Engine management systems in modern motor vehicles are becoming increasingly extensive and complex. The functionality of the control units which are the central components of such systems is determined by the hardware and software. They are the result of a lengthy development and production process.
Road testing of control units, together with testing them on the engine test bench, is very time consuming and costly. An alternative is to test control units away from their actual environment, in a virtual context. This involves operating the control unit on a Hardware-in-the-Loop test bench.
The control unit's large number of individual and interlinked functions necessitates a structured, reproducible test procedure. These tests can, however, only be conducted once an engine prototype has been completed, as the parameters for the existing conventional models are determined from the data measured on the test bench. The model depth is a further aspect: today, models are based on data that are averaged over a cycle. Investigation e. g. of the momentary acceleration of crankshafts is thus not possible.
This article, based on work conducted jointly by BMW AG, Munich, and the Technical University of Dresden, consequently investigates strategies for the model based testing of engine management systems. The potential for time-saving with a new model of considerable depth is also considered here. The representation of certain subsidiary functions of the real-time model by means of neural networks is in addition investigated.