Search Results

A neural network-based engine performance, fuel efficiency and emissions prediction system has been developed for both spark-ignited and compression ignition engines. Through limited training on an engine dynamometer, the neural network system is able to predict accurately real-time engine power output, fuel consumption and regulated exhaust emissions using readily measured engine parameters, across highly transient engine operating cycles. Applications for the models developed using this process include engine diagnostics, virtual sensing of unmeasured or unmeasurable engine emissions, engine control, and engine and vehicle modeling. Results from the prediction of the performance and emissions from a 300 hp CIDI engine and a 120 hp SI engine are presented, showing the potential of this newly developed approach.

A project has been undertaken to optimize the engine control software calibration of a modern heavy-duty diesel engine for operation with gas-to-liquids (GTL) diesel fuel, with the objective of developing an understanding of the scope for optimization with this fuel, which has different physical and combustion properties to that of conventional, crude-derived diesel. A data-driven, model-based calibration technique utilizing artificial neural networks was used to develop optimized transient and steady-state calibrations with both conventional diesel fuel, as well as neat GTL fuel. The engine control parameters that were optimized were injection timing, exhaust gas recirculation rate, rail pressure, and charge mass. The optimization aimed to minimize fuel consumption without deterioration in engine-out nitrogen oxide (NOx) and soot emissions. This paper reports on the calibration optimization methodology employed and the results achieved to date.

With the adoption of complex technologies such as multiple injections, EGR and variable geometry turbocharging, it has become increasingly onerous to develop optimal engine control calibrations for either light- or heavy-duty diesel engines. The addition of NOx and PM aftertreatment systems increases further the calibration burden, as both diesel particulate filters and NOx absorbers require regeneration initiated by the engine management system. There is significant interest in the industry in reducing development costs by moving as much of the engine calibration process as is feasible from the engine test cell to the virtual desktop environment. This paper describes the development of a model-based calibration optimization system that offers significant advantages in reducing the time and effort required to obtain certification-quality engine calibrations.

This paper explores the feasibility of using in-cylinder pressure-based variables to predict gaseous exhaust emissions levels from a Navistar T444 direct injection diesel engine through the use of neural networks. The networks were trained using in-cylinder pressure derived variables generated at steady state conditions over a wide speed and load test matrix. The networks were then validated on previously “unseen” real-time data obtained from the Federal Test Procedure cycle through the use of a high speed digital signal processor data acquisition system. Once fully trained, the DSP-based system developed in this work allows the real-time prediction of NOX and CO2 emissions from this engine on a cycle-by-cycle basis without requiring emissions measurement.