Detection of flame wrinkling in spherically expanding flames using time series classification with pressure traces

Abstract

Intrinsic instabilities in spherically expanding flames often manifest as cellular structures on the flame surface. In the case of spherically expanding flames, they have been extensively studied visually using high speed cameras in conjunction with optical diagnostic techniques. While pressure traces of the combustion event are often used to characterize non-intrinsic flame instabilities (e.g., thermoacoustic), they are significantly less explored with respect to intrinsic instabilities for spherical flames. This work investigates the potential for the detection of wrinkling of spherically expanding flames using time series classification (TSC) with pressure traces. The data set contained 386 time series samples (pressure traces) that were labeled based on the associated high-speed images. The dataset contained flames from mixtures that vary in terms of fuels, working fluids, initial pressures, and equivalence ratios. A high classification accuracy (> 93%) was obtained with an ensemble machine learning algorithm for TSC, though a preliminary exploration did not yield any clear classification features and indicates the need for further exploration.