Construction of a Cognitive Graph for Intelligent Manufacturing Robot Behavior

The forms of tasks in intelligent manufacturing are diverse, and their operating environment is also dynamically changing. Effective knowledge representation and reasoning are key foundations for enhancing robots' understanding and execution capabilities of complex tasks. Therefore, this article constructs a cognitive graph of robot behavior in the field of intelligent manufacturing based on BERT-BiLSTM-CRF. Firstly, based on the multi-level requirements in robotic operation, a robot behavior corpus is constructed, and the robotic behavior ontology layers Task, Skill, Action, Location, Agent (H), and Object are defined to standardize the semantic structure of behavior. Then, using the BERT model to extract language structure and semantic relationship information, deep-level feature representations are obtained from the text; Capture contextual semantic relationships of behavior through the BiLSTM network, and combine the CRF model to globally constrain and optimize the predicted behavior results. Finally, the extracted triplets are stored using neo4j to form a cognitive graph of intelligent manufacturing robot behavior (IMRBCG). The experimental results show that the P, R, and F1-score of the proposed method are 95.51%, 90.66%, and 93.02%, respectively, effectively extracting 8763 behavioral triplets. This cognitive graph provides technical support for the task of the use of planning and collaboration for intelligent manufacturing robots.