基于集成学习的空间驱动部件传动误差预测

In the multi-stage assembly process of space drive components, geometric errors and assembly deformation errors are transmitted in a coupled manner, and traditional accuracy analysis methods based on rigid assumptions are rendered inapplicable. Meanwhile, low computational efficiency is exhibited by the finite element analysis method. A high-precision prediction method for transmission errors of space drive components based on Stacking ensemble learning is presented to solve the above problem. Finite element models of space drive components with different assembly errors are established, and the motion processes of drive components are simulated to obtain transmission error curves. Four key factors, including cam eccentricity, cam tilt and sun gear eccentricity are screened out from eight assembly error parameters by means of orthogonal tests and analysis of variance. A Stacking ensemble learning model is constructed with support vector regression(SVR), XGBoost, random forest(RF) and K-nearest neighbors(KNN) as base learners, XGBoost as the meta-learner. Datasets are obtained by Latin hypercube sampling and applied to model training. Experimental results show that the coefficient of determination(R²) of the Stacking ensemble model is increased by 14. 3% and the root mean square error (RMSE) is decreased by 27. 4% compared with the optimal single model. The prediction accuracy and robustness of transmission errors for space drive components are significantly improved, which lays a foundation for the optimization of transmission performance and the improvement of motion accuracy.