苏晓云,汪建新,辛李霞.基于神经网络的铣削大理石表面粗糙度预测模型[J].表面技术,2017,46(8):274-279
基于神经网络的铣削大理石表面粗糙度预测模型
Neural Network-based Prediction Model for Surface Roughness of Milled Marble
投稿时间:2017-05-07  修订日期:2017-08-20
DOI:10.16490/j.cnki.issn.1001-3660.2017.08.044
中文关键词:  粗糙度  大理石  神经网络  粒子群  预测模型
英文关键词:roughness  marble  neural network  particle swarm  prediction model
基金项目:国家自然科学基金资助项目(51365033)
作者单位
苏晓云 包头钢铁职业技术学院,内蒙古 包头 014010 
汪建新 内蒙古科技大学,内蒙古 包头 014010 
辛李霞 包头钢铁职业技术学院,内蒙古 包头 014010 
AuthorInstitution
SU Xiao-yun Baotou Iron and Steel Vocational Technical College, Baotou 014010, China 
WANG Jian-xin Inner Mongolia University of Science and Technology, Baotou 014010, China 
XIN Li-xia Baotou Iron and Steel Vocational Technical College, Baotou 014010, China 
摘要点击次数: 54
全文下载次数: 30
中文摘要:
      目的 利用粒子群优化BP神经网络建立大理石加工表面粗糙度精确预测模型。方法 首先采用不同切削参数进行铣削大理石试验,测量加工表面粗糙度值,同时对粒子群算法进行改进,使惯性权重按指数形式递减,并增加速度扰动系数,利用改进粒子群算法优化BP神经网络,建立铣削大理石表面粗糙度神经网络预测模型。其次使用部分试验数据来训练预测模型,使得到的网络参数让网络可以精确预测表面粗糙度。最后利用其余试验数据验证神经网络预测模型的准确性与可靠性。结果 经过计算得到粒子群优化BP网络算法的预测模型归一化均方差为0.0501,最大相对误差为10.78%,且误差变化较为均匀。经验公式模型归一化均方差为0.1069,最大相对误差为39.64%,误差变化幅度较大。结论 将神经网络模型与经验公式相比较,结果表明,所建网络模型具有较高的预测精度与较强的鲁棒性,对合理选择切削用量以得到理想表面粗糙度有一定参考价值。
英文摘要:
      The work aims to establish a prediction model for surface roughness of processed marble based upon BP neural network in particle swarm optimization method. Firstly, various cutting parameters were tested on the milling marble to obtain roughness value of the machined surface. Meanwhile, the particle swarm optimization (PSO) algorithm was improved to decrease inertia weight progressively in exponential form and increase disturbance coefficient of the velocity. The BP neural network was optimized in modified PSO method to establish a neural network prediction model for surface roughness of milled marble. Secondly, the prediction model was trained by using some test data so that network parameters obtained could predict the surface roughness more accurately. Finally, accuracy and reliability of the improved BP network prediction model were verified based upon other test data. For the prediction model adopting particle swarm optimization BP network algorithm method, normalized mean square error was 0.0501, the maximum relative error was 10.78% and error changed uniformly. For empirical formula model, the normalized mean square error was 0.1069, the maximum relative error was 39.64% and the error variation changed more significantly. Compared with empirical formula, the neural network model exhibits better prediction accuracy and robustness, and it is of certain reference value to achieve better surface roughness by selecting cutting parameters in a reasonable manner.
查看全文  查看/发表评论  下载PDF阅读器
关闭

关于我们 | 联系我们 | 投诉建议 | 隐私保护 | 用户协议

您是第1513556位访问者    渝ICP备10003521号-3

版权所有:《表面技术》编辑部 2014 surface-techj.com, All Rights Reserved

邮编:400039 电话:023-68792193传真:023-68792396 Email: wjqkbm@163.com