摘要
针对螺栓装配过程出现的各类异常,提出一种基于改进XGBoost算法的扭矩角度变化曲线分类方法。针对经过预处理后的数据形成特征曲线,并根据其分布规则建立二分类模型,筛选出异常曲线。通过扭矩和角度曲线的变化关系优化传统的特征提取过程,进而形成螺栓异常状态曲线的特征工程方法,并通过主成分分析法降低了数据冗余。建立基于权值共享矩阵的多级异常状态XGBoost分类模型。实验结果表明,与传统方法相比,依据特征工程方法所建立的多级分类模型在精度方面提高了8%。
A classification method based on the improved XGBoost algorithm for torque angle variation curves is proposed for various types of anomalies in the bolt assembly process.A feature curve is formed for the pre⁃processed data,and a binary classification model is established according to its distribution rules to filter out the abnormal curves.The traditional feature extraction process is optimised by the variation relationship between the torque and angle curves,and then a feature engineering method for the bolt abnormal state curve is formed,and the data redundancy is reduced by principal component analysis.A multi⁃level abnormal,a multi⁃level abnormal state XGBoost classification model based on the weight sharing matrix was established.The experiments result show that the accuracy of the multi⁃level classific⁃ation model based on the feature engineering method is improved by 8%compared with the traditional method.
作者
徐英豪
朱习军
XU Yinghao;ZHU Xijun(School of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061,China)
出处
《电子设计工程》
2023年第16期86-90,共5页
Electronic Design Engineering
基金
山东省产教融合研究生联合培养示范基地项目(2020-19)。
关键词
螺栓异常
状态曲线
XGBoost优化
异常检测
多级分类
bolt abnormality
state curve
XGBoost optimization
anomaly detection
multi⁃level classif⁃ication
作者简介
徐英豪(1996-),男,山东淄博人,硕士研究生。研究方向为:数据挖掘。
