This paper describes a method for the recovering of software architectures from a set of similar (but unrelated) software products in binary form. One intention is to drive refactoring into software product lines and combine architecture recovery with run time binary analysis and existing clustering methods. Using our runtime binary analysis, we create graphs that capture the dependencies between different software parts. These are clustered into smaller component graphs, that group software parts with high interactions into larger entities. The component graphs serve as a basis for further software product line work. In this paper, we concentrate on the analysis part of the method and the graph clustering. We apply the graph clustering method to a real application in the context of automation / robot configuration software tools.