Actuating unit excited bearing rotor position and attitude prediction and calculation method

By combining numerical solution of air film mechanics with multi-input multi-output neural networks, the problems of large computational load and high time delay in solving the rotor position and attitude of air-bearing/air film bearings are solved, realizing fast and stable rotor position and attitude calculation and control, which is applicable to various bearings and actuation excitation methods.

CN121980898APending Publication Date: 2026-05-05LUOYANG BEARING RES INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511777763.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-28
Publication Date
2026-05-05

AI Technical Summary

Technical Problem

Existing technologies involve large computational loads and high time delays in solving the rotor position and attitude of air-bearing/film bearings, making it difficult to meet the requirements for rapid calculation and real-time control of multi-point trajectories. Furthermore, traditional approximation models have a narrow range of applicability.

Method used

A predictive model is constructed by combining numerical solution of air film mechanics with multi-input multi-output neural network. By establishing dimensionless Reynolds equations and discrete iterative solution, and training the predictive model with multi-input multi-output neural network, the rotor position and attitude can be solved quickly inversely.

Benefits of technology

It significantly reduces latency, improves control real-time performance and stability, has a wide range of applications, and can cover various bearings and different actuation excitation methods, enabling rapid calculation and real-time control of rotor position and attitude.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121980898A_ABST
    Figure CN121980898A_ABST
Patent Text Reader

Abstract

The invention relates to an actuating unit excited bearing rotor position and attitude prediction and calculation method, which is characterized in that gas film mechanics numerical solution is taken as a physical basis, a prediction model is constructed in combination with a multi-input multi-output neural network, and the precision of the prediction model is verified on a test set by using indexes such as multiple correlation coefficients and the like. According to the method, the parameters of the actuating unit can be quickly predicted and reversely solved, so that the rotor position attitude is quickly and reversely solved, the active regulation and control of the rotor position attitude are realized by changing the excitation parameters of the actuating unit and dynamically adjusting the distribution of a gas film pressure field, the time delay can be remarkably reduced, and the control real-time performance and stability are improved; moreover, the method can cover various bearings and different actuation excitation modes, gives consideration to the solving speed, the application range and the convenience, and is higher in practical value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of non-contact support and precision bearing technology, specifically to a method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit. Background Technology

[0002] Air film / air-floating bearings have low friction and high cleanliness. The bearing is excited by the actuation unit to form a micron-level air film and establish a pressure field between the rotors. The rotor position and attitude can be actively controlled by changing the excitation parameters of the actuation unit. It is widely used in high-cleanliness handling, precision positioning and high-speed rotation scenarios. The rotor position and attitude of the bearing are coupled by multiple factors: the amplitude of the actuation unit, the air film gap, the load, etc., and exhibit strong nonlinearity and strong coupling characteristics.

[0003] There are two main approaches to solving the rotor position and attitude of this bearing: one is to directly solve the Reynolds equations numerically, calculating the film pressure and load for each operating condition parameter, and then iteratively searching for the rotor position and attitude that satisfy the force / torque balance; this method has a large computational load and high time delay, making it difficult to meet the requirements for fast solution and real-time control of multi-point trajectories; the other is to use an approximate model based on small disturbance linearization or empirical formulas, which has higher computational efficiency but a narrow range of applications.

[0004] Therefore, there is an urgent need for a fast algorithm for solving the rotor position and attitude of bearings that has a wide range of applications and ensures physical consistency and accuracy constraints, so as to meet the requirements of rapid calculation and real-time control of rotor trajectory of air-bearing / film bearings. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit. Based on numerical solutions of air film mechanics, a prediction model is constructed by combining a multi-input multi-output neural network, thereby achieving rapid reverse calculation of rotor position and attitude. This method can significantly reduce time delay, improve control real-time performance and stability, and can cover various bearings and different actuation excitation methods, taking into account the solution speed, applicability and convenience.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit. The bearing is excited by the actuation unit to form an air film and establish a pressure field between the rotors. The position and attitude of the rotor are actively controlled by changing the excitation parameters of the actuation unit. The method for predicting and calculating the position and attitude includes the following steps: S1 Modeling and Dimensionless Conversion: Establish the relationship between the bearing structure geometric model and the gas film thickness composition, and derive a dimensionless Reynolds equation suitable for the bearing structure. The gas film thickness composition includes static position terms and dynamic deformation terms. S2 Discretization and Iterative Solution: The dimensionless Reynolds equations are discretized using finite difference or equivalent schemes, and the pressure field is obtained by nonlinear iteration of the discretized equations. S3 position and attitude solution: Based on the pressure field distribution of the bearing, calculate the bearing capacity and rotational torque of the bearing rotor, and obtain the static equilibrium position and attitude of the rotor according to the equilibrium criterion. S4 working area sampling and data construction: Select target points in the reachable working area of ​​the bearing rotor, sample and generate a dataset about "actuator unit parameters - rotor position and attitude", and divide the dataset into training set and test set; S5 Prediction Model Training and Validation: The prediction model is trained using a multi-input multi-output neural network, and the model accuracy is validated on the test set. S6 Application: Utilizing a trained and validated prediction model, the corresponding actuator parameters are quickly solved in reverse based on the rotor position and attitude, enabling active trajectory control of the air bearing.

[0007] Furthermore, the bearing type is at least one of radial bearing or thrust bearing, and the bearing structure is at least one or any combination of annular actuation, segmented multi-unit, tilting pad, conical surface or spherical surface, wherein the number of segments in the segmented multi-unit bearing is ≥3.

[0008] According to claim 2, the method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit is characterized in that, in step S1, the dimensionless Reynolds equation for the radial bearing in (θ, Z) is: ; In the formula For circumferential angle, For axial coordinates, For time-varying terms, The dimensionless film pressure denoted as dimensionless air film thickness, and T as dimensionless time; The dimensionless Reynolds equation for a thrust bearing in (R,θ) is: ; In the formula, R is the radial coordinate. For circumferential angle, For time-varying terms, Dimensionless film pressure denoted as dimensionless air film thickness, and T as dimensionless time.

[0009] Further, in step S1, the dynamic deformation term is summarized by the bearing structure deformation and air film thickness under the excitation of the actuation unit as follows: ; Where r is the polar radius. Let H be the polar angle, t be time, and H be the polar angle.s H is a static position term. d This is a dynamic deformation term, which is obtained by fitting the bearing's modal vibration modes.

[0010] Furthermore, in step S2, the nonlinear iteration adopts the Newton-Raphson method, the Gauss-Seidel method, or the over-relaxation iteration method; the convergence condition is that the pressure difference and position pose difference under adjacent iterations are both less than the threshold, and the deviation between the air film bearing and the external load is less than the threshold.

[0011] Furthermore, in step S3, the pressure is first integrated into the air film force of the bearing block and decomposed into each degree of freedom. The bearing stiffness model is then used to obtain the orientation correction. Finally, the static equilibrium position and attitude of the rotor are determined by matching the periodic average suspension force and torque with the external load and torque.

[0012] Furthermore, in step S4, the target point is selected by regular grid sampling, Latin hypercube sampling, or a combination of both.

[0013] Furthermore, in step S4, the data in the dataset is preprocessed by normalizing and standardizing in both the input and output dimensions, and then divided into training and testing sets.

[0014] Furthermore, in step S5, the multi-input multi-output neural network includes an input layer, a hidden layer, and an output layer. The input parameters of the input layer are the rotor position and attitude, and the output parameters of the output layer are the actuation unit parameters.

[0015] Furthermore, in step S5, the accuracy of the prediction model is measured based on the multiple correlation coefficient, which is calculated as follows: Calculate, where n is the number of samples in the test set. This represents the actual observed value of the i-th sample in the test set. This represents the predicted value of the i-th sample in the test set. This represents the actual observed mean of all samples in the test set. If the multiple correlation coefficient of the test set is >0.95, the model is considered to meet the accuracy requirements.

[0016] Beneficial effects: The method proposed in this invention is based on numerical solutions of air film mechanics. Following a combined approach of "numerical solution-neural network," it first uses a high-precision air film numerical method to generate "actuator unit parameters - rotor position / attitude" samples. Then, a neural network learns this mapping to quickly predict and inversely calculate the actuator unit parameters, thereby significantly reducing latency and improving control real-time performance and stability. The beneficial effects are mainly reflected in: Physical consistency: This invention is based on numerical solution of air film, which preserves the physical essence of the Reynolds equation and ensures the physical rationality of the model prediction results.

[0017] High efficiency: Using neural networks instead of iterative search significantly reduces the amount of computation; single-point solutions are reduced from numerical computation to milliseconds, and batch solutions for multiple points of the trajectory can be completed in near real time.

[0018] Accuracy and reliability: The accuracy of the prediction model is verified on the test set using indicators such as the multiple correlation coefficient, and it has been experimentally verified to have high reliability and high accuracy.

[0019] Wide applicability: The same process covers radial / thrust bearings and various actuation excitations, breaking through the limitations of the applicability of traditional approximate models.

[0020] The method of this invention enables active control of rotor position and attitude. By changing the excitation parameters of the actuation unit, the distribution of the air film pressure field is dynamically adjusted, thereby actively controlling the rotor position and attitude. It is suitable for scenarios such as high-cleanliness handling, precision positioning, and high-speed rotation. Attached Figure Description

[0021] Figure 1 A schematic diagram of the overall process of the method of the present invention; Figure 2 Schematic diagram of bearing structure / actuator structure layout; where... Figure 2 (b) is Figure 2 (a) Left view; Figure 3 Training data selection illustration and sampling strategy; among which, Figure 3 (b) is Figure 3 (a) Sampling diagram of the rotor working area; Figure 4 Schematic diagram of a neural network structure; Figure 5 Results of the circular trajectory experiment; Figure 6 Results of the triangle locus experiment; Figure 7 Results of the curved trajectory experiment. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, this invention proposes a method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit. The bearing forms a micron-level air film between the rotors through the excitation of the actuation unit. The air film pressure distribution constitutes a pressure field, providing load and position constraints for the rotor. By changing the excitation parameters of the actuation unit, the distribution of the air film pressure field is dynamically adjusted, thereby actively controlling the position and attitude of the rotor, meeting the requirements for rotor motion accuracy in scenarios such as high-cleanliness handling, precision positioning, and high-speed rotation.

[0024] The bearing type is at least one of radial bearing or thrust bearing, and the bearing structure is at least one or any combination of annular actuation, segmented multi-unit (≥3), tilting pad, conical surface or spherical surface.

[0025] This invention comprises two main parts: numerical solution of air film and neural network modeling. The detailed steps include the following steps S1-S6.

[0026] Step S1, Modeling and Dimensionlessization: Establish the bearing structure geometry and gas film thickness composition, and derive the dimensionless Reynolds equation suitable for the bearing structure.

[0027] Structure and Actuation: Considering practical acoustic suspension bearing structures (such as radial / thrust type, segmented type, tilting pad type, etc.), such as Figure 2 As shown, the bearing is driven by multiple actuation units (piezoelectric actuators) to generate high-frequency vibration; this creates a high-pressure gas film between the bearing and the rotor, and the generated gas film pressure is adjusted by the actuation structure parameters, thereby adjusting the position and attitude of the rotor.

[0028] Modeling assumptions: the gas is approximately isothermal and laminar; the gas film thickness is much smaller than the characteristic scale; the pressure gradient along the thickness direction is negligible, etc.

[0029] Film thickness: Decomposed using static morphology + dynamic deformation. The static term is determined by nominal clearance, rotor eccentricity, initial bearing bearing orientation, etc.; the dynamic term is approximated by the response of the bearing housing at the operating frequency of the actuation unit (usually expressed in mode shape fitting form).

[0030] Governing equations: The continuity equation and momentum equation for compressible gases are simplified and dimensionless in cylindrical / polar coordinates to obtain the Reynolds equation coupled with film thickness.

[0031] The dimensionless Reynolds equation for a radial bearing in cylindrical coordinates (θ, Z) is: In the formula, For circumferential angle, For axial coordinates, It is a time-varying term; Dimensionless film pressure; denoted as dimensionless air film thickness, and T as dimensionless time.

[0032] The dimensionless Reynolds equation for the thrust bearing in polar coordinates (R,θ) is: ; In the formula, R is the radial coordinate. For circumferential angle, It is a time-varying term; Dimensionless film pressure; denoted as dimensionless air film thickness, and T as dimensionless time.

[0033] The film thickness is composed of a static position term and a dynamic deformation term, wherein the dynamic deformation term is caused by the deformation of the bearing structure under the excitation of the actuating unit. The film thickness can be summarized as follows: ; Where r is the polar radius. Let Hs be the polar angle, t be time, Hs be the static position term, and H be the polar angle. d This is a dynamic deformation term, which is obtained by fitting the bearing's modal vibration modes.

[0034] Step S2, Discretization and Iterative Solution: Discretize the control equations using finite difference or equivalent schemes, and obtain the pressure field using the Newton-Raphson method, Gauss-Seidel method, or over-relaxation iteration.

[0035] Discretization method: The solution domain is divided into axial and circumferential grids, and the Reynolds equations are constructed using central difference to create discrete equations; Iterative framework: The pressure field is solved by Newton-Raphson equivalent nonlinear iteration or equivalent nonlinear iteration; the convergence condition is that the pressure difference and position pose difference under adjacent iterations are both less than the threshold, and the deviation between the air film bearing and the external load is less than the threshold.

[0036] Step S3, Position and Attitude Solution: Based on the pressure field distribution of the bearing, calculate the bearing capacity and rotational torque of the bearing rotor, and obtain the static equilibrium position and attitude of the rotor according to the equilibrium criterion.

[0037] First, the pressure is integrated into the air film force of the bearing blocks and decomposed into each degree of freedom. The bearing stiffness model is then used to obtain the orientation correction. Then, based on the matching of the periodic average suspension force and torque with the external load and torque, the static equilibrium position and attitude of the rotor are determined.

[0038] For example, if a radial bearing is used in cylindrical coordinates (θ, z), with a bearing radius of R and an effective axial length of L, then the resultant force components Fx and Fy on the radial section and the overturning moments Mx and My about the x and y axes are respectively: ; Fx is the result of the cosine component of the film pressure Δp(θ,z) along the θ direction (circumferential direction) in cylindrical coordinates combined with the radius R, and the integral covers the entire bearing surface Ω.

[0039] Fy depends on the circumferential sinusoidal component of the air film pressure. It is also magnified by the radius R and integrated to obtain the resultant force of the bearing perpendicular to the axial direction (y direction), which reflects the influence of pressure distribution on the lateral force.

[0040] Mx is the product of the film pressure Δp(θ,z) and the axial position z and the circumferential sinusoidal component. It is obtained by integrating after magnification with radius R. The negative sign indicates that the direction of the torque is related to the coordinate system definition, reflecting the pressure distribution's resistance to overturning around the x-axis.

[0041] My combines the film pressure, axial position z, and circumferential cosine component, and after integration, obtains the overturning moment about the y-axis, which reflects the supporting role of pressure distribution on bearing stability.

[0042] The resultant force components Fx and Fy directly reflect the supporting force of the air film pressure on the bearing and are the core indicators of the bearing's load-bearing capacity.

[0043] Overturning moments Mx and My reflect the bearing's ability to resist external torques and maintain a stable posture, which is crucial for scenarios such as high-speed rotation and precision positioning.

[0044] The above formula reveals the mechanical balance mechanism of radial bearings under complex loads. By adjusting the gas film pressure distribution Δp(θ,z) (such as through excitation by the actuation unit), these mechanical parameters can be actively controlled to achieve the optimization and dynamic adaptation of bearing performance.

[0045] Thrust bearing: Using polar coordinates (r, θ), the axial load capacity Fz and the pitch / roll moments Mx and My are: ; The axial bearing capacity Fz is the integral of the air film pressure Δp(θ,z) in the polar coordinate system with respect to the circumferential direction θ after weighting (rdr) along the radius r.

[0046] rdr: Reflects the contribution of the pressure distribution in the radial direction to the bearing capacity. The larger the radius, the greater the contribution of the same pressure to the bearing capacity (linear weighting).

[0047] Integral coverage Ω: Typically covers the entire bearing surface, ensuring that the film pressure in all areas is included in the calculation.

[0048] Pitching moment Mx is the ratio of film pressure to the square of radius (r) 2 The combined effect of the circumferential sinusoidal component (sinθ) reflects the pressure distribution's resistance to bearing overturning around the x-axis. Among them, r 2That is, the square of the radius amplifies the contribution of the outer region to the torque; that is, the farther away from the axis, the more significant the effect of pressure change on the pitching torque. The negative sign indicates that the direction of the torque is related to the coordinate system definition and is usually opposite to the positive direction.

[0049] Rolling moment is similar to pitching moment, but depends on the circumferential cosine component (cosθ), reflecting the pressure distribution's resistance to bearing overturning around the y-axis. The product of the square of the radius and the cosine component determines the contribution of pressure at different radii and circumferential positions to the rolling moment.

[0050] The difference between sinθ and cosθ makes the pitch and roll moments have different sensitivities to circumferential changes in pressure distribution, providing a theoretical basis for the dynamic control of bearings.

[0051] Step S4, Workspace Sampling and Data Construction: (e.g.) Figure 3 As shown, target points are selected within the reachable working area of ​​the bearing rotor, and a dataset of "actuator element parameters - rotor position and attitude" is generated by sampling using a regular grid or Latin hypercube. The data in the dataset are preprocessed by normalizing and standardizing in both the input and output dimensions, and then divided into training and test sets, as follows: Working area setting: The reachable static position area is determined with the bearing geometric center as the origin and under the constraints of parameters such as design clearance; Sampling strategy: Select several representative target points within this area according to a preset step size or the Latin hypercube scheme; Data generation: For each target point, steps S2 and S3 are called to perform numerical solutions to obtain the dataset formed by the corresponding actuation unit parameters.

[0052] Normalization and splitting: Split the dataset according to the ratio of training / test sets.

[0053] Step S5: Prediction model training and validation.

[0054] Training rules: Train a multi-input multi-output neural network (e.g., RBF or MLP) and optimize parameters such as weights, center and base width.

[0055] Accuracy metrics are validated on an external test set using indicators such as the multiple correlation coefficient. If the threshold is not reached, data sampling is increased or hyperparameters are adjusted before retraining. If the multiple correlation coefficient of the test set is >0.95, the accuracy requirement is considered met.

[0056] like Figure 4 As shown, the multiple input multiple output neural network includes an input layer, a hidden layer, and an output layer. The input parameters of the input layer are the rotor position and orientation, and the output parameters of the output layer are the actuator unit parameters.

[0057] The accuracy of a prediction model is measured by the multiple correlation coefficient, which is calculated using the following formula. ; Where n is the number of samples in the test set. This represents the actual observed value of the i-th sample in the test set. This represents the predicted value of the i-th sample in the test set. This represents the actual observed mean of all samples in the test set. A negative correlation coefficient > 0.95 is considered to meet the accuracy requirements.

[0058] Step S6, Verification and Application: The corresponding actuation unit parameters are quickly obtained using the trained prediction model. A rotor-bearing system is built for experimental verification. Then, using the verified model, the corresponding actuation unit parameters are quickly solved in reverse based on the rotor position and attitude to achieve active trajectory control of the air bearing.

[0059] Trajectory setting: Preset typical trajectories such as circles, triangles, and curves, and use a trained neural network prediction model to quickly obtain the parameters of the corresponding actuation unit.

[0060] Experimental platform: A rotor-bearing system was built to test the actuation unit parameters calculated by the predictive model to see if the rotor can achieve the expected position and attitude.

[0061] according to Figure 5-7 The experimental results of circular trajectory, triangular trajectory and curved trajectory show that the actuation unit parameters calculated by the prediction model can enable the rotor to reach the expected position and attitude with high accuracy.

[0062] In summary, this invention targets actuated film bearings, incorporating the modal deformation of the actuation structure into the film thickness and coupling it with the Reynolds equation solution. It uses a multi-input multi-output neural network to establish a high-dimensional mapping between "actuation parameters and rotor position / attitude," and proposes a fast inverse calculation scheme for "target position / attitude and actuation parameters." Based on the rapidly calculated actuation unit parameters, it achieves active control of the rotor position and attitude, significantly improving the real-time performance and applicability of active air bearing position and attitude calculation and control.

