A Neuro-Fuzzy Based Method for the Design of Combustion Engine Dynamometer Experiments

Because of increasing requirements for low emissions and fuel consumption, combustion engines are getting more and more control inputs, like multiple injection, exhaust gas recirculation (EGR), turbocharger valve position (TVP), variable valve timing (VVT), etc. With the addition of manipulated variables, the required measurement time for obtaining the steady-state characteristics and control look-up tables rises exponentially.

A comprehensive design of the measurement experiment is becoming more and more essential. The objective is to measure the engine characteristics and properties with a minimum number of measurement points (with firstly concentrating on the stationary behavior).

A new methodology is presented to automatically determine characteristic mappings by incorporating prior knowledge. Since physical modeling of the engine behavior is mostly not appropriate, prior knowledge for experimental design is derived by evaluating measurement data.