ISSN print edition: 0366-6352 ISSN electronic edition: 1336-9075 Registr. No.: MK SR 9/7 Published monthly

Mathematical modeling of drying kinetics of ground Açaí (Euterpe oleracea) kernel using artificial neural networks

Mohamad A. Bannoud, Beatriz P. Gomes, Marcela C. de S. P. Abdalla, Mariana V. Freire, Kaciane Andreola, Tiago D. Martins, Carlos A. M. da Silva, Luciane F. G. de Souza, and Matheus B. Braga

State University of Campinas (UNICAMP), Campinas, Brazil

E-mail: mohamad.bannoud@unifesp.br

Received: 13 July 2023  Accepted: 5 October 2023

Abstract:

The utilization of Açaí residues holds significant economic and environmental importance in Brazil. The drying technique is an alternative for preserving Açaí kernel from fresh fruit processing. This study employed diffusive, empirical, and Artificial Neural Network (ANN) models to simulate the convective drying of ground Açaí kernel, with air temperature and air velocity ranging from 80 to 120 °C and 0.5 and 0.9 m/s, respectively. To assess the robustness of the models, a validation step using experimental conditions distinct from those used in the training dataset was carried out. The impact of cross-validation on the generalization capacity of the ANN-based models was investigated. Furthermore, input importance techniques were employed to gain insights into the functioning of the ANN models. The diffusivity coefficient exhibited values between 9.22 × 10⁻¹⁰ and 2.26 × 10⁻⁹ m²/s. The Page was the best empirical model (R² = 0.9918). For the ANN-based models, five pairs of input–output variables were considered and produced models with comparable or superior performance compared to the diffusive and empirical models. The best ANN model achieved an R² > 0.9999. The cross-validation technique only enhanced the generalization capacity of ANN-based models that used delayed variables in the input instead of the drying time. Additionally, the feature importance analysis revealed that the best ANN model effectively captured the essential drying aspects: air temperature and velocity effects, the falling rate period, and higher drying rates at the beginning of the experiments.

Graphical abstract

Keywords: Drying kinetics; Euterpe oleracea; Artificial intelligence; Neural networks; Mathematical modeling

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-023-03142-2

Chemical Papers 78 (2) 1033–1054 (2024)