Parameter identification and state estimation of multibody systems – bringing and keeping simulation models closer to the reality

Abstract

Multibody models have proven their effectiveness in the virtual prototyping and testing of mechanical systems. However, increasing computational power and advancements in simulation techniques make it possible to evaluate models in real time also in embedded system applications. This new capability enables the models to mirror real system behaviors, however, implementation demands new levels of model accuracy and efficiency.

The aim of this thesis is to address these new requirements to strengthen the connection between a multibody model and its physical counterpart using two different approaches: Parameter identification to increase the accuracy of models and state estimation to constantly update the model with acquired measurements from the real system.

The underlying study advances estimation methods. Because of its relatively low computational burden, the work focused on the discrete extended Kalman filter. The objective was to enhance its accuracy to reach levels demonstrated by more advanced and slower estimator algorithms. A kinematic version of the estimator was applied to hydraulic machinery. For model identification, the research adopted linear regression to identify the complete set of model inertial parameters.

The new developments in discrete extended Kalman filtering were d emonstrated using a numerical four-bar mechanism example. In comparison, the proposed method outperformed the traditional forward Euler integration scheme. The kinematic Kalman filter approach was tested with a simple hydraulic crane setup and shown to give good estimates of crane kinematics. The identification algorithm was tested against both the simulationbased synthetic data of a slider-crank mechanism and experimental data of a comparable setup demonstrating the algorithm to be capable of accurately predicting the torque response and parameter values of the system. These advancements facilitate the tuning of multibody models for industrial equipment and enable integration of the estimation algorithms into embedded system applications.