One of the most critical challenges currently facing the diesel engine industry is how to improve fuel economy under emission regulations. Improvement in fuel economy can be achieved by precisely controlling Air/Fuel ratio and by monitoring fuel consumption in real time. Accurate and repeatable measurements of fuel rate play a critical role in successfully controlling air/fuel ratio and in monitoring fuel consumption.
Volumetric and gravimetric measurements are well-known methods for measuring fuel consumption of internal combustion engines. However, these methods are not suitable for obtaining fuel flow rate data used in real-time control/measurement.
In this paper, neural networks are used to solve the problem concerning discontinuous data of fuel flow rate measured by using an AVL 733 s fuel meter. The continuous parts of discontinuous fuel flow rate are used to train and validate a neural network, which can then be used to predict the discontinuous parts of the fuel flow rate. A comprehensive and detailed neural network (NN) modeling technique for engine application is presented. The experiments in this study are designed to collect informative data for steady states, where the speed and load are fixed. Different neural network structures are compared, including feed-forward NN, feed-forward NN with delays and non-linear autoregressive model with exogenous inputs (NLARX). The NLARX is determined to be the best structure for fuel flow rate prediction. The training process and input choices for catching fast and slow dynamics are illustrated in the paper. Additionally, for real-time control and measurement purposes, a fuel flow rate model is maintained to be as simple as possible by minimizing the structure based on NN identifiable properties. This study shows that NLARX could accurately predict fuel flow rate with a R-square above 0.99.
