Prediction of Dynamic Viscosity of Biodiesel Using Various Artificial Neural Network Methods, Response Surface Methodology, and Multiple Linear Regressions

In this paper, five machine learning models (feedforward neural network, Radial basis function neural network, Layer Recurrent Neural Network, Elman Neural Network, and Cascade Feed-forward Neural Network) and mathematical models (response surface methodology and multiple linear regressions) have been utilized to estimate the dynamic viscosity (DV) of biodiesel. In this work, the physiochemical properties including the kinematic viscosity (KV) and density (D) of biodiesel produced from various waste vegetable oils are determined according to ASTM standards. The biodiesel samples are stored for two years at 0 °C, room temperature (23 ± 1 °C), and 40 ± 1 °C, as well as the KV and D, are measured at various testing temperatures. The storage period storage temperature and testing temperature are utilized as input variables for the models. The results showed that all machine learning models have a good performance to estimate the DV of biodiesel compared to mathematical models. Among the developed models, the Layer Recurrent Neural Network is presented as the best model for the DV of biodiesel.