Identification of a data-driven dynamic model from virtual test bench results: an extended Hammerstein approach

Abstract

The JSAE & SICE Joint Research Committee on Control and Modeling virtual test engine was used as data source for a system identification and subsequent modeling attempt. The benchmark task was the identification of the air mass flow and the engine torque of an internal combustion engine over the whole operating range. The model represents the dynamics of a 4-cylinder Gasoline engine with a basic control unit with typical control inputs. After initial investigations into the dynamics of the engine model, initial data was acquired to gain insight into the system dynamics. Using DoE and dynamic modeling, a model was devised. It was shown that with reasonable modeling effort, the air mass flow could be modeled to an accuracy for the validation run of 1.2 g/s RSME, 2.9 g/s max, while the torque output could be estimated to within 4.6 Nm RMSE, 9.7 Nm max. Usage of a virtual test bench is beneficial to demonstrate benefits of new algorithms or methods inexpensively and efficiently. This exercise was undertaken by IAV within the benchmark challenge of the Japanese calibration user group, JCUG.