摘要：
针对单一传感器难以准确描述转子振动情况的问题,提出一种基于改进D-S证据理论的全信息故障诊断方法。结合小波分析和信息熵理论,提取各测点振动信号的全信息故障特征向量。以此作为相应BP神经网络的输入进行模式识别,获得各测点的原始证据。利用冲突性对各测点的原始证据进行预处理,再根据可信度对新证据进行加权平均融合,得到最终诊断结果。实验结果表明,该方法能有效获得原始证据,并降低冲突证据对合成结果的影响,诊断正确率为93%,高于常规BP方法和BP-D-S融合方法。
关键词:
特征提取,
全信息,
BP神经网络,
D-S证据融合,
冲突证据,
故障诊断
Abstract:
Aiming at the problem that a single sensor is difficult to describe the rotor vibration accurately,a full information fault diagnosis method of rotor based on improved D-S evidence fusion theory is proposed.Wavelet analysis and information entropy are combined to extract vibration signal full information fault feature vectors of each measuring point.And these feature vectors are made to be the input of the corresponding BP neural networks to get the original evidences of each measuring point.These original evidences are preprocessed by the conflict degree,and the new evidences are weighted average fused according to the credibility.Experimental results show that this method can obtain original evidences effectively and reduce the influence that comes from the evidence conflict,the accuracy rate of this method can reach 93%,and it is higher than that of the conventional BP method and BP-D-S fusion method.
Key words:
feature extraction,
full information,
BP neural network,
D-S evidence fusion,
conflict evidence,
fault diagnosis
中图分类号: