Primena ANN modela na proces uklanjanja bakra iz vodenih rastvora upotrebom agro-industrijskog otpada

Abstract

U radu je ispitana mogućnost upotrebe lokalno dostupnih lignoceluloznih materijala kao potencijalnih biosorbenata u svrhu uklanjanja jona bakra iz vodenih rastvora. Ispitivani materijali predstavljaju čvrst otpad koji nastaje nakon prerade kukuruza (oklasak kukuruza) i nakon prerade voća (koštice kajsije). Ovakav otpad ima malu ekonomsku vrednost a njegovo odlaganje predstavlja ekološki problem. U šaržnom sistemu ispitani su uticaj inicijalne koncentracije Cu(II) jona (Ci), količine biomase (m) i zapremine rastvora (V) na efikasnost biosorpcije i vrednost biosorpcionog kapaciteta. Utvrđeni su optimalni procesni parametri. Eksperimentalni rezultati su poređeni sa dva modela: SOP (second order polynomial regression models) i ANN (artificial neural networks), pri čemu je SOP model pokazao prihvatljiv determinacioni koefijent (0,842-0,997), dok je ANN pokazao visoku tačnost prognoze (0,980-0,986) u odnosu na eksperimentalne rezultate.

The purpose of this study was to investigate the adsorption properties of locally available lignocellulose biomaterials as biosorbents for the removal of copper ions from aqueous solution. Materials are generated from juice production (apricot stones) and from the corn milling process (corn cob). Such solid wastes have little or no economic value and very often present a disposal problem. Using batch adsorption techniques the effects of initial Cu(II) ions concentration (Ci), amount of biomass (m) and volume of metal solution (V) on biosorption efficiency and capacity were studied for both materials, without any pre-treatments. The optimal parameters for both biosorbents were selected depending on the highest sorption capability of biosorbent in removal of Cu(II). Experimental data were compared with second order polynomial regression models (SOPs) and artificial neural networks (ANNs). SOPs showed acceptable coefficients of determination (0.842-0.997), while ANNs performed with high prediction accuracy (0.980-0.986) in comparison to experimental results.