Development of Automated Mapping/Calibration Technique for Complex Engine Control Systems, 03-9323Printer Friendly Version
Inclusive Dates: 07/01/02 - Current
Background - In the pursuit to simultaneously meet the often conflicting emissions regulations and consumer demands, modern engines must incorporate a number of new subsystems; e.g., advanced fuel systems, exhaust treatment devices, camless valvetrains, etc. However, with the increasing number of subsystems, it is difficult to make a simple one-to-one association between actuators and performance measures. From the point of view of a control system designer, it is important that the cross coupling between actuators and performance measures be characterized (mapped). While such characterization is important in itself, it is but a step towards the final goal of determining setpoints for various actuators to achieve overall performance that is optimal in some sense.
The traditional full-factorial method for mapping/calibration can be shown to require years worth of effort for a modern enginea proposition impractical both from the time and cost points of view. This internal research program recognizes that it is the optimization stage that is important and aims to primarily address it without the expensive brute-force full-factorial method. The goal is to design efficient and scalable procedures to the subject optimization problem.
Approach - "Design of Experiments" (DOE) has been gaining wide acceptance as a structured method for determining the relationship(s) between factors affecting a process and the output(s) of that process. One of the approaches we are evaluating uses DOE to determine unknown parameters of physically based models of the engine processes. We then will use these "trained" models to perform searches in the space of actuator setting. Most of these searches will not require the engine to be run except in the final stages of verification. Failure at this stage is used to adaptively re-train the process models. The other methods we are exploring rely on techniques of global optimization that require no derivative information. Memetic and genetic algorithms are being considered.
Accomplishments - Exhaustive mapping at 8 modes specified for Type C1 of ISO 8178 standard is complete and will be used as a benchmark for the other model- and population-based search methods.