase:

[0010] Collect waveform data of magnetic levitation bearing winding current and actual rotor displacement under different operating conditions;

[0011] The rotor displacement prediction model is trained by taking the winding current values ​​and operating condition categories at times (t-1), (t-2), ..., (tn) as inputs and the rotor displacement value at time t as output; where n is an integer greater than or equal to 2.

[0012] Application phase:

[0013] Obtain the winding current values ​​and operating condition categories for the previous n time steps, input them into the trained rotor displacement prediction model, and predict the rotor displacement value at the current time step.

[0014] Furthermore, the operating condition category is characterized by the motor speed.

[0015] Furthermore, the operating condition categories include: suspension operating condition, rotation operating condition, and external disturbance operating condition;

[0016] If the motor speed is greater than the preset threshold, the operating condition is a rotating condition; otherwise, the operating condition is a suspended condition or an external disturbance condition.

[0017] Furthermore, the levitation condition is the process in which the rotor is levitated on the protective bearing by the electromagnetic force of the magnetic bearing stator. This process can be achieved by changing the PID parameters of the controller to obtain different levitation conditions.

[0018] Furthermore, the rotational operating condition refers to the process where the motor starts to rotate after the rotor is suspended in the air. This process can be modified by changing the speed of the motor to obtain different rotational operating conditions.

[0019] Furthermore, the external disturbance condition refers to random disturbances to the magnetic bearing system from the outside after the rotor is suspended. These disturbances can be obtained by striking the magnetic levitation bearing body with a wooden mallet at random angles and with random force.

[0020] Furthermore, the rotor displacement prediction model is a rotor displacement prediction model based on an artificial neural network;

[0021] The artificial neural network includes N1 hidden layers, each hidden layer including N2 nodes; the activation function is Sigmoid, Tanh or ReLU function, and the loss function is mean squared error loss function.

[0022] Furthermore, after collecting waveform data of the winding current of the magnetic levitation bearing under different operating conditions, the method further includes: performing digital filtering and normalization processing on the waveform data of the winding current, and then inputting it into the rotor displacement prediction model.

[0023] According to a second aspect of the present invention, a magnetic levitation bearing rotor displacement self-sensing system is provided, comprising: a magnetic levitation bearing body, a controller, and a power amplifier;

[0024] The controller uses the magnetic levitation bearing rotor displacement prediction method as described in the first aspect to predict the rotor displacement value at the current moment; and generates a corresponding current command based on the difference between the rotor displacement reference value and the predicted value.

[0025] The power amplifier outputs a corresponding current value according to the current command and acts on the magnetic levitation bearing body to control the rotor displacement of the magnetic levitation bearing.

[0026] According to a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the magnetic levitation bearing rotor displacement prediction method as described in the first aspect.

[0027] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0028] (1) This invention derives the mathematical relationship between the winding current and rotor displacement of a magnetic levitation bearing, discovering that the rotor displacement at the current moment is related to the winding current at historical moments, thereby guiding the selection of input and output data dimensions. Simultaneously, a new dimension is introduced to distinguish different operating conditions during the magnetic bearing's operation, enabling the model to autonomously differentiate between different operating conditions based on different inputs, resulting in more accurate predictions. Thus, this invention can predict rotor displacement information relatively accurately by extracting winding current information, eliminating the hardware dependence on displacement sensors, reducing the cost of the magnetic bearing system, facilitating a reduction in the size of the magnetic bearing system, and promoting the integration of the magnetic bearing with the motor.

[0029] (2) This invention characterizes the working condition category by motor speed and provides three different working conditions: suspension working condition, rotation working condition and external disturbance working condition. The operation is simple and reliable.

[0030] (3) Traditional self-sensing technology requires high model accuracy. In contrast, this invention uses machine learning to construct an artificial neural network model and determines the unknown parameters in the model through a data-driven approach. This eliminates the need to build a complex and accurate mathematical model for the system and also has a certain generalization ability, enabling it to respond quickly and accurately to unknown operating conditions. Furthermore, the algorithm proposed in this invention has a simple structure and can obtain the weight parameters in the model through offline training, greatly reducing the amount of online computation and facilitating real-time calculation and control. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of a magnetic levitation bearing rotor displacement prediction method provided in an embodiment of the present invention.

