Black box modeling for engine control and emissions
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
The 7th International Advanced Engine Control Symposium 2014
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
Many control and monitoring methods applied in automotive applications require models. This implies the availability of suitable models which can be utilized for control purposes. Besides models based on first principles, which are often time consuming to parameterize for complex processes, data based models can provide an alternative. These models are called black box models, because no physical structure is assumed. The models are determined directly by identification from suitable measurement data and of course the choice of identification data has strong effects on both, the achievable quality and the required amount of data and consequently measurement time.
Within this talk we will cover basics and challenges of data based model identification with links to applications in the engine field. The goal is to present a straightforward and efficient way to determine models applicable for control or virtual sensors. To this end, an iterative identification method for polynomial NARX models, based on an iterative DOE approach, will be presented. During the iteration steps, the model complexity is iteratively increased, starting from the simplest case, until a desired validation result is reached. This method will be presented in detail for two different real-world application examples, namely the modeling of a Diesel engine air system and NOx emissions. Further examples will provide an outlook on additional applications, like cylinder pressure based models, and the applicability of such models for MPC based air system control strategies.