Neural Model for Real-Time Engine Volumetric Efficiency Estimation 2013-24-0132

Increasing the degrees of freedom in the air path has become a popular way to reduce the fuel consumption and pollutant emissions of modern combustion engines. That is why technical definitions will usually contain components such as multi or single-stage turbocharger, throttle, exhaust gas recirculation loops, wastegate, variable valve timing or phasing, etc. One of the biggest challenges is to precisely quantify the gas flows through the engine. They include fresh and burnt gases, with trapping and scavenging phenomena. An accurate prediction of these values leads to an efficient control of the engine air fuel ratio and torque. Fuel consumption and pollutant emissions are then minimized.

In this paper, we propose to use an artificial neural network- based model as a prediction tool for the engine volumetric efficiency. Results are presented for a downsized turbocharged spark-ignited engine, equipped with inlet and outlet variable valve timing. The calibration process that is used in this study only requires steady-state operating points. The validation stage was conducted on both steady-state and vehicle transients. Model prediction is in very good agreement with experimental results while keeping a very low calibration effort and matching embedded computational requirements. The conclusion stresses that thanks to their generic structure, neural models offer an interesting potential for generalization to even more complex technical definitions.