Named entity identification method for manufacturing equipment faults based on improved Roberta

Abstract The recognition of manufacturing equipment fault named entities, as the foundation for constructing knowledge graphs in the field of fault diagnosis, has become an efficient identification method in fault diagnosis research. However, traditional methods for recognizing equipment fault named entities have shown insufficient performance in identifying professional vocabulary entities. This paper proposes a model for recognizing manufacturing equipment fault entities, which replaces the traditional Roberta model with the Roberta-wwm pre-trained model based on whole-word masking. The model improves its grasp of contextual information by integrating a bidirectional long short-term memory network and incorporating a multi-head attention mechanism to capture intricate interdependencies among fault entities in manufacturing equipment, and further optimizes the model’s performance. Experimental data from a large volume of fragmented fault texts generated during the operation and maintenance of certain resistance production equipment are used for validation. Through experiments, the F1 score for manufacturing equipment entity recognition is maximally increased by 2.58%, with evaluation metrics showing an F1 score of 98.76%, precision of 98.84%, and recall of 98.73%. This model provides an efficient and accurate solution for handling manufacturing equipment fault texts.