On the application of artificial neural networks for the prediction of NO_x emissions from a high-speed direct injection diesel engine

Abstract

This article considers the application and refinement of artificial neural network methods for the prediction of NO x emissions from a high-speed direct injection diesel engine over a wide range of engine operating conditions. The relative computational cost and performance of two backpropagation algorithms, Levenberg–Marquardt and Bayesian regularization, for this application are compared, with the Levenberg–Marquardt algorithm demonstrating a significant cost advantage. This work also assesses the performance of two alternative filtering approaches, a p-value test and the Pearson correlation coefficient, for reducing the required number of input variables to the model. The p-value test identified 32 input parameters of significance, whereas the Pearson correlation test highlighted 14 significant parameters while additionally providing a ranking of their relative importance. Finally, the article compares the predictive performance of the models generated by the two filtering methods. Overall, both models show good agreement to the experimental data with the model created using the Pearson correlation test showing improved performance in the low-NO x region.

Artificial neural network (ANN) assisted prediction of transient NOx emissions from a high-speed direct injection (HSDI) diesel engine

“…25 The detailed process of feature selection is highlighted in our previous study. 19 The results suggest 14 input features in the dataset are needed for the construction of an ANN NO x model. A brief discussion of each parameter chosen is included in this article.…”

“…However, a systematic overprediction of CoV was observed for low CoVs while higher CoVs were underpredicted by the ANN model suggesting missing physical parameters for the ANN input features. More recently Fang et al 19 In this study, we explore the applicability of the ANN method for the prediction of NO x emissions of a high-speed direct injection diesel engine undergoing transient load steps. The model is built from a substantial experimental dataset, which includes 7 months of engine testing (1108 individual experiments) from the University of Oxford single-cylinder diesel research engine running under various steady-state conditions.…”

“…The engine and the test cell have been designed to give the highest quality data, and much of this dataset has already been published. [20][21][22] The previously constructed ANN model based on this steady-state state dataset 19 is given transient input features including crank angle resolved cylinder peak pressure to predict the transient NO x emissions. The predicted transient NO x emissions are then quantitatively compared both with results from a conventional test-bed emissions analyser and a fast-NO x emissions analyser which have been published previously.…”

A new method to determine the causes of deviation in cylinder pressure curves of motored reciprocating piston engines

“…The optimisation results were then validated against test bench data. Fang et al 27 used filtering techniques to identify significant parameters in combination with neural networks to predict the nitrous oxide emissions of a diesel engine. Further, specialised approaches of modelling include the prediction of the cylinder peak pressure using a high frequency crank angle measurement in Dunne and Bennett 28 or a two stage calibration approach for a two step chemical reaction mechanism by Men et al 29 A comprehensive literature study on the model-based diesel engine calibration has been conducted by Nikzadfar and Shamekhi.…”

Prediction of NOx Emissions for a Range of Engine Hardware Configurations Using Artificial Neural Networks

“…The study focused on identifying the most important engine parameters for accurate NO x prediction over a wide range of operation. This study highlighted 14 parameters that have the biggest effect on NO x , based on a Pearson correlation analysis, and shown that ANN models are capable of accurately predicting NO x emissions even when tested outside the training range of the model [24] Despite their powerful generalisation abilities, like other empirical models, their performance deteriorates when deployed outside their training range. By using the widest limits of the data during training provides the best chance of extrapolating for an ANN model.…”