RBF神经网络在激光熔覆钴基合金涂层稀释率预测中的应用  

Application of RBF Neural Network in the Prediction of Dilution Ratio of Laser Cladding Cobalt Based Alloy Coating

作　　者：徐家乐 谭文胜[1] 胡增荣 王松涛 周建忠[3] Xu Jiale;Tan Wensheng;Hu Zengrong;Wang Songtao;Zhou Jianzhong(Department of Intelligent Equipment,Changxhou College of Information Technology,Changzhou,Jiangsu 213164,China;School of Rail Transportation,Sochow University,Suzhou.Jiangsu 215131,China;School of Mechanical Engineering,Jiangsu University.Zhenjiang,Jiangsu 212013,China)

机构地区：[1]常州信息职业技术学院智能装备学院,江苏常州213164 [2]苏州大学轨道交通学院,江苏苏州215131 [3]江苏大学机械工程学院,江苏镇江212013

出　　处：《应用激光》2021年第4期752-757,共6页Applied Laser

基　　金：常州市科技计划项目(CJ20210034);江苏省高校自然科学研究面上项目(20KJB460016);江苏省高职院校青年教师企业实践培训资助项目(2020QYSJ129);常州信息职业技术学院校级科研课题资助项目(CXKZ201906Y);常州信息职业技术学院2020年度校级科研平台项目(KYPT202003Z)。

摘　　要：应用光纤激光器在GCr15轴承钢表面激光熔覆制备钴基合金涂层,运用正交试验研究激光功率、扫描速度与送粉率等工艺参数对熔覆层稀释率的影响,通过极差分析确定影响稀释率的关键因素,基于正交试验结果采用RBF神经网络建立激光工艺参数与熔覆层稀释率之间的预测模型,并用测试样本对网络进行检验。结果表明:对稀释率影响最显著的因素为送粉率,由于粉末熔化存在所需能量阈值和"热屏蔽"效应,稀释率并非随着激光功率和送粉率的增大而一直增大或减小,而是存在波动现象;随着扫描速度的增大,稀释率不断变小,稀释率的变化由各熔覆工艺参数交互作用决定。经过试验数据训练后的RBF神经网络模型可以实现对不同激光工艺参数下制备的钴基涂层稀释率进行预测,预测值和试验测得值之间的相对误差都在6%以内,具有较高的预测能力。The Co-based alloy coating was prepared by laser cladding on GCr15 bearing steel with fiber laser. The effects of laser power, scanning speed and powder feeding rate on the dilution ratio of the coating were studied by orthogonal experiment. According to the results of orthogonal test, RBF neural network was used to establish the prediction model between the laser process parameters and the dilution rate of cladding layer, then the network was tested with test samples. The results show that the most significant factor affecting the dilution rate is the powder feeding rate. Due to the energy threshold and "heat shield" effect of powder melting, the dilution rate does not increase or decrease with the increase of laser power and powder feeding rate, there are some fluctuations;With the increase of scanning speed, the dilution ratio decreases gradually, and the variation of dilution ratio is determined by the interaction of various cladding process parameters. The RBF neural network model trained by the experimental data can predict the dilution ratio of the Co-based cladding layer prepared under different laser processing parameters. The relative error between the predicted value and the measured value is within 6%, which has a high prediction ability.

关 键 词：激光熔覆 钴基合金涂层 稀释率 正交试验 RBF神经网络 

分 类 号：TN249[电子电信—物理电子学]