[0032] Figure 2 A block diagram of a magnetic bearing control system with a displacement sensor provided in an embodiment of the present invention.

[0033] Figure 3 This is a schematic diagram of a neural network multidimensional input provided in an embodiment of the present invention.

[0034] Figure 4 A comparison is made between the neural network prediction results and the displacement sensor measurement results for the magnetic bearing suspension condition provided in the embodiments of the present invention.

[0035] Figure 5 A comparison is made between the neural network prediction results of the magnetic bearing rotation condition provided in the embodiments of the present invention and the measurement results of the displacement sensor.

[0036] Figure 6 The neural network prediction results of the magnetic bearing under external disturbance conditions provided in the embodiments of the present invention are compared with the displacement sensor measurement results. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be 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 illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0038] See Figure 1 This embodiment provides a method for predicting the displacement of a magnetic levitation bearing rotor, including a training phase and an application phase. The training phase includes operation S1 and operation S2.

[0039] Operate S1 to collect waveform data of the magnetic levitation bearing winding current and the actual rotor displacement under different operating conditions.

[0040] Data collection and acquisition are often the first steps in solving problems using machine learning methods. To determine parameters such as weights in the network model, a large amount of input and output data needs to be collected to train the network. For the specific application of a magnetic bearing system, different operating conditions of the system should be considered to increase the richness of the data samples and improve the network's generalization ability. In this embodiment, the operating conditions of a magnetic bearing system are divided into three categories: suspension condition, rotation condition, and external disturbance condition.

[0041] The levitation condition refers to the process where the rotor is levitated on the protective bearing by the electromagnetic force of the stator of the magnetic bearing. To ensure the richness of the sample data for this condition, different levitation dynamic responses can be obtained by changing the PID parameters of the controller. The rotation condition refers to the process where the motor starts to rotate after the rotor is levitated. Different rotation conditions can be obtained by changing the motor speed. The external disturbance condition refers to random disturbances to the magnetic bearing system from the outside after the rotor is levitated. This disturbance can be obtained by striking the magnetic levitation bearing body with a wooden mallet at random angles and with random force.

[0042] In this embodiment, the different operating conditions and data acquisition described above occur in a real magnetic bearing system. The operating conditions are constructed and data is collected through offline experiments. Specifically, the winding current of the magnetic levitation bearing can be acquired using a current sensor, and the actual displacement can be acquired using a displacement sensor; this displacement is considered the "desired" displacement. Thus, a current-displacement data sample set can be constructed, facilitating the implementation of subsequent solutions.

[0043] Operation S2 takes the winding current values ​​and operating condition categories at times (t-1), (t-2), ..., (tn) as inputs and the rotor displacement value at time t as outputs to train the rotor displacement prediction model; where n is an integer greater than or equal to 2.

[0044] Constructing a rotor displacement prediction model first requires determining the model's input and output. If the current i(t) collected at the current moment is simply used as the input and the displacement x(t) at the current moment as the output—a one-dimensional input, one-dimensional output scenario—a prediction phase delay will occur. In this embodiment, the physical knowledge of the magnetic levitation bearing model is incorporated into the design of the rotor displacement prediction model. Based on a linearized model, the mathematical relationship between the magnetic levitation bearing winding current and the rotor displacement is derived, thereby guiding the selection of the input and output data dimensions.

[0045] Specifically, according to Figure 2 The "controlled object" in the traditional magnetic bearing system control block diagram shown is, without considering the disturbance force f dis In this case, the transfer function between the winding current i and the displacement x can be obtained as follows:

[0046]

[0047] Where, k i k x These are defined as the current stiffness coefficient and the displacement stiffness coefficient, respectively. They are related to the fabrication of the magnetic bearing body and are considered constants. Equation (1) is a mathematical description of a continuous magnetic bearing system. In practical control systems, digital control is often used. Therefore, a discrete mathematical model of the magnetic bearing can better reflect the mathematical relationship between current and displacement in digital control. This invention uses the "zero-order hold (ZOH)" method to discretize the model in equation (1):

[0048]

[0049] Where T is the sampling period and m is the rotor mass.

