A novel bidirectional triangular graph neural network model for multi-domain parameter correlation reasoning in production lines
针对生产线多域参数关联推理中单向推理和跨域集成不足的问题，提出双向三角图神经网络模型，在标准数据集上MRR达0.883、Hits@3达97.18%，并通过手机装配线案例验证有效性。
As a complex manufacturing system, a production line exhibits intricate parameter correlations across various design dimensions. Some parameter correlations remain unclear, resulting in a slow design modification process and extensive change scopes. To overcome the limitations of unidirectional reasoning and inadequate cross-domain integration in current graph neural networks, a novel bidirectional triangular graph neural network model is proposed for multi-domain parameter correlation reasoning in production lines. First, a four-dimensional parameter correlation graph encompassing ‘structure-behavior-control-performance’ is constructed, and multi-source parameter feature fusion is achieved via a multi-head graph attention mechanism. Second, a bidirectional triangular reasoning architecture is designed, which incorporates breadth-first search to extract intermediate nodes along reasoning paths and introduces a weighting mechanism to enhance directional modelling. Furthermore, a self-consistency verification strategy is introduced to ensure logical coherence of the reasoning correlations. Inductive reasoning experiments on the WN18RR, FB15K-237, and NELL-995 datasets demonstrate that the proposed model achieves maximum MRR of 0.883 and Hits@3 of 97.18%, outperforming existing reasoning models. The effectiveness of the proposed method is further validated through a case study of a smartphone assembly line. This study offers a new paradigm of data- and knowledge-driven collaboration for deciphering cross-domain parameter coupling mechanisms in production lines, significantly enhancing the efficiency of production line variant design.