Computational tools for data-poor problems in turbulent combustion

Abstract

Computational modeling is undergoing a revolution due to the influx of data. In the field of propulsion and energy conversion systems, data-driven modeling is beginning to emerge as a new discipline, expanding the use of tools developed for uncertainty quantification. For turbulent combustion applications, the utility of data is determined by the specifics of the problem and the nature of quantities extracted from the simulations. In this regard, three different classes of problems can be identified, namely, data-rich, data-sufficient and datapoor problems. While data-rich problems are easily amenable to the machine learning tools emerging in other contexts, and data-sufficient problems form the overwhelming focus of computational modeling, there has been very limited focus on data-poor problems. This class of modeling problems is marked by the sparsity of observation data, where due to the nature of the physics, established experimental techniques are not readily applicable. In the discussion here, the source of this sparsity in data, and the nature of problems that give rise to such data-poor environments is discussed. One important class of problems is the occurrence of extreme or anomalous events that can lead to catastrophic failure of the system. High-altitude relight, flame flashback or engine unstart fall in this problem set. While these events all fall in the broad category of transient processes, it is shown that the causal mechanism can be fundamentally different. To develop the scientific inference process, a classification of such problems is used to determine specific modeling paths as well as computational tools needed. Research opportunities for the emerging field of extreme event prediction are highlighted in order to identify critical and immediate needs.