Generating planar distributions of soot particles from luminosity images in turbulent flames using deep learning

We report a computational method based on deep learning (DL) to generate planar distributions of soot particles in turbulent flames from line-of-sight luminosity images. A conditional generative adversarial network (C-GAN) was trained using flame luminosity and planar laser-induced incandescence (LII) images simultaneously recorded in a turbulent sooting flame with an exit Reynolds number of 15,000. Such a training built up the underlying relationship between the two types of images i.e., a predictive model which was then used to predict LII images from luminosity images and the accuracy was assessed using four different methods. Results show that the model is effective and capable of generating LII images with acceptable prediction accuracies of around 0.75. The model was also found to be applicable over a range of heights in the flames, as well as for the flames with a range of exit Reynolds numbers spanning from 8000 to 20,000. Besides, the probability density function (PDF) of LII signals in different flames can also be predicated using the model. This work, for the first time, demonstrates the feasibility of predicting planar signals from corresponding line-of-sight signals from turbulent flames, which potentially offers a much simpler optical arrangement for a modest trade-off in accuracy.