Using Chip Simulation to optimize Engine Control
Matthias Simons, Mihai Feier, Jakob Mauss · 2013
Calibration of engine controllers can greatly benefit from mathematical optimization. This requires however an executable model of the ECU functions. The following problem arises: Engine calibration is typically performed by an OEM, while the ECU code is owned by the supplier of the ECU. Therefore, the OEM is typically unable to set up an ECU simulation based on the original C code of the ECU. Instead, to set up optimization on PC, time consuming and error prone reverse engineering is needed to develop an 'equivalent model' of the ECU function of interest. To deal with this situation, we have implemented a novel method for automating the calibration of engine parameters. The method combines two ideas • simulation of ECU program code on PC using chip simulation • mathematical optimization based on the resulting executable model The simulation requires only the hex, ASAP2/a2l and map file, that an OEM typically has access to, but not the source code of the ECU functions of interest. This paper describes also a problem that we encountered when coupling chip simulation with optimization methods that require gradients to guide search for optima: partial derivatives of engine functions with respect to engine parameters are zero or infinite then, due to integer data types used by typical engine control code. As a result, gradient based optimization methods lack guidance and tend to terminate with sub-optimal solutions. The paper also sketches ideas how to overcome this problem and presents results of numerical experiments.