Abstract: There are many unfavorable factors in the process of aerospace products assembly such as deformation,rebound of thin-walled parts,multi-level coupling assembly,which will lead to large uncertainty of assembly deviation,so the assembly gross error is difficult to identify accurately.For this reason,a gross error identification method was proposed based on data mining and expert knowledge for the small sample aircraft structure assembly.The proposed method introduced the clustering analysis method of the measurement information theory and the intuitionistic fuzzy entropy method based on uncertainty theory.Using the system clustering method,a mathematical model was established for clustering analysis of assembly deviation measurement data.The similarity between the assembly quality detection data was calculated by Euclidean distance.The group average linkage was introduced to quantify the similarity between data classes for preliminarily identifying assembly gross errors.The weighted intuitionistic fuzzy entropy was proposed to judge the confidence interval of assembly deviation.The intuitionistic fuzzy similarity that accurate quantified of the information between experts was applied to evaluate the rationality of the assembly deviation confidence interval.Data mining and expert knowledge were combined to identify the assembly gross errors comprehensively.The assembly application case verified the accuracy and computational feasibility of the proposed method for the aircraft structure assembly.Compared with the classical Grubbs criterion,the accuracy of gross error identification was improved by 12.5%.

Key words: aircraft structure, assembly, small sample, gross error, knowledge, data mining

摘要： 航空航天产品铆接变形、多层级耦合装配等多因素影响导致其装配偏差不确定度大,装配粗差难以准确识别。为此,针对小批量飞机研制模式,引入测量信息论中的聚类分析法和基于不确定性理论的直觉模糊熵法,提出检测数据和专家知识混合驱动的小样本飞机结构件装配粗差判定方法。首先,利用系统聚类法建立装配偏差测量数据聚类分析数学模型,用欧式距离来量化装配质量检测数据之间的相似度,引入组平均连锁量化检测数据类之间的相似度,进行装配粗差的预筛选。其次,提出加权直觉模糊熵的装配偏差置信区间判定方法。用精确数量化专家之间判定信息的直觉模糊相似度,进行装配偏差置信区间的合理性评估。最后,对数据挖掘和专家知识进行融合,综合识别出装配粗差。通过装配应用案例验证了检测数据和知识混合驱动的飞机结构件装配粗差判定方法的准确性和计算可行性,与经典的格拉布斯粗差判定准则相比,粗差识别准确率提高了12.5%。

关键词: 飞机结构件, 装配, 小样本, 粗差, 知识, 数据挖掘