An Enhanced Adaptive Kalman Filter for Multibody Model Observation

The topic of state estimation using multibody models combined with Kalman filters has been an active field of research for more than 15 years now. Through state estimation, virtual sensors can be used to increase the knowledge of a system, measuring variables that cannot be obtained through conventional sensors. This is useful for control purposes or updating the state of a digital twin of a system. One of the most tricky questions with the different approaches tested in the literature is the parameter tuning of the filters, in particular, the covariance matrix of the plant noise. This work presents a new method which includes a shaping filter to whiten the plant noise combined with an adaptive algorithm to adjust the plant noise parameters. This new method is tested and compared with methods already described in the literature using the three-simulation method. The new method is at least as accurate as the best hand-tuned filters in most of the situations evaluated, and improves the accuracy of previously presented adaptive methods. All the methods and mechanisms tested in this paper are available in an open source library written in matlab called MBDE.