Multi-stage rotor inertia main shaft space pose regulation and control method based on reliability distribution

Abstract

The invention discloses a multi-stage rotor inertia main shaft space pose regulation and control method based on reliability distribution, and relates to a multi-stage rotor inertia main shaft space pose regulation and control method. The invention aims to solve the problems of insufficient regulation and control precision and lack of reliability distribution and self-adaptive capability in the prior art. The method comprises the following steps: firstly, selecting two parameters of inertia main shaft offset and inertia main shaft inclination angle, and representing the inertia main shaft space pose of a rotor; secondly, establishing a relationship between the reliability and the spatial pose of the inertial principal axis; then, based on a dynamic self-adaptive weight method, the established reliability indexes are distributed; and finally, designing a deep reinforcement learning agent, and realizing spatial pose regulation and control of the multi-stage rotor inertia main shaft by interacting with the environment to learn an optimal regulation and control strategy. According to the method, the pose regulation and control precision can be improved, the overall reliability of the system is ensured, the robustness of the rotor system is improved, and the regulation and control efficiency is improved. The invention belongs to the technical field of aero-engine assembly.

Description

Technical Field

[0001] This invention relates to a method for controlling the spatial orientation of a multi-stage rotor inertial spindle, belonging to the field of aero-engine assembly technology. Background Technology

[0002] As highly complex and precise power machinery, the assembly quality of the rotor system of an aero-engine determines its overall performance and lifespan. Under high-speed rotation, the misalignment between the rotor's principal axis of inertia and its geometric axis of rotation can induce significant unbalanced excitation forces and inertial moments, which is one of the main causes of severe engine vibration and even operational failure. To control this misalignment, scholars and research institutions both domestically and internationally have developed assembly methods based on geometric error propagation control. The core of this technology lies in obtaining the form and position errors of each stage of the rotor through precise measurement, predicting their cumulative effects at different assembly angles using a spatial coordinate transformation model, and finally optimizing the assembly angles of each stage of the rotor using geometric quantities such as centroidal coaxiality as the objective function, thereby minimizing the misalignment of the principal axis of inertia. This method improves the initial balance of the rotor to a certain extent and reduces vibrations introduced during assembly. However, with the continuous improvement of the performance and reliability requirements of aero-engines, the aforementioned control strategies centered on geometric accuracy have shown limitations.

[0003] First, current technologies lack steps to directly integrate reliability indicators into the assembly process, and a clear, quantifiable relationship has not yet been established between the optimization goals of the assembly process and the reliability of the system. For example, in pursuit of ultimate coaxiality, traditional optimization algorithms may result in an assembly angle that leads to uneven distribution of preload on some critical connecting bolts or localized stress concentration in specific thin-walled rotor sections. While this assembly state may be geometrically optimal, it could reduce the fatigue life of that component, making it a weak link in the overall rotor system's reliability and thus failing to meet the engine's design requirements for high reliability and long service life.

[0004] Secondly, current control strategies lack sufficient adaptability in terms of component manufacturing tolerances, material property dispersion, and the inherent uncertainties of the assembly process. Once a key parameter in the assembly deviates from the assumptions of the theoretical model, the original basis for optimal pose calculation no longer exists, and the system cannot dynamically adjust based on these real-time feedbacks, leading to an uncontrollable risk in the final assembly quality.

[0005] Therefore, there is a need in this field for a new method that can deeply integrate the reliability allocation process into the assembly control process, so as to realize the transformation from "precision assembly" to "reliable assembly" and ensure that the aero-engine rotor system not only meets the precision requirements during assembly, but also has a certain degree of reliability throughout its entire life cycle. Summary of the Invention

[0006] To address the problems of insufficient control precision, lack of reliability allocation, and lack of adaptive capability in existing technologies, this invention proposes a multi-level rotor inertial spindle spatial pose control method based on reliability allocation.

