摘要
针对航天结构件等复杂产品工时定额结果不准确的问题,提出了基于特征聚类的结构件数控加工工时预测方法。基于航天结构件的产品特点、材料特点和加工特点,分析了航天结构件数控加工工时影响因素,并提出了基于BERT模型和K-Means算法的工时影响因素特征向量分析方法。基于K-Means聚类算法对BERT模型提取的工艺特征向量进行分组,在此基础上,基于分组结果建立了不同的遗传算法优化的BP神经网络工时预测模型,进而从工时影响因素特征分析和网络结构优化两方面,提高工时定额的准确性。最后,基于历史工艺数据完成了模型训练和预测,验证了所提方法的有效性。
Aiming at the problem of inaccurate results of working-hours quotas for complex products such as aerospace structural parts,this paper proposes a method for predicting the working-hours of CNC machining of structural parts based on feature clustering.Based on the product characteristics,material characteristics and processing characteristics of aerospace structural parts,the factors influencing the CNC machining working-hours of aerospace structural parts are analyzed,and the feature vector analysis method of working-hours influencing factors based on BERT model and K-Means algorithm is proposed.Based on K-Means clustering algorithm,the process feature vectors extracted from BERT model are grouped,and based on this grouping result,different BP neural network working-hours prediction models optimized by genetic algorithm are established,and then the accuracy of working-hours quotas is improved from both working-hours influencing factor feature analysis and network structure optimization.Finally,the model training and prediction are completed based on the historical process data,and the effectiveness of the proposed method is verified.
作者
刘娟
刘检华
庄存波
徐磊
翟思宽
高庆霖
LIU Juan;LIU Jianhua;ZHUANG Cunbo;XU Lei;ZHAI Sikuan;GAO Qinglin(School of Mechanical Engineering,Beijing Institute of Technology,Beijing 100081;Yangtze River Delta Research Institute of Beijing Institute of Technology,Jiaxing 314000;Beijing Satellite Manufacturing Factory,Beijing 100094)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2023年第15期232-246,共15页
Journal of Mechanical Engineering
基金
国家自然科学基金(52005042)
国防基础科研(JCKY2020203B016)
装备预先研究领域基金(80923010101)资助项目。
