Thin-layer drying of fig (Ficus carica L.) slices is commonly described using empirical integer-order models that overlook the memory effects associated with moisture transport in dense, sugar-rich tissues. This study reanalysed forced-convection solar drying of five Moroccan fig cultivars (Kadota, Sarilop, Conidria, Nabout and Rey Blanche) under three temperatures (60, 70 and 80 °C) and two air-flow rates (150 and 300 m3 h⁻1) using a unified analytical framework combining Caputo time-fractional kinetics, functional principal component analysis (FPCA), hierarchical Bayesian Arrhenius inference and 1-Wasserstein optimal transport metrics. Rigorous model comparison across six classical thin-layer equations (Newton, Page, Henderson–Pabis, Logarithmic, Midilli–Kucuk and Weibull) and the Caputo fractional model was performed using Akaike information criterion (AIC), coefficient of determination (R2) and root-mean-square error (RMSE). Residual analysis (residual plots, Shapiro–Wilk normality tests and Q–Q plots) confirmed the absence of systematic bias in the Caputo fits. The fractional model accurately reproduced all moisture ratio curves (R2 = 0.706–0.999, mean 0.966). At 150 m3 h⁻1, the Caputo memory exponent α decreased from near-Markovian values at 60–70 °C to 0.92 ± 0.07 at 80 °C, with Conidria reaching α = 0.82, whereas α remained ≈1 at 300 m3 h⁻1, indicating suppression of anomalous diffusion under stronger convective conditions. FPCA reduced the 30 drying curves to two dominant modes explaining 93.1% of total variance, corresponding mainly to temperature and cultivar effects. Hierarchical Bayesian pooling showed that the between-cultivar variability in activation energy declined from 14.7 to 3.7 kJ mol⁻1 when the flow rate doubled, demonstrating the homogenising effect of air velocity. Optimal transport clustering resolved two cultivar groups independent of parametric model assumptions. All dried samples reached safe water activity levels (aw ≤ 0.455). The practical implications for dryer design are discussed: the Caputo framework provides a mechanistic basis for predicting drying time from measurable tissue and process parameters, and the optimal transport fingerprint enables cultivar classification without model-dependent bias. The proposed framework establishes an interpretable and reproducible analytical strategy for advanced drying kinetics analysis beyond conventional empirical models.

Memory-Aware Thin-Layer Drying of Fig Slices: Fractional Kinetics, Functional Decomposition and Optimal Transport Fingerprinting

期刊论文
发表日期：

2026年7月11日

查看全文

Lahcen Hssaini,
Rachida Ouaabou

摘要

Thin-layer drying of fig (Ficus carica L.) slices is commonly described using empirical integer-order models that overlook the memory effects associated with moisture transport in dense, sugar-rich tissues. This study reanalysed forced-convection solar drying of five Moroccan fig cultivars (Kadota, Sarilop, Conidria, Nabout and Rey Blanche) under three temperatures (60, 70 and 80 °C) and two air-flow rates (150 and 300 m³ h⁻¹) using a unified analytical framework combining Caputo time-fractional kinetics, functional principal component analysis (FPCA), hierarchical Bayesian Arrhenius inference and 1-Wasserstein optimal transport metrics. Rigorous model comparison across six classical thin-layer equations (Newton, Page, Henderson–Pabis, Logarithmic, Midilli–Kucuk and Weibull) and the Caputo fractional model was performed using Akaike information criterion (AIC), coefficient of determination (R²) and root-mean-square error (RMSE). Residual analysis (residual plots, Shapiro–Wilk normality tests and Q–Q plots) confirmed the absence of systematic bias in the Caputo fits. The fractional model accurately reproduced all moisture ratio curves (R² = 0.706–0.999, mean 0.966). At 150 m³ h⁻¹, the Caputo memory exponent α decreased from near-Markovian values at 60–70 °C to 0.92 ± 0.07 at 80 °C, with Conidria reaching α = 0.82, whereas α remained ≈1 at 300 m³ h⁻¹, indicating suppression of anomalous diffusion under stronger convective conditions. FPCA reduced the 30 drying curves to two dominant modes explaining 93.1% of total variance, corresponding mainly to temperature and cultivar effects. Hierarchical Bayesian pooling showed that the between-cultivar variability in activation energy declined from 14.7 to 3.7 kJ mol⁻¹ when the flow rate doubled, demonstrating the homogenising effect of air velocity. Optimal transport clustering resolved two cultivar groups independent of parametric model assumptions. All dried samples reached safe water activity levels (a_w ≤ 0.455). The practical implications for dryer design are discussed: the Caputo framework provides a mechanistic basis for predicting drying time from measurable tissue and process parameters, and the optimal transport fingerprint enables cultivar classification without model-dependent bias. The proposed framework establishes an interpretable and reproducible analytical strategy for advanced drying kinetics analysis beyond conventional empirical models.