ARTIFICIAL NEURAL NETWORK PREDICTION OF PERFORMANCE CHARACTERISTICS OF BIOFUEL PRODUCED FROM SWEET POTATOE (IPOMOEA BATATA)

Abstract

Fossil fuel depletion and the harm it causes to the environment has led to the development of alternative fuels. In this research, biofuel (ethanol) was produced and characterized from sweat potatoes. Blends of premium motor spirit with 0% (E0), 2% (E2), 4% (E4), and 10% (E10) of the produced biofuel at various percentages were separately used to power a four-stroke, single-cylinder SI engine on an engine test bed, and data of the engine performance - brake power, brake torque, brake mean effective pressure (BMEP), and the exhaust gas temperature reported in each test. The results of the physicochemical analysis revealed that the physical state of the biofuel is colorless, the viscosity at 30⁰C, density, calorific value, and pH level are 0.9834 mPa.s, 0.85 g/cm³,19 kJ/kg, and 1.82, respectively. It was observed that an increase in ethanol in the blend increases the performance of the engine, although the BMEP at E0 gave the highest value of 0.3 bar compared to other blends.  An artificial neural network (ANN) model for predicting engine performance characteristics was developed, trained, validated, and tested using the reported data. The result of the ANN model revealed that the Levenberg-Marquardt training algorithm (LMTA) with 10 hidden layer neurons offers the best fit for the features for both training, validation, testing, and overall. With the R for training equal 1, validation equal to  0.99468, testing equal to 0.90103, and overall R equal to  0.93842 as compared to the rest in terms of the number of neurons and training algorithms.