摘要
针对传统D-S证据理论对燃气轮机进行振动故障诊断时会出现一些悖论问题,提出D-S理论改进算法。首先,对采集的燃机振动信号分别从时域和频域进行特征提取,再利用三种不同类型的神经网络模型进行初步诊断,将初步诊断的结果经归一化构建原生证据,然后通过引入证据间的支持矩阵对原生证据进行修正,最后根据改进D-S规则进行决策融合。通过燃气轮机的振动故障诊断实验,证明了该算法能够充分利用各种信息,避免了传统方法出现的悖论现象,提高了燃气轮机振动故障诊断结果的准确性。
To solve the paradoxical problems in vibration fault diagnosis of gas turbines based on the traditional Dempster-Shafer( D-S)evidence theory,an improved D-S algorithm is proposed in this paper. Firstly,the collected gas turbine vibration signals are extracted from the time domain and frequency domain,respectively. Three different types of neural network models are used for preliminary diagnosis,whose results are normalized to construct the primary evidence. By introducing the support matrix between evidences,a primary evidence body is constructed. Finally,making the decision fusion based on the improved D-S rules. The gas turbine vibration fault diagnosis experiment proves that the algorithm can make full use of information,avoid the paradox phenomenon of the traditional method and improve the veracity of gas turbine vibration fault diagnosis.
出处
《电子测量与仪器学报》
CSCD
北大核心
2018年第7期171-179,共9页
Journal of Electronic Measurement and Instrumentation
基金
上海市青年科技英才扬帆计划(16YF1404700)
上海市“科技创新行动计划”社会发展领域项目(16DZ1202500)
上海市科学技术委员会工程技术研究中心(14DZ2251100)资助项目
关键词
燃气轮机
振动故障诊断
改进D-S证据理论
gas turbine
vibration fault diagnosis
improve Dempster-Shafer(D-S) evidence theory
