Shape idealization transformations are very common operations when adapting a CAD component to FEA requirements. Here, an idealization approach is proposed that is based on generative shape processes used to decompose an initial B-Rep solid, i.e., extru-sion processes with material addition are used to seg-ment a solid. The corresponding extrusion primitives form the basis of candidate sub domains for idealiza-tion and their connections conveyed through the gen-erative processes they belong to, bring robustness to set up the appropriate connections between idealized sub domains. This is made possible because the con-nections between extrusion primitives have an explicit geometric representation and can be used to bound the connections between idealized sub domains. Taking ad-vantage of an existing construction tree as available in a CAD software does not help much because it may be complicated to use it for idealization processes because this tree structure is not unique. Using generative pro-cesses attached to an object that are no longer reduced to a single construction tree but to a graph containing all non trivial construction trees, is more useful for the engineer to evaluate variants of idealization. From this automated decomposition, each primitive is subjected to a morphological analysis to define whether it can idealized or not. Subsequently, geometric interfaces be-tween primitives form also a graph that can be used to process the connections between the idealized sub do-mains generated from the primitives. These interfaces are taken into account to determine more precisely the idealizable sub domains and their contours when primi-tives are incrementally merged to come back to produce the global morphological analysis of the initial object. A user defined threshold is used to tune the morpho-logical analysis with respect to further user parameters. Finally, the idealizable sub domains and their connec-tions are processed to locate the mid-surfaces and con-nect them using generic criteria that the user can tune locally using complementary criteria.