This method is useful when no living species have the structure being studied.
The first step is to develop several mechanical abstractions or models(usually from physical and mechanical first principles) that can be used to predict what a structure should look like, given the constraints and a certain function. The created structures are called paradigm structures. The function of the real structure is that of the closest matching paradigm structure.
Problems arise due to difficulties in building a model that reflects biological reality or when a given model does not generate a sufficiently limited set of predicted structures(eg.multiple functions performed by one structure, or multiple structures which perform the same function). Other concerns are the bias involved from knowing the real structure beforehand, and the fact that only general behaviour can be predicted, not specific behaviour.