Evaluation of multilayer perceptron neural networks and adaptive neuro-fuzzy inference systems for the mass transfer modeling of Echium amoenum Fisch. & C. A. Mey.

BACKGROUND: Multilayer perceptron (MLP) feed-forward artificial neural networks (ANN) and first-order Takagi-Sugeno-type adaptive neuro-fuzzy inference systems (ANFIS) are utilized to model the fluidized bed-drying process of Echium amoenum Fisch. & C. A. Mey. The moisture ratio evolution is calculated based on the drying temperature, airflow velocity and process time. Different ANN topologies are examined by evaluating the number of neurons (3 to 20), the activation functions and the addition of a second hidden layer. Different numbers (2 to 5) and shapes of membership functions are examined for the ANFIS, using the grid partitioning method. The models with the best performance in terms of prediction accuracy, as evaluated by the statistical indices, are compared with the best fit thin-layer model and the available data from the experimental cases of 40 °C, 50 °C and 60 °C temperatures at 0.5, 0.75 and 1 ms-1 airflow velocity. RESULTS: The best performed ANFIS model, comprised by 5-2-2 of π-shaped andtriangular membership functions for time, temperature and airflow velocityinputs respectively, was able to describe the moisture ratio evolution of E. amoenum more precisely than the best ANN topology, achieving higher values of coefficientof determination (R2 ), root mean square error (RMSE) and sum of squared errors(SSE). The best thin-layer model involving six adjustable parameters, managedto describe experimental data most accurately with R2 = 0.9996, RMSE = 0.0057and SSE = 7.3·10-4 . CONCLUSION: The results of the comparative study indicate that empirical regression models with increased numbers of adjustable parameters, constitute a simpler and more accurate modeling approach for estimating the moisture ratio of E. amoenum Fisch. & C. A. Mey under fluidized bed drying. © 2021 Society of Chemical Industry.