This paper deals with active diagnosis in Timed Event Graphs represented as (Max,+)-Linear Systems. Based on the control theory of (max,+)-linear system, an offline algorithm is proposed that aims at better localizing the source of detected time failures within the supervised system. The proposed algorithm is divided into several steps that consists in successively synthesizing, testing and analyzing the injection of new input flows in the system to provide a better localization of the detected time failure.