It is unclear if an actual system existed at the time this article was written. There occurs at least one mention of an actual system in the equation set assembly section. It cites a limitation where qualitative differential equations must be derived manually from quantitative equations. The rest of the article gives the impression that it is suggesting techniques that system could utilize. A careful reading of the introduction does reveal that simple knowledge compilers were implemented. It is still difficult to tell where talk of those implemented parts stops and system proposal starts.

      I found that the description of the Reaction Wheel Assembly (RWA) overshadows the description of the knowledge compilation system. The RWA should have been used as an example with the compilation system in the forefront. Instead the reader is tossed back and forth between an article about a software system and one about Hubble Telescope parts. It made seeing the line between the two difficult.

      It would have been good if a short explanation of the benefits and limitations of using a bounding box spatial representation were given.

      It seems as though the unary constraints (specifically maximize quantity and minimize quantity) could also be thought of as goals. Since this is part of the representation of the device model and not necessarily part of a design system it is easy to see why the term "constraints" would be used. Depending on the task specific knowledge or resulting system, could they be treated as goals?

      In the section describing redesign goal tree generation an example equation
           [x'] + [y'] + [z'] = 0
is given. The cases where either y or z remains constant and the other must be decreased for x to increase are considered. Why is the case where both y and z can change not considered?