Force-Pose Nonlinear Mapping Algorithm for Assembling Micro-Mechanical System

Abstract

In order to precisely get the assembly statement when two micro-mechanical parts are touching, assembly methods mapping the nonlinear relationship between micro-force/torque and assembling pose are studied. The research platform is 12 degree-of-freedom (DOF) microassembly system with large-motion-range and precise-motion-accuracy. After assembly mechanism of micro- mechanical system is introduced, assembly methods utilizing both three-dimensional (3D) search algorithm and Back Propagation (BP) neural network are brought forward on the basis of micro-force/torque information feedback. In addition, two different neural network algorithms -- gradient descent and Levenberg-Marquardt (L-M) -- are further applied in the assembly process. Comparison among above algorithms was carried out in experiments, and the results indicated that BP neural networks were more reliable than 3D search algorithm in the relationship mapping process while the convergence rate of L-M algorithm was much faster than that of the gradient descent.