Abstract
To optimize control and/or state estimation of industrial processes, information about measurement quality provided by sensors is required. In this paper, a probabilistic scheme is proposed in discrete-time nonlinear state space with the purpose of sensor monitoring. A quantitative index representing the measurement quality, as well as satisfied state estimates, is obtained by estimating the probability density functions (PDFs) of the states and the measurement noise covariance considered as a random variable using the variational Bayesian approach. To solve the intractable integrals of nonlinear PDFs in real time, a set of weighted particles is generated to overlap an empirical density of state, while the PDF of the measurement noise is still derived analytically. An example of localization and an experiment with a rotary flexible joint are supplied to demonstrate that the proposed algorithm significantly improves the applicability of existing methods and can monitor correlated sensors satisfactorily.
Original language | English |
---|---|
Article number | 8678661 |
Pages (from-to) | 2294-2303 |
Number of pages | 10 |
Journal | IEEE Transactions on Industrial Electronics |
Volume | 67 |
Issue number | 3 |
DOIs | |
Publication status | Published - 2020 Mar |
Bibliographical note
Publisher Copyright:© 1982-2012 IEEE.
Keywords
- Nonlinear process
- particle approximation
- sensor monitoring
- state estimation
- variational Bayesian (VB) inference
ASJC Scopus subject areas
- Control and Systems Engineering
- Electrical and Electronic Engineering