Artificial Neural Network Based Predictive Emission and Fuel Economy Simulation of Motorcycles 2018-32-0030
As the number of different engine and vehicle concepts for powered-two wheelers is very high and will even rise with hybridization, the simulation of emissions and fuel consumption is indispensable for further development towards more environmental friendly mobility.
In this work, an adaptive artificial neural network based predictive model for emission and fuel consumption simulation of motorcycles is presented. The model is developed in Matlab and Simulink and is integrated to a longitudinal vehicle dynamic simulation whereby it is possible to simulate various and not yet measured test cycles. Subsequently it is possible to predict real drive emissions and on-road fuel consumption by a minimum of previous measurement effort.
The modelling approach is adaptive in terms of usability for different engine and exhaust gas treatment systems as the model does not require specific knowledge about technical vehicle parameters, which can be unknown due to manufacturers’ concealment. Backpropagation is used as supervised learning technique for training the neural networks and various learning inputs are investigated and evaluated.
The paper expands on previous research of possible measurement methodologies for real drive emissions for motorcycles to minimize the effort in estimating real world effects and to serve a tool for further improvement towards upcoming more stringent legislative emission limits. Therefore the adaption of an investigated Euro 3 motorcycle towards Euro 4 certification is presented as an example. Moreover there are presented tools to assign emission relevant scenarios to specific driving pattern and to assess them according to the vehicles driving dynamic.
Johannes Hiesmayr, Stephan Schmidt, Stefan Hausberger, Roland Kirchberger
Graz University of Technology
SAE/JSAE Small Engine Technology Conference