Title: Nonlinear Filter Tutorial, Part 2 (UQ/V&V Seminar)

Speaker: Nick West, Stanford University, ICME (Institute for Computational and Mathematical Engineering)

Date/Time: Tuesday, September 1, 2009, 2:00-3:00 (NM), 1:00-2:00 (CA)         

Location: CSRI Building, Room 90 (Sandia NM), Building 915, Room S145 (CA)           

Brief Abstract: This is the second part of tutorial on filtering methods.  In the previous part of this tutorial, we derived the optimal linear filter (the Kalman filter) and gave several examples of its use.  In this second part we discuss filtering techniques for non-linear systems with non-Gaussian noise terms.  We investigate two distinct types of filters: those that try to estimate the distribution of the state (Bayesian Filters, Exact Non-Linear Filter) and those that try to obtain the best estimate of the state (Maximum Likelihood Estimators, Extended Kalman Filters).  We present a third category of so-called "hybrid" filters which attempt to estimate both distributions and provide a "best estimate" (Ensemble Kalman Filters, Particle Filters).  This talk will be less grounded in the detailed mathematics (than the previous talk) and aims to present some of the "big picture" aspects of each method. 

As a running example we will consider the problem of estimating the state of flow through a hypersonic engine.  This problem is modeled by the 1D compressible Euler equations with a forcing term and our observations are of the state at select locations in the inlet.  We hope to track the shock that develops from fueling the engine.  We will demonstrate the effectiveness of several of the above filters on this problem.

CSRI POC: Laura Swiler, (505) 844-8093