Maintenance work in large-scale plants becomes the center of public attention, as the rise of cognitive engineering. A simulation method for maintenance training based on Distributed Virtual Environment (DVE) has proposed in this paper. It is composed of two methods: 1. Petri-net and signal model for low-load simulation of virtual objects' motion in VEs, and 2. hierarchical server model for effective management of whole-space data sharing. The methods are proposed for DVEs composed by separate sever topology, which enables flexible and load-balancing configuration. Two prototype training environments based on these methods have been also developed. Effectiveness of the proposed methods has confirmed by accomplishing scenario-based training.