摘要
建立了外圆纵向磨削表面粗糙度的模糊基函数网络(FBFN)预测模型,网络的训练采用自适应最小二乘算法(ALS)。ALS将最小二乘算法和遗传算法相结合,能够自主学习,不用人为干预,FBFN和粗糙度的分析模型相结合,只需少量实验数据便可完成网络的训练,自动产生模糊规则,确定隐含层的节点数。仿真和实验结果表明,FBFN网络结构非常适合粗糙度的预测和控制,采用ALS学习方法比BP算法、传统的遗传算法和正交二乘法等能产生更好的结果。
A framework for modeling cylindrical traverse grinding surface roughness using fuzzy basis function neural networks（FBFN） was constructed with adaptive least-square（ALS） training algorithm. The ALS algorithm, based on the least- square method and genetic algorithm（GA）, was proposed for autonomous learning and the construction of FBFN without any human intervention. Combining the FBFN with the surface roughness analytical model, the proposed algorithm would add a significant fuzzy basis function node at each iteration during the training process based on error reduction measure. Simulation and experimental studies were performed to demonstrate advantages of the proposed modeling framework with the training algorithm in modeling grinding processes. The resuits indicate that tailors the FNFB to predict and control the surface roughness and the new algorithms generate superior results over conventional algorithms such as backpropagation algorithms and conventional GA-based algorithm.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2006年第12期1223-1227,共5页
China Mechanical Engineering
基金
吉林省科技发展计划资助项目(20020632)
关键词
模糊基函数网络
自适应最小二乘法
表面粗糙度预测
外圆纵向磨削
遗传算法
fuzzy basis neural network （FBFN）
adaptive least-squares （ALS）
prediction of the surface roughness
cylindrical traverse grinding
genetic algorithm（GA）
