Optimization method and device for transmission error of space driving component

Abstract

The invention provides an optimization method and device for a transmission error of a space driving part, and belongs to the technical field of crossing of mechanical transmission and intelligent optimization. The scheme aims to solve the problems of inaccurate transmission error prediction and poor optimization effect caused by the fact that a space complex service environment is not fully considered in the prior art. The method in the scheme comprises the following steps: establishing a multi-physics coupling service environment model comprehensively considering microgravity, alternating temperature and vibration load; on the basis of the model, transmission error peak values under different tooth profile modification parameter combinations are obtained through dynamic simulation, so that a training data set is generated; training a proxy model by adopting the data set so as to establish a nonlinear mapping relationship between the modification parameter and the transmission error peak value; and with minimization of a transmission error peak value as a target, performing global search on the trained proxy model by adopting a global optimization algorithm to obtain an optimal tooth profile modification parameter combination. The invention further provides a corresponding optimization device. According to the method, the real service environment model is constructed and an intelligent optimization means is adopted, so that the prediction precision and the optimization effectiveness are improved, the optimization efficiency is improved, and the reliability of the model is enhanced.

Description

Technical Field

[0001] This application relates to the field of mechanical transmission technology, and in particular to a method and apparatus for optimizing the transmission error of a spatial drive component. Background Technology

[0002] Space propulsion components, such as planetary harmonic composite reducers used in critical equipment like space station robotic arms and satellite antenna pointing mechanisms, directly impact the on-orbit performance and service life of spacecraft. During on-orbit operation, spacecraft face a complex service environment involving microgravity, significant temperature cycles between sunlight and shadow, and multi-physics coupling such as structural vibrations. These factors can cause unexpected deformation and meshing interference in transmission systems designed for normal temperature and pressure conditions on Earth, leading to a significant amplification of transmission errors and affecting their on-orbit reliability.

[0003] Existing technical solutions typically base their simulation models on ideal or simplified ground conditions, leading to significant discrepancies between the predicted transmission errors and actual on-orbit conditions. Consequently, the optimized parameters obtained are not optimal solutions for real service environments, thus failing to fundamentally address the technical problem of on-orbit performance degradation of space drive components. Summary of the Invention

[0004] The purpose of this application is to provide a method and apparatus for optimizing the transmission error of drive components for space service environments, aiming to solve the technical problems of inaccurate transmission error prediction and poor optimization effect caused by the failure to fully consider the complex space service environment in the prior art.

[0005] To achieve the above objectives, this application provides a method for optimizing the transmission error of a space drive component. The method includes the following steps: Step 1: Establishing a multi-physics coupled service environment model for the space drive component, wherein the multi-physics coupled service environment model comprehensively considers the coupling effects of microgravity, alternating temperature, and vibration loads; Step 2: Based on the multi-physics coupled service environment model, obtaining the peak transmission error of the space drive component under different combinations of tooth profile modification parameters through dynamic simulation to generate a training dataset; Step 3: Using the training dataset, training a surrogate model to establish a nonlinear mapping relationship between the tooth profile modification parameters and the peak transmission error; Step 4: Taking the minimization of the peak transmission error as the optimization objective, using a global optimization algorithm to perform a global search on the trained surrogate model to obtain the optimal combination of tooth profile modification parameters.

[0006] Optionally, the step of establishing a multiphysics coupled service environment model specifically includes: applying a constant acceleration load to simulate microgravity, an alternating temperature load to simulate temperature cycling, and a vibration load to simulate on-orbit vibration to the finite element model of the space drive component.

[0007] Optionally, the proxy model is a blending ensemble learning model.

[0008] Furthermore, the blending ensemble learning model includes multiple base learners and one meta-learner; the base learners are selected from at least one of random forest, support vector regression, gradient boosting machine and radial basis function network; the meta-learner is a radial basis function network.

[0009] Optionally, the global optimization algorithm is a differential evolution algorithm.

[0010] Optionally, in the step of generating the training dataset, different combinations of tooth profile shaping parameters are generated using the Latin hypercube sampling method.

