Neural Predictive Controller Based Diesel Injection Management System for Emission Minimisation

Literature Review

The approach of employing neural networks for engine control applications for diesel powered vehicles was proposed earlier by Howlett et al. (1999). Neural network can control complex non-linear systems with little priori theoretical knowledge. The operating conditions of diesel engines are highly non-linear, due to rapid acceleration and deceleration. This requires quick and accurate control on fuel injection parameters such as injection duration, pressure and start of injection in a continuously varying manner, over a very short period of time. Use of neural networks for engine modelling and control was supported by Ouladsine et al. (2005) despite their strong dynamics and nonlinearities. A practical control technique based on neural network which is applied in online to engine was proposed by Omran et al. (2009).

Neural networks have wide applications in system identification and non-linear control applications. De Jesus et al. (2001) proposed a comparison of different neural network control algorithms like model predictive control, NARMA-L2 control, model reference control etc. An application for diesel engine speed control by varying fuel parameters was discussed. It shows that model predictive control is better to predict the behaviour of systems with variable time delay.

Artificial Neural Network (ANN) using Multi Layer Perceptron (MLP) network was proposed by Desantes et al. (2005) which has noticeable features to cope with problem of predicting emissions. As proposed by Obodeh and Ajuwa (2009) ANN based modelling does not require equations governing the engine performance but, the conventional map-based technique require. The above works illustrates that neural network is an effective tool for controlling emissions in diesel engine.