Study on the contact kinematics and optimization of recirculating planetary roller screw mechanism based on Bayesian learning model

The recirculating planetary roller screw mechanism (RPRSM) delivers ultra-precise transmission performance but still lacks theoretical studies on its kinematics and multi-body contact behavior. To address this, a contact kinematics model is developed to characterize interactions among the grooved roller, screw, nut, and cam ring. Moreover, a Bayesian learning model built upon variational inference and active learning Kriging (VI-ALK) strategy is introduced as an AI-driven approach to enable efficient adaptive surrogate modeling and optimization. The roller's recirculating mechanism and the effects of parameters on contact kinematics are thoroughly investigated. Results demonstrate that rolling-sliding contact occurs at the interfaces, with trajectory deviations effectively corrected via the cam ring. Furthermore, the proposed multi-objective optimization model significantly reduces stresses while precisely constraining contact positions.