Abstract: The aim of this study is to leverage knowledge inference techniques to enhance knowledge completion in the debugging and testing process of microwave components, thereby improving the completeness and accuracy of fault knowledge graphs, ultimately boosting testing efficiency and reliability. Addressing the challenge of incomplete knowledge in the field of microwave technology, in this paper a knowledge inference method based on the bidirectional encoder representations from Transformers (BERT) is introduced, named MicroReason-BERT. The model first undergoes pre-training on the fault knowledge graph of microwave components using a masked language modeling (MLM) task, enhancing its comprehension of the existing data. Subsequently, the encoder portion of the pre-trained model is used to encode the contextual representations of the head entity, relationship, and tail entity in the triples. On this basis, MicroReason-BERT employs deterministic classifiers and spatial measurement for representation and structural learning, optimizing model performance. To further enhance the model's performance, multiple loss functions are incorporated into the learning process. Experimental results show that MicroReason-BERT outperforms other knowledge inference models on the test set, significantly improving the completeness of the knowledge graph.