Multivariate calibration of carbon Raman spectra for quantitative determination of peak temperature history

Abstract Multivariate calibration methods are shown to yield quantitative estimates of peak sample temperatures of carbonaceous materials based on their Raman spectra. Partial least-squares (PLS) and principal component regression (PCR) multivariate calibration methods applied to uncorrected Raman spectra of carbon heat-shield materials yielded similar standard errors of predictions (SEP) of peak temperatures (SEP of 202°C for PLS and 191°C for PCR) for samples heated to temperatures ranging from 375 to 2700°C. Narrowing the range of peak temperatures of the calibration samples to 1200–2700°C improved the SEP for these samples by a factor of two. These analyses improved previous methods by eliminating tedious and subjective spectral baseline corrections and band fitting. Raman temperature determinations of future samples would, therefore, be more reliable and faster. The Raman analysis can proceed successfully in the presence of sample-dependent scattering efficiencies, baseline variations, and fluorescence. Useful qualitative information was also obtained by observing the pattern of temperature errors and by examining the loading vectors used to model the spectra. Finally, efficient outlier detection methods are available with these multivariate methods to identify problem samples. These same outlier detection methods could be used to improve the reliability of future analyses by identifying those samples which are different from the calibration samples and whose analysis by these multivariate methods might be considered suspect.