[0011] Optionally, the space drive component is a composite reducer consisting of a planetary reducer and a harmonic reducer; the tooth profile modification parameters include the tooth tip modification amount of the sun gear and planet gears of the planetary reducer, and the rigid gear and flexible gear of the harmonic reducer.

[0012] This application also provides an optimization device for transmission error of a space drive component, comprising: an environment modeling module for establishing a multi-physics coupled service environment model of the space drive component, wherein the multi-physics coupled service environment model comprehensively considers the coupling effects of microgravity, alternating temperature, and vibration load; a data generation module for obtaining the peak value of the transmission error of the space drive component under different combinations of tooth profile modification parameters through dynamic simulation based on the multi-physics coupled service environment model, so as to generate a training dataset; a surrogate model training module for training a surrogate model using the training dataset to establish a nonlinear mapping relationship between the tooth profile modification parameters and the peak value of the transmission error; and a parameter optimization module for performing a global search on the trained surrogate model with the minimization of the peak value of the transmission error as the optimization objective, using a global optimization algorithm to obtain the optimal combination of tooth profile modification parameters.

[0013] Optionally, the environment modeling module is configured to couple a constant acceleration load for simulating microgravity, an alternating temperature load for simulating temperature cycling, and a vibration load for simulating on-orbit vibration onto the finite element model of the space drive component.

[0014] Optionally, the proxy model training module is configured to train a blending ensemble learning model as the proxy model.

[0015] Compared with existing technologies, the technical solution provided in this application has the following beneficial effects: 1. Improved prediction accuracy and optimization effectiveness. By establishing a multiphysics model that couples microgravity, heat, and vibration effects, the on-orbit service environment of space drive components can be simulated more realistically, making the prediction of transmission errors more accurate, thereby ensuring the effectiveness of optimization results in real on-orbit applications. 2. Improved optimization efficiency. Using a high-precision surrogate model to replace the time-consuming finite element simulation for optimization iteration, and combining it with an efficient global optimization algorithm, the optimal tooth profile modification parameters can be found quickly and accurately, greatly shortening the optimization design cycle of space drive components. 3. Enhanced model generalization ability and robustness. By adopting the blending ensemble learning framework, which integrates the advantages of multiple base learners, compared with a single prediction model, it has better generalization performance and robustness to data noise, reducing the risk of overfitting in the surrogate model and ensuring the reliability of the optimization results. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating an overall method for optimizing transmission errors of drive components in a space service environment, as provided in this application embodiment;

[0018] Figure 2 A system functional block diagram of a drive component transmission error optimization device for space service environment provided in this application embodiment;

[0019] Figure 3 A schematic diagram of the finite element model of the space drive component provided in the embodiments of this application;

[0020] Figure 4 A schematic diagram of multiphysics coupling environment modeling provided in the embodiments of this application; Figure 5 This is a schematic diagram of the tooth profile modification method in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of the Blending ensemble learning model in the embodiments of this application; Figure 7 This is a comparison diagram of the transmission error before and after optimization in the embodiments of this application;

[0021] Figure labeling: 1-Finite element model; 2-Microgravity load; 3-Alternating temperature load; 4-Vibration load; 10-Sun gear; 20-Planetary gear; 30-Ring gear; 40-Cage; 50-Wave generator; 60-Flexible gear; 70-Rigid gear; 100-Environmental modeling module; 200-Dynamic simulation module; 300-Surrogate model training module; 400-Parameter optimization module; 310-Base learner; 320-Meta-learner; 330-Meta-feature; 340-Final prediction; S10-Step to establish the environment model; S20-Step to generate the dataset; S30-Step to train the surrogate model; S40-Step to global optimization; S50-Step to output the optimal parameters. Detailed Implementation

[0022] To make the above-mentioned objectives, technical solutions, and advantages of this application more apparent and understandable, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0023] Example 1

[0024] In one embodiment of this application, a method for optimizing the transmission error of drive components for space service environments is provided. This method, through systematic modeling, simulation, learning, and optimization, aims to accurately obtain tooth profile modification parameters that can suppress on-orbit transmission errors.

[0025] Please see Figure 1 This illustrates the overall flow of the optimization method provided in the embodiments of this application. The method mainly includes steps S10 of establishing an environment model, S20 of generating a dataset, S30 of training a proxy model, and S40 of global optimization, and finally outputs the optimal parameters in step S50.

[0026] Accordingly, please refer to Figure 2 This is a schematic diagram of the functional modules of the optimization device for implementing this method. The device includes an environment modeling module 100, a dynamic simulation module 200, a surrogate model training module 300, and a parameter optimization module 400. It should be noted that the functions of these modules are similar to... Figure 1 The methods and steps shown correspond one-to-one and together constitute a collaborative optimization system.

[0027] In this embodiment, the space drive component to be optimized is specifically a planetary harmonic composite reducer. Please refer to [link / reference]. Figure 3This is a schematic diagram of the finite element model of the composite reducer. The model can be built in commercial finite element analysis software (such as ANSYS), using fine meshing to accurately describe the reducer's geometry and material properties. The model mainly includes a planetary reducer and a harmonic reducer. The planetary reducer includes a sun gear 10, multiple planet gears 20, a ring gear 30, and a cage 40 supporting the planet gears 20. The harmonic reducer includes a wave generator 50 as the input, a flexible wheel 60 that generates elastic deformation, and a rigid wheel 70 as the output. This high-fidelity finite element model 1 is the basis for all subsequent analyses.

[0028] Specifically, the first step is to establish an environmental model, S10, which can be implemented by the environmental modeling module 100. The core of this step is to construct a multi-physics coupled service environment model that can realistically reflect the spacecraft's on-orbit operating environment. (See also...) Figure 4 The figure schematically illustrates the process of coupling multiple physical loads onto finite element model 1. This step specifically includes:

[0029] 1. Applying a microgravity load: To simulate the effect of the microgravity environment on the structure, a constant acceleration load of 9.8 × 10⁻³ m / s² is applied in the transient structural analysis module of the finite element model along a specific direction (e.g., perpendicular to the output axis of the reducer). It is understandable that although this load is numerically small, for a high-precision transmission system, the resulting minute deformation and center of gravity shift may still have a non-negligible impact on the meshing state.

[0030] 2. Applying an alternating temperature load 3: To simulate the drastic temperature changes experienced by the spacecraft as it travels between sunny and shadowed areas, an alternating temperature load is applied to finite element model 1. Specifically, by setting a third type of thermal boundary condition (i.e., convective heat transfer), the convective heat transfer coefficient between the reducer shell surface and the external space environment is defined, for example, 50 W / (mm²·℃). Subsequently, based on typical on-orbit temperature data, a periodically varying temperature load spectrum from -150℃ to +150℃ is applied. Through thermo-structural coupling analysis, the thermal stress and thermal deformation caused by temperature changes can be calculated, which are the main causes of gear meshing clearance changes and thermally induced errors.

[0031] 3. Applying a Vibration Load 4: To simulate the impact of structural vibrations generated by spacecraft platforms (such as solar panels or large antennas) under attitude adjustments or external disturbances on the transmission system, a complex vibration load is applied to the mounting base of the reducer through displacement loading. In this embodiment, this load is simplified to a sinusoidal vibration load with a frequency range of 0.1 Hz to 100 Hz and an amplitude of ±0.1 mm. This vibration load will cause dynamic responses in the internal components of the transmission system, potentially leading to impacts between the tooth surfaces and meshing instability.

[0032] The environmental modeling module 100 simultaneously couples the microgravity load 2, alternating temperature load 3, and vibration load 4 to the finite element model 1, thereby establishing a multi-physics coupled environment model that can reflect the actual service conditions.

