Adaptive neuro-fuzzy inference system approach to predict dynamic thermo-mechanical responses of poly (vinylidene fluoride) blend-based nanocomposites

The nonlinear nature of viscoelastic properties in polymeric materials makes it difficult to model and predict the dynamic mechanical responses. This experimental and computational research aims to establish a reliable model for predicting solid viscoelastic properties in PVDF/PEO blends and the corresponding nanocomposites with the technological importance in the fabrication of electrolyte membranes. In this regard, temperature-dependent dynamic mechanical properties were collected by the dynamic mechanical analyzer (DMA), and the variation of storage modulus (E '), loss modulus (E ''), dissipation factor (tan delta), and the glass transition temperatures (T-g) were monitored through the whole composition range. Next, an artificial intelligence network utilizing the (adaptive neuro-fuzzy inference system) (ANFIS) method was constructed for the prediction of E ' and tan delta as outputs, whereas the content of poly (ethylene oxide) (wt%) as the blending component, the loading of graphene nanosheets (wt%), and the temperature were assigned as inputs. To find the optimum number and combination of fuzzy rules, three different methodologies concerning grid partitioning (GP) with the diverse number of gaussian membership functions (2-7), subtractive clustering (SC) with different cluster radius (0.15-0.4), and fuzzy c-mean (FCM) with a various number of clusters (40-55) were considered. The results of root mean square error (RMSE), coefficient of determination (R-2), and mean absolute percentage error (MAPE) for different combination of fuzzy rules indicated that the GP methodology with the set of membership functions of [7 2 7] has a superior performance for prediction of various properties in highperformance materials.