artificial neural network-based viscosity study of polymeric solutions

In this study a supervised artificial neural network approach has been adopted to study the viscosity of polyvinylidene fluoride (pvdf)/dimethylacetamide (dmac) solutions. for this purpose, a total number of 1064 data for solution viscosity were measured over a broad range of conditions. after defining 6 factors, including temperature, shear rate, solvent, and three different types of additives as adjusted parameters, radial basis functions (rbf) were used to model the solution viscosity. the capability of the rbf model for describing the rheological behavior of the polymeric solution was examined under several operating conditions and favorable results were observed. the model developed by the rbf approach showed total average absolute relative error (aare) and r2 values of 1.29% and 99.86%, respectively. the results showed that the inclusion of 6 input factors led to an acceptable estimation of pvdf/dmac solutions viscosity. a sensitivity analysis accomplished based on the rbf model revealed that the fractions of organic and inorganic additives are the most effective factors on viscosity at low shear rates (γ ̇<1 s-1) which are respectively followed by the solvent fraction, water fraction, temperature , and shear rate. while the shear rate showed the highest level of impact at γ ̇ >1 s-1.