Prediction Of Engine Performance For F-T And Diesel Blended Fuel Using Artificial Neural Network

The requirements of automotive fuel increasingly demand with the growing of vehicles and emission regulations getting stringent increasingly. Coal to liquid is an effective way to develop alternative fuels based on the advantages of China’s energy structure. In this investigation, output characteristics of different mixing proportions of diesel and F-T diesel were comprehensive evaluation based on ANN models and multi-attribute decision system.In this paper, speed characteristics test has been done on4100QBZL by fueled with0#, F10, F-T under80%of throttle opening. The result shows that torque and fuel consumption decreased with the increase of F-T ratio, and NOx emission decreased significantly. Peak of cylinder pressure and heat release rate reduced with the increasing gradually of F-T proportion, frequency domain Eigen values of head vibration has the same variation trend. According to the parameters affected each other, training artificial neural networks model for power, economy, emission, combustion and head vibration characteristics. The result shows that the frequency domain model and PM model has a larger relative error while other models’error are less than8%which better able to fit the experimental data. Creating a unified evaluation to evaluate output performance of different fuels is necessary since the performance evaluation relates many aspects of fuel. The outputs of the different properties of engine burring different fuels were calculated in this paper based on multi-attribute decision system; also deeply dig the property indexes on the overall performance; and established a unified evaluation system combined with the emphasis on subjective decision makers on indicators. Output characteristic were predicted by artificial neural networks under maximum torque speed for F10, F50, F80, selecting F10as the optimal solution by multi-attribute decision system combines the makers’willing.Six-speed characteristics were predicted at80%throttle opening and engine bench test were done for verification. The result shows that the ANN models have great generalization with relative error less than13%, except opacity model and head vibration characteristic value model for frequency.Comprehensive view, a successful ANN model is likely to predict the output characteristics for different blended ratio, and get an optimization blended ratio combined multi-attribute decision system.