Resolving Coffee Roasting-Degree Phases Based on the Analysis of Volatile Compounds in the Roasting Off-Gas by Photoionization Time-of-Flight Mass Spectrometry (PI-TOFMS) and Statistical Data Analysis: Toward a PI-TOFMS Roasting Model.

Coffee beans of two cultivars, Arabica (Mexico) and Robusta (Vietnam), were roasted in a small-scale drum roaster at different temperature profiles. Evolving volatile compounds out of the roasting off-gas were analyzed by photoionization mass spectrometry at four different wavelengths, either with single-photon ionization (SPI) or resonance-enhanced multiphoton ionization (REMPI). The different analyte selectivities at the four wavelengths and their relevance for the examination of the roasting process were discussed. Furthermore, intensities of observed m/z were grouped by non-negative matrix factorization (NMF) to reveal the temporal evolutions of four roasting phases ("evaporation", "early roast", "late roast", and "overroast") from NMF scores and the corresponding molecular composition from the NMF factor loadings, giving chemically sound results concerning the roasting phases. Finally, linear classifiers were constructed from real mass spectra at maximum NMF scores by linear discriminant analysis to obtain quantities which are simple to measure for real-time analysis of the roasting process.