Adaptive neuro-fuzzy inference system and neural network in predicting the size of monodisperse silica and process optimization via simulated annealing algorithm

Document Type : Research Paper


Department of Materials Science and Engineering, Shahid Bahonar University of Kerman, Kerman, Iran


In this study, Back-propagation neural network (BPNN) and adaptive neuro-fuzzy inference system (ANFIS) methods were applied to estimate the particle size of silica prepared by sol-gel technique. Simulated annealing algorithm (SAA) employed to determine the optimum practical parameters of the silica production. Accordingly, the process parameters, i.e. tetraethyl orthosilicate (TEOS), H2O and NH3 were introduced to BPNN and ANFIS methods. Average mean absolute percentage error (MAPE) and correlation relation (R) indexes were chosen as criteria to estimate the simulation error. Comparison of proposed optimum condition and the experimental data reveal that the ANFIS/SAA strategies are powerful techniques to find the optimal practical conditions with the minimum particles size of silica prepared by sol-gel technique and the accuracy of ANFIS model was higher than the results of ANN. Moreover, sensitivity analysis was employed to determine the effect of each practical parameter on the size of silica nano particles. The results showed that the water content and TEOS have the maximum and minimum effect on the particle size of silica, respectively. Since, water acts as diluent and synthesis of monodisperse silica in diluent solution will decrease the growth probability of nucleate, leading to a the lower silica particle size.