Application of artificial neural network to classify fuel octane number using essential engine operating parameters

• Main Author: Ghanaati, Ali
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: TJ Mechanical engineering and machinery
• Online Access: http://eprints.utm.my/id/eprint/81797/1/AliGhanaatiPFKM2017.pdf; http://eprints.utm.my/id/eprint/81797/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:125970
• Institution: Universiti Teknologi Malaysia

Real-time fuel octane number classification is essential to ensure that spark ignition engines operation are free of knock at best combustion efficiency. Combustion with knock is an abnormal phenomenon which constrains the engine performance, thermal efficiency and longevity. The advance timing of the ignition system requires it to be updated with respect to fuel octane number variation. The production series engines are calibrated by the manufacturer to run with a special fuel octane number. Presently, there is no research which takes into account the fuel tendency to knock in real-time engine operation. This research proposed the use of on-board detection of fuel octane number by implementing a simple methodology and use of a non-intrusive sensor. In the experiment, the engine was operated at different speeds, load, spark advance and consumed commercial gasoline with research octane numbers (RON) 95, 97 and 100. The RON classification procedure was investigated using regression analysis as a classic pattern recognition methodology and artificial neural network (ANN) by executing combustion properties derived from in-cylinder pressure signal and engine rotational speed signal. The in-cylinder pressure analysis illustrated the knock-free, light-knock and heavy-knock regions for all engine operating points. The results showed a special pattern for each fuel RON using peak in-cylinder pressure, maximum rate of pressure rise and maximum amplitude of pressure oscillations. Besides, there is a requirement for pre-defined threshold or formula to restrict the implementation of these parameters for on-board fuel identification. The ANN model efficiency with pressure signal as network input had the highest accuracy for all spark advance timing. However, the ANN model with rotational speed signal input only had the ability to identify the fuel octane number after a specific advance timing which was detected at the beginning of noisy combustion due to knock. The confusion matrix for the ANN with speed signal input had increased from 68.1% to 100% by advancing the ignition from -10° to -30° before top dead centre. The results established the ability of rotational speed signal for fuel octane classification using the relation between knock and RON. The implication is that all the production spark ignition engines are equipped with engine speed sensor, thus, this technique can be applied to all engines with any number of cylinders.