Knowledge Graph-Based Language Question and Answer Model Design for Laser Welding

Considering that a large amount of process knowledge in laser welding process design is not effectively utilized and collected, this paper proposes a knowledge graph-based question and answering method for laser welding process design, and puts forward a text representation method combining sentence vectors and word vectors. A deep learning model that combines a Bidirectional Gated Recurrent Unit (BiGRU) with a Multi-Head Attention (MHA) mechanism is used to identify the intention of problem texts. Model comparison and ablation experiments were conducted on a constructed process problem dataset, which proved that the model achieved significantly better results and can provide valuable assistance to process personnel in laser welding process design.