Machine Learning Based Kinematic Modeling and Performance Optimization of Flexible Mechanisms

Abstract

Flexible mechanisms are widely applied in precision actuation and micro-manipulation due to their high accuracy and compact design, but their nonlinear large-deformation characteristics make kinematic modeling and performance optimization highly challenging. In this study, a machine learning–based framework is proposed to address these issues. A dataset of 5000 input–output samples was generated through finite element simulations and experimental measurements, and two predictive models—support vector regression (SVR) and deep neural networks (DNN)—were developed to approximate the nonlinear mapping between input forces and output displacements. Bayesian optimization was further integrated to search for optimal structural parameters. Results demonstrate that the proposed method achieves an average displacement prediction error below 0.02 mm, representing a significant improvement compared with analytical models (0.08 mm) and finite element extrapolation (0.05 mm). After optimization, the driving efficiency increased by 24% and system stability improved by 30%, confirming the effectiveness of combining machine learning with intelligent optimization. These findings highlight the potential of data-driven methods for advancing the intelligent design of flexible mechanisms and provide a scientific basis for future applications in precision positioning, biomedical devices, and micro-robotics.