Particle filtering for system identification in civil engineering

Abstract

In structural health monitoring and damage detection, particle filtering can potentially play a crucial role. The problems may be solved by posing them within the framework of revealing accurate statistical information from the imprecisely known model parameters and dynamical systems fed with experimentally observed noisy data. While it appears to be lucrative, weight-based particle filters typically experience severe performance failure. The major numerical bottleneck for such underperformance especially for higher dimensional systems happens to be the progressive particle impoverishment owing to weight collapse. In this chapter, we show that such difficulty can be conveniently bypassed by applying additive gain-type nonlinear particle filters. Specifically, we discuss ensemble Kalman filter (EnKF) and the ensemble Kushner- Stratonovich filter (EnKS). The discussion on EnKS encompasses both its non-iterative and iterative forms. We also numerically show that, in the identification of nonlinear and large-dimensional dynamical systems, a substantively superior performance of the non-iterative version of the EnKS may be observed vis-à-vis most existing filters. The costlier iterative version, though conceptually elegant, mostly appears to incorporate a marginal improvement in the reconstruction accuracy over its non-iterative counterpart. © 2022 World Scientific Publishing Company. All rights reserved.