This paper presents a fault diagnosis(detection and isolation) approach for the Ammunition Transport Automation system(ATAS). Due to limited time and information available during its cyclic operation, the on-line fault detection algorithm consists of sequential test logics referring to the normal states, which can be considered as a kind of expert system. If a failure were detected, the off-line isolation algorithm finds the fault location through trained ART2 neural network. By the results of simulations and some on-line field test, it has been shown that the presented approach is effective enough and applicable to related automation systems.