This paper proposes a diagnosis framework for distributed systems based on pomset languages. Diagnosis is performed by projecting these models on a collection of observable labels and then synchronization with an observation. This paper first proposes a new model called boxed pomset languages, which extends classical pomset-based languages as so called High-level Message Sequence Charts. It can describe infinite scenarios, and has good properties with respect to projections. We then give a solution for the event correlation problem (knowing whether two observed alarms are causally related) for pomset languages.