Kinematic viscosity estimation of fuel oil with comparison of machine learning methods

It is known that one of the important parameters affecting the emission and performance values of the liquid fuels used in thermal machines is the viscosity value. Therefore, many different studies have been carried out on the determination of dynamic and kinematic viscosities of liquid fuels. In this study, the kinematic viscosity value of Fuel Oil 4 at a constant temperature of 100 degrees C was estimated using machine learning methods. Extreme Learning Machine (ELM), Multi-Layer Perceptron (MLP), and K Nearest Neighbor (K-nn) methods were used to perform kinematic viscosity estimations. Two different distance metrics are considered in the K-nn algorithm. The experimentally obtained water content, density and flash point properties of the fuel were used as input data for machine learning approaches. Thus, four different models were developed for the kinematic viscosity estimation of Fuel oil fuel. The success rates of the predictions were compared using the Mean Relative Error (MRE) and Mean Squared Error (MSE). As a result, it was seen that all of the methods discussed provide predictive ability in accordance with standard values and the best prediction data is provided by using ELM.