Small engine load estimator for fuel injection system using two-stage neural network

• Main Author: Muslim, Mohd. Taufiq
• Format: Thesis
• Language: English
• Published: 2016
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utm.my/id/eprint/60719/1/MohdTaufiqMuslimPFKE2016.pdf; http://eprints.utm.my/id/eprint/60719/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:94102

Most motorcycles in developing countries use carburetor systems as fuel delivery method especially for models with cubic capacity of less than 350 cc. However, small gasoline carbureted engines suffer from low operating efficiency, high fuel consumption and high level of hazardous emissions. In recent years, Electronic Fuel Injection (EFI) technology has been applied to small engine motorcycles as well. EFI system has better fuel economy and can reduce harmful emissions by correctly calculating suitable amount of fuel to be injected into the combustion chamber. One way to achieve this is by accurately estimate the engine load by using the in-cylinder Air Mass Flow (AMF) rate of the engine. Most of the control schemes in modern system either approximate the AMF near the throttle plate using Mass Air Flow (MAF) sensor or in the intake manifold using Manifold Absolute Pressure (MAP) sensor. This work presents a more economical approach to estimate the AMF by using only the measurements of throttle position and engine speed, that is, without using the MAF sensor or the MAP sensor to estimate the AMF in intake manifold, resulting in lower implementation cost. The estimation is done via two-stage multilayer feed-forward neural network with combinations of Levenberg-Marquardt (LM) algorithm, Bayesian Regularization (BR) algorithm and Particle Swarm Optimization (PSO) algorithm. Based on the results in 20 runs, the second variant of hybrid algorithm yields a better network performance with a mean squared error (MSE) of 1.8308 by estimating the AMF closely to the simulated AMF values compared to using the first variant of hybrid algorithm (MSE of 2.8906), LM (MSE of 8.0525), LM with BR (MSE of 3.5657) and PSO (MSE of 133.7900) alone. By using a valid experimental training data, the estimator network trained with the second variant of the hybrid algorithm showed the best performance, with MSE of 1.9863, among other algorithms when used in an actual small engine fuel injection system.