Artificial neural network prediction of thermal and mechanical properties for Bi2O3-polybenzoxazine nanocomposites

Polybenzoxazine (BA-a) based nanocomposites with varying amounts of Bi 2 O 3 particles reinforcement (10, 20, and 30 wt%) were produced. The structures of the Bi 2 O 3 before and after surface silane treatment, as well as the structures of the BA-a matrix and its nanocomposites, were all evaluated using Fourier transform infrared spectroscopy (FTIR). The thermal stability of the samples was evaluated by thermogravimetric analysis (TGA) and the resistance to bending was studied using the three-point bending test. Then, an artificial neural network approach was used to estimate the thermal degradation and flexural strengths of the polybenzoxazine matrix and its nanocomposites. The results of the TGA analysis showed that when Bi 2 O 3 content increased, the char yield ( Y c ) of the nanocomposites also increased from 27.98% to 43.53% for the experimental data and from 27.95% to 43 0.57% for the prediction, with all SD s less than 0.02. Moreover, both the experimental and predicted values of the flexural strengths for the unfilled BA-a displayed 100.40 and 102.96 MPa, respectively and these values increased to 119.65 and 114.95 MPa when 10 wt% of fillers were incorporated into the matrix. However, for higher filler contents, 20 and 30 wt%, these values decreased to 112.94 and 108.01 MPa for the experiments, and 106.35 and 103.20 MPa for the prediction, respectively. The very small deviations (<0.05) between the experimental and predicted values demonstrated that the prediction approach using the artificial neural network is reliable and accurate.