Deep Artificial Neural Network Method for Magnetic Hysteresis Loop Prediction of Polyvinyl Alcohol@CoFe2O4 Nanocomposites

In this work, the magnetic hysteresis loop of the polyvinyl alcohol@CoFe₂O₄ nanocomposite has been predicted and simulated using a deep artificial neural network (ANN) and Monte Carlo (MC) methods. To increase the capability of the traditional neural networks in modeling and forecasting problems, the proposed deep ANN has two hidden layers that benefit from deep learning techniques to overcome well-known issues such as overfitting and gradient vanishing. The deep ANN predicted results were compared with the simulated and experimental hysteresis loops of the synthesized polyvinyl alcohol@CoFe₂O₄ nanocomposites obtained from the vibrating sample magnetometer and MC method. The interaction between polymer and nanoparticles, their structure, and morphology were analyzed employing Fourier transform infrared spectroscopy, X-ray diffraction spectroscopy, and field emission scanning electron microscopy. Comparison between the hysteresis loops revealed that the deep ANN method that has been trained with the previous published data was successful in the prediction of the shape and coercive field of particles in a polymer matrix relative to the MC method, which considered only the uniaxial anisotropy and Zeeman energy of the nanoparticles. The coercivity and remanence magnetization measured with the accuracy of about 93.33% and 62.23% for deep ANN method and 80.76% and 66.66% for MC method, respectively.