An adaptive assembly system for automotive applications.

Abstract

Contemporary automatic assembly systems used in the automotive industry are implemented using a deterministic task approach. The throughput of such systems is controlled mainly by the quality of the piece parts and the tooling. Wear and uncertain part geometry are the major sources disturbing the process, and causing dimensional variations. As a solution to these dimensional errors an adaptive system is developed--an active system, where the process state is monitored by a number of sensors and the measured information is used to adjust the process on-line. The major subsystems are: the non-contact (vision) measuring system, the fixturing setup, the parallel manipulator, and the distributed control system. The proposed approach depends partially on a priori error control through manipulator and sensor calibration. A parallel manipulator is designed and its repeatability assessed. The analytical solution of the inverse kinematics is derived, followed by numerical solution to the forward kinematics problem. A new method of mapping the workspace boundaries is proposed. Two original methods for obtaining repeatability maps are introduced and compared, indicating the capability of the parallel manipulator to achieve high repeatability; these results are then verified experimentally. An approach to parallel manipulator calibration is discussed. The calibration method is based on a non-parametric error representation. With calibration points for the experimental procedure being generated using the optimal design of statistical experiments, the error models are identified as the first order polynomials. The results from implementation of the error compensation scheme are also presented. Visual on-line sensing uses visual fixturing approach during which selected features on both parts are measured and then least-square fit to their nominal design data. The net effects of parts misalignment are then accommodated by recalculation of the assembly trajectory, fed into the parallel manipulator for every set of parts. Performance of the system was evaluated experimentally indicating its ability to reduce in-process variation by approximately 50%.