基于优化BP算法的涂装机器人喷涂漆膜厚度成长模型的研究  

A Modified BP Based Model of Thickness Growth of Film Sprayed with Painting Robot

作　　者：杨连生[1,2] 李爱平[1] 祁国栋[1] 罗曦[1] 刘雪梅[1] 

机构地区：[1]同济大学,上海201800 [2]上海外高桥造船有限公司,上海200137

出　　处：《中国造船》2016年第1期193-200,共8页Shipbuilding of China

基　　金：上海经信委资助项目(No.CXY-2013-25);上海科委资助项目(No.14111104400)

摘　　要：为了建立用于涂装机器人喷涂的船体漆膜厚度成长模型,搭建了专门的试验平台。根据正交试验方法设计三因素五水平试验方案,运用BP算法来训练已获取的数据,分别采用莱温伯格-麦夸特算法(LM-BP)和贝叶斯算法(BR-BP)对训练函数进行优化,采用遗传算法(GA)和粒子群算法(PSO)对权值、阈值进行优化。通过对比训练精度、验证精度、测试精度、运算时间以及迭代次数,确定采用GA-LM联合优化BP算法得出的涂装机器人喷涂漆膜厚度成长模型,以满足涂装机器人喷涂船体漆膜厚度控制的质量要求。A specialized experiment platform was established and the orthogonal experiment design with the scheme of 3 factors and 5 levels was used to develop a film thickness growth model in robotically-applied painting process applied in shipbuilding industry. BP algorithm was used to train the data obtained. Levenberg-Marquardt Algorithm（LM-BP）, Bayesian regularization（BR-BP）, and Genetic Algorithm（GA）, Particle Swarm Optimization Algorithm（PSO） was used to training weights and bias values in the BP network. By Comparing training precision, validation precision, test precision, running time and iteration times comprehensively, it is concluded that optimization by LM-BP is the best among the methods mentioned above.

关 键 词：涂装机器人喷涂 漆膜厚度 BP算法 

分 类 号：U671.99[交通运输工程—船舶及航道工程]