A Information Extraction Method for Electromechanical Equipment Fault Diagnosis Knowledge Graphs Based on Deep Learning

In the process of fault diagnosis for electromechanical equipment, a significant amount of equipment-related record information remains underutilized, leading to inefficiencies in fault reasoning and diagnosis. To address this issue, we propose a method for electromechanical equipment information extraction based on deep learning and knowledge graphs. This method extracts critical information from expert knowledge, equipment operational status, and maintenance logs to construct a knowledge graph suitable for fault diagnosis of electromechanical equipment. For information extraction, we propose an Enhanced BERT-BiLSTM-CRF model (EBBC-DAM-PATM), which enhances the traditional BERT-BiLSTM-CRF (BBC) deep neural network by introducing a Data Augmentation Module (DAM) and Projection Gradient Descent Adversarial Training Module (PATM). In the named entity recognition task for electromechanical equipment, this model demonstrates superior recall and F2 scores compared to the traditional BERT-BiLSTM-CRF model, along with better stability. Ultimately, we validated the effectiveness of this information extraction method and constructed a specialized knowledge graph for fault diagnosis of electromechanical equipment.