This thesis deals with the diagnosis of max-plus linear systems, a type of discrete event system with linear behavior in algebraic structures called dioids. These models are well suited for processes involving time, synchronizations and event rates, like those in transportation networks and manufacturing plants. They can also be represented as timed event graphs, a class of timed Petri net. Diagnosing max-plus linear systems consists in characterizing the presence and nature of abnormalities in their behavior. In this thesis, abnormal behavior is assumed to be caused by failures, and exclusive attention is given to those that generate delays: reduced capacities, slower assembling times, etc. The diagnosis process involves detecting, locating and estimating the extent of these failures by studying the available observations, that is, the event sequences generated by the system. Different subjects linked to the diagnosis of max-plus linear systems are discussed in this thesis. First, control is used to refine model-based diagnosis results in a process called active diagnosis; the system is modeled by a timed event graph and failure localization is improved by applying inputs specifically adapted to its structure. Then, a model acquisition method is proposed; it starts from event data (observations) presenting both normal and abnormal behavior and yields complete algebraic models for both cases. The evaluation of the quality of the available data in the context of model acquisition is discussed next; the proposed method uses a characterization of the solutions to a linear equation in a specific dioid with the goal of determining if the dynamics of the system are well represented in the data. Finally, a consistency-based method relying on acquired normal and abnormal algebraic models of a system is applied to diagnose its behavior. These processes are then brought together to create and illustrate a data-driven diagnosis technique for max-plus linear systems, providing a sound method to characterize the abnormal behavior of these systems without prior knowledge of their dynamics.