[0033] Next, the process proceeds to step S20, which is executed by the dynamic simulation module 200. This step aims to obtain the dataset needed for subsequent surrogate model training through dynamic simulation, where the optimized variable is the gear tooth profile modification parameters. (See also...) Figure 5 This is a schematic diagram of a tooth profile modification method. Tooth profile modification improves meshing performance by altering the geometry of the tooth profile to compensate for tooth deformation under load. Key modification parameters include, but are not limited to, the maximum modification amount Ca and the modification length La. In this embodiment, the sun gear 10 and planet gears 20 of the planetary reducer, as well as the rigid gear 70 and flexible gear 60 of the harmonic reducer, are selected for tooth tip straight-line modification. The modification amount is controlled by four independent parameters (Δ1, Δ2, Δ3, Δ4). These four parameters constitute the optimization design space of this method.

[0034] To efficiently collect samples within the design space, this embodiment employs the Latin hypercube sampling method. Compared to grid sampling or random sampling, Latin hypercube sampling can achieve more uniform coverage of the multidimensional parameter space with fewer sample points. The dynamic simulation module 200 uses this method to generate 30 different parameter combinations (Δ1, Δ2, Δ3, Δ4) within a preset parameter range (e.g., the value of each Δ ranges from 0 to 20 micrometers).

[0035] For the 30 sets of parameter combinations generated above, the dynamic simulation module 200 performs a complete dynamic simulation for each set. Specifically, the modified parameters are applied to the corresponding gears in the finite element model 1, and then the entire process of the reducer from startup to stable operation is simulated under the multiphysics coupling environment established in step S10. During the simulation, the theoretical rotation angle of the input end (e.g., sun gear 10) and the actual rotation angle of the output end (e.g., rigid gear 70) are recorded in real time, and the difference between the two is the transmission error. After the simulation, a transmission error curve that varies with time or rotation angle is obtained. The maximum value of the peak-to-peak value or absolute value is extracted from this curve as the "peak value of transmission error" under that set of modified parameters.

[0036] After repeating this process 30 times, the dynamic simulation module 200 generates a training dataset containing 30 samples. Each sample is a data pair in the form of ([Δ1, Δ2, Δ3, Δ4], [peak value of transmission error]), where the former is the input feature of the model and the latter is the output label of the model.

[0037] Next, step S30, training the surrogate model, is executed. This step is completed by the surrogate model training module 300. Considering that each dynamic simulation is time-consuming, it is difficult to directly call the simulation as the evaluation function in the optimization algorithm. Therefore, a surrogate model needs to be trained to replace the time-consuming physical simulation with fast mathematical calculations.

[0038] In this embodiment, the proxy model employs a blending ensemble learning model. Please refer to [link / reference]. Figure 6 This is a schematic diagram of the Blending ensemble learning model. The model has a two-stage hierarchical structure, designed to combine the advantages of multiple different models to achieve higher prediction accuracy and better generalization ability than a single model.

[0039] In the first stage, the surrogate model training module 300 selects multiple different types of machine learning algorithms as base learners 310. For example, in this embodiment, a random forest model, a support vector regression model, a gradient boosting machine model, and a radial basis function network model are selected. The surrogate model training module 300 divides the training dataset generated in step S20 into K-fold cross-validation (e.g., K=5). For each fold, the four base learners 310 are trained on K-1 parts of the data and make predictions on the remaining 1 part of the data. The prediction results of all folds are aggregated to form new features, namely meta-features 330.

[0040] In the second stage, the surrogate model training module 300 uses a meta-learner 320 to train the meta-features 330 generated in the first stage. The role of the meta-learner 320 is to learn how to optimally combine the prediction results of each base learner 310. In this embodiment, the meta-learner 320 uses a radial basis function network model. The meta-learner 320 is trained with the meta-features 330 as input and the actual transmission error peak value as output. After training, the entire blending ensemble learning model is established. That is, the model can receive a new set of tooth profile modification parameters (Δ1, Δ2, Δ3, Δ4) and quickly output a high-precision transmission error peak value 340.

[0041] Finally, global optimization is performed in step S40, which is implemented by the parameter optimization module 400. The goal of the optimization is to find a set of tooth profile modification parameters that minimize the peak value of the transmission error, which is essentially a global optimization problem.

