Combustion machine learning: Principles, progress and prospects

• DOI: 10.1016/j.pecs.2022.101010
• Publication Date: 2022-04-28T10:08:09Z
• Authors: Matthias Ihme; Wai Tong Chung; Aashwin Ananda Mi...
• Keywords: Interpretability; Robustness; Benchmark (surveying)

Progress in combustion science and engineering has led to the generation of large amounts data from large-scale simulations, high-resolution experiments, sensors. This corpus offers enormous opportunities for extracting new knowledge insights—if harnessed effectively. Machine learning (ML) techniques have demonstrated remarkable success analytics, thus offering a paradigm data-intense analyses scientific investigations through machine (CombML). While data-driven methods are utilized various areas, recent advances algorithmic developments, accessibility open-source software libraries, availability computational resources, abundance together rendered ML ubiquitous analysis engineering. article examines applications Starting with review sources data, techniques, concepts, we examine supervised, unsupervised, semi-supervised methods. Various examples considered illustrate evaluate these Next, past approaches problems combustion, spanning fundamental investigations, propulsion energy-conversion systems, fire explosion hazards. Challenges unique CombML discussed further identified, focusing on interpretability, uncertainty quantification, robustness, consistency, creation curation benchmark augmentation prior combustion-domain knowledge.