基于复合简化模糊ARTMAP的故障诊断

Fault Diagnosis by Voting Algorithm Based on Multiple Simplified Fuzzy ARTMAP

简单模糊ARTMAP(SFAM)是一种功能强大的神经网络,可以用于故障诊断等分类问题;然而,单独SFAM存在对样本训练次序敏感的性能缺陷。为改善SFAM的性能提出表决融合方法。首先把用于故障诊断的特征向量规格化并提交给数个SFAM,用求和规则表决算法融合多个SFAM诊断结果。在此处理过程中,用置信度作为融合的加权因子,融合结果作为故障诊断系统的最终判决。通过用轴承故障诊断案例进行实验测试,结果表明表决融合方法对故障诊断的有效性。

Simplified fuzzy ARTMAP （SFAM） is a powerful neural network which can be used as a classifier to solve the fault diagnosis problem. But the performances of SFAM are sensitive to the samples training sequence. In order to overcome the drawback of individual SFAM, a voting integrated method was proposed to improve the performance of the SFAM. The feature vector for fault diag- nosis was normalized and submitted to several SFAMs. The sum rule voting algorithm was performed to integrate the SFAMs diagnosis results. In this procedure, the credits were employed as weight factors. The integrated result was the final conclusion of the fault diagno- sis system. The experimental results show that the voting algorithm is valid for fault diagnosis.