[0007] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the present invention include: Step 1: Set up the measurement system, measure and calculate relevant parameters; Step 2: Perform inertial principal axis spatial pose characterization; Step 3: Establish the relationship between reliability and inertial principal axis spatial pose; Step 4: Perform reliability allocation; Step 5: Perform multi-stage rotor inertial spindle spatial pose control.

[0008] Furthermore, in step 2, two vectors, principal axis offset and principal axis tilt, are selected to characterize the spatial pose of the rotor's principal axis.

[0009] Furthermore, in step 3, the spatial pose reliability of the inertial principal axis is defined as the probability that both coaxiality and tilt angle meet the design requirements.

[0010] Furthermore, step 4 employs a reliability allocation method that considers dynamic adaptive weights, the specific steps of which include: Step 401: Decompose system reliability indicators; Step 402: Perform dynamic weight allocation; introduce dynamic weight factors and adjust the reliability allocation weights for real-time measurement data; Step 403: Perform adaptive reliability weight allocation. Based on real-time data, adaptively redistribute the reliability of each stage and unify the reliability data to... Inside.

[0011] Furthermore, step 5 includes the following steps: Step 501: Construct the state space. Including time t Overall posture System reliability Positional errors at each stage and dynamic weights ; Step 502: Construct the action space. Output phase of each stage of rotor assembly Adjustment amount This enables optimized indexing and assembly. Step 503: Set the reward function. It is responsible for guiding the intelligent agent to learn, thereby improving its reliability, reducing weighted pose error, avoiding excessive control actions, and ensuring a smooth control process. Step 504: Design a deep reinforcement learning network. The network is designed using the Actor-Critic framework, where the Actor network outputs actions based on the state, the Critic network evaluates the value of the actions, and the gradient ascent method is used to update the parameters of the Actor network. Step 505: Conduct training and optimization. By interacting with the environment, continuously learn and adjust error parameters to minimize the error. Step 506: Perform dynamic control by combining the physical entities of the multi-stage rotors with the trained virtual model to form a closed-loop control.

[0012] Furthermore, step 505 employs a phased training strategy. First, basic training is conducted by setting the number of training rounds, the maximum number of steps per round, and the learning rate parameters. Then, the optimization process continues based on the results of the basic training to reduce noise.

[0013] Furthermore, step 506 specifically includes: Step A: Acquire the spatial pose data of the rotor inertial spindle in real time through sensors and update relevant parameters synchronously; Step B, Decision-making and Execution. The deep reinforcement learning agent calculates the optimal control action, i.e., the phase adjustment amount, based on the current state, and sends the control command to the physical actuator; Step C: Implement dynamic weight updates and strategy optimization. Update the weights in real time based on the pose error of the rotor's inertial spindle. Continuously optimize the strategy using the updated data to adapt to changes in rotor state.

[0014] The beneficial effects of this invention are as follows: By combining dynamic reliability allocation with spatial pose control of a multi-stage rotor inertial spindle, and employing deep reinforcement learning for pose control execution, this invention achieves a technological innovation in pose control of a multi-stage rotor inertial spindle. This method innovatively establishes a bidirectional optimization mechanism for pose control and reliability allocation, and through continuous interaction between the agent and the environment, autonomously learns the optimal control strategy, significantly improving the system's adaptive capability. In practical applications, the system can dynamically adjust the reliability weights of each component according to real-time operating conditions, achieving the optimal balance between accuracy requirements and system reliability. Compared to traditional control methods, pose control accuracy is improved by approximately 15%, system reliability is improved by approximately 12%, and control time is shortened by approximately 26%, effectively improving control efficiency. This method effectively solves long-standing technical problems in traditional rotor assembly, such as over-reliance on expert experience, insufficient control accuracy, and lack of reliability allocation, providing a new method for the assembly of complex and precision equipment, and has significant application value and promising prospects for widespread application. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram of the operation steps of the reliability allocation method; Figure 3 This is a schematic diagram of the operation steps of the multi-stage rotor inertial spindle spatial pose control method; Figure 4 This is a schematic diagram of the convergence process of spatial pose error of a multi-stage rotor inertial spindle. Detailed Implementation

[0016] Specific implementation method one: as follows Figures 1 to 4 As shown, a multi-stage rotor inertial spindle spatial pose control method based on reliability allocation includes the following steps: Step 1: Set up the measurement system, measure and calculate relevant parameters; Step 2: Characterize the spatial pose of the inertial principal axis; select two vectors, inertial principal axis offset and inertial principal axis tilt, to characterize the spatial pose of the rotor's inertial principal axis; let the first... i The pose of the primary rotor's inertial spindle in the rotary coordinate system is as follows: , in, and These are the principal axis offset vector and the principal axis tilt vector, respectively. After the multi-stage rotors are assembled, the overall inertial spindle pose is: , in, For the first i The pose transfer matrix of the stage rotor; Step 3: Establish the relationship between reliability and spatial pose of the inertial principal axis; the reliability of the spatial pose of the inertial principal axis is defined as the probability that both coaxiality and tilt angle meet the design requirements. , in, This is the coaxiality function, representing the maximum offset of the overall principal axis of inertia; is the tilt angle function, representing the maximum tilt angle of the overall principal axis of inertia; and These represent the system's maximum allowable offset and maximum tilt angle, respectively. Step 4: Perform reliability allocation; adopt a reliability allocation method that considers dynamic adaptive weights, the specific steps of which include: Step 401: Decompose the system reliability index; let the target reliability of the system be... It needs to be allocated to m The key regulatory link, the first jThe reliability of each link is Assuming the various control links are connected in series, then: , in, Must meet: ; Step 402: Perform dynamic weight allocation; introduce dynamic weight factors. Furthermore, the reliability allocation weights are adjusted based on the real-time measurement data, and the calculation formula is as follows: , in, This is a sensitivity coefficient used to control the degree of influence of error on the weights. Its specific value is determined through simulation experiments. When selecting a value, it is chosen so that the weights can effectively distinguish between different error levels. for t Time of the first j The current pose error of each stage is calculated using the following formula: , After adjustment j The reliability of each link is: ; Step 403: Perform adaptive reliability weight allocation. Based on real-time data, adaptively redistribute the reliability of each stage and unify the reliability data to... within; No. j The reliability of each link is: , in, This represents the average pose error; The adjustment coefficient has a range of values. This is used to control the adjustment range and avoid violent oscillations; Traditional reliability allocation methods have certain limitations: fixed allocation weights cannot adapt to changes in system state; they lack feedback and cannot be adjusted based on actual operating data; they are overly conservative, over-designing to ensure the weakest link; and they waste resources, failing to achieve dynamic optimization of reliability resources. The method used in this implementation dynamically adjusts the reliability allocation weights based on the actual performance of each component, thereby improving fault tolerance and adapting to changes in operating conditions. Step 5: Perform multi-stage rotor inertial spindle spatial pose control. Specific steps include: Step 501: Construct the state space. Including time t Overall posture System reliability Positional errors at each stage and dynamic weights Specifically: ; Step 502: Construct the action space. Output phase of each stage of rotor assembly Adjustment amount To achieve optimized indexing and assembly; specifically: ; Step 503: Set the reward function. It is responsible for guiding the agent's learning, improving its reliability, reducing weighted pose error, avoiding excessive control actions, and ensuring a smooth control process; specifically: , in, , , Adjusting the coefficients of the reward function to balance the various weights; Step 504: Design a deep reinforcement learning network. The network is designed using the Actor-Critic framework, where the Actor network outputs actions based on the state, the Critic network evaluates the value of the actions, and the gradient ascent method is used to update the parameters of the Actor network. Step 505: Conduct training and optimization. By interacting with the environment, continuously learn and adjust error parameters to minimize the error. A phased training strategy is adopted. First, basic training is carried out by setting the number of training rounds, the maximum number of steps per round, and the learning rate parameters. Then, the optimization process continues to train based on the results of the basic training to reduce noise. Step 506: Perform dynamic control by combining the physical entities of the multi-stage rotors with the trained virtual model to form a closed-loop control. Step A: Acquire the spatial pose data of the rotor inertial spindle in real time through sensors and update relevant parameters synchronously; Step B, Decision-making and Execution. The deep reinforcement learning agent calculates the optimal control action, i.e., the phase adjustment amount, based on the current state, and sends the control command to the physical actuator; Step C: Implement dynamic weight updates and strategy optimization. Update the weights in real time based on the pose error of the rotor's inertial spindle. Continuously optimize the strategy using the updated data to adapt to changes in rotor state.

[0017] In step 5 of this implementation, a deep reinforcement learning agent is designed to learn the optimal control strategy through interaction with the environment. Compared to traditional methods, the unique advantage of deep reinforcement learning lies in its multi-objective collaborative optimization capability. The system can simultaneously consider multiple objectives such as pose accuracy and reliability indicators, seeking the optimal balance under complex constraints. This comprehensive optimization capability enables the system to maintain excellent performance in steady state while achieving rapid response and smooth transition in dynamic processes, realizing the collaborative optimization of pose control and reliability allocation.

[0018] Example Step 1: Set up a measurement system and measure and calculate relevant parameters; measure the form and position errors, center of mass offset, moment of inertia and product of inertia of each stage of rotor to obtain relevant parameters of the principal inertial shaft; Step 2: Perform spatial pose representation of the principal axis of inertia, selecting the principal axis offset vector and principal axis tilt vector for analysis; i The pose of the primary rotor's inertial spindle in the rotary coordinate system is as follows:

[0019] in, and These are the principal axis offset vector and the principal axis tilt vector, respectively. After the multi-stage rotors are assembled, the overall inertial spindle pose is:

[0020] in, For the first i The pose transfer matrix of the stage rotor is derived from the transfer model; Step 3: Establish the relationship between reliability and spatial pose of the inertial principal axis. The reliability of the spatial pose of the inertial principal axis is defined as the probability that both coaxiality and tilt angle meet the design requirements.

[0021] in, This is the coaxiality function, representing the maximum offset of the overall principal axis of inertia; is the tilt angle function, representing the maximum tilt angle of the overall principal axis of inertia; and These represent the system's maximum allowable offset and maximum tilt angle, respectively. Step 4: Perform reliability allocation; 1. Decompose the system reliability index, and let the target reliability of the system be... It needs to be allocated to m The key regulatory link, the first j The reliability of each link is Assuming that the various control links are connected in series, then we have:

[0022] Must meet:

[0023] 2. Perform dynamic weight allocation and introduce dynamic weight factors. The weights are adjusted based on the measurement data, and the calculation formula is as follows:

[0024] in, This is the sensitivity coefficient, used to control the degree of influence of error on the weights. Its specific value is determined through simulation experiments, and it needs to be selected so that the weights can effectively distinguish different error levels. for t Time of the first j The current pose error of each stage is calculated using the following formula:

[0025] After adjustment j The reliability of each link is:

[0026] 3. Perform adaptive reliability reallocation: Based on real-time data, adaptively reallocate the reliability of each stage and unify the reliability data to... within, no. j The reliability of each link is:

[0027] in, This represents the average pose error; The adjustment coefficient has a range of values. This is used to control the adjustment range and avoid violent oscillations; Step 5: Perform multi-level rotor inertial spindle spatial pose control, design a deep reinforcement learning agent, and learn the optimal control strategy through interaction with the environment; 1. Construct the state space. Including time t Overall posture System reliability Positional errors at each stage and dynamic weights The details are as follows:

[0028] 2. Construct the action space. Output phase of each stage of rotor assembly Adjustment amount To achieve optimized indexing and assembly, the following details are provided:

[0029] 3. Set the reward function. It is responsible for guiding the agent's learning, improving its reliability, reducing weighted pose error, avoiding excessive control actions, and ensuring a smooth control process, as detailed below:

[0030] in, , , Adjusting the coefficients of the reward function to balance the various weights; 4. Design a deep reinforcement learning network using the Actor-Critic framework, where the Actor network outputs actions based on the state, and the Critic network evaluates the value of the actions; the gradient ascent method is used to update the parameters of the Actor network. 5. Training and optimization are conducted, continuously learning and adjusting error parameters through interaction with the environment to minimize error. A two-stage training approach is adopted: first, basic training, followed by optimization. 6. Implement dynamic regulation; (1) The spatial pose data of the rotor inertial spindle is acquired in real time by the sensor, and the relevant parameters are updated synchronously; (2) The deep reinforcement learning agent calculates the optimal control action, i.e., the phase adjustment amount, based on the current state and sends the control command to the physical actuator, such as a precision rotating platform; (3) Realize dynamic weight update and strategy optimization. Update the weight in real time according to the pose error of the rotor inertial spindle. Continuously optimize the strategy using the updated data to adapt to changes in rotor state. 7. Output the optimal control result.

[0031] 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 substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A multi-stage rotor inertial spindle spatial pose control method based on reliability allocation, characterized in that, The specific steps include: Step 1: Set up the measurement system, measure and calculate relevant parameters; Step 2: Perform inertial principal axis spatial pose characterization; Step 3: Establish the relationship between reliability and inertial principal axis spatial pose; Step 4: Perform reliability allocation; Step 5: Perform multi-stage rotor inertial spindle spatial pose control.

2. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 1, characterized in that, In step 2, two vectors, principal axis offset and principal axis tilt, are selected to characterize the spatial pose of the rotor's principal axis.

3. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 1, characterized in that, In step 3, the spatial pose reliability of the inertial principal axis is defined as the probability that both coaxiality and tilt angle meet the design requirements.

4. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 1, characterized in that, Step 4 employs a reliability allocation method that considers dynamic adaptive weights. The specific steps include: Step 401: Decompose system reliability indicators; Step 402: Perform dynamic weight allocation; introduce dynamic weight factors and adjust the reliability allocation weights for real-time measurement data; Step 403: Perform adaptive reliability weight allocation. Based on real-time data, adaptively redistribute the reliability of each stage and unify the reliability data to... Inside.

5. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 1, characterized in that, Step 5 includes the following steps: Step 501: Construct the state space. Including time t Overall posture System reliability Positional errors at each stage and dynamic weights ; Step 502: Construct the action space. Output phase of each stage of rotor assembly Adjustment amount This enables optimized indexing and assembly. Step 503: Set the reward function. It is responsible for guiding the intelligent agent to learn, thereby improving its reliability, reducing weighted pose error, avoiding excessive control actions, and ensuring a smooth control process. Step 504: Design a deep reinforcement learning network. The network is designed using the Actor-Critic framework, where the Actor network outputs actions based on the state, the Critic network evaluates the value of the actions, and the gradient ascent method is used to update the parameters of the Actor network. Step 505: Conduct training and optimization. By interacting with the environment, continuously learn and adjust error parameters to minimize the error. Step 506: Perform dynamic control by combining the physical entities of the multi-stage rotors with the trained virtual model to form a closed-loop control.

6. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 5, characterized in that, Step 505 employs a phased training strategy. First, basic training is conducted by setting the number of training rounds, the maximum number of steps per round, and the learning rate parameters. Then, the optimization process continues based on the results of the basic training to reduce noise.

7. The multi-stage rotor inertial spindle spatial pose control method based on reliability allocation according to claim 5, characterized in that, Step 506 specifically includes: Step A: Acquire the spatial pose data of the rotor inertial spindle in real time through sensors and update relevant parameters synchronously; Step B, Decision-making and Execution. The deep reinforcement learning agent calculates the optimal control action, i.e., the phase adjustment amount, based on the current state, and sends the control command to the physical actuator; Step C: Implement dynamic weight updates and strategy optimization. Update the weights in real time based on the pose error of the rotor's inertial spindle. Continuously optimize the strategy using the updated data to adapt to changes in rotor state.

Images