Bayesian Online Changepoint Detection

Changepoints are abrupt variations in the generative parameters of a data sequence. Online detection of changepoints is useful in modelling and prediction of time series in application areas such as finance, biometrics, and robotics. While frequentist methods have yielded online filtering and prediction techniques, most Bayesian papers have focused on the retrospective segmentation problem. Here we examine the case where the model parameters before and after the changepoint are independent and we derive an online algorithm for exact inference of the most recent changepoint. We compute the probability distribution of the length of the current ``run,'' or time since the last changepoint, using a simple message-passing algorithm. Our implementation is highly modular so that the algorithm may be applied to a variety of types of data. We illustrate this modularity by demonstrating the algorithm on three different real-world data sets.

Curvature and wrinkling of premixed flame kernels-comparisons of OH PLIF and DNS data

The effects of curvature and wrinkling on the growth of turbulent premixed flame kernels were studied using both two-dimensional OH Planar Laser-Induced Fluorescence (PLIF) and three-dimensional Direct Numerical Simulation (DNS). Comparisons of results between the two approaches showed a high level of agreement, providing confidence in the simplified chemistry treatment employed in the DNS, and indicating that chemistry might have only a limited influence on the evolution of the freely propagating flame. The usefulness of PLIF in providing data over a wide parameter range was illustrated using statistics obtained from both CH4/air and H2/air mixtures, which show markedly different behavior due to their different thermo-diffusive properties. The results provided a demonstration of the combined power of PLIF and DNS for flame investigation. Each technique compensate for the weaknesses of the other, and to reinforce the strengths of both.

Curvature and wrinkling of premixed flame Kernels comparison between planar laser induced fluorescence (PLIF) and 3D direct numerical simulations (DNS)

A direct comparison between time resolved PLIF measurements of OH and two dimensional slices from a full three dimensional DNS data set of turbulent premixed flame kernels in lean methane/air mixture was presented. The local flame structure and the degree of flame wrinkling were examined in response to differing turbulence intensities and turbulent Reynolds numbers. Simulations were performed using the SEGA DNS code, which is based on the solution of the compressible Navier Stokes, species, and energy equations for a lean hydrocarbon mixture. For the OH PLIF measurements, a cluster of four Nd:YAG laser was fired sequentially at high repetition rates and used to pump a dye laser. The frequency doubled laser beam was formed into a sheet of 40 mm height using a cylindrical telescope. The combination of PLIF and DNS has been demonstrated as a powerful tool for flame analysis. This research will form the basis for the development of sub-grid-scale (SGS) models for LES of lean-premixed combustion systems such as gas turbines. This is an abstract of a paper presented at the 30th International Symposium on Combustion (Chicago, IL 7/25-30/2004).