This paper presents the model-based diagnosis in the framework of discrete-event systems. Timed automata are used as a discrete-event representation which is suitable for consistency-based diagnosis. The diagnostic statement is based on the observation whether the measured event-time sequences are consistent with the timed automata. The diagnostic algorithm can be applied online because it determines the fault occurrence recursively for the measured event-time sequences. The result is applied to diagnose valve faults in a chemical process.