An Evaluation of Popular Noise Reduction Techniques for Multi-scale Problems Involving Particle Flow Simulations
Speaker: Malgorzata Zimon, University of Strathclyde
Presentation
Abstract
Particle fluid simulations allow for investigation of phenomena occurring at the atomistic and microscopic level. The information generated by particle methods, e.g. molecular dynamics, or direct simulation Monte Carlo, is converted to macroscopic observables by means of statistical averaging. A significant drawback of nano- or micro-scale modeling is the substantial noise, which disturbs the analysis of the results. The uncertainty in the mean of the ensemble is due to fluctuations caused e.g. by additional forcing terms (thermostats). Proper orthogonal decomposition (POD), widely utilized for turbulent flows, has recently been explored for processing particle data. An extension of the method based on time-windows offers a useful approach for noise reduction in particle simulations. However, to successfully remove statistical noise from the system, large amounts of data need to be provided. Moreover, POD can fail to improve the quality of an ensemble mean (statistical average) when applied to steady-state simulations. In order to achieve a better efficiency of POD in processing non-stationary fields, we have combined the method with wavelet-based filtering. In this new procedure, the wavelet thresholding is performed within POD's domain. In case of stationary problems, we will show how effectively POD can be applied to a matrix constructed from the mean, following the application of singular spectral analysis (SSA). The combination of POD and SSA is shown to successfully smooth time-dependent observables. Investigation of another method, referred to as empirical mode decomposition, gives additional insight into possible improvement of filtering procedures for multi-scale applications. Simulations were undertaken to illustrate the performance of the new tools applied to noisy velocity and density fields. Numerical examples include molecular dynamics, dissipative particle dynamics simulations, and direct simulation Monte Carlo of force-driven fluid flows and phase separation phenomena.
Biography
Malgorzata completed her Master's degree in Mechanical Engineering & Applied Computer Science at the Technical University of Lodz in Poland. She took part in student exchange programmes at University of Strathclyde in Scotland, Osaka University in Japan, and at Manipal University in India. Malgorzata is currently a PhD student carrying out her research in the Department of Mechanical & Aerospace Engineering at the University of Strathclyde jointly with STFC Daresbury Laboratory. She is expected to finalize her current studies in April 2015 and begin working at the School of MACE in Manchester. Malgorzata¹s interests involve investigating de-noising algorithms in computational multi-physics modeling.