I.1. Subject matter

Global emissions of greenhouse gases covered by the Kyoto protocol reached approximately 49 billion tonnes of CO₂ equivalent in 2010. They increased by 80% between 1970 and 2010 and by 30% since 1990 [FRA 15]. The large majority of these emissions are obviously related to the combustion of fossil energies. The resulting pollution and the irreversible depletion of fossil resources should stimulate a reflection on our global energy system, which seems on its last legs. Many measures that have been taken through various programs, such as sustainable development, Horizon 2020 and even COP21, aim to decrease greenhouse gas emissions, reduce energy consumption, stabilize global warming, diversify primary energy sources or even develop renewable energies. Taking into account these elements forces the system into a new energy transition and tends to increase the share of renewable energies in the energy mix.

Power generation using wind and solar energies is reaching maturity, with ever-increasing production. Nevertheless, irregularity in the availability of these renewable energies gives rise to many problems in terms of power grid management. Managing a grid that day by day involves slightly more intermittent energies requires availability of reserve storing and release in order to cope with the differences between supply and demand in a given territory. It is indeed important to be able to deal with various time scales: daily (day/night), weekly (weekend/weekdays) and seasonal (summer/winter). Yet, it proves that hydrogen is ideal for storing this renewable energy. It can be produced using water electrolysis, which is itself fed by renewable energies. The resulting gas is then stored under various forms in order to be used at the appropriate time. Hydrogen can be combined with natural gas in existing networks, can be directly used in the industry or can even be used to again generate electric power by means of a fuel cell.

Since the fuel cell (FC) is at the center of the research activities presented in this book, it is worth our full attention. Indeed, this electrochemical generator converts the chemical energy of a reaction between a fuel (hydrogen) and an oxidizer (oxygen in the air) to generate electricity, heat and water. It can be used for stationary or nomadic applications or in transport. Moreover, in order to decarbonize the transport, the use of the hydrogen energy vector in combination with a fuel cell becomes an increasingly mature solution. As a proof, many global car manufacturers are manufacturing and marketing their own fuel cell vehicles. The most recent example is that of the Japanese manufacturer Toyota, which sold approximately 1,000 units in 6 months and is expecting to reach 30,000 fuel cell vehicles per year from 2020.

In Japan, except for stationary applications, more than 200,000 fuel cell systems have so far been installed in private homes to generate electricity and heat.

Hydrogen can be used in all sectors to store intermittent energy, supply mobile applications, meet the needs of an isolated site, which is disconnected from the power grid, propel a car, railway, naval or space flight transportation means and so on. In 1874, in his book The Mysterious Island, Jules Verne had predicted without an artificial intelligence algorithm that: “water will one day be employed as fuel, that hydrogen and oxygen which constitute it, used singly or together, will furnish an inexhaustible source of heat and light, of an intensity of which coal is not capable” [VER 74]. Half a century later, the rise of a hydrogen society is conceivable.

This book essentially deals with the problems related to fuel cell technology for energy generation. Even though the FC has experienced much progress in the last 20 years, the technical challenges have not yet been fully solved. The research work described in this book falls within this context, drawing on the works that relate to electrical engineering and, to a lesser extent, to the automation field.

I.2. Chapter breakdown

The works presented are centered on three axes organized in four chapters.

The first chapter starts with a brief overview of the current energy model. A review of the solutions that could enable the decarbonization of the energy mix is then provided. Among these solutions, a proposed focus is on the hydrogen vector, which is closely linked with the fuel cell generator. This chapter ends with the presentation of the fuel cell and its applications. Potential applications for transport and for the stationary field are highlighted. Finally, advantages and drawbacks of this technology are described. The FC presents problems related to cost, efficiency, lifetime, integration and hydrogen storage, but it has the advantage of releasing few or no pollutants, being silent, having high energy efficiency and offering substantial lifetime and reduced cost. Nevertheless, the FC is a multi-physical generator that becomes difficult to understand when considered with its auxiliaries.

