Biomass fuel identification using flame spectroscopy and tree model algorithms

This paper presents an identification method for types of fuel such as biomass by combining flame spectroscopic monitoring and tree model algorithms. The features of the flame spectra are extracted, including the spectral intensity of flame radicals [OH* (310.85 nm), CN* (390.00 nm), CH* (430.57 nm) and C₂* (515.23 nm, 545.59 nm)], flame radiation intensity and flame radiation energy (integration of spectral intensity). The identification models are built using four tree model algorithms, i.e., decision tree, random forest, extremely randomized trees, and gradient boost decision tree. The different type of biomass and spectra features of combustion flames are composed of sample pairs to train identification models. Experiments are carried out on a laboratory-scale biomass-air combustion test rig. Four different biomass fuels, including corncob, willow, peanut shell, and wheat straw are burnt. The results demonstrate that the identification models proposed is capable of identifying types of biomass fuels correctly with the average identification success rate of 98% in 10 trials.