Accuracy and Efficiency Comparison of Various Nonlinear Kalman Filters Applied to Multibody Models

Abstract

[Abstract] Multibody simulations are already used in many industries to speed up the development of new products. However, improvements in multibody formulations and the continuous increase in inexpensive computational power open new fields of applications for multibody simulations, such as using them as plant models for state observers. The present work introduces a novel state observer developed by combining a multibody model with an indirect Kalman filter. Together with other multibody-based Kalman filters already proposed in the literature, they have been applied to two mechanisms (four- and five-bar linkages) to assess their performance. The accuracy of the estimations and the computational cost are examined under several scenarios: using position or velocity sensors in different configurations, with different sampling rates, and considering different levels of errors in the models assumed by the filters. The aim of this work is to present different options to build a state observer based on a dynamic multibody model, and to provide guidance to select the most suitable option for future applications. All the methods have been implemented as a reusable MATLAB toolkit which has been released as Open Source in https://github.com/MBDS/mbde-matlab.