Exploiting structural abstractions for consistency based diagnosis of large configurator knowledge bases

Debugging, validation, and maintenance of configurator knowledge bases are important tasks for the successful deployment of product configuration systems, due to frequent changes (e.g., new component types, new regulations) in the configurable products. Model based diagnosis techniques have shown to be a promising approach to support the test engineer in identifying faulty parts in declarative knowledge bases. Given positive (existing configurations) and negative test cases, explanations for the unexpected behavior of the configuration systems can be calculated using a consistency based approach.

For the case of large and complex knowledge bases, we show how the usage of hierarchical abstractions can reduce the computation times for the explanations and in addition gives the possibility to iteratively and interactively refine diagnoses from abstract to more detailed levels. Starting from a logical definition of configuration and diagnosis of knowledge bases, we show how a basic diagnostic algorithm can be extended to support hierarchical abstractions in the configuration domain. Finally, experimental results from a prototypical implementation using an industrial constraint based configurator library are presented.