Application of ANN and RSM on fluoride removal using chemically activated D. sissoo sawdust

Response surface methodology (RSM) and artificial neural network (ANN) were used to generate a model for the optimization of fluoride removal using chemically activated Dalbergia sissoo sawdust (CADS). The single and collective effects of process parameters, i.e., solution pH, CADS dose, initial fluoride concentration, and contact time, were studied. The point of zero charge was found to be 4.2 with zeta potential analysis. In the first phase, a single-parameter study was performed to reveal dependency of fluoride removal on a particular process parameter. Positive effects of increment in CADS dose and contact time and negative effects of solution pH and initial fluoride concentration were observed. The second phase included RSM in which analysis of variance (ANOVA) was applied to test the feasibility of the mathematical model. The F value 1.91, R² value 0.87, and P value 0.11 show significance of the proposed model. Results obtained from the experiment set for central composite design (CCD) were used to predict the ANN response. Reasonable acceptable values of regression for training, test, and validation (0.76, 0.93, and 0.37) represent the suitability of the model. The ANN predicted 22.1% fluoride removal, which was close to the actual value (20.1%) and was comparable with CCD prediction (25.0%). BET surface area of CADS was found to be 76.33 m²/g. FTIR was performed to recognize the functional groups available for fluoride binding while SEM and EDX were conducted to ensure the changes in adsorbent surface morphology. Regeneration of CADS was feasible using an alkali medium. This study shows that CADS can be used for fluoride removal from aqueous stream in an efficient way.