Increasing requirements for the reduction of fuel consumption (CO2) and emissions require a precise electronic management of combustion engines. Engine-related measures to meet these requirements lead to an increase in variability and system complexity. To cope with increasing system complexity, model-based development methodology has proven effective. In this context, virtual development with real-time models plays an increasingly important role. The corresponding models can either be derived theoretically on the basis of known physical laws (white-box models) or obtained experimentally on the test bench by mathematically modeling the measured input and output behavior (black-box models). Both types of modeling have their advantages and disadvantages.

A semi-physical modeling methodology is presented that combines the advantages of theoretical and experimental modeling and overcomes their disadvantages. The goal is to find suitable, simplified equation structures and to determine their unknown parameters experimentally by real-time capable, recursive parameter estimation methods. This leads to physically interpretable real-time models that are able to adapt their parameters according to the current engine operating behavior and thus offer good transferability to other engines. The semi-physical modeling methodology is applied to the air system and combustion of a common rail diesel engine with a variable exhaust gas turbocharger and high- and low-pressure exhaust gas recirculation. The focus lies on the derivation of semi-physical real-time model for the combustion and its underlying processes inside the cylinder.

A semi-physical model approach for modeling the dynamics of combustion chamber processes is developed and combined with state and parameter estimation methods. This model approach enables crank angle-resolved calculation of the in-cylinder gas states and the determination of the characteristic combustion components of diesel combustion (premixed, diffusive combustion and burn-out). The technical implementation is realized close to the pressure indication system of the engine test bench, enabling a crankshaft-resolved model adaptation based on measured in-cylinder pressure. Model identification is performed using combined state and parameter estimation in steady-state engine operation. Model parameters are estimated perpetually for each duty cycle and converge to a constant value within 30-60 engine duty cycles. Final estimation results are stored as functions of engine operating point using experimental modeling. In this way, semi-physical real-time models are created directly online during the measurement.

The treated method is considered as an extension of the functionality of conventional pressure indication systems. Furthermore, the derived semi-physical models are used for real-time engine simulation in the context of hardware-in-the-loop testing of ECUs.

The research project (Project No. 1231) was financially and advisory supported by the Research Association for Combustion Engines (FVV) e.V. (Frankfurt am Main, Germany).