[0042] In this embodiment, the differential evolution algorithm is used for global optimization. This algorithm is a heuristic search algorithm based on swarm intelligence, and it is widely used due to its advantages such as simple implementation, fast convergence speed, and strong robustness. The parameter optimization module 400 first initializes a population containing several individuals (each individual is a set of candidate shaping parameters). In each iteration, the algorithm generates a new generation of a better population by performing mutation, crossover, and selection operations on the individuals in the population.

[0043] The key to this process lies in how to evaluate the "fitness" of each individual (i.e., its corresponding peak transmission error). The parameter optimization module 400 uses the blending surrogate model trained in step S30 as the fitness function. When the differential evolution algorithm generates a new candidate parameter combination, there is no need to run time-consuming dynamic simulations. Instead, the parameter combination is directly input into the surrogate model, which can return a predicted peak transmission error within milliseconds.

[0044] Through numerous rapid iterations, the differential evolution algorithm can efficiently perform a global search across the entire four-dimensional parameter space. When the termination condition is met (e.g., reaching the maximum number of iterations or the error no longer decreasing significantly), the algorithm stops searching and outputs the currently found individual with the best fitness in step S50, i.e., the optimal combination of tooth profile modification parameters (Δ1, Δ2, Δ3, Δ4).

[0045] To verify the optimization effect, the obtained optimal parameter combination can be applied to finite element model 1, and a complete multiphysics coupled dynamic simulation can be performed again. Please refer to [link / reference]. Figure 7The figure shows a comparison of transmission errors before and after optimization. Curve A represents the transmission error of the original design before optimization (e.g., unmodified or empirically modified), which fluctuates wildly and has a high peak value. Curve B represents the transmission error after optimization using the method of this embodiment, showing that both the fluctuation amplitude and peak value have been significantly suppressed. This result verifies the effectiveness of the method provided in this application embodiment, namely, that in complex space service environments, a systematic optimization process can significantly improve the transmission accuracy of drive components.

[0046] Example 2

[0047] This embodiment is an optional implementation of Embodiment 1, intended to illustrate that the optimization framework proposed in this application has good model compatibility, and its core idea is not limited to a specific type of proxy model.

[0048] In this embodiment, the implementation of the environmental model establishment step S10, the dataset generation step S20, and the global optimization step S40 is basically the same as in Embodiment 1. Specifically, the environmental modeling module 100 also establishes a multiphysics coupling model considering microgravity, alternating temperature, and vibration; the dynamic simulation module 200 also uses Latin hypercube sampling to generate 30 sets of sample data; and the parameter optimization module 400 also uses the differential evolution algorithm for global search.

[0049] The core difference in this embodiment lies in the surrogate model training step S30. The surrogate model training module 300 no longer uses a blending ensemble learning model, but instead employs another deep learning model—a gated recurrent unit neural network (GRN)—as the surrogate model. GRN is a variant of recurrent neural networks that addresses long-term dependency issues by introducing update and reset gates, demonstrating excellent performance in handling sequential data and nonlinear mapping problems.

[0050] Specifically, the surrogate model training module 300 constructs a neural network structure comprising an input layer, one or more gated recurrent unit (ROU) layers, and a fully connected output layer. The input layer receives four tooth profile modification parameters (Δ1, Δ2, Δ3, Δ4), and the output layer outputs a predicted peak value of the transmission error. This gated ROU network is then trained using the same training dataset generated by the dynamic simulation module 200 in Example 1. The training process adjusts the weights and biases in the network using a backpropagation algorithm and a gradient descent optimizer (e.g., the Adam optimizer) until the model's loss function converges on the validation set.

[0051] After training, this gated recurrent unit surrogate model can also achieve high-speed and high-precision fitting of the nonlinear relationship between "shaping parameters and error peak values". In the global optimization step S40, the fitness function called by the parameter optimization module 400 is this trained gated recurrent unit network.

