State Estimation in a Hydraulically Actuated Log Crane Using Unscented Kalman Filter

Multibody system dynamics approaches together with state estimation methods can reduce the need for a large number of sensors, especially in the digital twin of working mobile machinery. To demonstrate this, a hydraulically actuated machine was modeled using the double-step semi-recursive multibody formulation and lumped fluid theory in terms of system independent states. Next, because of the high non-linearity of the modeled system and with respect to the reported performance degradation of the Extended Kalman Filters (EKF), which are mostly related to the linearization procedure of this filter, the Unscented Kalman Filter (UKF) was implemented to achieve high accuracy and performance. The methodology of the proposed approaches was applied to a mobile log crane model PATU 655. The implementation of the proposed estimation algorithms is demonstrated with three different multibody based simulation models: the synthetic real system producing the artificial measurements, the simulation model, and the estimation model. Encoders and pressure sensors, installed on the synthetic real system, provided synthetic sensor measurement data. To mimic real-world conditions, the estimation and simulation models used in the processing of the state estimation algorithm were assumed to have errors in the initial conditions and force model. The proposed UKF was applied to the estimation model with the synthetic sensor measurement data. The minimum percent normalized root mean square errors in the associated measured and unmeasured states of case example were 0.11% and 1.86%, respectively. The UKF using the multibody system dynamics formulations is able to estimate the case example states despite 15% and 60% errors in mass and inertial properties of bodies and Payload, respectively, confirming the accuracy and performance of the algorithm.