Common-rail pressure estimation using a Neuro-Fuzzy architecture with local Hammerstein models

Hydraulic processes with turbulent flow are usually highly nonlinear and common rail (CR) systems make no exception. Since the performances of diesel CR engines are directly dependent on the rail pressure, and on its values used in control, a prediction model which can lead to better performances is presented. The prediction makes use of Hammerstein dynamic models integrated into a multilevel Neuro-Fuzzy structure. The process input space decomposition is performed axis orthogonal for a large region using Local Linear Model Tree (LOLIMOT) algorithm and the local dynamic models parameters are adapted using recursive last squares method. The practical final results are favorable.