Modelling and optimization of process parameters for silver nanoparticles synthesis: A comparison between response surface methodology and artificial neural network

• Main Authors: Chowdhury, Silvia, Yusof, Faridah, Sulaiman, Nadzril, Faruck, Mohammad Omer, Sidek , Shahrul Naim
• Format: Conference or Workshop Item
• Language: English
• Published: Kulliyah of Engineering, International Islamic University Malaysia 2016
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://irep.iium.edu.my/51712/1/51712.pdf; http://irep.iium.edu.my/51712/; http://www.iium.edu.my/icbioe/2016/

In recent decades, Response Surface Methodology (RSM) and Artificial Neural Network (ANN) have become the most used method for empirical modelling and optimization of process parameters in Chemical Engineering. This paper reports a comparative study between RSM and ANN data analysis methods for process parameter optimization of silver nanoparticles (AgNPs) synthesis. AgNPs yield was modelled as a function of four independent variables, i.e. AgNO3 concentration, initial pH of aqueous AgNO3, reducing agent (tri-sodium citrate) concentration and reaction time. The Design of Experiments via Face Centred Central Composite Design (FCCCD) was conducted using Design Expert® (Version 7.0) and quantitation of ‘yield’ was captured as the estimated area under the curve of UV-vis spectral scan of the product from 350 to 420 nm with Area* as the unit. As accorded by the software, with four parameters and three levels plus triplicate centre points, 27 experimental runs were conducted. Using both RSM and ANN to analyse the data for optimization and generalization ability, the predicted AgNPs yield were found to be in good agreement with the experimental results. Coefficient of Determination values (R2 ) for ANN and RSM were 0.9967 and 0.9584 respectively, implies that both developed models have good fitting with the experimental data. But in comparing the two models through the R2 and Absolute Average Deviation (AAD), ANN (R2 =0.9967, AAD=2.63 %) shows more superior values to the RSM (R2 =0.9584, AAD=5.91%), thus ANN produced better fitted models in predicting AgNPs yield production. In addition to comparing their modelling, sensitivity analysis and optimization ability, response surface plot were also constructed to evaluate the influences and interactions of input variables between the two models.