V. Karri, H.A. Hafez (Australia), and M. Kristiansen (Denmark)
Diesel, emissions, pollutant, predictions, empirical knowledge, environment, Bayesian network, ANN, PAH, Soot, PM, NOx, COx, SOx, HC.
A reliable estimation of diesel harmful emissions can provide useful information to scientists and environmentalists alike. Reliable attempts continue to progress toward predicting NOx, CO2, CO, Hydrocarbons (HC's), and particulate matter (MP), since the problem became known. Signing the Kyoto protocol by many developed countries will pressure efforts for Australian emissions reduction in general and diesel engine reduced emissions in particular. A new environmental standard requiring reduced emissions and particularly diesel Soot will come to effect in 2006. With Australian remote islands diesel infrastructure obliged to comply, a predictive model, of reliably estimating emissions, is becoming increasingly important. The quantitative predictive models available in the literature, uses empirical models and conventional approaches based on extensive experimental database. On the one hand, the validity of the empirical models is as good as the extent of the experimentation carried out. On the other hand, the cost of carrying out those experiments, in time and money, is becoming a burden for many researchers. The recent intelligent models based on artificial intelligence (AI), neural networks (ANN), and the Bayesian networks are thus becoming more popular. The latter models are used to model complex and intricate non-linear dynamic processes. In this paper, a Bayesian network is used to model emissions for a Lister-Petter diesel combustion engine. A limited number of experiment were performed, the data recorded and used to train, test and project emission levels, with probable accuracy, by the Bayesian network model. The obtained results were encouraging showing probable accuracy of 78% predicting the right levels of emissions. The model needs however, more refinements to obtain higher accuracy.
Important Links:
Go Back
