Modeling leaching behavior of solidified wastes using back-propagation neural networks

Abstract

In a previous study, treatment sludge obtained from a chemical industry, which contained potentially toxic heavy metals and organics, was characterized and solidified by solidification/stabilization (S/S). In this study, however, the prediction of leaching behavior of the sludge by linear regression method and neural networks (NNs) was discussed. NN analysis was used to construct models of leaching behavior as a function of mix composition (waste/binder ratio, W/B) using existing data from the previous study of cement-based S/S. The differences in leaching rate of each metal were also considered. The hazard characteristics of the waste were determined as defined in both Turkish and US EPA regulations, by means of Extraction Procedure Toxicity Test (EPTox) and DIN 38414-S4 Test. S/S studies were conducted using Portland cement to solidify the sludge containing high amount of Cr, Cu, Hg, Ni, Pb, and Zn. The W/B ratios of 36 specimens were kept between 0/100 and 40/100. The specimens were cured at room temperature for 7, 28, and 90 days. The heavy metal content of the extracts of each specimen was detected usually less than standard concentrations in EPTox and DIN 38414-S4 leaching procedures. By the use of NN, leaching behavior of the solidified wastes can be predicted and, therefore, optimum S/S technologies can be achieved.