摘要
针对KUKA机器人焊接质量监测工作量大、抽样样本小的特点,提出一种基于后验预测分布的贝叶斯动态监控方法。从历史数据中选择合适的数据,计算先验分布的超参数;再结合当前样本构建服从负二项分布的后验预测分布,实时计算控制限,实现对焊接质量的动态监测。结果表明:该方法优于传统似然估计法,有更强的异常检出力和稳健性。
Aiming at the characteristics of large workload and small sample size in KUKA robot welding quality monitoring,a Bayesian dynamic monitoring method based on posterior prediction distribution was proposed.The appropriate data were selected from the historical data to calculate the super parameters of the prior distribution;combined with the current samples,a posterior predictive distribution following the negative binomial distribution was constructed to calculate the control limits in real time to realize the dynamic monitoring of welding quality.The results show that the method is superior to the traditional moment estimation method,and has stronger anomaly detection power and robustness.
作者
吴姝
何雨飞
屈挺
胡楷雄
WU Shu;HE Yufei;QU Ting;HU Kaixiong(School of Logistics Engineering,Wuhan University of Technology,Wuhan Hubei 430063,China;School of Intelligent Systems Science and Engineering,Jinan University,Zhuhai Guangdong 519070,China)
出处
《机床与液压》
北大核心
2022年第9期7-12,共6页
Machine Tool & Hydraulics
基金
国家自然科学基金青年科学基金项目(11701437)
国家自然科学面上基金项目(51875251)
“广东特支计划”本土创新创业团队项目(2019BT02S593)
广州市创新领军团队项目(201909010006)。
关键词
焊接机器人
质量控制
贝叶斯理论
后验预测分布
Welding robots
Quality control
Bayesian theory
Posterior predictive distribution
