DENSITY MAPPING IN POLY(ETHYLENE-TEREPHTHALATE) USING A FIBER-COUPLEDRAMAN MICROPROBE AND PARTIAL LEAST-SQUARES CALIBRATION

Partial least-squares (PLS) analysis has been used to calibrate Raman microprobe spectra of poly(ethylene terephthalate) films in terms of d ensity, in order to give insight into changes in crystallinity through the film thickness. The microprobe utilizes a static multiplexed holo graphic grating to obtain the entire Raman spectrum (-1600-4000 cm(-1) ) in a ''single-shot'' at similar to 5-cm(-1) resolution. Because ther e are no moving parts, frequency registration and repeatability are ex cellent and ideally suited for multivariate calibration. In addition, the high spectral throughput allows the whole spectrum to be collected in a few seconds with high signal-to-noise ratio. With this equipment , cross-validated calibration precisions as low as similar to 0.0021 g cm(-3) were achieved. In this work we considered two ways of removing fluorescence backgrounds prior to carrying out multivariate calibrati on. The first involved manually fitting a baseline using a polynomial curve and subtracting it. The second approach simply takes the second derivative of the spectrum to attenuate the low-frequency components ( i.e., the curved baseline). It was found that either pretreatment gave good calibration precision provided that the resultant spectra were i ntensity-normalized to correct for variations in laser power, sample a lignment, and so on. Surprisingly, it was found that the best precisio n was obtained by grouping the spectral resolution elements into block s of eight data points, thereby improving the signal-to-noise but effe ctively degrading the spectral resolution by a factor of three. This w as especially important for the derivative spectra. Alternatively, Sav itsky-Golay smoothing of the second derivative data was applied to the same effect but also at the expense of degrading spectral resolution. The implication of this work is that instruments intended for multiva riate calibration applications could perhaps be designed to work at ra ther lower spectral resolutions (but higher signal-to-noise) than migh t otherwise be considered.