摘要
精准评估影响定制化印制电路板质量的关键工序,可更好地指导企业精益管理产品品质,但企业常用方法难以利用工序关联实体间的深层语义。为解决上述问题,构建工序关联实体知识图谱并提出GENI-SD模型评估工序重要性。首先,使用知识图谱表征工序关联实体间的关系,并采用图神经网络模型GENI对工序节点进行重要性评估;然后,引入基于改进谓词感知注意力机制的采样模块和考虑邻边方向的分数聚合,改进GENI容易聚合到噪声邻节点且未考虑邻边方向的不足,建立工序重要性评估新模型GENI-SD。最后,以PCB车间真实数据开展实验验证,结果显示GENI-SD优于其他对比模型,且得到的Spearman和NDCG@10指标值较GENI所得结果分别提高6.78%和0.71%。该研究为工序重要性评估提供了新的有效方法。
Accurately estimating the critical operations that affect the quality of customized printed circuit board(PCB)can better guide enterprises in lean management of product quality.However,the common methods of enterprises are difficult to use the deep semantics between operation-related entities.To solve this problem,this paper constructed a knowledge graph of operation-related entities and proposed GENI-SD to estimate the importance of PCB operation.Firstly,this paper employed the knowledge graph to represent the relationship between operation-related entities,and utilized the graph neural network model GENI to estimate the importance of operation nodes.Secondly,this paper improved GENI and established a novel model called as GENI-SD by introducing a sampling module based on an improved predicate-aware attention mechanism and a score aggregation considering the direction of adjacent edges,thus to resolve noisy neighbors and non-direction aggregation shortcoming of GENI.Finally,it conducted experiments on the real data of PCB workshop.The results demonstrate that GENI-SD outperforms SOTA and obtains 6.78%and 0.71%improvement in Spearman and NDCG@10 metrics comparing to that of GENI respectively.This study provides a novel and effective method for estimating the importance of operation.
作者
劳景春
金鸿
吕盛坪
李文强
Lao Jingchun;Jin Hong;Lyu Shengping;Li Wenqiang(College of Engineering,South China Agricultural University,Guangzhou 510642,China)
出处
《计算机应用研究》
CSCD
北大核心
2023年第5期1441-1446,共6页
Application Research of Computers
基金
国家自然科学基金面上项目(52275487)
广东省自然科学基金资助项目(2021A1515012395)。
关键词
印制电路板
知识图谱
图神经网络
采样
分数聚合
printed circuit board
knowledge graph
graph neural network
sampling
score aggregation
