Rangaraj Pandurangan
MS (2013)
Polynomial chaos in bootstrap particle filtering for system identification  Downalod

This study focuses on the use of particle filters, more specifically the bootstrap particle filter, for identification of the system parameters of dynamical systems from measurement data. The underlying principle of the bootstrap particle filters is based on Bayesian framework. The implementation of the bootstrap particle filter involves solving the forward problem a large number of times using Monte Carlo simulations. This is computationally intensive, especially in dealing with com- plex systems where the solution of a single forward problem requires significant com- putational time. The present study focuses on the development of a new approach of coupling the bootstrap particle filter with Polynomial Chaos Expansion to reduce the computational effort. Polynomial Chaos based methods are spectral uncertainty quantification tools based on projecting the uncertain parameters along random basis functions, which are in the form of polynomials. The central idea of the proposed method lies in projecting the forward problem into a space spanned by orthogonal functions and performing filtering in this space. This ensures an accelerated solution of the forward problem, requiring significantly less computational efforts. Moreover, performing the filtering in the random space bypasses the requirement of constructing the polyno- mial chaos expansion at later time steps in the identification algorithm. This further reduces the required computational effort.

Co-guide: Prof Abhijit Chaudhuri, Applied Mechanics
Placement: General Electric, Bangalore.
Present: Benz,  Bangalore