I've just found an interesting paper by Simeon Simoff and Mary Luo Maher, Deriving Ontology from Design Cases (2001), that provides some ideas on the implementation of a knowledge discovery process that derives "ontological" structured knowledge from a collection of design cases, each of them describe a solution for a different problem. Some thoughts:

• As far as I know, every practical methodology for buinding taxonomies and ontologies includes, perhaps implicitely, a phase related to the examination of a collection of "cases". (This is a sort of truism, if you consider the etymology of the word taxonomy, "method of arrangement").
• The proposed approach makes use of some basic statistical natural language processing techniques, for extracting a base vocabulary for the ontology. I think that this method could work in practice, but in my opinion it requires that the upper part of the domain level ontology is build beforehand: the semi-automatically generated vocabulary would then be useful for populating the lower part of the ontology. The first phase (the definition of the upper level) is often implicit, or hidden, if the user is a domain expert: in this case, the user simply arranges the terms of the vocabulary within his 'a priori' domain-specific core taxonomy.
• Developing or extracting the upper level domain ontology is the hard part, and it's not explicitly covered by this approach; however working with the vocabulary may be useful for making a pre-existing implicit upper level explicit, particularly if it has to be negotiated within a pool of experts, that need to reach an agreement on the whole subject.