Detection of Ethanol in Wines Using Optical-Fiber Measurements and Near-Infrared Analysis

Abstract

Optical-fiber measurements coupled with near-infrared analysis is used to predict the ethanol content of wine samples to within ±0.33 volume percent. Because telecommunications-grade optical fibers are used, the wavelength range is limited and the signal-to-noise ratio is an order of magnitude less than that for conventional near-infrared measurements. These features prompted the use of partial least-squares (PLS) as the calibration algorithm. PLS provides robust calibrations that are based upon the entire spectrum. The problem of extreme samples not being represented in the calibration model is discussed and illustrated. Also, the robustness of the PLS calibration is demonstrated. Spectral reconstruction is used to help assign the bands in the wavelength range used.