Using Neural Networks for Air-to-Fuel Ratio Estimation in Two- Stroke Combustion Engines

To be able to meet the demands of tomorrow on lower emissions from small two-stroke engines, used e.g. in chain-saws, there is a need to enhance the control over the combustion. One interesting parameter is the air-to-fuel ratio (A/F). If A/F can be measured, then it is possible to intelligently control the fuel, and thus obtain a desired A/F. This is interesting in combustion engine control, by the fact that the emissions and the power from the engine directly depend on the current A/F. Most of the larger engines mounted in automobiles today have a sensor (the so called lambda-sensor) measuring A/F in the exhausts. This sensor has a number of problems: it is expensive, slow, sensitive to pollution and, most important of all, it gives only a binary output (indicating whether A/F is above or below a factory set value). With this background, it is motivated to seek for other ways of measuring A/F. In this study a method using ion-current measurements and artificial neural networks to estimate A/F is developed and evaluated. The ion-current measurement is taken from the cylinder (via the spark plug) of a two-stroke chain-saw engine during the combustion cycle. This signal gives an instantaneous measure of the conductivity in the cylinder. Using this measure, two cascaded neural networks have been trained to estimate the air-to-fuel ratio. The tests have also shown that it is possible to extract other information from this signal, like misfiring, the fuel quality, and others. The result should be seen as a first step towards a complete, self-tuning engine control system.