Artificial neural network modeling of biosorption process using agricultural wastes in a rotating packed bed

• Published in: Applied thermal engineering 2018-07, Vol.140, p.95-101
• Main Authors: Liu, Zhi-Wei, Liang, Fang-Nan, Liu, You-Zhi
• Format: Article
• Language: English
• Subjects: Agricultural wastes; Artificial neural network; Artificial neural networks; Back propagation; Back propagation networks; Biosorption; Computational fluid dynamics; Correlation coefficients; Fluid dynamics; Fluid flow; Mathematical models; Neural networks; Optimization; Packed beds; Packing density; Parameters; Reynolds number; Rotating packed bed; Rotation; Waste materials
• ISSN: 1359-4311; 1873-5606
• DOI: 10.1016/j.applthermaleng.2018.05.029

[Display omitted] •Biosorption process of metal ions/dyes in the RPB was modeled by ANNs.•Optimum numbers of neurons were 6, 15 and 14 for FFBN, CFBN and EBN, respectively.•FFBN was better than EBN and CFBN in predicting biosorption in the RPB. Given the complexity of computational fluid dynamics models and the inaccuracy of semi-empirical models in modeling biosorption process in a rotating packed bed (RPB), an artificial neural network (ANN) based approach was proposed for modeling of biosorption process in the RPB with different biosorbents from agriculture wastes. The experimental data collected from previous studies were used for ANN modeling, and 82% of the data were used for training and 18% of the data were used for testing. The liquid Reynolds number (ReL), average high gravity factor (β), ratio of contact time to maximum contact time (t/tmax), ratio of particle size to bed depth (D/H) and ratio of initial concentration to packing density (C0/ρ) were set as input parameters; while the ratio of the biosorption amount at time t to the maximum biosorption amount (qt/qmax) was used as output parameters for each model. The optimum number of neurons in the hidden layer was determined based on the mean squared errors (MSE) and correlation coefficients (R2) by an optimization procedure. Compared with cascade-forward back-propagation networks (CFBN) and elman back-propagation networks (EBN), feed-forward backpropagation networks (FFBN) gave a lower MSE value and a higher R2 value, suggesting that FFBN had high prediction accuracy and generalization ability.