Advanced optimal sensor placement for Kalman-based multiple-input estimation

• Published in: Mechanical systems and signal processing 2021-11, Vol.160, p.107830, Article 107830
• Main Authors: Cumbo, R., Mazzanti, L., Tamarozzi, T., Jiranek, P., Desmet, W., Naets, F.
• Format: Article
• Language: English
• Subjects: Augmented Kalman Filter; Estimator bandwidth; Finite element method; Inverse load identification; Kalman filters; Load; Mechanical components; Observability; Observability (systems); Optimal Sensor Placement; Placement; Sensors; State estimation; Steady-state error covariance
• ISSN: 0888-3270; 1096-1216
• DOI: 10.1016/j.ymssp.2021.107830

•Optimal Sensor Placement in the framework of Augmented Kalman Filter estimation.•Two new metrics are proposed based on error covariance and estimator bandwidth.•The methods are compared with the existing Optimal Sensor Placement strategy.•Numerical validation on an industrial Finite Element Model is shown. The direct measurement of the external loads acting on a mechanical component represents often a challenge in many engineering applications. In this context, the attention of many researchers is focused on the field of inverse load identification. Several techniques are proposed in literature to address this problem by means of experimental methodologies, often coupled with simulation solutions. In this paper, a Kalman-based methodology is considered, which solves the problem of inverse load identification in a predictive manner. The common issue of most of the techniques is the selection of an optimal set of sensors which gives the best load estimation. In the Kalman-filtering framework, an Optimal Sensor Placement (OSP) strategy has been proposed by the authors and it aims to find the best set of sensors in terms of system observability, which is the minimum requirement of a stable estimation. However, this does not guarantee the most accurate load estimation. In this contribution, two alternative metrics are proposed, based on: i) steady-state error covariance of the estimation, ii) estimator bandwidth, with respect to the available set of measurements. Both criteria will be included in the existing OSP to improve the sensors selection. A comparison of the three strategies for multiple input/state estimation is discussed on an industrial-scale Finite Element Model, in order to show the improvement in the accuracy of the estimated quantities.