[0050] Furthermore, this equation in the "z-domain" form can be written in difference form:

[0051]

[0052] Here, k represents the value at the current moment, and (k-1) represents the value at the previous moment. It can be seen that the displacement x(k) at the current moment is related to the displacements x(k-1) and x(k-2) at the previous moment and the currents i(k-1) and i(k-2) at the previous moment. However, the displacement is our predicted value. Neither the displacement at the current moment nor the displacement at the previous moment can be obtained directly. Therefore, equation (3) needs to be further iterated and simplified until the direct relationship between the current displacement x(k) and the current i is found.

[0053] In practical magnetic bearing systems, the sampling time T is approximately 5 × 10⁻⁶. -5 Second, The order of magnitude is 10 2 ,therefore And replace it with A Equation (3) can be further simplified to:

[0054] x(k)-2x(k-1)+x(k-2)=A[i(k-1)+i(k-2)] (4)

[0055] Iterating over equation (4) yields:

[0056]

[0057] At this point, the relationship between the current displacement x(k) and the direct current is obtained.

[0058] According to equation (5), the displacement x(k) at the current moment is related to the current at all historical moments. However, in practical applications, it is impractical to store the current values ​​at all historical moments as input to the network model. Through experiments, as the amount of historical input current data increases, the error of the prediction result gradually tends to a stable value. Therefore, it is not necessary to use all historical current data as input. In order to balance the prediction accuracy and the complexity of the network structure, the historical current at the first five moments is selected as input in this embodiment.

[0059] Furthermore, the operation of a magnetic levitation system typically involves at least two processes: 1. Start-up. This is the process where the rotor is levitated by the electromagnetic force of the magnetic bearing at the protective bearing position. 2. Rotation. This is the process where, after the rotor is stably levitated, the motor is started, and the rotor rotates. Due to manufacturing tolerances, the rotor's mass distribution is unbalanced. This imbalance, when the rotor is rotating, causes the rotor to experience a centrifugal disturbance force related to its rotational speed and frequency. Figure 2 The middle is represented as f dis The magnitude of this disturbance is difficult to describe precisely, but when process 1 is initiated, f dis =0. Therefore, the difference between process 1 and process 2 lies in the disturbance force f. dis The difference lies in the operating conditions. If only current data is used as input, the neural network cannot distinguish between these two operating conditions, resulting in significant prediction errors. Therefore, this invention introduces a new dimension to differentiate the different operating conditions during the operation of the magnetic bearing, namely "Mode". The value of "Mode" can be generated from the motor speed ω reference, and its logical determination is as follows: Figure 3 As shown.

[0060] Specifically, Figure 3 The preset threshold ε is set to 0, f1 = 1, f2 = 2. Figure 3 The automatic judgment procedure shown is as follows: The motor speed ω is detected in real time. When the speed is 0, f1 = 1, and the model input is f1 and five historical current data points. When the speed is not 0, f2 = 2, and the model input is f2 and five historical current data points. The advantage of this setting is that, since labels f1 and f2 are also input into the model during training, the model can autonomously distinguish different operating conditions based on the different inputs during online operation, thus obtaining more accurate predictions.

[0061] Furthermore, before feeding the input data into the model, the original data needs to be digitally filtered to remove high-frequency noise; the input and output data need to be normalized to eliminate the difference in magnitude between the data.

[0062] After determining the dimensions of the input and output, the number of layers, nodes, activation functions, and objective functions of the neural network are determined based on experience.

[0063] For example, in this embodiment, the artificial neural network has N1 = 3 layers: one input layer, one hidden layer, and one output layer. The number of neurons in the hidden layer is N2 = 8. The activation function is the sigmoid function, and the loss function is the mean squared error loss function.

[0064] The constructed neural network model is trained using the backpropagation algorithm to determine the unknown parameters in the neural network. Training stops when the mean square error between the neural network's output and the actual (expected) result is less than a set value, and the neural network is tested using test set data. The parameters from the trained neural network are extracted and written into the constructed artificial neural network model in the real-time controller to predict the real-time displacement of the rotor, and the results are fed back to the control system of the magnetic levitation bearing.

[0065] Application phase:

[0066] Obtain the winding current values ​​and operating condition categories for the previous n time steps, input them into the trained rotor displacement prediction model, and predict the rotor displacement value at the current time step.

[0067] The technical effects achievable by the above embodiments will be further explained below in conjunction with experimental results.

[0068] The experiment collected dynamic and steady-state response data of the magnetic bearing under suspension, rotation and external disturbance conditions, and divided them into training and testing sets, which were then fed into the neural network for training and testing, respectively.