[0063] This invention uses a combination of numerical solution and neural network to reduce rotor pose calculation from the second level to the millisecond level, enabling online prediction and parameter inverse solution, significantly improving positioning accuracy and response speed, reducing debugging costs and downtime, with a wide range of applications and prospects for large-scale implementation.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit, wherein the bearing forms an air film and establishes a pressure field between the rotors by excitation from the actuation unit, and the position and attitude of the rotor are actively controlled by changing the excitation parameters of the actuation unit, characterized in that, The position and attitude prediction and calculation method includes the following steps: S1 Modeling and Dimensionless Conversion: Establish the relationship between the bearing structure geometric model and the gas film thickness composition, and derive a dimensionless Reynolds equation suitable for the bearing structure. The gas film thickness composition includes static position terms and dynamic deformation terms. S2 Discretization and Iterative Solution: The dimensionless Reynolds equations are discretized using finite difference or equivalent schemes, and the pressure field is obtained by nonlinear iteration of the discretized equations. S3 position and attitude solution: Based on the pressure field distribution of the bearing, calculate the bearing capacity and rotational torque of the bearing rotor, and obtain the static equilibrium position and attitude of the rotor according to the equilibrium criterion. S4 working area sampling and data construction: Select target points in the reachable working area of ​​the bearing rotor, sample and generate a dataset about "actuator unit parameters - rotor position and attitude", and divide the dataset into training set and test set; S5 Prediction Model Training and Validation: The prediction model is trained using a multi-input multi-output neural network, and the model accuracy is validated on the test set. S6 Application: Utilizing a trained and validated prediction model, the corresponding actuator parameters are quickly solved in reverse based on the rotor position and attitude, enabling active trajectory control of the air bearing.

2. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, The bearing type is at least one of radial bearing or thrust bearing, and the bearing structure is at least one of ring actuation, segmented multi-unit, tilting pad, conical surface or spherical surface or any combination thereof, wherein the number of units in the segmented multi-unit type is ≥3.

3. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 2, characterized in that, In step S1, the dimensionless Reynolds equation for the radial bearing in (θ, Z) is: ; In the formula For circumferential angle, For axial coordinates, For time-varying terms, The dimensionless film pressure Where is the dimensionless gas film thickness, and T is the dimensionless time; The dimensionless Reynolds equation for a thrust bearing in (R,θ) is: ; In the formula, R is the radial coordinate. For circumferential angle, For time-varying terms, Dimensionless film pressure denoted as dimensionless air film thickness, and T as dimensionless time.

4. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S1, the dynamic deformation term is determined by the bearing structure deformation under the excitation of the actuating unit, and the air film thickness is summarized as follows: ; Where r is the polar radius. Let H be the polar angle, t be time, and H be the polar angle. s H is a static position term. d This is a dynamic deformation term, which is obtained by fitting the bearing's modal vibration modes.

5. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S2, the nonlinear iteration adopts the Newton-Raphson method, the Gauss-Seidel method, or the over-relaxation iteration method; the convergence condition is that the pressure difference and position pose difference under adjacent iterations are both less than the threshold, and the deviation between the air film bearing and the external load is less than the threshold.

6. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S3, the pressure is first integrated into the air film force of the bearing block and decomposed into each degree of freedom. The bearing stiffness model is then used to obtain the posture correction. Finally, the static equilibrium position and attitude of the rotor are determined by matching the periodic average suspension force and torque with the external load and torque.

7. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S4, the target point is selected by regular grid sampling, Latin hypercube sampling, or a combination of both.

8. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S4, the data in the dataset is preprocessed by normalizing and standardizing in both the input and output dimensions, and then divided into training and test sets.

9. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S5, the multiple input multiple output neural network includes an input layer, a hidden layer, and an output layer. The input parameters of the input layer are the rotor position and attitude, and the output parameters of the output layer are the actuator unit parameters.

10. The method for predicting and calculating the position and attitude of a bearing rotor excited by an actuation unit according to claim 1, characterized in that, In step S5, the accuracy of the prediction model is measured based on the multiple correlation coefficient, which is calculated as follows: Calculate, where n is the number of samples in the test set. This represents the actual observed value of the i-th sample in the test set. This represents the predicted value of the i-th sample in the test set. This represents the actual observed mean of all samples in the test set. If the multiple correlation coefficient of the test set is >0.95, the model is considered to meet the accuracy requirements.