ANYL 143 |
| Vibrational spectroscopic techniques like Fourier transform infrared (FTIR), near infrared (FTNIR) and FT-Raman have been extensively used in the design of calibration models based on multivariate analytical tools such as PCA, PCR, PLS, etc. However, few reports on the effect of spectral resolution on the prediction errors of PLS calibration models are found in the literature. In this work we report the effect of the spectral resolution on the root mean square errors of cross validation (RMSECV) of PLS calibration models based on FTIR, FTNIR, and FT-Raman data for the same set of standard samples. A set of twenty-five samples corresponding to blends composed of diesel and biodiesel mixtures were prepared by weighing and used to design each calibration model. The biodiesel was a methyl ester of soybean oil and its purity was checked by CGFID. The nominal spectral resolutions used for each vibrational spectroscopic technique were 4, 8, 16, 32, and 64 cm-1 and for comparison purposes, the number of accumulated interferograms was constant for each resolution. The results have shown that the RMSECV values for the PLS/FTNIR model decrease exponentially as the spectral resolution increases. This result is well correlated to the increase in the signal-to-noise ratio as the resolution increases. In the cases were the selected spectral regions present IR or Raman signals well separated, similar behavior was observed for the RMSECV values as obtained from the PLA/FTIR and PLS/FT-Raman models. However, if the selected spectral regions contain bands that are too close, then the RMSECV values decrease as the spectral resolution varies from 4 to 8 cm-1 and then increase again for higher spectral resolutions. |
|
General Posters
7:00 PM-9:00 PM, Sunday, August 19, 2007 BCEC -- Exhibit Hall - B2, Poster
Division of Analytical Chemistry |