[0069] The test results under the suspension condition are as follows: Figure 4 As shown, Figure 4 In the middle (a), the winding current is... Figure 4 In Figure (b), the displacement output results are shown. The solid line represents the rotor displacement measured by the displacement sensor, and the dashed line represents the rotor displacement predicted by the neural network. The rotor displacement stabilizes after a series of oscillations. It can be seen that the predicted results match the actual results well, with an average error of 3.3 μm.

[0070] The test results under rotating conditions are as follows: Figure 5 As shown, Figure 5 In Figure (a), the winding current is shown under all operating conditions of the motor during startup and shutdown. Figure 5 (b) shows the displacement output result when the motor is running. Figure 5 (c) shows the displacement output when the motor is rotating stably. Figure 5 Figure (d) shows the displacement output under reduced motor speed. Due to rotor mass imbalance, the rotor displacement exhibits an approximately sinusoidal fluctuation in steady state. It can be seen that the predicted results agree well with the actual results, with an average error of 3.1 μm.

[0071] Test results under disturbance conditions are as follows Figure 6 As shown, Figure 6 In the middle (a), the winding current is... Figure 6 In Figure (b), the displacement output results are shown. The solid line represents the rotor displacement measured by the displacement sensor, and the dashed line represents the rotor displacement predicted by the neural network. When the rotor is in a stable suspended state, the stator of the magnetic bearing is struck with a small wooden mallet, and the rotor displacement exhibits irregular disturbances. It can be seen that the predicted results match the actual results well, with an average error of 0.9 μm.

[0072] Those skilled in the art will readily understand 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 within the scope of protection of the present invention.

Claims

1. A method for predicting the displacement of a magnetic levitation bearing rotor, characterized in that, include: Training phase: Collect waveform data of magnetic levitation bearing winding current and actual rotor displacement under different operating conditions; The operating condition category is characterized by motor speed, which includes: levitation condition, rotation condition, and external disturbance condition. If the motor speed is greater than a preset threshold, the operating condition category is rotation condition; otherwise, the operating condition category is levitation condition or external disturbance condition. The levitation condition is the process in which the rotor is levitated on the protective bearing by the electromagnetic force of the magnetic bearing stator. This process can be obtained by changing the PID parameters of the controller. The rotation condition is the process in which the motor starts to rotate after the rotor is levitation. This process can be obtained by changing the motor speed. The external disturbance condition is the random disturbance to the magnetic bearing system from the outside after the rotor is levitation. This disturbance can be obtained by using a wooden mallet to strike the magnetic levitation bearing body at random angles and with random force. The rotor displacement prediction model is trained by taking the winding current values ​​and operating condition categories at times (t-1), (t-2), ..., (tn) as inputs and the rotor displacement value at time t as output; where n is an integer greater than or equal to 2. Application phase: Obtain the winding current values ​​and operating condition categories for the previous n time steps, input them into the trained rotor displacement prediction model, and predict the rotor displacement value at the current time step.

2. The method for predicting the displacement of a magnetic levitation bearing rotor as described in claim 1, characterized in that, The rotor displacement prediction model is a rotor displacement prediction model based on artificial neural networks; The artificial neural network includes N One hidden layer, each hidden layer includes N Two nodes; the activation function is Sigmoid, Tanh, or ReLU, and the loss function is the mean squared error loss function.

3. The method for predicting the rotor displacement of a magnetic levitation bearing as described in claim 1, characterized in that, After collecting waveform data of the winding current of the magnetic levitation bearing under different operating conditions, the method further includes: performing digital filtering and normalization on the waveform data of the winding current, and then inputting it into the rotor displacement prediction model.

4. A magnetic levitation bearing rotor displacement self-sensing system, characterized in that, include: The magnetic levitation bearing body, controller, and power amplifier; The controller uses the magnetic levitation bearing rotor displacement prediction method as described in any one of claims 1-3 to predict the rotor displacement value at the current moment; and generates a corresponding current command based on the difference between the rotor displacement reference value and the predicted value. The power amplifier outputs a corresponding current value according to the current command and acts on the magnetic levitation bearing body to control the rotor displacement of the magnetic levitation bearing.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the magnetic levitation bearing rotor displacement prediction method as described in any one of claims 1-3.

Images