基于人工神经网络的微弧氧化膜层厚度预测模型的建立  

Establishment of Thickness Prediction Model of Micro Arc Oxidation Coating Based on Artificial Neural Network

作　　者：赵东山[1] 牛宗伟[1] 刘洪福[1] 

机构地区：[1]山东理工大学机械工程学院,淄博255049

出　　处：《材料导报》2013年第8期158-162,共5页Materials Reports

基　　金：国家自然科学基金(51005140);山东省自然科学基金(ZR2010EQ037);山东理工大学青年教师发展支持计划经费

摘　　要：在硅酸钠电解液体系中,采用微弧氧化技术在铝合金表面制得了均匀的陶瓷膜。将人工神经网络应用于微弧氧化工艺研究中,借助MATLAB神经网络工具箱,建立了具有4-12-1结构的BP神经网络模型,该模型很好地学习了微弧氧化电解液参数和膜层厚度之间的映射关系;对膜层的厚度进行了预测,并采用正交试验对电解液参数进行了优化。结果表明,该网络收敛速度较快,预测值与实际值基本吻合,平均预测误差仅为1.93%。当Na2SiO3质量浓度为6g/L、H3BO3质量浓度为1.5g/L、KOH质量浓度为0.5g/L、H2O2质量浓度为0.6g/L时,膜层的厚度达到最大值183μm。The uniform ceramic coatings were prepared on the aluminum alloy using micro arc oxidation tech- nique in silicate electrolyte. A BP artificial neural network with 4-12-1 architecture was designed by the MATLAB neural network tools. The relationship between electrolyte parameters and thickness was learned very well by this net- work, and the thickness was predicted using the network model, the electrolytic parameters were optimized using or- thogonal test. The results showed that the prediction values are similar to that of the experimental; the prediction method using BP artificial neural network has higher efficiency and accuracy. Furthermore, the coating thickness has a maximum value 183μm when the electrolyte is prepared by 6 g/L Na2SiO3,1.5 g/L H3BO3,0. 5 g/L KOH,0.6 g/L H2O2.

关 键 词：微弧氧化 BP神经网络 膜层厚度 预测模型 

分 类 号：TG178[金属学及工艺—金属表面处理] TP183[金属学及工艺—金属学]