Experimental Evaluation and Modeling of Strawberry Slices Drying Kinetics Based on Machine Learning - Original scientific paper

Aug 4, 2026

Upload by Eden

Research investigates the drying kinetics of strawberry slices by combining traditional mathematical modeling with machine learning. Experiments use 5 mm slices with an initial moisture content of 88.04% (wb) while varying key operating parameters including temperature, air velocity, and drying duration. Moisture ratio predictions are built from models derived from Fick’s Second Law and Newton’s Law of Cooling and are benchmarked against artificial neural networks and recurrent neural networks. Ten RNN configurations with three hidden layers (20/30/40 nodes) are trained and evaluated. RNN models, especially RNN04, achieve the best accuracy with maximum deviation up to about 2% versus experimental data, outperforming ANN results. The findings indicate that RNN-based approaches improve process understanding and enable more precise, efficient drying control for industry applications.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Motivation and background

Traditional drying models

Machine learning for drying kinetics

Materials and Methods

Materials

Drying equipment

ISSN 1330-3651 (Print), ISSN 1848-6339 (Online) [https://doi.org/10.17559/TV-20240723001875](https://doi.org/10.17559/TV-20240723001875) Original scientific paper Experimental Evaluation and Modeling of Strawberry Slices Drying Kinetics Based on Machine Learning Olivera EĆIM-ĐURIĆ*, Aleksandra DRAGIČEVIĆ, Rajko MIODRAGOVIĆ, Mihailo MILANOVIĆ, Andrija RAJKOVIĆ, Zoran MILEUSNIĆ, Vjekoslav TADIĆ Abstract: The research explores the drying kinetics of strawberry slices (5 mm thick with initial moisture content of 88 .04% wb) through the application of both traditional mathematical models and advanced machine learning method. The study aims to optimize the drying process by examining the effects of variables such as temperature, air velocity, and drying duration. Traditional models, derived from Fick's Second Law and Newton's Law of Cooling, were compared with artificial neural networks (ANN) andrecurrent neural networks (RNN) to predict moisture content during the drying process. Ten network models were formed, and each model had three "hidden" layers with 20, 30, and 40 nodes in each layer. Findings revealed that RNN models, particularly RNN04, surpassed traditional models in accuracy, with a maximum deviation of up to 2% from experimental data. RNN models showed lower deviations in the range of 0 .65% to 2%, while the ANN models had deviations in the interval of 2 .6% to 5.6% . The ANN and RNN models included parameters like temperature, air flow speed, and drying time, with RNN models exhibiting superior adaptability and precision. These results indicate that machine learning approaches, especially RNNs, can greatly improve the understanding and management of the drying process, providing more precise and efficient methods for the drying industry. Keywords: artificial neural networks; drying kinetics; machine learning; mathematical modelling; recurrent neural networks; strawberry slices 1 INTRODUCTION Drying is one of the oldest techniques for preserving food, and even today, convective drying remains one of the most commonly used methods in industry due to its simplicity. However, the relatively high temperatures to which the wet material is exposed, and the long drying times can lead to undesirable effects such as nutrient degradation, and changes in colour and shape [1, 2] . Pretreatment applications can shorten the process time [3] . The choice of the shape of the drying material can also strongly influence the drying process and the quality of the end product. In this context, dried fruit and vegetable slices have become increasingly popular in recent years and have also found their place on the market as healthy snacks. To understand the complex phenomena of mass and energy transfer that occur during convective drying between a hot air flow (acting as a working fluid) and a moist material, methods to determine the kinetics of drying by establishing functional relationships between drying time and moisture ratio are commonly used [4, 5] . Mathematical models and simulations of drying curves are very important tools for controlling the process itself and for determining the quality of the final product, attracting the interest of researchers for years and even in recent research on fruit drying [6, 7] . Theoretical models involve sets of differential equations that account for internal moisture movement mechanisms, external conditions, and material properties, making them challenging to solve. Simpler, semi-empirical models are derived either from Fickꞌs Second Law of Diffusion or Newtonꞌs Law of Cooling. Well-known mathematical formulations have so far been applied to many studied samples, representing the entire drying process through the dependence of moisture content on drying time, considering the shape and type of wet material and drying mode. The application of artificial intelligence and machine learning methods offers the possibility of investigating the kinetics of drying using a larger number

FAQs

Which variables are considered when studying strawberry slices drying kinetics?

The study evaluates effects of drying temperature, air velocity, and drying duration on moisture changes during convective drying.

How do traditional mathematical models compare with machine learning models?

Traditional models based on Fick’s Second Law and Newton’s Law of Cooling are compared against ANN and RNN approaches for predicting moisture content over time.

What model type achieved the highest prediction accuracy?

Recurrent neural networks performed best, with RNN04 showing the greatest accuracy and deviations up to about 2% from experimental data.

Aug 4, 2026

Upload by Eden