Developing a learning system capable of hypothesis justification [fault diagnosis]

Learning systems of all kinds are being investigated for potential uses in a wide variety of applications. These systems have a peculiar strength, that of abstracting a solution to a problem, increasing enormously the size of the domain of problems considerably. The wide variety of learning systems under investigation and development have individual benefits associated with their respective architectures, few however have a structure that permits an investigation into their processing by people other than their developers. Often the ability to offer some form of justification is a highly desirable feature. The paper offers some of the strong points of learning systems in general and proposes a learning system architecture with the ability to justify clearly the steps used to reach a result, in an industrial application. The application considered is fault diagnosis of a power network