Predicting filtration clogging behavior using machine learning and physicochemical parameters in final wine filtration processes

Otarola, Luis Lillo ; de la Fuente-Mella, Hanns ; Domarchi, Alonso Peña ; Marzo Navarro, Mercedes (Universidad de Zaragoza) ; Ceroni, José

Resumen: The wine bottling process includes a mechanical filtration to remove unwanted particles such as microorganisms and crystalline precipitates, thus ensuring product quality. This step involves high operational costs due to the variability in the quantity and type of particles, which obstruct the filtering medium, reducing flow and increasing pressure differential, negatively impacting processing and bottling time. Traditionally, clogging power could only be measured through laboratory analysis once the wine was ready for bottling, limiting efficient production planning. This study analyzes the variability of the filtration speed index through analytical and statistical methods and examines how certain physicochemical parameters influence these measurements. Using data from 689 records from a renowned Chilean vineyard, machine learning techniques are applied to evaluate different predictive models. The results indicate that the Gradient Boosting Machine model predicts the filtration speed index with an average absolute error of 10.66%, significantly lower than the analytical limit of 20%, thus facilitating timely and effective operational decision-making.

Idioma: Inglés

DOI: 10.1016/j.seppur.2026.137628

Año: 2026

Publicado en: Separation and Purification Technology 394 (2026), 137628 [13 pp.]
ISSN: 1383-5866

Tipo y forma: Artículo (PostPrint)

Área (Departamento): Área Comerci.Investig.Mercados (Dpto. Direc.Mark.Inves.Mercad.)