摘要
可靠性是反映再制造零/部件质量及其稳定性的重要指标。相比于新产品零/部件,影响再制造零/部件的可靠性因素多、机理关系复杂,且毛坯剩余价值高,难以开展大量的可靠性试验,使得可靠性数据样本量少,导致再制造零/部件可靠性预测精度低。为此,提出一种基于深度迁移学习的再制造零/部件可靠性预测方法。首先,分析制造阶段个性化工艺过程对再制造零/部件可靠性的影响,结合服役阶段运行数据对可靠性进行预测,通过引入新产品与同一产品不同型号零/部件服役阶段运行数据、制造阶段工艺质量数据进行样本扩充,利用主成分分析法获取影响可靠性的运行时间、状态信息及加工精度等关键特征数据并构建源域数据集;其次,采用卷积神经网络构建源域关键特征数据与可靠性之间的深度学习模型,以再制造零/部件数据为目标域数据集,通过自适应梯度算法将模型迁移应用到再制造零/部件可靠性预测中,以提高预测精度;最后,以某数控机床需要再制造的零/部件为例,对所提出预测方法的有效性进行验证。
Reliability is an important indicator of the quality and stability of remanufactured parts.Compared with new product parts,there are many reliability factors affecting remanufactured parts,complex mechanism relationships,and high residual value of blanks,which makes it difficult to carry out a large number of reliability tests,resulting in a small sample size of reliability data and low reliability prediction accuracy of remanufactured parts.Therefore,a reliability prediction method for remanufactured parts based on deep transfer learning is proposed.Firstly,the influence of personalized process in the manufacturing stage on the reliability of remanufactured parts is analyzed,the reliability is predicted by combining the operation data of the service stage,and the sample expansion is carried out by introducing the service stage operation data and manufacturing stage process quality data of new products and parts of different models of the same product,and the principal component analysis method is used to obtain key characteristic data such as running time,status information,processing accuracy and other key characteristics that affect reliability,and the source domain dataset is constructed.Secondly,a convolutional neural network is used to construct a deep learning model between key feature data and reliability in the source domain,taking the remanufactured parts data as the target domain dataset,and applying the model migration to the reliability prediction of remanufactured parts through the adaptive gradient algorithm to improve the prediction accuracy.Finally,taking the parts that need to be remanufactured in a CNC machine tool as an example,the effectiveness of the proposed prediction method is verified.
作者
郑晨
朱硕
邵智超
江志刚
张华
ZHENG Chen;ZHU Shuo;SHAO Zhichao;JIANG Zhigang;ZHANG Hua(Key Laboratory of Metallurgical Equipment and Control Technology,Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China;Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering,Wuhan University of Science and Technology,Wuhan 430081,China;Ordnance Non-commissioned Officer School,Army Engineering University of PLA,Wuhan 430075,China;Academy of Green Manufacturing Engineering,Wuhan University of Science and Technology,Wuhan 430081,China)
出处
《现代制造工程》
CSCD
北大核心
2024年第8期144-151,共8页
Modern Manufacturing Engineering
基金
国家自然科学基金资助项目(52075396)。
关键词
再制造零/部件
可靠性预测
深度学习
迁移学习
remanufactured parts
reliability prediction
deep learning
transfer learning
