This article reports on a study that used laser-induced breakdown spectroscopy (LIBS) for the discrimination of automobile paint samples.
Paint samples from automobiles of different makes, models, and years were collected and separated into sets based on the color, presence or absence of effect pigments, and the number of paint layers. Twelve LIBS spectra were obtained for each paint sample, each an average of a five single shot “drill down” spectra from consecutive laser ablations in the same spot on the sample. Analyses by a nonparametric permutation test and a parametric Wald test were performed to determine the extent of discrimination within each set of paint samples. The discrimination power and Type I error were assessed for each data analysis method. Conversion of the spectral intensity to a log-scale (base 10) resulted in a higher overall discrimination power while observing the same significance level. Working on the log-scale, the nonparametric permutation tests gave an overall 89.83 percent discrimination power with a size of Type I error being 4.44 percent at the nominal significance level of 5 percent. White paint samples, as a group, were the most difficult to differentiate, with the power being only 86.56 percent, followed by 95.83 percent for black paint samples. Parametric analysis of the data set produced lower discrimination (85.17 percent) with 3.33 percent Type I errors, which is not recommended for both theoretical and practical considerations. The nonparametric testing method is applicable across many analytical comparisons, with the specific application described here being the pairwise comparison of automotive paint samples. (publisher abstract modified)