Method of identifying burning material from its smoke using attenuation of light

In this study, it is verified that several materials can be accurately distinguished from their aerosols or from the smoke they emit when they are burnt individually. This is done by comparisons of transmitted and scattered light at various wavelengths using a Machine Learning Algorithm. Smoke was introduced in the paths of light of different wavelengths, simultaneously. The wavelengths were chosen from widest spectrum of radiation, for which LEDs and photodiodes were available commercially. These include UVC 275 nm, UVA 365 nm, Blue 405 nm, Red 620 nm and IR 960 nm. At least one photodiode was used to sense transmitted and at least one photodiode to sense scattered light from each wavelength of light. Each smoke or aerosol, from a single material, was tested many times to create large datasets. After a selection process, a Machine Learning Algorithm, namely Random Forest, was trained with the data from all materials burnt. It was found that a number of materials that are commonly involved in building fires can be identified with high accuracy using this model. The materials were identified with an accuracy of 99.6%–59%, which are N-Heptane, polyester carpet, Can smoke, PVC insulated wire, polyurethane foam, cotton fabric, cardboard, cigarette and polystyrene foam. The proposed method provides a model, whose accuracy is quantifiable, with easily trainable algorithm for new materials and can be tailored for certain materials of interest.