[0052] The final optimization result also finds a set of tooth profile modification parameters that significantly reduce transmission errors, with effects comparable to Example 1. This indicates that the overall technical framework of "environmental modeling-data generation-surrogate modeling-global optimization" proposed in this application is open. Those skilled in the art can flexibly select or replace other suitable surrogate models (such as Gaussian process regression, deep neural networks, etc.) according to the characteristics of the specific problem, the data scale, and the available computing resources, and achieve the same expected technical effect.

[0053] Example 3

[0054] Alternatively, in another embodiment, the optimization framework proposed in this application also has good algorithm compatibility, and its core idea is not limited to a specific type of global optimization algorithm.

[0055] In this embodiment, the implementation of steps S10 (establishing the environment model), S20 (generating the dataset), and S30 (training the agent model) is exactly the same as in Embodiment 1. That is, this embodiment also generates the same training dataset based on a multiphysics coupling model that includes microgravity, alternating temperature, and vibration, and trains the same blending ensemble learning agent model as in Embodiment 1.

[0056] The core difference in this embodiment lies in the global optimization step S40. The parameter optimization module 400 no longer uses the differential evolution algorithm, but instead uses the particle swarm optimization algorithm. The particle swarm optimization algorithm originates from the simulation of bird flock foraging behavior. It treats each candidate solution as a "particle" flying in a multidimensional search space, and each particle has its own position (representing a set of shaping parameters) and velocity.

[0057] Specifically, the parameter optimization module 400 first randomly initializes a group of particles in the parameter space. In each iteration, each particle updates its velocity and position based on two factors: first, its own historically found optimal position (i.e., individual optimal); and second, the current optimal position found by all particles in the entire population (i.e., global optimal). When evaluating the merits of each particle's current position, the fitness function called by the parameter optimization module 400 is the Blending proxy model trained in Example 1, which can quickly return the peak transmission error prediction value corresponding to any particle position (i.e., any set of shaping parameters).

[0058] By continuously learning and converging towards individual and global optimal positions, the entire particle swarm will eventually cluster around the global optimal solution. Once the preset number of iterations is reached or the population converges, the algorithm outputs the shaping parameter combination corresponding to the global optimal position.

[0059] The final optimization result can also efficiently find a set of optimal or suboptimal solutions that minimize the peak value of the transmission error, thus verifying the optimization effect. This shows that the technical framework proposed in this application can be seamlessly integrated with a variety of different global optimization algorithms (such as simulated annealing, genetic algorithms, etc.). Those skilled in the art can select the most suitable optimization tool according to the dimension and complexity of the optimization problem, as well as different requirements for convergence speed and global search capability, which enhances the universality and practicality of the method in this application.

[0060] Example 4

[0061] This embodiment aims to illustrate that the method proposed in this application has good modularity and scalability, and can flexibly adjust the complexity of the environment model according to specific application scenarios and needs.

[0062] In certain specific space missions, spacecraft may operate on a very stable orbit with extremely weak platform vibrations that have negligible impact on the propulsion system. However, the orbital characteristics may cause them to undergo extremely severe and frequent temperature cycles. For such applications, the environmental model can be simplified to improve the efficiency of modeling and simulation while ensuring that key influencing factors are considered.

[0063] The core difference in this embodiment lies in the environmental model establishment step S10. When constructing the multiphysics coupled service environment model, the environmental modeling module 100 only establishes a coupled model that comprehensively considers the microgravity load 2 and the alternating temperature load 3, without applying the vibration load 4. In other words, this is a simplified "thermo-mechanical" coupled environment model.

[0064] Based on this simplified environmental model, subsequent steps are performed sequentially. When performing dynamic simulation, the dynamic simulation module 200 only performs calculations under thermo-mechanical coupled loads. This significantly reduces computation time compared to the fully coupled thermo-mechanical-dynamic simulation that includes vibration in Example 1. Similarly, 30 sets of shaping parameters are generated through Latin hypercube sampling, and the simulation is run to obtain a training dataset for this specific working condition.

[0065] Next, the surrogate model training module 300 uses this new dataset and trains it using the same blending ensemble learning model as in Example 1 to obtain a surrogate model specifically for predicting the peak value of transmission error under "thermal-mechanical" coupling conditions.

