Knowledge graph technology is of great significance to the efficient fault diagnosis of automobiles. The existing construction of the knowledge graph of automobile faults has problems such as a poor entity recognition model and the inability to solve nested entities. In order to solve the above problems, an improved nested entity recognition model was proposed by adopting the pre-training semantic model of whole word mask, adding adversarial training, and improving the nested entity recognition model. The experimental results show that the proposed model is 3.56%, 4.08%, and 3.05% higher than the baseline model in terms of the F1 value (F₁), accuracy (P) and recall (R), and also has different degrees of improvement compared with other models, which verifies that the proposed model has a significant effect on entity recognition in the field of automobile maintenance. At the same time, based on the constructed automobile fault knowledge graph, the intelligent question answering prototype system of automobile fault knowledge is realized, and the application prospect of knowledge graph technology in the field of automobile fault diagnosis and maintenance is shown.