Over the past few years, laser-induced breakdown spectroscopy (LIBS) has earned a lot of attention in the field of online polymer identification. Unlike the well-established near-infrared spectroscopy (NIR), LIBS analysis is not limited by the sample thickness or color and therefore seems to be a promising candidate for this task. Nevertheless, the similar elemental composition of most polymers results in high similarity of their LIBS spectra, which makes their discrimination challenging. To address this problem, we developed a novel chemometric strategy based on a systematic optimization of two factors influencing the discrimination ability: the set of experimental conditions (laser energy, gate delay, and atmosphere) employed for the LIBS analysis and the set of spectral variables used as a basis for the polymer discrimination. In the process, a novel concept of spectral descriptors was used to extract chemically relevant information from the polymer spectra, cluster purity based on the k-nearest neighbors (k-NN) was established as a suitable tool for monitoring the extent of cluster overlaps and an in-house designed random forest (RDF) experiment combined with a cluster purity-governed forward selection algorithm was employed to identify spectral variables with the greatest relevance for polymer identification. Using this approach, it was possible to discriminate among 20 virgin polymer types, which is the highest number reported in the literature so far. Additionally, using the optimized experimental conditions and data evaluation, robust discrimination performance could be achieved even with polymer samples containing carbon black or other common additives, which hints at an applicability of the developed approach to real-life samples.
Keywords: Chemometrics; Identification; Laser-induced breakdown spectroscopy; Polymers.
© 2021. The Author(s).