[0066] Finally, the parameter optimization module 400 also uses the differential evolution algorithm, with this newly trained surrogate model as the fitness function, to perform a global optimization search, and finally obtain the optimal tooth profile modification parameter combination for this specific "thermal-mechanical" coupling condition.

[0067] The results of this embodiment demonstrate that, through modular adjustments to the environmental model, the method of this application can provide customized and precise optimization solutions for service environments of varying complexity. This flexibility makes the method applicable not only to the most complex operating conditions but also capable of completing optimization tasks with higher efficiency under relatively simple conditions, thus possessing significant engineering application value.

[0068] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method of optimizing spatial drive component transmission error, characterized by, The method comprises the following steps: Step 1: establishing a multi-physical field coupling service environment model of the space driving component, wherein the multi-physical field coupling service environment model comprises a microgravity factor, an alternating temperature factor and a coupling effect factor of a vibration load; Step 2: based on the multi-physical field coupling service environment model, obtaining a transmission error peak value of the space driving component under different combinations of tooth profile modification parameters through dynamic simulation, and generating a training data set based on the combinations of tooth profile modification parameters and the transmission error peak value; Step 3: training an agent model by using the training data set to establish a nonlinear mapping relationship between the tooth profile modification parameters and the transmission error peak value; Step 4: taking minimization of the transmission error peak value as an optimization objective, performing global search on the trained agent model by using a global optimization algorithm to obtain an optimal combination of tooth profile modification parameters.

2. The method of claim 1, wherein, The step of establishing the multi-physical field coupling service environment model specifically comprises: On the finite element model of the space driving component, a constant acceleration load for simulating microgravity, an alternating temperature load for simulating temperature cycles and a vibration load for simulating in-orbit vibration are coupled and applied.

3. The method of claim 1, wherein, The agent model is a Blending ensemble learning model.

4. The method of claim 3, wherein, The Blending ensemble learning model comprises a plurality of base learners and a meta-learner; The base learner is at least one of a random forest, a support vector regression, a gradient boosting machine and a radial basis function network; The meta-learner is a radial basis function network.

5. The method of claim 1, wherein, The global optimization algorithm is a differential evolution algorithm.

6. The method of claim 1, wherein, In the step of generating the training data set, different combinations of tooth profile modification parameters are generated by using a Latin hypercube sampling method.

7. The method of claim 1, wherein, The space driving component is a compound reducer composed of a planetary reducer and a harmonic reducer; The tooth profile modification parameters comprise tooth tip modification amounts of a sun gear and a planet gear of the planetary reducer and tooth tip modification amounts of a rigid gear and a flexible gear of the harmonic reducer.

8. An apparatus for optimizing spatial drive component transmission error, characterized by, The method comprises: an environment modeling module configured to establish a multi-physical field coupling service environment model of the space driving component, wherein the multi-physical field coupling service environment model comprises a microgravity factor, an alternating temperature factor and a coupling effect factor of a vibration load; a data generation module configured to, based on the multi-physical field coupling service environment model, obtain a transmission error peak value of the space driving component under different combinations of tooth profile modification parameters through dynamic simulation, and generate a training data set based on the combinations of tooth profile modification parameters and the transmission error peak value; an agent model training module configured to train an agent model by using the training data set to establish a nonlinear mapping relationship between the tooth profile modification parameters and the transmission error peak value; a parameter optimization module configured to take minimization of the transmission error peak value as an optimization objective, perform global search on the trained agent model by using a global optimization algorithm to obtain an optimal combination of tooth profile modification parameters.

9. The apparatus of claim 8, wherein, The environment modeling module is configured to: On a finite element model of the space driving component, a constant acceleration load for simulating microgravity, an alternating temperature load for simulating temperature cycling, and a vibration load for simulating on-orbit vibration are coupled to be applied.

10. The apparatus of claim 8, wherein, The proxy model training module is configured to train a Blending ensemble learning model as the proxy model.

Images