中国机械工程

• 智能制造 • 上一篇    下一篇

改进PSO_BP_Adaboost算法在尺寸超差故障诊断中的应用

姜春英;康玉祥;叶长龙;于苏洋   

  1. 沈阳航空航天大学机电工程学院,沈阳,110136
  • 出版日期:2018-10-25 发布日期:2018-10-19
  • 基金资助:
    沈阳市科技计划资助项目(F16-216-6-00)

Applications of Improved PSO_BP_Adaboost Algorithm in Fault Diagnosis of Dimension Out-of-tolerance

JIANG Chunying;KANG Yuxiang;YE Changlong;YU Suyang   

  1. School of Mechatronics,Shenyang Aerospace University,Shenyang,110136
  • Online:2018-10-25 Published:2018-10-19

摘要: 利用改进的粒子群优化算法优化BP神经网络,组合多个优化BP网络构成Adaboost强分类器,采用“一对一”分类思想建立了改进PSO_BP_Adaboost多分类器算法,并在部分UCI数据集上进行了有效性验证。实例中,将某零件上相同尺寸、不同位置的4个孔的直径作为BP网络的输入值,利用真实数据进行验证,该算法的分类正确率达到98%,表明提出的改进多分类器算法可有效用于尺寸超差故障诊断。

关键词: 故障诊断, 粒子群优化, 神经网络, Adaboost算法

Abstract: An improved PSO algorithm was used to optimize a BP neural network,and a number of BP networks can form a kind of Adaboost strong classifier.Therefore,an improved PSO_BP_Adaboost multi-classifier algorithm was established with the idea of “one-to-one”,and the validity was proved by some of UCI dataset.The same diameters of 4 holes in different positions on one workpiece were chosen as the input values of the BP network.It was verified by real data that the correct rate of the algorithm reaches 98%.The results show that this improved PSO_BP_Adaboost algorithm may be used effectively for dimension fault diagnosis of out-of-tolerance.

Key words: fault diagnosis, particle swarm optimization(PSO), neural network, Adaboost algorithm

中图分类号: