The distribution similarity measure–based Gaussian mixtures model CS modelling scheme is employed in modelling the captured wrench signals for different CSs. The wrench (Cartesian force and torque) signals of the manipulated object are captured for different states of the given assembly process. This article proposes the distribution similarity measure–based Gaussian mixtures model for the contact-state (CS) modelling in force-guided robotic assembly processes of flexible rubber parts. Classification Success Rate (CSR) and computational time will be considered as comparison indices and the EM-GMM will be shown to have a superior CSR performance with reduced a computational time. Furthermore, we compare the performance of the EM-GMM with that obtained when using available CS monitoring approaches. For both experiments, the excellent monitoring performance will be shown.
Two experiments are considered in the first experiment we use the EM-GMM in monitoring a typical peg-in-hole robotic assembly process and in the second experiment we consider the robotic assembly of camshaft caps assembly of an automotive powertrain and use the EM-GMM in monitoring its CS's. In order to see the performance of the suggested CS monitoring scheme, we installed a test stand that is composed of a KUKA Lightweight Robot (LWR) doing a typical peg-in-hole tasks. The use of the GMM would accommodate the signals non-stationary behavior and the EM algorithm would guarantee the estimation of the optimal parameters set of the GMM for each signal and hence the modeling accuracy would be significantly enhanced. In this article, the concept of the Gaussian Mixtures Models (GMM) is used in building the likelihood of each signal and the Expectation Maximization (EM) algorithm is employed in finding the GMM parameters. they have non-normal distribution that would result in performance degradation if using the available monitoring approaches. It will be shown that the captured signals are non-stationary, i.e. The captured signals are employed in building a model (a recognizer) for each CS and in the framework of pattern classification the CS monitoring would be addressed. In order to perform such a monitoring target, the wrench (Cartesian forces and torques) and pose (Cartesian position and orientation) signals of the manipulated object are firstly captured for different CS's of the object (peg) with respect to the environment including the hole. This article addresses the problem of Contact-State (CS) monitoring for peg-in-hole force-controlled robotic assembly tasks.
0 kommentar(er)
