The Application of Artificial Neural Network in Predicting and Optimizing Power and Emissions in a Compressed Natural Gas Direct Injection Engine 2007-01-4264

This paper describes the application and capability of neural network as an artificial intelligence tool to determine the performance and emissions in a compressed natural gas direct injection (CNG-DI) engine. A feed-forward back-propagation artificial neural network (BPANN) approach is explored to predict the combustion performance in terms of indicated power and emissions in the appearance of CO and NO emissions level. A series of numerical computations by means of computational fluid dynamics (CFD) code were carried out based on the statistics-based design of experiment method. The data for combustion process under various engine operating parameters at the fixed speed at 1000 rpm were obtained to train the developed artificial neural network (ANN). The operating conditions employed to represent the combustion parameters for controlling the injection and ignition event are start of injection (SOI), end of injection (EOI) and spark advance (SA) timing, which affects to the combustion processes, performance as well as emissions formation. The developed ANN was identified as a black box model with input and output data, which does not require priori knowledge. There were 15 data acquired from the CFD combustion simulation to characterize the combustion behavior of such engine used for training process of ANN, and 5 test data used for verifying the model. The study results showed that the predicted results were in good agreement with the full CFD simulation. This circumstance proves that the developed ANN model has the competence to successfully predict the performance and emissions of CNG-DI engine.