Naslov
Characterization of arsenic immobilization in zeolite - lime - cement blends using artificial neural networks

Autori
Bolanča, Tomislav ; Šipušić, Juraj ; Ukić, Šime ; Šiljeg, Mario ; Ujević, Magdalena

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
3rd Workshop Eureka "Purewater" 4208!E : Book of abstracts / Margeta, Karmen - Zagreb : Fakultet kemijskog inženjerstva i tehnologije Sveučilišta u Zagrebu, 2010, 29-29

ISBN
978-953-6470-49-5

Skup
Workshop Eureka "Purewater" 4208IE (3 ; 2010)

Mjesto i datum
Zagreb, Hrvatska, 05.05.2010. - 06.05.2010

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
arsenic immobilization; zeolite; lime; cement; artificial neural networks

Sažetak
The global epidemic of arsenic poisoning, especially from ground waters, has become a matter of grave environmental concern in recent years. water treatment technologies for arsenic removal usually are based on ion exchange and/or adsorption (e.g. iron oxide). However, after the adsorbent medium is completely exhausted, the disposal of the spend medium is a major consideration, since toxic levels of arsenic which may leach out into the environment and thus has to be disposed of safely according to prevailing environment regulations. Arsenic waste immobilization technology using portland cement are currently recognized as a most promising to prevent the free movements of arsenic in the waste surrounding media. The major objective of this study is to develop an effective model, based on sodification - stabilization technique, to treat toxic arsenic rich spent adsorbent for its safe disposal. For this purpose artificial neural network model was develop to predict the characteristics of material used for stabilization. Zeolite - lime - cement - water - arsenic spent adsorbent ratio was modeled in relation with mechanical strength and leaching of arsenic and iron ions. Developed artificial neural network model was based on feed forward error back propagated methodology. In order to increase the predictive ability of the model gradient descent, Broyden-Fletcher-Goldfarb-Shanno and scaled conjugate gradient training algorithms were tested in combination with tangent hyperbolic, logistic and exponential activation function. Number of hidden layer nezrons was optimized to prevent overtraining and ensure good generalization. The developed artificial neural network model was validated showing satisfactory performance characteristic.

Izvorni jezik
Engleski

Znanstvena područja
Kemija, Kemijsko inženjerstvo

POVEZANOST RADA