基于BP神经网络的电弧熔丝增材制造数据库系统  

Database system of wire and arc additive manufacturing based on BP neural network

作　　者：吴悠[1] 潘建刚 廖明潮[1] 张亮 Wu You;Pan Jiangang;Liao Mingchao;Zhang Liang(Wuhan Polytechnic University,Wuhan 430023,China;Xi’an Shiyou University,Xi’an 710065,China;Shenzhen Polytechnic,Institution of Intelligent Manufacturing Technology,Shenzhen 518055,Guangdong,China)

机构地区：[1]武汉轻工大学,武汉430023 [2]西安石油大学,西安710065 [3]深圳职业技术学院,智能制造技术研究院,广东深圳518055

出　　处：《焊接》2022年第11期35-39,共5页Welding & Joining

摘　　要：通过对电弧熔丝增材制造(Wire and arc additive manufacturing, WAAM)单道焊缝试验数据的分类整理,分析用户需求和使用需要,基于python编程语言下的Django框架,采用B/S架构开发了一个电弧熔丝增材制造数据库系统。试验结果表明,该系统采用数据库与算法预测模型结合的方式开发而成,主要设置了用户权限管理、基本打印数据和焊缝形貌预测三大模块,具有存储扩展打印试验数据功能和预测未知工艺参数下焊缝形貌的功能。不同的打印工艺方法引入不同的BP神经网络结构,使用时数据库系统自动读取库内已有的算法模型或根据已有的试验数据训练新的模型,之后录入试验数据会自动对模型重新训练,实现随数据库内试验数据扩展或修正自动适应的参数预测,能够预测未知工艺参数下的焊缝形貌尺寸。最后,基于MIG工艺设计了1组验证试验对数据库的预测功能效果进行检验,熔宽预测误差为1.3%,余高预测误差为1.5%,说明了数据库系统预测功能的可行性。Based on the classification of single pass weld experiments data of wire arc additive manufacturing, this paper analyzes the needs of users and uses them. Based on the Django framework under python programming language, a database system of wire and arc additive manufacturing is developed with B/S architecture. The system is developed by the combination of database and algorithm prediction model. It mainly sets three modules: user authority management, basic experiments data management and weld geometry prediction. It has the function of storing and expanding printing experiment data and predicting weld geometry under unknown welding process parameters. Different printing process methods introduce different BP neural network structures. When in use, the database system automatically reads the existing algorithm model in the database or trains a new model according to the existing experimental data. After entering the experimental data, the model will be retrained automatically, so as to realize the parameter prediction automatically adapted with the expansion or correction of the experimental data in the database, it can predict the weld geometry and size under unknown process parameters. Finally, a set of validation tests are designed based on MIG process to test the effect of the prediction function of the database. The prediction error of melt width is 1.3% and the prediction error of reinforcement is 1.5%, which shows the feasibility of the prediction function of the database system.

关 键 词：数据库 电弧熔丝增材制造 打印工艺 焊缝形貌预测 BP神经网络 

分 类 号：TG44[金属学及工艺—焊接]