The aim of this study is to realize a virtual combustion sensor, that is, a “grey-box” model able to forecast the rate of heat release (ROHR) in a common rail diesel engine, supplied by a modulated injection rate. The model has the following inputs: engine speed, injection pressure, environment conditions and control parameters of the fuel split injection. The idea behind model development is to research ROHR's discriminating features on a Wiebe functions basis using evolutionary algorithms. After this we used a clustering algorithm to find the optimal data set with which we trained the neural network which represents our “grey-box” model.
The ROHR model could be used as a virtual combustion sensor in a model based control system for the real-time updating of control parameters. Moreover, it can be used to develop hardware-in-the-loop diesel engine simulation systems. The ROHR model is global, portable, and multi-resolution. Global means the model can forecast the time averaged ROHR for each engine operating condition. Portable means it is applicable to any diesel engine. Multi-resolution means the user can improve model accuracy by increasing the number of Wiebe functions used to break down the ROHR signal into the main components.
As a test case, we applied ROHR models to forecast combustion processes in a small displacement mono-cylinder diesel engine with a common rail injection system. Results are shown on forecasting the time averaged cylinder pressure with a very low error.