Virtual Sensors for Safe Operation of Electrolyser and Hydrogen-powered Car

The rapid depletion of finite fossil fuel resources and the increase in greenhouse gasses due to fossil fuel use has led to greater efforts into the search for alternative fuels for transportation. Many studies of alternative fuels have shown that hydrogen is one of the clean and inexhaustible fuels that can become a part of the overall solution to the transportation fuel debate. This thesis covers the research of a totally closed loop system of producing hydrogen from electrolysis as well as the use of hydrogen as a fuel in a converted standard gasoline internal combustion engine car. Due to the increase in use and application of hydrogen as an energy carrier for stationary and mobile applications, there is increasing pressure to ensure the safe handling and monitoring of this highly combustible gas. The associated equipment to monitor and measure the explosion limit of any leakage together with the pressure and flow rate is very expensive. Any reliable mathematical or empirical means to estimate and predict those safety features of hydrogen will greatly assist in avoiding expensive instrumentation. In this thesis, Artificial Intelligent predictive models including Artificial Neural Networks (ANNs) and Adaptive Neuro-Fuzzy Inference System (ANFIS) are presented as virtual sensors for the accurate estimation of hydrogen parameters such as the percentage lower explosive limit, hydrogen pressure and hydrogen flow rate. These parameters are shown as a function of different input conditions of the power supplied (voltage and current), the feed of de-ionised water and various system parameters of a proton exchange membrane electrolyser. All results are obtained from experimental tests of the Hogen¬¨vÜ20 electrolyser. This includes the installation of de-ionised water and electrolyser sensors; National Instrument data acquisition system; LABVIEW user interfaces, as well as utilising Windows Diagnostic Software Operation. It was shown that, using neural networks, hydrogen safety parameters were predicted with an inaccuracy of less than 5% of percentage average root mean square error. The outcomes of the system are useful, as virtual sensors, for operators to monitor hydrogen safety on the shop floor. This is also an innovative way of avoiding expensive instrumentation for hydrogen monitoring. A good understanding and parallel progress of the hydrogen internal combustion engine technology is essential when the unit cost of hydrogen production is affordable. One of the v major aspects of an internal combustion engine running on hydrogen is system calibration. It is required to build appropriate engine tuning maps for ignition and injection timing to achieve smoother knock-free combustion. This aspect is particularly exasperated when tailormade engine management systems are integrated into the hydrogen internal combustion engine without prior matrix of data to refer to for online tuning. While the process of exhaustive experimentation and subsequent table-filling processes are time-consuming, an intelligent system-estimating ignition and injection timing with respect to the engine desired performance parameters will be of great use in the eventual fine-tuning. The research presents the implementation of different intelligent fine-tuning methods for the conversion of a Toyota Corolla four-cylinder, 1.8-litre hydrogen-powered car. This is done through the investigation of past research, theoretical analysis and the creation of a fuzzy expert system as well as Model-based Calibration and Calibration Generation techniques that use engine parameters and control variables. While traditional control methodology does not cope with the exponential increases in control parameters and the algorithm complexity of hydrogen internal combustion engines, the research methodology determines the optimal of hydrogen engine performances in terms of power, fuel economy and emissions, in a smart process that can be seen as a generic application for all current and future hydrogen internal combustion engine control system parameters. All results achieved from the real tests are used to investigate the application of artificial intelligent models as virtual sensors to predict relevant emissions of the converted hydrogenfuelled car. Emissions were predicted with an inaccuracy of less than 4% of percentage average root mean square error. These predictions are based on the study of the qualitative and quantitative effects of engine-operating parameters on the harmful exhaust gas emissions. This study is a step towards a full understanding of hydrogen production, modelling of virtual sensors on a sound mathematical basis including hydrogen-fuelled engines, their performances and associated environmental benefits.