Improving the performance of cable-driven parallel robots through control, modeling, and calibration

Abstract

Cable-driven parallel robots (CDPRs) are a class of robots offering a unique combination of properties: a large workspace, high dynamics, and high payloads. This enables them to be used in various applications where other types of robots are unsuitable. Unfortunately, the potential of CDPRs is hindered by a gap between their theoretical capabilities and their performance in practice. The models used to describe and control. CDPRs often cannot capture their true behavior, due to errors in their assumptions and structure or inaccuracies in their parameters. This reduces the accuracy of CDPRs and impedes their ability to access their full theoretical workspace in practice. To address these problems, this thesis introduces new control, modeling, and calibration approaches for CDPRs, which are aimed at closing the gap between theory and practice. Two new force control methods are developed, which keep the cable forces feasible and close to a specified level. Experiments on two CDPRs show that they can access a similar workspace as predicted in theory. Compared to a purely kinematic method they achieve a 300 % increase in accessible workspace volume. On dynamic trajectories, the new methods exhibit a similar performance as a state-of-the art model predictive control approach, with 79 % less computational costs. A catenary-pulley model for CDPRs is introduced as the combination of two widely used models, accounting for the cables’ sagging due to their weight and the pulleys that guide them. The model increases the accessible workspace volume of a suitable CDPR by 132.7 % and predicts a 17 % lower platform stiffness than an elastic-pulley model. This highlights the importance of using a suitable model for a given CDPR. A unified framework for CDPR models is developed, which enables their optimization and comparison based on measurement data. It can be used to determine the best CDPR model for a given application in terms of accuracy, computational costs, and available measurement data. The framework’s capabilities are demonstrated on a large-scale CDPR for which eleven different models are compared based on two distinct datasets.