Automated Production of Information Models for Use in Modelbased Diagnosis

The most difficult part of system maintenance is diagnosing failures. System expertise is needed, but manuals cannot cover all the unique situations and in-house experts are difficult to retain because of today’s budgets. The resulting “expertise gap” has led to the development of “intelligent” diagnostic aids to compensate for the missing expertise. The first computer-assisted diagnostic aids attempted to capture and emulate the reasoning process of experts by encoding heuristics they used for diagnosis. Despite some early successes, capturing expertise is a high-risk process. An alternative approach to system maintenance uses a model of the system as a foundation for its knowledge base. This model-based approach differs from the rule-based approach in the way it incorporates and processes knowledge about the system. A particular type of model, the information flow model, is very useful for diagnosing problems. This approach, however, suffers from a complex model building process that is labor-intensive. Recently, the model building process has been automated by a tool that enables both top-down and bottom-up modeling. As discussed in this paper, this type of tool has the potential to enable widespread usage of the information modeling approach to system maintenance. We also discuss how the automated modeling approach can be used with machine learning techniques to enable the process to adapt with changing systems.