Determination of biomass drying speed using neural networks

Abstract

[EN] The difficulty of measuring the drying rate of biomass under hot air convection conditions due to the influence of multiple factors, such as environmental conditions and material properties; and the problems associated with the variability of desiccation curves under changing conditions makes the use of mass transfer models based on diffusion and convection generally quite inaccurate. The research proposes the use of neural networks to determine the average drying speed (g removed water in unit of biomass material (kg) in unit time (s)), highlighting its ability to handle complex and variable data, as well as its adaptability and robustness. After 62 iterations, the R 2 of the training process reached values of 0.93. Subsequent validation provided an R 2 of 0.88. The mean square error was less than 10 -3 g dryed water kg -1 biomass s -1 . Traditional mass transfer models applied to drying processes were compared with experimental data. It has been proven that the values of the convection coefficient in mass transfer are overestimated when obtained from the Sherwood number. Values of this coefficient applied to wood are 30 times lower due to capillary phenomena and electrostatic forces between the material and the water particles.