Prediction of a diesel engine exhaust gases physical properties with artificial neural network

Abstract

In recent years, ANN (artificial neural network) method has been used as an effective method for analyses of the characteristic parameters in internal combustion engines. Also, determination of the best network structure is an important part of the research work in this branch. So, this subject is the main idea of the current study. The most reliable network structure has been determined for prediction of two important engine after-treatment parameters. These parameters are pressure and temperature of the gases at EVO (exhaust valve opening) time. Outputs of four ANN models have been compared with the results of a reliable developed multi-zone combustion model. The ANN models, which have been considered in this research work, are MLP (Multi Layer Perception), RBF (Radial Basis Function), SOM (Self Organized Map) and GFF (Generalized Feed Forward) with training algorithms of LM (Levenberg Marquart)and MOM (Momentum), respectively. Finally, the MLP-LM model has been proposed as the most appropriate model. Keywords-diesel engine; multi-zone combustion model; aftertreatment; artificial neural networkI.

Methodology for Virtual Prediction of Vehicle-Related Particle Emissions and Their Influence on Ambient PM10 in an Urban Environment

“…Both systems provide online measurements with a sampling rate of 1 Hz. Since exhaust emissions mainly occur in a size range below 100 nm, exhaust-PM was calculated based on the number concentration assuming a mean particle diameter of 100 nm and a density of 1 g/cm 3 .…”

“…The particle tracking is based on the discrete phase model (DPM). Therefore, spherical model particles (parcels) in a size range between 1 and 10 μm are tracked within the flow field while a particle density of 𝜌 𝑝 = 1 g/cm 3 is assumed (aerodynamic diameter). To account for turbulent dispersion, stochastic particle tracking was enabled (discrete random walk model).…”

“…Domingues et al used an artificial neuronal network (ANN) as well as symbolic regression in order to predict exhaust emissions such as O 2 , NOx and particle number [2]. Ghiasi et al applied an ANN for the prediction of parameters for an exhaust after-treatment system [3]. Ricciardi presented a semi-empirical brake wear model.…”