Fuel Type Recognition by Classifiers Developed with Computational Intelligence Methods Using Combustion Pressure Data and the Crankshaft Angle at which Heat Release Reaches its Maximum

A compression ignition engine powered by different fuels requires, for ecological reasons, adjusting the engine work to the injected fuel type. This paper presents a methodology for building classifiers that recognize the type of fuel injected to engine cylinders to accuracy sufficient in practical engine applications. These classifiers were developed using various computational intelligence methods: non-fuzzy classification trees, the proposed method combining several non-fuzzy classification trees into one fuzzy rule-based system, and another method that uses swarm algorithms to optimize classification tree parameters. Analysis of the in-cylinder pressure data measured for the engine running at full load allows calculating the angle at which maximum heat is released. The classifiers were built based on this angle and on other descriptors of pressure changes in the cylinder. Compared features included classifier accuracy, clearness and response time. The methods proposed required developing learning data sets based on experimental data. The measurement data from the tests conducted on an engine test bench were for the engine powered by five different fuels: diesel, fatty acid methyl esters of rapeseed oil, and three blends of these two fuels. The minimum number of consecutive engine cycles necessary to recognize the fuel type was discussed.