To support the decision-making for the maintenance of a bridge, its knowledge graph (KG) has become an effective approach for aggregating information from different sources. However, the construction of a large-scale KG is time-consuming and costs lots of expert labor, even resulting in incomplete entity nodes and relationship expressions, due to the bridge being a complex giant system. Thus, constructing a large scale KG requires a more intelligent way. In this paper, an innovative method for the construction of a large-scale KG with the assistance of large language models (LLMs) is proposed. At the beginning of all, the pre-trained LLMs are involved in extracting the entities and relationships from the documents, which are relative to the maintenance of bridges. Then, an approach for named entities disambiguation is introduced to generate an accurate KG. Finally, a federated method of LLMs ensemble is used to evaluate the validity of obtained KG. The results of experiments indicate that proposed methods are effective in constructing a high quality KG with 80\% accuracy with a dataset including 2.5M Chinese words(1.5M tokens), which is of great significance for the research of bridge knowledge engineering and the development of bridge knowledge graph application technology.