Noise, ill-conditioning and sensor placement analysis for force estimation through virtual sensing

Abstract

The knowledge of loads acting on machines and components is crucial in many application fields. Input estimation though is a very challenging inverse problem with which researchers have struggled over the last decades. Several issues related to ill-conditioning and ill-posedness are not sufficiently understood nor solved. This paper proposes an in-depth numerical study covering some of the aspects that can make the difference between a reliable and an unreliable force estimation. A states-input estimation algorithm for multiple force estimation is implemented as a linear augmented Kalman filter/smoother coupled with a reduced order model of a complex ill-posed mechanical structure, namely a twistbeam rear suspension. The influence of different noise levels, measurements scaling and time horizon used for the estimation is thoroughly analyzed. Finally, an observability-based optimal sensors placement strategy is implemented showing robustness improved accuracy of the estimated quantities.