The second chapter presents various existing types of FCs and identifies two technologies, namely the PEMFC (proton exchange membrane fuel cell) and the SOFC (solid oxide fuel cell), which are considered to be the most promising solutions for transportation and stationary applications. Regardless of the technology used, many auxiliaries are needed to operate the FC and it is important to control the whole system. Hence, all the auxiliaries are described and particular attention is given to the compressor unit that allows the air supply of the PEMFC, which is one of the largest energy consumers of the system (about 10–15%). With regard to the SOFC, which mainly differs from the PEMFC by its operating temperature (above 800°C), an integrated power generator is presented. For both technologies, experimental implementation is a key point, as it facilitates the understanding of how the FC and its system operate. These works can be considered the first axis, on which the research activities needed to obtain an augmented durability FC system rely. The second axis deals with the energy optimization of various hybrid energy sources, which enables an increase in FC lifetime. Finally, the third axis is dedicated to FC diagnosis and prognosis, which moreover enable the expansion of lifetime and an increase in generator reliability.

Thus, the third chapter deals with the hybridization of energy sources. If a transport application is considered, it is important to hybridize the FC with other sources, as the overall dynamics of the FC system is rather slow and causes difficulties in the case of rapid transients within the mission. Furthermore, in order to reduce the overall energy consumption of a vehicle, braking energy should be recovered, which is impossible with a single FC, as it is a non-reversible source. Finally, hybridization of this FC with energy-storing devices should be considered, in order to obtain high overall efficiencies. In such hybrid systems, an energy management strategy must be developed so that power demands are allocated to the right sources, at the right times, depending on the mission profile to be followed. Two approaches are proposed in this chapter. The first approach, which relies essentially on the wavelet transform combined with neural networks, enables the development of online energy management applied to a heavy hybrid vehicle composed essentially of a FC system, accumulators and ultracapacitors. It is worth noting that this energy distribution takes into account the frequency ranges of various sources using only the current and previous data of a univariate signal of the vehicle power demand. This will eventually allow real-time implementation as well as an increase in FC lifetime. The second approach relies on type-2 fuzzy logic and genetic algorithms. This solution is applied to a hybrid locomotive, composed of accumulators, ultracapacitors and a diesel engine, in order to minimize fuel consumption. The results show that the characteristics and dynamics of various sources have been taken into account thanks to artificial intelligence tools. High-performance energy management tools are proposed here, allowing optimal use of sources. Nevertheless, our efforts should focus on FC lifetime and reliability, in order to have a hybrid system with even higher performances.

For this purpose, FC diagnosis and prognosis have been developed in the fourth chapter. These two disciplines should enable us to reach our objective of increasing the lifetime and reliability of the FC. Diagnosis methods are used to identify the origin of failure and to determine the FC state of health, thus allowing for decision-making that ensures proper operation of the generator. PHM (Prognostics & Health Management) helps to predict the evolution of FC behavior in order to estimate a future failure. There are many approaches to the development of diagnosis and prognostic tools. Nevertheless, in order to find effective solutions, first it is important to have in-depth knowledge of the degradation mechanisms of the FC and its system. Thus, in the first part of Chapter 4, degradation mechanisms and the failures that the FC and its system may present are reviewed. In the second part, two diagnosis methods are presented. The first one (data-based) is a supervised classification method called “k-nearest neighbors”. The second (signal-based) relies on a wavelet transform approach. These approaches have in particular enabled the diagnosis of various FC system failures that may occur with good classification rates reaching 90%. As will be shown, it is also possible to estimate the FC state of health using the wavelet transform coupled with energy indicators. Finally, in the last part of this chapter, methodologies related to the data-based prognostic are developed. The tools used to monitor the state of health and estimate the future behavior of the FC are based on neural networks. The first works are carried out using the ANFIS (Adaptive Neuro-Fuzzy Inference System). This model enables us to predict the behavior of voltage evolution. Nevertheless, the learning bases prove to be substantial. To reduce the required data and achieve better prediction performances, we have subsequently used the ESNs (echo state networks), which are new systems of neural networks. The conventional neural networks are often costly in terms of computation time due to algorithmic complexity. For the ESNs, algorithmic complexity is replaced by structural complexity, so that the learning phase is faster than that of conventional neural networks. Thanks to the use of ESNs of genetic algorithms and the wavelet transform, the results obtained are very encouraging. Indeed, they allow us to predict cell voltages until the end of life of the FC with an error below 10%.