At present, many problems of intelligent manufacturing are uncertain and complex, which cannot be solved effectively by expert experience and mechanism models. Therefore, the human-cyber-physical data fusion modeling in complex manufacturing environments is studied, and the ontology is proposed as the approach of the human-cyber-physical data fusion. In the extraction of triplets, a model named ErBERT based on entity-relation joint extraction is proposed, which is different from the traditional pipeline extraction. After word serialization by the pre-training model BERT, the model completes entity recognition and relationship classification by max pooling, fully connection and Softmax, and obtains the extracted human-cyber-physical triplets. The extracted triplets are mapped to OWL files according to rules, and finally stored in the graph database to realize ontology construction. The experimental result shows that the triplets extracted by the ErBERT have good accuracy and achieve the goal of fusion of human-cyber-physical data through ontology, which provides theoretical method support for realizing the ternary collaborative decision-making and optimization of human-cyber-physical data.