Identification and Validation of an Air Mass Flow Predictor Using a Nonlinear Stochastic State Representation

The control of an optimal combustion requires an estimation of cylinder air mass (m_air), and the calculation of the injected fuel mass. However, the low bandwidth of the engine process and the fuel injector impose a 2 Top Dead Center (TDC) predictive computation of m_air. The elaboration of an accurate predictor requires the determination of an exact and robust « intake manifold » model. Thus, the model has to be validated and qualified experimentally, even though no m_air sensor and no physical model currently exist. From a behavioural model tending toward the real « intake manifold » behaviour, this study describes the determination of a stochastic state model. The non linearity of the process is treated, as well as the « no computation » characteristic of the internal combustion engine map. The numeralization method of the state system respects the real time constraint of the engine (engine speed interval [Niddle; 7000rpm]). This model is identified and validated on a PSA engine, as a Xk=manifold pressure estimator. The identification of state noise covariance and measurement noise covariance allows definition of an acurate and fast respond time estimator. Then, a 2 step predictor is elaborated, by integration of the stochastic state and measurement noise variables: average and covariance. The identification and validation has been done on the engine, and leads to a quality equivalent to that of the estimator. Thus, the “change base” operator allows, using the model, construction of m_air estimator/predictor with similar qualities.