A Mass Spectrometry-Machine Learning Approach for Detecting Volatile Organic Compound Emissions for Early Fire Detection

Abstract

Mass spectrometry in parallel with real-time machine learning techniques were paired in a novel application to detect and identify chemically specific, early indicators of fires and near-fire events involving a set of selected materials: Mylar, Teflon, and poly(methyl methacrylate) (PMMA). The volatile organic compounds emitted during the thermal decomposition of each of the three materials were characterized using a quadrupole mass spectrometer which scanned the 1–200 m/z range. CO2, CH3CHO, and C6H6 were the main volatiles detected during Mylar thermal decomposition, while Teflon’s thermal decomposition yielded CO2 and a set of fluorocarbon compounds including CF4, C2F4, C2F6, C3F6, CF2O, and CF3O. PMMA produced CO2 and methyl methacrylate (MMA, C5H8O2). The mass spectral peak patterns observed during the thermal decomposition of each material were unique to that material and were therefore useful as chemical signatures. It was also observed that the chemical signatures remained consistent and detectable when multiple materials were heated together. Mass spectra data sets containing the chemical signatures for each material and mixtures were collected and analyzed using a random forest panel machine learning classification. The classification was tested and demonstrated 100% accuracy for single material spectra and an average of 92.3% accuracy for mixed material spectra. This investigation presents a novel technique for the real-time, chemically specific detection of fire related VOCs through mass spectrometry which shows promise as a more rapid and accurate method for detecting fires or near-fire events.