In this paper, we present a dynamic analysis approach to increase the understandability of a large softwareintensive system, more particularly to enable the identification of dependencies between its execution entities. This approach analyzes the execution of a software system in a top-down fashion to cope with complexity and uses execution entities such as scenarios, components, and processes rather than code artifacts such as modules, classes, or objects. The approach synchronizes and analyzes two sources of execution information (logging and process activity), and builds architectural views of the system execution, according to a specific metamodel. We have validated this approach on an MRI scanner, a representative large software-intensive system, enabling the identification of dependencies in the execution of its software subsystem.