Towards low-cost fault diagnosis in large component-based systems

Abstract

The problem known as fault diagnosis in discrete-event systems is to determine a method for monitoring large systems and efficiently performing diagnosis given the flow of observations. The classical model-based approach for monitoring discrete-event system is the diagnoser approach. However, its computation is based on the global model of the system, so it is exponential to the number of components of the system. This space complexity makes this diagnoser infeasible for large component-based systems that are more and more common in real-world applications. This chapter proposes a new diagnoser approach for component-based systems based on a set of specialized diagnosers whose computation is less complex. The diagnose is devoted to the diagnosis of one type of fault only (one diagnoser per fault). Instead of taking into account the system as a whole, the chapter proposes analyze the system to detect a subsystem that is sufficient for diagnosing this particular type of fault. In practice, the purpose of this approach is to drastically decrease the computation cost of any monitoring agent for component-based systems.