摘要
针对螺栓拧紧过程滑牙、疲劳以及螺栓拧断等异常模式复杂,同时拧紧序列尺度不一致,拧紧点位繁多导致的异常识别困难问题,采用动态时间规整DTW算法来控制拧紧序列数据的尺度平衡,采用K-means聚类方法划分子数据集,针对拧紧序列长度冗长对梯度计算造成的干扰,应用Attention注意力机制。将该方法用于上汽大众车间的拧紧数据集异常识别,结果表明,分类准确率为92.04%,能有效的识别异常和低质量拧紧曲线,具有重要的应用价值。
In view of the complex abnormal patterns in the bolt tightening process,such as slippage,fatigue,and bolt breakage,and inconsistencies in the scale of the tightening sequence and difficulties in identifying the abnormal paterns due to the large number of tightening points,the dynamic time warping DTW algorithm is used to control the scale balance of the tightening sequence data.The K-means clustering method divides the sub-data set,and applies the Attention mechanism to the interference to the gradient calculation caused by the length of the tightening sequence.The method is used for abnormality identification of tightening data set in the SAIC Volkswagen workshop.The results show that the classification accuracy rate is 92.04%,which can effectively identify abnormal and low-quality tightening curves,and has important application value.
作者
成铭
郑宇
杨文强
CHENG Ming;ZHENG Yu;YANG Wenqiang(School of Machinal Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《机械设计与研究》
CSCD
北大核心
2023年第2期107-111,共5页
Machine Design And Research
基金
国家自然科学基金资助项目(No.52075338)。
关键词
拧紧异常
聚类方法
动态时间规整
LSTM分类
abnormal tightening
clustering methods
dynamic time warping
LSTM classification
