Thesis (M. Eng.) - Asian Institute of Technology, 2011

A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Mechatronics, School of Engineering and Technology

Neural networks for identification and control of a spark ignition engine using natural gas can be improved its dynamic. Basically, converted engines result in a significant power loss. The fuel quality variation and low volumetric efficiency of natural gas require numerous modification and calibration efforts, but the optimum operations can not be guaranteed. The vehicle was op~rated on both a test bench and practical driving cycles. The parameters such as throttle position and fuel injection duration were excited, while an oxygen concentration was observed over the whole operating range. The neural network has capability to identify the relation between natural gas injection duration and the engine states. According to these experiments, the correctness of neural network is between 91.2% to 93.4%.Therefore, the injection duration for each operating point was calculated from the neural network model. A interpolated look-up table is suitable for a fuel injection control. In this case, a test vehicle is Toyota Altis 1.6 1, fourcylinder engine equipped with compressed natural gas components. In additional, I adopt an ARM 32-bit microcontroller as a rapid prototype. The embedded controller contained the look-up table that used during a transient period. For steady-state responses, the system was combined with a PI controller. Thus, the experimental results can be confirmed the effectiveness of vehicle responses and also reduction of the calibration efforts. Finally, this similar methodology can be applied to other kinds of engines or fuels and allow inexpensive adjustments.