سلامت و محیط (Mar 2016)
Application of artificial neural network (ANN) in Biosorption modeling of Chromium (VI) from aqueous solutions
Abstract
Background and Objectives: In this work, biosorption of hexavalent chromium from aqueous solution with excess municipal sludge was studied. Moreover, the performance of neural networks to predict the biosorption rate was investigated. Materials and Methods: The effect of operational parameters including initial metal concentration, initial pH, agitation speed, adsorbent dosage, and agitation time on the biosorption of chromium was assessed in a batch system. A part of the experimental results was modeled using Feed-Forward Back propagation Neural Network (FFBP-ANN). Another part of the test results was simulated to assess the model accuracy. Transfer function in the hidden layers and output layers and the number of neurons in the hidden layers were optimized. Results: The maximum removal of chromium obtained from batch studies was more than 96% in 90 mg/L initial concentration, pH 2, agitation speed 200 rpm and adsorbent dosage 4 g/L. Maximum biosorption capacity was 41.69 mg/g. Biosorption data of Cr(VI) are described well by Freundlich isotherm model and adsorption kinetic followed pseudo-second order model. Tangent sigmoid function determined was the most appropriate transfer function in the hidden and output layer. The optimal number of neurons in hidden layers was 13. Predictions of model showed excellent correlation (R=0.984) with the target vector. Simulations performed by the developed neural network model showed good agreement with experimental results. Conclusion: Overall, it can be concluded that excess municipal sludge performs well for the removal of Cr ions from aqueous solution as a biological and low cost biosorbent. FFBP-ANN is an appropriate technique for modeling, estimating, and prediction of biosorption process If the Levenberg-Marquardt training function, tangent sigmoid transfer function in the hidden and output layers and the number of neurons is between 1.6 to 1.8 times the input data, proper predication results could be achieved.
