Infrared-Photoacoustic Spectroscopy and Multiproduct Multivariate Calibration to Estimate the Proportion of Coffee Defects in Roasted Samples
Rafael Dias,
Patrícia Valderrama,
Paulo Março,
Maria Scholz,
Michael Edelmann,
Chahan Yeretzian
Affiliations
Rafael Dias
Coffee Excellence Center, Institute of Chemistry and Biotechnology, Zurich University of Applied Sciences (ZHAW), Einsiedlerstrasse 31, CH-8820 Wädenswil, Switzerland
Patrícia Valderrama
Post-Graduation Program in Food Technology—PPGTA, Federal Technological University of Paraná State—UTFPR, Via Rosalina Maria dos Santos, 1233, Campo Mourão 87301-899, Paraná, Brazil
Paulo Março
Post-Graduation Program in Food Technology—PPGTA, Federal Technological University of Paraná State—UTFPR, Via Rosalina Maria dos Santos, 1233, Campo Mourão 87301-899, Paraná, Brazil
Maria Scholz
Instituto Agronômico do Paraná—IAPAR, Technical Scientific Board, Rod. Celso Garcia Cid, Km 375, Postal Code 481, Londrina 86001-970, Paraná, Brazil
Michael Edelmann
Coffee Excellence Center, Institute of Chemistry and Biotechnology, Zurich University of Applied Sciences (ZHAW), Einsiedlerstrasse 31, CH-8820 Wädenswil, Switzerland
Chahan Yeretzian
Coffee Excellence Center, Institute of Chemistry and Biotechnology, Zurich University of Applied Sciences (ZHAW), Einsiedlerstrasse 31, CH-8820 Wädenswil, Switzerland
Infrared-photoacoustic spectroscopy (IR-PAS) and partial least squares (PLS) were tested as a rapid alternative to conventional methods to evaluate the proportion of coffee defects in roasted and ground coffees. Multiproduct multivariate calibration models were obtained from spectra of healthy beans of Coffea canephora and C. arabica (Arabica) and blends composed of defective and healthy beans of Arabica in different proportions. The blends, named selections, contained sour, black, broken, whole beans, skin, and coffee woods. Six models were built using roasted and ground coffee samples. The model was optimized through outlier evaluation, and the parameters of merit such as accuracy, sensitivity, limits of detection and quantification, the inverse of analytical sensitivity, linearity, and adjustment were computed. The models presented predictive capacity and high sensitivity in determining defects, all being predicted with suitable correlation coefficients (ranging from 0.7176 to 0.8080) and presenting adequate performance. The parameters of merit displayed promising results, and the prediction models developed for %defects can be safely used as an alternative to the reference method. Furthermore, the new method is fast, efficient, and suitable for in-line application in quality control industrial coffee processing.
