航空学报 > 2000, Vol. 21 Issue (4): 372-375

双BP神经网络在磨损颗粒自动识别中的应用

左洪福1, 吴振锋1, 杨忠2   

  1. 1. 南京航空航天大学机电工程学院,江苏南京210016;2. 南京电力高等专科学校,江苏南京210008
  • 收稿日期:1999-04-20 修回日期:1999-09-03 出版日期:2000-08-25 发布日期:2000-08-25

APPLICATION OF DOUBLE BP NETWORK IN DEBRIS IDENTIFICATION

ZUO Hong fu1, WU Zhen feng1, YANG Zhong2   

  1. 1. Department of Mechanical Engineering, Nanjing Univ. of Aero. and Astro., Nanjing 210016, China;2. Nanjing Institute of Electric Power, Nanjing 210008, China
  • Received:1999-04-20 Revised:1999-09-03 Online:2000-08-25 Published:2000-08-25

摘要: 引入了一套磨粒形态学描述子来提取磨损颗粒的显微形态特征 ,然后以此为输入参数提出了一套BP神经网络 ,对磨损颗粒进行自动识别分类。针对本网络输入参数多 ,网络训练耗时长的问题 ,尝试采用因子模糊化的网络训练方法 ,大幅度提高了神经网络的训练速度 ,并取得了较好的应用效果。

Abstract: A set of morphology descriptors of debris is presented to describe the micro feature of wear particles, and the program to auto identify wear particles by means of artificial neural network (ANN) technique is compiled. During training the network, the fuzzified factor based training technique given out in this paper is used, and the training process is accelerated rapidly. When the network is used to identify the wear particles, the identifying accuracy is higher than 90%, and the identifying speed is very fast. The method by far excels the traditional ones.

中图分类号: