Evaluation of Predictive Capabilities of Regression Models and Artificial Neural Networks for Density and Viscosity Measurements of Different Biodiesel-Diesel-Vegetable Oil Ternary Blends


Gülüm M. , Kutlu Onay F., Bilgin A.

ENVIRONMENTAL AND CLIMATE TECHNOLOGIES, cilt.22, ss.179-205, 2018 (ESCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 22
  • Basım Tarihi: 2018
  • Doi Numarası: 10.2478/rtuect-2018-0012
  • Dergi Adı: ENVIRONMENTAL AND CLIMATE TECHNOLOGIES
  • Sayfa Sayıları: ss.179-205

Özet

Nowadays, biodiesel and vegetable oils have received increasing attention as renewable clean alternative fuels to fossil diesel fuel because of decreasing petroleum reserves and increasing environmental concerns. However, the straight use of biodiesel and vegetable oils in pure form results in several operational and durability problems in diesel engines because of their higher viscosity than fossil diesel fuel. One of the most used methods for solving the high viscosity problem is to blend them with fossil diesel fuel or alcohol. The reliable viscosity and density data of various biodiesel-diesel-alcohol ternary blends or biodiesel-diesel binary blends are plentifully available in existing literature, however, there is still the scarcity of dependable measurement values on different biodiesel-diesel-vegetable oil ternary blends at various temperatures. Therefore, in this study, waste cooking oil biodiesel (ethyl ester) was produced, and it was blended with fossil diesel fuel and waste cooking oil at different volume ratios to prepare ternary blends. Viscosities and densities of the ternary blends were determined at different temperatures according to DIN 53015 and ISO 4787 standards, respectively. The variation in viscosity with respect to temperature and oil fraction and the change of density vs. temperature were evaluated, rational and exponential models were proposed for these variations, and these models were tested against the density and viscosity data measured by the authors, Nogueira et al. and Baroutian et al. by comparing them to Gupta et al. model, linear model, Cragoe model and ANN (artificial neural networks) previously recommended in existing literature.