Modelling the ageing process: a novel strategy to analyze the wine evolution towards the expected features

• Autor(a) principal: Pereira, Ana C.
• Data de Publicação: 2016
• Outros Autores: Carvalho, Maria J., Miranda, Andreia, Leça, João M., Pereira, Vanda, Albuquerque, Francisco, Marques, José C., Reis, Marco S.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.13/3726
• topic: Wine ageing; Chemical characterization; Feature extraction; Classification models; Process monitoring and evaluation; .; Faculdade de Ciências da Vida; Escola Superior de Tecnologias e Gestão; Faculdade de Ciências Exatas e da Engenharia

In this work we present a new strategy to monitor the wine evolution during the ageing process. More specifi cally, we validate a procedure for analyzing how wine evolves during the ageing process in relation to the desired and expected quality features and we apply the proposed methodology to the case of a Portuguese fortified wine, the Madeira wine, where we compare the wine evolution under two different ageing processes. The approach developed consists on modeling samples labeled as aged reference wines (5 year old Madeira wines), produced from four different grape varieties, and then analyze how and in which extent young wines (up to 3 years old) come closer to the reference data set. The analysis is based on a comprehensive set of chemical data, including: polyphenolic composition, organic acids, reducing sugars, color and oenological parameters, commonly used as routine quality control information. The study considers several feature extraction methods, such as: Principal Components of Analysis (PCA), Independent Component of Analysis (ICA) and Partial Least Squares (PLS). The classification methodologies tested were: Linear Discriminant Analysis (LDA), nearest neigh bor (k-NN) and Soft Independent Modelling by Class Analogy (SIMCA). The different options of preprocessing/ feature extraction/classification were evaluated and compared using a Monte Carlo approach. From our analysis, the best combination of feature extraction/classification methodologies was PLS/LDA, which presented a classification performance of approximately 90% for three out of the four classes modeled, and of about 78% for the remaining one. Regarding the wines monitored during the first 3 years, our analysis revealed that they indeed mature in relation to the five year old reference wines. Furthermore, for some wines, it is possible to detect differences between the two ageing processes analyzed. This study is of particular importance for this type of wines, where the ageing process plays a central role for attaining the expected quality levels, implying significant risks and costs for local and industrial producers. Notwithstanding the specific case study presented, the strategy outlined can be extrapolated to other products with similar characteristics in terms of their monitoring and process control